Electrophysiological Recording Techniques (Neuromethods, 54)

Neuromethods Series Editor Wolfgang Walz University of Saskatchewan Saskatoon, SK, Canada For other titles publish...

Author: Robert P. Vertes | Robert W. Stackman Jr.

30 downloads 1004 Views 21MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Neuromethods

Series Editor Wolfgang Walz University of Saskatchewan Saskatoon, SK, Canada

For other titles published in this series, go to www.springer.com/series/7657

Electrophysiological Recording Techniques Edited by

Robert P. Vertes Center for Complex Systems and Brain Sciences, Florida Atlantic University, Boca Raton, FL, USA

Robert W. Stackman Jr. Department of Psychology and Center for Complex Systems and Brain Sciences, Florida Atlantic University, Boca Raton, FL, USA

Editors Robert P. Vertes, Ph.D. Center for Complex Systems and Brain Sciences Florida Atlantic University 777 Glades Road Boca Raton, Florida 33431-0991 USA [email protected]

Robert W. Stackman, Jr. Ph.D. Department of Psychology and Center for Complex Systems and Brain Sciences Florida Atlantic University 777 Glades Road Boca Raton, Florida 33431-0991 USA [email protected]

ISSN 0893-2336 e-ISSN 1940-6045 ISBN 978-1-60327-201-8 e-ISBN 978-1-60327-202-5 DOI 10.1007/978-1-60327-202-5 Springer New York Dordrecht Heidelberg London Library of Congress Control Number: 2010938917 © Springer Science+Business Media, LLC 2011 All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Humana Press, c/o Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. While the advice and information in this book are believed to be true and accurate at the date of going to press, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect to the material contained herein. Printed on acid-free paper Humana Press is part of Springer Science+Business Media (www.springer.com)

Preface to the Series Under the guidance of its founders Alan Boulton and Glen Baker, the Neuromethods series by Humana Press has been very successful since the first volume appeared in 1985. In about 17 years, 37 volumes have been published. In 2006, Springer Science + Business Media made a renewed commitment to this series. The new program focuses on methods that are either unique to the nervous system and excitable cells or which need special consideration to be applied to the neurosciences. The program strikes a balance between recent and exciting developments like those concerning new animal models of disease, imaging, in vivo methods, and more established techniques. These include immunocytochemistry and electrophysiological technologies. New trainees in neurosciences still need a sound footing in these older methods in order to apply a critical approach to their results. The careful application of methods is probably the most important step in the process of scientific inquiry. In the past, new methodologies led the way in developing new disciplines in the biological and medical sciences. For example, Physiology emerged out of Anatomy in the 19th century by harnessing new methods based on the newly discovered phenomenon of electricity. Nowadays, the relationships between disciplines and methods are more complex. Methods are now widely shared between disciplines and research areas. New developments in electronic publishing also make it possible for scientists to download chapters or protocols selectively within a very short time of encountering them. This new approach has been taken into account in the design of individual volumes and chapters in this series. Wolfgang Walz

v

Preface The application of neurophysiological methods to the study of the brain–behavior relationships represented a major advance to the field of neuroscience when it was in its infancy. Modern neuroscientists now have a great deal more technology available to them than ever before; and consequently the field of neurophysiology has grown considerably. Following a tradition set by the original Neuromethods series, this book presents a current view of the widespread application of electrophysiological methods to the study of the brain and to the problem of brain–behavior relationships. The book has been organized to display the research topics to which modern neurophysiological methods have been applied. Such applications range from recordings of single neurons and ensembles of neurons to recordings of field potentials within discrete brain regions to field potential recordings across multiple brain areas. In the interest of continuity, the present volume begins with a chapter by Stan Leung who contributed to an earlier volume (Volume 15) of the series. The present chapter by Leung describes the basic principles of field potential recording/analysis and current source density (CSD) analysis. Using experimental data as well as model systems of hippocampal pyramidal cells, Leung nicely illustrates the differential patterns of current flow (sources/sinks) along pyramidal cell dendrites and soma to the activation of different segments of the neuron (basal dendrites, soma, proximal or distal regions of apical dendrites), depicting averaged evoked potentials and their derived CSD profiles. Complementing Leung’s chapter, Ding, Schroeder, and colleagues (Chen, Dhamala, Bollimunta, Schroeder, Ding) describe an in vivo procedure for CSD analysis of ongong, non-triggered, neural activity. The method termed ‘phase realigned averaging technique (PRAT)’ extracts generally low amplitude signals from continuous streams of activity. The procedure involves parceling, phase realigning, and averaging ongoing activity at select frequencies to determine spatiotemporal properties, such as peak current flow within defined cortical fields from awake animals. Among other things, the method allows for a determination of the relationship of endogenous activity (e.g., alpha rhythm) to behaviorally relevant events, such as sensory or motor responses to external stimuli. The chapter by Pinault describes a technique that permits the discrete labeling of individual neurons during simultaneous extracellular recording, an important tool for defining the discrete connectivity of neurons whose physiological properties have been identified. Four chapters, those from Fenton, Jeffery and Donnett; Kuang and Tsien; Hampson, Simeral, Berger, Song, Chan and Deadwyler; and Stackman provide details of new strategies that apply to in vivo single-unit recording from freely moving rodents. Fenton, Jeffery, and Donnett present the challenges that face the design of wireless recording systems. They describe the advantages of their new digital telemetry (DT) system over other analog wireless systems and outline two applications for DT – tetherless recordings from freely moving rodents during truly unrestricted behavioral performance and an epilepsy monitoring system for use in humans. Kuang and Tsien’s chapter addresses two of the exciting challenges emerging in the field, that of how to acquire high-density ensemble neuronal activity from wild type and genetically engineered mice, and second, how to

vii

viii

Preface

analyze these data. The chapter by Stackman addresses the inherent challenges of relating neuronal firing patterns of limbic neurons to distinct behavioral sequences. The chapter focuses on the rodent head direction cell as a model system to delineate the degree to which directional correlates of single-unit activity relate to spatial navigation. In a computational vein, the chapter by Bressler is devoted to event-related potentials (ERPs) while that by Albo and colleagues (Albo, Viana Di Prisco, Vertes) examines spike– field interactions. As Bressler points out, local field potentials (LFPs) recorded with depth electrodes (intracortical) are 1–2 mV in amplitude, whereas those recorded from the scalp (e.g., EEG recordings in humans) are 10–50 mV in amplitude. The latter generally requires special procedures for detection and analysis, particularly if internally or endogenously generated. The chapter by Bressler, then, provides an in-depth description of various procedures for characterizing event related potentials (triggered and nontriggered), with special attention to state of the art time and frequency domain analytical methods that are particularly useful in situations in which standard ensemble averaging techniques may be inappropriate. The chapter by Albo et al. describes current methods for unit-field (and field-field) analysis, or the application of spike-field coherence techniques to the study of unit-field oscillations. The chapter provides a nice overview of the advantages/disadvantages of various approaches to assessing functional interactions among synchronously occurring signals (spike trains and field potentials) across the brain. As a direct application of some of the techniques, they describe their findings showing a three way interaction (coherence) between theta rhythmic units in the anterior thalamus and theta oscillations in the hippocampus and retrosplenial cortex, suggesting that hippocampus may drive the anterior thalamus, which in turn rhythmically paces the retrosplenial cortex, with implications for the role of theta in limbic functions. Recording from ensembles of hippocampal neurons (15-35 cells of CA1/CA3), Deadwyler, Berger, and colleagues (Hampson, Simeral, Berger, Song, Chan, Deadwyler) describe a “closed loop system” which distinguishes among the separate behavioral components of a two choice delayed nonmatch to sample (DNMS) task, and then uses ensemble activity at phases of the task to both predict choice behavior and modify it during task performance. In effect, ensemble activity (or codes) was used to adjust delay times (between sample and choice) during ongoing trials to improve performance on those trials. Specifically, depending on the relative strength (or efficacy) of the ensemble code in the sample phase of the task, the delay between sample and nonmatch task phases could be shortened or lengthened, thereby improving performance. It is well recognized that the septum and hippocampus are strongly interconnected and together serve as a functional unit generating the hippocampal theta rhythm. Theta serves a well-recognized role in mnemonic functions. In a major advance in examining septo-hippocampal interactions, Williams and colleagues (Goutagny, Jackson, Williams) have developed a remarkable in vitro preparation in which the septum and hippocampus are simultaneously dissected (with connections between them intact) and kept viable for at least 8 h. In addition, with a barrier placed between them, the two structures can be independently manipulated to assess the effects of one on the other. Using this preparation, Williams and colleagues have confirmed the pronounced septal influence on the hippocampus in the modulation of theta, and further showed that hippocampal theta activity, in turn, exerts a strong driving influence on the septum. Helen Mayberg is a pioneer in the use of deep brain stimulation (DBS) to treat depression. The chapter by Mayberg and Holtzheimer begins with a review of background material (mainly imaging studies) that led to the use of DBS for major depressive disorders (MDD). In effect, they (and others) found that certain regions of the

Preface

ix

frontal cortex, such as the subcallosal cingulate cortex (SCC) were hyperactive in MDD, whereas other areas such as dorsolateral prefrontal or posterior cingulate cortices were hypoactive in MDD, and that successful antidepressant treatment normalized activity in these regions. This suggested a critical role for the frontal/prefrontal cortex (particularly the SCC) in MDD. Mayberg and Holtzheimer proceed to describe in detail the specific procedures used for DBS of the SCC and summarize the extremely promising results that have been obtained to date with the technique with two groups of patients with treatment resistant depression. DBS is not only a cutting edge procedure for the treatment of depression, but also used in conjunction with other methods and has the potential to define an extended circuitry responsible for MDD. It is daunting to take on the challenge of describing the “current” state of any field of science. This is especially the case for neurophysiological techniques since this area is in a near constant state of improvement. This collection of chapters provides a clear indication of how modern technological advances have influenced the study of the neurophysiological substrates for behavior. As one will find in reviewing this volume, current challenges in electrophysiological techniques will most certainly be conquered by the next generation of improvements in technology and analysis. Robert P. Vertes Robert W. Stackman, Jr.

Contents Preface to the Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii 1 Field Potential Generation and Current Source Density Analysis . . . . . . . . . . . . . L. Stan Leung 2 Current Source Density Analysis of Ongoing Neural Activity: Theory and Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yonghong Chen, Mukesh Dhamala, Anil Bollimunta, Charles E. Schroeder, and Mingzhou Ding 3 The Juxtacellular Recording-Labeling Technique . . . . . . . . . . . . . . . . . . . . . . . . . Didier Pinault 4 Neural Recording Using Digital Telemetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . André A. Fenton, Kathryn J. Jeffery, and James G. Donnett 5 Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hui Kuang and Joe Z. Tsien 6 Behavioral Correlates of Neuronal Activity Recorded as Single-Units: Promises and Pitfalls as Illustrated by the Rodent Head Direction Cell Signal . . . . Robert W. Stackman Jr. 7 Event-Related Potentials of the Cerebral Cortex . . . . . . . . . . . . . . . . . . . . . . . . . Steven L. Bressler 8 Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow Within Papez’s Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Zimbul Albo, Gonzalo Viana Di Prisco, and Robert P. Vertes 9 Cognitively Relevant Recoding in Hippocampus: Beneficial Feedback of Ensemble Codes in a Closed Loop Paradigm . . . . . . . . . . . . . . . . . . . . . . . . . . Robert E. Hampson, John D. Simeral, Theodore W. Berger, Dong Song, Rosa H.M. Chan, Vasilis Z. Marmarelis, and Sam A. Deadwyler 10 An Intact Septo-Hippocampal Preparation for Investigating the Mechanisms of Hippocampal Oscillation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Romain Goutagny, Jesse Jackson, and Sylvain Williams 11 Targeted Modulation of Neural Circuits: A New Treatment Strategy for Neuropsychiatric Disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Helen S. Mayberg and Paul E. Holtzheimer Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xi

1

27

41 77

103

127 169

191

215

241

257 281

Contributors Zimbul Albo • Department of Neurology, Baylor College of Medicine, Baylor University, Houston, TX, USA Theodore W. Berger • Department of Biomedical Engineering, Viterbi School of Engineering, University of Southern California, Los Angeles, CA, USA Anil Bollimunta • J. Crayton Pruitt Family Department of Biomedical Engineering, University of Florida, Gainesville, FL, USA Steven L. Bressler • Center for Complex Systems and Brain Sciences, Charles E. Schmidt College of Science, Florida Atlantic University, Boca Raton, FL, USA Rosa H.M. Chan • Department of Biomedical Engineering, Viterbi School of Engineering, University of Southern California, Los Angeles, CA, USA Yonghong Chen • J. Crayton Pruitt Family Department of Biomedical Engineering, University of Florida, Gainesville, FL, USA Sam A. Deadwyler • Department of Physiology and Pharmacology, School of Medicine, Wake Forest University, Winston-Salem, NC, USA Mukesh Dhamala • Department of Physics and Astronomy, and Georgia State University Neuroscience Institute, Georgia State University, Atlanta, GA, USA Mingzhou Ding • J. Crayton Pruitt Family Department of Biomedical Engineering, University of Florida, Gainesville, FL, USA James G. Donnett • Axona Ltd., St. Albans, UK André A. Fenton • Department of Physiology and Pharmacology, The Robert F. Furchgott Center for Neural and Behavioral Science, SUNY Downstate Medical Center, Brooklyn, NY, USA Romain Goutagny • Department of Psychiatry, McGill University, Montreal, QC, Canada Robert E. Hampson • Department of Physiology and Pharmacology, School of Medicine, Wake Forest University, Winston-Salem, NC, USA Paul E. Holtzheimer • Department of Psychiatry, Emory University School of Medicine, Atlanta, GA, USA Jesse Jackson • Department of Psychiatry, McGill University, Montreal, QC, Canada Kathryn J. Jeffery • Division of Psychology and Language Sciences, Department of Cognitive, Perceptual and Brain Sciences, Institute of Behavioural Neuroscience, University College London, London, UK Hui Kuang • The Key Laboratories of MOE and STCSM and College of Life Sciences, Shanghai Institute of Brain Functional Genomics, East China Normal University, Shanghai, China L. Stan Leung • Department of Physiology and Pharmacology, The University of Western Ontario, London, ON, Canada Vasilis Z. Marmarelis • Department of Biomedical Engineering, University of Southern California, Los Angeles, CA, USA

xiii

xiv

Contributors

Helen S. Mayberg • Departments of Psychiatry and Neurology, Emory University School of Medicine, Atlanta, GA, USA Didier Pinault • Faculté de Médecine, INSERM U666, Physiopathologie Clinique et Expérimentale de la Schizophrénie, Université Louis Pasteur, Université de Strasbourg, Strasbourg, France Charles E. Schroeder • Nathan Kline Institute for Psychiatric Research, Orangeburg, NY, USA; Columbia University College of Physicians and Surgeons, New York, NY, USA John D. Simeral • Division of Biology and Medicine, Department of Neuroscience, Brown University, Providence, RI, USA Dong Song • Department of Biomedical Engineering, Viterbi School of Engineering, University of Southern California, Los Angeles, CA, USA Robert W. Stackman, Jr. • Department of Psychology, and Center for Complex Systems and Brain Sciences, Charles E. Schmidt College of Science, Florida Atlantic University, Boca Raton, FL, USA Joe Z. Tsien • Brain and Behavior Discovery Institute, School of Medicine, Medical College of Georgia, Augusta, GA, USA; Department of Neurology, School of Medicine, Medical College of Georgia, Augusta, GA, USA Robert P. Vertes • Center for Complex Systems and Brain Sciences, Charles E. Schmidt College of Science, Florida Atlantic University, Boca Raton, FL, USA Gonzalo Viana Di Prisco • Department of Neurology, Baylor College of Medicine, Baylor University, Houston, TX, USA Sylvain Williams • Department of Psychiatry, McGill University, Montreal, QC, Canada

Chapter 1 Field Potential Generation and Current Source Density Analysis L. Stan Leung Abstract The basic principles underlying field potential generation and the application of current source density (CSD) analysis are outlined in this chapter. Currents in the brain are mainly derived from synaptic or action currents flowing in a closed loop, traversing both intracellular and extracellular media. Extracellular currents generate the field potentials, with a spatial organization of an open or a closed field that may be standing or traveling. CSD analysis is the method used to derive the macroscopic sources and sinks that generate a potential field. Assuming that the medium is homogeneous and resistive, CSD can be approximated by a second-order derivative of the field potential. When the activation is spatially extensive, the current may essentially flow in one or two dimensions, and the CSD may be approximated using one- or two-dimensional mapping. The field potentials should first be mapped regularly at an adequate interval, over an appropriate spatial extent. A multichannel electrode array offers accurate sampling intervals, and the field potentials can be sampled simultaneously in one or two dimensions. Examples of potential fields and CSDs in a layered cortical structure (hippocampal CA1 area) are illustrated, with different fields generated by basal or apical dendritic excitation, proximal and distal dendritic excitation, proximal inhibition, and synchronous action potentials (population spikes). Generation of field potentials from sinks and sources of neuronal cables, arranged in a particular geometry, may be used to predict the CSD profiles. Successful application of CSD analysis would facilitate the understanding of neuronal dynamics, synaptic transmission, and plasticity in cortical structures. Key words: Current source density, Field potential, Dendritic excitation, Population spike, Inhibitory field, Silicon probe, Cable theory, Electroencephalogram, Field mapping

1. Introduction As indicated by its name, field potential is a measure not localized to a small volume. Instead, it is spatially distributed in a conductive medium, so called a volume conductor. For neural activity, the volume conductor includes the brain, blood vessels and meninges, skull, scalp, and the body. The types of field potentials Robert P. Vertes and Robert W. Stackman Jr. (eds.), Electrophysiological Recording Techniques, Neuromethods, vol. 54, DOI 10.1007/978-1-60327-202-5_1, © Springer Science+Business Media, LLC 2011

1

2

Leung

include spontaneous, induced, or evoked electrical activities that are recorded on the scalp, on the surface of the brain, or inside a particular brain structure. An electroencephalogram (EEG) is a spontaneous electric activity conventionally recorded on the scalp of humans. Local field potentials are extracellular recordings in a particular brain structure. Changes in currents or electric field also give rise to a magnetic field that is recorded as a magnetoencephalogram (MEG). With the advent of recordings of single cells and patches of dendrites (1), it may be asked whether field potentials have a role to play in modern neuroscience. The answer is affirmative. Human EEG continues to have a major importance in diagnosis of neurological disease and epilepsy, in particular. Evoked potentials also have diagnostic value for disorders of the sensory or motor pathways. Animal and human EEG/MEG and local field potentials have played a major role in revealing network oscillations and synchrony. There are many highlights in field potential research. Hans Berger’s discovery of the alpha rhythm in the 1920s was followed by Edgar Adrian’s confirmation. Adrian subsequently reported an induced rhythm of 30–90Hz (now called the gamma rhythm) in the olfactory bulb. The gamma rhythm has been recognized as a means of synchronizing distributed neurons within and across brain structures. Oscillatory field potentials and synchrony have been intensively studied since the landmark reports of gamma in the visual cortex (2) and thalamus (3), following pioneering studies in the olfactory system (4). The theory of field potentials is not covered in introductory or advanced neurophysiology textbooks, with the exception of Johnston and Wu (5) and a section written by Brinley (6). A basic coverage of field potentials was given by Hubbard et al. (7), a more specialized treatment was given by Freeman (4) and Niedermeyer and Lopes da Silva (8), and a decidedly more biophysical coverage was given by Plonsey (9) and Nunez (10). A major neurophysiological problem is to find the sources and sinks of the currents that generate the field potential. The current source per volume is defined as the current source density (CSD). In general, the inverse problem of finding the CSDs from the potential field has no definitive solution. However, with the knowledge of the anatomy, geometry, and the possible neural sources and sinks, it is possible to postulate a likely solution. Direct recording of field potentials from the brain regions that generate the potentials is important for refining the solution. CSD analysis can be readily applied to reveal the sites and time courses of the dendritic or somatic synaptic activation or action currents in a population of synchronously activated neurons. CSD analysis has been the subject of previous reviews (11, 12). In recent

Field Potential Generation and Current Source Density Analysis

3

years, advances in computer and electrode fabrication technology have made potential mapping and CSD analysis more precise, and with an increasing capability to address more research questions than ever before. Multichannel field potential recording and CSD analysis will be the focus of the present review.

2. Field Potential Theory 2.1. Current Flow in a Cable

A common example used to illustrate the relationship among intra and extracellular potentials and membrane currents is an axon. An axon is modeled by a cable, which is an engineering terminology for a long metallic wire that carries voltage across long distances. Below the threshold of an action potential, an axon can be considered as an electric cable that conducts passively or electrotonically. Dendrites and dendritic trees can also be represented as cables. The passive flow of current down an axon (cable) is illustrated in Fig. 1a. Assume currents are injected at x = 0, the currents flow down the long axis of the axon in both positive and negative directions. We also assume that extracellular currents can only flow along a thin layer of extracellular fluid, such as along the long axis of a moistened axon suspended in air. For each unit length, the extracellular resistance is small in comparison to intracellular resistance, and the currents are mainly determined by the intracellular resistance/length assumed to be ri [ohm/cm]. For the locations 0, 1, and 2 inside the axon, the intracellular (longitudinal) current between two points is determined by Ohm’s Law (voltage = current × resistance), such that I 0 = −(V1 − V0 )/(ri ∆x ),

(1)

where ri Dx is the resistance across a small segment of axon of length Dx, and the first negative sign on the right side indicates a positive direction of current flow when V0 is greater than V1, as shown in Fig. 1a. Similarly,

I 1 = −(V 2 − V1 )/(ri ∆x ).

(2)

To conserve charges (per time), the membrane current is J1 = I 0 − I 1 ,

= −(V1 − V0 )/(ri ∆x ) + (V2 − V1 )/(ri ∆x ),

(3)

= (V0 + V2 − 2V1 )/(ri ∆x ).

Equation (3) determines the membrane current per length from the spatial distribution of the intracellular potential.

4

Leung

Fig. 1. Current flow in an axon suspended in air or in a volume conductor. (a) Injection of current into the axon suspended in air sets up local current loops. Injected current at the origin (x = 0) moves to the right and left, and a decrease in longitudinal intracellular currents (I ) gives a membrane current (J ). At steady-state following constant current injection, the voltage decays exponentially with distance (x) in an infinite cable model of the axon. (b) Schematic extracellular potential field during an action potential in an axon suspended in a conductive medium (after (13)). The maximal current sink (inward Na+ current) is at the origin, and the current sources (current outflow) surround the sink, giving rise to a spatial sequence of positive–negative–positive fields. The extracellular (field) currents are perpendicular to the potential contours, and only one major current loop is drawn for simplicity.

If the differential form of the above (Equations 1–3) is used, i.e., when Dx → 0, then

I = −(1/ri )

J =−

∂V ∂x

∂ ∂ 2V I = (1/ri ) 2 . ∂x ∂x

(4)

Field Potential Generation and Current Source Density Analysis

5

Adding the relation that the membrane current J = cm

∂V V + , ∂t rm

where rm is the membrane resistance per length and cm is the membrane capacitance per length, then

cm

∂V V 1 ∂ 2V + = . ∂t rm ri ∂x 2

∂ 2V ∂V . + (ri /rm )V = c m ∂x 2 ∂t

(5)

When the steady-state condition is considered, that is, at a long time after continuous injection of current at x = 0 (when membrane capacitance currents can be ignored), then

∂ 2V + (ri /rm )V = 0 ∂x 2 and for an infinite cable with current injection at x = 0, the solution for x > 0 is

V = V0 e − x /λ ,

(6)

where V0 = intracellular voltage at x = 0 and l = (rm/ri)0.5. Equation (6) gives the familiar steady-state solution to an infinite cable equation, which is an exponential decay with space constant l. 2.2. Theory of Extracellular (Field) Potentials

Field potential generation is instantaneous or quasi-stationary, that is, it depends on the extracellular current ( J ) distribution, or the current field, at one instant of time. In other words, it is a snapshot of the changing world, independent of the future or the past. The reason for an instantaneous generation of potential field is because extracellular currents are mainly resistive, and capacitance and inductive currents that would impose time shifts do not make a significant contribution. In addition, currents flow in a closed loop, and there is no gain or loss of currents around the loop (when capacitance currents are included). For solving the intracellular potentials, the extracellular potential (or resistance) is ignored as a first approximation (Sect. 2.1). Similarly, extracellular potentials are assumed to be generated from a set of membrane currents impressing on the extracellular medium, without explicit reference to the intracellular potentials (9). Since the general case is a three-dimensional medium, we define the CSD, or Iv, as the sum of the net membrane currents in a local volume. Thus, current source is the net outward membrane current within a volume, while current sink is the net inward membrane current within a volume.

6

Leung

In the generation of field potentials, we may assume that the extracellular medium contains sources and sinks (CSDs) that give rise to the extracellular current vector ( J ), electric field vector ( E ), and scalar potential field (f). The fields are described by three standard electric field equations (12, 14) as follows. E = −∇�

(7)

J = σE = −σ∇� f

(8)

∇·J = −I v

(9)

and

where s = conductivity tensor  ∂ ˆ ∂ ˆ ∂ ˆ  ∇ = ∂x i + ∂y j + ∂z k  gradient operator

and

 ∂ ∂ ∂ ˆ k divergence operator ∇ =  ˆi + ˆj + ∂y ∂z   ∂x Combining the equations above gives the Poisson’s equation ∇·(s∇ � f) = − I v

(10)

and assumption of s that is homogeneous and isotropic gives

∂ 2f � ∂ 2f � ∂ 2f � I v + + =− , ∂x 2 ∂y 2 ∂z 2 s

(11)

which gives a solution of

f (r ) =

Io 4psr

(12)

when a single current source of Io is placed at the origin at distance r from where the potential is measured. The generation of the potential field by CSDs is mathematically identical to the generation of the potential field by electrostatic charges, since a common equation, the Poisson’s equation, describes both generation processes. 2.3. Qualitative Aspects of Potential Fields Generated by Action and Synaptic Currents 2.3.1. Action Currents in Single Cells

The analysis of an injected current in a cable (Fig. 1a) is similar to that of an action potential with an inward current at the origin (x = 0). If the axon is inside a volume conductor (e.g. in a large bath of saline or in the brain), the potential will spread through the volume and result in “volume conduction” (13). The site of the current flow from the extracellular into the intracellular medium corresponds to a current sink, i.e., where the current disappears from the extracellular medium. A current sink generates

Field Potential Generation and Current Source Density Analysis

7

a local negative potential. An outward membrane current is a current source for the extracellular medium and it generates a local positive potential. As indicated by Fig. 1b, the current sources are located at sites (left and right) away from the origin (sink), where the current flows out of the membrane; only the main current loops are drawn in Fig. 1b. The spatial sequence of CSD is source–sink–source, corresponding to positive–negative–positive potential field. Extracellular potential field lines are of decreasing magnitude away from the axon. Zero isopotential lines separate regions of positive and negative potentials. Since the one sink is relatively more intense than each of the two sources, the extracellular medium is occupied more by negative potentials. 2.3.2. Synaptic Currents in Single Cells

Synaptic excitation is assumed to occur at the distal apical dendrite of a pyramidal cell in the cerebral cortex (Fig. 2a). The distal apical dendrites are depolarized by an excitatory postsynaptic potential (EPSP) that has an inward current (sink) at the synaptic site. Since the distal apical dendrite is more depolarized (or more positive in potential) than the rest of the cell, longitudinal current flows away from the apical dendrite toward the cell body. For long apical dendrites, most currents would flow out of the more proximal location dendrites and then extracellularly to complete current loops. The outflow at the more proximal apical dendrites is a current source, and it is passive (without an electromotive force). Distal excitation generates a dipole field, with negative potential at the distal apical dendrite, positive potential at the proximal apical dendrite, and a zero isopotential in between (Fig. 2a).

Fig. 2. Qualitative potential fields in a pyramidal cell (population assumed to line up in parallel) following excitation at different levels and somatic inhibition. The pyramidal cell schematic includes two basal dendrites (top) and a single apical dendrite (bottom). (a) Excitation of the distal apical dendrites gives a distal dipole (positive–negative) field separated by a zero isopotential (horizontal dotted line). (b) Proximal apical excitation generates a proximal dipole field. (c) Basal dendritic excitation makes the basal dendritic layers negative and the soma layer positive. (d) Somatic inhibition sets up a field that is positive at the soma and possibly negative at the apical dendrites; inhibitory currents and potentials have a slow time course.

8

Leung

If excitatory synapses are activated at the proximal dendrites (Fig. 2b), the inward current at the dendrite causes main current to flow proximally toward the soma (thick arrow) and a smaller current flowing distally (thin arrow). The latter assumes that the somatic loop has a lower resistance than a distal current loop. The zero isopotential is expected to be somewhere along the path of the major current loop, i.e., between the proximal dendritic synaptic site and the soma. When the basal dendrites are synaptically excited (Fig. 2c), the current flows to the soma and exit there to complete an extracellular loop. The sink is at the basal dendrites and the main source will be at the soma. The dipole field (negative at the basal dendrites, positive at the soma) will have a zero isopotential between the activated synapses and the soma. If an inhibitory postsynaptic potential (IPSP) occurs at the soma, it may be Cl− entering through GABA-A receptors, or K+ exiting through GABA-B receptors. In either case, the inhibitory postsynaptic current is an outward current and constitutes a current source at the soma. The current source drives currents extracellularly to passive sinks on either side of the soma. The reversal of the extracellular potential field depends on the magnitude of the basal and apical dendritic sources. In Fig. 2d, it is assumed that the apical dendritic current loop is the major current loop, and the isopotential line is at the proximal apical dendritic layer (15). A field generated by inhibitory synaptic currents is generally smaller than the one generated by excitatory synaptic currents for several reasons. The main reason is that inhibitory synaptic currents are normally smaller than the excitatory synaptic currents because of the following: (1) at rest (~ −60 mV), the electromotive force of an inhibitory current (reversal potential ~ −72 mV) is smaller than that of an excitatory current (reversal potential ~0 mV); (2) capacitative currents are larger for excitation than inhibition, since excitatory currents rises 5–10 times faster than inhibitory currents. The above does not suggest that synaptic inhibition is less effective than excitation, since inhibition can occur with shunting (conductance change without voltage change). In general, excitatory and inhibitory fields could not be distinguished without additional neurophysiological data. For example, the somatic inhibitory field (Fig. 2d) may not be readily distinguished from a proximal dendritic excitatory field (Fig. 2b). Thus, intracellular recording or unit recording is necessary to confirm whether the dominant event is excitation or inhibition. 2.3.3. Geometrical and Temporal Considerations: Open and Closed Field, Standing and Traveling Wave

Individual pyramidal cells are lined up in palisades (parallel) in the cerebral cortex (Fig. 3a). When the cells are synchronously activated during an evoked potential or a spontaneous EEG, the potential field of individual cells will sum to give a large signal. The large signal results from summation of extracellular currents, or potentials (Equation 12), from individual neurons.

Field Potential Generation and Current Source Density Analysis

9

Fig. 3. Open and closed fields, and volume for CSD measurement. (a) Schematic open field generated by distal dendritic excitation of palisades of pyramidal cells in a cerebral cortex. The potential field has positive and negative poles, and potential field lines are crossed by current flows with arrows. (b1) Schematic closed potential field with a single polarity at the center, generated by (b2) a single cell with distal excitation indicated by arrows at the periphery of the cell, or by (b3) a radially symmetric cluster of cells each excited at the distal dendrite (indicated by arrows). (c) Volume element of dimension Dx, Dy, and Dz used for CSD estimate. Current (J ) outflow subtracting the inflow in each dimension gives the current source (or sink) for that dimension. An increase in current in the z direction is depicted, indicating a net source for the volume shown.

Figures 2 and 3a illustrate open fields. An open field is the one that is not contained in space, and its effect can be detected at large distances away from the source. Because of the volume conductor, the potential can be observed far from the current sources and sinks that generate the potential (Fig. 3a1). This is why the EEG can be recorded on the scalp, where there are no neuronal sources and sinks. Theoretically, a single source gives potentials that decrease ultimately with 1/r (r = distance away from the source; 12). Given that currents flow in a closed loop in the brain, a source is always accompanied by a sink and a source-sink pair (dipole) is the minimal unit of extracellular potential generation. Dipole potentials decrease ultimately with 1/r 2. Concatenation of two dipoles, a “+ − − +” sequence (or effectively a “+ − +” sequence)

10

Leung

generates a field that decays ultimately with 1/r 3. The latter configuration resembles the source–sink–source structure of an action potential (Fig. 1b), and explains, in part, the faster decay of an action potential with distance, when compared with a dipole field (e.g. distal excitatory synaptic potential). Although the potential field (and the accompanying current flow field) ultimately decreases with distance, it is still present at large distances. Indeed, potentials recorded on the scalp are caused by volume conduction. A closed field is a potential field that can only be detected within the structure itself. A single cell may have a symmetrical spherical geometry such that uniform activation of its dendrites (or cell bodies) will only generate radial currents (Fig. 3b2). Distal sinks (excitation) are balanced by proximal passive sources. Radially symmetric source–sink gives a potential field only within the spatial domain of the sinks and sources (Fig. 3b1). Other than a single cell, activation of a structure of many cells may result in radially symmetric sources and sinks that generate a closed field (Fig. 3b3). If the spatial potential field is recorded at different times, the field associated with a propagating action potential (or compound action potential from a population of axons) will move with time. The shifting of the field with time gives a traveling (propagating) potential field. For synaptic excitation, there is also some delay in charging of the membrane capacitance. However, for most general purposes, the potential field associated with synaptic currents can be considered as standing since the shift in field polarity is small in space and relatively rapid in time (within 10 ms).

3. General Application of CSD Analysis 3.1. Single vs. Multichannel Electrode Mapping 3.1.1. Pros and Cons of Single vs. Array Electrode Mapping

Typically, the potential field is sampled (mapped) at discrete locations, in a Cartesian (rectangular) coordinate system, before the application of CSD analysis. There are two main methods for mapping – by a single electrode sequentially or by an array of electrodes simultaneously. Practically, there are advantages and disadvantages of each method (Table 1). Since the silicon probe is the most frequently used multichannel probe that is commercially available, acquisition is focused on using the silicon probe. There are many advantages of using an electrode array for simultaneous registration of potentials in the brain (Table 1). Linear arrays of 8–64 electrodes, of various inter-electrode distances, are now commercially available (see Appendix). Mapping an instantaneous potential field by simultaneous recordings in an array is theoretically rigorous. Some events in the brain occur only once,

Field Potential Generation and Current Source Density Analysis

11

Table 1 Advantages of sequential single-electrode mapping vs. simultaneous multichannel mapping by a silicon probe Single electrode recording

Multichannel probe recording

Area of coverage

Flexible

Limited

Depth sampling interval

Flexible

Fixed but accurate

Technical requirements

Few

Many

Disruption of the brain

Small

Moderate

Duration of mapping

Slow

Fast

Nonstationary events

No

Yes

Overall signal to noise

Adequate

Superior

for example, seizures, epileptiform, and other transients, and simultaneous recordings are needed to capture these events. A rapid acquisition of the potential profile (and CSDs) will allow studies in which the responses change with time, such as experiments on synaptic plasticity and pharmacological responses. A practical disadvantage in using multichannel probes is the necessity to maintain viability and calibration of multiple recording channels. Special hardware (e.g. multiple analog-to-digital converters and sample-and-hold circuits) and software are needed to achieve simultaneous recordings for multiple channels (Sect. 3.1.2). The area of mapping is limited by the availability of multichannel probes. As described in Sect. 4.1, the spatial sampling interval depends on the spatial frequency of the potential field. The most useful probes are those with electrodes lined up in a single dimension in depth. It is impractical to sample potentials by many depth probes for several reasons. First, if each probe maps the z direction, multiple probes have to protrude from a surface grid in the x–y plane, and these types of multishank probes are costly. Second, neighboring probes cannot be too close to each other, otherwise serious distortion and lesion of the brain tissue would occur. Mapping by a single electrode is flexible in terms of mapping intervals, in x, y, and z-dimensions. Considering a glass microelectrode, serious disruption of the brain occurs at the widest part of the micropipette (typically 1–1.5 mm outer diameter) and less so at the tapering end of ~1-mm diameter. A single microelectrode track inflicts little detectable damage to the tissue, and multiple tracks of different spatial intervals are possible. Averaging of the event enhances the signal-to-noise ratio. Equipment requirements for single-electrode mapping are relatively simple, and only

12

Leung

an amplifier and an analog-to-digital converter are needed. However, an accurate micromanipulator is important (Sect. 3.2.2). A major assumption of single-electrode mapping is that an event, usually an evoked potential, is stationary, such that the event recorded at one location and time is the same as that recorded at another location and time. Whether the event is stationary or not should be confirmed. One way is to record the same event from a nonmoving (fixed) electrode to show that the event is stable in time. Another way is to return the mapping electrode to the same location, and repeat the recording at different times. In general, single-electrode mapping is counter-indicated for nonstationary events, such as seizure and one-time transients (e.g. interictal spikes). Single-electrode mapping can be automated for speed and accuracy. In one particular implementation (16, 17), a programmable microdrive pulled up a microelectrode (in a z-axis), paused for a time delay (for brain tissue to settle), and then acquired an average-evoked potential (AEP) before the cycle repeated itself. With additional x and y-axis controls of a micromanipulator, the electrode can be placed at any Cartesian coordinates (x, y, and z), and mapping over a volume can be achieved. Single track or one-dimensional mapping may make use of recordings acquired by surface arrays of electrodes, such as those used to map the surface distribution of potentials on a cortical surface (4, 15). A surface array may have electrodes separated by 200–300 mm, and its recordings may be used to justify singledimensional mapping or to provide an estimate of the lateral current spread. Surface arrays can be applied to determine potential and CSD profiles on the surface of a brain slice in vitro. It is possible to apply a mixed strategy of moving the silicon probe in order to map a larger extent of the brain. A small movement (25–100 mm) of a silicon probe was found to not significantly disrupt the brain tissue, such that in the regions that overlapped before and after movements, the CSDs remained essentially the same (18). 3.1.2. Specifics of Multichannel Recording

A silicon probe consists of an array of electrodes made by thinfilm technology on a silicon micro-machined substrate and insulated by dielectric (silicon dioxide/nitride) deposits. The commercially available probes were originally developed at the University of Michigan (19). Commercial availability depends on reliable batch processing. A typical recording area (electrode site) is made of iridium, rectangular or square in shape, of 5–20 mm dimension. Adjacent electrodes can be separated by 30–200 mm. Fabrication dimensional accuracy is <1 mm (19). The insulated shank is 15-mm thick, 30- to 200-mm wide, and with a length varying from 3 to >10 mm. Multiple-shank probes are also available.

Field Potential Generation and Current Source Density Analysis

13

Impedance of each electrode in the silicon probe is about 2 MW at 1 kHz. Electrode impedance should be matched by preamplifiers of high input resistance impedance (>0.2 GW). The preamplification, main amplification, filtering, and analog-to-digital conversion should be identical for all the electrodes in the array. Identical signal processing for each channel is important for CSD analysis, since a small distortion at a single channel will be magnified. Gain calibration of the complete multichannel system can be made by placing a silicon probe in a large saline bath and passing square stimulating current pulses across a pair of long silver wires placed at a large distance (>2 cm) away from the recording electrode. The goal is to achieve recordings of the same waveform and identical magnitude at all electrodes. Recording from chronically implanted animals provides an important application of silicon probes. In one application, a silicon probe is connected by means of a flexible ribbon cable to a small connector embedded in dental cement to the skull (20). Bragin et al. (21, 22) used a microdrive to drive a single probe slowly through the brain. Single neuron activity can be recorded from a silicon probe. The single unit signals from electrodes well within the shank (~200-mm wide) were reported to be lower than those from electrodes near the tip (22). This occurs likely because of tissue disruption that was more severe at the shank than the tip of the probe. Even when an electrode is placed in a cell-free conductive medium (such as a saline bath), the electrode would inevitably distort the potential and the current flow field. The distortion is typically larger for the larger silicon probe than a single microelectrode. Drake et al. (19) documented the distortion of the field of an action potential by a silicon probe. They reported up to 60% increase of the potential (compared with a theoretical condition with no recording probe) on the recording face of the probe, and up to 100% attenuation of the potential on the nonrecording side of the probe. The field distortion is considered acceptable when the dimension of the probe is smaller than the inter-electrode interval. We confirmed a possible distortion of the field potentials recorded by a silicon probe (16 electrodes at 50 mm intervals) in the hippocampus. The latter probe has a nonconducting flat shank of 15-mm thickness and it tapers from the main shank of 200 mm thickness to 33 mm at the tip. We (Wu and Leung, unpublished data) found that the maximal orthodromic population spike amplitude recorded by a silicon probe was smaller than that recorded by a single glass micropipette. The probe-recorded potential may be small because extracellular currents flowing from one side (the “blind” side) of the silicon probe cannot reach the recording electrode. Accentuation of the field on the recording side, an effect of a nonconductive boundary condition, apparently did not fully compensate for the loss of signal from the “blind” side.

14

Leung

Several techniques are used to confirm the location of a articular electrode on a silicon probe. Recording of field potenp tials evoked by stimulation of known afferents will provide information on line. Some of these responses in CA1 area of the hippocampus are described later (Sect. 4). Bragin et al. (21, 22) sectioned the brain with the implanted probe in place. Using a special electronic circuit that allows for both recording and passing current, lesion can be made at a specific electrode of the probe and then verified histologically (18). 3.2. Experimental CSD Analysis 3.2.1. Second-Order Differencing

On the basis of (Equation 11), CSDs (Iv) in a homogeneous medium are derived from a second-order derivative of the potential field (Ø) in three dimensions. Since it is necessary to map the potential field at discrete locations, the second-order differentiation is replaced by second-order differencing. In other words,

∂ 2f ≈ [f (x +∆x , y , z, t ) + f (x − ∆x , y , z, t ) − 2f (x , y , z, t )]/(∆x )2 ∂x 2

∂ 2f ≈ [f (x , y + ∆y , z, t ) + f (x , y − ∆y , z, t ) − 2f (x , y , z, t )]/(∆y )2 ∂y 2

∂ 2f ≈ [f (x , y , z + ∆z, t ) + f (x , y , z − ∆z, t ) − 2f (x , y , z, t )]/(∆z)2 ∂z 2 and

I v ≈ s {[2f (x , y , z , t ) − f (x + ∆x , y , z , t ) − f (x − ∆x , y , z , t )]/(∆x )2 +[2f (x , y , z , t ) − f (x , y + ∆y , z , t ) − f (x , y − ∆y , z , t )]/(∆y )2 +[2f (x , y , z , t ) − f (x , y , z + ∆z , t ) − f (x , y , z − ∆z , t )]/(∆z)2} (13)

A general consideration before attempting CSD analysis for a new structure is to decide on the extent of mapping, the mapping interval, and the electrodes to be used for mapping. Although many CSD studies used one-dimensional mapping, the underlying assumption should be experimentally tested. For an unknown structure and input, the extent of the potential field should be mapped by multiple tracks or by means of cortical surface arrays. Even if three-dimensional CSD analysis may not be practical, the three-dimensional potential field gives an estimate of the error of using one-dimensional CSDs. In general, whether the conductivity is homogeneous or isotropic will affect the results. However, accurate measurement of conductivity for a particular preparation is difficult, and theoretical and experimental studies have determined that small (even several folds) differences in isotropicity or layer conductivity did not significantly affect the locations of the CSDs. A nonisotropic medium was found in the cerebellum (23) and a lower conductivity (inhomogeneity) was reported in the pyramidal cell layer than other dendritic layers in CA1 (24).

Field Potential Generation and Current Source Density Analysis

15

With the assumption that there is negligible change in currents in the x and y directions, I v ≈ s [2f (z, t ) − f (z + ∆z, t ) − f (z − ∆z, t )]/(∆z)2

≈ s /(∆z)2 {[(f (x , y , z, t ) − f (x , y , z + ∆z , t )] −[ f (x , y , z − ∆z, t ) − ( f (x , y , z, t )]} I v ≈ {Jz + ∆z / 2 − Jz − ∆z / 2} / (∆z)

(14) (15)

The equation above indicates that the CSD Iv is the difference of the Jz leaving (at z + Dz/2) and entering (at z − Dz/2) a particular volume, as illustrated in Fig. 3c, with Jz ≈ −s (∆f / ∆z) . When outflow is larger than inflow, Iv is positive or there is a source. When outflow is smaller than inflow, Iv is negative or there is a sink. 3.2.2. Accuracy of the CSD Estimates

The spatial interval used for deriving CSD is a parameter that needs to be experimentally determined. Sampling of a signal requires a rate that is at least double that of the frequency of the signal, as required by the sampling theorem. A higher sampling rate will give better accuracy. Thus, the spatial sampling interval should be at least double that of the main spatial frequency of the potential function in each dimension. In field mapping of the cerebellum and hippocampus, 20–50 mm was determined to be an adequate spatial sampling interval (18, 25, 26). Since the CSD estimate is based on differencing, it will amplify or increase the noise of a single channel. This suggests that if different amplifiers and analog-to-digital converters are involved in sampling of potentials at different depths, the calibration of different channels must be accurate. In addition, (Equation 14) indicates that a major error may come from the inaccuracy of the interval Dz (on account of the square of Dz). An accurate micromanipulator, calibrated to 1 mm or less, is important for accuracy. Spatial smoothing of the CSDs has been extensively discussed by Freeman and Nicholson (23). A simple smoothing formula uses a spatial interval of N = 2 instead of N = 1 in the following equation (an extension of Equation 14): I v ≈ s [2f (z, t ) − f (z + N ∆z, t ) − f (z − N ∆z, t )]/(N ∆z)2 . (16) The two-step differencing formula is equivalent to doing a running spatial average of three consecutive depth potentials f ′i = (¼fi−1 + ½fi + ¼fi+1); [fi−1, fi and fi+1 are the raw data at Dz intervals] and then applying (Equation 14) (N = 1 differencing) to f′i (23). Polynomial fit to the potential profile was also used. Smoothing reduces the noise (that is unrelated to the signal) and improves the reliability of the CSDs. The down side of smoothing is that accuracy is degraded.

16

Leung

4. CSD Analysis of Hippocampal Field Potentials

4.1. Synaptic Excitation and Inhibition in CA1

The following gives examples of CSD analysis applied to hippocampal CA1 area. Similar analysis applied to the dentate gyrus has been reported by Golarai and Sutula (26), Canning and Leung (27), and Canning et al. (28). CSD analysis had also been done in the visual cortex (11) and olfactory cortex (29). The layered structure of the afferent excitation of hippocampal CA1 area is readily revealed by potential mapping and CSD analysis. The following results were obtained by mapping of the potentials evoked in CA1 in the rat under urethane (1.5 mg/kg i.p.) anesthesia. Stimulation of the medial perforant path that synapses on the distal apical dendrites of CA1 gave a maximally negative potential at the distal apical dendrites (Fig. 4a1, (16)). However, because of volume conduction from the dentate gyrus (DG), there was no positive potential as predicted in Fig. 2a. CSD analysis (Fig. 4a2) gives a pattern of sinks and sources that is consistent with the current flows in Fig. 2a. The sink was located at the distal apical dendrites, accompanied by distributed proximal sources that gradually diminished in amplitude toward the soma. Smoothing by the two-step differencing (N = 2 in (Equation 16)) smoothed out the dendritic source distribution and decreased the magnitude of the maximal sink. Volume conduction from the DG to CA1 is illustrated by the AEPs following medial perforant path stimulation (Fig. 5a). The initial field potential (<10 ms latency) was dominated by the EPSPs generated in the DG, giving negative potentials that extended throughout CA1. Despite the presence of a distal dendritic sink in CA1, no positive potential was found in CA1 (16). In addition, the EPSPs fired a DG population spike that was volume-conducted to CA1 (asterisk in Fig. 5). Both EPSP and population-spike fields generated by the DG were effectively removed by calculating CSDs in CA1 (Fig. 5b). Part of the EPSP source in the DG is shown in Fig. 5b (Upward slant arrow). Thus, CSD analysis is effective in removing volume conduction. Excitation of the proximal apical dendrites, through the Schaffer collaterals (association fibers) from ipsilateral CA3b, resulted in soma positive and apical-dendritic negative potentials, with a zero isopotential near the soma (Fig. 4b1, (31)). The sink occupies the proximal apical dendrites, accompanied by a source at the soma and a smaller source at the distal apical dendrites (Fig. 4b2). An example of proximal apical dendritic excitation is shown by the temporal traces in Fig. 6a, where the vertical dotted line (labeled fEPSP) indicates the EPSP field before the onset of population spiking.

Field Potential Generation and Current Source Density Analysis

17

Fig. 4. One-dimensional average evoked potential (AEP) (column 1) and current source density (CSD) (column 2) in CA1 area of the dorsal hippocampus of rats, acquired by simultaneous recording from a 16-channel silicon probe with 50-mm intervals. Top inset, histological section of the hippocampus showing the location of the recording probe (Rec), and schematically the stimulation of the basal or apical dendrites, alveus, or medial perforant path (MPP). CSDs are shown without (one-step, solid lines) and with smoothing (two-step, dotted lines). (a–d) corresponds to the corresponding part in Fig. 2 that predicts the potential field qualitatively. (a) Distal apical dendritic excitation at 5.9 ms latency after stimulation of the ipsilateral MPP at 300 mA (N = 4 sweeps averaged). (b) Proximal apical excitation at 5.2 ms after stimulation of ipsilateral CA3b stratum radiatum at 140 mA, below the population spike threshold (N = 16). (c) Basal dendritic excitation at 11.7 ms latency after stimulation of stratum oriens of contralateral CA3b at 100 mA (N = 4); Asterisks indicates a small apical dendritic excitation, more apparent in the CSD profile. (d) Inhibitory field recorded at 15 ms after low-intensity (25 mA) alveus stimulation (N = 64). Inset in D, intracellular recording of a CA1 pyramidal cell in vivo (30) shows an antidromic spike followed by an inhibitory postsynaptic potential; ext = extracellular field response. (b–d) derived from one rat, and A from another rat. Zero depth (0 mm) indicates the pyramidal cell layer.

18

Leung

Fig. 5. Current source density (CSD) analysis of the CA1 field evoked by medial perforant path (MPP) stimulation removes volume conduction from the dentate gyrus (DG). (a) Average-evoked potential (AEP) and B, CSD of the MPP-evoked potential profile in CA1. Note that the negative potentials generated by the field excitatory postsynaptic potential (fEPSP) in the DG and the population spike in the DG (asterisk) were removed in the CSD profile (b). Upward slant arrow indicates distal apical sink in CA1; Upward slant arrow source at the distal dendrites of granule cells in the DG, Hash # polysynaptic apical excitation; Hat ^ basal dendritic excitation after MPP stimulation. Filled circle indicates stimulus artifact.

Excitation of the basal dendrites resulted from stimulation of the basal dendritic layer in the contralateral CA3a (32). As expected, the basal dendrites show a maximal sink and negative potential, while the soma and the rest of the dendrites serve as passive sources (Fig. 4c). There is a small proximal apical dendritic sink (asterisk in Fig. 4c) that accompanies the main basal dendritic sink, because the apical dendrites also receive commissural CA3 projection. At 15 ms following an alveus stimulation that evoked an antidromic spike, the dominant intracellular event is an IPSP (Fig. 1.4d inset). The inhibitory potential field was maximally positive near the cell body layer, and reversed to a negative potential at the distal apical dendrites (Fig. 4d1, (15)). The CSD profile indicates a source at the cell body layer and proximal dendrites,

Field Potential Generation and Current Source Density Analysis

19

Fig. 6. One-dimensional potential field and current source density following activation of an orthodromic (a) or antidromic (b) population spike in CA1. Top traces, CSD vs. time profiles in depth (mm) with respect to pyramidal cell layer (0 mm). Bottom contour map (original in color) with schematic pyramidal cell showing arrow with propagation direction, and zero isopotential (dotted line) separating source (+) and sink (−) regions. (a) Mid-apical dendritic excitation by the Schaffer collaterals initiated an orthodromic population spike at the proximal dendrites that propagated to the soma (solid arrow) and then the basal dendrites (dotted arrow). Light vertical line indicates rising phase of the field excitatory postsynaptic potential (fEPSP), with the fEPSP-population spike inflection point indicated by Hat at 150-mm depth. (b) Population antidromic spike evoked by stimulation of the alveus initiated an initial segment spike (asterisk) followed by somatic spike that propagated to the proximal apical dendrites (solid arrow) and also to the basal dendrites (dotted arrow). Filled circle indicates stimulus artifact.

with the main sinks at the distal apical dendrites and a small sink at the basal dendrites (Fig. 4d2). The two-step CSD formula appears quite effective in smoothing out the CSD depth profile. The inhibitory field/CSD was smaller than the excitatory field/CSD (note axis calibrations in Fig. 4). The inhibitory synaptic currents are normally smaller than the excitatory synaptic currents as explained in Sect. 2.3.2 above.

20

Leung

A somatic inhibitory field, by itself, may not be readily distinguished from a dendritic excitatory field. Thus, accompanied data of intracellular recording (Fig. 4d inset) or unit recording is necessary to confirm whether the dominant event is excitation or inhibition. The above examples illustrate that the relation between experimental potential and CSD profiles (Fig. 4) is more complex than that predicted using basic assumptions (Fig. 2). In general, because of volume conduction and overlapping fields, it is not possible to state that a negative potential corresponds to a sink and a positive potential to a source. 4.2. Population Spike: Orthodromic and Antidromic

An orthodromic action potential results when excitation exceeds the spike threshold. A population spike is a compound action potential caused by the synchronous firing of action potentials in CA1 (33). As shown by CSD analysis (Fig. 6a), following Schaffer collateral excitation of the apical dendrites, an orthodromic population spike (sink) started at the proximal dendrites, about 150 mm from the cell body layer. The slanted arrow in the CSD time traces (top) and in the CSD contour map (bottom) illustrates the propagation of the orthodromic population spike from the apical dendrites toward the cell body, and then from the cell body to the basal dendrites (dotted arrow). A population spike inevitably started in the proximal apical dendrites (31, 34) in vivo, while threshold excitation in vitro activates an action potential originating from the axon hillock (35). The difference may be explained in part by the necessity to use higher than threshold excitation to induce an orthodromic population spike in vivo. The initiation of a spike from the axon hillock is reproduced by stimulation of the axons of a CA1 pyramidal cell in the alveus. The latter results in a population antidromic spike that backpropagates to invade the initial segment (asterisk in Fig. 6b), soma, and the proximal apical dendrites in succession. The population spike sink shows a progressive increase in peak latency with depth, as indicated by the solid arrow in the CSD traces and contour map in Fig. 6b, (31). A population antidromic spike sink could be detected in the apical dendrites, up to 150 mm away from the soma. After invasion of the soma, a spike sink propagating from the soma to the basal dendrites is also apparent (dotted arrow in Fig. 6b).

4.3. Spontaneous Events in the EEG

Various spontaneous events in the hippocampal EEG, including the theta rhythm, gamma rhythm, 200-Hz ripples, and transient sharp waves, have been mapped using silicon probes (21, 36, 37). A slow oscillation (<1 Hz) has recently been mapped and analyzed by Wolansky et al. (38).

Field Potential Generation and Current Source Density Analysis

21

5. Modeling of Field Potentials CSD analysis derives sources and sinks from the potential field. It is also important to do the inverse process of synthesizing (modeling) the field potentials from presumed sources and sinks. The latter process complements CSD analysis. Often, CSD analysis makes simplifying assumptions, such as single dimensional current flow in an isotropic medium, and some of these assumptions can be examined by the modeling approach. A spatiotemporal pattern of CSDs may have different underlying mechanisms, and explicit modeling is necessary to distinguish among these mechanisms. For example, alternate hypotheses were made for a spatiotemporal pattern of CSDs in the cerebellum, whether it was caused by electrotonically spreading synaptic currents or propagating dendritic action currents in Purkinje cells (39). Both experimental and modeling studies support dendritic action currents. A three-step process of generating field potentials has been proposed (12): 1. A compartment model of individual neurons is made (40, 41). Each part of a neuron is represented by one compartment, and the inputs may be synaptic excitation and inhibition, and voltage-dependent currents (described later). Effectively, each compartment represents a discrete, lumped space. Compartments are then connected together as a real neuron. The set of equations of the intracellular voltages in different compartments are solved by a computer, assuming negligible extracellular potentials. 2. The membrane current per compartment is calculated at each time. The net outward (or inward) membrane current per volume is the CSD, and it depends on the packing of the neurons in space. 3. The field potentials are generated from the CSD distribution over volume by integration or superposition (Equation 12). Typically, simplifying assumptions of the extracellular medium (e.g. conductivity, isotropicity) and its boundary conditions are made. In the ith compartment, the “spatial” current is described by an equation similar to (Equation 3),

J i = −(Vi − Vi −1 )/(ri ∆x ) + (Vi +1 − Vi )/(ri ∆x ).

(17)

In Sect. 2.1, current is injected only at one point, such that the rest of the cable (compartments) will only have a passive membrane current. In general, the membrane current of a compartment consists of passive current (of the form c m (∂V / ∂t ) + (V / rm ) ),

22

Leung

active currents from EPSP, IPSP, action potential, and other voltage-dependent currents, that is, Ji = −(Vi − Vi−1)/(riDx) + (Vi+1 − Vi)/(riDx) = sum of passive, synaptic, and voltage-dependent currents for the ith compartment. In a recent model, the CA1 pyramidal cell is represented by 38 compartments presenting axon, cell body, basal and apical dendrites (Fig. 7a, (42)). In the model, an antidromic action potential is activated by an axonal action potential, and it propagates to active sites in the soma and dendrites of a population of CA1 pyramidal cells. The CA1 pyramidal cells are assumed to form an incomplete shell subtending 0.25 radians at the center of the shell 4 mm away from the surface of CA1 (43), i.e., about a 2 × 2 mm surface activation of CA1 is assumed. The potentials are generated at the z-axis, and CSDs are calculated in one-dimension (z-direction) as for the experimental data. The model results

Fig. 7. Modeling the potential field and current source density (CSD) by a compartment model of pyramidal cells and volume conduction. (a) A 38-compartment model of a CA1 pyramidal cell. Compartments with a thick circle are capable of generating action currents; compartment 23 (apical dendrite) is approximately at depth 200 mm, and compartment 14 (basal dendrite) at –100 mm. (b) Model-generated CSDs of a population antidromic spike, initiated by stimulation (stim) of axon compartment 38. Action potential propagates to the soma, and the movement of the current sink is shown from the soma to the proximal apical dendrites (solid arrow) and basal dendrites (dotted arrow ). Adapted from Leung and Peloquin (42). Compare with experimental data of antidromic population spike shown in Fig. 6b.

Field Potential Generation and Current Source Density Analysis

23

(Fig. 7b) show an action current sink traveling from the soma to the proximal apical dendrites (solid arrow), and a smaller one traveling from the soma to the basal dendrites (dotted arrow). Initially positive potentials at the basal dendrites (−150 mm) and distal apical dendrites (>250 mm) are similar between experimental and model data. The present model assumes that the axon segments generate negligible extracellular currents. Thus, the initial-segment sink in the experimental data (asterisk in Fig. 6b) is not detected in the model results.

6. Conclusion The availability of mutlichannel silicon probes facilitated the application of potential mapping and CSD analysis in cortical structures. This review covers the basic assumptions of field potential and CSD analysis, with application examples that revealed somatic and dendritic synaptic and action currents in the hippocampus. Changes of synaptic and action currents is critical to understanding synaptic plasticity and neurological diseases, and CSD analysis will contribute to these studies.

Acknowledgments I thank the Canadian Institutes of Health Research and the Canadian Natural Sciences and Engineering Research Council for their funding of my research on field potentials for many years, and Pascal Peloquin for collecting and analyzing the data and helping with the figures.

7. Appendix Vendor for Silicon probes (multiple electrode arrays); single and multiple shanks, 8–64 electrodes: NeuroNexus Technologies 3985 Research Park Dr. Suite 100 Ann Arbor, MI 48108 http://www.neuronexustech.com The following vendors sell multichannel acquisition and analysis systems, with silicon probe headstage Neurolynx 4055 Valley Commons Drive Suite G Bozeman, Montana 59718

24

Leung

Phone: (406) 585-4542 http://www.neuralynx.com Plexon Inc 6500 Greenville Ave., Ste 730 Dallas TX 75206 Phone: (214) 369-4957 http://www.plexoninc.com Tucker-Davis Technologies 11930 Research Circle Alachua, FL 32615 Phone: (386) 462-9622 http://www.tdt.com References 1. Johnston D, Magee JC, Colbert CM, Christie BR (1996) Active properties of neuronal dendrites. Annu Rev Neurosci 19:165–186. 2. Singer W, Gray C (1995) Visual feature integration and the temporal correlation hypothesis. Annu Rev Neurosci 18:555–586. 3. Ribary U, Ioannides AA, Singh KD, Hasson R, Bolton JP, Lado F, Mogilner A, Llinas R (1991) Magnetic field tomography of coherent thalamocortical 40-Hz oscillations in humans. Proc Natl Acad Sci USA 88:11037–11041. 4. Freeman WJ (1975) Mass action of the nervous system. Academic Press, New York. 5. Johnston D, Wu SM (1995) Extracellular field recordings. In: Foundations of cellular neurophysiology, MIT Press, Cambridge, MA, pp. 423–439. 6. Brinley FJ (1968) Volume conductor theory. In: Medical physiology, Vol 1, Mountcastle VB, (ed), CV Mosby Company, St. Louis, MO, pp. 247–251. 7. Hubbard JI, Llinas R, Quastelo MJ (1969) Electrophysiological analysis of synaptic transmission. Williams & Williams, Baltimore, MD. 8. Niedermeyer E, Lopes da Silva FH (2005) Electroencephalography (5th ed), Lippincott, Williams & Wilkins, Philadelphia. 9. Plonsey R (1969) Bioelectric phenomena. McGraw Hill, New York. 10. Nunez PL (1981) Electric fields of the brain: the neurophysics of EEG. Oxford University Press, Oxford. 11. Mitzdorf U (1985) Current source-density method and application in cat cerebral cortex: investigation of evoked potentials and EEG phenomena. Physiol Rev 65:37–100. 12. Leung LS (1990) Field potentials in the central nervous system: recording, analysis and modeling. In: Neurophysiological techniques: appli-

13.

14. 15. 16.

17.

18.

19.

20.

21.

cations to neural systems. Boulton AA, Baker GB, Vanderwolf CH, (Eds), Neuromethods, Vol 15, Humana Press, Clifton, NJ, pp. 277–312. Lorento de No R (1947) Analysis of the distribution of the action currents of nerve in volume conductors. Stud Rockefeller Inst Med Res 132:384–477. Halliday D, Resnick R (1962) Physics, part II., Wiley, New York. Leung LS (1979) Potentials evoked by the alvear tract in hippocampal CA1 of rats: II. Spatial field analysis. J Neurophysiol 42:1571–1589. Leung LS, Roth L, Canning K (1995) Entorhinal inputs to the hippocampal CA1 and dentate gyrus in the rat: a current-sourcedensity study. J Neurophysiol 73:2392–2403. Wu K, Leung LS (2003) Increased dendritic excitability in hippocampal CA1 in vivo in the kainic acid model of temporal lobe epilepsy: a study using current source density analysis. Neuroscience 116:599–616. Townsend G, Peloquin P, Kloosterman F, Hetke JF, Leung LS (2002) Recording and marking with silicon multichannel electrodes. Brain Res Brain Res Protoc 9:122–129. Drake KL, Wise KD, Farraye J, Anderson DJ, BeMent SL (1988) Performance of planar multisite microprobes in recording extracellular single-unit intracortical activity. IEEE Trans Biomed Eng 35:719–732. Hetke JF, Lund JL, Najafi K, Wise KD, Anderson DJ (1994) Silicon ribbon cables for chronically implantable microelectrode arrays. IEEE Trans Biomed Eng 41:314–321. Bragin A, Jando G, Nadasdy Z, Hetke J, Wise K, Buzsaki G (1995) Gamma (40–100 Hz) oscillation in the hippocampus of the behaving rat. J Neurosci 15:47–60.

Field Potential Generation and Current Source Density Analysis 22. Bragin A, Hetke J, Wilson CL, Anderson DJ, Engel J Jr, Buzsaki G (2000) Multiple site silicon-based probes for chronic recordings in freely moving rats: implantation, recording and histological verification. J Neurosci Meth 98:77–82. 23. Freeman JA, Nicholson C (1975) Experimental optimization of current source-density technique for anuran cerebellum. J Neurophysiol 38:369–382. 24. Holsheimer J (1987) Electrical conductivity of the hippocampal CA1 layers and application to current-source-density analysis. Exp Brain Res 67:402–410. 25. Nicholson C, Freeman JA (1975) Theory of current source-density analysis and determination of conductivity tensor for anuran cerebellum. J Neurophysiol 38:356–368. 26. Golarai G, Sutula TP (1996) Bilateral organization of parallel and serial pathways in the dentate gyrus demonstrated by current-source density analysis in the rat. J Neurophysiol 75:329–342. 27. Canning KJ, Leung LS (1997) Lateral entorhinal, perirhinal and amygdala-entorhinal transition projections to hippocampal CA1 and dentate gyrus in the rat: a current source density study. Hippocampus 7:643–655. 28. Canning KJ, Wu K, Peloquin P, Kloosterman F, Leung LS (2000) Physiology of the entorhinal and perirhinal projections to the hippocampus studied by current source density analysis. Ann NY Acad Sci 911:55–72. 29. Ketchum KL, Haberly LB (1993) Membrane currents evoked by afferent fiber stimulation in rat piriform cortex: II. Analysis with a system model. J Neurophysiol 69:261–281. 30. Leung LS, Yim CY (1986) Intracellular records of theta rhythm in hippocampal CA1 cells of the rat. Brain Res 367:323–327. 31. Kloosterman F, Peloquin P, Leung LS (2001) Apical and basal orthodromic population spikes in hippocampal CA1 in vivo show different origins and patterns of propagation. J Neurophysiol 86:2435–2444. 32. Leung LS, Shen B, Kaibara T (1992) Longterm potentiation induced by patterned stim-

33. 34.

35.

36.

37.

38.

39.

40.

41.

42.

43.

25

ulation of the commissural pathway to hippocampal CA1 region in freely moving rats. Neuroscience 48:63–74. Andersen P, Bliss TVP, Skrede KK (1971) Unit analysis of hippocampal population spikes. Exp Brain Res 13:208–221. Herreras O (1990) Propagating dendritic action potential mediates synaptic transmission in CA1 pyramidal cells in situ. J Neurophysiol 64:1429–1441. Golding NL, Spruston N (1998) Dendritic sodium spikes are variable triggers of axonal action potentials in hippocampal CA1 pyramidal neurons. Neuron 21:1189–1200. Ylinen A, Bragin A, Nadasdy Z, Jando G, Szabo I, Sik A, Buzsaki G (1995) Sharp waveassociated high-frequency oscillation (200 Hz) in the intact hippocampus: network and intracellular mechanisms. J Neurosci 15:30–46. Csicsvari J, Jamieson B, Wise KD, Buzsaki G (2003) Mechanisms of gamma oscillations in the hippocampus of the behaving rat. Neuron 37:311–322. Wolansky T, Clement EA, Peters SR, Palczak MA, Dickson CT (2006) Hippocampal slow oscillation: a novel EEG state and its coordination with ongoing neocortical activity. J Neurosci 26:6213–6229. Nicholson C, Llinas R (1971) Field potentials in the alligator cerebellum and theory of their relationship to Purkinje dendritic spikes. J Neurophysiol 36:509–531. Rall W (1964) Theoretical significance of dendritic trees for neuronal input-output relations. In: Neural theory and modeling, Reiss RF (Ed), Stanford Univ Press, Stanford, pp. 73–97. Rall W (1977) Core conductor theory and cable properties of neurons. In: Handbook of physiology: the nervous system I, Brooks VB (Ed), Am Physiol Soc., Bethesda, MD, pp. 39–97. Leung LS, Peloquin P (2006) GABAB receptors inhibit back propagating dendritic spikes in hippocampal CA1 pyramidal cells in vivo. Hippocampus 16:388–407. Leung LS (1984) Model of gradual phase shift of theta rhythm in the rat. J Neurophysiol 52:1051–1065.

Chapter 2 Current Source Density Analysis of Ongoing Neural Activity: Theory and Application Yonghong Chen, Mukesh Dhamala, Anil Bollimunta, Charles E. Schroeder, and Mingzhou Ding Abstract Current source density (CSD) is the second spatial derivative of the local field potential (LFP). CSD analysis has been used extensively to localize the pattern of transmembrane current flow in neuronal ensembles. For brain responses to repeated external stimulation, the LFP data are epoched and averaged across an ensemble of trials, from which the CSD profile is then derived. For spontaneous brain activity, however, the lack of an external triggering event makes ensemble average difficult, hampering the investigation of such important cognitive functions as anticipatory attention and working memory. In this chapter, we describe a new method called phase realigned averaging technique (PRAT), which can overcome this difficulty and achieve CSD profiles on a frequency-by-frequency basis. The method is first validated on simulation examples and then applied to LFP recordings from a monkey performing an intermodal selective attention task. Key words: Local field potential, Current source density analysis, Alpha oscillations, Cortical column

1. Introduction Local field potentials (LFPs) are an important index of brain activity. Generated by electrical currents flowing across cell membranes, LFPs, together with population spikes, provide complementary measures of ensemble neural dynamics at both the input and the output level. Typically, LFPs are measured against a distant reference and are thus vulnerable to volume-conducted far-field effects. The second spatial derivative of LFPs, called the current source density (CSD), eliminates this problem and has the ability of more precisely localizing transmembrane currents than LFPs. To date, CSD analysis has been mainly applied to stimulus-evoked neural Robert P. Vertes and Robert W. Stackman Jr. (eds.), Electrophysiological Recording Techniques, Neuromethods, vol. 54, DOI 10.1007/978-1-60327-202-5_2, © Springer Science+Business Media, LLC 2011

27

28

Chen et al.

responses. Following repeated experiments, LFP data are epoched and averaged across an ensemble of trials, triggered on some external event such as stimulus onset. The CSD profile is then computed on the averaged LFPs. This approach, widely practiced, has yielded valuable insights into the neuronal mechanisms behind sensory as well as higher order cognitive processes (1–4). Neurons in the brain are spontaneously active even in the absence of sensory stimulation. This ongoing neural activity, rich in oscillatory content, provides a window into such important brain functions as anticipation, working memory, and top-down deployment of attention (5, 6). To date, characterization of ongoing brain activity has mainly relied on time series techniques. How to extend CSD analysis to ongoing neural activity remains a challenge. The lack of an external trigger makes ensemble averaging difficult to achieve. Meanwhile, single-trial CSDs are too noisy to be a reliable indicator of meaningful neural events. In this chapter, we seek to develop a novel method called the phase realigned averaging technique (PRAT) to overcome this problem. The method is formulated in the spectral domain and can reveal CSD profiles in depth recordings on a frequency-by-frequency basis. Simulated examples are used to demonstrate its effectiveness. It is then applied to laminar LFPs sampled with a multi-contact electrode placed in the inferotemporal cortex of a macaque monkey performing an intermodal selective attention task.

2. Theoretical Background CSD analysis was first introduced in 1950s (7, 8). Subsequent publications elucidated its theoretical basis and range of applicability (1, 3, 4, 9). A standard CSD analysis has several simplifying assumptions: ohmic conductive medium, constant extracellular conductivity, homogeneous in-plane neuronal activity, and equidistant laminar electrode contacts (1). As a result of the continuity condition in current flow, the CSD I(x,y,z) in a small volume element is defined as the divergence of current flow density J from the surface of that element. Under the assumption of a purely ohmic conductive medium, it can be related to the negative of the Laplacian of the field potential F(x,y,z) (1, 3):

I = − — (s — Φ),

(1)

where s is the conductivity tensor, positive I represents an outgoing current (source), and negative I an incoming current (sink). This relationship also incorporates quasi-static approximations in the Maxwell’s equations (1, 10). Since s is symmetric, it can be made diagonal through a linear transformation (9). If we further

Current Source Density Analysis of Ongoing Neural Activity

29

assume that (1) the dendrites are elongated along the z-direction, (2) s is homogeneous, and (3) the dominant current flows are along the elongated structures only, (2.1) reduces to:

I = −s z ∂ 2 Φ / ∂z 2 ,

(2)

where the z-direction is perpendicular to the cortical surface and sz = s is a constant. Experimentally, the spatiotemporal LFPs denoted by y(z, t) are generally recorded using a linear array electrode with multiple equally spaced recording contacts sampling activity from all six layers of the cortex. The second spatial derivative in (2.2) can be estimated by the following three-point finite-difference approximation:

∂ 2y (z, t ) y (z + h , t ) − 2y (z, t ) + y (z − h , t ) ≈ , ∂z 2 h2

(3)

where h is the inter-electrode spacing. The CSD profiles are obtained by the negative of this derivative at the N–2 electrode contact positions. Here N is the number of electrode contacts, which are also referred to as channels in this chapter. The profiles at the first and the last electrode contact position can be estimated by extrapolating potential fields, but are usually left undefined.

3. Phase Realigned Averaging Technique

Physiological data are noisy. Single-trial CSD profiles are generally not very informative in identifying the precise patterns of transmembrane current flow. The signal-to-noise ratio can be improved by averaging over an ensemble of trials. For ongoing brain activity, this averaging procedure is not readily implementable because of the absence of an external trigger (e.g. stimulus onset) for trial alignment. Here we propose a method in the frequency domain to overcome this problem. Consider a LFP dataset recorded with a multicontact linear electrode. Let ym(z, t) be the data from the m-th trial, where the variable z denotes the electrode contact, t the time, and m = 1, 2, 3, …, M with M being the total number of trials. According to Fourier’s theorem, ym(z, t) can be written as the linear superposition of sinusoids of different frequencies. The phase information at frequency f is obtained by fitting x (t ) = a1 sin(2pft ) + a 2 cos(2pft ) to ym(z, t). After extracting the parameters a1 and a 2 , the phase (qm) for the m-th trial at frequency f is given by tan −1 (a1 / a 2 ). Designating certain value of the phase as the trigger, all trials can be realigned according to this trigger. Averaging the realigned trials leads to the averaged LFP for this electrode contact. This is the essence of the method.

30

Chen et al.

To realign the data from the whole electrode array, apply the following algorithm: 1. Choose a reference electrode contact or channel and estimate the phase at frequency f for all M trials. The reference channel is usually the channel with maximum signal power at f. 2. Shift each trial m by its delay qm/2pf for the data in all the channels. The relation among the data from different channels is preserved. 3. Repeat step 2 for all trials. 4. Average over the realigned trials to obtain < y (z, t * ) >:

< y (z, t * ) > =

1 M

M

∑y

m =1

m

(z, t − qm / 2pf ),

(4)

where t* is the adjusted time as the shifting operation disrupts the original physiological time. 5. Compute CSD profiles at frequency f as the negative of the second spatial derivative of < y > :

I ∝ − < y > ′′ = −

∂2 < y > . ∂z 2

(5)

The second derivative can be approximated by the finite difference formula in (3). This method will be henceforth referred to as the PRAT. The resulting LFP and CSD profile will be referred to PRAT-LFP and PRAT-CSD, respectively. By scanning across the frequency spectrum of interest, one can analyze the current sources at different frequencies. See Csicsvari et al. (11) for a similar method based on band-pass filtering.

4. Simulations Mathematical models are used to generate spontaneous spatiotemporal signals similar to that recorded in physiological experiments. The PRAT algorithm is then used to perform CSD analysis. By comparing the PRAT-CSD pattern with the corresponding mathematical functions, we are able to assess the validity of our method. Obviously, such a direct cross-validation is not possible in the analysis of real physiological data where the answer is not known a priori. 4.1. The Model

Experimentally, the data to be analyzed come from a cortical column. Here a cortical column is represented by the unit interval.

Current Source Density Analysis of Ongoing Neural Activity

31

The transmembrane current is assumed to have a sinusoidal profile containing a source and a sink: az cos(2pf z z + q z ), where az, fz, qz (=−p/2) are amplitude, frequency, and phase. This term is referred to as the spatial dynamics term. Solving the differential equation ∂ 2j = az cos(2pf z z + q z ), ∂z 2

we obtain

f (z) = C1z + C 2 − az cos(2pf z z + q z ) / (2pf z )2 , where C1 and C2 are constants. The temporal dynamics Y(t) is modeled in two different ways: (1) a 10 Hz sinusoidal function at sin(2pf t t + qt ) and (2) a second-order autoregressive process [AR(2)]: Y (t ) = aY (t − 1) + bY (t − 2) + x (t ) , with a = 0.6, b = −0.9, and x(t) being a white Gaussian noise. Like the 10-Hz sinusoid, the AR(2) process chosen this way also has a spectral peak at 10 Hz. Multiplying the space- and time-dependent functions and adding noise, we generate spatiotemporal LFP signals y(z, t):

y (z, t ) = j (z)ψ (t ) + h(z, t ),

(6)

where h(z, t) is a stochastic process with long-range power law (power a 1/frequency) correlation in all channels as well as random amplitude Gaussian noise in different channels (12). This choice is motivated by the observation that 1/f spectra are commonly observed in EEG and LFP recordings from the mammalian cortex (13–15). 4.2. CSD Analyses with PRAT

The above models were simulated and these data were assumed to be acquired by a multi-electrode with 14 equally spaced contacts. Figures 1 and 2 show the results for the two different temporal functions, 10-Hz sinusoid and AR(2), respectively. In both figures, panel (a) gives the color-coded power spectra at different contacts or channels (vertical axis) where oscillation at 10 Hz is clearly seen; panel (b) displays the superposition of 500 trials with random initial phases at the reference contact denoted by zk; panel (c) shows the same 500 trials after phase realignment; panel (d) is the PRAT-LFP. The time label is from −100 to 0 ms. This label is motivated by the fact that, in the next section, experimental data from the prestimulus time period will be considered where the stimulus onset is defined as 0 ms. It is evident that the PRATLFPs in Figs. 1d and 2d are not able to reveal the underlying current/source pattern in the mathematical model, suggesting that LFPs have limited ability to precisely localize generators of transmembrane current flow. The PRAT-CSD profiles obtained by taking the second spatial derivative on the PRAT-LFPs are

32

Chen et al.

Fig. 1. Simulation example where the temporal dynamics is defined by a 10 Hz sinusoid. (a) Power spectra (5–80 Hz) as a function of electrode contact. (b) Trials from the reference contact before phase realignment. (c) The same trials after phase alignment. (d) The PRAT-LFP profile. (e) The PRAT-CSD profile. (f) The spectrum of the total transmembrane current flow. The horizontal axis is frequency (Hz) in (a) and (f). It is time otherwise in the unit of millisecond.

shown in Figs. 1e and 2e. The source and the sink pairs in the CSD profiles are clearly seen at around electrode contact 4 and 11. Temporally, the source-sink pair oscillates at a frequency of 10 Hz. Thus, the PRAT method recovers the dynamics built into the mathematical model. The above analysis can be carried out for each frequency from 5 to 80 Hz. Integrating the rectified PRAT-CSD profile over space and a given time interval yields the amount of transmembrane current at frequency f. The results are plotted in panel (f) and are called the CSD spectra. From the CSD spectrum, the greatest amount of transmembrane current is seen to occur at10 Hz. This again is in agreement with the conditions implemented in the mathematical model.

Current Source Density Analysis of Ongoing Neural Activity

33

Fig. 2. Simulation example where the temporal dynamics is defined by an AR(2) process; the conventions are otherwise the same as in Fig. 1.

5. Application to Experimental Data Field potential oscillations are ubiquitous in the nervous system. Depending on the signal rhythmicity, these oscillations are classified according to the following approximate nomenclature: delta (1–3 Hz), theta (4–7 Hz), alpha (8–12 Hz), beta (13–25 Hz), and gamma (25–90 Hz). The alpha rhythm is a prominent oscillatory activity in the 8–12 Hz band in EEG recordings over the occipital and parietal areas during wakefulness (16, 17). Nearly 80 years after its discovery (18), its genesis, cellular mechanisms, and functions remain unclear. Early work emphasized the pacemaking

34

Chen et al.

role of the thalamus (19). More recent evidence suggests that it might be of a cortical origin (20). This problem is considered in this chapter by recording LFP and multiunit activity (MUA) from the inferotemporal cortex of a behaving macaque monkey. The PRAT method, in conjunction with other methods such as CSDMUA coherence, is applied to address two problems: (1) laminar location of alpha current generators and (2) effect of prestimulus alpha oscillation on stimulus-evoked processing. 5.1. Experimental Paradigm

The data considered here is part of a previously published study (21, 22). A male macaque monkey was trained to discriminate stimuli in both visual and auditory domains. There are two conditions. In Condition 1, the monkey was presented with a mixed stream of auditory and visual stimuli. In each sensory modality, a standard stimulus occurred 86% of the time and an oddball stimulus 14% of the time. Selective attention was manipulated by instructing the monkey to respond to the oddball stimulus in the attended modality only. Task difficulty was balanced between the modalities. In Condition 2, the monkey performed the oddball detection task in the auditory domain in the absence of visual stimulation. The reason for analyzing activity in visual cortices during auditory discrimination was that the discrimination kept the monkey verifiably alert without using visual stimuli, so that we could study spontaneous ongoing neural activity.

5.2. Recordings

LFP and MUA were sampled with a linear array electrode with 14 contacts spanning all six cortical layers in the inferotemporal cortex. Data from one penetration, collected during periods of adequate task performance (i.e. >80% target detection), were analyzed to demonstrate the method presented above. Problem 1: Laminar generators of ongoing alpha oscillation. Data from Condition 2 were analyzed. The length of a contiguous segment of spontaneous ongoing activity was on average 30 s long and there were five such segments for the penetration. After highpass filtering (3 Hz, zero phase-shift) and down-sampling to 200 Hz, the LFP data were further divided into epochs of 200 ms in duration, which were considered trials or realizations of an underlying stochastic process. The power spectrum of each recording contact was estimated and the contact showing the highest power spectral density at 10 Hz was chosen as the reference channel. The PRAT method was used to obtain PRAT-LFP and PRAT-CSD. Figure 3a shows the results where the reference contact is channel 6. The PRAT-LFPs (solid lines) exhibit clear oscillation at 10 Hz. The PRAT-CSD (color coded) revealed an alpha current generator in the supra-granular layers (around contacts 5–7) with an underlying source/sink/source configuration. The alpha current generator in the infra-granular layers (around contact 10) was relatively weak. No alpha current generator was seen in the granular layer. The generator around contact three is believed to reflect

Current Source Density Analysis of Ongoing Neural Activity

35

Fig. 3. Analysis of spontaneous alpha activity in inferotemporal cortex. (a) PRAT-CSD profile displayed as a color-coded plot, which is the second spatial derivative of the PRAT-LFPs (solid traces). A single epoch of MUA from two contacts is superimposed. (b) Laminar distribution of the peak (10 Hz) LFP power across recording contacts. (c) CSD-MUA coherence spectra at different contacts.

dendritic backpropagation. In Fig. 3b, the alpha band power is plotted as a function of the recording contact. The highest power occurs around channels 5 and 6, suggesting that the alpha current generator in the supra-granular layers may play an important role in the overall organization of alpha activity in the column. The respective roles of the alpha current generators can be further delineated by examining the concomitant MUA data. In Fig. 3a, an epoch of MUAs at channels 6 and 10 are overlain on the CSD profile. The MUA near the supra-granular layer alpha generator varies rhythmically with the underlying current, while the MUA near the infra-granular alpha generator is not modulated by the current. This suggests that the alpha current generator in the supra-granular layers is possibly the pacemaker of the alpha rhythm in the column. We confirmed this impression by calculating the CSD-MUA coherence. The MUA data were epoched the same way as the LFP data and down-sampled from 2 kHz by taking a temporal average in nonoverlapping windows of 5 ms duration to achieve effectively the same sampling resolution of 200 Hz as the down-sampled LFPs. The coherence between single-trial CSDs around alpha current generators identified by the PRAT-CSD method and the corresponding mean-centered single-trial MUAs was calculated by the multivariate autoregressive

36

Chen et al.

(MVAR) spectral analysis method (23). Figure 3c gives the CSDMUA coherence at channels 5, 6, 7, and 10 where relatively high transmembrane current flows were found. In the supra-granular layers, the CSD-MUA coherence is relatively strong, reaching values close to 0.16, while in the infra-granular layers, the CSDMUA coherence is close to zero. This supra-granular bias is consistent with the single epoch data in Fig. 3a. The above results demonstrate that the inferotemporal cortex contains an alpha pacemaker in the supra-granular layers, in agreement with the suggestion that the alpha rhythm might be of a cortical origin. A more thorough analysis of this problem has been carried out by Bollimunta et al. (24). The alpha pacemaker in the supra-granular layers has a source/sink/source configuration. In light of the substantially enlarged basal dendritic arbor reaching the size of 400 mm in IT (25), this alpha generator most likely reflects the activity of superficial pyramidal neurons. The CSD-MUA coherence further suggests basal dendritic excitation. It is worth noting that Lukach et al. (26), in an in vitro slice study, have shown that the supra-granular layers contain the pacemaker of alpha range oscillations in the entorhinal cortex in rats. Problem 2: Effect of prestimulus alpha oscillation on stimulus evoked response. The data recorded under Condition 1 were considered. The continuous LFP recordings were divided into 600 ms (−200, 400 ms) epochs based on the standard visual stimulus triggers. The prestimulus interval was defined to be from −200 to 0 ms, where 0 ms denotes stimulus onset. After data preprocessing, approximately 2,000 trials during which the monkey paid attention to the visual stimulus were made available for further analysis. The stimulus-evoked CSD from contact 5 in the supragranular layers was computed using the conventional method and shown in Fig. 4. From this figure, an early stimulus processing

CSD

late

early

0.1

0

−0.1

0

50

100

150

200

250

time (ms)

Fig. 4. Stimulus evoked current source density from a contact in the supra-granular layers. Early and late evoked responses are marked.

37

Current Source Density Analysis of Ongoing Neural Activity

period (50–100 ms) and a late processing period (100–200 ms) were defined. Figure 5a shows the CSD profile of the ongoing prestimulus alpha oscillation determined by the PRAT method. Note the similarity between the CSD profile in Fig. 5a and that in Fig. 3a. Although obtained under different experimental conditions, these transmembrane current flow patterns are likely to reflect the same physiological generating mechanisms. To examine the relation between prestimulus alpha oscillation and stimulus-evoked response, the magnitude of the prestimulus oscillation at 10 Hz was estimated on a trial-by-trial basis. A template matching method was used for this purpose. For single-trial LFP data, the second spatial derivative see (3) was calculated to yield single-trial CSD profiles. The PRAT-CSD in Fig. 5a was used as a template and moved along a given single-trial a

b

c

d late

1

normalized poststimulus TTCF

early

0.95

early: R=0.1 late: R=0.9

0.9

0.85

0.8

0.75

0.2

0.4 0.6 0.8 normalized prestimulus TTCF

1

Fig. 5. Effect of prestimulus alpha oscillation on stimulus-evoked response. (a) PRAT-CSD profile at 10 Hz during the prestimulus time period, which is used as a template for measuring the strength of single-trial alpha activity. (b) A trial having a strong match index value with the template (note the CSD pattern between the two solid lines). (c) Stimulus evoked CSD profile. (d) Early (blue) and late (red) evoked responses plotted against the normalized magnitude of prestimulus alpha oscillation. Here TTCF stands for total transmembrane current flow.

38

Chen et al.

CSD profile to find the best pattern match defined in terms of cross correlation. The correlation coefficient at the best match location was retained as an index of the magnitude of the prestimulus alpha activity in that trial. The procedure was repeated for all trials. Figure 5b shows a single-trial CSD profile with a high matching index value; the similarity between the transmembrane current flow pattern that is between the two vertical lines and the PRAT-CSD in Fig. 5a is noticeable. All the trials were sorted into groups of 500 trials, according to the values of the template matching index, each group having 90% of the trials overlapped with the previous one, starting from the lower matching index value to the highest. For the 500 trials in each group, a PRAT procedure was performed to yield a group PRAT-CSD profile. The total transmembrane current flow (TTCF) was computed by integrating the rectified ongoing laminar CSD over time (100 ms) and space (depth). For the same group of trials, the stimulus-evoked CSD profile, shown in Fig. 5c, was calculated in the conventional way. The TTCF during the early (50–100 ms) and the late (100–200 ms) poststimulus time period were obtained and plotted against the prestimulus alpha TTCF in Fig. 5(d) (blue for early and red for late). All quantities were normalized to a maximum value of 1. The solid straight lines represent least squares fits. The correlation coefficient between prestimulus alpha activity and the early evoked response is 0.1, while the correlation coefficient between prestimulus alpha activity and the late component is 0.9. The above results demonstrate that the magnitude of the prestimulus alpha oscillation can affect stimulus processing. In particular, the prestimulus alpha oscillation is shown to be more strongly correlated with the late evoked component than the early evoked component. This observation appears to contradict intuitive expectations and thus calls for a possible explanation. In the cortex, excitatory neuronal information transmission is mediated by the release of the neurotransmitter glutamate. There are two major classes of glutamate receptors: AMPA and NMDA. On the one hand, experimental evidence suggests that the early evoked component reflects the fast response to stimulus input and is mainly mediated by AMPA receptors (27, 28). On the other hand, the late evoked component is apparently related to neuronal responses to feedback input from higher order areas and is thought to engage the NMDA receptors (28, 29). Spontaneous field potential oscillations in the theta and alpha range before the onset of stimulus reflect the cyclical variation in the excitability of neuronal ensembles (30–33). They involve the potentiation of NMDA receptors (34–36). The NMDA-mediated increase in excitability is an essential ingredient in recent theories of attention and memory (29, 37). This differential involvement of glutamate receptors with distinct stages of information processing

Current Source Density Analysis of Ongoing Neural Activity

39

and prestimulus facilitation of NMDA receptors may underlie the correlation pattern observed between prestimulus ongoing alpha activity and stimulus evoked response.

Acknowledgment This work was supported by NIH grants MH070948, MH079388, and MH060358. References 1. Mitzdorf U (1985) Current source density method and application in cat cerebral cortex: investigation of evoked potentials and EEG phenomenon. Physiol Rev 65:37–100. 2. Nakagawa H, Matsumoto N (2000) Current source density analysis of ON/OFF channels in the frog optic tectum. Prog Neurobiol 61:1–44. 3. Nicholson C, Freeman JA (1975) Theory of current source-density analysis and determination of conductivity tensor for anuran cerebellum. J Neurophysiol 38:356–368. 4. Schroeder CE, Steinschneider M, Javitt DC, Tenke CE, Givre SJ, Mehta AD, Simpson GV, Arezzo JC, Vaughan HG (1995) Localization of ERP generators and identification of underlying neural processes. Electroencephalogr Clin Neurophysiol Suppl 44:55–75. 5. Liang H, Bressler SL, Ding M, Truccolo WA, Nakamura R (2002) Synchronized activity in prefrontal cortex during anticipation of visuomotor processing. Neuroreport 13:2011–2015. 6. Zhang Y, Wang X, Bressler SL, Chen Y, Ding M (2008) Prestimulus cortical activity is correlated with speed of visuomotor processing. J Cogn Neurosci 20(10):1915–1925. 7. Pitts W (1952) Investigations on synaptic transmission. In: Cybernetics: transactions of the ninth conference. von Foerster H (Ed), Josiah Macy Jr. Foundation, New York, pp 159–166. 8. Howland B, Lettvin JY, McCulloch WS, Pitts W, Wall PD (1955) Reflex inhibition by dorsal root interaction. J Neurophysiol 18:1–17. 9. Freeman JA, Nicholson C (1975) Experimental optimization of current source-density technique for anuran cerebellum. J Neurophysiol 38:369–382. 10. Hämälainen M, Hari R, Ilmoniemi RJ, Knuutila J, Lounasmaa OV (1993) Magnetoen cephalography – theory, instrumentation, and

11.

12.

13. 14.

15.

16. 17.

18.

19. 20.

applications to noninvasive studies of the working human brain. Rev Modern Phys 65:413–497. Csicsvari J, Jamieson B, Wise K, Buzsaki G (2003) Mechanisms of gamma oscillations in the hippocampus of the behaving rat. Neuron 37:311–322. Rangarajan G, Ding M (2000) Integrated approach to the assessment of long range correlation in time series data. Phys Rev E 61:4991–5001. Buzsaki G, Draguhn A (2004) Neuronal oscillations in cortical networks. Science 304: 1926–1929. Freeman WJ, Rogers LJ, Holmes MD, Silbergeld DL (2000) Spatial spectral analysis of human electrocorticograms including the alpha and gamma bands. J Neurosci Methods 95:111–121. O’Connor SC, Robinson PA (2004) Unifying and interpreting the spectral wavenumber content of EEGs, ECoGs, and ERPs. J Theor Biol 231:397–412. Shaw JC (2003) Brain’s alpha rhythm and the mind. Elsevier, Amsterdam Niedermeyer E (2005) The normal EEG of the waking adult. In: Electroencephalography: basic principles, clinical applications and related fields. Niedermeyer E (Ed), Williams and Wilkins, Baltimore, pp 149–173. Berger H (1929) Über das elektroencephalogramm des menschen (On the electroencephalogram of man). Arch Psychiatr Nervenkr 87:527–570. Andersen P, Andersson SA (1968) Physiological basis of the alpha rhythm. Appleton-CenturyCrofts, New York Lopes da Silva FH, Vos JE, Mooibroek J, Van Rotterdam A (1980) Relative contributions of intracortical and thalamo-cortical processes in the generation of alpha rhythms, revealed by

40

21.

22.

23.

24.

25.

26.

27.

28.

Chen et al. partial coherence analysis. Electroencephalogr Clin Neurophysiol 50:449–456. Mehta AD, Ulbert I, Schroeder CE (2000a) Intermodal selective attention in monkeys. I. Distribution and timing of effects across visual areas. Cereb Cortex 10:343–358. Mehta AD, Ulbert I, Schroeder CE (2000b) Intermodal selective attention in monkeys. II. Physiological mechanisms of modulation. Cereb Cortex 10:359–370. Ding M, Bressler SL, Yang W, Liang H (2000). Short-window spectral analysis of cortical event-related potentials by adaptive multivariate autoregressive modeling: data preprocessing, model validation, and variability assessment. Biol Cybern 83:35–45. Bollimunta A, Chen Y, Schroeder CE, Ding M (2008) Neuronal mechanisms of cortical alpha oscillations in awake-behaving macaques. J Neurosci 28(40):9976–9988. Elston GN, Tweedale R, Rosa MGP (1999) Cortical integration in the visual system of the macaque monkey: large-scale morphological differences in the pyramidal neurons in the occipital, parietal and temporal lobes. Proc R Soc Lond 266:1367–1374. Lukatch HS, MacIver MB (1997) Physiology, pharmacology, and topography of cholinergic neocortical oscillations in vitro. J Neurophysiol 77:2427–2445. Hackley SA (1993) An evaluation of the automaticity of sensory processing using eventrelated potentials and brain-stem reflexes. Psychophysiology 30:415–428. Roelfsema PR, Ooyen A (2005) Ation-gated reinforcement learning of internal representa-

29.

30. 31.

32.

33.

34. 35.

36.

37.

tions for classification. Neural Comput 17: 2176–2214. Schroeder CE, Mehta AD, Foxe JJ (2001) Determinants and mechanism of attention modulation of neural processing. Front Biosci 6:672–684. Bishop G (1933) Cyclic changes in excitability of the optic pathway of the rabbit. Am J Physiol 103:213–224. Steriade M, Nunez A, Amzica F (1993) A novel slow oscillation of neocortical neurons in vivo: depolarizing and hyperpolarizing components. J Neurosci 13:3252–3265. Sanchez-Vives MV, McCormick DA (2000) Cellular and network mechanisms of rhythmic recurrent activity in neocortex. Nat Neurosci 3:1027–1034. Lakatos P, Shah AS, Knuth KH, Ulbert I, Karmos G, Schroeder CE (2005) An oscillatory hierarchy controlling neuronal excitability and stimulus processing in the auditory cortex. J Neurophysiol 94:1904–1911. Silva LR, Amitai Y, Connors BW (1991) Intrinsic oscillations of neocortex generated by layer 5 pyramidal neurons. Science 251:432–435. Flint AC, Connors BW (1996) Two types of network oscillations in neocortex mediated by distinct glutamate receptor subtypes and neuronal populations. J Neurophysiol 75:951–956. Placantonakis DG, Welsh JP (2001) Two distinct oscillatory states determined by the NMDA receptor in rat inferior olive. J Physiol 534:123–140. Vertes RP (2005) Hippocampal theta rhythm: a tag for short term memory. Hippocampus 15:923–935.

Chapter 3 The Juxtacellular Recording-Labeling Technique Didier Pinault Abstract The single-cell juxtacellular recording–labeling technique makes it possible to label the neuron recorded extracellularly. It is a very useful tool for achieving single-cell structure–function correlation studies in living, intact neural networks and for determining their phenotype and genotype. It can reveal the overall picture of the smallest neurons, including interneurons. It can be combined with other electrophysiological techniques (e.g. electro-encephalographic recordings and intracerebral electrical stimulation), electron microscope, immunohistochemical, molecular and/or genetic techniques. Its principle consists in iontophoresing tracer molecules into the membrane of the neuron being recorded. This is done under continuous electrophysiological monitoring, allowing the retrieval of the neuron labeled in more than 85% of attempts. Continuous DC recordings suggest that the juxtacellular filling “or tickling” procedure produces a transient micro-electroporation, which allows the internalization of the tracer into the intracellular space. Since this procedure allows single neurons to be recorded for long periods, many electrophysiological features can be collected, and the finest and remotest axonal ramifications can be marked. In spite of some limitations and pitfalls, the juxtacellular technique remains the high standard for investigating the genetic, molecular, physiological, and architectural bases of cell–cell communication. It is also a very versatile and useful tool when it comes to decipher, for instance, the molecular, cellular, and network mechanisms of brain state, physiological, and pathological oscillations. Key words: Neuronal tract tracing, Iontophoresis, Biocytin, Neurobiotin, Single-cell labeling, Extracellular, Structure-function, Neural network, Ultrastructure

1. Overview and Scope of this Chapter

The purpose of this chapter is to gather practical information and advice for acquiring expertise in employing the single-cell juxtacellular recording/marking procedure devised and developed in whole animal preparations (1, 2). Some of our results obtained with this single-cell recording/labeling technique are illustrated. Its advantages and limitations are presented with reference to other approaches employing recording glass micropipettes.

Robert P. Vertes and Robert W. Stackman Jr. (eds.), Electrophysiological Recording Techniques, Neuromethods, vol. 54, DOI 10.1007/978-1-60327-202-5_3, © Springer Science+Business Media, LLC 2011

41

42

Pinault

Also, this chapter provides insight into the kinds of questions that can be addressed using the juxtacellular technique considering future trends.

2. Historical Considerations 2.1. The Golgi Impregnation Method

To understand how the brain works, modern neuroscientists seek to correlate the dendritic and axonal architectures of single nerve cells with their biophysical, molecular, synaptic, and genetic properties, which are specifically associated with a brain operation or a neurobiological disorder. For a long period, physiological and morphological data of single nerve cells have been obtained separately. It all started with the silver impregnation method discovered by Camillo Golgi in 1873, which had laid the foundations of the connectionist view of neural organization (3, 72, 73). The Golgi method, however, generally failed to impregnate myelinated axons in adult brains (4). In addition, not all axons that were impregnated could be followed to either their sites of termination or cells of origin. Therefore, the limitations and capriciousness of the Golgi method represent serious drawbacks. Furthermore, it cannot be combined with immunocytochemical procedures. Thus, the postmortem Golgi technique is used less and less nowadays.

2.2. Iontophoretic Application of Neuronal Tracers

Meanwhile, numerous neuroanatomical tract-tracing methods have been contrived and developed (5–7). Most of the purely anatomical techniques currently available employ a tracer-filled microsyringe or coarse glass pipette to expel by pressure, or by iontophoresis, a large amount of tracer into the brain tissue. Once captured by and internalized in the neuronal elements, the neuronal tracer is transported anterogradely and/or retrogradely to the dendritic and axonal ramifications. An appropriate histological procedure allows the tracer molecules that are internalized into the neuronal elements to be revealed. These tracing methods can be combined with immunocytochemical and electron microscopic techniques. Thereby, anatomical and synaptic relationships between brain regions have been assessed. However, the overall picture of a given dye-filled neuron is usually precluded. In addition, these tracing methods do not allow precise connectional studies of local neural networks and of direct correlation with physiology.

2.3. Combining Electrophysiological Recording with Neuronal Tracing

Some of the drawbacks of the pure histological and anatomical methods have been overcome by electrophysiological techniques that employ injectable dyes. It all started in 1949 when (71) manufactured the first electric recording glass micropipette, which contained a suitable electrolyte. This major breakthrough has been

The Juxtacellular Recording-Labeling Technique

43

generating the great, still increasing population of electrophysiologists, who study the physiological properties of single neurons at both extracellular and intracellular levels (8). The electrophysiological data obtained so far have tentatively been correlated with the architecture of the Golgi-impregnated neurons. This deductive reasoning lasted till the advent of injectable tracers (9, 74, 75). Dissolved in an electrolyte, such a substance, once intracellularly injected into the recorded neuron, is made visible after processing the nervous tissue histologically. During the last quarter of the twentieth century, neuroscientists wishing to examine the functional architecture of single nerve cells had been succeeding in studying intrinsic and synaptic membrane properties of single neurons in combination with direct intracellular injection of tracer. The outcome of this technical breakthrough has thus deepened our understanding of the structure of physiologically identified neurons. 2.4. Limitations of Direct Intracellular Application of Tracers

The use of tracer-filled intracellular micropipettes remains a powerful means for structure–function correlation studies. In spite of its unique advantages, direct intracellular iontophoresis of markers may damage or kill the neuron under study, yielding a relatively low rate of success. Indeed, having injected the recorded neuron, the experimenter rapidly withdraws the micropipette that is being sealed with the recorded neuron’s membrane without controlling its electrical behavior. These conditions would not allow us to know whether the injected neuron is still viable at this stage. Maintaining the intracellularly injected neurons alive in an in vivo preparation remains a difficult task as long as they cannot be visually or electrophysiologically watched over after the staining procedure. Therefore, it is important to have a reliable control not only before but also and especially at the end of the filling procedure. Furthermore, because of the complications in getting stable penetrations, marking neurons with intracellular micropipettes remains difficult to apply for labeling small cells and exploring deep brain regions especially in whole animal preparations. In addition, unless applying intra-axonal dye injection (10), it often precludes the whole picture of the axonal arbor of the recorded neurons (11), particularly those forming long-range connections. Therefore, the intracellular approach, which remains irreplaceable for studying membrane events that are responsible for the firing patterns, may be discouraging when it comes to unraveling the wiring of cells and the set of connections in in vivo preparations. Single-cell labeling has also been attempted using extracellular, marker-filled, recording micropipettes with some success, but it is impossible to ensure that the labeled neuron is the one that was being recorded (2, 12, 13). These pipettes have a tip diameter of a few micrometers and the intensity of the current pulses, which are delivered in a blind manner by a high compliance

44

Pinault

iontophoresis device, ranges from 50 to 2,000 nA. With such intensity, the current pulses usually produce damage at the recording site, and the surrounding neuronal elements can subsequently take up the applied marker (2). The main advantage of this extracellular technique is to identify one cell or more on the basis of electrophysiological criteria and afterwards to stain a few (rarely one) neurons at the recording site. As a corollary, some (or all) of the tracer-filled cells including their entire axonal projection can be traced individually, allowing one to elaborate some fundamental principles of anatomical organization (14–18). However, great care must be taken when analyzing the histochemical material as passing fibers can also be stained at the application site (19). Therefore, blind extracellular iontophoresis of marker molecules with recording pipettes may be reliable for tracing neuronal pathways, but it is not quite suitable for structure–function correlation studies, since it is still a challenge to ascertain that a tracer-filled neuron was the recorded unit, simply because the application procedure is made without electrophysiological monitoring. The so-called extracellular technique is neither more nor less than the multiunit juxtacellular labeling technique (2).

3. Principle of the Juxtacellular Technique 3.1. Overview

The extracellular technique has been refined such as to stain the recorded neuron at will under direct electrophysiological control (1, 2). Thereby, the powerful and versatile single-cell juxtacellular recording-labeling technique, which requires the use of a tracerfilled sharp micropipette (tip diameter: 0.5–1.5 mm; Fig. 1a), has been devised to stain long-axon projection neurons as well as interneurons (Fig. 2). The micropipette tip must physically and electrically be brought in juxtaposition with the membrane of the neuron being recorded (Fig. 1b, c2). This is achieved under continuous electrophysiological control (Fig. 1c1–c4). The principle of the single-cell juxtacellular filling procedure rests on the electrical stimulation of the recorded neuron’s membrane with iontophoretic positive rectangular nanocurrent pulses (0.5–8 nA; 200 ms duty cycle; Fig. 1c2), which are delivered through the bridge circuitry of an intracellular preamplifier (Fig. 1b) still under continuous electrophysiological monitoring of the electrode tip-cell

Fig. 1. (continued) It is achieved with a tracer/saline filled micropipette under continuous electrophysiological monitoring. (c1) Extracellular recording. (c2) Physical approach and juxtacellular recording and tracer iontophoresis with square pulses of nanocurrents (0.5–8 nA, 200 ms on, 200 ms off). Note that the current pulses drive the firing of the recorded neuron. (c3) Back to the extracellular recording. (c4) Gentle physical withdrawal of the micropipette from the recorded neuron. Note that the withdrawal is correlated with a progressive diminution of the action potential amplitude. See text for more details. (a) Is adapted from Pinault (25).

The Juxtacellular Recording-Labeling Technique

45

Fig. 1. Experimental design and principle of the juxtacellular technique. (a) Photomicrographs showing the main physical features of a typical sharp-tip glass micropipette for juxtacellular or intracellular recording. The micropipette’s tip is shown below at a higher magnification. (b) Schematic diagram of a conventional intracellular preamplifier. It is used to record either intracellular, extracellular, or juxtacellular voltage signals (V) and to inject nanocurrent (I) through the tip of the glass micropipette while recording. The Wheatstone bridge is used to determine the resistance of the micropipette by adjusting a known resistance (Rb) so that the measure current is zero. A resistance in series (Rs ³ 100 megohms) must be quite larger (×10–100) than the micropipette resistance (Rm < 100 megohms) to measure the full signal at the micropipette tip and draw negligible current from the signal source. The low pass filtering properties of the micropipette and its amplifier are due to inevitable capacitances between the micropipette and the preparation (ground). That is why a feedback circuit (Rc) is required to compensate the capacitance effects (Cp). (c1–c4) The four steps of the juxtacellular filling procedure.

46

Pinault

Fig. 2. Juxtacellular labeling with biocytin of projection (a) and local-circuit (b) neurons. (a) Dendritic and axonal ramifications of a thalamic reticular nucleus (TRN) cell projecting to the thalamus. This photomicrograph shows parts of its somatodendritic and axonal arborizations within the same horizontal 100-micrometers thick horizontal section. The right inset shows dense axon terminations at higher magnification. (b) A basket cell of the cerebellar cortex that had its main axonal and dendritic projections within the same saggital 100-mm thick section. Its axon gave off seven basket-like arborizations, each known to enclose the perikaryon of a Purkinje cell.

membrane interaction. The intensity of the delivered current must be high enough to entrain the firing of the recorded neuron (Fig. 1c2) but not too high so as to injure or damage the neuron. These conditions allow the internalization of the tracer molecules into the neuron being recorded. The intensity of the juxtacellular current pulses is 10–100 times lower than that usually applied with the extracellular technique. The single-cell juxtacellular recording–labeling technique well circumvents the specific problems encountered with the intracellular and extracellular methods. Although relatively simple, the juxtacellular recording–labeling technique requires intracellular equipment and conditions, and a good deal of electrophysiological experience. We will see that this single-cell labeling technique makes investigating the physiological and structural bases of cell–cell communication very reliable for structure– function correlation studies, provided care is taken to avoid the possible drawbacks and pitfalls that are described later.

The Juxtacellular Recording-Labeling Technique

47

From an electrochemical point of view, the tracer molecules that are internalized during juxtacellular iontophoresis with positive current pulses are, theoretically, those that are positively charged and/or those that are cotransported with the ions of the saline solution. As the clearest possible evidence, the internalization of the tracer molecules is firing-dependent since negative current pulses, which inhibit cellular firing, remain ineffective in labeling the recorded neuron (1, 2). 3.2. Internalization of the Tracer into the Recorded Neuron

The following properties of the juxtacellular technique provide the experimenter with convincing evidence that the recorded neuron is labeled: 1. The fact that the juxtacellular current pulses systematically increase firing frequency unambiguously indicates that the tracer injection is targeted to the neuron being recorded (Fig. 3a1–a3). Iontophoresing marker molecules onto the target neuron’s membrane under continuous display of its electrical behavior require constant adjustment of the ejecting current, which must entrain its firing without provoking injurious discharges. Controlling the electrical behavior of the target neuron for the duration of the filling procedure allows the experimenter to keep the recorded neuron alive until the complete withdrawal of the micropipette tip from the recorded neuron, which is not possible (or extremely difficult) with the intracellular method, and the retrieval of the neuron’s histological silhouette in more than 85% of filling attempts. 2. When expelling the neuronal tracer only in the immediate extracellular space of the recorded neuron, namely with currents subthreshold to those that modulate its firing, the ejected tracer fails to be internalized into neuronal elements (2). Therefore, to have the label internalized into the recorded cell, the juxtacellular iontophoretic current must be of threshold or suprathreshold intensity to “tickle,” but subthreshold to injure, the target neuron. 3. An excess of tracer deposit can be seen against the membrane of the labeled neuron, especially after long-lasting juxtacellular application (>5 min) and after a relatively short survival time (Fig. 3a4). 4. As a control experiment, following a voluntary electrical killing of the neuron that had just been recorded and entrained juxtacellularly with a tracer-filled micropipette (Fig. 3b1), the histochemical reaction product usually discloses a badly damaged neuron (Fig. 3b2). 5. The micropipette often causes hemorrhage along its descent into the brain tissue. So, after the histochemical process, the track left by the micropipette can often be seen just above the tracer-filled neuron’s cell body (2, 20) (Fig. 3a4).

48

Pinault

Fig. 3. The juxtacellular technique labels the neuron being recorded. (a1) Extracellular recording of a typical thalamic reticular nucleus (TRN) cell that exhibits a short-latency high-frequency burst in response to an electrical stimulation of the internal capsule (IC stim). (a2) Juxtacellular recording/labeling of the same TRN cell with a Neurobiotin-filled micropipette. The intensity of the rectangular current pulses (200 ms on/200 ms off; lower trace: current monitor) is ~8 nA. The asterisk indicates the moment when the background noise starts to increase (about 50 s after the beginning of pulse application), meaning that the current passing through the micropipette tip starts to excite the cell membrane; moreover, the same current pulses suddenly and strongly entrain the firing of the recorded neuron (arrow). (a3) Spontaneous activity recorded extracellularly after juxtacellular tracer application. (a4) After an application of 10 min and a postinjection survival period of 1 h, one TRN perikaryon with long dendritic profiles and fine protrusions is found at the intended stereotaxic aim in the TRN (horizontal section photographed from a dorsal view). Arrows indicate portions of its axon trunk.

The Juxtacellular Recording-Labeling Technique

49

6. A reversible electrically-induced microporation is the most likely mechanism responsible for the internalization of the tracer molecules during the juxtacellular filling procedure. Indeed, during this procedure, a hyperpolarizing DC shift of a few millivolts of the membrane potential can be recorded, and a microlesion of the tracer-filled neuron’s membrane can be observed after a relatively short survival time (2). 3.3. Required Equipment

The material requirements for successful application of the juxtacellular recording–labeling technique are the same as for intracellular recordings: (1) A heavy stereotaxic frame that must be well isolated from any source of mechanical vibration; (2) an intracellular preamplifier, a signal conditioner (gain and bandpass), an oscilloscope, and an analog-to-digital converter to store voltage and current signals on a computer disk for on-line and off-line analyses; (3) a reliable stepping micromanipulator whereby the micropipette can be advanced in a well-controlled mode with steps ranging from 1 to 3 micrometers; (4) a stereotaxic brain atlas (1) for the accurate insertion of the recording/labeling micropipette and (2) to follow the trajectory of its tip during the descent, allowing to the first approximate identification of the spatial location of the recorded neurons. Conventional intracellular amplifiers provide the source of iontophoretic nanocurrent passing through the micropipette tip, while the electric activity (voltage signals) of the neuron is being recorded. The major elements constituting such an amplifier are illustrated in a schematic diagram (Fig. 1b). For greater details, the reader may refer to the literature (21). In brief, a microelectrode amplifier requires a main control unit with: (1) a bridge balance, (2) a capacitance neutralizer, and (3) a current pump. Injecting constant current requires a very large resistance (Rs ³ 100 megohms) in series with the micropipette whose resistance must be much lower (Rme < 100 megohms). The voltage output (e.g. ×10) of the intracellular preamplifier is then connected to a signal conditioner, which allows the observation of the recorded extracellular/juxtacellular activity in satisfactory conditions (e.g. total gain: ×500; bandpass: 100 Hz–6 kHz). Cutting the high-pass

Fig. 3. (continued) The two inset photomicrographs are the surface (left) and depth (right) of the same horizontal 80-mm thick section containing the perikaryon that shows the marks (indicated by two arrowheads) left by the final position of the micropipette tip. Note the presence of tracer deposit close the cell membrane. (a5) Composite two-dimensional reconstruction of this tracer-filled TRN cell from superimposed camera lucida drawings (under oil immersion at 100×) from serial sections to show its patterns of dendritic and axonal ramifications. (b1) Recording of the electrically induced death of an identified TRN neuron just after having applied the juxtacellular filling procedure. (b2) Afterwards, the histochemical reaction product reveals a cell body with absent or damaged dendritic profiles. It has been observed at the expected stereotaxic locus and looks like a dead or badly injured cell. Abbreviations: A anterior; L lateral; P posterior; M median; VL ventrolateral nucleus of the thalamus; Po posterior complex of the thalamus. Adapted from Pinault (1).

50

Pinault

filter at ³100 Hz is a useful strategy to follow the state of the target neuron without balancing the bridge. Two approaches are available to establish the settings of the juxtacellular 200 ms current pulses: (1) In our laboratory, the settings are those that are produced directly by the preamplifier’s dye injection system, which is very easy to use because of the manual potentiometer for current intensity adjustment in the front panel of the preamplifier. (2) The other approach is to use a stimulator that is connected to the current-command on the rear panel, similar to those used during any intracellular recording experiment. This later approach offers the advantage that the experimenter can change both the duration and the frequency of the current pulse. 3.4. Anesthesia and Surgery

Juxtacellular labeling can be achieved either under nonanesthetized conditions (22, 23) or, most often, under anesthesia. The type of anesthesia may be important especially for labeling axons. For instance, from our own experience with Neurobiotin, the labeling of thalamic axons was better in quality and in quantity under urethane anesthesia than under neuroleptic analgesia (>50% vs. <50% of individual thalamic reticular nucleus axons stained; Pinault, unpublished observation). One likely interpretation of this difference might be due to the cellular activity/metabolism, which is modulated in a different way from one anesthetic condition to the other, and for a given type of nerve cell. Therefore, when the aim is to study the properties of the single axon architecture of a given type, it would be reasonable to test more than one anesthetic condition for this neuronal type. The choice should also take into consideration the state of the corresponding network when it comes to studying the electrophysiological properties of single nerve cells. The success rate of in vivo single-cell electrophysiology experiments depends primarily on mechanical stability at the submicron scale of the living brain. Most studies describe the same specific methodological approaches, which makes it possible to stabilize the living brain of any anesthetized animal held in a stereotaxic frame (e.g. 76–78): (1) Cisternal drainage (to prevent the development of edema); (2) large craniotomies and duratomies with the brain surface covered with a saline-based agar gel during the recording session; (3) a pneumothorax with suspended hips (to minimize cardiac and/or respiratory pulsations); (4) body temperature maintained at least 1 or 2°C below the physiological mean and elevation of the animal’s body (to minimize the vascular pressure differential between the body and the brain). The opening of the cisterna magna and the large craniotomy permit all of the cerebrospinal fluid to flow out, making the brain slump down progressively into the cranium in the course of experiments. These conditions reduce the precision

The Juxtacellular Recording-Labeling Technique

51

with which single neurons can be reached stereotaxically in a target region. Under these conditions, undesirable vascular and respiratory pulsations may still be present in single-cell recordings. Furthermore, standard large cranioduratomy may be regarded as improper opening because of bleeding or inadequate maintenance of the brain surface. Bleeding, drying, or edema are all potential causes of clogging or damage of the sharp tip of the recording glass micropipette. Regarding the intracranial pressure, two different sorts of problems should be considered: (1) excess pressure producing edema, which may be dependent upon level of anesthesia along with the damage caused to the brain, and (2) inadequate pressure producing collapse of the brain, which is presumably due to artificial drainage of the cerebrospinal fluid. These conditions might result in pulsatile artifacts (due to heart or respiratory pulsations) in single-cell recordings. Consequently, to avoid these inconveniences, we have devised a new micro-cranioduratomy (Fig. 4), which prevents significant outflow of the cerebrospinal fluid and makes it possible to keep the living brain totally free of unwanted pulsations (25). This allpurpose surgical procedure prevents brain collapse and edema, making it possible to enhance stereotaxic accuracy and to increase the success rate and quality of single-cell recordings/labelings (79, 80). 3.5. Micropipettes and Neuronal Markers

Micropipettes are prepared from 1.2 to 1.5 mm glass microfilament-containing capillaries with a standard puller. To avoid tissue damage and to get excellent recordings, it is better that the micropipettes have a long thin shank (Fig. 1a). They are filled with solution containing the marker molecules and, just before use, the tip is rubbed on the edge of a slide under microscopic observation until an external diameter of 0.4–1.5 micrometers is achieved. The micropipette is connected to an appropriate intracellular preamplifier with Ag/AgCl wires. Such a dye-filled micropipette has a DC resistance of 25–70 MW and has excellent recording characteristics, as well after as before the filling procedure, and passes tracer readily. How to fill the micropipette without bubble and with a high success rate (>90%)? In our laboratory, we first inject a small column (~1 mm) of pure water at the top of the capillary. Thanks to the microfilament and because of the lack of viscosity, the water quickly runs into the tip and the thin shank in less than 10–15 min. Then, it is easy to backfill the micropipette with the tracer-containing saline solution, using micropipette filler. It is important to note that these micropipettes are usable at least 4–5 h after the filling of the micropipette, the time that is required for the ions and tracer molecules to diffuse up to the micropipette tip.

52

Pinault

Fig. 4. Stabilizing micro-cranioduratomy for single-cell anatomo-electrophysiological exploration of living intact brain networks. In this experiment, the target is the thalamic reticular nucleus (TRN). (a) Dorsal view of the craniotomy made on the right side of the cranium, above the dorsal thalamus. The white circle indicates the minute hole (or micro-craniotomy; diameter <0.8 mm) for insertion of the micropipette under stereotaxic guidance. The dotted lines indicate the frontal plane, b, which contains the target recording site for the micropipette. (b) Caudal view of the corresponding frontal plane. It is redrawn from the atlas of Paxinos and Watson (24). Note that the craniotomy includes a large drilled bone area, which is characterized by a thin bone membrane in which a small micro-craniotomy is made to allow the insertion of the micropipette. Also note the minute incision in the meninges, which include the dura and pia maters. Small surgical sponges soaked with artificial cerebrospinal fluid are laid down on the cranium opening. (c1–c2) Dorsally taken macrophotographies showing a typical micro-cranioduratomy (CD, shown at higher magnification in C2). Note that the transparency of the bone membrane makes visible veins (V) and arteries (A). Also note the presence of an insignificant bone hemorrhage, which does not interfere with the quality of the brain surface. CPu caudate putamen; EP entopeduncular nucleus; HPC hippocampus; MD medial dorsal thalamic nucleus; VB ventrobasal complex of the thalamus; VL ventral lateral thalamic nucleus; VM ventral medial thalamic nucleus. Adapted from Pinault (25).

Neuronal markers that are soluble in a salt aqueous solution can be used with a moderate concentration of the salt because diffusion of hypertonic solutions from the electrode tip may damage the target neuron. Lucifer yellow has been used successfully

The Juxtacellular Recording-Labeling Technique

53

in the macaque retina (81). Three well-known biotin-containing compounds, which are transported both anterogradely and retrogradely (26–28), have been employed with success at 1.5% in 0.5 M of CH3COOK or NaCl: (1) the biotin–lysine complex (biocytin; Ne-(+)-Biotinyl-l-lysine; Sigma), (2) Neurobiotin (N-(2 aminoethyl) biotinamide hydrochloride; Vector Labs), and (3) Lysine-fixable biotin-dextran amin (BDA, Molecular Probes). One fundamental difference between the first two biotin-containing compounds and BDA is that this latter dye does not degrade over time (29). After a 10–15 min juxtacellular application of biocytin or Neurobiotin, excellent stainings have been obtained with a survival time ranging from a few min to 12 h, but the labeling intensity usually faded during the postinjection period with a total extinction at 48 h. As BDA is not handled by the metabolism of the injected neuron, it may be used to visualize long-axon neurons after a survival time superior to 24 h (28). Nevertheless, the percentage of tracer-filled thalamic neurons retrieved after a juxtacellular application is lower with biotin-dextran (<50%, from a total of 23 individual attempts) than with biocytin or Neurobiotin (>80%, from a total of 171 single-cell attempts). About half of the BDAfilled neurons had remarkable signs of degeneration after a survival time of 24 h and especially after long-lasting juxtacellular application (³20 min). In spite of this, we ended up revealing nice dendritic and axonal stainings of thalamocortical cells after a survival time of a few days (2, 14). Under urethane-anesthesia, the rate of the anterograde axonal transport, which could be determined for biocytin and Neurobiotin and which varies qualitatively and quantitatively from one cell to another, is approximately of 1.5–3 mm/h. In some instances, we have observed axonal labeling at a distance more than 6–8 mm after a survival time of less than 1 h (under urethane anesthesia or neuroleptic analgesia; Pinault, unpublished observation). The labeling intensity is roughly proportional to the amount of juxtacellularly applied current (2). This rule should depend on many factors such as type of cell, anesthesia, survival time, etc. In our laboratory, surprisingly, we have sometimes observed, after a survival time less than 1 h, better labeling following a 5-min than a 10-min driving/filling procedure. We think that the labeling quality also depends on the physiological and/or metabolic state of the recorded neuron. 3.6. Histological Revelation

Following the completion of the recordings/labelings and a survival time of a few min to a few hours, the rats are euthanized (without regaining consciousness) by intravenous injection of a lethal dose of pentobarbital and then transcardially perfused with 100 ml of 0.9% saline followed by 500 ml of 4% paraformaldehyde (PFA) solution in 10 mM phosphate buffered saline (PBS)

54

Pinault

and 0.1% glutaraldehyde (pH = 7.4). Frontal, saggital, or horizontal brain sections were cut at 40–100 micrometers with a vibrating microtome and serially collected in PBS. They were then thoroughly washed in PBS before being incubated with avidin–biotin–peroxidase complex (ABC; Vector Labs) overnight. The injected tracer, bound to this enzymatic complex, was revealed with the appropriate substrate and chromogen, the hydrogen peroxide and the 3,3¢-diaminobenzidine tetrahydrochloride, respectively, according to the method described elsewhere (26) and nickel intensification (30). Nowadays, we use the DAB kit (Vector Labs). This approach permits a permanent visualization of the juxtacellularly-filled neuron by the enzymatic reaction product (Fig. 3a4).

4. Comparison of Juxtacellular Staining with Neurobiotin and BDA

The low success rate of single-cell labeling obtained with BDA was hypothesized to be due to a neurotoxic process. To test this hypothesis, we attempted to compare the stainings of thalamic neurons following juxtacellular application of BDA3000 (3,000 MW) or Neurobiotin (322.85 MW) under the same experimental conditions. Dual single-cell recordings were conducted of neurons in the thalamic reticular nucleus of anesthetized adult rats (N = 23) with two glass micropipettes, one containing Neurobiotin (1.5%) and the other BDA3000 (1.5%) in 0.5 M K-Acetate. At the end of the recording session, each neuron of the last recorded pair was subjected to the juxtacellular filling procedure (4–10 min; 0.5–8 nA). After a survival period of less than 1 h, the animals received an overdose of anesthesia and were transcardially perfused with a fixative for processing the brain tissue with standard histological techniques. Two lots of BDA (6591-2 and 65A1-1) and one lot (K0225) of Neurobiotin were tested. The somatodendritic arborization was observed using a light microscope. These experiments yielded the following results (unpublished data): With the first lot of BDA (6591-2), the success rate of labeled neurons, whatever the labeling quality, was 37% (three retrieved neurons from eight juxtacellular filling attempts). Also without taking into consideration the labeling quality, the success rate was 92% (12/13) with the second lot of BDA (65A1-1) and of 100% (13/13) with Neurobiotin. On the contrary, the number of poorly stained neurons was much higher with BDA (in 5 out of the 12 retrieved cells) than with Neurobiotin (only in 1 out of 13), as ascertained by obvious signs of neuronal degeneration (Fig. 5a–c). Photomicrography D shows a thalamic neuron that was nicely marked with BDA. Its labeling quality was comparable to that of nearly all of the thalamic neurons that were well stained

The Juxtacellular Recording-Labeling Technique

55

Fig. 5. Biotin dextran amin (BDA) is neurotoxic at the single-cell level. (a–c) Damaged thalamic reticular nucleus neurons (TRN) following juxtacellular application of BDA. (d) Nondamaged TRN cell following juxtacellular application of BDA. S soma; D dendrite.

with Neurobiotin (N = 12) under the same experimental conditions. These results demonstrate that BDA3000 is toxic for rat living neurons in a short time period (<1 h) following a juxtacellular application. Such toxicity has rarely been observed with Neurobiotin. In conclusion, after a survival time of >1 h, the probability to retrieve satisfactory neuronal labeling after a juxtacellular application of a few min of BDA is significantly lower than that obtained with Neurobiotin (54% vs. 92%).

5. Tips for the Application of the Juxtacellular Filling Procedure

We offer at least nine tips to increase the success rate of juxtacellularly labeling the recorded neuron: 1. The whole animal preparation should be stable at the submicron scale, as mentioned earlier. The stability can be increased with the use of a new micro-cranioduratomy (25). 2. The single-cell extracellular recording should be of excellent quality. Positive or biphasic action potentials of a few millivolts are required to get the target neuron driven by the

56

Pinault

juxtacellular current pulses. The amplitude of the action potentials usually increases gradually as the tip is brought progressively to the target neuron, which permits one to decrease the intensity of the juxtacellular current. If the action potentials of the target neurons are only negative going (usually of low amplitude (<1 mV)), then the neuron will never be driven by the juxtacellular current pulses. Therefore, the proper conditions for juxtacellular labeling will never be reached, and of course, this neuron will never be stained. This means that the micropipette tip remains very close to the cell body, even when moving it up or down. Thus, to apply the juxtacellular filling procedure under such conditions would be in vain. In this case, it is preferable that the experimenter ends up finding another cell. Nevertheless, whatever the extracellular conditions, many electrophysiological characteristics can be recorded. On the basis of the well-known electrophysiological features and/or following orthodromic and/ or antidromic activation, the target neuron can be identified during the recording session. 3. Starting the juxtacellular filling procedure is often a gradual process. Indeed, in most instances, the intensity of the applied current pulses is progressively increased so as to entrain the firing of the neuron being recorded. If one finds that they are unable to modulate the firing of the target cell with a reasonable amount of current (~10 nA), then the micropipette tip should be brought physically closer to the cell with steps of 1 micrometers and the juxtacellular current pulses reapplied with increasing intensity. If the microelectrode tip is close enough to drive the cell, then a good first index of a successful juxtacellular condition is an increase in the amplitude of the background noise during the current application (Fig. 3a2, asterisk). At that moment, the current initiates the juxtacellular condition. In contrast, sometimes the juxtacellular condition is already reached before starting the filling procedure, as ascertained by driving the firing at the start of the current pulse application with intensity of a few nA (sometimes less than 1 nA). 4. When the target neuron displays a firing pattern that is the signature of a deeply hyperpolarized state (e.g. rhythmic high frequency burst firing), it is usually difficult to get the neuron entrained with juxtacellular nanocurrent. In this case, increasing the amount of juxtacellular current applied carries the risk of damaging the cell definitively, especially when it spontaneously and quickly gets back to a more depolarized state. To drive the recorded neuron by the juxtacellular filling procedure with a reasonable amount of current (0.5–8 nA) under such circumstances, we advise the experimenter to find a way to provoke either brain arousal or activation of the corresponding

The Juxtacellular Recording-Labeling Technique

57

network (e.g. stimulation of the receptive field, gently “tickling” the rat, pinching its tail, etc.). 5. Electrophysiological monitoring allows one to adjust the amount of current in a continuous manner so as to not injure the recorded neuron. If the neuron races out of control, current application should be stopped immediately till recovery. Sometimes, applying a holding hyperpolarizing current makes recovery easier. In some instances, on the one hand, the cell is too severely injured and is subsequently going to die (no recovery, increased AP duration, and decreased AP amplitude). Therefore, it is better to fix the brain as soon as possible (a few tens of seconds is enough to partially label the recorded neuron). On the other hand, when the cell recovers, the tickling procedure can be resumed. 6. In some cases, the amplitude of the target neuron’s action potentials quickly decreases during current pulse application, which is characteristic of a depolarization block. Under such circumstances, we either decrease the current intensity or simultaneously apply a holding hyperpolarizing current of a few nA. 7. It is an asset to record in addition the DC shift of the membrane potential, which should be done with Ag/AgCl wires that link the recording micropipette to the preamplifier. Of course, the membrane potential is at the level “0 mV” during extracellular recording. During the juxtacellular filling procedure, a shift of a few millivolts can be recorded, which is a strong argument in favor of the hypothesis that the tracer is internalized through a micro-electroporation of the membrane (2). When the shift becomes significantly much higher (£10 mV), the neuron is or close to being injured. Under such conditions, it is better to stop current pulse application and to apply a holding hyperpolarizing current in an attempt to allow the cell to recover. At this stage, experienced electrophysiologists would try with some success to switch into the intracellular mode. That is why we chose to work with CH3COOK-filled micropipettes. 8. At the end of the tickling procedure, the micropipette should be withdrawn (Fig. 1c4) with the greatest care. Because the tip of the micropipette can be tied up with the membrane, it should be removed up very gently and slowly, with steps of 1–2 micrometers. If the amplitude of the action potentials increases during the physical withdrawal, it is advised to wait (sometimes a couple of min) to let the cell recover between the ascent steps. In our laboratory, we sometimes use a holding hyperpolarizing current to decrease occasional subsequent injury discharge and to help the cell to recover from the unseal procedure.

58

Pinault

Fig. 6. Posterior (a), medial (b) and dorsal (c) views of two 3D-reconstructed ventral lateral (VL)-projecting thalamic reticular nucleus (TRN) neurons of the same sector. These two neurons were successively, juxtacellularly stained with biocytin along the same micropipette trajectory. Their somatodendritic and axonal fields are shown in the corresponding stereotaxic frontal (a) and saggital (b) plates. Note that their dendritic domains had no significant overlap, and that their axons displayed distinct arbors in immediate neighboring zones in the VL nucleus. Abbreviations: A anterior; D dorsal; L lateral; P posterior. Adapted from Pinault and Deschênes (48).

The Juxtacellular Recording-Labeling Technique

59

9. To avoid false positive stainings, which might be prejudicial to the interpretation of the histochemical material, the experimenter must better be aware that a juxtacellular application as short as a few seconds (e.g. when exciting a silent neuron for a very short while with positive current pulses) can be sufficient to label a cell. Therefore, each attempt and the corresponding stereotaxic coordinates must always be noted. Thereby, nerve cells that have individually been tickled by the current pulses along the same descent can be retrieved with their own axonal and/or dendritic ramifications (Fig. 6), and their morphology can be correlated with their electrophysiological features.

6. Widespread Applicability and Versatility

The widespread applicability and versatility of the juxtacellular technique in whole animal preparation were first demonstrated during systematic and pilot studies (1, 2). As predicted, it can be combined at will with a broad spectrum of techniques, ranging from molecular to behavior. Since then, many comprehensive studies from all over the world have definitely established its general applicability and usefulness in all fields of neuroscience.

6.1. “Full” Success Rate

The juxtacellular technique is less invasive and less traumatizing than the intracellular technique and allows one to electrophysiologically control the four steps of the filling procedure (Fig. 1c1–c4). Thereby, the experimenter can be sure to maintain the recorded neuron alive and thus to retrieve it with a success rate superior to 85%, for not to say 100%, of the filling attempts, as approved by many teams. The neurons that are not labeled may have undergone a flawed juxtacellular application (ineffective micropuncture), a local hemorrhage, or any other possible pre or postmortem complication (e.g. osmotic or metabolic knock out occurring subsequent to an overload of tracer). Indeed, when developing the juxtacellular technique, we could observe only a residual tracer deposit in the extracellular space at the intended stereotaxic target (Fig. 3a4). Furthermore, we could visualize hemorrhages along the course of micropipette tracks and even sometimes at stereotaxic targets. Such vascular accidents could have been produced through ischemia neuronal death, since we have observed no labeled neuron at hemorrhage sites even at those induced by “extracellular” applications.

6.2. Everywhere in the Brain

The juxtacellular recording/labeling technique can be used in many species, including in mice (31, 32) and in non-human

60

Pinault

primates (82, 83). It has been used to study anatomo-functional properties of single nerve cells in nearly all brain regions, including the basal forebrain (20, 84, 85), the basal ganglia (33–37), the brainstem (38), the cerebellum (32, 39), the zona incerta, (40), the neocortex (41, 42), the pretectum (86), the rostral medulla (43), the mesencephalon (87), the hippocampus (44), the septum (45), the thalamus (17, 88, 89) in the retina (81), in the hypothalamus (31), and in the pons (46). 6.3. Exploring the Smallest Nerve Cells

Since its development, the juxtacellular technique makes it possible to discover the detailed structure of brain networks. As a matter of fact, comprehensive studies of the anatomo-functional properties of the smallest neurons, the interneurons, or localcircuit cells have been achieved (39, 44). Also, new organization principles of neuronal connections have been discovered with the juxtacellular technique in the thalamus (47–51) and in the basal ganglia (34, 35). For instance, a mapping and a 3D graphic and morphometric analysis of the axonal projection of individual thalamic reticular nucleus axons was achieved, allowing us to model the anatomical relations between the dorsal thalamus and the thalamic reticular nucleus (48). Using the single-cell and multiunit juxtacellular techniques, we also demonstrated that reciprocal connections between the dorsal thalamus and the related thalamic reticular nucleus do not apply at the single-cell level (47). Instead, we provided the first anatomical demonstration of a principle of lateral inhibition mechanism (Fig. 7).

6.4. Juxta-Somatic or Juxta-Dendritic Labeling

The finesse of the juxtacellular labeling procedure allows the filling of the neuron being recorded not only from its perikaryon but also from one of its dendrites. For instance, both electrophysiological and histochemical evidence that the neuron being recorded could be filled following juxta-somatic or juxta-dendritic iontophoresis of tracer molecules were obtained when labeling pyramidal neurons of the neocortex or Purkinje cells along their somatodendritic domain (2). In the cerebellar cortex, depth measurements and the continuous display of both somatic and dendritic spikes were used to ascertain the somatic or dendritic location of the micropipette tip (Fig. 8). These experiments confirmed the differences observed in vitro between the electric properties of the Purkinje cell’s dendrites and soma (52–54). For some purpose, spontaneous or evoked, fast and/or larger action

Fig. 7. (continued) terminal axonal buttons of the retrogradely labeled TC neuron (in blue) clustered to fit in with the spatial arrangement of dendritic profiles and of the soma of the TRN cell (in red). Scale bar in (a) is valid for (b). Adapted from Pinault and Deschênes (47).

The Juxtacellular Recording-Labeling Technique

61

Fig. 7. Extraordinary simultaneous anterograde and retrograde juxtacellular labelings. In this experiment, Neurobiotin was applied juxtacellularly on an electrophysiologically identified thalamic reticular nucleus (TRN) cell. The thalamocortical (TC, in black) and TRN (in red) cells are marked from a simultaneous juxta-axonal and juxta-somatodendritic biocytin iontophoresis into the TRN with the same micropipette, respectively. (a, b) The 3D-reconstruction of this neuronal pair is represented from a dorsal and posterior view, respectively. These two nerve cells are shown in the most appropriate horizontal (bregma -5.32 mm) and frontal (bregma -1.80 mm) stereotaxic planes (adapted from those of the atlas of Paxinos and Watson (24)). (c) Part of the drawing of another pair, which is presented at higher magnification, shows that some

62

Pinault

Fig. 8. Juxtasomatic (a) or juxtadendritic (b) filling of electrophysiologically-identified purkinje neurons with neurobiotin. (a) The activity recorded at the cell body of a Purkinje cell is characterized by high fast-rising spikes sometimes followed by complex dendritic spikes of smaller amplitude (see traces (a) and (b); lower trace = current monitor). These somatodendritic spikes are usually followed by a short-lasting inhibition. Photomicrograph: after a juxta-somatic iontophoresis of 11 min and a survival period of 30 min, the histochemical reaction reveals only one tracer-filled Purkinje cell (the injection site indicated by an arrowhead). (b1, b2) A Purkinje cell labeled from a proximal dendrite where both dendritic and somatic spikes can be recorded. At the beginning of the injection (b1) current pulses of 3 nA (lower trace = current monitor) trigger rhythmic dendritic activities; at the end of the 10-min injection period, the recording is dominated by the spontaneous occurrence of small amplitude somatic spikes (b2; see also the expanded traces a–d corresponding to the periods indicated in (b1) and (b2)), suggesting that the recording site is located just above the Purkinje cell layer. The dendritic spikes are usually followed by a short-lasting inhibition of the somatic discharges. Photomicrograph: After a survival period of 90 min, the histochemical reaction product discloses one Neurobiotin-filled Purkinje neuron with a clear-cut decreasing gradient from the injection site, which is located at a proximal dendrite (indicated by an arrowhead) and from which a passing fiber has been filled simultaneously. The scale of the photomicrographs is given by their width: (a) = 130 mm; (b) = 200 mm. Adapted from Pinault (2).

potentials could be recorded juxtacellularly from fine neuronal processes, such as axons or dendrites, providing conspicuous information relative to membrane properties of the target neuron. This makes the juxtacellular technique an invaluable complement to intracellular studies.

The Juxtacellular Recording-Labeling Technique

63

6.5. Ultrastructure

As a given neuron can be stained faintly after short-lasting juxtacellular iontophoresis of marker molecules and/or after long survival periods, one can examine the ultrastructure and synaptic organization of identified neuronal elements using the electron microscope to make additional immunolabelings. The ultrastructure of juxtacellularly labeled neurons is well preserved (55). Indeed, when correlating light and electron microscopic analyses, we observed that the finest axonal and dendritic ramifications were stained without significant loss of ultrastructural features. Thereby, ultrastructural properties of synaptic connectivity were discovered in diverse structures (33, 37, 40, 51). For instance, we could reliably carry out correlated light and electron microscopic analyses of the finest, identified tracer-filled axonal and dendritic elements of physiologically-characterized thalamic reticular nucleus neurons (55). We could observe that a minority (about 10%) of thalamic reticular nucleus neurons is endowed with poorly ramifying, varicose, intrinsic axon collaterals (Fig. 9) that could be further demonstrated as being postsynaptic structures contacted by numerous GABA-negative terminals (55). Similarly, we could provide evidence that the so-called “axon-like processes” are also postsynaptic dendrites.

6.6. Molecular Phenotype

Biotin-containing compounds are tracers that have a high binding affinity for avidin. They can thus be visualized with avidin conjugated to various, solid or fluorescent, molecular probes (26). This makes it possible to combine neuronal labeling with immunohistochemical techniques to reveal immunoreactive proteins of the recorded neuron (33, 43, 44). Thereby, the molecular content of the recorded neuron can be correlated with its anatomo-functional properties (56). Furthermore, single-cell gene expression of juxtacellularly labeled neurons can now be studied, since Guyenet’s team has demonstrated that the juxtacellular technique can be combined with in situ hybridization histochemistry (57).

6.7. Pharmacology

Electrophysiological, anatomical, immunohistochemical, and pharmacological properties of single nerve cells can be studied in vivo in a given experimental paradigm (58). Drug application can be performed through multibarreled micropipettes, which are attached to the juxtacellular recording/labeling micropipette (Fig. 10).

6.8. A Single Nerve Cell Can Drive the Behavior

Juxtastimulation of individual neurons in the somatosensory cortex of conditioned awaked rats can influence behavioral responses (23). The rats were first trained to respond to microstimulation of the corresponding cortical region with low current intensities. Afterwards, they responded significantly more often during single-cell nano-juxtastimulation, which gave rise to trains of action

64

Pinault

Fig. 9. A biocytin-filled thalamic reticular nucleus cell with two thin varicose processes emerging from a common dendritic shaft. One is the axon that projects to the thalamus, whereas the other is an “axon-like” profile that gives rise to a few intrinsic ramifications. (a, b) Caudal view of the somatodendritic complex and of the two axon-like processes (shown separately in b). The framed areas (a–d) are shown at higher magnification in the corresponding photomicrographs. The common source of the two thin profiles is indicated by an arrow in (a) and an arrowhead in (b). The axon displays varicosities of different sizes before penetrating the thalamus (arrowheads in b). In (c), the arrows indicate the direction of the two processes, upward for the axon and downward for the intrinsic axon-like process. The initial portion of these two thin processes is similar, but it is quite different from that of a dendrite (b, c). On the contrary, the distal portion of the intrinsic “axon-like” profile and distal dendrites are indistinguishable (d). Scale bar, 10 mm in d (valid for a–c). Abbreviations: D dorsal; L lateral. Adapted from Pinault et al. (55).

The Juxtacellular Recording-Labeling Technique

65

Fig. 10. Diagram of seven-barreled compound glass micropipettes used for extracellular recording, juxtacellular labeling, and iontophoresis. Bending the tip of a seven-barreled glass micropipette and gluing it to a single-barreled recording micropipette fabricates the juxtacellular micropipette. This was reinforced with epoxy resin and metal bars. Tips are separated by 0–10 micrometers. The multibarreled electrodes are broken back to between 7 and 15 mm and the single-barreled to around 1 mm. Scale bar, 10 mm. Adapted from Jones et al. (58).

potentials. However, the cellular and network mechanisms underlying the link between single-cell juxtastimulation and behavioral change remain to be determined. 6.9. Spatio-Temporal and Anatomical Properties of Cellular Interactions, Oscillations, and Synchronizations

The juxtacellular technique can be combined with many other electrophysiological techniques, including electroencephalography, extracellular field potential and multiunit recordings, and intracerebral electrical stimulation. It makes it possible to correlate the anatomo-functional properties of single nerve cells with the corresponding cellular and network activities. For instance, by performing our micro-cranioduratomy (25), we could combine paired juxtacellular recording/labeling with electroencephalographic recordings in corticothalamocortical systems of anesthetized rats (Fig. 11). Then after having combined the juxtacellular findings with intracellular data, we could demonstrate that, in a genetic model of absence epilepsy, layer VI corticothalamic neurons of the somatosensory cortex initiate epileptic discharges (Fig. 12). Therefore, the juxtacellular technique is very useful when it comes to deciphering the molecular, cellular, and network mechanisms of brain state, physiological and pathological oscillations.

66

Pinault

Fig. 11. Multisite single-cell juxtacellular recording/labeling in whole animal preparation. This experiment is designed to study the spatiotemporal dynamics of cellular interactions, oscillations, and synchronizations in the corticothalamic (CT) system (circuit shown in Fig. 12). (a1, a2) Juxtacellular staining of simultaneously recorded CT and thalamocortical (TC) neurons, respectively. (a3) Partial reconstruction of the CT neuron of A1. (b) Schema illustrating the location of the recording micropipettes in the somatosensory thalamus (ventral posteromedial thalamic nucleus and the corresponding thalamic reticular nucleus (TRN) sector) and of the Ag/AgCl ECoG electrode in the related frontoparietal cortex. The stereotaxic plates (in millimeter from bregma) are drawn from the Paxinos and Watson’s atlas (24). Note that minimal craniotomies are made for the insertion of the micropipettes, and that the frontoparietal cortex is not directly injured by the corresponding

The Juxtacellular Recording-Labeling Technique

67

Fig. 12. Layer VI corticothalamic (CT) neurons of the somatosensory cortex initiate epileptic discharges in a genetic model of absence epilepsy. This schematic diagram shows likely spatiotemporal cellular interactions within the CT system occurring during natural medium-voltage 5–9 Hz oscillations in Genetic Absence Epilepsy Rats from Strasbourg. At least two types of thalamocortical (TC) neurons coexist, of which one (TC2) is endowed with a presumed H-current. Note that thalamic, relay and reticular, discharges occur in synchronous and phase-locked manners during the electro-encephalographic (ECoG) epileptic spike-wave (SW) complex. From top to bottom: (a) SW complex (ECoG), an extracellular CT discharge, an intracellular TRN discharge, and two typical intracellular TC discharges. The second TC cell (TC2) exhibits a presumed H-current, coinciding with an EPSP barrage. The ramp-shaped depolarization, which includes a presumed Ih, can trigger a low-threshold Ca2+ spike (LTS). In TRN cells, the EPSP barrage can trigger voltage-dependent components (V-components). (Center) Schematic drawing of the anatomical relationships between the three main elements that make up the CT system. (b) Means and standard deviations of the time relationship between the CT, TC, and TRN action potential discharges and the SW complex. Adapted from Pinault, J Physiol (2003).

Fig. 11. (continued) craniotomy. (c1, c2) Juxtacellular marking of simultaneously recorded TC and TRN neurons, respectively. (d1) 3D reconstructions of the two thalamic neurons recorded in (d2). Note that they are not connected. (d2) Simultaneous TC and TRN extracellular activities associated with a spontaneous electrocorticographic (ECoG) epileptic discharge. (d3) Time relationship of the TC and TRN discharges with the related ECoG SW complex (superimposition of five successive SW complex-related cellular events). (d4) Cross-correlogram of the corresponding TC and TRN discharges. (d5) Dot raster displays of the simultaneously recorded TC and TRN neurons (a few hundreds of successive ECoG/cellular events) and the corresponding peri-event time histograms. Abbreviations: CL central lateral; CM central medial; CPu caudate putamen; ep entopedoncular; ic internal capsule; LD lateral dorsal; MD medial dorsal; Po posterior thalamic nuclear group; VL ventral lateral; VM ventral medial; VPl ventral posterolateral; VPm ventral posteromedial; WM white matter; ZI zona incerta. Adapted from Pinault, J Physiol (2003).

68

Pinault

7. Disadvantages and Pitfalls The juxtacellular recording/labeling technique is, like many other techniques, not perfect. However, we will see that some of its disadvantages, when well identified, can turn out to be advantages. 7.1. Physical and Electrical Stability

The major limitation encountered during the juxtacellular filling procedure may be the mechanical stability of the electrode tip-cell membrane juxtaposition because any vibration (due to respiration, vascular pulsations, and/or environmental mechanical noise) and/or too much iontophoretic current can definitively injure the neuron being recorded (2). Therefore, the juxtacellular contact of the micropipette tip needs to be as stable as possible on a submicron scale.

7.2. Axonal Labeling

There is no compelling evidence to indicate that a single nerve cell that had been well entrained juxtacellularly will have its axon completely filled throughout all of its terminations. Complete dendritic and axonal stainings have been obtained at least with local-circuit neurons (2, 44, 50). An overall picture of a given type of long-axon neurons can be obtained after a series of many systematic neuronal fills. For instance, we could ascertain, in urethane-anesthetized rats, the completeness of the axonal labeling in at least 60% of thalamic reticular nucleus neurons (N = 100) that had been subjected to a juxtacellular application of biocytin or Neurobiotin for about 10–15 min. It is worth mentioning again that achieving complete fills depends on several factors (species, anesthetic, tracer, amount of current, and survival time) that should be well chosen and adjusted when it comes to studying a given neuronal type. When heavily myelinated, the axon trunk of a labeled neuron often appears as an interrupted line (Fig. 3a4). This drawback is not specific to the juxtacellular technique since this pattern is also observed in intracellularly stained neurons (17). The dashed appearance of the axon is due to the weak penetration of the avidin–biotin–peroxidase complex through the heavy myelinated segments. Indeed, nodes of Ranvier, branch points, fine collaterals, and terminations were always darkly stained as any ramification of the somatodendritic complex. Furthermore, electron microscope observation has shown that the labeling intensity of the histochemical reaction product rapidly and progressively decreases from the nodes of Ranvier and that the nonstained axon segments were morphologically intact (55). Therefore, at the single-cell level, this drawback does not cause any major problem to trace the labeled neuron and further provides morphometric data such as the number of nodes of Ranvier.

The Juxtacellular Recording-Labeling Technique

69

7.3. Multiunit Labeling

As stated earlier, a proper juxtacellular application of tracer molecules should, as expected, not contaminate nerve elements nearby the neuron that has been isolated for individual labeling. However, in some instances, one or a couple of neurons can simultaneously be stained with the recorded/labeled neuron, which sometimes renders it difficult (especially when they are very close to each other) to find the exact cell that had been recorded. This happened at a maximum of 7% of the successful labelings of thalamic neurons (2). In rare instances, the activity of two neurons had been juxtacellularly recorded and entrained, and then two neurons were retrieved after the histological process. When a difference in the quality of labeling is observed, it is reasonable to assume that the darker cell was the one that had been more entrained by the juxtacellular current. In spite of this drawback, simultaneous double anterograde labelings may be very useful to demonstrate, for instance, divergent axonal projections from a given point in the brain (48). Passing fibers and/or glial cells are rather rarely marked; on close inspection, they are usually distinguishable from all the labeled elements of the target neuron. On the contrary, additional retrograde filling of a presumed presynaptic neuron to the cell being recorded can rarely occur (4%, (2)). Such a single-cell retrograde filling apparently resulted from a juxta-axonal application onto one of its fine axonal profiles, in which part or a button was sandwiched between the micropipette tip and the recorded target neuron (Fig. 7). These simultaneous putative presynaptic and postsynaptic stainings undoubtedly offer an important means when deciphering neural networks. Thereby, we could demonstrate the principle of a mechanism of lateral inhibition in the dorsal thalamus (48). This principle has further been supported by several experimental functional data (59).

7.4. Miscellaneous

Other limitations that are not specific to the juxtacellular technique itself may be due to the biological or chemical material employed (see section about histochemical markers).

8. Perspectives The widespread applicability and versatility of the juxtacellular technique should yield a tremendous improvement of single-cell recording/labeling experiments, offering singular and miscellaneous advantages. First, employing it in whole animal preparations represents the unique advantage for the structural and morphometric analyses of the functional architecture of “entire” single living nerve cells with, for instance, a computer aided threedimensional reconstruction (e.g. Neurolucida, MBF Bioscience – MicroBrightField, Inc.) from successive serial sections (48, 60);

70

Pinault

once digitized, the tracer-filled neuron can be viewed from any angle (Figs. 6 and 7). One may seek to stain individually a few identified neurons in a single preparation and then to combine them to obtain a topological map of the network under study. Simultaneous intracellular recordings of presynaptic and postsynaptic neurons, such as those performed in neocortical (61) or hippocampal (90, 91) slices and in vivo in brainstem (62), allow the correlation of morphological and physiological properties of identified single-axon, excitatory or inhibitory synaptic connections; additionally, pharmacological properties of these synapses can be correlated with their ultrastructure as well as with the morpho-functional data of the corresponding neurons (61). The combination of intracellular recording and dye injection can thereby yield more precise information about cell–cell communication within local circuits than can be achieved by anatomical and physiological studies alone. Nevertheless, the probability of recording a synaptically connected cell pair remains relatively low (63). Juxtacellular modulation of the recorded neuron’s firing with a tracer-filled sharp micropipette provides a reliable presynaptic stimulation, which can be further recovered histologically, to investigate single-axon postsynaptic effects, for instance with a combined simultaneous intracellular recording. One great advantage of applying current pulses onto the recorded neuron’s membrane is the ability to drive its firing at will. Indeed, previous electrophysiological studies performed in anaesthetized (e.g. (64–67)) or awake animals (e.g. (68)) and in brain slices (69) have shown that juxtacellular currents can excite or inhibit the firing of the neuron under study. Thereby, the juxtacellular approach ought to facilitate the double-recording/marking of connected presynaptic and postsynaptic nerve cells. Thus, with such a powerful and reliable single-cell labeling tool, one may expect the following: (1) to examine the overall dendritic and axonal architectures of functionally-identified nerve cells, (2) to perform structural and morphometric analyses of the tracer-filled neurons, (3) to provide a catalogue of all the neurons that make up a given network, (4) to examine identified neuronal elements in the electron microscope, (5) to correlate physiological, pharmacological, and morphological properties of identified single-axon, excitatory or inhibitory, synaptic connections, (6) to have subsequent morphological identification of neurons whose pharmacological properties have been tested, (7) to correlate physiological and morphological characteristics of task-related neurons, (8) to probe the spatio-temporal properties of cellular interactions, oscillations, and synchronizations in living, large scale as well as local neural networks, (9) to examine plastic morphological processes, (10) to identify specific molecules or proteins of tracer-filled neurons, (11) to study direct cell–cell communication via gap junctions, for instance with Lucifer Yellow

The Juxtacellular Recording-Labeling Technique

71

(9) or Neurobiotin (70), (12) to study the juxtacellular iontophoresis of substance onto the recorded neuron (13, 92–94), and (13) to inject macromolecules (e.g. proteins, DNA, RNA, antisense RNA/oligonucleotides) for studying protein synthesis, lineage tracing, or their functional properties.

9. Conclusion Nowadays, it is rational to say that modern neuroscientists should succeed in correlating the dendritic and axonal architectures of single nerve cells with their biophysical, molecular, synaptic, and genetic properties, which are specifically associated with a given behavior and/or a physiological or pathological brain state. “Tickling” under continuous electrophysiological monitoring the recorded neuron with iontophoretic nanocurrents delivered through the tip of a marker-containing sharp micropipette allows with a high success rate the visualization of the neuron’s architecture after having collected its extracellular functional properties. This novel single-cell labeling procedure preserves the ultrastructure of the studied neuron and can further be combined with immunohistochemical and molecular methods. Combining the juxtacellular recording/staining method with emergent molecular techniques (e.g. DNA microarrays) to determine the cellular phenotypes and genotypes of neural networks is, nowadays, no more a dream. The juxtacellular technique is expected to be very useful for injecting macromolecules (e.g. proteins, DNA, RNA, anti-sense RNA/oligonucleotides), for studying protein synthesis, lineage tracing, or their functional properties. Therefore, the juxtacellular technique represents a powerful technique for delineating the fundamental physiological and pathological, structural and molecular bases of cell–cell communication, helping to gain profound insights into how nervous systems are generated, and how they integrate and distribute nerve information.

Acknowledgments I thank Professor Martin Deschênes (Laval University, Quebec, Canada) with whom I had been learning the anatomofunctional approach, using especially intracellular recordings, of neuronal networks in living intact brain. Without learning the intracellular technique I would not have developed the juxtacellular technique. The present study is supported by the French Institute of Health and Medical Research (INSERM), by the Université of Louis Pasteur and The Université de Strasbourg, Faculté de Médicine, Strasbourg. I also thank Thomas Zheng for critical reading of the manuscript.

72

Pinault

References 1. Pinault D (1994) Golgi-like labeling of a single neuron recorded extracellularly. Neurosci Lett 170:255–260. 2. Pinault D (1996) A novel single-cell staining procedure performed in vivo under electrophysiological control: morpho-functional features of juxtacellularly labeled thalamic cells and other central neurons with biocytin or Neurobiotin. J Neurosci Methods 65: 113–136. 3. Ramón Y, Cajal S (1911) Histologie du système nerveux de I’homme et des vertébrés. Paris: Maloine 1. 4. Nauta WJH, Ebbesson SOE (1970) National Institute of Neurological Diseases and Stroke., Puerto Rico. University. Contemporary research methods in neuroanatomy. New York: Springer. viii, 386 p. 5. Björklund A, Hökfelt T (1983) Handbook of chemical neuroanatomy. Amsterdam; New York: Elsevier. v. <1, 4, 6–16, 18, 20 > p. 6. Heimer L, Záborszky L (1989) Neuroanatomical tract-tracing methods, 2: recent progress. New York: Plenum. xv, 408 p. 7. Bolam JP, editor. (1993) Experimental neuroanatomy: a practical approach. New York: Oxford University Press. 8. Kater SB, Nicholson C (1973) Intracellular staining in neurobiology. New York, Springer. xii, 332 p. 9. Stewart WW (1978) Functional connections between cells as revealed by dye-coupling with a highly fluorescent naphthalimide tracer. Cell 14:741–759. 10. Wilson P, Kitchener PD, Snow PJ (1996) Intraaxonal injection of neurobiotin reveals the long-ranging projections of A beta-hair follicle afferent fibers to the cat dorsal horn. J Neurophysiol 76:242–254. 11. Uhlrich DJ, Cucchiaro JB, Humphrey AL, Sherman SM (1991) Morphology and axonal projection patterns of individual neurons in the cat perigeniculate nucleus. J Neurophysiol 65:1528–1541. 12. Lynch G, Deadwyler SA, Gall C (1974) Labeling of central nervous system neurons with extracellular recording microelectrodes. Brain Res 66:337–341. 1 3. Benson TE, Voigt HF (1995) Neuron labeling by extracellular delivery of horseradish peroxidase in vivo: a method for studying the local circuitry of projection and interneurons at physiologically characterized sites. J Neurosci Methods 57:81–91.

14. Deschenes M, Bourassa J, Pinault D (1994) Corticothalamic projections from layer V cells in rat are collaterals of long-range corticofugal axons. Brain Res 664:215–219. 15. Bourassa J, Pinault D, Deschenes M (1995) Corticothalamic projections from the cortical barrel field to the somatosensory thalamus in rats: a single-fibre study using biocytin as an anterograde tracer. Eur J Neurosci 7:19–30. 16. Pinault D, Bourassa J, Deschenes M (1995a) The axonal arborization of single thalamic reticular neurons in the somatosensory thalamus of the rat. Eur J Neurosci 7:31–40. 17. Pinault D, Bourassa J, Deschenes M (1995b) Thalamic reticular input to the rat visual thalamus: a single fiber study using biocytin as an anterograde tracer. Brain Res 670:147–152. 18. Murphy PC, Sillito AM (1996) Functional morphology of the feedback pathway from area 17 of the cat visual cortex to the lateral geniculate nucleus. J Neurosci 16: 1180–1192. 19. Chevalier G, Deniau JM, Menetrey A (1992) Evidence that biocytin is taken up by axons. Neurosci Lett 140:197–199. 20. Kirouac GJ, Pittman QJ (1999) Identification of barosensitive neurons in the mediobasal forebrain using juxtacellular labeling. Am J Physiol 276:R1766–R1771. 21. Purves RD (1981) Microelectrode methods for intracellular recording and ionophoresis. London; New York: Academic Press. 22. Lee MG, Manns ID, Alonso A, Jones BE (2004) Sleep-wake related discharge properties of basal forebrain neurons recorded with micropipettes in head-fixed rats. J Neurophysiol 92:1182–1198. 23. Houweling AR, Brecht M (2008) Behavioural report of single neuron stimulation in somatosensory cortex. Nature 451:65–68. 24. Paxinos G, Watson C (1986) The rat brain in stereotaxic coordinates. Sydney; Orlando: Academic Press. xxvi, [237] p. of plates p. 25. Pinault D (2005) A new stabilizing craniotomy-duratomy technique for single-cell anatomo-electrophysiological exploration of living intact brain networks. J Neurosci Methods 141:231–242. 26. Horikawa K, Armstrong WE (1988) A versatile means of intracellular labeling: injection of biocytin and its detection with avidin conjugates. J Neurosci Methods 25:1–11. 27. Lapper SR, Bolam JP (1991) The anterograde and retrograde transport of neurobiotin in the central nervous system of the rat: comparison

The Juxtacellular Recording-Labeling Technique

28.

29.

30. 31.

32. 33.

34.

35.

36.

37.

38.

39.

with biocytin. J Neurosci Methods 39: 163–174. Brandt HM, Apkarian AV (1992) Biotindextran: a sensitive anterograde tracer for neuroanatomic studies in rat and monkey. J Neurosci Methods 45:35–40. Veenman CL, Reiner A, Honig MG (1992) Biotinylated dextran amine as an anterograde tracer for single- and double-labeling studies. J Neurosci Methods 41:239–254. Adams JC (1981) Heavy metal intensification of DAB-based HRP reaction product. J Histochem Cytochem 29:775. Takahashi K, Lin JS, Sakai K (2006) Neuronal activity of histaminergic tuberomammillary neurons during wake-sleep states in the mouse. J Neurosci 26:10292–10298. Barmack NH, Yakhnitsa V (2008) Functions of interneurons in mouse cerebellum. J Neurosci 28:1140–1152. Bevan MD, Booth PA, Eaton SA, Bolam JP (1998) Selective innervation of neostriatal interneurons by a subclass of neuron in the globus pallidus of the rat. J Neurosci 18:9438–9452. Wu Y, Richard S, Parent A (2000) The organization of the striatal output system: a singlecell juxtacellular labeling study in the rat. Neurosci Res 38:49–62. Mailly P, Charpier S, Mahon S, Menetrey A, Thierry AM, Glowinski J, Deniau JM (2001) Dendritic arborizations of the rat substantia nigra pars reticulata neurons: spatial organization and relation to the lamellar compartmentation of striato-nigral projections. J Neurosci 21:6874–6888. Mallet N, Le Moine C, Charpier S, Gonon F (2005) Feedforward inhibition of projection neurons by fast-spiking GABA interneurons in the rat striatum in vivo. J Neurosci 25:3857–3869. Sadek AR, Magill PJ, Bolam JP (2007) A single-cell analysis of intrinsic connectivity in the rat globus pallidus. J Neurosci 27: 6352–6362. Koyama Y, Honda T, Kusakabe M, Kayama Y, Sugiura Y (1998) In vivo electrophysiological distinction of histochemically-identified cholinergic neurons using extracellular recording and labelling in rat laterodorsal tegmental nucleus. Neuroscience 83: 1105–1112. Simpson JI, Hulscher HC, Sabel-Goedknegt E, Ruigrok TJ (2005) Between in and out: linking morphology and physiology of cerebellar cortical interneurons. Prog Brain Res 148:329–340.

73

40. Bartho P, Slezia A, Varga V, Bokor H, Pinault D, Buzsaki G, Acsady L (2007) Cortical control of zona incerta. J Neurosci 27: 1670–1681. 41. Zhang ZW, Deschenes M (1997) Intracortical axonal projections of lamina VI cells of the primary somatosensory cortex in the rat: a single-cell labeling study. J Neurosci 17: 6365–6379. 42. Cruikshank SJ, Weinberger NM (2001) In vivo Hebbian and basal forebrain stimulation treatment in morphologically identified auditory cortical cells. Brain Res 891:78–93. 43. Schreihofer AM, Guyenet PG (1997) Identification of C1 presympathetic neurons in rat rostral ventrolateral medulla by juxtacellular labeling in vivo. J Comp Neurol 387:524–536. 44. Klausberger T, Magill PJ, Marton LF, Roberts JD, Cobden PM, Buzsaki G, Somogyi P (2003) Brain-state- and cell-type-specific firing of hippocampal interneurons in vivo. Nature 421:844–848. 45. Bassant MH, Simon A, Poindessous-Jazat F, Csaba Z, Epelbaum J, Dournaud P (2005) Medial septal GABAergic neurons express the somatostatin sst2A receptor: functional consequences on unit firing and hippocampal theta. J Neurosci 25:2032–2041. 46. Song G, Yu Y, Poon CS (2006) Cytoarchitecture of pneumotaxic integration of respiratory and nonrespiratory information in the rat. J Neurosci 26:300–310. 47. Pinault D, Deschenes M (1998a) Anatomical evidence for a mechanism of lateral inhibition in the rat thalamus. Eur J Neurosci 10:3462–3469. 48. Pinault D, Deschenes M (1998b) Projection and innervation patterns of individual thalamic reticular axons in the thalamus of the adult rat: a three-dimensional, graphic, and morphometric analysis. J Comp Neurol 391: 180–203. 49. Bartho P, Freund TF, Acsady L (2002) Selective GABAergic innervation of thalamic nuclei from zona incerta. Eur J Neurosci 16:999–1014. 50. Borhegyi Z, Varga V, Szilagyi N, Fabo D, Freund TF (2004) Phase segregation of medial septal GABAergic neurons during hippocampal theta activity. J Neurosci 24: 8470–8479. 51. Lavallee P, Urbain N, Dufresne C, Bokor H, Acsady L, Deschenes M (2005) Feedforward inhibitory control of sensory information in higher-order thalamic nuclei. J Neurosci 25:7489–7498.

74

Pinault

52. Llinas R, Nicholson C (1971) Electrophysio‑ logical properties of dendrites and somata in alligator Purkinje cells. J Neurophysiol 34:532–551. 53. Llinas R, Sugimori M (1980a) Electrophysiological properties of in vitro Purkinje cell dendrites in mammalian cerebellar slices. J Physiol 305:197–213. 54. Llinas R, Sugimori M (1980b) Electrophysiological properties of in vitro Purkinje cell somata in mammalian cerebellar slices. J Physiol 305:171–195. 55. Pinault D, Smith Y, Deschenes M (1997) Dendrodendritic and axoaxonic synapses in the thalamic reticular nucleus of the adult rat. J Neurosci 17:3215–3233. 56. Pilowsky PM, Makeham J (2001) Juxtacellular labeling of identified neurons: kiss the cells and make them dye. J Comp Neurol 433:1–3. 57. Schreihofer AM, Stornetta RL, Guyenet PG (1999) Evidence for glycinergic respiratory neurons: Botzinger neurons express mRNA for glycinergic transporter 2. J Comp Neurol 407:583–597. 58. Jones GA, Llewellyn-Smith IJ, Jordan D (2002) Physiological, pharmacological, and immunohistochemical characterisation of juxtacellularly labelled neurones in rat nucleus tractus solitarius. Auton Neurosci 98:12–16. 59. Pinault D (2004) The thalamic reticular nucleus: structure, function and concept. Brain Res Brain Res Rev 46:1–31. 60. Glaser EM, Tagamets M, McMullen NT, Van der Loos H (1983) The image-combining computer microscope – an interactive instrument for morphometry of the nervous system. J Neurosci Methods 8:17–32. 61. Thomson AM, Deuchars J (1994) Temporal and spatial properties of local circuits in neocortex. Trends Neurosci 17:119–126. 62. Charpier S, Behrends JC, Triller A, Faber DS, Korn H (1995) “Latent” inhibitory connections become functional during activitydependent plasticity. Proc Natl Acad Sci U S A 92:117–120. 63. Miles R, Poncer JC (1996) Paired recordings from neurones. Curr Opin Neurobiol 6: 387–394. 64. Strumwasser F, Rosenthal S (1960) Prolonged and patterned direct extracellular stimulation of single neurons. Am J Physiol 198: 405–413. 65. Spehlmann R, Kapp H (1964) Direct extracellular polarization of cortical neurons with multibarreled microelectrodes. Arch Ital Biol 102:74–94.

66. Fregnac Y, Shulz D, Thorpe S, Bienenstock E (1988) A cellular analogue of visual cortical plasticity. Nature 333:367–370. 67. Cruikshank SJ, Weinberger NM (1996) Receptive-field plasticity in the adult auditory cortex induced by Hebbian covariance. J Neurosci 16:861–875. 68. Brons JF, Woody CD, Allon N (1982) Changes in excitability to weak-intensity extracellular electrical stimulation of units of pericruciate cortex in cats. J Neurophysiol 47: 377–388. 69. Andrew RD, Fagan M (1990) A technique for controlling the membrane potential of neurons during unit recording. J Neurosci Methods 33:55–60. 70. Peinado A, Yuste R, Katz LC (1993) Extensive dye coupling between rat neocortical neurons during the period of circuit formation. Neuron 10:103–114. 71. Ling G, Gerard RW (1949) The normal membrane potential of frog sartorius fibers. J Cell Physiol 34:383–396. 72. Lorente de Nó, R (1949) Cerebral cortex: Architecture, intracortical connections, motor projections. In: Physiology of the nervous system. J.F. Fulton, editor. Oxford University Press, London. 288–313. 73. Scheibel ME, and Scheibel AB (1966) The organization of the nucleus reticularis thalami: a Golgi study. Brain Res. 1:43–62. 74. Jankowska E, Rastad J, Westman J (1976) Intracellular application of horseradish peroxidase and its light and electron microscopical appearance in spinocervical tract cells. Brain Res 105:557–562. 75. McCrea RA, Bishop GA, Kitai ST (1976) Intracellular staining of Purkinje cells and their axons with horseradish peroxidase. Brain Res 118:132–136. 76. Deschênes M, Hu B (1990) Electrophysiology and pharmacology of the corticothalamic input to lateral thalamic nuclei: An intracellular study in the cat. Eur J Neurosci 2:140–152. 77. Konopacki J, Bland BH, Dyck R (2003) Intracellular recording and labeling of neurons in midline structures of the rat brain in vivo using sharp electrodes. J Neurosci Methods 127:85–93. 78. Lipski J, Duffin J, Kruszewska B, Zhang X. Upper cervical inspiratory neurons in the rat: an electrophysiological and morphological study. Exp. Brain Res., 1993; 95:477–487. 79. Pinault D (2003) Cellular interactions in the rat somatosensory thalamocortical system

The Juxtacellular Recording-Labeling Technique

80.

81.

82.

83.

84.

85.

86.

87.

during normal and epileptic 5-9 Hz oscillations. J Physiol 552:881–905. Pinault D, Vergnes M, Marescaux C (2001) Medium-voltage 5-9-Hz oscillations give rise to spike-and-wave discharges in a genetic model of absence epilepsy: in vivo dual extracellular recording of thalamic relay and reticular neurons. Neuroscience 105:181–201. Bordt AS, Hoshi H, Yamada ES, PerrymanStout WC, Marshak DW (2006) Synaptic input to OFF parasol ganglion cells in macaque retina. J Comp Neurol 498:46–57. Joshi S, Hawken MJ (2006) Loose-patchjuxtacellular recording in vivo--a method for functional characterization and labeling of neurons in macaque V1. J Neurosci Methods 156:37–49. Parent A, Charara A, Pinault D (1995) Single striatofugal axons arborizing in both pallidal segments and in the substantia nigra in primates. Brain Res 698:280–284. Manns ID, Alonso A, Jones BE (2000) Discharge properties of juxtacellularly labeled and immunohistochemically identified cholinergic basal forebrain neurons recorded in association with the electroencephalogram in anesthetized rats. J Neurosci 20:1505–1518. Pang K, Tepper JM, Zaborszky L (1998) Morphological and electrophysiological characteristics of noncholinergic basal forebrain neurons. J Comp Neurol 394:186–204. Bokor H, Frere SG, Eyre MD, Slezia A, Ulbert I, Luthi A, Acsady L (2005) Selective GABAergic control of higher-order thalamic relays. Neuron 45:929–940. Hajos M, Allers KA, Jennings K, Sharp T, Charette G, Sik A, Kocsis B (2007)

88.

89.

90.

91.

92.

93.

94.

75

Neurochemical identification of stereotypic burst-firing neurons in the rat dorsal raphe nucleus using juxtacellular labelling methods. Eur J Neurosci 25:119–126. Cetas JS, Price RO, Crowe J, Velenovsky DS, McMullen NT (2003) Dendritic orientation and laminar architecture in the rabbit auditory thalamus. J Comp Neurol 458:307–317. Monconduit L, Lopez-Avila A, Molat JL, Chalus M, Villanueva L (2006) Corticofugal output from the primary somatosensory cortex selectively modulates innocuous and noxious inputs in the rat spinothalamic system. J Neurosci 26:8441–8450. Buhl EH, Han ZS, Lorinczi Z, Stezhka VV, Karnup SV, Somogyi P (1994) Physiological properties of anatomically identified axoaxonic cells in the rat hippocampus. J Neurophysiol 71:1289–1307. Gulyas AI, Miles R, Hajos N, Freund TF (1993) Precision and variability in postsynaptic target selection of inhibitory cells in the hippocampal CA3 region. Eur J Neurosci 5:1729–1751. Baranyi A, Chase MH (1983) Effects of juxtaand intracellular microinjection of ethanol on trigeminal motoneurons in the chronic cat. Brain Res 269:159–164. Soja PJ, Pang W, Taepavarapruk N, Cairns BE, McErlane SA (2001) On the reduction of spontaneous and glutamate-driven spinocerebellar and spinoreticular tract neuronal activity during active sleep. Neuroscience 104:199–206. Yamuy J, Fung SJ, Xi M, Chase MH (2004) Hypocretinergic control of spinal cord motoneurons. J Neurosci 24:5336–5345.

Chapter 4 Neural Recording Using Digital Telemetry André A. Fenton, Kathryn J. Jeffery, and James G. Donnett Abstract Digital telemetry (DT) offers a method of collecting the electrical signals produced by neural activity and transmitting them wirelessly to a receiver/decoder for analysis and storage. The wirelessness means that activity can be recorded from a subject that is behaving relatively normally, which opens up a number of research and therapeutic opportunities – for example, in the study of spatial encoding, or in pre-seizure activity in an epileptic subject. In this chapter we first review the history of neural recording and describe the classic analog method of data processing, outlining the technical problems that need to be solved in collecting and transmitting tiny electrical signals within a noisy environment. We then outline digital signal processing together with the basic principles of telemetry, describing how DT solves these problems in a way that preserves signal fidelity while allowing subjects to move around in an unconstrained way. We finish by describing several situations in which DT is enabling advances to occur both in the laboratory and in the clinic. Key words: Single-unit, Neuronal ensemble, Freely moving, Wireless, Analog, Digital signal processing, Epilepsy, Clinical study, Spatial correlate, Cognition

1. Introduction A major aim in neuroscience is to determine what information is being encoded by ongoing neural activity by recording the activity and then, in essence, decoding it. This enterprise is rooted in the assumption that the collective activity of individual neurons encodes biologically meaningful signals including sensations, perceptions and actions, and also more subjective mental phenomena like feelings, memories, and thoughts and it has two main aims. On the one hand, deciphering the “neural code” (or rather codes) will advance the intellectual endeavor of understanding how the brain works. On the other hand, sufficiently accurate decoding has enormous potential to advance clinical practice,

Robert P. Vertes and Robert W. Stackman Jr. (eds.), Electrophysiological Recording Techniques, Neuromethods, vol. 54, DOI 10.1007/978-1-60327-202-5_4, © Springer Science+Business Media, LLC 2011

77

78

Fenton, Jeffery, and Donnett

especially the treatment of neurological disorders and mental dysfunction for which treatment options often depend on rapid and accurate diagnosis. Recording meaningful signals from the brain’s vast panoply of neurons is a far from trivial technological enterprise. Because many areas of the brain are simultaneously active, and because large neural networks often encode information, it is often necessary to record from hundreds of neurons simultaneously in order to determine the kind of information being processed, and the way in which this processing occurs. Animal studies of high-level knowledge structures such as the “cognitive map” (1) benefit from ensemble recordings of large numbers of neurons made using arrays of electrodes. Clinically, our best current guess is that decoding discrete events like an impending seizure, specific aberrant feelings, and bizarrely inappropriate thoughts will require the ability to record the simultaneous ensemble action potential activity of many thousands of neurons. Furthermore, several decades of research have revealed that the brain has a mostly modular organization and so it is necessary not only to be able to record the signals, but also to be able to determine, with a reasonable degree of accuracy, their anatomical source. And finally, for sophisticated animal experiments, and also for clinical use in humans, it is necessary that these recordings be undertaken not in an anaesthetized and immobile preparation but in an awake and preferably ambulatory subject. These three requirements – high bandwidth, anatomical localization, and portability – pose enormous challenges for the designers of neural recording systems. Recent years have seen significant advances in technologies for decoding brain signals. Field potential recordings, such as 12-channel EEG, provided early electrophysiologists with a means to record pre-ictal or frank seizure activity in epileptic patients, while intracranial recordings, undertaken before or during surgery, have provided neurosurgeons with the ability to determine seizure foci with a high degree of precision. More recently, brain imaging techniques like functional magnetic resonance imaging and optical tomography, which can detect the global metabolic activity of particular brain regions, have been able to pinpoint the source of brain activity with resolution as precise as 1 mm. However, neither the spatial nor temporal resolutions of such methods can yet match the spatial and temporal scales of individual neurons and their action potentials. These technologies have been, to date, clearly limited in their bandwidth, their anatomical precision, or their portability. Electrophysiological single-neuron recordings will be central to the decoding effort described above because of their high anatomical specificity, and also because electrical signals are lightning fast events that can be immediately captured and which provide high-resolution information about neural activity changes.

Neural Recording Using Digital Telemetry

79

Furthermore, electrical signals can be recorded from sites that are up to several tens (or even hundreds) of microns distant from the recording electrode, allowing collection of information from multitudes of simultaneously active neurons. Such high-speed data collection necessitates high-speed processing power. This has become increasingly feasible over the past decade, where advances in digital signal processing (DSP) technology have made it possible to process vastly more data than before using increasingly more compact recording systems. Whereas an early single-neuron recording system would typically occupy a large 19-in. rack standing in the corner of a laboratory or operating theatre and provide only a few channels of data, requiring a large computer for collection and off-line analysis, modern systems offering more than 100 channels can be as small as a laptop computer and, indeed, require only a laptop for processing. It can safely be assumed that further miniaturization will continue to make recording systems increasingly compact and high bandwidth, thus meeting two of the three criteria for clinical and research utility. What about the third criterion – that of portability? A compact multi-channel single-neuron recording system is undoubtedly better than a large one, but if the subject needs to be tethered to it by a recording cable then the range of laboratory experiments is highly constrained, and clinical utility remains limited to the bedside. Increasingly, however, scientists would like to record from behaving subjects in real-world settings, and clinicians would like to be able to record from ambulatory patients, either during routine daily-life monitoring or in emergency situations in the field. Although acquiring and storing data with a subject-mounted memory device may be an option in some cases, the ideal solution to the tethering problem would be to devise a means of wirelessly transmitting multi-channel neuronal data at high speed. Recent developments in wireless signal transmission, in the form of wireless digital telemetry (DT), are beginning to make this possible. In the present chapter we review the principles of DT recording, and then describe a DT device that we have developed that we think has the capacity to advance both basic research and clinical practice. The digital telemeter is fundamentally an inexpensive, miniature, battery-operated, multi-channel, portable digital system to record wideband bioelectric signals. DT can be designed to record both fast and slow brain potentials such as action potentials (APs; 0.3–6 kHz) and local field potentials (LFPs). Its digital technological nature makes it ideally suited to the introduction of a radio link, which obviates the need for wires between the subject and the data-storage machine. DT can be made inexpensive and small because it exploits the billion-dollar market for portable audio applications, which drives chip manufacturers to perfect these

80

Fenton, Jeffery, and Donnett

circuits by continuously reducing noise, power consumption, size, and price while increasing fidelity. DT measures biopotentials with high precision because it digitizes signals on the subject at high (e.g. 24-bit) resolution. The signals are immune to electromagnetic distortion because digitized data are transmitted in an interference-resistant, error-correcting, wireless digital protocol. Below, we first briefly outline the history of single-neuron recording in behaving subjects, before reviewing the conventional analog strategy and then the DT strategy for making these recordings. Then we focus on the technical features that are now attainable with DT technology, including the capacity to multiplex and wirelessly transmit high-bandwidth signals between subject and recording system. Finally, we will briefly consider three application scenarios in which DT has been deployed. 1.1. Historical Overview

Single-unit recording from the brains of freely-behaving subjects began in the late 1960s with the use of implanted microwires to collect extracellular action potential signals, coupled with fieldeffect transistors (FETs) placed close to the electrode on the animal’s head, to provide the first stage of processing. James Ranck Jr., a key innovator of the extracellular technique in behaving animals, has recounted the history of single-unit recording in the context of hippocampal place cell studies (2). We provide a very selective and brief version of the story we have learned from Ranck. Single-neuron recordings in freely-behaving subjects have always, to date, relied on extracellular techniques because intracellular recordings (such as are used in brain slices) are impractical due to the tremendous mechanical difficulty of impaling a neuron and then maintaining the delicate intracellular connection on a moving subject for the duration of behaviorally-relevant episodes. Extracellular recording does not depend on charge transfer between the brain and electrode, instead measuring the electric field that is created by extracellular current flow. The voltage at the electrode conductor depends on the strength of the field and thus the proximity of the electrode to the field source. This capacitive voltage is typically small, on the order of a few hundred microvolts, meaning that recording systems need to be sensitive, and also that electrical noise poses a significant challenge. Single mammalian central neurons were recorded from the hippocampus of the anesthetized cat a full quarter century before FETs enabled recordings from freely-moving animals (3). In those days, signals were recorded as varying analog voltages on an oscilloscope. Throughout the 1950s, single neurons were recorded using glass microelectrodes with fine tips, which are advantageous for getting close to neurons with minimal damage to other parts of the brain, but which have very high impedance. High electrode impedance (Ze) causes several major problems for recording the small single-unit

Neural Recording Using Digital Telemetry

81

potentials that are picked up by extracellular electrodes. The signal is attenuated by the relative magnitude of the amplifier input impedance (Za) according to the ratio Za/(Za+Ze). The electromagnetic noise pickup is a positive function of signal impedance in the cable that connects the animal to the amplifier. Perhaps most ruinous is “movement artifact,” in which moving the recording cable causes microphonic noise, which arises from multiple sources, including the piezoelectric and triboelectric effects, and which is also a positive function of signal impedance. In 1957, Hubel used sharpened tungsten for extracellular recordings of single neurons in the anesthetized cat (4). Tungsten is more robust than glass but it is still fragile. The next year Strumwasser showed that single neurons could be recorded with cut 25 mm insulated wire (5). Unlike tungsten, the cut-wire electrode was not brittle, allowing it to withstand the violence of head movements. Olds used the cut-wires to record from awake rats (6), but to get useful recordings he had to train the rats to remain motionless because the impedance of 25-mm cut-wires is still large (about 1 MW). Everything changed in the late 1960s when FETs became available that were both small and resistant to failure. Able to be mounted on an animal’s head, the FET made it possible to record extracellular action potentials from single neurons in freelymoving subjects by lowering the signal impedance directly at the connection to the electrode. The much reduced impedance in the cable from the animal to the recording system meant that movement artifact was substantially attenuated and recordings could be made in normally-behaving and locomoting subjects. Using this method, O’Keefe was able to observe that neurons in the hippocampus are spatially selective (7), thus igniting the study of internal high-level cognitive structures in the brain. Extracellular recording technology has come a long way since FETs were first introduced, and measurements made by eye from oscilloscope traces, but most recording systems are still fundamentally analog, with a digital stage that is remote from the subject. Electrode signals are initially subjected to a number of operations such as amplification and filtering for frequency bands of interest. Only when all such pre-processing has been completed are the signals sampled and digitized for storage and display on a computer. In analog systems, therefore, the main advance in recording technology has been to replace the conventional oscilloscope by a software counterpart running on a computer. The analog processing still used in most conventional recording systems is in contrast with many other types of instrumentation, and virtually all consumer electronic devices, which these days are fully digital. In digital devices, the original analog signals are converted to a string of binary numbers early in the signal pathway and all the subsequent processing, such as filtering, are

82

Fenton, Jeffery, and Donnett

done numerically, in software. The digital telemetry concept described in this chapter applies these techniques of early digitizing and digital signal processing to the processing of electrophysiological signals. Even ignoring its “wirelessness,” the DT approach thus departs in important ways from the conventional recording strategy. Below, we detail each step in the processing pathway, describing the conventional method for achieving such processing and then comparing this with DT.

2. Steps in the Recording Process: Conventional Analog Vs. DT Systems

The new DT technology described here offers advantages over conventional recording systems at each step of the process from the signal in the subject’s brain to the computer. Electrophysiological signals from mobile subjects are processed via the following steps (Fig. 1): Collection of the signal, buffering, digitization, amplification/filtering, and transmission to the data acquisition system. A major difference between conventional analog recording and DT is that the steps typically take place in a different sequence (though this is not essential) – whereas in analog recordings the signal is first buffered, then transmitted, then amplified and filtered, and finally digitized, in DT the buffering, amplification, and digitization all occur locally, on the subject, and then the digitized signal is multiplexed (to convert the multiple channels of data to a single stream of binary digits) before being sent by radio link to the recording system. This process is detailed in the next section.

Fig. 1. Different stages of a conventional recording system. Simultaneous local field (LFP) and action potential (AP) signals are detected with high-impedance electrodes and buffered to reduce the signal impedance. The wideband signals from each electrode are transmitted by a galvanic conductor to an amplifier for band-pass filtering and amplification (~10 K gain). Here the scheme depicts that the AP frequency band (300–6,000 Hz) has been selected and the LFP signal is lost. The filtered signal is now ~1 V and relatively immune to electromagnetic interference. The signal from each amplifier is sent to an analog-to-digital converter (ADC) where it is digitized (typically at 12–16 bits resolution), stored, and/or displayed under computer control.

Neural Recording Using Digital Telemetry

83

2.1. Signal Collection and Buffering

In modern-day extracellular recording systems, neuronal signals are collected by intracerebral high-impedance, fine microelectrodes, configured either singly or in bundles, and usually comprising either insulated microwires (e.g. ceramic-coated 25 mm platinum–iridium alloy) or silicon probes. The first buffering stage of processing usually takes place on the subject, as close to the electrode connector as possible. Operational amplifier integrated circuits that incorporate large numbers of FETs in convenient packages are used for signal buffering. Although the signal amplitudes are typically small (for example, action potentials are a few hundred microvolts and local field potentials are about 1 mV), the primary purpose of this stage is to reduce the impedance of the signal, which is typically a few hundred kiloohms and depends on the properties of the electrode. Reducing impedance ensures that the source impedance is lower than the impedance of the amplifier, which allows the signals to be relatively immune to electromagnetic noise and movement artifact. Remarkably, clinical EEG systems, and many animal EEG systems, do not use a buffering stage, or if they do it is not until the vulnerable signals have traveled up through a few meters of unbuffered wire. This is a major reason that EEG recordings in animals and patients are artifact-prone. Buffering at the electrode interface allows field and action potentials as small as 40 mV to be recorded from freelymoving rats – even while the animals are jumping or being dropped (8, 9). In DT, buffering does not occur as a separate stage but is part of the signal-conditioning process that takes place prior to transmission (Fig. 2).

2.2. Filtering and Amplifying

Neural signals, which are tiny voltages of the order of tens or hundreds of microvolts, need to be amplified, and they also need to be filtered to remove noise and to isolate the frequencies of interest. In a conventional system, amplification and filtering take place after the signal has been transmitted to the recording system, as described later. This means that any distortion of the signal that occurs en route is amplified along with the signal itself. If the signal can be filtered and amplified on the rat, as in the case of the DT, the amplified signal is far more resistant to subsequent transmission distortion. (It is also the case that because digitization has taken place on the rat too, as we describe later, even more robustness is added to the signal.) Amplifiers are typically used in a differential configuration, in which a reference signal from an electrode implanted elsewhere in the brain is subtracted from the signal of interest in an effort to remove unwanted signals that are common to both electrodes, such as electrical mains noise (“common-mode rejection”; see Fig. 3). Signals are also typically AC-coupled to remove DC potentials. In our DT, the electrode signals are fed to a differential instrumentation amplifier. As with conventional systems,

84

Fenton, Jeffery, and Donnett

Fig. 2. Stages of our DT recording system. Simultaneous LFP and AP signals at high-impedance electrodes are amplified five times, and the common mode interference from an indifferent electrode (black) is subtracted in the amplifying stage, which also reduces the signal impedance. The wide-band output from pairs of amplifiers is low-pass (<6 kHz) filtered and digitized by an audio delta–sigma ADC that digitizes and combines the data from two electrodes in a standard digital audio data format. The binary digital data signal is now a few volts and relatively immune to electromagnetic interference. The outputs from a set of ADCs are combined into one digital signal by digital multiplexing (MUX). Only two ADCs are depicted in the schematic, but any number can be used so long as the aggregate digital data streams can be managed by the next stage of digital signal processing. The digital data are processed (filtered, reordered, multiplied, ignored, etc.) by a microcontroller unit (MCU). The processed digital signal is fed to a radio transceiver integrated circuit and transmitted as a frequency-modulated 2.4 GHz radio signal. A remote receiver converts the radio signal to a digital signal and the data are reorganized into separate stereo data streams conforming to a digital audio standard by an MCU within the receiver. The digital audio stereo signals are processed by digital signal processing (DSP) units, which can filter the data into any band for storage and display under computer control via the USB. Both AP and LFP signals from each electrode can be captured. (Inset) A photograph of a DT transmitter stage.

a signal from a reference electrode is subtracted to remove common-mode noise, and a small amount of gain (200–300×) is applied. The function of subsequent filtering is to remove unwanted frequencies in the signal so that the frequencies of interest can be analyzed separately. Action potential waveforms, because of the sharp nature of their waveforms, have high-frequency components between 300 Hz and 6 kHz and conventionally require high-pass filtering so that the slow undulations of the EEG can be removed. By contrast, local field potentials (LFPs) such as hippocampal sharp wave-associated ripples (10, 11) tend to have maximum frequencies of about 300 Hz, and normally are

Neural Recording Using Digital Telemetry

85

Fig. 3. Effects of filtering on a rat brain signal recorded with HM-L-coated 25 mm platinum–iridium wires. (a) The raw, unfiltered, and unreferenced signal. (b) Effect of referencing: the same signal after a signal from a nearby wire has been subtracted from it, removing the common mode noise. (c) The referenced signal after application of a 50 Hz notch filter to remove mains noise. (d) The referenced and notch filtered signal after being low-pass filtered (1–250 Hz). Note that the slow undulations persist but the highfrequency, spiky elements have been removed. (e) The referenced signal after being high-pass filtered. Notice that the low-frequency undulations have been removed but the spiky, high-frequency elements remain: these are due to background neural and other electrical activities. Traces recorded on an Axona DacqUSB system by Robin Hayman.

high-pass filtered so that the superimposed AP spikes can be removed (Fig. 3). This is necessary in conventional systems due to the limited resolution at which signals are digitized. By contrast, DT allows both high- and low-frequency signals, from APs and LFPs, respectively, to be recorded simultaneously. This is because the DT digitizes the signal at a very high resolution, as described in the next section. This has a number of advantages, one of which is that DT requires only minimal filtering (specifically, only the input AC-coupling plus a simple two-pole anti-alias filter before the digitization stage). Analog filters, such as the commonly used Butterworth-type, work by reducing the amplitude of those components of the signal that are outside the frequencies of interest, with the attenuation gradually increasing for signals further from the filter’s passband. For a typical filter design, if the high-pass frequency is set to 6 kHz, the amplitude of signal components at 12 kHz may only be reduced by 60%, thus leaving a considerable residue of high-frequency signal energy. Thus, it would be necessary to sample such a signal at a rate much higher than 12 kHz, even though the highest frequency of interest is only 6 kHz. In principle, it should only be necessary to sample a signal at two times the highest frequency of interest (the so-called Nyquist limit). 2.3. Digitization

Digitization involves taking the continuous, time-varying voltage signal produced by neurons and converting it to a stream of numerical values that can be subjected to analysis in software. In conventional analog recording systems, digitization takes place

86

Fenton, Jeffery, and Donnett

right at the end of the signal processing pathway, after the signal has been transmitted from the subject to the recording apparatus. The reason that wireless recording of single neurons has become feasible in recent times is that miniaturized digital signal processing (DSP) chips have become available that are so small that the digitizing can take place on the subject itself, rather than in the recording system. This offers a number of advantages. Perhaps the most important of these is that, for reasons described in the next section, digital signals can be more reliably transmitted from the subject to the rest of the recording system. The signal is thus highly immune to movement artifact or electrical noise that might otherwise be introduced in the transmission pathway. In converting a continuous and smoothly varying signal into a discrete set of samples, digitization causes a loss of information in the spaces between the samples; it introduces constraints on resolution both vertically (i.e. in the amplitude domain) and horizontally (in the time domain). In the amplitude domain, the number of bits used to encode the signal amplitude constrains how finegrained the amplitude representation can be, which can be problematic if both the high-amplitude (mV) LFP signal and the low-amplitude (mV) AP signal are simultaneously of interest. For example, hippocampal 4–12 Hz, q oscillations (11) tend to have 1 mV amplitudes, and large irregular activity (LIA) is several millivolts, nearly two orders of magnitude greater than APs (11). While many investigators avoid this problem by focusing on only one or the other signal-type, in recent years there has been a substantial interest in studying the relationship between concurrent LFP (population) and AP (single cell) signals because it seems that this relationship may be used to encode information (12–16). Thus, the problem of how to simultaneously record both highamplitude LFPs and low-amplitude APs needs to be solved. In conventional analog recording systems, digitizing takes place right at the end of the processing sequence, and is usually done by PC-based analog-to-digital converters (ADCs). By this stage the signal has already been magnified, and the frequencies of interest isolated, by analog components, as described in the section on amplification and filtering. Although there are PC-based 16-bit ADC cards, commercial ADCs that operate at above 1 MHz (allowing 32 AP channels to be digitized at 32 kHz) typically operate at 12-bit resolution. To maximize the resolving power of the ADC, the frequency band for the signal of interest is selected in the filtering stage and then amplified to utilize most of the bits in the digitization stage. In conventional systems, this can be done by setting the gain so that signal peaks do not saturate the ADC; the result of this is that on average, the signal is represented by up to 4 bits less than the maximum resolution of the ADC, effectively reducing the resolution of a 12-bit ADC to only 8 bits. Since LFP amplitudes are greater than AP amplitudes

Neural Recording Using Digital Telemetry

87

by an order of magnitude, only the low-frequency LFP band or the high-frequency AP band can be digitized with sufficient resolution, but not both. Consequently, to record both the LFP and the AP bands from one electrode requires doubling the number of main amplifiers and ADC channels, so that the signal can be processed in parallel, in both ways. An alternative approach, rather than filtering the signal two different ways, is to use the capability of modern ADCs to sample the signal at vastly higher resolution. In our DT, the ADCs digitize at 18–24 bits, which – even with minimal amplification of the input signal – allows for a wide dynamic range: this means that both APs in the microvolt range and large LFPs in the millivolt range are able to be captured simultaneously from the one signal. In the time domain, digitization also produces problems when a continuously varying signal is sampled at discrete time intervals because a problem known as aliasing arises (Fig. 4), whereby the undersampling of high-frequency components in the signal introduces spurious additional low-frequency components, and/or a shift in phase of some components. Given the likely importance of phase in neural encoding, this is obviously a serious problem and so it is necessary to avoid aliasing by making sure that the signal is sampled at a frequency at least twice that of the highest frequency in the input (the Nyquist limit previously alluded to). For neural signals, digitization is usually done at a rate of about 32 kHz/channel for action potential data, while local field potentials are digitized at 1 kHz or less. As discussed in the section on filtering and amplification, the highest frequency of interest even in APs is around 6 kHz, so that it should only be necessary to sample at 12 kHz, but the limitations on conventional

Fig. 4. Aliasing caused by undersampling of a signal. (a) The original signal is a regular sine wave of constant peak amplitude. The dots show the points at which samples were taken. (b) When a curve is reconstructed from the samples, using interpolation to fill in the spaces, the reconstructed signal is distorted in frequency, phase, and amplitude relative to the original.

88

Fenton, Jeffery, and Donnett

filters to attenuate frequency components above 6 kHz quickly enough require that a much higher sampling rate be used. An alternative is to use almost no filtering, but sample the signal at a very high frequency, so that only very high-frequency signals (tens to hundreds of kHz) are at risk of being aliased back into the brain signal frequency band (17). However, very high rate digitization brings with it a number of problems. As we describe later, our implementation of DT instead uses a particular kind of ADC (delta–sigma), which has a very high “internal” sampling rate, but an easier-to-work-with output sample rate. This allows us to have a very simple filter, but nevertheless to sample at only 12 kHz, exactly twice the 6 kHz highest frequency of interest. 2.4. Transmission to the Data Acquisition System

The next step in the signal-processing sequence is to transmit the data from the animal to the data acquisition system for further processing. This process faces two challenges. First, the signal needs to maintain integrity (e.g. preserved signal-to-noise ratio) over the often meters-long distance between the animal and the recording system. Second, the medium of signal transmission – usually wires, in the case of a conventional system – places limitations on the amount of data that can be carried. For example, to collect many channels of data simultaneously in a conventional system, it is necessary to run at least as many wires from the subject to the recording system as there are electrode channels, which limits the number of channels that can be recorded at once due to the weight of the cabling (especially for small subjects like rats and mice). It also forces the subject to be “tethered” and thus restricted to a relatively small spatial area. While tethered recordings have been (and continue to be) enormously useful, they limit the kinds of investigations that can be conducted. Multiplexing the signals can alleviate the tethering problem, which involves combining the multi-channel data into a single channel that can be transmitted via just one wire for demultiplexing by the recording system. An alternative, or additional solution, is radio transmission of the signal, which eliminates the need for heavy cabling but which renders the signal potentially vulnerable to electromagnetic interference. Multiplexing can be used with both analog and digital signals, and with either wired or wireless signal transmission. The signals from multiple channels are broken up into short fragments which are spliced together into a single signal, transmitted to the recording system and then demultiplexed into individual channels again at the other end. This introduces data loss in the temporal domain for each channel due to the time needed to transmit the remaining channels. For analog demultiplexing, the signals from each channel are sampled and held as voltages by capacitors during the intervals while the remaining channels are being transmitted.

Neural Recording Using Digital Telemetry

89

This means that the amplitude of the signals can continue to be rendered with high resolution, but because they are being held by capacitors, which naturally leak over time, they are inherently less accurate. With digital signals, a similar fragmenting and splicing process occurs, but demultiplexing is electronically simpler because the data stream was, and continues to be, a string of bits (with signals represented by large discrete voltage levels) rather than a smoothly-varying voltage. The signal then needs to be transmitted to the recording system. Multiplexing greatly reduces the need for multiple wires between the subject and the recording system: for example, our DT multiplexes the signals from 16 channels onto one wire, and needs only two more wires for power and ground, making three in total. However, the number of wires can be reduced to zero by transmitting the information as a radio signal. Radio transmission of neural signals was originated decades ago (a review by (18)), and an early device was even capable of transmitting single-unit signals (19). The natural place to insert a wireless link is just after the buffering stage. In a conventional analog recording system, this is usually done by taking the buffered analog voltages, multiplexing them as described above, and using the multiplexed signal to modulate a radio frequency carrier wave. The original signals are recovered at the receiver by demodulating the received radio signal and demultiplexing the signal back into the individual channels comprising it. This is “analog telemetry.” In principle, this carrier-wave modulation could even be used to transmit multiple unmultiplexed signals in the same way that continuous stereophonic audio signals are mixed into a single wave, and then separated again at the receiving end. In practice, this is very complicated to achieve for multiple signals, and multiplexing is preferred, even with the slight data loss it entails. Radio signal transmission has the problem that the signal between the transmitter and receiver is vulnerable to distortion or fading. As the distance between the transmitter and receiver varies, the received signal strength varies as well, and at larger distances the signal may drop below the level at which it can be successfully received, causing data loss. Since the radio signal is transmitted at a single frequency, reflections of the signal in the recording environment can interfere destructively with the original signal, causing reception to fail. Furthermore, other sources of radio interference can affect reception, such as wireless telephones and networks, computers with wireless peripherals, microwave ovens, engines, etc. These problems are serious with analog systems because once the signal has been corrupted it cannot be recovered, and it is impossible to know that such corruption has occurred. By contrast, digital signals can be protected by means of error detection/correction

90

Fenton, Jeffery, and Donnett

algorithms (detailed below) which enable the received signal to be checked for error. If these occur, the receiver can request the retransmission of data “packets” (stored locally on the subject after digitization). Thus, radio transmission of digital data – digital telemetry – offers the great advantage that data can be collected in a robust and high fidelity manner without the impediment of recording cables. Our DT wireless transmission exploits a radio transmitter integrated circuit that was originally intended for the consumer digital audio market. This device offers a number of features that make it particularly attractive. It is small (15 × 15 mm total circuit area including all components) and low-powered, which is important for a portable battery-operated device, particularly if it is to be carried by a small animal. It accepts and transmits standard audio-format signals, but also provides a parallel bi-directional data channel that can be used to communicate between the transmitter’s built-in miniature microcontroller unit (MCU) and the recording system. And most importantly, it offers various mechanisms for overcoming radio frequency (RF) noise, reflections, and other problems that might threaten the transmission of acquired electrode data. It does this by sacrificing more than half of its native 4 Mb/s radio link bandwidth to “Quality of Service” (QoS) protocols – in particular, it packages up the sampled electrode data and attaches “checksums,” which are essentially periodic summaries of the foregoing data stream, that are transmitted along with it. If the signal is corrupted in-air, the received checksum will no longer match its corresponding data stream and so the receiver will request a retransmission. Furthermore, both transmitter and receiver are able to dynamically change their operating frequency (“frequency hop”) to optimize radio link reliability. Even in cluttered lab environments where other wireless networking components are often present, the device is able to work at several meters’ range, often out of line-of-sight, without any radio link dropouts.

3. A Specific Implementation of DT

In the foregoing sections, we have outlined the general characteristics of conventional neuronal recording systems, and hinted at particular ways in which the digital telemetric approach we have developed can improve upon this conventional approach. For the past several years we have been utilizing emerging DSP and telemetry technology to design a DT that specifically enables highspeed transmission of many channels of single-neuron data. In this section, we describe a specific implementation of DT, discussing particular choices of components and configurations and the

Neural Recording Using Digital Telemetry

91

Table 1 A summary table of our DT technical specifications Feature

Value

Data rate

1.536 Mb/s

Analog/digital conversion

18–24 bits

Max voltage resolution

1–38 nV/bit (at gain = 200)

Input range

± 5 mV (at gain = 200)

Bandwidth

0.64 Hz–6 kHz (at 12 kHz ADC)

Crosstalk

90 dB

Noise level

1 mVpp

Input impedance

1.0e12 Ω

CMRR

115 dB

Current consumption

80 mA

reasons for these, together with an evaluation of performance of the DT. The technical specifications are summarized in Table 1. 3.1. Description of DT

As previously described, our DT implementation integrates signal buffering, differential recording, filtering, amplification, digitizing, multiplexing, and a radio transmission stage into one tiny unit. The buffering and differential referencing is accomplished by high-precision instrumentation amplifiers at the DT’s inputs. Signals are then AC-coupled (0.05 Hz highpass). Minimal amplification (200–300×) is necessary, since the digitization stage utilizes a 24-bit ADC and so can resolve down to 1 nV per bit (or 38 nV/ bit with an 18-bit ADC). In order to avoid aliasing problems but still be able to sample at the Nyquist limit, we use a particular kind of analog-to-digital converter, the “delta–sigma”-type, which offers several advantages. At 12 kHz digitization, the delta–sigma ADC has a linearphase anti-aliasing filter that attenuates frequencies between 5.8 and 6.0 kHz by 70–90 dB. The linear phase property means that the introduced phase-shift changes linearly within the passband. This is unlike the standard Butterworth-type filter that is used in most conventional analog main amplifiers. The filter roll-off is very steep (approx. 70 dB/octave) meaning that frequencies near 12 kHz are attenuated by about 3.5 orders of magnitude. This sharp roll-off allows DT to sample 6 kHz signals (APs) at 12 kHz without aliasing. A second feature of the delta–sigma ADC which supports this is that by internally oversampling the signal (by an order of 100 times the output sample rate) it allows the DT’s

92

Fenton, Jeffery, and Donnett

output sample rate to approach the Nyquist limit. This means that high-frequency APs can be safely digitized at a rate that is a fraction of the conventional recording system data rate, dramatically reducing the bandwidth required to transmit the signals and the space required to store them. 3.2. Bandwidth Usage Optimization

As discussed previously, the native bandwidth of our DT’s radio transmitter is 4 Mb/s. After application of the QoS mechanisms that guarantee robust signal transmission, the device offers a residual bandwidth of 1.536 Mb/s for the actual data to be carried. The actual contents of this data stream are not constrained by the radio device, which allows us to use the DT’s MCU to structure them in a large variety of ways. The MCU contains timing circuits, which allow the sampling rates of the ADCs to be under software control. Since the MCU receives the digitized data streams from the ADCs, it can reorganize these in a wide variety of ways; specific channels can be selected for amplification, downsampling, and transmission, while others are ignored. Again, this selection process is fully programmable. Finally, the MCU can choose how many of the ADCs’ native bits of resolution are actually kept (i.e. it can transmit only the top 8 or top 12, or 16 as necessary). This constellation of features allows combinations of different numbers of channels at different sampling rates and resolutions to be processed and transmitted. Since the bandwidth available to any radio communications device has some limit, this ability of the DT to have its data stream configured for purpose on the fly allows it to optimize the use of the available bandwidth. For example, in a situation where more electrodes have been implanted than the radio would have bandwidth to transmit at a high sample rate, an experimenter can select those electrodes with the best signals, and transmit only those. Alternatively, it might be preferable to transmit more signals simultaneously, but at lower resolution.

3.3. Power Management

In order to operate completely wirelessly, a DT requires an independent source of power. Typically, this will be a rechargeable battery, carried either on a head-mounted device for short recordings and/or on a larger subject, or on the animal’s back for longer recordings on a small animal such as a rat or mouse. A battery weighing only 3.5 g is sufficient to power the DT for 1.5 h. However, an alternative for animal studies might be to locate an electromagnetic power source under the animal’s cage, and use inductive pickup within the DT transmitter to extract power from the surrounding electric field as in inductively-coupled power transfer (ICPT) technology. In principle, such an approach can suffer from the problem of what to do to control the power level; excessive induced power would need to be dissipated as heat, which could be problematic. However, the DT’s radio link

Neural Recording Using Digital Telemetry

93

allows it to establish a feedback connection over the air with the inductive power source and instruct it to maintain appropriate output levels. Thus, DT can be the basis for a continuous recording system. 3.4. Testing DT

To compare DT rigorously with conventional analog recording, we used a test signal with the waveshape properties of a continuous burst of APs (high-frequency peaks at 3,555 and 5,587 Hz and peak amplitudes ~105 mV; interspike interval 3.5 ms). The first thing to note is that the DT-recorded waveform properties were superior to those measured from a commercial analog recording system having 12-bit digitization (Fig. 5). The voltage recorded at the sample that most often contained the peak voltage was much more variable in the analog system. The coefficient of variation was 40% compared to only 11% with DT. Even though the analog system is digitized at 32 kHz, and the DT at 12 kHz, this result illustrates that DT more reliably captured the peak of the AP-like signal. This is a property of the delta–sigma-modulated ADC in DT. Waveform properties can be more accurately calculated by reconstructing the signal from the digitized values. Even so, parameter estimates from cubic spline reconstructions of the DT recordings were less variable than the spline estimates from the conventional analog recordings. Note also that the waveform properties were virtually identical when recorded by DT via galvanic and radio transmission. This occurs because the transmitted signal is digital and thus not subject to distortion. Either the signal is transmitted faithfully, or there is a dropout. If a bit is lost and cannot be recovered by error correction the whole signal is lost until the reading frame of the formatted bit stream is recovered. 99.7% of the signals were

Fig. 5. Waveform properties recorded by an analog 12-bit system, and DT with galvanic and radio transmission. Since the gains and filtering in the analog (10,000 times, 300–6,000 Hz) and DT (100 times, 50–6,000 Hz) systems differed, the coefficients of variation (CV) for three waveform properties are plotted. The voltage at peak sample describes the voltage at the ADC sample, which was most often the largest voltage. This parameter describes how well the rapidly changing peak voltage was sampled. Oversampling by the delta–sigma-modulated ADC in DT measured the peak more reliably. This digitization error can be attenuated by reconstructing the continuous waveform from digitized values. A cubic spline was used to represent the continuous waveform, and the positive and negative peaks from the spline function were calculated. The variability of peaks in the DT waveforms was still smaller. Waveforms were identical for galvanic and radio transmission.

94

Fenton, Jeffery, and Donnett

transmitted wirelessly as the experimenter moved within 10 m of the receiver without regard for maintaining line-of-sight between the transmitter and the receiver. Transmission did not degrade when competing radio frequency interference was produced by Wi–Fi transmission from either a wireless Ethernet router or laptop placed 1 m from the receiver. These simple tests demonstrate that the DT’s radio communication is robust.

4. Practical Uses of DT As discussed earlier, the great advantage of DT is that it frees the subject from the constraints of a recording cable, allowing recording of neuronal activity in far more behaviorally relevant settings (see Fig. 6). It greatly enhances the ability to record from the brain. So far there are three broad domains in which such technology is proving to be particularly useful: experiments in animal epilepsy, animal cognition studies, and clinical use. 4.1. Animal Epilepsy Monitoring

Epilepsy is a devastating disorder and anti-convulsants are often limited in efficacy or have intolerable side effects. Surgical intervention is an undesirable last resort that is not even always possible. Therefore, rapid progress in understanding epilepsy depends on use of animal models to develop new approaches to treatment or even cures. Detailed observations and therapeutic assessments that are initially not clinically possible can be performed in animals. However, in the animal as well as in the clinical setting, it is difficult to correlate electrographic and behavioral seizure manifestations, necessitating continuous video monitoring and analysis.

Fig. 6. DT offers improved discrimination even at the input stage. The digital signal was converted to analog and recorded by a commercial analog system for 5 min then the electrode was recorded by the conventional system. Plotting peak action potential voltage against time reveals that the discriminated unit (black) is less variable. The unit was also more distinct from other APs (gray) in the DT recording (i.e. the peak was more separate from other events).

Neural Recording Using Digital Telemetry

95

Furthermore, seizures in animals, as in humans, may manifest electrographically but not be accompanied by overt behavioral signs. This means that epilepsy research needs chronic, 24-h EEG in order to test potential anti-convulsant therapies with accuracy. In addition, to understand mechanisms, the high-frequency (80–500 Hz) oscillations and other aspects of the EEG may be important clues that a seizure is about to occur. Current technology is limiting because it involves low-resolution recordings of only the lowest frequencies (<70 Hz), and animals must be attached to cables for recordings outside their home cage. DT, being inexpensive, miniature, wireless, and portable, has the potential to solve many of these issues. Accordingly, we have merged DT and digital video technologies (20) to create what we call the animal Epilepsy Monitoring Unit (aEMU) to permit standardized, video-synchronized, continuous wideband, multi-site electrophysiological recording in the rat home cage. This enables the study of the prediction, treatment, and basic science of spontaneous seizures in animal models of epilepsy. In clinical epilepsy, the Epilepsy Monitoring Unit (EMU) has enormous diagnostic and therapeutic utility. It is used for identifying seizure type and severity, monitoring therapeutic efficacy, and determining the targets for surgical intervention. Given the centrality of the modern EMU, not only in clinical practice but in the development of our current understanding of epilepsy, it is remarkable that there are currently few comparable systems for animal experimentation that permit such long-term (days to weeks) recordings. In addition to direct consequences for epilepsy research, the lack of an EMU has indirect consequences – only the models with the most obvious and severe motor convulsions are typically studied. Less obvious seizure-manifestations cannot be studied easily. Thus technical limitations preclude the more direct study of the human disease. Current animal monitoring protocols that do use EEG have additional problems. Animals usually require transportation to a remote chamber, pressure to the head to connect cables, and tethering of the head for recording: these can be substantial stressors. Stress can exacerbate seizures and alter their manifestation (21, 22). Stress to the head can cause irritation or pain, disturb electrodes, and lead to problems that confound the assessment of seizures. DT recordings, however, allow an animal to move in its home cage uninfluenced by experimenters, providing accurate estimates of epilepsy-related events with reduced stress. DT is ideally suited to investigate the electrical origins of spontaneous seizures and the mechanisms of epilepsy. The focus of most epilepsy research has been on field potentials (0.5–500 Hz). The wideband abilities of DT will also make it easy for researchers to investigate whether information about seizure development and evolution is also contained in the fast AP activity of

96

Fenton, Jeffery, and Donnett

single-unit ensembles of individual neurons, or in their spatio-temporal relationships to field potentials. As discussed earlier, the relationship of spike discharge of individual neurons to LFPs is intensely investigated in several areas of neuroscience. One of these is concerned with understanding the organization and balance of excitation and inhibition (23, 24). Excitation/ inhibition imbalance has long been a central hypothesis for seizure genesis, but these spike-firing patterns have not been investigated due to bandwidth restrictions and to the restricted numbers of available channels. The aEMU will permit study of these unit/field relations. High-frequency oscillations are now also suspected to be a key aspect of epilepsy and epileptogenesis (25). Current animal EEG systems do not provide the signal-tonoise ratio or bandwidth required to detect these small events (~100 mV; ~10% the amplitude of slower oscillations), which the high voltage and temporal resolutions of DT is readily able to detect. Figure 7 shows a photograph of a rat with a DT transmitter for continuous recordings in its home cage. The rat has epilepsy, which DT was crucial in discovering. The rat received an ibotenic acid lesion as a 7-day-old pup, which is a widely used preparation to create animals with schizophrenia-related characteristics (26). In some of these animals, we noticed what might have been mild seizure-like events that were very brief, lasting 2–3 s. Subsequent, chronic monitoring using DT revealed that these animals were indeed having spontaneous seizures that occurred every day or so. These electrographic seizures were often not accompanied by overt behavioral manifestations. Since schizophrenia researchers typically observe one of these animals only for a few tens of minutes, it is not surprising that epilepsy has not been reported in these animals. However, this provides a clear illustration of how use of DT can advance and refine investigations into the neurobiology of epilepsy and other neurological and mental dysfunctions. 4.2. Animal Cognition Studies

A second major potential use for DT is in neurobiological studies of animal cognition. Early studies of animal behavior focused on simple behaviors such as reinforcement learning and perceptual discrimination, exemplified by the early and influential School of Behaviorism, which ignored (or even denied) the existence of internal knowledge structures such as the “cognitive map.” However, as the field has advanced, scientists have become increasingly interested in such cognitive structures, particularly as the development of chronic single-neuron recording made the demonstration of these structures indisputable. The consequent rise of “Cognitivism” has meant that behavioral studies have become increasingly sophisticated, and it is becoming necessary to increase the sophistication of recording systems to match. DSP technology has made possible the recording of many neurons simultaneously, and now DT makes this possible in subjects that are behaving in situations that approximate natural conditions.

Neural Recording Using Digital Telemetry

97

Fig. 7. (Top) Photograph of a rat wearing a DT transmitter in its home cage for chronic recordings. In this configuration, with power supplied to the transmitter electronics wirelessly by magnetic induction and wireless radio transmission to a remote receiver, an animal can be recorded continuously for arbitrarily long periods of study. This is especially valuable for monitoring spontaneous events that are rare, like electrographic seizures in animal models of epilepsy. (Bottom) Forty-second traces from an eight-channel DT recording of a spontaneous, generalized seizure in an adult rat that received a ventral hippocampal excitotoxic insult as a 7-day-old neonate. The voltage traces are local field potentials at eight locations indicated to the left of each trace. Chronic DT recordings were used to discover that these animals have approximately one spontaneous seizure per day. This seizure frequency is considered low by experimental criteria but high by clinical criteria. Furthermore, this neonatal lesion preparation is not an established epilepsy model; it is, however, closely related to a well-established developmental model of schizophrenia (26). Seizures in this model have not been reported, probably because of the low seizure frequency and the corresponding need for chronic monitoring to detect seizure events (recordings by Hsin-Yi Kao and Heekyung Lee).

One of the areas in which this technology promises to be most useful is in spatial behavior, in which animals traverse large regions of the environment in order to solve a spatial task such as navigating to a goal or returning home from a foraging expedition. With tethered recording systems it has been possible to build up a picture of how neurons represent small, two-dimensional spaces;

98

Fenton, Jeffery, and Donnett

of course, space is not two-dimensional, it is three-dimensional, and naturally-behaving animals will typically move considerable distances vertically as well as horizontally in their daily lives – when, for example, climbing through burrow systems or trees. The study of spatial encoding in large or complex environments is only just a beginning, but initial results suggest that this field of enquiry is poised to take off in the next few years. Already such studies are revealing aspects of the neural representation of space that were not appreciated by recordings from standard, small experimental spaces. One example is the grid-cells, recently discovered in the medial entorhinal cortex. These cells discharge in multiple, regularly spaced locations that tile an environment in a hexagonal grid pattern of spatial action-potential discharge (27). Although the grid pattern is striking, it was not initially recognized in recordings from medial entorhinal cortex as rats explored standard experimental spaces of about 1 m (28–31). More recently, these authors have reported data from very extended 18-m long environments (32) – an enormous technical challenge when the subject is tethered to a recording cable. Another experiment in a large-scale space recorded place neurons from the hippocampus as rats explored a 1.4 × 1.5 m chamber, finding that place fields became multiple, in a pattern that had never been recognized in recordings from small environments (8). Thus, recordings while rats explored larger than typical environments have corrected our notions of the fundamental firing pattern of cells in the hippocampus and medial entorhinal cortex. It is certain that such experiments are just the beginning of our attempts to understand representation in large-scale space, but further development of this line of enquiry will require telemetry. DT will make it possible, for the first time, to record in (relatively) more naturalistic settings such as vivarium colonies or three-dimensional mazes. Such recordings will rapidly permit us to build up a picture of how neurons in different brain areas are naturally active during normal daily behaviors such as social interaction, foraging, mating, nestbuilding, rearing of offspring, and fighting with intruders. 4.3. DT in Clinical Settings

As well as advancing animal studies, DT also promises to substantially advance clinical electrophysiology applications. While many applications are envisaged, perhaps the most obvious is seizure monitoring in epileptic patients from scalp and intracranial electrodes. Specific properties of DT, especially the high digital resolution and range, wideband frequency response, small size, portability, and battery-power converge to enable, also for the first time, recording APs and LFPs from the brains of freelymoving people. Such recordings can be accomplished before, during, and after seizure, with good chances of maintaining stable detection of the same set of neurons and neuronal potentials. Human intracranial recordings have not typically been made using buffering amplifiers at the electrode connectors because of

Neural Recording Using Digital Telemetry

99

their large size, mechanical coupling to large batteries, and heat generation, which made it impractical to store such equipment under the wound bandages on the subject’s head. Instead, the buffering amplifiers have been located at the level of the subject’s chest and unbuffered wires that carry the high impedance signals from the electrodes make the connection. As a result, changing electric fields in the vicinity of the subject induces noise currents in the unbuffered pathway and this prevents recordings during movement, seizure, and even after seizure when clinical staff is attending to the patient. Since DT is miniature and low-powered, it can be placed under a bandage with only a lead from the battery carried somewhere on the subject’s person. DT digitizes at the electrode interface and transmits the digital signal without wires, so these problems are avoided and recordings that were previously impossible become straightforward. As an enabling technology, DT has substantial potential within the next decade to establish whether high-frequency LFPs and AP spike trains from single and multi-unit intracranial recordings have clinical utility for diagnosing neurological and psychiatric disorders (33, 34).

5. Conclusion This chapter has reviewed the principles of chronic single-neuron recording and introduced the new technology of digital telemetry. Neither digital signal processing nor telemetric transmission are new; what is new is the use of recently developed high-speed DSPs and ADCs to enable digitization and processing of multiple, highbandwidth neuronal channels locally, at the source (i.e. on the subject). This means that digitized single-neuron data, which are much more robust to degradation and are also more easily processed than its analog counterpart, can be feasibly transmitted by radio link in a way that has never been possible previously. This, in turn, opens the door to new types of recording scenarios which we have detailed in the final section; such scenarios include recording in complex environments where a cable would become entangled or prolonged recording in clinical situations where a human subject does not wish to be tethered for long periods of time. What is next for telemetric neuronal recording? Since chip technology continues to evolve at a rapid rate, it is likely that it will soon be possible to record from many more channels simultaneously, meaning that truly large-scale ensemble neuronal recording will finally be possible. Currently, one of the biggest limitations in technology is the need for a power source to be carried by the subject. Since battery life is a universal problem in modern electronic applications, battery technology is intensively researched for the consumer market and so it is likely that the advent of miniature, long-lasting high-capacity rechargeable batteries will

100

Fenton, Jeffery, and Donnett

eventually mean that prolonged neuronal recordings can be undertaken even on rodents. This is an exciting possibility, for it means that truly naturalistic studies of neuronal activity, in which neurons are recorded over long periods of time as animals live their daily lives without experimenter intrusion, will become possible. If the range can be extended, and combined with GPS localization technology, it may even become possible to record neuronal activity from animals in the wild – for example, from navigating or even migrating animals. On the clinical side, it may mean that the development of indwelling neural pacemakers (i.e. long-term implanted electrodes that detect abnormal neural activity and then correct it) will at last become feasible. It seems likely that in the not-so-distant future, recording cables will be regarded as a quaint and archaic feature of old-fashioned recording systems, and the next generation of neurophysiologists will wonder how this one ever managed to do any experiments at all!

Acknowledgments The initial DT concept was developed in collaboration with Imre Szabo and Kalman Mathe as part of an EC Framework V Project (QLG3-CT-199-00192, “N APPY”). Portions of this work were supported by NINR Grant R41 NR009877-01A1, and NINDS Grants R43 NS057839-01 and R42 NS064474-02A1 to A.A.F, and by an EC Framework 7 Project (HEALTH-F2-2007-200873, “SPACEBRAIN”) to K.J.J and J.G.D. A.A.F. and J.G.D. are founders of Bio-Signal Group Corp., a company that develops digital telemetry for commercial applications. J.G.D. and K.J.J. founded Axona Ltd., a company that sells digital electrophysiology recording systems. References 1. O’Keefe J, Nadel L (1978) The hippocampus as a cognitive map. New York: Clarendon Press. p 570. 2. Ranck JB, Jr., Kubie JL (2008) Historical perspective: place cells in Ann Arbor and Brooklyn. In: Mizumori SJ, editor. Hippocampal place fields: relevance to learning and memory. Oxford: Oxford University Press. xvii. 3. Renshaw B, Forbes A, Morison BR (1940) Activity of isocortex and hippocampus: electrical studies with micro-electrodes. J Neurophysiol 3:74–105. 4. Hubel DH (1957) Tungsten microelectrode for recording from single units. Science 125:549–550.

5. Strumwasser F (1958) Long-term recording’ from single neurons in brain of unrestrained mammals. Science 127:469–470. 6. Olds J (1965) Operant conditioning of single unit responses. Proceedings of the XXIII International Congress of Physiological Science, Tokyo. 7. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Res 34:171–175. 8. Fenton AA, Kao HY, Neymotin SA, Olypher A, Vayntrub Y, Lytton WW, Ludvig N (2008) Unmasking the CA1 ensemble place code by exposures to small and large environments: more place cells and multiple, irregularly

Neural Recording Using Digital Telemetry

9.

10. 11.

12.

13. 14.

15.

16. 17. 18.

19.

20.

21.

arranged, and expanded place fields in the larger space. J Neurosci 28:11250–11262. Lenck-Santini PP, Fenton AA, Muller RU (2008) Discharge properties of hippocampal neurons during performance of a jump avoidance task. J Neurosci 28:6773–6786. Buzsaki G, Leung LW, Vanderwolf CH (1983) Cellular bases of hippocampal EEG in the behaving rat. Brain Res 287:139–171. Vanderwolf CH (1969) Hippocampal electrical activity and voluntary movement in the rat. Electroencephalogr Clin Neurophysiol 26:407–418. Arieli A, Sterkin A, Grinvald A, Aertsen A (1996) Dynamics of ongoing activity: explanation of the large variability in evoked cortical responses. Science 273:1868–1871. Barth DS (2003) Submillisecond synchronization of fast electrical oscillations in neocortex. J Neurosci 23:2502–2510. Bragin A, Jando G, Nadasdy Z, van Landeghem M, Buzsaki G (1995) Dentate EEG spikes and associated interneuronal population bursts in the hippocampal hilar region of the rat. J Neurophysiol 73:1691–1705. Csicsvari J, Hirase H, Mamiya A, Buzsaki G (2000) Ensemble patterns of hippocampal CA3-CA1 neurons during sharp wave-associated population events. Neuron 28:585–594. O’Keefe J, Recce ML (1993) Phase relationship between hippocampal place units and the EEG theta rhythm. Hippocampus 3:317–330. Galton I (2002) Delta-sigma data conversion in wireless transceivers. IEEE Trans Microw Theory Tech 50:302–315. Mackay RS (1968) Bio-medical telemetry; sensing and transmitting biological information from animals and man. New York, Wiley. xi, 388. Skutt HR, Beschle RG, Moulton DG, Koella WP (1967) New subminiature amplifiertransmitters for telemetering biopotentials. Electroencephalogr Clin Neurophysiol 22: 275–277. Fenton AA, Muller RU (1996) Using digital video techniques to identify correlations between behavior and the activity of single neurons. J Neurosci Methods 70:211–227. Eggers AE (2007) Temporal lobe epilepsy is a disease of faulty neuronal resonators rather than oscillators, and all seizures are provoked, usually by stress. Med Hypotheses 69:1284–1289.

101

22. Rhodes ME, Harney JP, Frye CA (2004) Gonadal, adrenal, and neuroactive steroids’ role in ictal activity. Brain Res 1000:8–18. 23. Dragoi G, Buzsaki G (2006) Temporal encoding of place sequences by hippocampal cell assemblies. Neuron 50:145–157. 24. Klausberger T, Magill PJ, Marton LF, Roberts JD, Cobden PM, Buzsaki G, Somogyi P (2003) Brain-state- and cell-type-specific firing of hippocampal interneurons in vivo. Nature 421:844–848. 25. Bragin A, Wilson CL, Engel J (2003) Spatial stability over time of brain areas generating fast ripples in the epileptic rat. Epilepsia 44:1233–1237. 26. Lipska BK, Jaskiw GE, Chrapusta S, Karoum F, Weinberger DR (1992) Ibotenic acid lesion of the ventral hippocampus differentially affects dopamine and its metabolites in the nucleus accumbens and prefrontal cortex in the rat. Brain Res 585:1–6. 27. Hafting T, Fyhn M, Molden S, Moser MB, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. 28. Frank LM, Brown EN, Wilson M (2000) Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27:169–178. 29. Fyhn M, Molden S, Witter MP, Moser EI, Moser MB (2004) Spatial representation in the entorhinal cortex. Science 305: 1258–1264. 30. Hargreaves EL, Rao G, Lee I, Knierim JJ (2005) Major dissociation between medial and lateral entorhinal input to dorsal hippocampus. Science 308:1792–1794. 31. Quirk GJ, Muller RU, Kubie JL, Ranck JB, Jr. (1992) The positional firing properties of medial entorhinal neurons: description and comparison with hippocampal place cells. J Neurosci 12:1945–1963. 32. Kjelstrup KB, Solstad T, Brun VH, Hafting T, Leutgeb S, Witter MP, Moser EI, Moser MB (2008) Finite scale of spatial representation in the hippocampus. Science 321:140–143. 33. Bragin A, Engel J, Jr., Wilson CL, Fried I, Buzsaki G (1999) High-frequency oscillations in human brain. Hippocampus 9:137–142. 34. Uhlhaas PJ, Singer W (2006) Neural synchrony in brain disorders: relevance for cognitive dysfunctions and pathophysiology. Neuron 52:155–168.

Chapter 5 Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis Hui Kuang and Joe Z. Tsien Abstract One of the fundamental goals in neuroscience is to uncover, in real-time, the formation and retrieval of the brain’s associative memory traces. Here, we describe methodology we have developed to permit large-scale recording and analysis of neuronal activity from ensembles of neurons. We have constructed a lightweight multi-channel recording microdrive that permits long-term recording from multiple neurons, or several brain regions simultaneously, from freely behaving mice. Our device is capable of acquiring up to 128 channels of neuronal activity data simultaneously from freely moving mice. The recording and decoding of such highdensity signals can be combined with the acquisition of behavioral responses of mice in elegant paradigms in order that one might define the firing patterns of multiple neurons and their relationships with behavioral performances as memory traces are formed or recalled. It is well known that startling events are often encoded as episodic memories that are remembered well for years. We have recorded hundreds of individual CA1 units using our high-density recording technique in mice while subjecting them to repetitions of particular startling stimuli. By decoding simultaneously acquired hippocampal network activity our analyses have revealed functional coding units, that we have termed neural cliques. Our data indicate that any episodic event is represented and encoded by the activation of a set of neural clique assemblies that are organized in a categorical and hierarchical manner. The neural clique assemblies’ organization represents a network-level mechanism capable of vast storage capacity, and permits identification of common patterns from individual behavioral episodes and their application to abstract concepts necessary for intelligence and adaptive behaviors. The decoding and deciphering of these real-time ensemble-recording technologies offer great promise for application to multiple brain regions and will significantly impact the development of brain–machine interface technology. Key words: Single-unit, Neuronal ensemble, Hippocampus, Behavioral correlate, Mouse, Highdensity, Episodic memory, Cognition, Neural representation, Neural network, Startle

1. Introduction Mouse models are one of the most powerful systems to study the relationship between genes and behavior. As approximately 60–70% of genes have been estimated to be either brain-specific Robert P. Vertes and Robert W. Stackman Jr. (eds.), Electrophysiological Recording Techniques, Neuromethods, vol. 54, DOI 10.1007/978-1-60327-202-5_5, © Springer Science+Business Media, LLC 2011

103

104

Kuang and Tsien

or highly enriched in the brain, it is important to study these genes’ functions in regulating cognition and behaviors. The rapid development of a series of inducible and region-specific gene knockouts (28, 43, 51), and more recently, inducible protein knockout techniques (5, 18, 20, 55) have permitted ever more precise investigations of the relationship between genes and cognition. For example, a series of conditional gene knockout experiments have allowed us to show that the knockout of the NMDA receptor in the CA1 region of the hippocampus impairs the CA1 synaptic plasticity and leads to profound memory deficits (52). Moreover, genetic enhancement of NMDA receptor coincidencedetection function through the up-regulation of the NR2B subunit in the mouse forebrain can lead to significant enhancement in both learning and memory (45, 62), thereby stringently validating Hebb’s learning rule (48). Furthermore, the latest studies have also shown that memory can be selectively erased using a rapid and inducible manipulation of alpha-CaMKII at the time of memory recall (5). Thus, various mouse genetic techniques provide powerful ways to dissect the molecular and genetic mechanisms of cognition in the mammalian species. One major technical bottleneck in our study of the relationship between genes and behaviors lies at our limited ability to measure neural network properties and dynamical patterns associated with genetic and behavioral changes. Over the past several decades, neuroscientists have obtained valuable insights by using EEG to map global brain responses or by recording the activity of one or a few neurons at a time. However, neither approach provides a direct means to investigate the network mechanisms underlying information processing. Encouragingly, in recent years simultaneous monitoring of activities of many neurons has become more feasible in rats (15, 16, 32, 40). With the recent technical advances, researchers have gained a greater capacity for recording many neurons from various mammalian species, ranging from rats to cats to monkeys (15, 19, 32, 40, 58, 59). Since mice are typically only about 1/10–1/15 of the body weight of rats (20–30 vs. 300–450 g of body weight), many of the ensemble-recording microdrives designed for rats are often too big to be used for recording in mice. A mouse version of such microdrives has been reported that is able to carry as many as 24 channels that can simultaneously record approximately 20–30 individual neurons in the brains of freely behaving mice (31). In this chapter we describe the design and construction of a high-density microdrive system which can hold up to 128 channels and allows for the measurement of activities of over 200 individual neurons in the brains of freely behaving mice. This high-density ensemble-recording array, as illustrated in the later part of the chapter, provides a valuable tool for the study of realtime memory-encoding patterns in freely behaving mice.

Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis

2. Design and Construction of a Large-Scale EnsembleRecording Microdrive 2.1. Considerations for Chronic Recording from the Hippocampus in Mice

105

The recording microdrive described here is mainly for recording neural activity from a large number of individual CA1 cells in the hippocampus of freely behaving mice. We are interested in the hippocampus because anatomically the hippocampus, and especially its CA1 sub-region, is known to be a crucial site for the formation of long-term memories (39, 42, 44, 52). The hippocampus has been a major focus of systematic investigations, which have produced extremely valuable insights into the molecular and neural mechanisms of learning-related behaviors (9, 12, 13, 24, 35, 46, 61). In fact, individual hippocampal neurons in rats have been shown to respond to many external inputs (3, 8, 10, 35, 56, 60, 61). Yet the response variability at the level of individual neurons poses a theoretical obstacle to understanding how the brain achieves its robust real-time neural coding of the stimulus representation (1, 13, 23). One good example is place cells in the hippocampus which show “location-specific” firing when an animal navigates through familiar environments (35). The discharge of place cells is shown to be extremely variable during individual passes through their place fields (13), thereby making the prediction of real-time place fields unreliable on each single trial. The traditional way to deal with the response variability of single neurons is to average spike discharge of the neurons over repeated trials. Although data averaging across trials permits the identification of tuning properties of the individual neurons, this practice invariably loses crucial information regarding real-time-encoding processes in the brain. Moreover, place cells’ spatial selectivity seems to be strongly dependent on the running motion of the animals; if animals simply sit on that “place field” location, the place cell’s firing selectivity would simply degenerate, which is different from our motion-independent perception and memory of location or space. These variability features of “place cells” require researchers to remove those recording files corresponding to the periods in which animals do not reach certain running speeds or are immobile. It has long been thought that mnemonic encoding of information may involve the coordinated activity of large numbers of individual neurons (17). However, the small size of a mouse has greatly constrained the number of individual neurons that researchers can record from using the traditional electrophysiological methods. Their small brain size requires us to develop lightweight high-density recording techniques so that sophisticated neural analyses of cognition in mice can be feasible.

106

Kuang and Tsien

2.2. Design and Construction of a Multi-Channel Recording Microdrive

Our high-density microdrive can be constructed in 32, 64, 96, 128 channels, or even more, and the electrodes on the microdrive can be formatted in the single electrode, stereotrode (two wires), or tetrode format (four wires). Take the 128-channel recording electrodes as an example – the electrodes can be formatted as either 32 tetrodes or 64 stereotrodes corresponding to 32 or 64 recording sites, respectively (see top insets in Fig. 1e). The electrodes consist of two independently movable bundles of 32 stereotrodes or 16 tetrodes (64 channels on each side of the hippocampi). The foundation for the microdrive was prepared from four 36-pin connector arrays positioned in parallel; one array was secured with epoxy glue (5 min epoxy system, ITW

Fig. 1. High-density ensemble-recording array for mice. (a–e) Construction of the high-density ensemble-recording microdrive. (a) The base foundation for the microdrive. (b) Four 36-pin connector arrays were positioned at the base of the microdrive in parallel. Each bundle of 32 pieces (for stereotrodes) or 16 pieces (for tetrodes) of polyimide tubing was glued to an independently movable screw nut on the microdrive base. (c) A microdrive on the assembly stage. The free ends of electrode wires are wrapped around to adjacent connect pins. (d) A fully assembled, adjustable 128-electrode microdrive. (e) 128 channels can be formatted with either tetrodes (left circle) or stereotrodes (right circle) on each bundle. The tip of the two electrode bundles was cut to a certain angle (10–20°) to fit the contour of the dorsal CA1 cell layer. Black scale bars in circles of E are 100 µm. White scale bars in a–c are 3 mm. (f, g) High-density in vivo ensemble recording in freely behaving mice. (f) An example of a freely behaving mouse implanted with a completed 128-channel microdrive in bilateral hippocampi. (g) This ultra-light microdrive, even after connected to 128-channel headstages and cables, allows the mouse to move freely in various situations, such as running, exploring, eating, grooming, sleep, performing learning tasks, etc.

Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis

107

Performance Polymers, Riviera Beach, FL) to both sides of the microdrive base, and the third and the fourth arrays were separated from the middle array with a rectangular plastic spacer (Fig. 1a, b). A bundle of 16 (for tetrodes) or 32 (for stereotrodes) pieces of polyimide tubing (TSP 075150: inner diameter 75 µm, outer diameter 150 µm, Polymicro Technologies, Phoenix, AZ) were glued to each of the two independently movable screw nuts on the microdrive base. The distance between the two bundles was pre-calculated to allow bilateral recording from the right and left side of the dorsal hippocampus (2.0 mm lateral to bregma and 2.3 posterior to bregma on the both sides). After the glue had dried, the polyimide was trimmed to ensure that at least 1–2 mm of tubing would protrude from either end of the microdrive base throughout the advancing range of the microdrive (Fig. 1b). Each stereotrode or tetrode was constructed by twisting a folded piece of two or four wires (STABLOHM 675, H-FORMVAR, 25 µm for stereotrode and 13 µm for tetrode, California Fine Wire), securing the two strands together with a low-intensity heat source and removing the insulation from the tips of the free ends over an open flame. Each completed stereotrode or tetrode was threaded through one of the polyimide tubes secured to the microdrive screw nuts. After all electrodes had been inserted into separate polyimide tubes, the twisted ends of the wires were cut to a length that extended 3–4 mm beyond the end of the polyimide bundle, and the wires were then secured to the polyimide tubing with glue. The angle for the 64 wires on each side was about 10–20° that would maximally follow the contour of CA1 pyramidal cell layer. Of course, depending on the region and contour of loci, each investigator can modify the angle of the electrode bundle. The free end of each stereotrode or tetrode (insulation had been removed) was wrapped around adjacent connector pins (Fig. 1c). In addition, each wrapped connector pin was individually coated with silver paint to enhance conduction (Silver Print II, GC Electronics). The silver-coated connector pin arrays were then coated with nail enamel (Chanel, Inc., New York) for insulation. A reference wire (magnet wire, 0.01 mm2, Belden electronic division) was soldered to the four pins on ends of each connector array. In the final stages of microdrive construction, the looped stereotrode were secured to the foundation of the microdrive (Fig. 1d) to reduce the potential for accidental damage following surgery. In addition, the tips of the tetrode were plated with gold (Cyanida Gold solution, SIFCO Selective plating) to a final impedance of 500–800 kW.

108

Kuang and Tsien

3. Surgical Procedure We used wild-type hybrid B6CBA/J mice because this strain gives good behavioral performances than many of the inbred mouse strains. Mice were typically handled for several days prior to surgery to minimize the potential stress of human interaction. On the day of surgery, the mouse was anesthetized with an i.p. injection of 60 mg/kg ketamine (Bedford Laboratories, OH) and 4 mg/kg Dormitor (Pfizer Animal Health, NY). The mouse’s head was immobilized in a stereotaxic frame, and its eyes were coated with sterile ocular lubricant (PuraLube Vet Ointment, Pharmaderm, Melville NY). After the hair above the skull had been shaved, Betadine solution was applied to the skin surface and an incision was made along the midline of the skull. The edges of the cut skin were held to the sides with small clips, and the membranous layer was removed to expose the skull. Hydrogen peroxide was applied to the skull surface to permit visualization of the bregma position along the midline. The positions for the two bundles (2.0 mm lateral to bregma and 2.3 mm posterior to bregma on the both right and left sides) were then measured and marked. Four holes were drilled in a rectangular array surrounding the coordinates designed for the stereotrode or tetrode bundles, and small screws were secured in each of these holes and fixed with dental cement. Holes for the stereotrode or tetrode bundles were then drilled and the dura was carefully removed. The stereotaxic apparatus was then used to lower the stereotrode or tetrode bundles into these holes and into the mouse’s cortex. The gaps surrounding the stereotrodes or tetrode were filled with softened paraffin, and the microdrive was stabilized with dental cement. The reference wire attached to the two posterior head screws was soldered to the reference wire affixed to the connector pin arrays of the microdrive, and copper mesh was wrapped around the entire microdrive to protect the wires from potential damage. The mouse was then aroused with an injection of 2.5 mg/kg Antisedan and returned to its home cage.

4. In Vivo Recording and Behavior Paradigm 4.1. In Vivo Recording and Localization of the Electrodes

The mouse is usually allowed to recover for several days before advancing the electrodes. The connector pin arrays on the microdrive were first attached to pre-amplifiers with extended cables to allow for the monitoring of neuronal signals using the 128-channel Plexon system (Dallas, TX) in stereotrode or tetrode format. The extracellular signals were recorded by the 128-channel Plexon

Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis

109

system, and then the signals were filtered through microelectrode pre-amplifiers to separate neuronal activity and field potentials. A helium-filled Mylar balloon was tied to the cables to alleviate the weight of the apparatus and cables, thereby enabling the mouse to move freely in various situations (e.g., running, exploring, eating, grooming, sleep, performing learning tasks, etc.) (Fig. 1f, g). Typically 4–5 days after surgery we begin to advance the electrodes (the mice were gently held-still by hand). The stereotrode or tetrode bundles were advanced slowly toward the hippocampal CA1 region, in daily increments of about 0.07 mm, until the tips of the electrodes had reached the CA1 as deduced from an assessment of field potential and neuronal activity patterns. The ensemble activities of a large number of individual neurons were then subsequently recorded during freely behaving states, such as running or sleep (25). At the end of experiments, the mouse was anesthetized and a small amount of current was applied to four channels in the microdrive to mark the positioning of the electrode bundle. Histological staining, with 1% Cresyl Echt violet, was used to confirm the electrode positions. 4.2. Ensemble Recording of CA1 Single Units’ Activity in Response to External Stimuli

While most of the hippocampal physiology laboratories have been studying spatial encoding and place cells in the rat hippocampus and many excellent articles and chapters have been written on this topic, here we describe some of our experiments aimed at examining the neural encoding of episodic experiences. It is well known that the human brain can produce robust memories of startling episodic events (e.g., devastating earthquakes, roller coaster rides, shark attack, etc.), even upon a single exposure (21, 22). Such episodic events are likely to involve large numbers of neurons, thereby greatly increasing the chance of finding them simultaneously and, consequently, facilitating the analysis of network level real-time-encoding patterns in the brain. Therefore, we have correspondingly designed a set of simple, yet robust, behavioral paradigms using three different types of startling episodic stimuli delivered to the mice as a way of creating discrete episodic startle memories: air blow – a sudden blow of air to the animal’s back (mimicking an owl attack from sky); drop – a short vertical free fall inside a small elevator (recreating the mouse’s experiences inside a cookie jar that falls from a kitchen shelf); and shake, an unexpected brief earthquake-like shaking of the mouse’s cage. We used computer-controlled mechanical devices for controlling the precise timing and intensity of these startling stimuli. Such robust episodic events are capable of producing strong episodic memories (25, 27). The spike activities in freely behaving mice in response to various startling episodes such as an air blow, drop, and shake were recorded by the Plexon MAP System (26).

110

Kuang and Tsien

5. Data Analysis 5.1. Off-Line Spike Sorting and Neuron Classification

Overall, the spike activities during various behavior paradigms were recorded and then sorted off-line by using the MClust 3.0 and KlustaKwik 1.5 programs as described (1–4). Only stable units (for at least 6 h) with clear boundaries and <0.5% of spike intervals within a 1-ms refractory period are included in the analysis. First, the spike waveforms and their associated time stamps for each of the 128 channels were stored in data files using Plexon system format (*.plx). The artifact waveforms were removed and the spike waveform minima were aligned using the Offline Sorter 2.0 software (http://www.plexon.com Dallas, TX), which resulted in more tightly clustered waveforms in principal component space. The Plexon system data files (*.plx) were then converted to Neuralynx system format (*.nst) and spike-sorted with the MClust3.3 program (http://www.cbc.umn.edu/~redish/mclust David Redish). This program permits classification of multidimensional continuous data. Its cluster-splitting feature (Buzsaki lab) yields superior accuracy in comparison to the other available spike-sorting software and is therefore particularly suitable for spike sorting of hippocampal signals. Principal component analysis was used to extract defining features from the spike wave shapes that are used as part of the input for the MClust3.3 spike-sorting program. The first two principal components, as well as the peak height, valley value, FFT, and total energy of spike waveform parameters, were calculated for each channel, and units were identified and isolated in highdimensional space through the use of an autoclustering method (KlustaKwik 1.5) (16). After autoclustering, the clusters containing non-spike waveforms were deleted using the “KlustaKwik Selection” function, and then the units were further isolated using a manual cluster-cutting method in MClust. Only units with clear boundaries and less than 0.5% of spike intervals within a 1 ms refractory period are included in the analysis. Our microdrive system can provide stable recordings in vivo as evident from the similar spike-sorting pattern obtained from the same electrode over 6 h (Fig. 2a, b), and in a few cases, even over one month (Fig. 2c). Both tetrodes and stereotrodes are capable of providing reliable separation of individual units (Fig. 2d, e). On occasions, up to 14 stable units can be recorded and separated on a single stereotrode (Fig. 2e). The quantitative measures of cluster quality in our spike sorting were obtained by measuring the L-Ratio and Isolation Distance (MClust3.3 software). These two measures, recently introduced by Schmitzer-Torbert et al., quantify how well-separated the spikes of one cluster (putative unit) are from other spikes recorded simultaneously on the same multichannel electrode (41). The good separation of units by our

Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis

111

Fig. 2. Recordings of single units in mice. (a) Automatic spike sorting was performed using the KlustaKwik method and was followed by the MClust method for manual cluster cutting and merging. Six sorted units detected by a stereotrode are presented here in different colors. The stereotrode waveforms (the waveforms of each unit detected by each tip of the stereotrode, shown side-by-side) of the individual units are shown along the corresponding clusters. (b) The right panel shows the same six stable units at the completion of 6-h recording. (c) The panels show the same two single units recorded from a stereotrode remained stable for 1 month. The letters on the top left corner indicate the date of the recording, July 24th, August 1st, and August 27th. The insets at the bottom left corner and top right corner show the waveforms of the single units detected by two channels of the stereotrode. (e, f) Separation of multiple single units by either stereotrodes or tetrodes. (e) Nine single units were detected by a single tetrode. Average waveforms of nine separated units and the corresponding energy spike distributions were used for this classification. The insets show four waveforms sideby-side detected by the four channels of the tetrode. (f) Fourteen single units were detected by a single stereotrode. Average stereotrode waveforms of the putative separated units and the corresponding energy spike distributions were used for this classification. The insets show that two waveforms were detected by the two channels of the stereotrode.

112

Kuang and Tsien

stereotrodes and tetrodes is evident from the measurements of L-Ratio and Isolation Distance. Consistent with the previous studies on rats (2, 4, 14, 37), CA1 units in the mouse hippocampus can also be divided into two classes: principal units (putative pyramidal neurons) and theta units (putative interneurons), distinguishable by the width of the waveform, firing rates, and the inter-spike interval (Fig. 3). Putative pyramidal cells have low mean firing rates and have wider and asymmetrical wideband waveforms (top left insets of Fig. 3), whereas putative interneurons have, on average, higher discharge rates and narrower spike width (bottom left insets of Fig. 3). In addition, discharge dynamics of pyramidal cells and interneurons also differed as reflected by their autocorrelograms (the right panels in Fig. 3). Pyramidal cells are known to fire complex-spike bursts with 3–10 ms inter-spike intervals (14, 16, 37). Consequently, the autocorrelograms of pyramidal cells typically

Fig. 3. Putative excitatory and inhibitory neurons. Discharge dynamics has characteristic difference between pyramidal cells and interneurons, reflected by their histograms (left-side panels) and autocorrelograms of inter-spike intervals (right-side panels). The putative pyramidal cell shown in the upper panels has lower mean firing rate and wider and more asymmetrical wideband waveform than the interneuron shown in the bottom panels. As known by the inter-spike-intervals plot, pyramidal cells have complex-spike bursts with 3–10 ms inter-spike intervals. Consequently, the autocorrelogram of pyramidal cells typically shows a characteristic peak at 3–5 ms, followed by a rapid exponential decay, whereas putative interneurons exhibited a much slower decay. The left panels show the histograms of inter-spike intervals. The y-axes represent the frequency of the spike occurrence. This putative pyramidal cell exhibited typical burst firing, with the peak intraburst firing rate reaching 120 Hz, whereas the putative inhibitory cell had the mean firing rate at the 22 Hz. The insets within the left panels show waveforms. The right panels show the autocorrelograms of inter-spike intervals of these two cells.

Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis

113

show a characteristic peak at 3–5 ms, followed by a rapid exponential decay (top right panel in Fig. 3), whereas putative interneurons exhibited a much slower decay (bottom right panel in Fig. 3). In general, the number of pyramidal cells constitutes the majority of the recorded cells in the CA1 region. 5.2. Ensemble Patterns of CA1 Single Units’ Activity Triggered by Startling Events

By using our high-density recording technique, we can record hundreds of individual CA1 units in mice while subjecting them to seven repetitions of each of the above-mentioned startling stimuli. These stimuli produced collective changes in firing rates and activity patterns within a subset of the recorded neuronal populations. As an example, the spike rasters of 260 simultaneously recorded single units from a mouse showed dynamical changes in the firing patterns of many CA1 neurons after the occurrence of single startling episodes of air blow, drop, or shake (Fig. 4). Although a significant proportion of the simultaneously recorded CA1 cells did not respond to any of the startling episodic stimuli, the remainder exhibited significant changes in firing rates. In general, based on their temporal response duration, dynamical changes triggered by startling episodes can be divided into four major firing modes: transient increase, transient decrease, prolonged increase, and prolonged decrease. We then analyzed the response-selectivity of these CA1 cells. Spike-raster plots and peri-event histograms revealed that some of these CA1 neurons responded to all three types of startling events (Fig. 5, general episodic cells), whereas other cells appeared to only respond to air blow, drop, or shake (Fig. 5b–d). Importantly, we have also observed many CA1 cells that reacted to combinations of two different types of startling episodes, responding to both startles either equally or differentially (Fig. 5e, f; sub-general episodic cells), thereby reflecting the hippocampal binding function of cortical inputs. Interestingly, there are also cells, which would fire specifically to only one class of episode (e.g. drop cells, earthquake cells, or air-puff cells) (see Fig. 5g, h). This diversity of response-specificity suggests that the startling events are likely to be represented in CA1 by activity patterns of unique ensembles of neurons. Because the hippocampus is involved in the formation of episodic memory that contains not only ‘‘what’’ information but also ‘‘where’’ information (7, 35, 36, 39, 38, 44), we next asked to what extent the firing patterns of CA1 cells triggered by startling events are influenced by the environmental contexts in which the startles occur. To address this question, we repeated the sudden air blow stimuli in two distinct cages and the drop stimuli in two different elevators. Although a given type of startling stimulus triggered similar responses in many of the responsive CA1 units regardless of the environmental context (Fig. 5g, h), some

114

Kuang and Tsien

Fig. 4. Startle-induced ensemble of single-unit firing patterns in CA1. Startle-induced ensemble of single-unit firing patterns in CA1. Spike rasters of 260 simultaneously recorded single units from mouse A during a period of 1 s before and 2 s after the occurrence of single startling episodes of air blow (a), drop (b), and shake (c) (t = 0 marked with vertical grey line). x-axis, timescale (seconds); y-axis, the numbers of simultaneously recorded single units (n = 260). The startle stimulus durations are indicated as a bar next to the vertical grey line above the spike raster. Although many neurons did not respond to startling stimuli, a significant portion of recorded units exhibited dynamical changes in their firing rates.

Fig. 5. (continued) All histograms use a time bin of 200 ms. Please note the response variability of each individual neuron across the repeating seven trials (y-axis). x-axis indicates the timescale (seconds). The vertical grey lines in spike rasters indicate the time marker for the occurrence of startling events.

Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis

115

Fig, 5. Diversity and selectivity of CA1 neuronal responses to different startle stimuli. Spike-raster plots and peri-event histograms (time bin, 100 ms) for representative units are shown. In (a–f), top, middle, and bottom correspond to air, drop, and shake events, respectively. (a) A unit responsive to all three types of startles (general-startle unit). (b) A unit that increased its firing selectively to air blow (air-blow unit). (c) Dropselective unit. (d) Shake-selective unit. (e) Air- and drop-responsive unit. (f) Drop- and shake-responsive unit. (g) A unit that responded to air blows in both contexts. (h) A unit that responded only in context B. (i) A unit that responded to drop similarly in elevator A and in elevator B. (j) A unit that exhibited a strong drop response only in Elevator B.

116

Kuang and Tsien

CA1 cells exhibited context-specific firing changes (Fig. 5i, j). Thus, these contextual experiments demonstrate that the startling episode triggered firing changes in some CA1 neurons that are not only stimulus-dependent, but also depend on the context in which the event occurs, thereby reflecting a clear neural integration of both what and where information in the hippocampal region, a hallmark mnemonic function of the hippocampus. 5.3. Ensemble Patterns Classification and Visualization of CA1 Network Encoding

The existence of a variety of responsive individual neurons suggests that startling events may be represented through distinct activity patterns at the network level by an ensemble of individual neurons. What are those ensemble-encoding patterns? How can these encoding patterns be mathematically described and even visualized? To provide an intuitive solution that would facilitate a search for the relevant network-encoding patterns that might be hidden among the activity of the hundreds of simultaneous recorded neurons, we used MDA (11) to compute a highly informative low-dimensional sub-space among the firing patterns of responsive neurons. MDA is a supervised dimensionality reduction method that is well suited for identifying and integrating the classification-significant statistical features of neural population responses to distinct types of known stimuli. In our implementation of MDA, a startle trial was represented as a high-dimensional feature vector of normalized neural responses Rn= (fpoststartle, n−fpre)/(fo+ fpre). Each vector has k = n× m dimensions, where n is the number of CA1 neurons (e.g., n= 260) and m is the number of time bins (e.g., m= 2 for 500 ms over a 1-s response period). Responses that fell below a threshold criterion were eliminated, leading to a sparse set of neural features included in the analysis (e.g., in mouse A of the 260 × 2 features, 185 features with R> 0.5 were used, with 160 features selected from the first 500-ms post-stimulus bins and 25 feature vectors from the second 500ms bins). In other words, the non-responsive neurons were excluded because they are unlikely to carry neural information; as a result only the responsive neurons are used for MDA analysis. Next, we calculated the low-dimensional MDA sub-space that is maximally discriminating for the response matrix. Projecting the neural population responses to given startle events onto single points in this sub-space shows that repeated startle responses indeed form clear, well-separated clusters (Fig. 6a), which are distinct from the cluster formed by the rest projections. In other words, CA1 ensemble activity patterns elicited by various startling stimuli can be mathematically described and conveniently visualized as specific startle clusters in a low-dimensional encoding sub-space (here in 3Ds), achieving levels of startle discrimination not seen in individual CA1 neuron responses. We also confirmed that non-responsive cells indeed did not contribute to these classifications, as adding them to MDA analysis did not

Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis

117

Fig. 6. Classification, visualization, and dynamical decoding of CA1 ensemble representations of startle episodes by multiple-discriminant analysis (MDA). (a) Ensemble patterns during epochs of awake rest (gray ellipsoid), air-blow (green ellipsoid), drop (blue ellipsoid), and earthquake (shake; magenta ellipsoid) are shown in a 3D sub-encoding space obtained using MDA for a mouse in which 260 CA1 cells were simultaneously recorded; MDA1, MDA 2, and MDA3 denote the discriminant axes. Three representative dynamic trajectories of network patterns during encoding are shown for each type of startling event. (b) Dynamic monitoring of post-learning spontaneous reactivations of network traces during and after the actual startling events. 3D sub-space trajectories of the ensemble-encoding patterns during drop and air-blow episodes in the same mouse are shown. The initial response to an actual air-blow or drop event (black lines) is followed by spontaneous reactivations (red and green lines for two air-blow reactivations, and blue lines for drop reactivation), characterized by co-planar, geometrically similar lower-amplitude trajectories (directionality indicated by arrows). (c) The same trajectories of reactivation traces after rotation to show that the trajectories are highly specific towards their own startle clusters. These post-learning dynamic trajectories are typically smaller in amplitude than the initial trajectories and take place on the same timescale of those triggered by the actual events. Reactivations within the first several minutes after the startle event seem to number between one and five, with random intervals.

significantly change the MDA patterns (data not shown). The encoding patterns in CA1 ensembles can be independently confirmed by another statistical pattern classification such as Principal Component Analysis (PCA) (26). 5.4. Monitoring of Real-Time-Encoding and Dynamical Post-Event Processing of CA1 Network Traces

MDA provides a sensitive and mathematical means of measuring and visualizing the ensemble neural activity patterns in a highly informative encoding sub-space. Furthermore, this dimensionality reduction method permits one to dynamically monitor the population firing patterns by using a sliding-window technique (1-s width). Using the fixed matrix coefficients produced by the MDA method, we can compute the instantaneous projection of neural responses during the entire experiment. As such, the temporal evolution of the ensemble activity patterns can be directly visualized as dynamical trajectories in the encoding sub-space. For example, during the baseline state before startles, the instantaneous projection was confined to the rest ellipsoid; however, upon the actual startling stimulus we observe a planar trajectory that begins in the rest cluster, quickly visits the corresponding startle cluster, and then returns to rest (Fig. 6a). Interestingly, post-event processing of newly formed ensemble patterns can also be directly detected and precisely quantified (Fig. 6b, c). These spontaneous reactivations of memory traces,

118

Kuang and Tsien

represented by dynamic trajectories with similar geometric shapes but smaller amplitudes, seem to occur at intervals ranging from several seconds to minutes after a discrete startling event. Existence of these reactivations suggests that the memory formation is a highly dynamic process, and that the reactivations could be crucial in the immediate post-learning fixation of newly formed memory traces (Fig. 6b, c). Previous studies, based on comparing firing covariance values of place cells with overlapping fields between the running sessions and the post-running sleep period, imply that place cells participate in reactivations during sleep (60). The detection of awake-state reactivations of memory-encoding patterns immediately following the startling events is generally consistent with those interpretations, and further illustrates the unprecedented sensitivity of this new decoding method. Thus, the combination of large-scale recording and new decoding algorithms begins to open a door to direct visualization and quantitative measurement of network-level memory traces and their dynamic temporal evolution. 5.5. Identification of Functional Coding Units, Neural Cliques, in the CA1 Network

Our finding that the representations of startle events form tight, well-separated clusters in a low-dimensional encoding sub-space prompted us to examine in detail which neurons in the CA1 population are responsible for encoding the different events, and what essential features of the neural signals are used to accomplish that. Thus, we used agglomerative hierarchical clustering (11), a pattern classification method that can aggregate units by iteratively grouping together neurons with minimally distant responses. The clustering results reveal that the network-encoding power is actually derived from a set of functional coding units in the CA1 cell population; these units are termed neural cliques, and each is a group of neurons with similar response properties and selectivity. These neural cliques exhibited an increase in firing rate to all three types of startles (i.e., a “general startle clique”), to one type of startle (i.e., “air-blow clique,” “drop clique,” or “shake clique”), and to a subset of mixed startles (i.e., drop/ shake clique, drop/air-blow clique, or air-blow/shake clique), respectively (Fig. 7a). One can mathematically evaluate the contribution of these neural cliques to the CA1 representations by repeating the MDA analysis while sequentially adding clique members to an initial set of non-responsive neurons. For example, a random selection of 40 non-responsive cells as an initial set provides no discriminating power, yielding only overlapping representations (26). By contrast, inclusion of the ten most responsive cells from the general startle clique leads to good separation between the rest state and the startle states, but not among startle types. The selective discrimination of “drop” startle events is obtained by the addition of as few as the ten most responsive neurons from the drop clique.

Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis

119

Fig. 7. Hierarchical organization of the memory-encoding neural clique assembly. (a) Hierarchical clustering analysis of responses of 757 CA1 neurons from four mice to the three different types of startling episode. (b) Hierarchical and categorical organization of memory coding units into the feature-encoding pyramid.

Similarly, inclusion of the ten most responsive air-blow-specific clique neurons and the ten most responsive earthquake-specific clique neurons lead to full discrimination between all startle types. Thus, these neural cliques indeed constitute the basic functional units encoding the identity of different startling episodes. Through examining the overall organization of neural clique assembly involved in startle memory encoding, it is clear that the internal CA1 representation of any given startle episode involves a combination of neural cliques, invariably consisting of the general startle clique, a sub-general startle clique, a startle identityspecific clique, and a context-specific startle clique. Thus, each clique assembly is organized in a categorical hierarchy forming a “feature-encoding pyramid” (Fig. 7b). The neural clique representing or extracting the most general features (common to all categories of startle event) is at the bottom, followed by neural cliques responding to less-general features (covering multiple, but not all, common categories). Moving gradually up the cliques respond to increasingly specific and discriminating (responding to a specific category), and the most discriminating feature

120

Kuang and Tsien

clique (corresponding to context-specificity) is at the top of the feature-encoding pyramid. This invariant feature-encoding pyramid of neural clique assemblies reveals basic principles for the organization of memory encoding in the brain (26). 5.6. Converting the Activation Pattern of Neural Clique Assembly into a Real-Time Binary Code

The MDA-encoding sub-space we have examined so far provides an efficient separation of the startle episodes based on the recorded ensemble activity patterns. However, in contrast to the specificity exhibited by neural cliques, this MDA sub-space does not show event-selectivity along any of its discriminant axes. To translate the ensemble responses into an event-selective encoding coordinate system, we assigned new positions for the cluster centers so that they are linearly mapped into a clique-space, where each axis directly corresponds to a particular clique, thus projecting specific activation patterns to 1 and the absence of activation to 0. This mathematical operation achieves the reorientation of the main axes of the low-dimensional encoding sub-space by inverting the matrix containing the centers of startle representations in that space. By projecting the ensemble patterns directly into this cliquespace, the recorded neural activities are now mapped onto a set of highly reproducible and selective responses. Each clique-space projection vector (columns in Fig. 8a) is clearly selective to a specific combination of startling episodic events, and does not respond to additional features. In addition, this selectivity also extends to the representation of contexts. Furthermore, the weights in the projection vectors strongly correlate with the responsiveness of neurons in the corresponding clique (data not shown). As a result, the activated cliques can be directly detected by using simple threshold crossing and, consequently, their collective identity uniquely codes for any given startle. For example, based on a predefined sequence of clique assembly (general-startle/air blow/ drop/shake/air sub-context/drop sub-context), the activation code 110010 corresponds to the internal representation of the air blow in context A, 110000 to air blow in context B, 101000 to drop elevator A, 101001 to drop elevator B, and 100100 to shake. Importantly, these binary codes can be dynamically implemented to detect the occurrence of the internal representations of startling episodes. For example, using the thresholded responses of these cliques, we can compute the “hit ratio” for correctly matching activation patterns with the binary codes. Identification of neural cliques as the functional coding units for internal representation has prompted us to further look into the robustness of real-time encoding by neural cliques. This finding is a particularly important issue, as it is well known that a single neuron often shows large variations in both spike discharge and inter-spike intervals in response to repetition of identical stimuli (1, 13, 23). We found that these individual members within each clique fired tightly together in close temporal

Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis

121

Fig. 8. Conversion of activation patterns of neural clique assembly into real-time binary codes. Neural responses were weighted by using a remapping procedure and smoothed with a Gaussian filter (s = 100 ms) as shown on the Left (±1 standard deviation confidence intervals are shown in corresponding colors). Rows correspond to the different startling episodes, whereas columns indicate the different neural cliques (general startle, air blow, drop, shake, air blow context A, and drop context B). The binary activation patterns corresponding to each event can be mathematically converted to a set of binary codes (bottom right, after the defined sequence of the cliques). As such, the clique activation codes are as follows: 110010 for air blow in context A, 110000 for air blow in context B, 101000 for drop in elevator A, 101001 for drop in elevator B, and 100100 for shake. Sparse membership distributions of CA1 neural cliques are illustrated (top left).

proximity during startle episodes. This collective co-spiking feature allowed the neural cliques to overcome the response variability of their individual members and, thus, to achieve real-timeencoding robustness. For example, the neural clique formed by the 10 drop-neurons consistently produced robust response to drops, but not to air blow or shake events (26). Therefore, the co-spiking of clique neurons is capable of providing a network-level mechanism for real-time-encoding robustness and can act as a robust internal timer to reliably signal the occurrence of external events.

6. Technical Considerations and Future Perspective

The ability to monitor the real-time activity patterns of large numbers of individual neurons in freely behaving animals is beginning to provide crucial insights into how the brain encodes memory (27, 49). Over the past decade, the development and application of molecular biology and genetics have made mice an

122

Kuang and Tsien

ideal model organism to study the molecular and neural basis of cognitive behaviors (54). However, the network-level analysis of neural mechanisms of cognitive behaviors has lagged behind, largely due to technical difficulties. The small size of mice makes it difficult for a systematic investigation of activity patterns of neural networks in vivo. We developed a high-density ensemble-recording technique with up to 128 recording channels that can be formatted as single electrodes, stereotrodes, or tetrodes (25, 26). This high-density recording array is capable of recording from over 200 individual neurons simultaneously in the hippocampus of freely behaving mice. This large-scale in vivo ensemble-recording technique, once it is coupled with mouse genetics, should be valuable to the study of the complex relationship between genes, the neural network, and cognitive behaviors (5, 53). Although the described design is currently made for recording in the mouse CA1 region, the basic construction can be easily customized and modified to fit the specific need for recording in other brain sites. Importantly, our microdrive system can be formatted in single, stereotrode, or tetrode format. While the tetrode format may offer the best separation of single units, the stereotrode format is still capable of achieving reliable separations of multi-units on most recording electrodes. On average, we can record and separate 5–7 units per stereotrode (per site) in the hippocampus. This is highly consistent with the number of recorded units in the rat hippocampus using stereotrodes as reported by others (30, 32–34). Thus, given the same number of channels available the tetrode format would lead to a 50% reduction in the numbers of recording sites in the CA1 region in comparison to the stereotrode format. A major advantage of tetrodes comes when the potentially recorded cells are located within equal distance of the two recording channel tips of the stereotrode and happen to exhibit the same waveform characteristics. Under this unique scenario, the stereotrode would then have difficulty in discriminating them. By using two additional channels (relying on the differences in spatial location), a tetrode can differentiate between the waveforms of these “identical cells” (15, 16). Based on our examinations of 99 single units measured by the tetrodes, as long as the amplitudes of the detected waveform by the two channels (stereotrode format) are sufficiently above the noise basal level, the stereotrodes can provide highly comparable isolation with 100% accuracy. In addition, we have also assessed a second unusual scenario in which recorded cells with identical waveforms happen to be located in the middle-line “plane” of the two recording channel tips of the stereotrode. We have calculated the occurrence probability of such cases where middle-line “plane” cells were unable to be resolved by stereotrodes as constituting only a tiny fraction (about 1.5% of our recordings in such cases, those ambiguous

Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis

123

units were removed from further analysis). Although the vast majority of the recorded cells usually are not located on the middleline “plane” between two recording tips of the stereotrode (primarily because the CA1 pyramidal cell layer typically consists of only 2–3 rows), we did indeed observe those middle-line “plane” cells on some occasions. While it is still possible to distinguish them based on the subtle differences in peak amplitudes and wave shapes, in our practice we simply discard them, thereby only selecting the four well-separated single units from this stereotrode for further analysis. Therefore, using our above stringent criteria during the spike-sorting procedures, we can ensure that all units used for subsequent data analysis are well separated with high confidence. It is noteworthy that although our current microdrive is relatively lightweight, the connecting cables carrying 96 or 128 wires to the data acquisition systems can drastically increase the total weight, thereby restraining free movement of the animals. To solve this problem, we currently use a helium-filled Mylar balloon tied to the cables to alleviate the weight of the apparatus and cables. This method is simple and efficient and enables the mouse to move freely during recording. In the near future, it might be desirable to further combine our microdrive system with a miniature telemetry system, which can transmit signals via radio frequency in a wireless fashion. Indeed, such a device was recently reported (6); see the digital telemetry system described in this volume (Fenton, Jeffery and Donnett, Chap. 4. Neural recording using digital telemetry.). As an example of the usefulness of this large-scale ensemblerecording technique, we described our recent experiments in which we decoded and visualized the CA1 network-level ensemble patterns in response to startling episodic experience. Our analysis allowed us to identify network-level functional coding units that are capable of overcoming the response variability of individual neurons and achieving real-time network representation of startling episodic experiences. It would be of interest to investigate how the individual neurons that comprise a functional coding clique are anatomically connected and how they are modulated by synaptic plasticity (29, 47, 57), as well as to further dissect to what extent they reflect the sensory, emotional, mnemonic, or even conceptual aspects of the events (3, 21). Our identification of network-level functional coding units, termed neural cliques, in the hippocampus also allows us to suggest that any given episodic event is represented and encoded by the activation of a set of neural clique assemblies which are organized in a categorical and hierarchical manner (27, 49, 50). This hierarchical feature-encoding pyramid invariably consisted of the general feature-encoding clique at the bottom, sub-general feature-encoding cliques in the middle, and highly specific

124

Kuang and Tsien

feature-encoding cliques at the top. This hierarchical and categorical organization of neural clique assemblies provides the networklevel mechanism capable of not only achieving vast storage capacity, but also generating commonalities from the individual behavioral episodes and converting them to the abstract concepts and generalized knowledge that are essential for intelligence and adaptive behaviors (49). Furthermore, activation patterns of the neural clique assemblies can be mathematically converted to strings of binary codes that would permit universal categorizations of the brain’s internal representations across individuals and species. It is conceivable that such universal brain codes can also potentially allow for unprecedented brain–machine interface communications. References 1. Abbott LF, Dayan P (1999) The effect of correlated variability on the accuracy of a population code. Neural Comput 11:91–101. 2. Alonso A, Garcia-Austt E (1987) Neuronal sources of theta rhythm in the entorhinal cortex of the rat. I. Laminar distribution of theta field potentials. Exp Brain Res 67:493–501. 3. Berger TW, Alger B, Thompson RF (1976) Neuronal substrate of classical conditioning in the hippocampus. Science 192:483–485. 4. Buzsaki G, Rappelsberger P, Kellenyi L (1985) Depth profiles of hippocampal rhythmic slow activity (‘theta rhythm’) depend on behaviour. Electroencephalogr Clin Neurophysiol 61:77–88. 5. Cao X, Wang H, Mei B, An S, Yin L, Wang LP, Tsien JZ (2008) Inducible and selective erasure of memories in the mouse brain via chemical-genetic manipulation. Neuron 60:353–366. 6. Chien CN, Jaw FS (2005) Miniature telemetry system for the recording of action and field potentials. J Neurosci Methods 147:68–73. 7. Cohen NJ, Eichenbaum H (1993) Memory, amnesia, and the hippocampal system. Cambridge, MA: MIT. xii, 330 p. p. 8. Deadwyler SA, Bunn T, Hampson RE (1996) Hippocampal ensemble activity during spatial delayed-nonmatch-to-sample performance in rats. J Neurosci 16:354–372. 9. Disterhoft JF, Coulter DA, Alkon DL (1986) Conditioning-specific membrane changes of rabbit hippocampal neurons measured in vitro. Proc Natl Acad Sci U S A 83:2733–2737. 10. Dragoi G, Harris KD, Buzsaki G (2003) Place representation within hippocampal networks is modified by long-term potentiation. Neuron 39:843–853.

11. Duda RO, Hart PE, Stork DG (2001) Pattern classification. New York: Wiley. xx, 654 p. p. 12. Eichenbaum H, Dudchenko P, Wood E, Shapiro M, Tanila H (1999) The hippocampus, memory, and place cells: is it spatial memory or a memory space? Neuron 23:209–226. 13. Fenton AA, Muller RU (1998) Place cell discharge is extremely variable during individual passes of the rat through the firing field. Proc Natl Acad Sci U S A 95:3182–3187. 14. Fox SE, Ranck JB, Jr. (1981) Electrophysiological characteristics of hippocampal complex-spike cells and theta cells. Exp Brain Res 41:399–410. 15. Gray CM, Maldonado PE, Wilson M, McNaughton B (1995) Tetrodes markedly improve the reliability and yield of multiple single-unit isolation from multi-unit recordings in cat striate cortex. J Neurosci Methods 63:43–54. 16. Harris KD, Henze DA, Csicsvari J, Hirase H, Buzsaki G (2000) Accuracy of tetrode spike separation as determined by simultaneous intracellular and extracellular measurements. J Neurophysiol 84:401–414. 17. Hebb DO (1949) The organization of behavior; a neuropsychological theory. New York: Wiley. xix, 335 p. p. 18. Hedou G, Mansuy IM (2003) Inducible molecular switches for the study of long-term potentiation. Philos Trans R Soc Lond B Biol Sci 358:797–804. 19. Hoffman KL, McNaughton BL (2002) Coordinated reactivation of distributed memory traces in primate neocortex. Science 297:2070–2073. 20. Kida S, Josselyn SA, de Ortiz SP, Kogan JH, Chevere I, Masushige S, Silva AJ (2002)

Large-Scale Neural Ensembles in Mice: Methods for Recording and Data Analysis

21. 22. 23.

24.

25.

26.

27.

28.

29. 30.

31.

32.

33.

CREB required for the stability of new and reactivated fear memories. Nat Neurosci 5:348–355. Koch M (1999) The neurobiology of startle. Prog Neurobiol 59:107–128. LeDoux JE (1994) Emotion, memory and the brain. Sci Am 270:50–57. Lestienne R (2001) Spike timing, synchronization and information processing on the sensory side of the central nervous system. Prog Neurobiol 65:545–591. Leutgeb S, Mizumori SJ (1999) Excitotoxic septal lesions result in spatial memory deficits and altered flexibility of hippocampal single-unit representations. J Neurosci 19:6661–6672. Lin L, Chen G, Xie K, Zaia KA, Zhang S, Tsien JZ (2006) Large-scale neural ensemble recording in the brains of freely behaving mice. J Neurosci Methods 155:28–38. Lin L, Osan R, Shoham S, Jin W, Zuo W, Tsien JZ (2005) Identification of networklevel coding units for real-time representation of episodic experiences in the hippocampus. Proc Natl Acad Sci U S A 102:6125–6130. Lin L, Osan R, Tsien JZ (2006) Organizing principles of real-time memory encoding: neural clique assemblies and universal neural codes. Trends Neurosci 29:48–57. Mack V, Burnashev N, Kaiser KM, Rozov A, Jensen V, Hvalby O, Seeburg PH, Sakmann B, Sprengel R (2001) Conditional restoration of hippocampal synaptic potentiation in GlurA-deficient mice. Science 292:2501–2504. Malenka RC, Nicoll RA (1999) Long-term potentiation – a decade of progress? Science 285:1870–1874. McEchron MD, Weible AP, Disterhoft JF (2001) Aging and learning-specific changes in single-neuron activity in CA1 hippocampus during rabbit trace eyeblink conditioning. J Neurophysiol 86:1839–1857. McHugh TJ, Blum KI, Tsien JZ, Tonegawa S, Wilson MA (1996) Impaired hippocampal representation of space in CA1-specific NMDAR1 knockout mice. Cell 87:1339–1349. McNaughton BL, O’Keefe J, Barnes CA (1983) The stereotrode: a new technique for simultaneous isolation of several single units in the central nervous system from multiple unit records. J Neurosci Methods 8:391–397. Moita MA, Rosis S, Zhou Y, LeDoux JE, Blair HT (2003) Hippocampal place cells acquire location-specific responses to the conditioned stimulus during auditory fear conditioning. Neuron 37:485–497.

125

34. Moita MA, Rosis S, Zhou Y, LeDoux JE, Blair HT (2004) Putting fear in its place: remapping of hippocampal place cells during fear conditioning. J Neurosci 24:7015–7023. 35. O’Keefe J, Nadel L (1978) The hippocampus as a cognitive map. New York: Clarendon. 570 p. 36. Rampon C, Tang YP, Goodhouse J, Shimizu E, Kyin M, Tsien JZ (2000) Enrichment induces structural changes and recovery from non-spatial memory deficits in CA1 NMDAR1-knockout mice. Nat Neurosci 3:238–244. 37. Ranck JB, Jr. (1973) Studies on single neurons in dorsal hippocampal formation and septum in unrestrained rats. I. Behavioral correlates and firing repertoires. Exp Neurol 41:461–531. 38. Rudy JW, Barrientos RM, O’Reilly RC (2002) Hippocampal formation supports conditioning to memory of a context. Behav Neurosci 116:530–538. 39. Sara SJ (2000) Retrieval and reconsolidation: toward a neurobiology of remembering. Learn Mem 7:73–84. 40. Schmidt EM (1999) Electrodes for many single neuron recordings. In: Nicolelis MA, editor. Methods for neural ensemble recordings. Boca Raton, FL: CRC. p 1–23. 41. Schmitzer-Torbert N, Jackson J, Henze D, Harris K, Redish AD (2005) Quantitative measures of cluster quality for use in extracellular recordings. Neuroscience 131:1–11. 42. Scoville WB, Milner B (1957) Loss of recent memory after bilateral hippocampal lesions. J Neurol Neurosurg Psychiatry 20:11–21. 43. Shimizu E, Tang YP, Rampon C, Tsien JZ (2000) NMDA receptor-dependent synaptic reinforcement as a crucial process for memory consolidation. Science 290:1170–1174. 44. Squire LR (1987) Memory and brain. New York: Oxford University Press. xii, 315 p. p. 45. Tang YP, Shimizu E, Dube GR, Rampon C, Kerchner GA, Zhuo M, Liu G, Tsien JZ (1999) Genetic enhancement of learning and memory in mice. Nature 401:63–69. 46. Thompson RF (2005) In search of memory traces. Annu Rev Psychol 56:1–23. 47. Tsien JZ (2000) Building a brainier mouse. Sci Am 282:62–68. 48. Tsien JZ (2000) Linking Hebb’s coincidencedetection to memory formation. Curr Opin Neurobiol 10:266–273. 49. Tsien JZ (2007) The memory code. Researchers are closing in on the rules that the brain uses to lay down memories. Discovery of this memory code could lead to the design

126

50. 51.

52.

53.

54. 55.

Kuang and Tsien of smarter computers and robots and even to new ways to peer into the human mind. Sci Am 297:52–59. Tsien JZ (2007) Real-time neural coding of memory. Prog Brain Res 165:105–122. Tsien JZ, Chen DF, Gerber D, Tom C, Mercer EH, Anderson DJ, Mayford M, Kandel ER, Tonegawa S (1996) Subregion- and cell typerestricted gene knockout in mouse brain. Cell 87:1317–1326. Tsien JZ, Huerta PT, Tonegawa S (1996) The essential role of hippocampal CA1 NMDA receptor-dependent synaptic plasticity in spatial memory. Cell 87:1327–1338. Wang H, Feng R, Phillip Wang L, Li F, Cao X, Tsien JZ (2008) CaMKII activation state underlies synaptic labile phase of LTP and short-term memory formation. Curr Biol 18:1546–1554. Wang H, Hu Y, Tsien JZ (2006) Molecular and systems mechanisms of memory consolidation and storage. Prog Neurobiol 79:123–135. Wang H, Shimizu E, Tang YP, Cho M, Kyin M, Zuo W, Robinson DA, Alaimo PJ, Zhang C, Morimoto H, Zhuo M, Feng R, Shokat KM, Tsien JZ (2003) Inducible protein knockout reveals temporal requirement of CaMKII reactivation for memory consolidation in the brain. Proc Natl Acad Sci U S A 100:4287–4292.

56. Weiss C, Kronforst-Collins MA, Disterhoft JF (1996) Activity of hippocampal pyramidal neurons during trace eyeblink conditioning. Hippocampus 6:192–209. 57. Wigstrom H, Gustafsson B (1985) On longlasting potentiation in the hippocampus: a proposed mechanism for its dependence on coincident pre- and postsynaptic activity. Acta Physiol Scand 123:519–522. 58. Wills TJ, Lever C, Cacucci F, Burgess N, O’Keefe J (2005) Attractor dynamics in the hippocampal representation of the local environment. Science 308:873–876. 59. Wilson MA, McNaughton BL (1993) Dynamics of the hippocampal ensemble code for space. Science 261:1055–1058. 60. Wilson MA, McNaughton BL (1994) Reactivation of hippocampal ensemble memories during sleep. Science 265: 676–679. 61. Wirth S, Yanike M, Frank LM, Smith AC, Brown EN, Suzuki WA (2003) Single neurons in the monkey hippocampus and learning of new associations. Science 300: 1578–1581. 62. Wong RW, Setou M, Teng J, Takei Y, Hirokawa N (2002) Overexpression of motor protein KIF17 enhances spatial and working memory in transgenic mice. Proc Natl Acad Sci U S A 99:14500–14505.

Chapter 6 Behavioral Correlates of Neuronal Activity Recorded as Single-Units: Promises and Pitfalls as Illustrated by the Rodent Head Direction Cell Signal Robert W. Stackman Jr. Abstract The purpose of this chapter is to provide an overview of a current approach in defining the relationships between the firing patterns of groups of neurons recorded from the freely behaving rodent. The design and construction of a 16-channel headstage and its typical integration with a commercially available multichannel data acquisition system is described in detail. Next, the chapter examines the strategies necessary to test the behavioral significance of firing patterns of individual neurons recorded from behaving animals. A discussion is included that surveys the limitations and advantages inherent in testing how discrete behavioral events are represented in single-unit activity recorded from rodents engaged in spatial and nonspatial behaviors. To expand on the technological and theoretical approaches to establishing the relationship between single-unit activity and behavior, the rodent head direction cell system is described as one useful model system. The head direction cell is considered by many to represent the directional sense of the organism and to be essential for spatial navigation. The chapter provides a critical overview of the studies that have been conducted to date to test this relationship. In short, these are the record while the rodent performs some navigational task experiments. Examination of the results from the studies reveals that there is little clear evidence to support the view that head direction cells guide spatial navigation. The ensuing discussion addresses putative reasons for the division between the empirical data and the theory of head direction cell – behavior relations. The chapter concludes with clear suggestions for alternative experimental approaches that might better address brain–behavior relationships. A plea for increased open dialogue between the fields of behavioral analysis and systems/cellular neurophysiology emerges. Improved understanding of the contributions to the neuroscience field made by behavioral neuroscientists and those by electrophysiologists should raise the bar of modern approaches to brain–behavioral relationship studies using in vivo electrophysiological methods. Key words: Single-unit, Neuronal ensemble, Spatial cognition, Anterior thalamus, Postsubiculum, Hippocampus, Place cell, Head direction cell, Mouse, Rat, Spatial navigation, Video-tracking, Behavioral correlate

Robert P. Vertes and Robert W. Stackman Jr. (eds.), Electrophysiological Recording Techniques, Neuromethods, vol. 54, DOI 10.1007/978-1-60327-202-5_6, © Springer Science+Business Media, LLC 2011

127

128

Stackman Jr.

1. Introduction 1.1. Brief Overview of the Study of Behavioral Correlates of Single-Unit Activity

The purpose of this chapter is to present a description of current methodology used to examine the behavioral significance of singleunit activity recorded from freely moving rodents. This is of course a remarkably broad area of interest; therefore, the focus here will be restrained to the neuronal activity that represents spatial information necessary for such behaviors as spatial memory and navigation. Indeed, the most commonly employed neurophysiological strategy has been to record individual neuronal activity from the rats or mice, as they freely move about open fields or mazes. This strategy has led to some of the most exciting demonstrations of the brain’s representation of spatial information namely, the discovery of place cells in the hippocampus by John O’Keefe in 1971. Place cells are principal neurons of the hippocampus that discharge at a peak rate when the rodent occupies some particular circumscribed location of the floor surface of a maze or open field. The place field, or the area of the maze, where the place cell exhibits its highest firing rate, is influenced by a variety of extra-maze and intra-maze cues and by self-motion cues. The full report of the discovery of hippocampal place cells came complete with an exquisite treatise on their place within the broader context of the role of the hippocampus in learning and memory, John O’Keefe and Lynn Nadel’s, The hippocampus as a cognitive map (42). The discovery of place cells made such an impression on the field of behavioral neuroscience that several laboratories employing similar techniques turned their attention to spatial information as a guide to brain–behavior relationships. Certainly, the determination of head direction cells in the dorsal presubiculum by Jim Ranck in 1984, and the more recent discovery of grid cells in the entorhinal cortex by Edvard and May-Britt Moser and colleagues followed the approach originated by John O’Keefe. In all of these phenomenological descriptions, the spatially tuned properties of these respective neurons revealed themselves by discharging at specific locations and/or directional headings as the rodent moved about an open field. On the basis of the behavioral correlates reported for neurons recorded from a number of regions comprising the hippocampal formation (hippocampus proper, dentate gyrus, subiculum, and entorhinal cortex), and associated regions of the limbic system (cingulate cortex, anterior thalamus, mammillary nuclei), it would seem that these place cells, head direction cells, and grid cells would naturally be integral components of a neural network supporting spatial navigation. Although this hypothesis is certainly straightforward, the results of experiments testing it are not. This chapter includes discussion of issues related to the functional significance of well-defined behavioral correlates of single-unit

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

129

activity recorded from freely behaving rodents. Observing these spatially tuned neurons discharge during recording sessions can be awe-inspiring. Yet when one asks the intriguing but simple question: why a place cell fires where it does issues such as the one I raised above come to the foreground. The firing properties of place cells, head direction cells, and now grid cells have for the most part been determined during recordings in which rodents moved freely about a very familiar enclosure. These free-running sessions place little demand on the animal and so it is unclear why an extended limbic circuit might even be concerned with “keeping track of the animal’s position” under these circumstances. Simply, it is unclear what the motivation would be for hippocampal or anterior thalamic neurons (e.g.) to represent the momentto-moment trajectory or directional heading of the rodent as it walks around an enclosure it has been exposed to many times before and in which the delivery of food is not generally contingent on the position of the animal. This argument will be fully developed in the latter portion of this chapter.

2. Construction of Microelectrode Arrays for Recording Multiple Channels of Single-Unit Activity Data from Freely Moving Mice 2.1. Design Considerations

The advent of molecular biological technology merging with neuroscience has provided exquisite tools for testing specific hypotheses regarding the molecular mechanisms that underlie complex behaviors. Genetically engineered mice that contain region-specific or experimentally controlled promoters to knockout or overexpress a given gene coding for a specific protein offer great promise for the field of brain–behavioral relationships (12). Although the bulk of single-unit recordings have been conducted in rats, technology has proven fairly easy to modify for implementation in mice. Doing so though is an important consideration – simply the availability of mice should not be perceived as reason enough to retool one’s laboratory to shift from rat to mouse. However, there are clear advantages to the use of genetically engineered mice for testing specific hypotheses that may not be approachable with pharmacological agents. Having decided to pursue single-unit recording in mice, one finds that there are numerous options available now for lightweight miniature headstages and advance-able microelectrode arrays that can be chronically implanted on to the mouse skull for subsequent recordings during behavior. The microelectrode array that we have begun to use is essentially a modification of John Kubie’s drive-able 10-wire microelectrode array designed for rats (23). This microelectrode array can be configured as a 16-channel system (16 single wires, 8 stereotrodes, or 4 tetrodes) depending on the region of interest. The advantages of this design are that it requires fairly routine

130

Stackman Jr.

procedures, readily modifiable and modular, and uses economical components. Being that our design is a modified Kubie-style microdrive, this model comprises tetrodes, stereotrodes, or a bundle of single wires that are passed through a cannula and positioned above a region of interest. Turning the drive screws advances the bundle of wires in unison through the brain tissue, thereby limiting the design a bit. Those interested in microdrives with tetrodes that can be individually positioned in mouse brain regions are encouraged to refer the chapter in this volume by Kuang and Tsien (Chap. 4: Large Scale Ensemble Recording in Mice and Neural Data Analysis). Our modified Kubie drive is built on a Mill-Max double row Amphenol connector array. A typical bilateral configuration might be composed of 2 tetrodes implanted dorsal to CA1 region of each hippocampus or 4 stereotrodes implanted dorsal to the anterodorsal thalamic nuclei in each hemisphere. 2.2. Microwire Bundles

Stereotrodes (31) (or tetrodes (15, 52)) are constructed from one piece of 25-mm Nichrome wire (Stablohm 675, California Fine Wire, cut to 30-cm length) (or 2 pieces of 13-mm Nichrome wire, each 30 cm in length). Care should be taken to ensure that the workspace is clean of dust, oils, and debris prior to beginning work with the fine-electrode wire. One should also limit as much as possible the manipulation of the Stablohm 675 wire as it is easily kinked and broken. Handling the wire using Dumont #5 finetipped forceps limits premature breakage of the wire and eliminates transmitting oils and debris from fingers to the fine wire. Dedicating a sharp pair of fine iris scissors for cutting the California fine wire also ensures clean cuts without kinking the wire or abrading the insulation. Once cut, the 30-cm length(s) of wire is looped and the two ends brought together under a small piece of paper tape. The paper-taped ends of the wire are then clamped in the tips of a small mosquito forceps. The loop formed by the folded wires is gently placed over a suspended smooth rod, and the mosquito forceps clamp is allowed to hang straight down. The clamped wire ends are then twisted either by gently spinning the mosquito forceps by hand or by using a magnetic stirring plate that can achieve a very slow turning speed (~1 rev/s) to achieve approximately 50 turns for a 30-cm electrode wire. Once twisted sufficiently, the bundle should be allowed to untwist for 20 revolutions or so. The twisted bundle is then heated by passing a heat gun in an up and down motion along its length. The goal is to bond the plastic insulation of the two or four wires together so that the bundle does not untwist. Three or four passes with the warm heat gun from varying angles is generally sufficient to bond the wires of the tetrode. It is also possible to bond the wires by running a bead of Krazy Glue along the length of the twisted wires. The tetrode can now be removed from the clamp

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

131

and rod, trimmed and stored away until ready for loading into the microdrive. If the microelectrode array is to be composed of 16 single wires, then these can be cut at the requisite length just prior to loading the microdrive. 2.3. Microdrive Design and Assembly

The Mill-Max double-row 50-pin connector strip (Mill-Max Manufacturing Corp., Oyster Bay, NY; Product number: 853-43100-10-001000, Interconnect, 0.050 Grid Straight Socket) is cut down to a double row 9-pin (18 total pins). Cutting the desired length of amphenol strip is facilitated by holding the 50-pin connector strip, counting off nine columns of pins, and then pushing out the pins of the adjacent tenth column. The narrow kerf of the blade of a fine jewelers saw or coping saw can then easily cut through the amphenol. The pins on the resulting Mill-Max connector strip to be designated as ground or reference are removed, and the male ends of the remaining pins are cut down to ~50% of their original length with a stone disk cut-off wheel powered by a Dremel motor tool. Shortening the length of the pins will facilitate the wrapping of electrode wires in a later step. Examine the shortened ends of the pins with a dissecting microscope and remove any burrs by burnishing the ends with a pair of Dumont #5 forceps with slightly roughened tips. Next, the reference pins are replaced into the Mill-Max connector.

2.4. Microelectrode Carrier

This next step is to build the cannulae that will hold the electrode wires. The cannulae are cut from stainless steel hypodermic tubing (Small Parts, Miami Lakes, FL); the gauge of which depends on how many microelectrodes the device will carry. Hypodermic tubing from Small Parts that is 26-gauge can easily hold eight strands of 25-mm Nichrome electrode wire (i.e., 2 tetrodes, 4 stereotrodes, or 8 single electrode wires). One might want to experiment with smaller diameter stainless steel tubing or tubing of other material such as micro-diameter polyimide tubing. However, since our typical design is to build the cannulae from 26-gauge hypodermic stainless steel tubing that is what is described here. Cut two 12-mm lengths of 26-gauge by scoring tubing with a fresh razor blade or cutting it with sharp wire cutters. With a medium and then a fine grit ruby stone (Small Parts, Miami Lakes, FL) remove any burrs from the tubing ends, trim, and polish to final length. Cut the male ends off of the two sacrificed Mill-Max pins and pass the length of the pin along the surface of the medium ruby stone to form a slightly flattened side on both pins. Next, solder the end of one of the finished 26-gauge stainless steel tubing pieces to the modified Mill-Max pin. The desired position of the solder joint connecting the pin and tubing should be adjusted before soldering so that the female end of each Mill-Max pin extends parallel to the tubing and beyond the end of the tubing by a 1.5 mm or so (see Fig. 1a). Repeat this

132

Stackman Jr.

Fig. 1. The components of a lightweight 16-channel microelectrode array for chronic recording from freely moving C57BL6J mice. (a) This panel depicts the 26-gauge stainless steel cannula tubing after it has been soldered to the modified Mill-Max pin. The arrowhead indicates the female end of the Mill-Max pin. The asterisk indicates the length of polyethylene tubing (PE-50) that covers the cannula tubing and extends beyond the end of the cannula. (b) This panel depicts a side view of the Mill-Max double-row 18-pin connector (indicated by the white asterisk) loaded with two cannulae each containing two tetrode arrays. The cannulae are each covered with a length of PE-50 tubing to protect the fine microelectrode wire projecting beyond the tips of the cannulae (indicated by the black arrowhead). (c) This panel depicts the tooled white Teflon sheet (left side) that will comprise a platform to support the Mill-Max connector and the two drive screws. A notch (indicated by the arrowhead) has been cut into the Teflon sheet to fit the finished Mill-Max connector array. The right side of this panel depicts one of the drive screws. The threads of the screw have been turned into the blue shrink tubing cuff (indicated by the white asterisk). This microdrive design represents a modification of that originally presented by (23). Kubie’s design incorporated three drive screws in a triangular configuration around a central connector. The present design uses two screws to reduce overall weight of the device. (d) A male C57BL/6J mouse with a 16-channel microelectrode/microdrive array implanted over the anterodorsal thalamic nuclei. This particular array contains 4 tetrodes (2 per cannula). The female ends of the Mill-Max pins can be observed in this view just prior to connection to the running recording cable. Note the two drive screws on the top of the white Teflon that have been turned to advance the microelectrodes into the brain. This mouse is 60 days postoperative.

solder procedure with the second tubing piece and modified MillMax pin. Verify that the 26-gauge tubing is open and clean of any debris by sonicating or passing a 32-gauge stainless steel wire through the tubing lumen. The 26-gauge tubing will comprise the cannulae and the end opposite the soldered pin will extend into the brain. It is critical that those ends are carefully polished with the fine ruby stone before proceeding.

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

2.5. Loading Microelectrode Wires

133

The next step is to load the tetrodes, stereotrodes, or single wires into the 26-gauge stainless steel tubing soldered to the Mill-Max pin. Cut a length of polyethylene tubing (i.e., Intramedic PE50 or equivalent) that is ~1 cm longer than the 26-gauge stainless steel cannula tubing and then push the PE50 tubing over the end of the cannula tubing (see asterisk in Fig. 1a). The PE50 tubing will act as a sheath over the cannula tubing and protect the microelectrode wires that project from the tip of the cannula tubing after they are properly loaded. Each tetrode, stereotrode, or the bundle of eight single wires (cut to the desired length) can be passed through the cannula tubing beginning at the end nearest the soldered pin. To avoid damaging the microelectrode wires, it is best to clamp the cannula tubing (+ sheath) with the soldered pin end pointing up and then load the microelectrode wires into the cannula by viewing the work under a dissecting microscope. Grasp the tetrodes, stereotrodes, or single wires with Dumont #5 forceps and slowly pass the microelectrode wire through the cannula until the microelectrode wires project beyond the tip of the cannula tubing. Be sure to leave at least 2–3 cm of tetrode (or stereotrode/electrode) wire on the side of the cannula tubing soldered to the Mill-Max pin. These remaining lengths of microelectrode wires should next be permanently fixed in place by applying a drop of 5-min epoxy cement, or gel Krazy Glue, to the top of the cannula tubing where the electrode wires emerge. Once the epoxy is set, the untwisted ends of the tetrodes, stereotrodes, or single wires are brought into electrical contact with the dedicated pins of the Mill-Max double-row connector. These procedures are similar to that described in a number of papers and so will only be described briefly here. The cannula containing the tetrodes (or other electrode wires) is pushed down over one of the full length Mill-Max pins. Caution must be taken to avoid crimping or kinking the electrode wires when the cannula is positioned on the Mill-Max connector. The electrode wire ends are trimmed to an appropriate length, and then stripped of insulation by passing the wire over the flame of an alcohol burner or by gentle abrading with the roughened tips of a Dumont #5 forceps. The stripped end of each electrode wire is then wrapped around the respective Mill-Max pin and secured with conductive silver print paint (GC Electronics, Rockford, IL). Care should be taken to ensure that wire is sufficiently coiled around the pin. Excess lengths of electrode wire not coiled around the respective pin should be pushed into the setting silver print paint. It is critical that any excess silver print paint is scraped off of the amphenol to avoid electrical shorts between electrode pins. Alternatively, we have found that removing the Mill-Max pins, passing the end of the stripped electrode wire up through the hole of the Mill-Max connector strip, and then pushing the Mill-Max pin back through the hole effectively traps the electrode wire and establishes a very

134

Stackman Jr.

good electrical contact. A working length of insulated copper ground wire is soldered to the ground pin on the Mill-Max connector. Once all wires are wrapped and electrical contact verified, then a loop of scotch tape is wrapped around the Mill-Max connector to form a potting mold. The bottom of the length of tape should be flushed with the top of the Mill-Max amphenol connector (the top side contains the female ends of the incorporated connector pins). With the Mill-Max connector clamped into a vise and oriented top down (female end of Mill-Max connector pins pointing down), the electrical contacts of the electrode wires to the connector pins are potted by filling the mold formed by the scotch tape with fluid dental acrylic. Once the dental acrylic is sufficiently hardened, the tape is removed and the microelectrode-Mill-Max connector array can be tooled with a stone cut-off wheel in a Dremel tool to remove and sharp edges. Figure 1b depicts a photograph of a finished microelectrodeMill-Max connector array. 2.6. Building the Microdrive

The final step is to build the actual drive components of the microdrive. We modified our drive slightly from that originally described by (23). Drive screws are modified 1–72 fillister head slotted drive machine screws (1.27 cm length, Ref MX-017208FL-M, Small Parts, Miami Lakes, FL). A 0.7–0.8 mm length of shrink tubing is cut and fitted over the bottom of each drive screw, so that no threads project beyond the tubing. The shrink tubing is then shrunk to fit tight over the screw threads to form a cuff by passing the screw over a warm heat gun. Next, the outside surface of the shrink tubing cuff is painted with Krazy Glue or Loctite, and then dental acrylic is applied to increase the girth of the cuff (see Fig. 1c). A platform to hold the two drive screws and the microelectrode-Mill-Max connector array is next cut from a 2.5-mm thick sheet of white Teflon (PTFE). A notch (3-mm wide × 5-mm long) is cut in the Teflon to fit the Mill-Max connector. Two holes, just smaller in diameter than that of the 1-72 machine screws, are drilled through each end of the Teflon sheet. Figure 1 depicts the Teflon sheet with the notch cut out for the Mill-Max connector and with the holes drilled for the drive screws. Remove the cuffs from the drive screws and then pass the threads of each drive screw through the hole, so that the bottom of the fillister head of the drive screw rests against the Teflon sheet. Return the cuffs to the drive screws and advanced them as far as they can travel into the cuff, then back each screw out of the cuff at least 6–7 full turns. The fillister head of each drive screw should remain flush with the top of the Teflon sheet. With both the Teflon sheet and the microelectrode-Mill-Max connector array secured in a positioning clamp, epoxy the microelectrode-Mill-Max connector array into the notch of the Teflon sheet (see Fig. 1). Now the microelectrode/microdrive array is completely assembled and

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

135

should be stored safely prior to surgical implantation. The microelectrode wires are trimmed to the desired length just prior to surgery, and the wire tips are cleaned by soaking in a Betadine solution for 10 min and then rinsed with sterile water. We use standard stereotaxic surgical procedures to implant these microelectrode arrays onto the skulls of C57BL/6J mice (see Fig. 1) under isoflurane anesthesia. The microdrive array is fixed to the skull by miniature screws (0–80, 0.25″ binding head machine screws, Small Parts, Miami Lakes, FL) and dental acrylic. The mice tolerate the microelectrode arrays for extended periods of time (e.g., several months) without failure.

3. Behavioral Correlates of Single-Unit Activity from Limbic Brain Regions

3.1. Identifying Appropriate Behavioral Variables

Single-unit recording, or ensemble recordings of multiple single units together with local field potentials, from freely moving rodents represents an exquisite technique for defining the contributions of brain regions and the firing profiles of their intrinsic neurons, to distinct behavioral processes. Many laboratories that record individual neurons from the hippocampus and other limbic regions of freely moving rodents have focused on the relationship of spatial information to the firing properties of these neurons. This dominant focus relates to the tremendous volume of behavioral data indicating the functional significance of the rodent hippocampus and its allied structures to spatial memory and navigation. This focus of course ignores considerable evidence that the hippocampus of the rat and mouse is also involved in processing nonspatial information (5, 6, 19, 46, 55, 86). Notably, several laboratories have examined the relationship of nonspatial information and hippocampal single-unit activity including, the Eichenbaum lab’s focus on examining the degree to which hippocampal neurons represent memory processes required for various odor discrimination tasks (9, 27, 85); the Disterhoft lab’s focus on nonspatial correlates of hippocampal neurons recorded from rats during trace fear conditioning (30, 79); and recent efforts by the Tsien lab to define how episodic memories can be represented in ensembles of neurons recorded across several brain regions of freely moving mice ((24, 25) and Chap. 4: Large-Scale Neural Ensembles in Mice: Recording Methods and Data Analysis by Kuang and Tsien this volume). Whether spatial or nonspatial information processing is being investigated, the approach of these studies generally entails the determination of cell spiking times and rates together with explicit behavioral responses (e.g., position, arm entry, choice accuracy, dwell time, lever pressing rate, etc.). The spike activity of neurons

136

Stackman Jr.

is acquired from chronically implanted electrodes as the animal engages in the performance of a specific behavioral task. The time stamps of the neural data and behavioral measures are aligned to a synchronizing pulse. Neuronal responses are identified as being those cells active during some component of task performance and those cells that were inactive during the task. Peri-event time histograms are constructed for each cell recorded for each of the time-stamped task events or behavioral responses. Units can then be classified according to the given event characteristic occurring when that unit exhibits an increased (or decreased) discharge (e.g., unit response to reward, cue approach, goal approach, match/nonmatch, delay, correct response, etc.). It is then relevant to verify that task-related responses of individual neurons are not present when the same animal performs a control task or during some baseline assessment period before the trial is initiated. The firing rates of the ensemble of recorded neurons with likebehavioral correlates can be determined for a given task event episodes as well. The spike count for each cell of the ensemble is averaged and then converted to a z-score. Normalized z-scores for each cell can then be averaged to construct peri-event spike histograms for the collection of recorded neurons classified as comprising the ensemble. Task-related responses for individual units or the ensemble can be further analyzed to determine the degree of variability in these firing patterns on a trial-to-trial basis and whether these response patterns are sensitive to the phase of learning of the animal. Often in such studies, the subset of trials in which the animal commits an error in responding offer much to the determination of whether the neuronal representation guides the animal’s choice behavior. For example, on the one hand, if task-related responses of individual neurons or the population activity are consistent across correct and incorrect trials, then it will be difficult to make the claim that the firing patterns reflect the animal’s intended choice. On the other hand, if the activity pattern produced by of these neurons is distinct for correct and incorrect trials, then one might be able to distinguish which firing features of the neurons during correct trials is absent or altered on incorrect trials. Often, however with such studies, the behavior of the experimental subjects is so accurate – at near ceiling level of performance – that incorrect trials are rare. In this case, it may be necessary to alter the task parameters to increase the error rate in order to test the altered firing patterns that might represent incorrect trials. 3.2. Experimental Design Issues to Consider

Several issues arise in designing experiments to test the behavioral correlates of single-unit neuronal activity. For instance, should recordings begin after the animal is well trained on the particular task? Or should recordings begin with the initial trial to capture the encoding of the new memory? Most of the

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

137

microelectrode-microdrive-headstage configurations currently being used in rodent work are remarkably durable. Meaning that they are tolerant to the mild forces placed on them by numerous connections and disconnections of the recording cable and by the movements of the animal while executing the behavior. Even so, if the design entails training the animal on a task that requires weeks to acquire, then it may be difficult to maintain isolation of the same set of neurons over the complete course of training to define how the behavioral correlates of the firing of those neurons are altered by learning. In this case, the clear choice would be to implant the microelectrodes after acquisition of the behavioral task. Another issue to consider is whether the particular behavioral paradigm will yield sufficient opportunities for the animal to engage in the behavior of interest. In this case, the number of trials imposed is relevant since the pattern of neuronal activity that may represent a particular behavioral response should be robust and be present each time the animal engages in that behavior. Quite simply, most of these issues relate to the concern that sufficient sampling is needed to be certain that the pattern of neuronal responses during the behavioral response of interest is replicable. 3.3. Spatial Correlates of Limbic Neuronal Activity 3.3.1. Tracking the Rodent

3.3.2. Issues Encountered When Video-Tracking LED Positions

In the case of examining spatial determinants of unit activity in rodents, the common method has been to couple the acquisition of neuronal responses with animal position data – generally acquired by video tracking one or more light-emitting diodes (LEDs) that are attached to the headstage at end of the recording cable. Robert Muller and John Kubie, of SUNY Downstate Medical Center in Brooklyn, NY, first developed the technology to incorporate position data via video-tracking with single-unit recording to facilitate their efforts to examine hippocampal place cells from rats freely exploring environments (37, 38). The Muller and Kubie approach was later augmented to track a red LED and a green LED that were fixed to a boom on the end of the recording cable, with the red LED position over the rodent’s snout and the green LED over the rodent’s shoulders. Tracking the position of at least two differently colored or sized LEDs permitted the detection of head direction in addition to head position. A digital video tracking system converts the tracked positions of the LEDs on the video input signal to X and Y position coordinates, which are then timestamped with the neuronal responses to determine whether neuronal single-unit activity is correlated with the rodent’s spatial location or directional heading. Examples of head direction x firing rate tuning curves that can be generated with the behavioral position and electrophysiological data streams are depicted in Fig. 2. The inclusion of LED video tracking offers a great advantage for identifying spatial correlates of neuronal activity. However, there are some minor issues that arise namely that the LEDs can at

138

Stackman Jr.

Fig. 2. Regional differences in head direction cell firing properties. Representative polar plots depicting direction by firing rate tuning curves for a head direction cell recorded from postsubiculum (left), anterodorsal thalamus (center), and lateral mammillary nuclei (right). Data were acquired while rats moved freely about a high-walled cylindrical environment. Note that the three polar plots have distinct radial scales for firing rate. The peak firing rate (in Hz) for the postsubiculum cell was 19.94; for the anterodorsal thalamus cell was 46.90; and for the lateral mammillary nuclei was 106.86. The preferred firing direction for the postsubiculum cell was 6°; for the anterodorsal thalamus cell was 204°; and for the lateral mammillary nuclei was 150°. The range of directional firing (in degree) for these three cells was 72, 100, and 172, respectively. Directional information content (DIC), a quantitative measure of the information in bits each spike conveys about the directional heading of the animal, was calculated for each cell as DIC = ∑pi (li/l) log2(li/l), where pi is the probability of the head pointing in the ith bin, li is the firing rate when the head is pointed within the ith bin, and l is the overall mean firing rate of the cell for all bins. An information content value of 0 indicates no relation between HD and firing rate, and a value ³1 indicates a strong relation between HD and firing rate. DIC values for these three cells were 1.49, 1.07, and 1.21, respectively.

times become obscured by the recording cable or by recesses in the experimental apparatus resulting in missed video samples. The frequency of missed video samples must be determined as an indicator of the quality of the data from the recording session. In addition, the LEDs are often tracked by an overhead camera system, limiting the resolution of behavior to essentially 2-D horizontal movements. Tracking behavior in the third dimension would require at least one additional camera and a device to switch video feed from the overhead camera to the alternativeviewing camera. We employed this multi-camera approach to determine the responses of head direction cells in rats as they moved in the vertical plane between two horizontal surfaces (90). We essentially captured single-unit activity from rats exploring the floor of an open field and then during bouts when the rats climbed to and from an elevated platform using a vertical ladder. Our two-camera system required switching cameras during recordings and video samples corresponding to the transitions between the horizontal and vertical surfaces were often lost. Current technology available from Noldus Information Technology (Ethovision XT 6.0, Wageningen, Netherlands) permits video tracking from multiple cameras simultaneously, which would solve some of the problems associated with single-unit

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

139

recording and behavioral tracking with a single overhead camera. In a study of the spatial correlates of single-units recorded from the lateral mammillary nuclei, we adapted the top–down video tracking software to detect neurons that discharged according to the pitch of the rat’s head (65). We modified our standard approach to permit a coarse detection of head pitch by evaluating differences in the apparent distance between the two tracked LEDs positioned relative to the animal’s head at 60 Hz; the LED–LED distance is reduced when the rat rears or pitches its head upward or downward. A more thorough analysis of mammillary head pitch cells would require multiple cameras (e.g., side-view and top-view) as well as additional LEDs to provide a higher resolution of head orientation in 3-D. Another more general limitation is simply that having the rodent tethered to the recording cable and tracking LEDs restricts the behavioral complexity of the task imposed and the environment. It is likely that this limitation will be eliminated with the advent of wireless digital technologies (see Chap. 3: Neural Recording Using Digital Telemetry by Fenton, Jeffrey and Donnett, this volume) that permit high-density recordings for extended time periods in a wide range of behavioral applications. The combination of wireless telemetry and multiple cameras would permit examining singleunit activity in rats engaged in more naturalistic navigational and behavioral tasks and permit the tracking of rats moving in three dimensions during neuronal recordings. As the digital telemetry technology gains acceptance in the field, one hopes that it could be further miniaturized for application to recordings of mice as well. 3.3.3. Alternatives to Video-Tracking LEDs

A number of commercial systems for acquiring single-unit neurophysiological data from freely moving animals incorporate video-tracking capability such as that from Axona Ltd (St. Albans, Herts, UK) and Neuralynx (Cheetah 5.1 and Digital Lynx systems, Bozeman, MT). In addition, several companies offer tracking packages as add-ons to their modular high-density recording systems to facilitate tracking location and directional heading of a rodent. One example is the Cineplex software system (Plexon, Inc., Dallas, TX) that is capable of simultaneous acquisition and extraction of X and Y position coordinates from tracked LEDs to determine in real-time the nose-point and centerof-head position of a rodent in an environment. In addition, independent commercial systems are available that permit the determination of a rodent’s nose-point in real-time from a video feed at a sampling rate of 30 Hz. The Ethovision XT 7.0 software package (Noldus, Information Technology, Wageningen, Netherlands) is capable of detecting the nose-point of the tracked animal, using video feed from an overhead camera and thereby determining the head direction of the animal in any environment.

140

Stackman Jr.

The Ethovision system is also capable of synchronizing the video feed data together with external analog and digital inputs such as electrophysiological data with its Physiology Integration module. Other programs such as the ObjectScan system (CleverSys, Reston, VA) and the Videotrack 3.0 package (ViewPoint Life Sciences, Montreal, Canada) offer similar capabilities but appear to be designed to automatically calculate behavior variables related to the tracked nose-point of the rodent such as the exploration of toy objects by “sniffing.” Such packages will be very helpful in determining the degree to which hippocampal and limbic neurons represent nonspatial information as well as spatial information. An interesting advantage of the stand-alone packages from Noldus, CleverSys, and ViewPoint Life Sciences is that these systems do not require the tracking of LEDs on or relative to the rodent’s head. The determination of the nose-point is made relative to tracking multiple body points to determine the head position and head direction. All three of the packages allow the user to define regions of interest or zones within the experimental arena, which then can be analyzed to determine dwell times, entry times, etc. – behavioral determinants against which neuronal firing rates can easily be examined. Depending on the behavioral task imposed during a recording session, the Ethovision XT package would appear to eliminate the need for tracking LEDs worn by the rodent, which in turn opens up many possibilities for examining the firing correlates of neurons during the performance of more complicated behaviors than simple exploration of testing arenas. Despite what appears to be a number of commercial options available to integrate unique computerized videotracking packages with high-density neurophysiological systems, there appears to be little evidence that such combinatorial approaches are being used currently. 3.3.4. Behavioral Requirements of Subjects

In any case, if one is interested in examining spatial correlates of single-unit activity, then one only needs the rodent to visit specific locations in an environment (e.g., an open field or elevated radialarm maze) to determine whether the neuron or neurons in question exhibit location-specific or direction-specific firing patterns. Rats can be encouraged to explore the entire floor surface of an environment by mildly restricting their daily food ration and then offering some highly palatable food in the recording arena. Depending on the strain of mice examined, it may not be necessary to food restrict the animals so as to encourage exploration of the floor of an open arena. We have found that non food-deprived male C57BL/6J mice will readily explore maze surfaces and arena floors in spite of being relatively familiar with these environments. Chocolate cake sprinkles can be introduced into the arena to elevate the degree of arena exploration by the mice during recording sessions. The only minor drawback to the chocolate sprinkles is

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

141

that they appear identical to the fecal boli that mice produce during recording sessions. Thus, your recording arena can be mistakenly judged as being “dirty” by well meaning observers. Regardless, the size of the arena and the activity level of the animal will tend to dictate how long a recording session must be to generate sufficient sampling of the environment for a hippocampal place cell study. Recording suspected head direction cells requires only that the rodent orient its head in 360° for sufficient sampling to be achieved; the degree of sampling of the floor surface is secondary. Indeed, one can be fairly accurate in judging the preferred firing direction of a head direction cell of a rat or mouse after just a few minutes of recording, provided the animal has turned its head through 360°. Thus, sessions in which head direction cell activity is recorded can be made of shorter duration than those of place cell activity, perhaps making the study of such cells more amenable to determining their relationship to behavioral responses in tasks with changing stimulus features. However, restricting the length of the recording session limits the sheer volume of recording data, which may affect the statistical power to draw meaningful conclusions of the relationship between the spatially tuned neurons and the animal’s behavioral responses. 3.4. Nonspatial Correlates of Limbic Neuronal Activity 3.4.1. Choice of Nonspatial Variables to Examine

In the case of nonspatial information, the task of establishing the degree of association between neuronal firing and the presentation of some behavioral stimulus or a behavior choice, for example, can be quite complicated depending on the behavioral task imposed. For instance, if one is interested in the manner in which hippocampal neuronal firing reflects the acquisition, maintenance, and subsequent extinction of a nonspatial auditory trace fear memory, then this would require recording hippocampal units before and during multiple trials of an auditory trace fear conditioning paradigm. During auditory trace eye blink conditioning, a tone conditioned stimulus (CS) is presented for some duration (e.g., 300 ms) followed by a stimulus-free (trace) interval after which a corneal air puff or weak periorbital electric shock is delivered to the animal. To associate the CS and the US across the trace interval and exhibit an eye blink conditioned response (CR) that precedes the air puff, theoretically the animal must encode the duration of the trace interval. Auditory trace conditioning of the eye blink response in rabbits, rats, and mice is dependent upon the hippocampus (60, 75, 78). As McEchron and colleagues (29, 30) demonstrated, rabbit hippocampal CA1 neurons represent the relationship between the auditory tone CS and the subsequent air puff to the eye US in a heterogeneous manner. Some cells gradually increased their firing during the trace interval after the tone but before the air puff – activity that might reflect the learning of the temporal relationship between the CS and US. Other cells appeared to decrease their activity during the trace

142

Stackman Jr.

interval; and other cells appeared to exhibit learning-induced increases in response to the air puff US. However, it is difficult to reconcile their findings that as training continued the strength of these trace-conditioning correlates of CA1 neuronal activity tended to diminish. Such investigations clearly demonstrate the involvement of the CA1 hippocampal neurons in the encoding of the auditory trace conditioning memory. However, the heterogeneous profiles of the hippocampal representation raise numerous further questions about how these distinct profiles of activity gain influence over the behavior of the animal. A clear advantage in this case is the choice of a well-characterized Pavlovian conditioning paradigm. Several control conditions can be employed to dispel notions that the alterations in firing properties that appear to occur with learning are not instead a reflection of non-associative influences on hippocampal activity (such as the simple experience of the animal being presented with the tone and corneal air puff stimuli). In this case, explicit control conditions can be employed, such as an explicitly unpaired condition where CA1 neuronal responses are recorded from an animal exposed to the same number of tone CS stimuli and air puff stimuli as a conditioned animal, yet for this control subject, the CS and US are not paired. Another approach would be to present one auditory stimulus (CS+) paired with the air puff in a trace-conditioning configuration and a second auditory stimulus (CS−) presented without the air puff. Learning-related changes in hippocampal neuronal activity would be seen during subsequent probe tests where the CS+ was presented alone. Nonassociative influences on hippocampal neuronal activity would be observed as firing property changes that were common to probe presentations of both the CS+ and CS−. 3.4.2. Space as the Unfortunate Contaminant in Studies of Nonspatial Correlates

The argument is perennially made that it is nearly impossible to eliminate the contribution of spatial information from most “nonspatial” tasks – the animal typically goes to some location to make a response, or stimuli are delivered to the animal when it occupies some particular location in the testing chamber (40). Simply, there is no need to infer that hippocampal activity reflects the encoding of nonspatial information, if one cannot be certain that spatial information has been eliminated. Therefore, it is important to examine whether spatial location influences the firing properties of neurons in animals performing nonspatial tasks. As will be discussed later, the same philosophical argument can be applied to studies in which place cell or head direction cell activity is recorded. It is clear that rodent hippocampal neurons exhibit location-specific firing and that a subset of limbic neurons exhibit direction-specific firing properties. However, the degree to which such spatially tuned neurons guide spatial behavior for the animal remains in question. Specifically, demonstrating that limbic neuronal activity is so closely associated with spatial information does not necessarily

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

143

prove that the rodent is utilizing this spatial code to solve spatial problems. This argument will be developed more thoroughly in the subsequent section of this chapter. Considering the auditory trace-conditioning paradigm discussed earlier, it would be relevant to determine whether spatial location modulates the profiles of learning-related changes in activity. In the case of the (30) studies, their design involved recording neurons from restrained rabbits, which essentially eliminated this concern. However, it would have been important to determine whether the absolute position of the restrainer and rabbit within the experimental environment affected the behavioral correlates of neuronal activity. Interestingly, results similar to (30) of CA1 neurons modeling the tone CS in a trace-conditioning paradigm have been reported from recordings of freely moving Long-Evans rats as well (16). Even so, the advantage of a trace-conditioning paradigm is that the temporal relationship between the CS and US is the critical factor; the spatial environment or the context within which the learning or performance takes place is less relevant. Specifically, the spatial context is not predictive of the delivery of the aversive US stimulus, and the animal can be anywhere in the arena and be presented the CS followed by the US. One interpretation then would be that trace conditioning-induced changes in hippocampal CA1 neuronal activity should not reflect spatial information processing, save for encoding the irrelevancy of the context to the relationship between the CS and US. Regardless, it would be important to define any conditioninginduced alterations in the spatial firing correlates of CA1 neurons, and the degree to which spatial location of the animal (provided not restrained) modulates the neuronal representations of the trace conditioning. Furthermore, as suggested earlier, the learned irrelevance of the spatial context as a predictor of the periorbital shock or the corneal air puff might be expected to be reflected in the activity profile of individual CA1 neurons. In fact, it would be interesting to examine how spatial correlates of hippocampal single-unit activity would be altered as the rodent subsequently acquires trace auditory fear conditioning in the same environment. 3.4.3. Dissociating Spatial and Nonspatial Influences on Single-Unit Activity

An alternative approach to determining whether hippocampal neuronal activity can represent nonspatial information would be to dissociate the contribution of spatial information to the behavioral correlate of neuronal activity by requiring the animal to engage in the same nonspatial behavioral task in several locations within some environment. This is precisely what Wood and colleagues did when examining the behavioral correlates of hippocampal neurons in rats performing a nonspatial odor discrimination task involving successively presented stimuli (85). Rats were trained to approach cups of scented sand on an elevated platform surrounded by a rich array of fixed distal spatial cues.

144

Stackman Jr.

On a given trial, a cup containing sand scented with one odor (out of nine possible odors) was presented to the rat in one of nine locations on the platform. The cup contained a buried Froot Loop reward, if the odor presented was different from that presented on the preceding trial (i.e., nonmatch). In this case, the rat would readily dig in the cup to retrieve the food reward. The cup did not contain a buried Froot Loop reward if the odor was the same as that presented on the preceding trial (i.e., match). Presented with a match odor, the rat was expected to “turn away from the cup” or simply to resist digging. Therefore, for the rat to perform the task in an efficient manner and to avoid impulsive digging in each presented cup, it would have to retain a memory of the odor just presented and then compare the subsequent odor to that remembered odor. The elegant and simple design entailed varying odors, locations where odors were presented, and the degree to which an odor was a match or nonmatch. The authors reported recording a total of 127 hippocampal neurons: 25 cells were reported to exhibit nonspatial firing correlates, 14 cells appeared to exhibit characteristic location-specific firing, and nearly as many cells exhibited firing related to whether the cup contained a nonmatch or match odor. Taken together these results indicate a slight preference for rat hippocampal neurons to encode the nonmatch/match characteristic of odor cues over location-specific firing when the rats are engaged in a task requiring them to recognize odors encountered previously. These results offer strong support for the view that the hippocampal formation plays a more generalized role in memory processes – principal neurons encode both spatial as well as nonspatial information. Such a demonstration brings the rodent hippocampal literature in line with that of the human hippocampal field. Moreover, it is notable that based on the description of the given behavioral environment (an elevated platform surrounded by distinct fixed array of spatial cues), these hippocampal neurons would be expected to exhibit characteristic location-specific firing as place cells if the rats were not trained to perform the successive odor discrimination task. In fact, it would be interesting to determine whether the same hippocampal neurons that exhibited nonmatch or match related firing would adopt location-specific firing during free exploration of the same elevated platform in the absence of odors being presented. These results would indicate that a given hippocampal neuron is capable of exhibiting both spatial and nonspatial firing correlates, as previous studies have suggested (8, 43, 80). The nonspatial correlates of hippocampal neurons described by Wood et al. (85) offer hope for bridging the gap between the proponents of an exclusively spatial view of hippocampal function and those of a spatial + nonspatial view of hippocampal function. Such a demonstration brings the rodent hippocampal literature in line

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

145

with that of the human hippocampal field. However, although it stands to reason that the rat would use match/mismatch signals to guide its performance in the odor discrimination task, these data remain correlational. There is no clear evidence that these hippocampal patterns of neuronal activity are essential for the accurate performance of the animal in the task. Let me be clear that this argument can be levied against the majority of experiments where hippocampal units are recorded while the rodent performs some behavioral task. It would be helpful to demonstrate that the hippocampus is necessary for the odor task performance, or that the hippocampus is engaged during the task. Frustratingly, followup studies have revealed that hippocampal lesions fail to significantly impair performance of rats on similar nonspatial recognition tasks (88, 89). In reconciling these discrepancies, it appears that there may be at least two behavioral strategies to solve odor recognition memory tasks – that is by using recollection of the episodic memory of the original experience or by judging the familiarity of stimuli presented previously (10). The studies reviewed in this section have been used to address a number of issues related to the strength of examining single-unit activity as it reflects behavioral responses in rodents. A number of limitations outlined here can affect the thoroughness to which this approach can be applied. Some technical limits related to tracking LEDS, the tethered cable, etc. are likely easier to eliminate with new technology. Importantly, it is the behavioral controls that must be applied to these studies of the behavioral significance of single-unit activity, which stand to have a substantial impact to our understanding of the manner in which behavioral events related to the task in question, are represented in the brain. The following section examines the issue of single-unit activity and behavior, using an exquisite example of the connection between the rodent’s behavior and neural activity: the head direction cell system. The head direction cell is defined and then a critical review of its assumed relationship to spatial navigation is examined. Lastly, suggestions are made as to how one might bridge the gap between the study of the basic firing properties of head direction cells and the study of how such cells may relate to spatial navigation.

4. Head Direction Cells 4.1. Primer

In January 1984, Dr. James Ranck Jr. of SUNY Downstate Medical Center made the discovery of head direction cells while recording single-units from a freely moving rat (51). For many years, Ranck had been conducting studies of the behavioral correlates of single-units recorded from hippocampal areas in the rat brain. Ranck’s approach was to position advance-able

146

Stackman Jr.

25-µm diameter metal electrodes into discrete brain areas and record extracellular single units from awake-behaving rats. This refined combinatory approach of simultaneous comparison of the on-going behavior of the animal and the activity of singleunits had been used by John O’Keefe in his investigations of hippocampal neurons (41). Two years after O’Keefe’s paper and years before the discovery of head direction cells, Jim Ranck published a lengthy report of the relationship between the behaviors he observed in rats that were present when individual hippocampal neurons fired (50). Three years later, O’Keefe published a thorough analysis and interpretation of rat hippocampal neurons, using essentially the same approach (39). The main difference was that John O’Keefe recognized that many of the hippocampal neurons he recorded were discharging in a location-specific manner, and he referred to them as place cells. Although Jim did not realize the spatial correlate of hippocampal neurons, it was his view that the hippocampus was likely part of a larger network essential for spatial behavior. To test this notion Ranck had begun to take his behavioral analysis of single-units to regions outside the hippocampus proper, an effort that led to the discovery of head direction cells. As is the case with many landmark findings in science, his discovery was serendipitous. Although Ranck believed that he had positioned his microelectrodes in the subiculum, the later histological reconstructions would reveal that the first head direction cells were instead recorded from the deep cell layers of the dorsal presubiculum or postsubiculum. If it were not for this stereotaxic error, perhaps we would not have known about the existence of head direction cells until much later. Jim published his initial report on the head direction cell as a Society for Neuroscience abstract that same year (48) and then as descriptive report in an edited book (49). Dr. Jeffrey Taube, then a postdoctoral fellow at SUNY Downstate Medical Center with Jim Ranck and Bob Muller, published the full quantitative characterization of the firing properties of postsubiculum head direction cells several years later (72, 73), using Muller and Kubie’s technological advance of tracking of the positions of two LEDs mounted on the recording cable via an overhead video camera. The relative positions of the two LEDs permitted calculation of the animal’s head direction at a 60-Hz resolution. Jeff Taube’s lab and the lab of Dr. Patricia Sharp went on to demonstrate the existence of head direction cells in several interconnected limbic regions of rat brain and report on numerous fundamental properties of the head direction cell signal. The reader is encouraged to consult the following reviews for more thorough discussion of the findings of the Sharp and Taube labs (56, 57, 58, 67, 68, 70).

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

147

4.2. Firing Characteristics of Head Direction Cells

As Ranck first described, a head direction cell fires during the time that the rodent’s head is oriented in a particular direction in the yaw or horizontal plane. This “preferred firing direction” of the head direction cell varies from cell to cell and is not influenced by the location of the animal within an environment (see Fig. 2). Therefore, the head direction cell signal is an interesting complement to the place cell: the activity of a hippocampal place cell is location-specific but head direction independent, whereas the activity of a head direction cell is direction dependent but location independent. The range of directional firing of a head direction cell that is its above-background firing also varies from cell to cell but generally represents approximately 90° of arc. Typically a head direction cell recorded from the postsubiculum or anterodorsal thalamus has little if any “background firing” meaning that when the rat’s head is oriented away from the cell’s preferred firing direction, the cell is quiet. Peak firing rates and the range of above-background directional firing vary from cell to cell and efforts to analyze and interpret differences in peak firing rates have not revealed much about differences between low rate and high rate head direction cells. After the initial full reports of the activity of postsubiculum head direction cells, the Taube and Sharp laboratories independently began programmatic analyses to determine the network of brain regions responsible for the generation of the head direction cell signal. The anterodorsal nuclei of the thalamus projects heavily to the postsubiculum and receives a dense projection from the postsubiculum. The view that the directional signal is conveyed through these connections was supported by Taube’s report in 1995 of finding of head direction cells in the anterodorsal thalamic nuclei (66). The same year Tad Blair and Pat Sharp used a time-slide analytical approach to determine that anterior thalamic head direction cells exhibited anticipatory directional firing – the peak firing rate of head direction cells best matched the future directional heading of the rat by ~25 ms. Subsequently, the Sharp and Taube laboratories published essentially simultaneous reports of head direction cells recorded from the lateral mammillary nuclei (1, 65). Comparison of directional tuning curves for postsubiculum, anterodorsal thalamic, and lateral mammillary neurons (see Fig. 2) revealed that lateral mammillary head direction cells exhibited the highest peak firing rates, while those in the anterodorsal nucleus of thalamus and postsubiculum exhibit progressively lower peak firing rates (65). To date, head direction cells have been recorded from mice, rats, chinchillas, and nonhuman primates, suggesting that the directional signal may be a common mammalian neural code for spatial information processing.

4.3. An Interconnected Network of Head Direction Cells

The brain regions, in which head direction cells have been recorded, comprise an interconnected network within the limbic system that likely influences spatial navigation and memory.

148

Stackman Jr.

Fig. 3. Major anatomical connections amongst brain regions containing head direction cells and regions containing place cells and grid cells. This diagram is not meant to be comprehensive in its presentation of all known neuroanatomical connection amongst these brain regions. The brain regions containing head direction cells are depicted as progressively shaded rectangles with rounded corners in the center of the diagram. Brainstem projections to the head direction cell circuit originating from the medial vestibular nuclei are shown at the bottom of the diagram while outputs of the circuit to the hippocampal place cell circuit are shown at the top of the diagram. The brain regions containing place cells are depicted as dark oval shapes. Note that the entorhinal cortex is depicted as a progressively shaded oval shape meant to illustrate that the medial entorhinal cortex contains grid cells and head direction cells, while the lateral entorhinal cortex contains place cells. The arrows illustrate primary interconnections between structures; the strength of these connections is suggested by the weight of the lines. Double-headed arrows indicate projections to and from a given brain region. Note that projections between the retrosplenial cortex and the laterodorsal thalamus have been omitted for clarity.

Figure 3 depicts the circuit comprising the brain regions where head direction cells have been recorded in rats. Numerous computational models have been proposed to delineate the differential contributions of distinct structures within the head direction cell circuit. In general, these models have proposed an attractor network of interconnected neurons whose collective “hill” or “bump” of activity represents the head direction signal. Some of these models attempt to account for the anticipatory responses of head direction cells in the anterior thalamus or lateral mammillary nuclei (14, 53, 61, 76). Although these models are beyond

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

149

the scope of this chapter, they have provided intriguing hypotheses regarding the generation and modulation of the direction signal as it is conveyed from region to region. The pervasive view of those studying the head direction cell system is that the collective activity of all head direction cells within a given brain region, the ensemble activity, provides a neuronal representation of the moment-to-moment directional heading of the animal as it moves through environmental space. The coordination of regional head direction cell ensembles may be fundamental for directional path integration and spatial problem solving. Directional responses of postsubiculum, anterodorsal thalamic, and lateral mammillary neurons are each influenced by external landmark cues and self-motion cues. Specifically, the angular rotation of a familiar landmark cue causes a similar shift in the preferred firing directions of head direction cells. Typically all recorded head direction cells will respond equivalently after such a cue rotation manipulation, as predicted by the neural attractor network models. Manipulation of internal cues such as optic flow, vestibular signals, and motor efference copy also influences the preferred firing directions of head direction cells (2, 22, 62, 64). Together these data demonstrate that cue manipulations known to influence spatial navigation also influence head direction cell activity in a predictable manner. These findings have lent support to the view that head direction cells are a fundamental brain mechanism of navigation.

5. The Behavioral Significance of Head Direction Cells 5.1. Introduction

To witness the discharge patterns of a well-isolated head direction cell with its compass-like firing as the animal turns its head into and out of the cell’s preferred firing direction is truly something to behold. In his overview of his initial discovery, Ranck described the head direction cell discharge as akin to that of a sensory afferent (51). This is an interesting analogy since the robust on/off nature of head direction cell firing when the animal abruptly turns its head away from the preferred firing direction of the cell certainly resembles a sensory afferent’s response to the presentation and removal of its appropriate stimulus. However, in the case of the head direction cell, this begs the question, what is the sensory stimulus? Such a strong example of a brain–behavior connection as the head direction cell must have some significance to spatial behavior. In the first set of studies providing a quantitative analysis of head direction cells, Jeff Taube hinted at the fact that rats appear to have a direction sense based on an overview of three behavioral studies and then suggested that head direction cells provided the underlying mechanism for such a sense (73).

150

Stackman Jr.

The drumbeat of the view that head direction cell activity is the neural representation of the direction sense grew louder in the years since 1990 – perhaps reaching its height with the publication in 2005 of an edited volume devoted entirely to the head direction cell (81). In contrast to hippocampal principal neurons that can discharge in a location-specific manner or fire in response to nonspatial information reflecting cognition or perception (such as that of odors, matching or nonmatching rules, and cueapproach) (4, 7, 85), head direction cell activity appears in a sense to be unidimensional – meaning that to date there is no evidence to suggest that the activity of head direction cells can also represent other nonspatial information. One of the largest questions in neuroscience concerns how single neurons and populations of neurons represent behaviorally relevant information and how such information might be utilized by the subject to guide its behavior, the final output of networks of interconnected synapses, cells, and brain regions. Therefore, the head direction cell system seems like an excellent model system to establish the cellular mechanism of the mammalian direction sense. 5.2. Do Head Direction Cells Guide Spatial Behavior?

Indeed, the most commonly employed strategy has been to record units from rats or mice as they freely move about open fields or mazes. This strategy has led to some of the most exciting demonstrations of the brain’s representation of spatial information namely, the determination of place cells in the hippocampus by John O’Keefe in 1970. Place cells are principal neurons of the hippocampus that discharge at a peak rate when the rodent occupies some particular circumscribed location of the floor surface of a maze or open field. The place field, or the area of the maze, where the place cell exhibits its highest firing rate, is influenced by a variety of extra-maze and intra-maze cues and by self-motion cues. The full report of the discovery of place cells in the hippocampus came complete with an exquisite treatise on their place within the broader context of the role of the hippocampus in learning and memory, John O’Keefe and Lynn Nadel’s, The hippocampus as a cognitive map (42). The discovery of place cells was so impressive on the field of behavioral neuroscience that several laboratories employing similar techniques turned their attention to spatial information as a guide to brain–behavior relationships. Certainly, the determination of head direction cells in the dorsal presubiculum by Jim Ranck in 1984 and their subsequent more complete characterization by Taube et al. (73); and the more recent demonstration of grid cells in the entorhinal cortex by Edvard Moser and colleagues follows the approach originated by John O’Keefe. In all of these phenomenological descriptions, the spatially tuned properties of these respective neurons revealed themselves by discharging at appropriate locations and/or directional headings as the rodent moved about the open field.

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

151

Based on the behavioral correlates reported for neurons recorded from a number of regions comprising the hippocampal formation (hippocampus proper, dentate gyrus, subiculum, and entorhinal cortex) and associated regions of the limbic system (cingulate cortex, anterior thalamus, mammillary nuclei), it would seem that these place cells, head direction cells, and grid cells would naturally be integral components of a neural network underlying spatial navigation. Although this notion is certainly straightforward, the results of experiments testing this hypothesis provide a much less clear picture. The final part of this chapter examines this issue by focusing on the head direction cell circuit and its relationship to spatial navigation as a model system for establishing the relationship between single-unit activity and a clearly defined behavior. Although the focus is on the head direction cell, a great deal of the issues raised here applies to the investigations of place cells and likely to grid cells as well. 5.3. Empirical Strategies Used to Test the Relationship of Head Direction Cells to Behavior

The final part of this chapter examines this issue by focusing on the head direction cell circuit and its relationship to spatial navigation as a model system for establishing the relationship between single-unit activity and a clearly defined behavior. Although the focus is on the head direction cell, a great deal of the issues raised here applies to the investigations of place cells and likely grid cells as well. One clear advantage to focusing this discussion on head direction cells is the fact that evidence to date indicates that head direction cells respond in register to manipulations of stimuli. That is, the rotation of a familiar cue card, for example, will cause all head direction cells recorded to shift their preferred firing directions by a consistent degree. Simply, if one head direction cell responds to a cue manipulation by shifting its preferred direction, then all the other cells will respond equally. The registration of all head direction cells is strong evidence that some attractor-like neural network underlies the connectivity of the head direction cell system in the brain. This attractor network would then appear to limit individualized responses of head direction cells to manipulations of the environmental stimuli.

5.4. Possible Interpretations Based on Empirical Outcome

A number of studies have been conducted to specifically address the behavioral significance of the directional firing correlates of limbic system neurons in laboratory rats. Typically, these experiments involve recording head direction cells while the rat performs some spatial task. Some manipulation is imposed (i.e., disorientation, cue rotation, cue removal, etc.) and a determination is made of the resulting influence on spatial firing of neurons and on the spatial behavior. For the sake of simplicity, the outcomes that follow assume that the rodent is performing a radial-arm maze task in which it must choose a baited arm of the maze as dictated by extra-maze visual cues. Head direction cell activity is acquired

152

Stackman Jr.

simultaneously with performance of the radial maze task. This experimental design provides one with several possible outcomes. 5.4.1. Possible Outcome 1: Congruent, Stability of Spatial Firing and Behavior

The preferred firing directions of head direction cells and the accuracy of performance on the spatial task remain stable after the experimental manipulation. What this means for the navigating rat or mouse is that under present conditions the animal chooses the correct arm of the maze and the preferred firing directions of recorded cells remain stable. This outcome falls into a category that will be referred to here as congruent. Congruency between spatial behavior and the preferred firing directions of head direction cells would be consistent with the view that head direction cells guide spatial behavior.

5.4.2. Possible Outcome 2: Congruent, Shifts in Spatial Firing and Behavior

Here, in response to the experimental manipulation, the preferred firing directions of head direction cells undergo a shift in responding or what is commonly referred to as a remapping event. For example, recordings would reveal that head direction cells had undergone a significant shift in their respective preferred firing directions within the familiar environment. For example, the manipulation caused the preferred firing directions of the head direction cells to shift by ~90°. Consistent with the changes in directional firing properties, one would observe a corresponding shift in the behavioral choice of the animal on the radial maze task. As the Possible outcome 1 above, this outcome would also be classified as congruent. This congruency between the spatial behavior and the firing of head direction cells would provide support for the view that the spatial behavior is in some manner guided by the directionally tuned neurons.

5.4.3. Possible Outcome 3: Incongruent, Shifts in Spatial Firing, but no Shift in Behavioral Response

This outcome would be frustrating for a model in which directionally tuned neuronal activity guides on-going spatial behavior. Here, the experimental manipulation would cause the preferred firing directions of head direction cells to shift or remap. Despite such changes in head direction cell activity, the behavioral choice of the animal in the radial maze task would remain stable (meaning the rodent would choose the correct arm of the maze). That is, the animal would continue to make the correct spatial choice in the given task. In such a case, the spatially tuned network of neurons reflects one view while the animal’s behavioral responses reflect another view. This may simply be the manifestation of “going with one’s gut”!

5.4.4. Possible Outcome 4: Incongruent, Stable Spatial Firing, but a Shift in Behavioral Response

This outcome is the complement of outcome 3: head direction neurons exhibit stable preferred firing directions but behavioral responses on the radial-arm maze task are disrupted or variable. The disconnection between head direction cell responses and behavioral responding would be taken as evidence that these directionally tuned neurons do not guide spatial behavior.

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

153

An alternative conclusion could of course be made. That being that the activity of a few head direction cells or place cells may not in fact be predictive of the global behavioral responding of the animal, since behavior is likely more a reflection of the head direction cell population code or the place cell population code. To date, the studies of how individual hippocampal place cells respond to various experimental manipulations reveal considerable variability in response. Specifically, some hippocampal neurons remap in response to certain manipulations while others do not. These findings of variability would tend to support the view that studying the place cell population as a whole may be more informative for establishing truths about spatial choice behavior and the firing of place cells. In contrast, studies of anterior thalamic and postsubiculum head direction cells have shown much more consistency in firing responses to experimental manipulations. To put it simply, if the manipulation causes one head direction cell to shift its preferred firing direction by 90°, then all other simultaneously recorded cells also exhibit a 90° shift in their preferred firing directions. The consistency in the effect that manipulations have across the preferred firing directions of multiple head direction cells recorded simultaneously is remarkable. Such findings suggest that there is limited advantage in recording ensembles of head direction cells over recordings of just a few cells simultaneously. 5.5. What is the Evidence that Head Direction Cells Guide Spatial Behavior?

Clearly, to date, the thorough analysis of head direction cells and place cells recorded while rodents are freely exploring cylindrical arenas has been essential for understanding so much about this unique neural signal. However, under such behavioral conditions, little can be said regarding the degree to which head direction cells are guiding this “foraging behavior.” It is not clear at all whether there is any incentive for the rat to keep track of its location or directional heading in the arena during such “free foraging” sessions. Why should it be? The behavioral experience of the rodent up to that point has been perpetual removal from its home cage, placement into an arena wherein sweetened food pellets rain down from the heavens while the rat is tethered to the recording cable for periods of varying length. The delivery of food is not contingent in any way on the behavior of the animal. If the animal so desires, it can consume pellets or choose not to. At the end of the recording session, regardless of its spatial location or directional heading, the animal is retrieved and returned to its home cage. Thus, it is not clear why we should be so quick to infer that because neurons discharge accordingly to directional heading or spatial location that these cell signals must be monitored by the animal to guide its behavior or to maintain its spatial orientation. To truly test the relationship between head direction cell activity and spatial orientation, one needs to impose a more demanding

154

Stackman Jr.

behavioral condition on the animal than simply retrieving randomly located food pellets. A handful of studies have been conducted though that have tested the relationship between head direction cell activity and spatial memory and navigation and it is useful to briefly summarize them with the above outcome interpretations in mind. A more thorough review and analysis of head direction cells and place cells and their respective contribution to spatial behavior was published some years ago (35). However, what follows here is an analysis of the literature that reaches some distinct conclusions about the state of the relationship between the directionally tuned single-units recorded from rodent brain and the degree to which they guide spatial behaviors. 5.5.1. Dudchenko and Taube (1997)

The first study to thoroughly examine head direction cell responses during the acquisition and performance of a spatial task tested single-unit activity from rodents in a radial-arm maze task (87). Water restricted rats were trained to run to one arm of a radialarm maze. Correct arm choice behavior was rewarded by delivery of water in the cup at the end of the correct arm. A floor-toceiling black curtain surrounded the maze and a white curtain subtending ~48° of arc was provided as a primary polarizing visual cue. The correct arm of the maze was consistently positioned relative to the white curtain cue across all training trials. Head direction cells were recorded from the anterodorsal thalamus of seven rats and from the postsubiculum of one rat before and during maze performance under standard conditions and conditions in which the white curtain cue was rotated by 90 or 180° out of the rats’ view. This cue rotation manipulation reliably shifts the preferred firing directions of head direction cells in most environments. Here, the manipulation enabled testing the correspondence of head direction cell activity and spatial choice behavior. Rotation of the white curtain cue by 90 or 180° caused a corresponding shift in the preferred firing directions of all recorded head direction cells, and the rats chose the rotationally appropriate arm of the radial maze. A minority of cases in which head direction cells did not shift their preferred firing directions, the rat’s behavioral choice did not shift either. Further analyses of these data revealed a strong positive correlation between the degree of shift in preferred firing direction of head direction cells and the degree of shift in maze arm choice. There was no apparent difference between responses of head direction cells recorded from the anterodorsal thalamus and those from the postsubiculum. The results of Dudchenko and Taube (1997) provided a clear indication that the white cue curtain exerted stimulus control over both the spatial responding on the radial maze and the preferred firing directions of the head direction cells. The consistency of the single-unit responses and the behavioral choice data do suggest that head direction cells reflect the rodent’s spatial

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

155

orientation in an environment. These data match the definition of Possible Outcome 2, Congruent, Shifts in Spatial Firing and Behavior defined earlier. The interpretation of this outcome would be that due to the equivalent influence of the cue manipulation on both determinants, these data support the view that the head direction cells supported the spatial behavior. However, the difficulty here is that these data are correlational and whether the spatial behavior of the rats was truly guided by the head direction cells remains uncertain. 5.5.2. Golob et al. (13)

Golob et al. (13) sought to address this issue more directly during a subsequent recording study of head direction cells from rats performing a spatial task. We imposed a condition known to produce a large shift in preferred firing directions of head direction cells to ask what the consequence of such a shift in head direction cell network would have on spatial behavior (13). Waterrestricted rats were trained to locate water reward in one corner of a high-walled square enclosure containing a single polarizing cue card on one wall. Behavioral training entailed picking the rat up from a holding cage outside of the square enclosure and releasing it at the center of the enclosure floor. The rat was removed after selecting the correct corner of the enclosure and consuming the water reward. Head direction cells were recorded from the anterodorsal thalamus during these standard trials and trials in which the cue card was rotated 90° before the rat was released into the enclosure. Behavioral performance of the rats during standard trials was very accurate (77% correct) and stable over days. Despite the consistency in behavioral responding, head direction cells were observed to shift their preferred firing directions often by ~90° during ~23% of standard trials. Behavioral performance was also accurate on cue rotation trials, meaning that the rats selected the rotationally appropriate corner during these cue rotation trials. As during standard trials, despite the stable behavioral responding, most but not all head direction cells shifted their preferred firing directions during cue rotation trials. A third condition was also imposed in which the enclosure was converted into a novel high-walled rectangular enclosure by doubling the length of two of its parallel walls between behavioral trials. Such a change in enclosure geometry is known to induce a significant shift in the preferred firing direction of head direction cells (73). The analysis of head direction cells and spatial responses during rectangular enclosure trials were equally revealing. Head direction cell preferred firing directions were found to undergo a clockwise or counterclockwise shift of 90° when the rat was in the rectangular enclosure. Interestingly, all rats generalized from their performance in the square enclosure to their corner choice during the trials in the rectangular enclosure. Simply, what this means is that although head direction cells shifted their preferred firing

156

Stackman Jr.

direction by large amounts, the accuracy of the behavioral response was not changed from that observed during the standard trials. To summarize, the data of Golob et al. (13) match the possible outcome 3, Incongruent, Shifts in Spatial Firing, but No Shift in Behavioral Response defined earlier. The interpretation of this outcome has to be that these data do not provide support for the view that head direction cells were guiding spatial behavior under these conditions. 5.5.3. Ragozzino et al. (47 )

Ragozzino, Leutgeb, and Mizumori studied the responses of dorsal striatal head direction cells and hippocampal place cells in rats performing a hippocampal-dependent spatial working memory radial-arm maze task (47). The behavioral task required the rats to use a win-shift strategy in which each maze arm or spatial location was to be visited only once in a session. The authors imposed a number of distinct manipulations of the stimuli (rotating cues, extinguishing room lights, introducing the rodents to the maze in a novel room, and cue conflicts between extra and intra-maze cues) to examine the degree of influence environmental changes had on spatial working memory and the firing of spatially tuned neurons in the striatum and hippocampus. Consistent with previous work, the dorsal striatal head direction cells shifted their preferred firing directions in response to cue rotations, maze rotations, and the novel environment, and place cells tended to respond equivalently. Under the cue conflict conditions where the relationship between intra-maze and extra-maze cues was disrupted, the responses of head direction cells and place cells tended to be quite distinct. The lack of correspondence in responses to cue conflict between head direction cells and place cells reported here is distinct from the general agreement between the two classes of spatially tuned neurons previously reported (21). Although the focus of their work was not to explicitly test whether head direction cells guide spatial behavior, Ragozzino et al. (47) did find a significant correlation between directional firing of striatal neurons and the number of errors committed by the rats during performance of the working memory task. Specifically, reduced directional-specific firing of the striatal neurons was associated with an increase in errors made on the maze. These data then match the outcome 2 defined earlier, Congruent, Shifts in Spatial Firing and Behavior, providing support for the view that head direction cells may guide spatial responses under these conditions. Caution should be applied when interpreting the results of Ragozzino et al. (47), since these head direction cells were recorded from the dorsal striatum and there is little evidence that performance on a win-shift spatial working memory task is dependent upon the striatum (28, 44). It is not clear why striatal cell activity would correlate with performance on a win-shift task in this way. Finally, the rats of the Ragozzino et al. (47) study were

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

157

trained to complete at least eight radial-arm maze trials within a 1-h period. The rationale for requiring eight trials within a recording session was most likely to ensure that there was adequate sampling in spatial locations while single-units are recorded. Running multiple trials of a win-shift spatial working memory task within a 1-h period raises a concern that behavioral performance may degrade with repeated trials due to the potential for proactive interference (memory for arms entered on trial X, affects choice accuracy on trial X + 1, etc.). 5.5.4. Muir and Taube (36)

In a brief report, Muir and Taube (36) described the responses of head direction cells in four rats performing a spatial navigation task in a replica of Edward Tolman’s sunburst maze (36). In a classic study, Tolman and colleagues had trained 53 rats to run from a start location on an elevated maze, straight across a circular arena, and through an alleyway comprising three 90° turns to arrive at a goal location. Once well trained, the rats were given a probe condition where access to the familiar alleyway from the circular arena was blocked but 18 novel alleyways radiating out in different directions were available. The idea being that given that the learned route was blocked, would the rats choose the most efficient short cut to the goal? Tolman’s results revealed that 36% of the rats chose the alleyway leading directly to the previously learned location of the goal (74). This result was interpreted as support for Tolman’s contention that rats use cognitive maps to perform spatial tasks. Interestingly, the next most popular response of the rats (17% of the rats) was to choose the alleyway that pointed in the direction the rats traveled to the goal location during training. Muir and Taube built a modified replica of Tolman’s sunburst maze apparatus to permit recording of anterodorsal thalamic and postsubicular head direction cells from waterrestricted rats during training and sunburst probe tests. Rats were well trained on the standard trials before implantation of electrodes; however, the performance of Muir and Taube’s rats during the sunburst probe trials did not match with that of Tolman’s original study – in fact only 1 of the 4 rats chose the alleyway leading directly to the learned goal location. The recording data revealed that the preferred firing directions of head direction cells remained stable (shifted on average <1°) across multiple standard trials and the preferred firing directions of the cells did not shift during the sunburst probe trials. Despite the remarkable stability of the directional firing of the neurons, the behavioral performance was punctuated by the rats committing numerous errors during the sunburst trials (average number of errors = 5.5 ± 1.9) (36). These results indicate that head direction cell responses were stable across two behavioral conditions, one in which rats performed very accurately (standard training trials) and another in which the rats were impaired (sunburst probe trials). Therefore, this experimental

158

Stackman Jr.

outcome matches with that of outcome 4, Incongruent – Stable Spatial Firing, but a Shift in Behavioral Response, defined earlier. These data do not support the view that the head direction cells that were monitored during task performance guided the spatial responses of the rats in any way. 5.5.5. Summary of Effort to Date

Taken together, the four studies explicitly testing head direction cell responses in rats performing some spatial memory or navigation task present an inconclusive picture regarding the degree to which head direction cells guide spatial behavior. Table 1 summarizes the main results of the four studies reviewed earlier. There may be a number of reasons for the lack of consistency in findings from these studies – different behavioral tasks used in each study, low number of animals used in some studies, and the fact that the two studies supporting the relationship between head direction cell activity and spatial behavior provide only correlational evidence. Although it does appear that the results of two of the studies are congruent with the view that head directions guide spatial behavior, several issues were raised for each of these studies that are problematic for such an interpretation of the data. One clear message here should be that additional experiments are needed to explicitly understand the conditions under which head direction cell signals might be used by rodents to solve a spatial problem.

5.6. Where Do We Go from Here?

Based on the review of the literature about the relationship between head direction cell activity and spatial behavior, it is certain that a fresh empirical approach is needed. Three suggestions are made here to bridge this divide. Although the preceding section has focused on directional behavior and the firing of head direction cells, the following suggestions are applicable to the broader field of behavioral neuroscientists interested in testing the significance of behavioral firing correlates of neuronal singleunit activity.

5.6.1. Does the Behavioral Task Engage the Brain Region from which Head Direction Cells Are Recorded?

The data from studies of head direction cell discharges from rats actively engaged in some spatial memory or navigation task have failed to adequately support the view that such directionally tuned neurons guide spatial behavior. In most of the studies, head direction cells were recorded from the postsubiculum or anterodorsal thalamus where directionally tuned neurons are relatively plentiful. As indicated in the column Relevance of recording site? of Table 1, what is not clear though is whether the postsubiculum or anterodorsal thalamus is required for performance of the task employed. Another way to address this issue is to ask whether performance on the task employed engages either the postsubiculum or anterodorsal thalamic nuclei? Large lesions that compromise the entire anterior thalamus (including anterodorsal, anteromedial, anteroventral, and often laterodorsal nuclei) have been found

Anterodorsal thalamus and postsubiculum

Dudchenko and Taube (1997)

Stable Shifted w/cue Stable Stable Stable

Impaired

Unstable Unstable Unstable

Stable Shifted w/cue

Stable Shifted w/cue Stable Stable

w/cue stable

Novel rectangle

Standard Dorsal striatum dorsal Working memory hippocampus win-shift radial-arm CUE rotation maze Novel rectangle Anterodorsal Reference memory/ Standard thalamus and Tolman sunburst Sunburst probe Postsubiculum maze

Stable Shifted

Standard Cue rotation

Reference memory water-finding task

Stable Shifted w/cue

No evidence to support involvement

Dorsal striatum not likely involved, HPC is involved

No evidence to support involvement

Both ADN and PoS may be involved

Head direction Relevance of Spatial behavior cell PFD recording site?

Standard Cue rotation

Condition

Reference memory radial – arm maze

Task

4

2

3

1

Outcome

Note: Head direction cell PFD refers to the preferred firing direction of head direction cells recorded. Relevance of recording site? refers to whether there is evidence to indicate that the brain region from which head direction cells were recorded is necessary for the behavioral task used. Outcome refers to which of the four possible experimental outcomes (as described in the text of Sects. 5.4.1–5.4.4) that the results of the experiment match. stable indicates that the spatial behavior or the preferred firing directions of head direction cells was not altered by the condition imposed; unstable indicates that the preferred firing direction of head direction cells shifted between trials in the task; shifted w/cue indicates that the preferred firing direction of the cell exhibited an appropriate shift after rotation of the cue. ADN, anterodorsal nucleus of thalamus; PoS, postsubiculum; HPC, hippocampus

Muir and Taube (36)

Ragozzino et al. (47)

Golob et al. (13) Anterodorsal thalamus and postsubiculum

Recording site

Reference

Table 1 Summary of empirical studies testing relationship of spatial behavior and head direction cells Behavioral Correlates of Neuronal Activity Recorded as Single-Units 159

160

Stackman Jr.

to impair performance of rats on a win-shift spatial working memory radial-arm maze task (26). This hippocampal-dependent win-shift radial maze task is comparable to that used by (47) to test the relationship of striatal behavior to the activity of head direction cells recorded from the dorsal striatum. Lesions of the anterior thalamus also impair the postoperative acquisition of spatial working memory and reference memory radial maze tasks (32, 33, 84). Discrete lesions of the anterodorsal thalamic nuclei produce relatively mild deficits in an allocentric spatial memory water maze task (77). Of the two radial-arm maze tasks used in the studies that support a link between head direction cell activity and spatial navigation (Dudchenko and Taube, 1997; (47)), only performance of the task used by (47) has been clearly shown to be impaired by lesions of the anterior thalamus. Rather than record from the anterior thalamus though (47) recorded head direction cells from the striatum, a structure unlikely to be principally involved in the task (28, 44). To be thorough, there is no evidence available to suggest that the anterodorsal thalamic nuclei are involved in the performance of the water-finding task in the square enclosure (13) or the performance of the Tolman sunburst maze (36). Interestingly, acquisition of our water-finding task is impaired in rats with bilateral lesions of the vestibular apparatus (63), a lesion that also compromises directional firing of rat anterodorsal thalamic neurons (64). By extension then, performance of the water-finding task may be influenced by lesions of the anterodorsal thalamic nuclei. Regarding the involvement of the postsubiculum in spatial behavior, there is but one published report. Jeff Taube (71) found that lesions of the postsubiculum impaired spatial learning of rats in the Morris water maze (71). It is remarkable how relatively little is known about the unique contribution of the postsubiculum to the spatial behavior when compared with that of the anterior thalamus. To move closer to an understanding of the contribution of anterior thalamic and postsubiculum head direction cells to spatial behavior, it would seem that the neurophysiologists could benefit from listening to the behavioral neuroscientists. Determining whether the brain region from which the head direction cells are recorded from is actually relevant for, or engaged by, the behavioral task would seem to be an appropriate first step in the process. If it is not clear that the brain region in question is involved in directional responding in a spatial task, then why should we expect that neurons from that brain region should be guiding the animal’s performance in the task? 5.6.2. Recording Ensembles of Head Direction Cells May Be more Informative

The lack of clear support for head direction cells guiding spatial behavior from these studies may have arisen from the fact that these studies examined the activity of just a few head direction cells during task performance. The activity of relatively few head

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

161

direction cells may not in fact be predictive of the global behavioral responding of the animal, since behavior is likely more a reflection of the head direction cell population code. To date, the studies of how individual hippocampal place cells respond to various experimental manipulations reveals considerable variability (11, 45). Specifically, some hippocampal neurons remap in response to certain manipulations while others do not. These findings of variability would tend to lend support to the view that studying the place cell population as a whole may be more informative for establishing truths about spatial choice behavior and the firing of place cells. In contrast, studies of anterior thalamic and postsubiculum head direction cells have shown much more consistency in firing responses to experimental manipulations (62, 66, 69, 73). The fact that head direction cells seem to respond in register to cue manipulations would then seem to suggest that the response of the head direction cell ensemble or population to some experimental manipulation could be estimated by examining how that manipulation affects one or a few head direction cells. To date, there has been one published report of recordings of head direction cell ensembles (20). It will be interesting to determine whether measures of the coherency of head direction cell ensembles correlates with the accuracy of behavioral responding of rodents in a spatial task. However, given the consistency in response to simultaneously recorded head direction cells, it would be surprising to find that recordings of ensembles of head direction cells would provide a distinct result to that provided by recordings of individual head direction cells. 5.6.3. First Find a Behavior that Is Directional in Nature, Then Test Whether it Is Dependent upon Head Direction Cells

This suggestion is simple and straightforward. Rather than recording head direction cells in rats performing all sorts of different tasks, each with distinct response requirements and cue conditions, why not begin with a behavioral task that requires the rodent to use directional information or attend to its directional heading? Once identified, then one could begin to determine whether head direction cells guided this directional behavior. I had originally thought that this would have been easier said than done. However, in reviewing the now classic literature by Tolman and his contemporaries, together with more recent work, it appears that rodents display a preference for directional responding in several spatial tasks. Furthermore, converging evidence suggests that rodents acquire directional response tasks more readily than they do place response tasks. Recall that in the sunburst maze trials, Tolman reported that 17% of his rats chose the alleyway that enabled the rats to travel in the same direction as that traveled to the goal location during training (74). A thorough follow up experiment by (3) tested groups of rats in several T-maze tasks and found that rats trained to run in a particular direction on the maze to find food acquired the task in significantly

162

Stackman Jr.

fewer trials than rats trained to place or by response (e.g., always turn right) (3). Interestingly, in an influential review of these and other classic studies of the 1930s and 1940s Restle (54), mistakenly interpreted the results of the study by (3) as evidence that rats predominantly learn about places rather than responses (or direction). Clearly the views of Restle, Tolman, and others promoted the view of rodents as predominate place learners and influenced the interpretation of hippocampal place cells as components of a cognitive mapping system by (42), which supported place learning. The demonstration that place learning in the Morris water maze task is dependent upon the hippocampal formation was seen as a clear definition of the functional significance of hippocampal place cells (34). 5.7. “Directional” Responding of Rodents in Spatial Tasks

The water maze and the place learning and memory therein are now considered by many to be the gold standard test of hippocampal-dependent place navigation. However, a number of recent studies have found that rats tend to exhibit a preference for directional responding in the Morris water maze, rather than place responding (17, 18). Specifically, rats were trained to locate a submerged platform in a pool surrounded by extra-maze visual cues. During no-platform probe tests, the water maze was translated linearly in the testing room such that the absolute spatial location of the platform in the room was centered in the opposite quadrant of the pool. Hamilton et al. (18) found that rats swam in the direction of the relative location of the platform rather than to the learned absolute location of the platform as dictated by the extra-maze cues. We recently demonstrated that male C57BL/6J mice exhibit an identical preference for directional responding over place responding in the water maze under similar translated probe test conditions (83). Similar results indicating directional response preferences of rodents and evidence that rats acquire directional tasks faster than place or response tasks have been found in other paradigms as well (59), suggesting that rodents may be predisposed to using directional information. What remains to be seen is whether (1) such directional behavior is dependent upon brain regions such as the postsubiculum and anterodorsal thalamic nuclei; and (2) head direction cells guide such directional behavior. Taking this experimental approach, our laboratory has recently found that directional responding of male C57BL/6J mice in the Morris water maze task is disrupted by the temporary inactivation of the anterodorsal thalamic nuclei, but not by temporary inactivation of the dorsal CA1 region of the hippocampus (82). These encouraging results are consistent with the view that head direction cells in the anterodorsal thalamic nuclei are important for directional navigation. Certainly, more work remains to be done to explicitly test whether head direction cells guide such direction-based navigational behavior, and to

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

163

firmly establish that the behavioral responding is truly directional. However, our data demonstrate the feasibility of this fresh alternative approach to testing a fundamental question in brain–behavioral relationships.

6. Conclusion This chapter has reviewed issues related to the behavioral significance of single-unit activity recorded from limbic system regions. The chapter has emphasized the recent development of chronic single-unit recording in freely moving mice and identified issues and limitations in the approach toward defining the behavioral significance of limbic single-unit activity for spatial and nonspatial learning and memory. The final section of the chapter explored the relationship of single-unit activity recorded simultaneously with the acquisition of behavioral responses, using the head direction cell signal as a model system. The focus on the head direction cell system limited the discussion to four studies, each with clear advantages and limitations. It is clear from this discussion that there is presently little support for the notion that head direction cells represent the direction sense, as has been suggested. Reviewing the literature of place cells and spatial behavior, as well as other tests of the relevance of neuronal firing correlates to given behaviors also suggest limited support for the relationship. Several reasons are offered for why the empirical studies to date have failed to clearly define the relationship between spatial navigation and the firing of head direction cells. It should be noted that the suggestions made for bridging the gap between the single-unit data and its behavioral significance certainly hold true for nonspatial behavioral correlates as well. For instance, the strategy of first determining whether the behavioral task engages the brain region, from which the single-units are to be recorded from, is fundamental. One would imagine that time and effort would be conserved greatly if the significance of a given brain region to a given behavior were determined initially. In addition, a plea is made that both neurophysiologists and behavioral or systems neuroscientists could benefit by increased dialogue between what at times appear to be insular camps. Converging evidence is appearing from independent behavioral neuroscientists to suggest that rodents exhibit a preference for directional navigation over place navigation in several spatial tasks. Responding to this emerging evidence, we recently found evidence to suggest that directional navigation by C57BL/6J mice is dependent upon the anterodorsal nuclei of the thalamus. It will be exciting to explicitly examine whether head direction cells per se guide such directional responding of the mice. Overall, the

164

Stackman Jr.

convergence of chronic single-unit recording together with refined behavioral analytical approaches and the increasing use of mice as experimental subjects represent an exciting development in the field of behavioral neurophysiology. The availability of genetically engineered mice offers great promises for examining the molecular mechanisms that underlie spatial navigation and perhaps the molecular signature of the head direction cell.

Acknowledgments Many of the ideas described in this chapter stem from conversations and experiments conducted with current and former students, advisors, and collaborators. Special thanks in particular are extended to Michael Guidi, Shiao-Ying Chow, and Sidney Williams, who tolerated pressure to run “one last experiment” time and time again as we tested the differential influence of the anterior thalamus and dorsal hippocampus on directional responding in the Morris water maze task. Portions of this work were supported by a grant from the National Institutes of Health (AA014407) and from developmental funding provided by the Charles E. Schmidt College of Science at Florida Atlantic University. References 1. Blair HT, Cho J, Sharp PE. 1998. Role of the lateral mammillary nucleus in the rat head direction circuit: A combined single unit recording and lesion study. Neuron 21:1387–1397. 2. Blair HT, Sharp PE. 1996. Visual and vestibular influences on head-direction cells in the anterior thalamus of the rat. Behav Neurosci 110:643–660. 3. Blodgett HC, McCutchan K, Mathews R. 1949. Spatial learning in the T-maze: the influence of direction, turn, and food location. J Exp Psychol 39:800–809. 4. Breese CR, Hampson RE, Deadwyler SA. 1989. Hippocampal place cells: stereotypy and plasticity. Journal of Neuroscience 9:1097–1111. 5. Broadbent NJ, Squire LR, Clark RE. 2004. Spatial memory, recognition memory, and the hippocampus. Proc Natl Acad Sci USA 101(40):14515–14520. 6. Clark RE, Zola SM, Squire LR. 2000. Impaired recognition memory in rats after damage to the hippocampus. J Neurosci 20(23):8853–8860.

7. Deadwyler SA, Breese CR, Hampson RE. 1989. Control of place-cell activity in an open field. Psychobiology 17(3):221–227. 8. Deadwyler SA, Bunn T, Hampson RE. 1996. Hippocampal ensemble activity during spatial delayed-nonmatch-to-sample performance in rats. J Neurosci 16(1):354–372. 9. Dudchenko PA, Eichenbaum H. 2004. Hippocampus: cognitive processes and neural representations that underlie declarative memory. Neuron 44(1):109–120. 10. Fortin NJ, Wright SP, Eichenbaum H (2004) Recollection-like memory retrieval in rats is dependent on the hippocampus. Nature 431:188–191. 11. Fenton AA, Muller RU. 1998. Place cell discharge is extremely variable during individual passes of the rat through the firing field. Proc Natl Acad Sci U S A 95(6):3182–3187. 12. Gaveriaux-Ruff C, Kieffer BL. 2007. Conditional gene targeting in the mouse nervous system: Insights into brain function and diseases. Pharmacol Ther 113(3):619–634. 13. Golob EJ, Stackman RW, Wong AC, Taube JS. 2001. On the behavioral significance of head

Behavioral Correlates of Neuronal Activity Recorded as Single-Units direction cells: neural and behavioral dynamics during spatial memory tasks. Behav Neurosci 115(2):285–304. 14. Goodridge JP, Touretzky DS. 2000. Modeling attractor deformation in the rodent head-direction system. J Neurophysiol 83(6): 3402–3410. 15. Gray CM, Maldonado PE, Wilson M, McNaughton B. 1995. Tetrodes markedly improve the reliability and yield of multiple single- unit isolation from multi-unit recordings in cat striate cortex. J Neurosci Methods 63(1–2):43–54. 16. Green JT, Arenos JD. 2007. Hippocampal and cerebellar single-unit activity during delay and trace eyeblink conditioning in the rat. Neurobiol Learn Mem 87(2):269–284. 1 7. Hamilton DA, Akers KG, Johnson TE, Rice JP, Candelaria FT, Sutherland RJ, Weisend MP, Redhead ES. 2008. The relative influence of place and direction in the Morris water task. J Exp Psychol Anim Behav Process 34(1):31–53. 18. Hamilton DA, Akers KG, Weisend MP, Sutherland RJ. 2007. How do room and apparatus cues control navigation in the Morris water task? Evidence for distinct contributions to a movement vector. J Exp Psychol Anim Behav Process 33(2):100–114. 19. Hammond RS, Tull LE, Stackman RW. 2004. On the delay-dependent involvement of the hippocampus in object recognition memory. Neurobiol Learn Mem 82(1):26–34. 20. Johnson A, Seeland K, Redish AD. 2005. Reconstruction of the postsubiculum head direction signal from neural ensembles. Hippocampus 15(1):86–96. 21. Knierim JJ, Kudrimoti HS, McNaughton BL. 1995. Place cells, head direction cells, and the learning of landmark stability. J Neurosci 15:1648–1659. 22. Knierim JJ, Kudrimoti HS, McNaughton BL. 1998. Interactions between idiothetic cues and external landmarks in the control of place cells and head direction cells. J Neurophysiol 80:425–446. 23. Kubie JL. 1984. A driveable bundle of microwires for collecting single-unit data from freely moving rats. Physiology and Behavior 32:115–118. 24. Lin L, Chen G, Kuang H, Wang D, Tsien JZ. 2007. Neural encoding of the concept of nest in the mouse brain. Proc Natl Acad Sci U S A 104(14):6066–6071. 25. Lin L, Osan R, Shoham S, Jin W, Zuo W, Tsien JZ. 2005. Identification of networklevel coding units for real-time representation

26.

27.

28.

29.

30.

31.

32.

33.

34. 35.

36.

37.

38.

165

of episodic experiences in the hippocampus. Proc Natl Acad Sci USA 102(17):6125–6130. Mair RG, Burk JA, Porter MC. 2003. Impairment of radial maze delayed nonmatching after lesions of anterior thalamus and parahippocampal cortex. Behav Neurosci 117(3):596–605. Manns JR, Howard MW, Eichenbaum H. 2007. Gradual changes in hippocampal activity support remembering the order of events. Neuron 56(3):530–540. McDonald RJ, White NM. 1993. A triple dissociation of memory systems: hippocampus, amygdala, and dorsal striatum. Behav Neurosci 107(1):3–22. McEchron MD, Disterhoft JF. 1997. Sequence of single neuron changes in CA1 hippocampus of rabbits during acquisition of trace eyeblink conditioned responses. J Neurophysiol 78(2):1030–1044. McEchron MD, Tseng W, Disterhoft JF. 2003. Single neurons in CA1 hippocampus encode trace interval duration during trace heart rate (fear) conditioning in rabbit. J Neurosci 23(4):1535–1547. McNaughton BL, O’Keefe J, Barnes CA. 1983. The stereotrode: a new technique for simultaneous isolation of several single units in the central nervous system from multiple unit records. J Neurosci Methods 8(4):391–397. Mitchell AS, Dalrymple-Alford JC. 2005. Dissociable memory effects after medial thalamus lesions in the rat. Eur J Neurosci 22(4):973–985. Mitchell AS, Dalrymple-Alford JC. 2006. Lateral and anterior thalamic lesions impair independent memory systems. Learn Mem 13(3):388–396. Morris RGM, Garrud P, Rawlins JNP, O’Keefe J. 1982. Place navigation impaired in rats with hippocampal lesions. Nature 297:681–683. Muir GM, Taube JS. 2002. The neural correlates of navigation: do head direction and place cells guide spatial behavior? Behav Cogn Neurosci Rev 1(4):297–317. Muir GM, Taube JS. 2004. Head direction cell activity and behavior in a navigation task requiring a cognitive mapping strategy. Behav Brain Res 153(1):249–253. Muller RU, Kubie JL. 1987. The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J Neurosci 7(7):1951–1968. Muller RU, Kubie JL, Ranck JB, Jr. 1987. Spatial firing patterns of hippocampal complex-spike cells in a fixed environment. J Neurosci 7(7):1935–1950.

166

Stackman Jr.

39. O’Keefe J. 1976. Place units in the hippocampus of the freely moving rat. Exp Neurol 51:78–109. 40. O’Keefe J. 1999. Do hippocampal pyramidal cells signal non-spatial as well as spatial information? Hippocampus 9(4):352–364. 41. O’Keefe J, Dostrovsky J. 1971. The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Res 34:171–175. 42. O’Keefe J, Nadel L. 1978. The Hippocampus as a Cognitive Map. UK: Clarendon Press, Oxford. 43. Olton DS, Wible CG, Pang K, Sakurai Y. 1989. Hippocampal cells have mnemonic correlates as well as spatial ones. Psychobiology 17(3):228–229. 44. Packard MG, Hirsh R, White NM. 1989. Differential effects of fornix and caudate nucleus lesions on two radial maze tasks: evidence for multiple memory systems. J Neurosci 9(5):1465–1472. 45. Poucet B, Lenck-Santini PP, Hok V, Save E, Banquet JP, Gaussier P, Muller RU. 2004. Spatial navigation and hippocampal place cell firing: the problem of goal encoding. Rev Neurosci 15(2):89–107. 46. Raffaele KC, Olton DS. 1988. Hippocampal and amygdaloid involvement in working memory for nonspatial stimuli. Behavioral Neuroscience 102:349–355. 47. Ragozzino KE, Leutgeb S, Mizumori SJ. 2001. Dorsal striatal head direction and hippocampal place representations during spatial navigation. Exp Brain Res 139(3):372–376. 48. Ranck JB. 1984. Head direction cells in the deep cell layer of dorsal presubiculum in freely moving rats. Society for Neuroscience Abstracts 10:599. 49. Ranck JB. 1985. Head direction cells in the deep cell layer of dorsal presubiculum in freely moving rats. In: Buzsaki G, Vanderwolf CH, editors. Electrical Activity of the Archicortex. Budapest: Hungarian Academy of Sciences. p 217–220. 50. Ranck JB, Jr. 1973. Studies on single neurons in dorsal hippocampal formation and septum in unrestrained rats. I. Behavioral correlates and firing repertoires. Exp Neurol 41(2):461–531. 51. Ranck Jr. JB. 2005. Foreword: History of the discovery of head direction cells. In: Wiener SI, Taube JS, editors. Head direction cells and the neural mechanisms of spatial orientation. Cambridge, MA: The MIT Press. p xi–xiii. 52. Recce ML, O’Keefe J. 1989. The tetrode: An improved technique for multi-unit extracellular

53.

54. 55.

56. 57.

58.

59.

60.

61.

62.

63.

64.

65.

recording. Society for Neuroscience Abstracts 15:1250. Redish AD, Elga AN, Touretzky DS. 1996. A coupled attractor model of the rodent head direction system. Network: Computational Neural Systems 7:671–685. Restle F. 1957. Discrimination of cues in mazes: a resolution of the place-vs.-response question. Psychol Rev 64(4):217–228. Rondi-Reig L, Libbey M, Eichenbaum H, Tonegawa S. 2001. CA1-specific N-methylD-aspartate receptor knockout mice are deficient in solving a nonspatial transverse patterning task. Proc Natl Acad Sci USA 98(6):3543–3548. Sharp PE, editor. 2002. The neural basis of navigation: evidence from single-unit recording. Norwell, MA: Kluwer. Sharp PE. 2005. Regional distribution and variation in the firing properties of head direction cells. In: Wiener SI, Taube JS, editors. Head direction cells and the neural mechanisms of spatial orientation. Cambridge, MA: The MIT Press. p 3–15. Sharp PE, Blair HT, Cho J. 2001. The anatomical and computational basis of the rat head-direction cell signal. Trends Neurosci 24(5):289–294. Skinner DM, Etchegary CM, Ekert-Maret EC, Baker CJ, Harley CW, Evans JH, Martin GM. 2003. An analysis of response, direction, and place learning in an open field and T maze. J Exp Psychol Anim Behav Process 29(1):3–13. Solomon PR, Vander Schaaf ER, Thompson RF, Weisz DJ. 1986. Hippocampus and trace conditioning of the rabbit’s classically conditioned nictitating membrane response. Behav Neurosci 100(5):729–744. Song P, Wang XJ. 2005. Angular path integration by moving “hill of activity”: a spiking neuron model without recurrent excitation of the head-direction system. J Neurosci 25(4): 1002–1014. Stackman RW, Golob EJ, Bassett JP, Taube JS. 2003. Passive transport disrupts directional path integration by rat head direction cells. J Neurophysiol 90(5):2862–2874. Stackman RW, Herbert AM. 2002. Rats with lesions of the vestibular system require a visual landmark for spatial navigation. Behav Brain Res 128:27–40. Stackman RW, Taube JS. 1997. Firing properties of head direction cells in the rat anterior thalamic neurons: Dependence on vestibular input. J Neurosci 17(11):4349–4358. Stackman RW, Taube JS. 1998. Firing properties of rat lateral mammillary single units:

Behavioral Correlates of Neuronal Activity Recorded as Single-Units

66. 67. 68. 69.

70.

71.

72.

73.

74. 75.

76.

77.

78.

head direction, head pitch, and angular head velocity. J Neurosci 18(21):9020–9037. Taube JS. 1995. Head direction cells recorded in the anterior thalamic nuclei of freely moving rats. J Neurosci 15(1 Pt 1):70–86. Taube JS. 1998. Head direction cells and the neurophysiological basis for a sense of direction. Prog Neurobiol 55:225–256. Taube JS. 2007. The head direction signal: origins and sensory-motor integration. Annu Rev Neurosci 30:181–207. Taube JS, Burton HL. 1995. Head direction cell activity monitored in a novel environment and during a cue conflict situation. J Neurophysiol 74(5):1953–1971. Taube JS, Goodridge JP, Golob EJ, Dudchenko PA, Stackman RW. 1996. Processing the head direction cell signal: a review and commentary. Brain Res Bull 40(5–6):477–484. Taube JS, Kesslak JP, Cotman CW. 1992. Lesions of the rat postsubiculum impair performance on spatial tasks. Behav Neural Biol 57(2):131–143. Taube JS, Muller RU, Ranck JB, Jr. 1990. Head-direction cells recorded from the postsubiculum in freely moving rats. I. Description and quantitative analysis. J Neurosci 10(2):420–435. Taube JS, Muller RU, Ranck JB, Jr. 1990. Head-direction cells recorded from the postsubiculum in freely moving rats. II. Effects of environmental manipulations. J Neurosci 10(2):436–447. Tolman EC, Ritchie BF, Kalish D. 1946. Studies in spatial learning. I. Orientation and the short-cut. J Exp Psychol 36:13–24. Tseng W, Guan R, Disterhoft JF, Weiss C. 2004. Trace eyeblink conditioning is hippocampally dependent in mice. Hippocampus 14(1):58–65. van der Meer MA, Knierim JJ, Yoganarasimha D, Wood ER, van Rossum MC. 2007. Anticipation in the rodent head direction system can be explained by an interaction of head movements and vestibular firing properties. J Neurophysiol 98(4):1883–1897. van Groen T, Kadish I, Michael Wyss J. 2002. Role of the anterodorsal and anteroventral nuclei of the thalamus in spatial memory in the rat. Behav Brain Res 132(1):19–28. Weiss C, Bouwmeester H, Power JM, Disterhoft JF. 1999. Hippocampal lesions prevent trace eyeblink conditioning in the freely moving rat. Behav Brain Res 99(2):123–132.

167

79. Weiss C, Kronforst-Collins MA, Disterhoft JF. 1996. Activity of hippocampal pyramidal neurons during trace eyeblink conditioning. Hippocampus 6(2):192–209. 80. Wible CG, Findling RL, Shapiro M, Lang EJ, Crane S, Olton DS. 1986. Mnemonic correlates of unit activity in the hippocampus. Brain Res 399(1):97–110. 81. Wiener SI, Taube JS, editors. 2005. Head direction cells and the neural mechanisms of spatial orientation. Cambridge, MA: The MIT Press. 82. Williams SB, Guidi M, Stackman RW. Heading in the right direction: Directional navigation in the Morris water maze is dependent on the anterodorsal thalamus, but not the dorsal CA1 of the hippocampus.; 2009; Chicago, IL: Society for Neuroscience. 83. Williams SB, Stemme M, Vick KA, Stackman RW. Do C57BL/6J mice exhibit directional or place navigation in the Morris water maze?; 2008; Washington, DC.: Society for Neuroscience. 84. Wolff M, Gibb SJ, Cassel JC, Dalrymple-Alford JC. 2008. Anterior but not intralaminar thalamic nuclei support allocentric spatial memory. Neurobiol Learn Mem 90(1):71–80. 85. Wood ER, Dudchenko PA, Eichenbaum H. 1999. The global record of memory in hippocampal neuronal activity. Nature 397(6720):613–616. 86. Yee BK, Rawlins JN. 1994. The effects of hippocampal formation ablation or fimbriafornix section on performance of a nonspatial radial arm maze task by rats. J Neurosci 14(6):3766–3774. 87. Dudchenko, P. A. & Taube, J. S. 1997. Correlation between head direction cell activity and spatial behavior on a radial arm maze. Behavioral Neuroscience 111, 3-19. 88. Dudchenko, P. A., Wood, E. R. & Eichenbaum, H. 2000. Neurotoxic hippocampal lesions have no effect on odor span and little effect on odor recognition memory but produce significant impairments on spatial span, recognition, and alternation. J Neurosci 20, 2964-2977. 89. Mumby, D. G., Gaskin, S., Glenn, M. J., Schramek, T. E. & Lehmann, H. 2002. Hippocampal damage and exploratory preferences in rats: memory for objects, places, and contexts. Learn Mem 9, 49-57. 90. Stackman, R. W., Tullman, M. L. & Taube, J. S. 2000. Maintenance of rat head direction cell firing during locomotion in the vertical plane. J Neurophysiol 83, 393-405.

Chapter 7 Event-Related Potentials of the Cerebral Cortex Steven L. Bressler Abstract The event-related potential (ERP) is a major methodological tool used to investigate the functioning of the cerebral cortex of humans and other mammalian species. The cerebral cortex is the part of the mammalian brain that is most critical for sensory perception, motor action, and cognition. The ERP yields vital information about neuronal population activity in the cortex in relation to sensory, motor, or cognitive events. The temporal resolution of the ERP is ideally suited for capturing the timing of neurophysiological processes in the cortex, and hence it provides a sensitive electrophysiological signature of neurocognitive activity. In fact, the ERP is the principal recording methodology currently in use for investigating the fine temporal structure of human cognitive processing. Like the closely linked eventrelated magnetic field, the ERP can be recorded noninvasively, and is thus of great value for the study of cognition in normal human subjects. The ERP is also valuable in the clinical setting for determining the temporal integrity of cortical processes in neurological patients. First, this chapter discusses the neurophysiological origins of the cortical ERP. Next, the different techniques by which the ERP is measured, and the many forms that it takes, are examined. Finally, the ways in which the ERP is modeled, and the procedures by which it is analyzed, are described. Key words: Brain, Electromagnetic activity, Electroencephalogram (EEG), Magnetoencephalogram (MEG), Event-related field (ERF), Electrocorticogram (ECoG), Local field potential (LFP), Pyramidal cell

1. Introduction It has long been known that electrical activity can be recorded from the living brains of humans (3) and other mammals (10). From the earliest days of recording this activity, researchers have sought to understand its relation to brain function and to use it to monitor and assess brain state. Continuous records of brain activity, examined without regard to particular points in time, are often useful for determining brain state. However, more detailed knowledge of brain function depends on temporal registration of Robert P. Vertes and Robert W. Stackman Jr. (eds.), Electrophysiological Recording Techniques, Neuromethods, vol. 54, DOI 10.1007/978-1-60327-202-5_7, © Springer Science+Business Media, LLC 2011

169

170

Bressler

the activity to specific events, either in the external environment or self-generated. The event-related potential (ERP) is a temporal signature of brain electrical activity that occurs in relation to a sensory, motor, or cognitive event (5, 12). The event-related field (ERF) is the magnetic correlate of this activity1. ERPs and ERFs have an advantage over indices of brain function that monitor blood flow or metabolism in that their time course more closely follows the activity of neuronal populations in the brain. According to one point of view, the ERP refers only to a transient waveform that results from averaging multiple brain electrical potential time series, all precisely time-locked to an external event. However, there are many examples of event-related electrophysiological phenomena that are temporally related to an external event but not precisely time-locked to it, and do not require averaging to be observed (21). Therefore, the perspective taken in this chapter is a broader one, in which any brain electrical potential waveform that reliably occurs in relation to a sensory, motor, or cognitive event qualifies as an ERP. Thus, for example, oscillatory phenomena that reliably occur in the brain either before or after an event are considered to be ERPs, even if the timing is imprecise, whereas oscillatory phenomena that arise spontaneously without relation to an event are not. ERPs provide a window to the dynamics of neuronal population activity in the brain in relation to sensory, motor, and cognitive processes. Neuronal population activity results from the cooperative interactions of individual neurons. The fraction of any single neuron’s total activity that represents its involvement in cooperative interaction may be exceedingly small, yet it is the cooperative activity of populations that carries influences between different parts of the brain. Population activity originates at the local neuronal circuit level, but becomes coordinated across widely distributed brain systems. Thus, the ERP derives from cooperative interactions in local neuronal populations, and is modulated by long-range interactions between neuronal populations transmitted over axonal pathways.

2. Generation of Electromagnetic Activity in the Cerebral Cortex

Electromagnetic activity is generated by neurons throughout the brain, in their soma, axons, and dendrites (2, 21, 54). The dendritic activity of neurons in the cerebral cortex is responsible for the macroscopic electrical and magnetic activity observed with extracranial

General considerations regarding the ERP throughout this article also apply to the ERF, except where noted.

1

Event-Related Potentials of the Cerebral Cortex

171

sensors (50). The cortical pyramidal cell is an important class of excitatory neuron that is critically involved in the generation of cortical electrical field potentials and the corresponding magnetic fields. The dendrites of pyramidal cells, as most other neurons, are specialized to receive excitation and inhibition at chemical synapses where postsynaptic ion channels are gated by ionotropic receptors. At these synapses, the release of neurotransmitter from the presynaptic axon terminal causes the ion channels to open, and electromotive forces (EMFs) cause current to flow through the channels. As a result, current circulates in closed loops across the cell membrane and through the intracellular and extracellular spaces. The excitatory postsynaptic potential is a depolarization of the dendritic transmembrane potential due to a net inward flow of positive current across the postsynaptic membrane. Dendritic excitatory synapses create loop currents consisting of net positive charge that flows inward across the postsynaptic dendritic membrane, passes through the intracellular compartment, flows outward across passive membrane with a strength that decreases with distance from the sites of influx, and finally completes the loop through the extracellular space (Fig. 1). The dendritic inhibitory

Fig. 1. Schematic membrane potential recordings (MP1–MP3) from intracellular microelectrodes (ME1–ME3) in superficial and deeper parts of a cortical pyramidal neuron, and simultaneous field potential recordings (FP1–FP3) from extracellular electrodes (E1–E3), following apical dendrite depolarization resulting from afferent fiber stimulation. Loop currents are indicated by dashed lines. (Modified from Speckmann (61) ).

172

Bressler

postsynaptic potential is a hyperpolarization of the dendritic transmembrane potential at inhibitory synapses due to a net outward flow of positive current across the postsynaptic membrane. Loop currents are also created by inhibitory synapses, but the flow of current is in the opposite direction to that created by excitatory synapses (21, 61). When both excitatory and inhibitory synapses are active, the net balance of all active synapses determines the direction of flow of loop currents. Loop currents cause pyramidal cells to generate trains of pulses (action potentials). They do so by establishing a gradient of transmembrane potential that continuously varies in time and space along the pyramidal cell dendrites as a function of current strength. The sum of currents contributed by all the active synapses on the dendritic tree produces a resultant transmembrane potential at the cell body and the initial segment of the axon. When this resultant potential exceeds the firing threshold, the initial segment responds by generating a pulse train that actively propagates along the axon, diverges into the axonal branches, and, by causing the release of neurotransmitter at the branch terminals, leads to the excitation or inhibition of other neurons. Thus, the loop currents generated by a pyramidal cell determine its effect on other neurons. Loop currents generated by neighboring pyramidal cells summate in the extracellular space when they flow in the same direction, and cancel otherwise. The passage of extracellular current across the resistance in this space is manifested as the extracellular electrical field of potential or field potential (61). The extracellular components of the net loop currents generated by a population of active neighboring pyramidal cells give rise to the population mean field potential (22). The population mean field potential of pyramidal cells can be recorded by an electrode of appropriate size and position in the extracellular space of the cortical tissue (the local field potential or LFP), on the cortical surface (the electrocorticogram, ECoG) or even at the scalp surface (electroencephalogram, EEG). The ability to detect the cortical field potential at a distance from the cortical tissue depends on it being an open field (see later), and summating over large populations of pyramidal cells. The intracellular components of the same closed-loop currents giving rise to the field potential are primarily responsible for the closely related magnetic field recorded extracranially as the magnetoencephalogram (MEG) (32, 53). The magnitude of the LFP recorded by an extracellular microelectrode at any instant in time depends on multiple factors, including the number of nearby synchronously active pyramidal cells, the strength and directions of their currents, their morphology and alignment, and the position of the electrode in the field. In general, for any population of neurons to generate a strong field potential, it is not sufficient that the neurons actively generate

Event-Related Potentials of the Cerebral Cortex

173

strong extracellular currents. The morphology and alignment of those neurons must also promote the summation of the currents in the extracellular space. For example, the field potential generated by a population of neurons in which the orientations of the dendrites are uniformly distributed in all directions is zero, on average, because of cancellation of extracellular currents, even if the individual dendrites are all maximally excited. On the contrary, parallel alignment of the dendrites, as with the pyramidal cells, promotes extracellular current summation if the same portion of each dendrite (e.g., the distal end) is excited. However, cancellation may still occur if the location of the excitation is randomly distributed along the dendrites. The cortical pyramidal cells have a single long apical dendrite aligned in parallel across the population and perpendicular to the cortical surface. The population typically receives concurrent excitation or inhibition at the same dendritic locale, for example, distal or proximal end, and thus tends to generate extracellular currents that maximally summate and augment the field potential. They are densely interconnected with each other and with neighboring neuron types, both excitatory and inhibitory, to form local neuronal circuits that are complex but similarly organized throughout the cortex. Pyramidal cells are also targets for synaptic inputs from other cortical and subcortical areas, and likewise send long, myelinated axons to those areas. The pyramidal cell population is called a dipole generator because the field it generates is a distributed dipole field, meaning that the summated loop currents that emerge from one end (pole) of the cells are detected by an extracellular electrode there as a current source, and the currents that enter into the other end (pole) are detected by an extracellular electrode there as a current sink (21). An important property of the dipolar source-sink population geometry is that it generates a field that is open, meaning that the currents spread in the volume of the brain and can be detected at a distance from the generating population (41, 43). Thus, a local population of synchronously active pyramidal cells can generate an open dipole field that is recordable either locally or at a distance from the population (Fig. 2). A superficial cortical sink may be recorded as a negative potential (and a superficial source as a positive potential) by an electrode in the superficial cortical layers, at the cortical surface or at the scalp. Synchronously active dipole fields of multiple local populations tend to summate and thus be detectable at a greater distance than those of the individual local populations alone, unless cancellation occurs due to surface folding (30). The question of what causes field potentials to change over time is central to understanding the relation of ERPs to brain function. Although the determinants of temporal variation of the field potential are diverse, and their effects are not well understood,

174

Bressler

Fig. 2. Left: Dipolar electrical field resulting from depolarization of the apical dendrites of a single cortical pyramidal neuron. Isopotential surfaces are represented by dashed lines, and extracellular field currents by solid lines. Right: Electrical field created in the surrounding volume by a synchronously active sheet of pyramidal cell dipole generators resulting from depolarization of their apical dendrites. Isopotential surfaces are represented by solid lines. (Modified from Gloor (30)).

some basic aspects of pyramidal cell population activity that bring about temporal variation of its generated field potential may be identified (19). One important consideration is that the number of single neuron generators contributing to the population activity, and the magnitudes of the currents that they generate, may change over time. Another crucial aspect is that neuronal synchrony, the tendency for the unitary generators in the population to be similarly active, may also change over time. Factors such as inputs from other populations and intrinsic changes in the excitability of the population can affect both the total magnitude of currents generated by the neurons of the population and their degree of synchrony. These factors thus influence the time course of the population field potential and ultimately determine the dynamics of ERP generation in relation to brain function.

3. Measurement of the Cortical ERP The cerebral cortex is uniquely positioned to make the principle contribution to brain activity recorded extracranially, and is also a common target of intracranial recording, because cortical population activity is thought to be fundamentally related to cognitive processes (27). Depending on the size of the recording and reference electrodes, and the location within or outside the

Event-Related Potentials of the Cerebral Cortex

175

head, recorded cortical field potentials integrate neural activity over a range of spatial scales. The nature, size, and location of both the recording and reference electrodes are important for determining the spatial scale of integration of the field potential that is represented by a recorded voltage trace, as well as the recorded voltage range. Microelectrodes placed directly within the cortex record field potentials, integrated on a submillimeter scale, that are predominantly generated by local neuronal populations. To truly localize the field potential to a restricted population, two closely spaced electrodes are inserted into the same cortical region, and the potential difference between the two is recorded as the LFP. Although the amplitude range can vary considerably, intracortical field potentials are generally no larger than 1,500 µV in peak-to-peak amplitude. Electrodes placed on the surface of the brain integrate over a larger submillimeter to millimeter scale. Since the cerebral cortex makes up most of the brain’s surface, and the non-cortical surfaces are difficult to access, the intracranial brain surface field potential is almost always recorded from the cortex as the ECoG. The ECoG recording may be bipolar (i.e., the difference in field potential activity recorded from two nearby cortical surface electrodes) or monopolar (i.e., from one cortical surface electrode with respect to a distant neutral reference electrode). ECoG peak-to-peak amplitudes are normally in the range of several hundred microvolts. Integrating over an even greater spatial extent, the extracranial EEG is recorded from the surface of the head, again either bipolarly or monopolarly. Scalp-recorded EEGs are greatly attenuated due to the high resistivity of the skull and scalp, and peakto-peak amplitudes usually lie between 10 and 50 µV in the adult human. Finally, the MEG records the magnetic field with sensors located just outside the head. MEG signals from a third-order gradiometer are commonly less than 1–2 picoTesla in peak-topeak amplitude. Both the EEG and MEG integrate over a centimeter spatial scale. Measurement of ERPs from within the cortex, at the cortical surface, or at the scalp involves the detection of summated dipole fields of extracellular currents generated by cortical pyramidal cell populations that have become synchronously active as a result of a sensory, motor, or cognitive event. Measurement of the ERF involves detection of magnetic fields generated by intradendritic current flow of cortical pyramidal cell populations (53). The generator populations that are detected by the ERP and ERF generally overlap, but are not identical because of differences between the two measures in sensitivity to generator orientation and depth. The time-varying ERP or ERF waveform is commonly treated as a signal to be detected in the presence of noise. The following discussion explicitly treats the ERP as a signal, but similar considerations also apply to the ERF.

176

Bressler

Noise refers to any contribution to the potential difference (voltage) recorded from two electrodes that is not from the signal source. Common sources of noise in brain electrical recordings include the following: (1) potentials from the brain (cephalic noise); (2) potentials from the head muscles and skin, eyes, and tongue (extracephalic cranial noise); (3) potentials from parts of the body other than the head, such as the heart (extracranial physiological noise); (4) random microscopic fluctuations at the electrodes (thermal noise); (5) noise from movement of the person or animal (movement artifact); (6) fluctuations introduced by electronic recording components (electronic noise); (7) radiated contamination from other electrical equipment (environmental noise); and even (8) fluctuations due to imprecision in the discrete digitization of the continuously varying recorded voltage for storage in a digital computer (quantization noise). The ability to detect the ERP waveform signal in the presence of noise depends on the relative strengths of signal and noise, as measured by the ratio of signal power (magnitude squared) to noise power. If this signal-to-noise ratio is large then the signal may be observable in the digitized time series of just a single voltage trace. If it is small, however, some procedure is required for signal detection. A simple and effective signal detection technique is to average over an ensemble of realizations (also called trials) of the voltage time series. This is a reasonable procedure if each individual time series realization is registered to a common time marker representing the occurrence of an event. The cortical sensory-evoked potential is an example of a signal that is commonly detected by ensemble averaging (Fig. 3). Repeated stimuli (e.g., flashes of light or brief tones) are presented to a subject while voltages are recorded from arrays of monopolar or bipolar electrodes placed within or near the corresponding sensory

Fig. 3. A visual event-related potential derived by averaging an ensemble of single-trial LFPs recorded from the posterior parietal cortex of a macaque monkey. The LFPs were recorded from a chronically implanted bipolar transcortical electrode consisting of 51-µm-diameter platinum wires with 2.5-mm tip separation. (Modified from Bressler (5)).

Event-Related Potentials of the Cerebral Cortex

177

cortex. The voltage records are digitized and broken into time segments corresponding to successive stimulus presentations called trials. The resulting time series segments from the individual trials are collected, temporally registered with respect to the time of each stimulus, and averaged separately for each electrode. When using ensemble averaging to detect the transient cortical sensory-evoked potential, it is generally assumed that a dipole generator population of pyramidal cells in the sensory cortex responds to each stimulus in a characteristic manner by generating a reproducible waveform signal. This waveform may not be detectable in the single-trial time series if the signal-to-noise ratio is too small. It is further assumed that the signal occurs with a fixed amplitude and latency with respect to the stimulus. The noise, on the contrary, is deemed to be temporally unrelated to the stimulus. These premises embody the standard model of the ERP, which is discussed later. The standard model predicts that ensemble averaging of single-trial time series maintains the magnitude of the stimulus-evoked signal while decreasing the magnitude of the noise by destructive waveform cancellation. The signal-tonoise ratio is proportional to the number of trials averaged. If the assumptions of the standard model hold or are not too severely violated, ensemble averaging is a simple method for ERP estimation.

4. Varieties of the Cortical ERP The cortical ERP is an electrical signal generated by neuronal populations in relation to a sensory, motor, or cognitive event. The corresponding event-related magnetic field (ERF) has many of the same dynamic and functional properties as the ERP. Two general classes of ERP are distinguished by whether the relevant event is continuous or discrete. In the case of continuous events, such as when sensory stimuli are presented rapidly and repetitively, the steady-state ERP takes the form of a continuous periodic response, and is analyzed in long epochs. The steady-state ERP shows the same repetition frequency as the stimulus, within limits, and preferred frequencies at which the steady-state response is maximal have been suggested to represent the natural resonant frequencies of oscillating neuronal populations in the sensory cortices (33). Steady-state visualevoked potentials have proven useful in the assessment of cognitive function (49, 59, 60). The study of steady-state ERPs depends on a variant of frequency analysis (see later). Field potentials recorded during periodically modulated sensory stimulation are narrowband filtered around the frequency of the driving periodicity to

178

Bressler

derive the steady-state (periodic) ERP. Variations in the amplitude and phase of the steady-state ERP are interpreted in terms of driving frequency, spatial location, and behavioral state. In the case of discrete events, the associated transient ERP is analyzed in short epochs time-locked to the event. Transient ERPs have engendered a great deal of interest because of their potential for revealing the dynamics of cognitive processing by the brain. They may occur with either relatively fixed or variable latency (or phase) in relation to the repeated event. If the latency is variable, then ensemble averaging is destructive since components of opposite polarity on successive trials tend to be cancelled, and hence may not reveal the ERP. Nonphase-locked ERPs are referred to as “induced” when they reliably occur following a stimulus, and as “anticipatory” when they reliably arise in the period prior to a stimulus or motor response. This type of ERP may be effectively analyzed by averaging the frequency content of single-trial time series, rather than averaging the time series themselves. Nonphase-locked transient event-related phenomena are detected as frequency-specific changes in the ERP time series. These phenomena may consist either of an event-related increase or decrease of power in a particular frequency range, typically the alpha (8–13 Hz) or beta (14–30 Hz) bands. Because the level of ERP power is typically considered to reflect the degree of synchrony within local neuronal populations, a power increase is called event-related synchronization and a power decrease is called event-related desynchronization (55). Phase-locked transient ERPs show temporal variation on a subsecond time scale that is conducive to measurement of the rapidly changing dynamics of cognition (57). For this reason, a long lasting effort in the study of ERPs has been to identify components that span brief periods of time in the ERP before or after a measurable event. The ERP waveform consists of a series of positive and negative wave-like components that are identified by their time of occurrence and polarity. Thus, for example, the P300 component occurs as a positive wave, which peaks at or near 300 ms after a stimulus event. Components sometimes are also designated simply according to their serial order, so that the P300 component might also be called the P3 component, meaning the third positive wave following the stimulus. Other components are named based on event properties. For example, the contingent negative variation (CNV) is a slow negative wave that appears in the interval between two stimuli after a contingency of the second stimulus on the first has been established. ERP components are also categorized as relating to sensory, motor, or higher-order cognitive processes of the brain. Sensory ERPs may be recorded by electrodes placed in sensory brain

Event-Related Potentials of the Cerebral Cortex

179

structures, on the surface of sensory cortices or on the overlying scalp. They are typically extracted from noise by ensemble averaging with respect to an external stimulus in one of the sensory modalities. The early poststimulus components are directly related to stimulus-evoked sensory processing, and because their characteristics depend on the physical properties of the stimulus, they are called exogenous. Olfactory ERPs are usually large with respect to the background, and thus do not require averaging to be observed. They also occur as oscillatory bursts that do not have readily identified components (21). In the auditory and somatosensory modalities, early components generated by sensory relay nuclei in the brain stem are revealed in extracranial recordings by the ensemble averaging of large numbers of trials. Since these early components are considered obligatory, they have clinical value as a test of the integrity of the subcortical sensory pathways. In the visual modality, the brain stem nuclei apparently generate closed potential fields, and thus the earliest components observable from the scalp are generated within the cortex (12). In animal studies, sensory nerve stimulation produces a positive–negative wave complex at the cortical surface. The positive deflection may arise from a dipole field with a superficial source generated by depolarization of layer 4 neurons in primary cortex, and the negative deflection from a superficial-sink dipole generated by depolarization in the superficial layers (58). The latencies of early sensory cortical components are of great interest to researchers who study cortical information processing (38, 52). Motor ERPs are extracted from noise by ensemble averaging with respect to a movement-related event rather than a sensory stimulus. Since the characteristics of motor ERP components do not depend on external stimulus properties, but rather on a subject’s internal state, they are called endogenous. The most wellknown motor component is the readiness potential (RP), a slow, ramp-like negative potential shift that begins as early as 1.5 s before the production of voluntary limb movements. The RP magnitude grows larger in recordings over the sensorimotor cortex contralateral to the movement, when compared with the ipsilateral side, as the time of the movement approaches. Together with the observation that the RP has a somatotopic organization in the contralateral sensorimotor cortex, this finding indicates that the RP component is related to preparation for limb movement. The RP (also designated N1) is terminated by a positive– negative (P1–N2) complex prior to muscle contraction, which is then followed by a late positive (P2) component. The P1 deflection is often absent unless the movement is brisk and forceful. Cognitive ERP components are related to cognitive, rather than sensory or motor, processes of the brain. They can provide valuable information about the spatial organization of large-scale cortical network activity underlying a cognitive function, as well as

180

Bressler

the temporal organization on a subsecond time scale (4, 6). By definition, cognitive ERP components are considered to be endogenous. The aforementioned CNV is an endogenous cognitive ERP component that occurs in the interval between two stimuli (S1 and S2), presented in any sensory modality, for which a contingency has been established by their prior pairing. Most often, the subject is required to execute a motor response to the S2. The CNV arises as a ramp-like, negative-going wave that peaks shortly after S2. It appears maximally in the EEG over frontal and central regions, and can be as large as 20 µV at the scalp. When the S1–S2 duration is sufficiently long, the CNV resolves into early and late subcomponents, the early one related to the sensory processing of S1, and the late one associated with anticipation of S2 and motor preparation. The late subcomponent is thought to be generated by the prefrontal cortex (56) in relation to that structure’s role in mediating cross-temporal contingencies (26). The late CNV occurs prior to a stimulus, reflecting anticipation and preparation, whereas other cognitive effects appear following a stimulus. The mismatch negativity (MMN) is an early poststimulus ERP component that may reflect the maintenance of sensory working memory (51). It is thought to be elicited by stimuli having physical properties that deviate from prior (standard) stimuli registered in sensory memory. Occurring between 80 and 200 ms after presentation of deviant stimuli, thus overlapping the N1 and P2 components, the MMN is revealed by computing the difference wave between averaged ERPs evoked by deviant and standard stimuli. The MMN is subserved by a large-scale network that includes the dorsolateral prefrontal cortex in addition to sensory cortical areas (1). The question of whether the MMN is automatic or whether it may be affected by attention has been controversial. Modulation of the early poststimulus ERP components has been reported in relation to spatial or nonspatial attention in the auditory, visual, and somatosensory modalities (45). The P3 component appears as a positive deflection between 300 and 900 ms poststimulus that is related to the cognitive context of the stimulus (64). Context is usually established by presenting a subject with a series of events of one class interspersed with rarer (oddball) events of a second class to which the subject must respond. Two P3 components are distinguished (62): an earlier, frontal P3a component and a later, parietal P3b component. The P3a is elicited by unexpected novel stimuli. The amplitude of the P3b depends on the relative event probability, and the latency reflects the degree of difficulty in categorizing the stimulus. A widely distributed cortical network underlying the P3b is thought to be involved in the categorization of stimuli as significant events, with network strength reflecting the degree of “consonance” resulting from comparison of stimulus attributes with a maintained “expectation” (36).

Event-Related Potentials of the Cerebral Cortex

181

Finally, an ERP component related to semantic memory is the negative-going N400. It occurs between 200 and 500 ms after presentation of a potentially meaningful information-bearing stimulus when that stimulus is incongruent with the prevailing semantic context established by previous stimuli. The amplitude of the N400 is directly related to the degree of semantic deviance of the word from its sentence context, and is attenuated by prior priming with semantically related words. The N400 amplitude is reduced as a function of associative, semantic, and repetition priming within or across sensory modalities (37). Variation of its scalp-recorded topographic distribution with task and stimulus type suggests that the N400 reflects the construction of meaning by cross-modal interactions in a large-scale cortical network. This view is supported by intracranial evidence that the N400 arises from similar waves of activity in multiple brain areas, particularly in the temporal and prefrontal cortices, during the retrieval of information from semantic memory.

5. Models of the Cortical ERP The traditional approach to the analysis of transient ERPs is to consider the ERP as a characteristic waveform that occurs in relation to a behaviorally significant discrete event, and to consider the recorded single-trial field potential as composed of activity that is both associated (“ERP signal”) and not associated (“noise”) with the event. Thus, averaging single-trial field potential time series, in temporal registration with the event, is commonly employed to extract the ERP from the nonevent-related noise. Justification for this analytic procedure is provided by a standard model, entailing three important assumptions (15). First, although the amplitude and latency of a component may be affected by a host of different conditions, once the conditions are fixed the component itself is considered to be an entirely reproducible signal whose waveform does not vary from one trial to the next. Second, the component is considered to be unitary in form, meaning that it is not composed of more basic waveforms. Third, the signal is considered to be completely independent of any ongoing neural processes, which if they exist are simply treated as additive uncorrelated noise. Despite the fact that this standard model has proven to be remarkably robust, evidence suggests that each of these three assumptions requires modification (66, 67). First to be considered are findings of variability in both component amplitude and latency across trials, even under controlled conditions. Trial-totrial variability of component amplitude does not seriously affect signal detection by ensemble averaging as long as the polarity

182

Bressler

remains constant: the averaged component retains the latency and shape of the single-trial signal despite variations in amplitude. Ensemble averaging is also robust to latency variability, providing that the signal waveform does not contain polarity reversals: the averaged component degrades smoothly as the degree of latency variability increases. Changes in component polarity over trials, or latency variability of component containing polarity reversals, may result in destructive cancellation of the signal during the ensemble averaging process. This possibility is particularly relevant in the case of high-frequency oscillatory field potential responses of the brain to sensory stimulation (65). These oscillatory components are said to be induced rather than evoked because the oscillatory waves within the component have variable latency (or phase), although the onset times of the component itself may be relatively constant with respect to the stimulus event. The high-frequency oscillatory waveform of these induced signals guarantees that polarity reversals will destructively cancel with even a small degree of trial-totrial latency variability. Therefore, the standard model cannot be assumed to hold in the case of induced high-frequency oscillations, and methods other than ensemble averaging are required for signal detection. Other studies have brought into question the second assumption from the standard model that ERPs represent fundamental, indecomposable waveforms. Some investigators have argued that evoked oscillations are fundamental to brain function, and that ERPs arise from the summation of evoked oscillations of different frequencies (34). Others have suggested instead that evoked Gaussian potentials are fundamental basis functions that combine to form ERP components (47). Both the second and third assumptions of the standard model are violated by evidence indicating that ongoing activity at the time of a stimulus may contribute to the poststimulus ERP (44). It is generally agreed that the EEG and MEG contain ongoing oscillatory activity at specific frequencies, and that the phases of ongoing oscillations at the time of the stimulus are random from trial to trial. If the stimulus acted to reset the phase of these oscillations to the same value on each trial, summation of poststimulus phase-aligned waves would result when the trials were averaged. To distinguish phase resetting models from the standard model requires techniques that allow comparison of prestimulus and poststimulus activity on a single-trial basis. In summary, a wealth of knowledge about ERPs has been obtained from the ensemble averaging of single-trial time series, under assumptions of the standard model of ERP generation. Nonetheless, substantial evidence indicates that this model is not completely adequate, and that exclusive reliance on ensemble averaging may obscure important issues relating to the genesis

Event-Related Potentials of the Cerebral Cortex

183

and functional relevance of ERPs. To provide a more detailed understanding of ERP component structure than is available by ensemble averaging, additional methods are required to derive information from the entire ensemble of single-trial time series. Frequency domain analysis is a class of methods that quantifies oscillatory activity in the trial ensemble in terms of its frequency, amplitude, and phase. It may be advantageous for addressing issues such as the occurrence of induced oscillations and the phase reorganization of ongoing oscillations. These considerations emphasize the need for the more sophisticated forms of ERP analysis reviewed in the next section.

6. Analysis of the Cortical ERP A general problem in the investigation of ERPs is that field potential recordings, whether LFP, ECoG, or EEG, most often contain a combination of potentials, in unknown proportions, from multiple sources. (The same problem applies to the ERF.) Thus, in addition to the ERP, which is derived from specific neuronal populations associated with a behavioral event, the field potential typically also contains potentials derived from the more general field activity of larger populations. Thus, a primary task of all ERP studies is to extract invariant features of the event-related activity from a set of field potential recordings. This section deals with different methodologies by which this is accomplished. 6.1. Time Domain Analysis

The most common form of ERP analysis is performed in the time domain by measurement of the amplitudes and latencies of components identified in the waveform averaged over an ensemble of single-trial time series. This approach necessarily involves a loss of information about the distributional properties of the ensemble of single trials. Various alternative time domain techniques have been devised for analysis of the entire ensemble of single-trial time series rather than the single-ensemble average. One general class of time domain analysis operates by application of a weighting function, or filter, directly to the single-trial time samples. The filter is applied either to the entire time series of each trial or to segments of it. One such weighting function, called the minimum mean square error filter (or “Wiener” filter), is designed to selectively enhance the ERP signal while suppressing the noise (28). Of course, to implement such a filter requires knowledge of the signal, which may be problematic. A related technique is matched filtering, in which the average ERP is used as a template that is “matched” by correlation with the single-trial time series. Matched filtering has been used to improve the average ERP by recreating it from “latency-corrected” trials (71), as

184

Bressler

well as to correlate the distributions of single-trial latency and amplitude values with behavioral or other parameters (40, 66, 67). Another weighting function is the band-pass filter, which is used to extract ERP components on the basis of their frequency characteristics. This approach is related to frequency domain analysis, which is described later. Similar to band-pass filtering are single-trial filtering techniques based on wavelet transforms (8). Time domain filtering techniques have been used in a variety of applications to investigate ERP function and composition. Statistical feature extraction represents another approach to understanding the ERP from the analysis of ensembles of singletrial time series data. Principal components analysis (PCA) is one such method that has been used to extract overlapping (principal) ERP components based on the inherent variability structure of the data set (16, 18). This variability occurs over time in the trial, over the ensemble of trials, over simultaneously sampled locations, and over different experimental conditions. The method involves computing the eigenvalues and eigenvectors of the covariance matrix of the original time series data, with the principal components being derived from the eigenvectors that are located along the directions of maximal data variance. Thus, the principal components reflect the intrinsic morphology of the single-trial waveforms, rather than a predetermined set of basis functions as in most filtering techniques. However, because of the orthogonality of the components that comes from the eigenvector decomposition, a potential problem with the PCA procedure is the misallocation of variance between the components (70). Various proposals have been made for overcoming this and other problems with the application of PCA to ERPs (35). A related approach that avoids the imposed component orthogonality constraint of PCA is independent component analysis (ICA) (63). 6.2. Frequency Domain Analysis

A second general class of analysis operates in the frequency domain by converting a series of amplitude values over time into a series of amplitude values over frequency through application of the discrete Fourier transform. Frequency domain analysis, or spectral analysis, provides greater insight into the organization and function of ERPs than is available from time domain techniques alone. It can distinguish a time period during which the ERP consists of oscillatory activity at particular frequencies, i.e., narrow-band, from that in which the activity is distributed over a wide range of frequencies, i.e., broad-band. Since the use of frequency domain analysis assumes that the time series is stationary, additional procedures are required to analyze transient ERPs, which are inherently nonstationary. Combined time-frequency analysis is performed using a short time window that is moved over a longer time period of

Event-Related Potentials of the Cerebral Cortex

185

interest, for example, the delay period in a working memory task. Time-frequency analysis seeks to provide an adequate representation of the temporal evolution of frequency components in the ERP. A number of time-frequency analysis methods have been advanced, including multitaper (48), wavelet or Hilbert transform (9), and parametric modeling techniques (13, 17, 20, 42). 6.3. Spatial Analysis

The neural processes underlying ERP generation are extended in space within the brain. Spatial analysis is therefore an important tool for understanding the relations between ERPs recorded at different spatial locations within or outside the brain (39). When simultaneous recordings are obtained with a sufficiently dense grid of electrodes on the brain or at the scalp, a basic form of spatial analysis, called topographic mapping, may be performed. The topographic distribution is an important feature of ERP components that complements other features such as latency, amplitude, polarity, and frequency content. Components that may otherwise be difficult to disambiguate, in some cases, may be easily distinguished by their topographic distributions. In fact, topographic patterning of the ERP may be fundamentally related to brain state at spatial scales from microscopic to macroscopic (23). Beyond simple topographic mapping, spatial analysis of the ERP takes the form of spatial spectral analysis (24), spatiotemporal PCA analysis (25), and inverse transformation to obtain estimates of cortical sources (14).

6.4. Interdependency Analysis

The ERP may be defined not only by its timing, frequency content, and spatial distribution, but also by higher-order features. Interdependency measures attempt to characterize the similarity structure of the waveform morphology of ERPs recorded at different spatial locations. Interdependency analysis is performed in either the temporal or frequency domain. In either case, stationarity considerations dictate that when applied to ERPs the analysis be performed in short time windows. A relatively simple tool for interdependency analysis in the time domain is the cross-correlation function, which measures linear relations between different time series. Nonlinear relations are characterized by measures based on the concept of mutual information (46). A useful technique for interdependency analysis in the frequency domain is parametric spectral estimation (13, 17, 20). This method, which allows spectral interdependency quantities such as multiple coherence and ordinary coherence to be computed from autoregressive model parameters, has been proven advantageous in the investigation of oscillatory network interdependencies when applied to simultaneously recorded ERPs from a distributed set of recording sites in the brain. The multiple coherence assesses the interdependency of each recording site with the group of all the other sites, whereas the ordinary

186

Bressler

coherence gauges the interdependency between two specific sites. Time-frequency analysis allows measurement of event-related interdependency between ERPs recorded at different cortical sites. ERP interdependency in different frequency ranges has been identified as a neural correlate of basic sensory and motor processes, as well as higher cognitive processes such as perception and recall of semantic entities (6, 68). Another approach to interdependency analysis derives from the concept of causal influence. Wiener (69) proposed that, for two simultaneously measured time series, one series may be considered causal to the other if knowledge of the first allows better prediction of the second. This concept was adopted and formalized by Granger (31) in the context of autoregressive models of stochastic processes. Specifically, if the variance of the prediction error for the second process at a given time is reduced by including past measurements from the first in the second’s autoregressive model, then the first process is considered to have a Granger causal influence on the second process. Causal influences may be characterized in the time domain, or in the frequency domain, using a spectral decomposition for Granger causality derived by Geweke (29). The Geweke spectral measure is useful for revealing important aspects of event-related oscillatory network dynamics ((7); Fig. 4). The possibility that the measured Granger causality from one process to a second is actually the result of common driving from a third process cannot be excluded when that third process is not recorded. However, when a third process is recorded, the technique of conditional Granger causality analysis can be used to unequivocally determine whether that third process is responsible for Granger causality between the first two (11).

Fig. 4. Interdependency analysis of the sensorimotor cortex of the right hemisphere of a macaque monkey Network interactions during a maintained manual postural position are shown by graphs of ordinary coherence (left) and Granger causality (right) based on oscillatory LFPs in the beta frequency range. (Modified from Brovelli et al. (7)).

Event-Related Potentials of the Cerebral Cortex

187

7. Discussion The cortical ERP reflects the coordinated behavior of large numbers of neurons in relation to a meaningful externally or internally generated event. It is an important neural signal that provides a window onto the dynamics of sensory, motor, and cognitive processing in the brain, and its usefulness can be heightened by careful applications of the analytic techniques described earlier. Single neurons are coordinated in the operations of neuronal populations, whose dynamics may effectively be studied at the mesoscopic and macroscopic levels of organization by ERP analysis. ERP analysis is thus an indispensable complement to single-cell neurophysiology and whole-head neuroimaging techniques. In animal recording, ERPs allow access to the dynamics of neuronal population activity that cannot be achieved with unitrecording techniques. In human recording, ERPs allow access to changes in brain activity on the order of milliseconds that cannot be achieved with hemodynamic-based neuroimaging techniques. Other recording methodologies with temporal resolution comparable to the ERP, such as voltage-sensitive optical imaging and direct magnetic resonance imaging of neuronal magnetic fields, are currently under development. However, even if these other modes of recording should eventually supplant the ERP, the same types of functional consideration and analytic technique discussed in this article will nevertheless apply to them as well. References 1. Alain C, Woods DL, Knight RT (1998) A distributed cortical network for auditory sensory memory in humans. Brain Res 812:23–37. 2. Basar E (1980) EEG-brain dynamics, Elsevier, Amsterdam. 3. Berger H (1929) Über das elektrenkephalogramm des menschen (On the human electroencephalogram). Archiv f Psychiatrie U Nervenkrankheiten 87:527–570. 4. Bressler SL (2002) Understanding cognition through large-scale cortical networks. Curr Dir Psychol Sci 11:58–61. 5. Bressler SL (2002) Event-related potentials. In: The handbook of brain theory and neural networks. Arbib MA (Ed), MIT, Cambridge, MA. pp 412–415. 6. Bressler SL, Kelso JA (2001) Cortical coordination dynamics and cognition. Trends Cogn Sci 5:26–36. 7. Brovelli A, Ding M, Ledberg A, Chen Y, Nakamura R, Bressler SL (2004) Beta oscillations

8.

9.

10. 11.

12.

in a large-scale sensorimotor cortical network: directional influences revealed by granger causality. Proc Natl Acad Sci USA 101:9849–9854. Browne M, Cutmore TR (2000) Adaptive wavelet filtering for analysis of event-related potentials from the electro-encephalogram. Med Bio Eng Comput 38:645–652. Bruns A (2004) Fourier-, Hilbert- and wavelet-based signal analysis: are they really different approaches? J Neurosci Meth 137: 321–332. Caton R (1875) The electric currents of the brain. Br Med J 2:278. Chen Y, Bressler SL, Ding M (2006) Fre‑ quency decomposition of conditional Granger causality and application to multivariate neural field potential data. J Neurosci Meth 150:228–237. Coles MGH, Rugg MD (1997) Event-related brain potentials: an introduction. In: Electrophysiology of mind: event-related

188

13.

14.

15.

16.

17.

18.

19. 20.

21. 22. 23. 24.

25.

Bressler brain potentials and cognition, Rugg MD, Coles MGH (Eds), Oxford University, Oxford, pp 1–26. Cui J, Xu L, Bressler SL, Ding M, Liang H (2008) BMSMART: a MATLAB/C toolbox for analysis of multichannel neural time series. Neural Networks 21:1094–1104. Darvas F, Pantazis D, Kucukaltun-Yildirim E, Leahy RM (2004) Mapping human brain function with MEG and EEG: methods and validation. Neuroimage 23:S289–S299. Dawson GD (1954) A summation technique for the detection of small evoked potentials. Electroenceph Clin Neurophysiol 6: 153–154. Dien J, Beal DJ, Berg P (2005) Optimizing principal components analysis of event-related potentials: matrix type, factor loading weighting, extraction, and rotations. Clin Neurophysiol 116:1808–825. Ding M, Bressler SL, Yang W, Liang H (2000) Short-window spectral analysis of cortical event-related potentials by adaptive multivariate autoregressive modeling: data preprocessing, model validation, and variability assessment. Biol Cybern 83:35–45. Donchin E, Heffley E (1979) Multivariate analysis of event-related potential data: a tutorial review. In: Multidisciplinary perspectives in event-related potential research (EPA 600/9-77-043), Otto D (Ed), US Government Printing Office, Washington, DC, pp 552–572. Elul R (1972) The genesis of the EEG. Int Rev Neurobiol 15:227–272. Franaszczuk PJ, Blinowska KJ, Kowalczyk M (1985) The application of parametric multichannel spectral estimates in the study of electrical brain activity. Biol Cybern 51:239–247. Freeman WJ (1975) Mass action in the nervous system. Academic Press, New York. Freeman WJ (2000) Mesoscopic neurodynamics: from neuron to brain. J Physiol Paris 94:303–322. Freeman WJ (2003) The wave packet: an action potential for the 21st century. J Integr Neurosci 2:3–30. Freeman WJ, Rogers LJ, Holmes MD, Silbergeld DL (2000) Spatial spectral analysis of human electrocorticograms including the alpha and gamma bands. J Neurosci Methods 95:111–121. Fuchs A, Jirsa JK, Kelso JA (2000) Theory of the relation between human brain activity (MEG) and hand movements. Neuroimage 11:359–369.

26. Fuster JM (1985) The prefrontal cortex, mediator of cross-temporal contingencies. Hum Neurobiol 4:169–179. 27. Fuster JM (2003) Cortex and mind: unifying cognition. Oxford University Press, Oxford. 28. Gevins AS (1987) Overview of computer analysis. In: Methods of analysis of brain electrical and magnetic signals, Gevins AS, Rémond A (Eds), Elsevier, Amsterdam, pp 31–83. 29. Geweke J (1982) Measurement of linear dependence and feedback between multiple time series. J Am Stat Assoc 77:304–313. 30. Gloor P (1985) Neuronal generators and the problem of localization in electroencephalography: application of volume conductor theory to electroencephalography. J Clin Neurophysiol 2:327–354. 31. Granger CWJ (1969) Investigating casual relations by econometric models and crossspectral methods. Econometrica 37: 424–438. 32. Hamalainen MS, Hari R (2002) Magnetoencephalographic characterization of dynamic brain activation: basic principles and methods of data collection and source analysis. In: Brain mapping: the methods, Toga AV, Mazziotta JC (Eds), Academic Press, San Diego, pp 227–254. 33. Herrmann CS (2001) Human EEG responses to 1–100 Hz flicker: resonance phenomena in visual cortex and their potential correlation to cognitive phenomena. Exp Brain Res 137:346–353. 34. Karakaş S, Erzengin OU, Başar E (2000) The genesis of human event-related responses explained through the theory of oscillatory neural assemblies. Neurosci Lett 285:45–48. 35. Kayser J, Tenke CE (2005) Trusting in or breaking with convention: towards a renaissance of principal components analysis in electrophysiology. Clin Neurophysiol 116: 1747–1753. 36. Kok A (2001) On the utility of P3 amplitude as a measure of processing capacity. Psychophysiology 38:557–577. 37. Kutas M, Federmeier KD (2000) Electrophysiology reveals semantic memory use in language comprehension. Trends Cogn Sci 4:463–470. 38. Ledberg A, Bressler SL, Ding M, Coppola R, Nakamura R (2007) Large-scale visuomotor integration in the cerebral cortex. Cereb Cortex 17:44–62. 39. Lehmann D (1987) Principles of spatial analysis. In: Methods of analysis of brain electrical

Event-Related Potentials of the Cerebral Cortex

40.

41.

42.

43.

44. 45.

46.

47.

48. 49.

50.

51.

and magnetic signals Gevins AS, Rémond A (Eds), Elsevier, Amsterdam, pp 309–354. Liang H, Bressler SL, Ding M, Truccolo WA, Nakamura R (2002) Synchronized activity in prefrontal cortex during anticipation of visuomotor processing. Neuroreport 13:2011–2015. Lopes da Silva F (1991) Neural mechanisms underlying brain waves: from neural membranes to networks. Electroenceph Clin Neurophysiol 79:81–93. Lopes da Silva FH, Mars NJI (1987) Parametric methods in EEG analysis. In: Methods of analysis of brain electrical and magnetic signals, Gevins AS, Rémond A (Eds), Elsevier, Amsterdam, pp 243–260. Lopes da Silva F, Van Rotterdam A (1982) Biophysical aspects of EEG and MEG generation. In: Electroencephalography: basic principles, clinical applications and related fields Niedermeyer E, Lopes da Silva F (Eds), Lippincott Williams & Wilkins, Baltimore, MD, pp 15–26. Makeig S, Debener S, Onton J, Delorme A (2004) Mining event-related brain dynamics. Trends Cogn Sci 8:204–210. Mangun GR, Hillyard SA (1997) Mechanisms and models of selective attention. In: Electrophysiology of mind: event-related brain potentials and cognition. Rugg MD, Coles MGH (Eds), Oxford University Press, Oxford, pp 40–85. Mars NJI, Lopes da Silva FH (1987) EEG analysis methods based on information theory, In: Methods of analysis of brain electrical and magnetic signals. Gevins AS, Rémond A (Eds), Elsevier, Amsterdam, pp 297–307. Melkonian D, Gordon E, Bahramali H (2001) Single-event-related potential analysis by means of fragmentary decomposition. Biol Cybern 85:219–229. Mitra PP, Pesaran B (1999) Analysis of dynamic brain imaging data. Biophys J 76:691–708. Müller MM, Andersen S, Trujillo NJ, ValdesSosa P, Malinowski P, Hillyard SA (2006) Feature-selective attention enhances color signals in early visual areas of the human brain. Proc Natl Acad Sci USA 103:14250–14254. Murakami S, Okada Y (2006) Contributions of principal neocortical neurons to magnetoencephalography and electroencephalography signals. J Physiol 575:925–936. Näätänen R (2008) Mismatch negativity (MMN) as an index of central auditory system plasticity. Int J Audiol 47:S16–S20.

189

52. Nowak L, Bullier J (1998) The timing of information transfer in the visual system. In: Cerebral cortex. Kaas JH, Rockland K, Peters A (Eds), Plenum, New York. pp 205–241. 53. Okada YC, Wu J, Kyuhou S (1997) Genesis of MEG signals in a mammalian CNS structure. Electroencephalogr Clin Neurophysiol 103:474–485. 54. Pantev C, Elbert T, Lutkenhoner B (1994) Oscillatory event-related brain dynamics. Plenum, New York. 55. Pfurtscheller G, Lopez da Silva FH (1999) Event-related EEG/MEG synchronization and desynchronization: basic principles. Clin Neurophysiol 110:1842–1857. 56. Rosahl SK, Knight RT (1995) Role of prefrontal cortex in generation of the contingent negative variation. Cereb Cortex 5:123–134. 57. Rugg MD, Coles MGH (1997) Electrophysiology of mind: event-related brain potentials and cognition. Oxford University, Oxford. 58. Schroeder CE, Mehta AD, Givre SJ (1998) A spatiotemporal profile of visual system activation revealed by current source density analysis in the awake macaque. Cereb Cortex 8:575–592. 59. Silberstein RB, Ciorciari J, Pipingas A (1995) Steady-state visually evoked potential topography during the Wisconsin card sorting test. Electroencephalogr Clin Neurophysiol 96:24–35. 60. Silberstein RB, Nunez PL, Pipingas A, Harris P, Danieli F (2001) Steady state visually evoked potential (SSVEP) topography in a graded working memory task. Int J Psychophysiol 42:219–232. 61. Speckmann EJ (1997) Generation of field potentials in the brain. J Clin Pharmacol 37:8S–10S. 62. Squires NK, Squires K, Hillyard SA (1975) Two varieties of long-latency positive waves evoked by unpredictable stimuli in man. Electroencephalogr Clin Neurophysiol 38:387–401. 63. Stone JV (2002) Independent component analysis: an introduction. Trends Cogn Sci 6:59–64. 64. Sutton S, Braren M, Zubin J, John ER (1965) Evoked-potential correlates of stimulus uncertainty. Science 150:1187–1188. 65. Tallon-Baudry C, Bertrand O, Delpuech C, Pernier J (1996) Stimulus specificity of phaselocked and non-phase-locked 40 Hz visual responses in human. J Neurosci 16:4240–4249.

190

Bressler

66. Truccolo WA, Ding M, Knuth KH, Nakamura R, Bressler SL (2002) Trial-to-trial variability of cortical evoked responses: implications for the analysis of functional connectivity. Clin Neurophysiol 113:206–226. 67. Truccolo WA, Knuth KH, Shah A, Bressler SL, Schroeder CE, Ding M (2003) Estimation of Single-trial multicomponent ERPs: differentially variable component analysis (dVCA). Biol Cybern 89:426–438. 68. Varela F, Lachaux JP, Rodriguez E, Martinerie J (2001) The brainweb: phase synchronization and large-scale integration. Nat Rev Neurosci 2:229–239.

69. Wiener N (1956) The theory of prediction. In: Modern mathematics for engineers, Series 1 Beckenback EF (Ed), McGraw-Hill, New York. 70. Wood CC, McCarthy G (1984) Principal component analysis of event-related potentials: simulation studies demonstrate misallocation of variance across components. Electroencephalogr Clin Neurophysiol 59:249–260. 71. Woody CD (1967) Characterization of an adaptive filter for the analysis of variable latency neuroelectric signals. Med Biol Eng 5:539–553.

Chapter 8 Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow Within Papez’s Circuit Zimbul Albo, Gonzalo Viana Di Prisco, and Robert P. Vertes Abstract Understanding the temporal and spectral structure of neural coding with especial attention to regional specificity and behavioral function is one goal of system neuroscience. Several methodological approaches have been used to analyze signals arising from multisite or distributed probes placed in distant areas to study synchrony and interaction in various species. Neuronal synchronization may be a way to enhance neuronal interactions among neural ensembles but the exact nature of this process remains largely unknown. Spike-field coherence has recently become a popular method when functional integration analysis is being considered for simultaneously collected hybrid signals, such as local field potentials and spike trains. In this chapter, we review some of the most recent approaches and applications of this methodology to address neural circuitry function and behavioral significance. Many authors have contributed extensively to our current understanding of synchronous signals in relation to neural interaction but far more is to occur in future years when both data acquisition and analysis techniques continue to expand. One intriguing and fascinating process to address using these techniques is undoubtedly the hippocampal theta rhythm. Its relevance to brain information processing and behavior makes it both an excellent target process to understand neural states in relation to behavioral significance and a source of physiologically complex integrative signal as many brainstem, diencephalic, and cortical structures appear to contribute to its generation and maintenance. At the end of this chapter, we discuss our own work on neuronal synchronization and resonance within three structures of Papez’s circuit namely, hippocampus, anterior thalamus, and retrosplenial cortex, and discuss its importance for mnemonic function. Key words: Spike-field coherence, Synchrony, Papez’s circuit, Memory, Theta rhythm, Hippocampus, Anterior thalamus, Retrosplenial cortex, Multisite coherence, Information flow, Local field potential, Spike train

1. Introduction Brain functioning relies on the communication and interactions between neuronal ensembles within and across brain regions, that is cortical areas and subcortical nuclei. Brain nuclei and areas Robert P. Vertes and Robert W. Stackman Jr. (eds.), Electrophysiological Recording Techniques, Neuromethods, vol. 54, DOI 10.1007/978-1-60327-202-5_8, © Springer Science+Business Media, LLC 2011

191

192

Albo, Viana Di Prisco, and Vertes

interact directly or indirectly in the orchestration of normal and pathological behavior (1). To understand how the central nervous system (CNS) collects and processes information within appropriate time scales to produce an organized output (behavior), one needs to have knowledge about input and output relations of brain regions. Neural interactions within the limbic system are crucial for processes such as learning and memory (2–4), but the exact nature of such interplay is largely unknown. Novel detailed analysis of these neurophysiological processes studied through the use of multisite recording techniques has been proved very valuable for the understanding of complex circuits and behaviors in animals and humans. Current interest in the study of neural connectivity goes beyond the anatomical study of connections aimed at defining functional connectivity underlying neural processing. Different methods are used to approach the study of functional or effective neural connectivity. The development of techniques for massive unit recordings has called for methods to analyze simultaneous multiple unit recordings also known as spike train analysis (5–7). Similarly new techniques have been developed to study multisite recordings of brain oscillations, such as wavelet time-frequency analysis and phase synchronization (8). These methods will not be dealt with in this chapter. We will review mostly methodological approaches aimed at investigating the coherence between mixed or hybrid time series. Such signals are generally recorded in experiments in which some channels record unit activity (discrete point data processes) and other channels (sometimes with the same probe) record continuous local field potentials (LFPs). When both signals (spike trains and LFP) are recorded in a given site with the same microelectrode using different filter settings, LFP-spike coherence can be computed (9, 10). In this chapter, we will also discuss methodological approaches used to analyze signals arising from multisite or distributed probes, that is, one or more electrodes recording the LFP in a given structure such as the hippocampus and distant microelectrodes recording the spike train in another structure like the thalamus. In this case, we refer to this methodology as far field potential (FFP)-spike coherence (Fig. 1). This long-range communication between brain areas, as for example between the hippocampus and the cortex, is of utmost importance for functional integration but its nature is largely unknown (11). Classical methods which are currently used to study interactions such as pairwise crosscorrelation and spectral coherence give only limited information about a link between two structures. They cannot determine whether one structure is driving another or assess feedback between them. Rosenberg et al. (12) studied the effects of neural interaction in simulated data of up to six elements, using multivariate regression analysis for spike train data.

Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow

193

LFP-spike coherence

FFP-spike coherence

Fig. 1. The figure illustrates the difference between the coherence measured locally between the spike train of a neuron or groups of neurons and the local-field potential recorded with the same or with an adjacent microelectrode, which we refer as LFPspike coherence. And the one measured at a distance between the spike train of a neuron or groups of neurons in one structure, i.e. anterior thalamus, and the far field potential recorded in another structure, i.e. hippocampus or cortex, which we refer as FFP-spike coherence.

Their results were consistent with the idea of partial coherence (PC) analysis as a robust and powerful method to identify plausible patterns for neural circuits. The measurement of PC provides strong evidence for functional relations between sites when their coherences survive the partialization process. PC analysis per se measures the degree of interdependency but not directionality. Also, as we have recently shown that PC analysis is very sensitive to measurement noise (13). New methods of analysis have been employed to address these issues. The most popular linear model for time series interactions is the multivariate linear autoregressive model (MVAR). In the MVAR, the time evolution of the state of the system is described as a function of the number of past states known as the model order. MVAR not only takes into account the correlation within and between time series but also allows a measure of prediction or causality. Wiener (14) established that one time series can be considered causal if a second simultaneous measured time series can be better predicted by taking into account information from the first one. In statistical terms, we have the concept of

194

Albo, Viana Di Prisco, and Vertes

Granger causality when the variance of prediction error for the second time series decreases when past measurement of the first time series are included in the linear regression (15). Measurements with a spectral domain representation (16) lead to methods of directed coherence applied to EEG time series (130), partial directed coherence (PDC) (17), and the directed transfer function (DTF) (18). The statistical properties of PDC are discussed in (19). Sameshima and Baccala (17) studied multiunit thalamocortical activity in the rat using linear directed partial coherence (DPC) and found a predominant cortical driving over the thalamus with episodic thalamic feedback and increased coactivation. Local linear, nonlinear autoregressive models (LNLAR) (20) have also revealed directional influence of interactions between recording sites. Both studies were implemented for bivariate data.

2. Spike-Field Coherence The activity of a single neuron in an ensemble is usually coordinated with the activity of the ensemble or neurons in the neighborhood such as cortical columns. In many instances, the activity of a neuron becomes synchronized with the ensemble. It is generally believed that neuronal synchronization is a way to enhance neuronal interactions among neural ensembles (21, 22). When a volley of excitatory synaptic inputs arrives closer in time to a neural ensemble, those neurons will be depolarized and fire synchronously. Neural synchronization can be assessed by analyzing spike trains from several neurons in the ensemble. The synaptic currents generated by the afferent volley give rise to extracellular field potentials that can be recorded locally with the same microelectrode used to record spikes. The signal from the amplifier only needs to be split and filtered with appropriate bandpass settings. LFPs are believed to originate from excitatory and inhibitory postsynaptic potentials arising from volleys of incoming action potentials. LFPs are an average of synaptic field potentials from a large number of neurons (23–27). When a group of neurons receive a dense projection from a distantly located neural structure, there is likely to be some degree of coherence between a given cell spike train and the LFP recorded in the target structure. Similarly if a neuron fires in synchrony with many others in a given structure, the volley can elicit a correlated LFP in a target structure. Some initial approaches to long-range communication used spike-triggered averages of EMG potentials to relate motorrelated activity of neurons in various brain regions such as the red nucleus (28) and motor cortex (29). Coherence analysis has also

Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow

195

become a popular tool to study long-range interactions (30, 31). For instance, Baker and collaborators have measured coherence in hybrid signals within the brain by computing the coherence between single units in somatosensory/parietal areas and LFPs in the motor cortex (32, 33) or between single units recorded in the cerebellum with LFPs in motor cortex (34). In this chapter, we highlight studies of coherence of “hybrid” signals from two or more interconnected distant structures. Great advances have occurred in the field of neuroscience regarding the relationship between oscillatory activity in the EEG and simultaneously recorded single-unit discharge during behavior (35–38). This not only represents an intriguing event but may also be an important link between large scale neuronal population activation and action potentials as the fundamental computational elements of neuronal communication. Oscillatory inputs produce a rhythm for setting spike phases, which are potentially useful for coding (39–46). Aravamuthan et al. (47) have recently reported increases of spectral power in spike trains (at the pedunculopontine nucleus, PPN) as well as the coherence between PPN spiking and PPN LFP activity in the ~1 Hz range after unilateral 6-hydroxydopamine lesions of the medial forebrain bundle in rats. Using a urethaneanesthetized preparation, the authors showed increased spike-field coherence (SFC) of PPN spike trains and PPN and motor cortex LFPs in lesioned but not in intact rats, thus highlighting the importance of oscillations abnormalities within neural structures in animal models of Parkinson’s disease (Fig. 2). Another interesting application of coherence analysis comes from Castellanos et al. (48). Instead of SFC, the authors used measurements of tactile stimulus-driven coherence and demonstrated that electrical stimulation of the SI cortex increased stimulus-evoked coherence in about 60% of cells of nucleus gracilis in rats. They found no significant correlation between increments in firing rates and stimulus coherence, but there was a positive correlation with the amplitude of the peristimulus time histogram. They concluded that the cortical facilitation of activity may involve an appropriate ordering of the stimulus-evoked firing pattern, and that the enhancement of coherence was more relevant than an increase in the number of spikes elicited by the tactile stimulus to better account for the experimental data. Neuronal processing during behavioral performance in nonhuman primates has been also recently addressed using SFC analysis by several investigators. Behavior-related modulations in unit/multiunit activity and/or LFPs are best evaluated by correlating coupling between signals that are on different spatial scales. Wu et al. (38) analyzed SFC with a sliding window multitaper method from recordings from 16 microelectrodes in the monkey prefrontal cortex, examining performance-related changes in

196

Albo, Viana Di Prisco, and Vertes

Fig. 2. Coherence between pedunculopontine nucleus (PPN) neuron spiking and LFP in both the PPN and the FFP in the motor cortex (MCx) in both intact and 6-OHDA lesioned rats. Note the change in coherence in the ~1 Hz range. (Reprinted from Aravamuthan et al. (47 )).

time-varying SFC. Their multitaper method minimized spectral leakage, and its reliability for overcoming limitations of low spike rates was tested with a temporal coordination spike/LFP model. SFC was analyzed in sliding windows of 200-ms length, 20 ms offset correcting for spectral leakage and computing the grand average SFC across all pairs of individually movable platinumtungsten fiber microelectrodes (excluding signal pairs at the same electrodes) across the frequency of interest (5–70 Hz, frequency steps 5 Hz). SFC was then matched to behavioral performance generating smoothed time-frequency maps. They found that changes in SFC that most directly correlated with the behavioral performance of monkeys on their short-term memory task were those most strongly exhibited during transitional states. The most prominent modulation was observed in the gamma frequency band (25–70 Hz) during stimulus processing at the early part of the delay period for correct trials. Modulation in the lower frequencies (5–20 Hz) minimally reflected the stimulus processing and mainly consisted of a much earlier response to sample stimuli and a clear peak in the middle of the delay period during performance of correct trials.

Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow

197

In another recent study, Anderson and Sheinberg (9) examined the effects of temporal context and expectancy in attentional processing on neuronal activity in the inferior temporal cortex of macaque monkeys. Using a simple paradigm involving interchanging images as both cues and targets, the authors found increased SFC in the beta–gamma range for expected targets. Neuronal spiking and evoked LFPs taken simultaneously from the same recording microelectrodes allowed the computation of the SFC. In this case, SFC was calculated as the ratio of the power of the spike-triggered average to the average of the power spectra for each of the LFP samples comprising the spike-triggered average. Confidence intervals were assessed using the bootstrap technique. SFC and LFP analysis have also been used to study the characteristics in the temporal structure of neural coding of the parietal reach region (PRR) of the posterior parietal cortex of macaque monkeys, during planning and execution of reach movements and saccades (49). LFP and spikes were obtained simultaneously from the same recording electrode. The authors found that LFP signals were better suited in predicting the behavioral state of the animal when factored with spike activity. SFC was significantly enhanced in the 20–40 Hz range for reaches and in the <10 Hz for saccades. They utilized the multitaper spectral method used by other authors (5, 37, 50, 51) in their estimation of temporal structure in the LFP and the SFC. Theta oscillations have been found to have a systematic effect on single neuron activity in extrastriate visual cortex (area V4) during the performance of a working memory task in awake behaving monkeys involving the presentation of sample stimuli with varying degrees of contrast (52). In their spectral analysis of LFPs, trial-average spectrograms were computed for each contrast level to compare the fixation vs. the delay period. Energy increases in the delta, theta, and gamma band occurred for highcontrast stimuli. It was also found that V4 neurons discharged more spikes near their preferred angle of each theta cycle. The authors described a direct relationship between theta oscillations and the timing of discharge of single neurons during the delayed period such that single-unit activity varied systematically with the angle of the LFP theta oscillation and this effect correlated well with stimulus selectivity. Changes in firing rate of V4 neurons and an associated increase in coherence between spikes and the LFP in the gammafrequency range (30–50 Hz) have been found in area V4 when the focus of attention on a visual stimulus occurs within the receptive field of the neurons (53). More recently, Fries et al. (54) demonstrated zero-phase gamma-band coherence among spike trains of V4 neurons and also changes in the prestimulus pattern of synchronization with a relative increase in the gamma band (top–down interaction) occurring during selective attention.

198

Albo, Viana Di Prisco, and Vertes

It is known that neurons driven to fire at rates near gamma frequency tend to phase lock with gamma oscillations to enhance signal discrimination. Broadband gamma modulation enhances signal discrimination when the gamma band is not excessively broad and the noise coming from other sources is weak. Soteropoulos and Baker (34) also postulated that extensive oscillatory synchronization was critically important in somatosensory processing, pointing to the significant coherence between deep cerebellar nuclei (DCN) units and M1 motor cortex LFP oscillations bilaterally (contralateral M1:25/87 units; ipsilateral: 9/87 units) at 10–40 Hz in macaques. Averaged coherence between DCN units and contralateral M1 LFP showed a prominent 17-Hz coherence peak and an average phase of approximately -pi/2 radians, implying that the DCN units fired around the time of maximal depolarization of M1 cells. The lack of a time delay between DCN and M1 activity suggested that the cerebellum and cortex form a pair of phase coupled oscillators. Synchrony and coherence in neural circuits have also been approached through modeling. Using computer simulations of a Hodgkin-Huxley type neuron, Tiesinga et al. (55) observed that a neuron’s firing rate together with coherence of its synaptic inputs and outputs (spike trains) could be modulated by the synchrony of the inhibitory inputs. Effects of attention could be attributed to changes in the synchrony of local interneuron networks. When inhibitory synchrony increased, the degree of coherence between spiking neurons and synaptic input increased, but the firing rate either increased or remained the same. The firing rate modulation with inhibitory synchrony was highest when the input network oscillated in the gamma frequency range. Neurons in the visual cortex could be controlled by top–down inputs that regulated coherence in the activity of a local inhibitory networks paced at gamma frequencies. Coherent theta oscillatory activity has also been proposed to originate from synchronization of interconnected layer V intrinsic bursting cells by recurrent excitation using Hodgkin-Huxley-type simulations (56). Using an elegant model of network dynamics, Masuda and Doiron (57) set out to mathematically describe the role of gamma oscillations (using binomial statistics on each cell on the population) to enhance signal discrimination during behavioral paradigms of attention. The authors proposed that oscillatory activity of presynaptic neuronal populations enhances coding in a much faster time scale than pure decoding and integration of incoming spikes, therefore affecting downstream information. They also showed that additive or multiplicative attention paradigms, believed to modulate neuronal response properties, will improve signal discrimination by shifting from Poisson to binomial spike statistics. The fact that single neurons represent information not only in terms of firing rate but also phase-locking to field oscillations

Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow

199

has been explosively expanded in the last few years. Neuronal temporal phase coding in relation to memory in various regions of the human brain was also described recently (58). The authors analyzed spiking activity and LFPs from the same set of nine microwire electrodes at multiple intracranial depths and found various patterns of phase locking in different areas, suggesting oscillatory phase resetting as a mechanism to synchronize widespread brain regions at different frequency ranges during behavioral tasks. They propose that slow oscillations (theta/delta range) facilitate phase coding and higher frequency oscillations (gamma range) help decode combinations of simultaneously active neurons.

3. On Causality and Directionality The study of cause-effect relationships lies at the root of the scientific inquiry in natural sciences. The philosophical meaning of causality is subjected to debate, particularly in the neurosciences (59). Causality, in its most general term, denotes a directional relationship between one event (called cause) and another event (called effect), which is the consequence (result) of the first event. Causation could be interpreted as a deterministic, probabilistic, or a dynamic relationship. Time series analysis methods have become the main stream in the neurosciences. The notable mathematician and brain theorist, Nobert Wiener, proposed that for simultaneously recorded time series, one time series can be considered causal with respect to other if one can better predict the second time series by taking into account the first time series. Granger (15) took Wiener’s definition of causality and formalized it in mathematical terms, based on linear regression modeling of stochastic processes. Granger causality can be examined in the spectral domain by using Fourier analysis (16). During the last decade, there has been an increased interest in the notion of Granger causality in the neurosciences ((60, 61), also see chapters by Ding and Bressler in this volume). In the early 1970s, Gersh (129) proposed a new definition of causality: “Our definition of causality is a frequency domain application of the ideas of causality determined by partial regression analysis techniques…. Our definition of causality is that a single times series is causal relative to other times series if it uniquely explains the pairwise linear relationship between other time series over a relevant interval in the frequency domain, so that if the influence of a driving or causal time series is removed from the other time series that the residual time series are pairwise uncorrelated and incoherent over that frequency interval. We assert that however causality is defined, that when a time series has this

200

Albo, Viana Di Prisco, and Vertes

Fig. 3. (a) Ordinary (solid lines) and partial (dotted lines) coherence between hippocampal field potential (Hipp), retrosplenial cortex field potential (RCx), and two cells recorded in the anterior thalamus (ATh). In each row, the partial coherence is indicated for the two time series indicated after removing the influence of the third. Below Gersch causality diagrams indicate the identified driver. In the first case (left panels) is the Hipp whereas in the second case (right panels) is the RCx. (b) Population results for 16 cases showing the relationship between the normalized relative power for RCx (filled triangles-solid line) and for Hipp (filled squares-dashed line), and the partial coherence between RCx and ATh after removing Hipp (open squares-dashed line), and the partial coherence between Hipp and ATh after partialling out RCx (open triangles-solid line). The shadowed area corresponds to three cases where the relative power of RCx became more predominant than the relative Hipp power. Below on the left the partial coherence between RCx and ATh as function of the normalized Hipp power, and on the right the partial coherence between Hipp and ATh against normalized relative RCx power.

property of explaining the linear relationship between other time series it should be identified as being causal to the other time series.” We recently examined the effectiveness of Gersh causality (i.e. PC analysis) in situations in which data consists of signal + noise (Fig. 3). We showed that PC-based Gersh causality is extremely sensitive to signal-to-noise ratio with the most noise-free series being often identified as the driver. Our results have been reproduced and confirmed by others (62, 63). We stated that Gersh causality might be inappropriate (or not viable) since it does not deal with time structure whereas “causal inference is fundamentally a concept involving the temporal order of events… this is another reason why PC is not suited for identifying causal influence or direction of driving” ((13) see also (63)).

Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow

4. Time-Varying Activity, Nonlinearity, and Noise

201

There are three very important concepts that should be considered when using a particular method to analyze brain signals: timevarying activity, nonlinearity, and noise. Most of the traditional methods such as Fourier-based coherence depend on the stationarity of the signals during data acquisition but very often in many situations such as stimulus-dependent activity, brain signals are dynamically changing during their measurement. Applying methods that assume stationarity in those cases may lead to erroneous results. Also, most methods assume that the system behaves linearly; however, it is known that neural processes are nonlinear. A very comprehensive methodological review of nonlinear multivariate analysis of neurophysiological signals can be found in Pereda et al. (64). Although in many cases it is considered that a linear approximation is justified, one has to be careful not to neglect important nonlinearities. On the contrary, nonlinear methods can not be applied blindly and a justification is needed in most cases. In addition, nonlinear methods are known to be extremely sensitive to noise. Care should be taken to factor in noise levels, especially when one is dealing with hybrid signals with dissimilar signal-to-noise ratios. Noise can easily lead to erroneous conclusions not only when applying partial-coherence analysis (13), but also with Granger causality based methods. A trendy and acceptable way to evaluate the power of different methods is to apply several methods to the same data set and compare the results among them (65–69). A significant amount of information can be gained when the synthetic data are derived from a model in which different parameters such as noise levels can be systematically manipulated. In a recent series of papers, Schelter, Winterhalter, Timmer, and collaborators addressed these issues by comparing different multivariate signal processing methods on simulated and experimental data. For example, these authors compared PC analysis, a Granger causality index (GCI) in the time domain, partial directed coherence (PDC), and the DTF (65). They tested the methods on a simple linear vector autoregressive model with different noise levels, on a nonlinear chaotic Roessler system, and on a time-variant (nonstationary) model system. They reported that PDC showed the best performance in revealing the multivariate interaction structure. DTF could not distinguish between direct and indirect interactions. Since it is a symmetric measure, PC cannot resolve directions unless it is coupled with phase spectral analysis. However, if the spectral coherence is weak or restricted to a small frequency range, phase spectral analysis is of little use. PC cannot deal with time-variant interactions.

202

Albo, Viana Di Prisco, and Vertes

Gourevitch et al. (66) have also compared linear vs. nonlinear methods of Granger causality on simulated and experimental data sets. The presence of noise in the measurement of time series not only affects the assessment of causal influences using Gersh causality (13), but also using methods based on Granger causality. Schelter et al. (68) stated that partial directed coherence (PDC) cannot cope with observation noise. PDC relies on modeling the time series by a multivariate autoregressive (MVAR) process. Estimation of the MVAR parameters is a key step in the process. Noise can lead to a serious underestimation of the MVAR para‑ meters (67, 70). Schelter et al. (68), in a study of PC and partial directed coherence, pointed out that differences in the variance of the stochastic driving noise can lead to spurious interactions (see Schelter at al. (68) and Fig. 4b). They also showed an example

Fig. 4. (a) Spiking of a median raphe (MR) neuron that increased in rate of discharge and fired rhythmically to a tailpinched elicited theta in a urethane anesthetized rat. (b) Superimposed extracellular action potentials showing a narrow spike of ~1 ms. (c) ISI histogram showing clustering at two intervals (~20 and ~200 ms), reflecting inter and intraburst frequencies during theta. (d, e) Auto and crosscorrelograms depicting the rhythmical discharge of the cell (d) locked to theta (e). (f) Spectral and coherence plots showing peaks in the EES and unit channels at theta frequency (~5.6 Hz) and significant coherence (0.64) between EEG and unit signals at theta frequency.

Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow

203

from a tremor time series where the value of the PDC is affected by the signal-to-noise ratio (see Fig. 7 of Schelter at al. (68)). As an alternative, Schelter and collaborators propose a combination and extension of PDC with Kalman filtering-based state space modeling, which includes observation noise and explicitly also allows for detection of time-varying changes (67).

5. Hippocampal Theta Rhythm One neurophysiological process that has attracted considerable interest because of its relation to behavior, mainly memory and learning, is the hippocampal theta rhythm (71–75). Theta rhythm appears in the hippocampus when the animal is engaged in locomotion or during the processing of novel information from the environment, and it disappears during automatic behaviors or quiet resting. Although most of the experiments in hippocampal theta rhythm have been performed in rats and cats, brain wave oscillations at theta frequency have been recorded in primates including humans during tasks that suggest its involvement in encoding and retrieval of information about the environment (76–78). There is ample evidence that the hippocampus is vital for spatial cognition and memory (41, 79), and the contributions of theta phase coding to hippocampal place cell processing has been well established (80). The discovery of theta modulated grid (81) and border cells (46) in the entorhinal cortex has inspired many models of hippocampal/entorhinal processing of spatial information (82–84). However, how this information flows in a circuit to produce such behavioral output is largely unknown. The hippocampus is at the crossroad of many brain pathways including the “classic” closed circuit proposed by James Papez more than 50 years ago (hippocampus > mammillary body > anterior thalamus > limbic cortex (mostly cingulate and entorhinal) > hippocampus). Although Papez’s circuit was originally proposed as the anatomical substrate for emotion, there is now solid evidence that it plays a role in spatial memory. A complete understanding of this circuit requires the detailed knowledge of how different brain structures interact in the space, time, and frequency domains to produce a functionally organized output. The generation of the theta rhythm in the hippocampal formation depends to a large extent of its afferent input. GABAergic and cholinergic neurons in the medial septum (MS) fire in rhythmic fashion and drive neurons in the hippocampal formation. Cholinergic neurons fire during the positive phase of the dentate gyrus theta wave, whereas GABAergic neurons fire during the negative phase (85). The temporal orchestration of septal rhythmically bursting cells allows for excitation and inhibition to take

204

Albo, Viana Di Prisco, and Vertes

place in both hippocampal principal (pyramidal and granule) cells and interneurons (86). Fluctuations in the membrane potential of the pyramidal cells that result from the rhythmic bursting of GABAergic and cholinergic cells of the MS and the vertical limb of the diagonal band (MS/DBv) generate extracellular current flow and accordingly the oscillatory field potential pattern known as the theta rhythm. The MS/DBv network is under the control of brainstem and diencephalic nuclei (86) in particular the nucleus reticularis pontis oralis that provides a tonic drive (87) and the supramammillary nucleus (SUM) of the hypothalamus, which provides a phasic rhythmic theta drive to both the MS/DBv and to the hippocampus proper (88). As such theta cells (i.e. cells that fire in synchrony with the hippocampal theta rhythm) are not only present in the septohippocampal system (89–91) but also in the mammillary bodies (MB) and SUM (92–94), as well as in the posterior hypothalamic area (88, 95), median raphe nucleus (MR) (96), see Fig. 4), Gudden’s ventral tegmental nucleus (VTG) (97), posterior cingulate (98–102), and entorhinal cortices (103–105). Thetarelated cells have also been found in the superior colliculus (106) and in the lateral amygdala (107–110). Intrinsic theta-frequency oscillations of hippocampal CA1 interneurons located near the border between stratum lacunosum-moleculare and stratum radiatum described in the slice as interplay between inward Na+ currents and outward K+ currents could serve to rhythmically inhibit and synchronize pyramidal neurons during theta activity (111). The same authors further described a muscarinic induction of theta-frequency oscillations in this same region (CA1) in slices (112). Perirhinal neurons also exhibit intrinsic properties that could assist in the entrainment and synchronization of theta-frequency oscillations to possibly enhance the information transfer to the hippocampus (113). Theta intrinsic oscillations have also been recently found in parasubiculum (114) as well as subiculum (115). The more we understand theta rhythmic cell properties the closer we are in underpinning the neural significance of these cells in orchestrating system synchronization and behavior.

6. Theta Rhythmicity Within Papez’s Circuit

Eccles described the hippocampus as the dominant structure in the limbic circuit of Papez (116). Eccles also indicated that this circuit could function as a self-reexciting loop that provides selective background information to the hippocampus. Alonso and Llinas (93) stated that the actual generation of theta activity might be a distributed property of a network and might be located at

Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow

205

many sites both within Papez circuit and outside of it, including entorhinal cortex (Fig. 5). As mentioned earlier, Dickson et al. (103) described theta-related cells in EC, and showed that they were controlled by the same ascending brainstem synchronizing pathways that control “theta cells” in the hippocampus. Alonso and Garcia-Austt (104) had reported earlier that medial EC units fired rhythmically and synchronously with the hippocampal theta rhythm. There was a constant phase relationship between EC units and theta, and most theta-associated cells were located in the superficial layers (II–III) of this structure. They also described a “theta generator” in the medial EC (105) and therefore the possibility of multiple generators in Papez’s circuit is not unrealistic. Theta rhythmically firing neurons of the anteroventral nucleus of the thalamus (AV) (see further) might be part of the distributed rhythmic network as proposed for other structures of Papez’s circuit. Although Alonso and Garcia-Austt did not examine processes in the spectral domain, their results in the temporal domain are remarkably similar to those we found for the anterior thalamus (118). We examined theta-related single-unit activity in the anterior thalamus while simultaneously recording LFPs in the retrosplenial cortex and the dorsal hippocampus. About 70% of the recorded units were classified as theta-on cells. The majority of these cells (63%) showed rhythmicity at theta frequency and phase-locked firing with hippocampal theta (Fig. 5). In a subset of these cells, we examined the interrelation between AV unit activity and theta in the hippocampus and retrosplenial cortex (118). Furthermore, we analyzed the flow of information among these three structures, using PC analysis (Fig. 2) and the direct transfer function (DTF) (13). The concept of reentrant signaling among neuronal groups leading to coherent oscillatory activity, as proposed by Sporns et al. (119), may apply to the structures of the limbic system – or more specifically those of Papez’s circuit. Resonant interactions between the thalamus and cortex may explain some of the intriguing facets of these systems. In discussing the relevance of synchrony and oscillatory activity in limbic cortex (with especial emphasis on theta rhythm), Bland and Colom (72) emphasized the importance of resonant signaling on theta information flow. Entraining of a large subset of neural circuitry in limbic structures into a common processing mode is thought to selectively tune them for the reception of particular kinds of information, thereby, reinforcing the resonant hypothesis of theta rhythm. Perhaps theta-related behaviors require such a complex synchronous encoding. The proposed theory of multiple oscillators extends the MS pacemaker hypothesis since the theta signal might also resonate in non-septal structures such as EC, AV, MB, VTG, and others.

206

Albo, Viana Di Prisco, and Vertes

Fig. 5. Circuit diagram of the main connections between the brain regions that underlie episodic memory following Aggleton and Brown (3). Inset shows a theta modulated border cell recorded in the entorhinal cortex (Reprinted from Solstad et al. (46)). Below, discharge characteristics of a single cell in the AV of the ATh that fired rhythmically in bursts synchronous with the theta rhythm of the hippocampus (Reprinted Fig. 2 from Vertes et al. (117)) (a) Upper traces: recordings of the hippocampal EEG and unit activity before and during theta elicited with a tail pinch (horizontal bar). Lower traces: expanded record (from A) of a period after tail pinch showing a strong correlation between unit bursts and theta. (b, c) Autocorrelograms and crosscorrelograms (spike-triggered averaging) depicting the rhythmical discharge of the cell (b) locked to the theta rhythm (c) during theta but not control conditions. (d) spectral and cross-spectral (coherence) plots showing peaks in the EEG and unit signals at theta frequency and significant coherence between EEG and unit signals at theta frequency during theta (solid lines) but not during control conditions (dotted lines). Abbreviations: AD anterior dorsal thalamic nucleus; AM anterior medial thalamic nucleus; AV anterior ventral thalamic nucleus; MB mammillary bodies; MD medial dorsal thalamic nucleus.

Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow

207

Multiple oscillators along a circuit(s) may prove beneficial in preventing the decay in time and the lost of spectral power of the theta signal. Reverberating activity within reentrant neural circuits could result in the rhythmic activation of normally nonoscillatory neurons within well-defined frequency bands and also from coupling of intrinsically oscillatory structures – as has been demonstrated by nonlinear analyses (120). Resonant phenomena may have a functional role for the integration of behavior (121, 122) and functional integration evolving from fairly widespread synchronous activity has been recently proposed in large-scale brain interactions (123). Following this line of reasoning, it might be the case that the anterior thalamus is a semiautonomous theta resonator and not simply a relay. Reverberating activity not only enhances signal communication but can also complement the dynamic flow of information within circuits during relevant behaviors. Recently, Kocsis and Kaminski (124) found that the interactions between the SUM and the septohippocampal system can change in direction during different situations. Current theories of large-scale neuronal communication have emphasized the role of frequency synchronization on information transfer (21) to explain the flexible interaction needed for cognition. Axmacher et al. (125) recently addressed the issue of synchronization of neural assemblies at different frequency ranges as a possible explanation for stages of memory formation. We have demonstrated neuronal synchronization through coherence analysis in three structures of Papez’s circuit. In addition, a circuit with a high degree of topographic connectivity (i.e. the anterior thalamus) would allow for an anatomical–physiological integration of neural signals in space. Specifically, the anterior thalamus might receive theta rhythmic information from the septohippocampal complex (through MB and/or directly from the subicular complex) and relay it to cingulate cortex (101) and other association cortices (11, 126–128). Also, the idea of independent resonators in a large circuit is not only tantalizing but also integrates the concept of sharing neural information across scales – while accounting for cellular mechanisms. As mentioned earlier, MB neurons are considered as resonators in “theta rhythmic circuit” (93), and in addition MB receives projections from the ventral tegmental nucleus of Gudden (VTG). The medial MB is reciprocally connected with VTG, and VTG units have also been shown to be coherent with hippocampal theta (97) and consequently may feed oscillatory activity via MB to AV. The VTGMB-AV network could prove to be pivotal in the generation and maintenance of rhythmic theta oscillations. Moreover, the topographical relationship between MB and the Gudden’s nuclei could explain, at least in part, some of the differences in rhythmicity found in the different subdivisions of the anterior thalamus. The dorsal tegmental nucleus of Gudden (DTG) projects to the lateral MB and from there to AD, whereas VTG is interconnected

208

Albo, Viana Di Prisco, and Vertes

primarily with the medial MB and then to AV. Circuit connectivity, as the one described here, with superimposed spatial, temporal, and spectral resolution would enhance our knowledge of the physiological and behavioral functions it subserves.

7. Conclusion Hippocampal theta waves can have an entrainment effect of different cortical and subcortical structures and may play a key role in the orchestration of neural information flow underlying the organization and generation of behavior. Synchronization and coherence between cortical and hippocampal field potentials and spike firing in distant brain regions have proved valuable to understand dynamic interactions in Papez’s circuit. SFC is becoming an exciting methodology to study neural interactions in brain circuits. It has not only deepened our understanding of dynamic interactions among brain structures but also provided additional insight of the relation among different types of neural signals within a given circuit. More studies of this type are needed to advance our understanding of brain function.

Acknowledgments This work was supported in part by NIMH grant MH63519 and NSF grant IOS 0820639. References 1. Schnitzler A, Gross J. (2005) Normal and pathological oscillatory communication in the brain. Nat Rev Neurosci 6:285–296. 2. Vertes RP. (2006) Interactions among the medial prefrontal cortex, hippocampus and midline thalamus in emotional and cognitive processing in the rat. Neuroscience 142:1–20. 3. Aggleton JP, Brown MW. (2006) Interleaving brain systems for episodic and recognition memory. Trends Cogn Sci 10:455–463. 4. DeCoteau WE, Thorn C, Gibson DJ, Courtemanche R, Mitra P, Kubota Y, Graybiel AM. (2007) Learning-related coordination of striatal and hippocampal theta rhythms during acquisition of a procedural maze task. Proc Natl Acad Sci USA 104:5644–5649.

5. Jarvis MR, Mitra PP. (2001) Sampling properties of the spectrum and coherency of sequences of action potentials. Neural Comput 13:717–749. 6. Brown EN, Kass RE, Mitra PP. (2004) Multiple neural spike train data analysis: state-of-the-art and future challenges. Nat Neurosci 7:456–461. 7. Tetzlaff T, Rotter S, Stark E, Abeles M, Aertsen A, Diesmann M. (2008) Dependence of neuronal correlations on filter characteristics and marginal spike train statistics. Neural Comput 20:2133–2184. 8. Le Van Quyen M, Bragin A. (2007) Analysis of dynamic brain oscillations: methodological advances. Trends Neurosci 30:365–373. 9. Anderson B, Sheinberg DL. (2008) Effects of temporal context and temporal expectancy

Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow on neural activity in inferior temporal cortex. Neuropsychologia 46:947–957. 10. Berens P, Keliris GA, Ecker AS, Logothetis NK, Tolias AS. (2008) Feature selectivity of the gamma-band of the local field potential in primate primary visual cortex. Front Neurosci 2:199–207. 11. Sirota A, Montgomery S, Fujisawa S, Isomura Y, Zugaro M, Buzsaki G. (2008) Entrainment of neocortical neurons and gamma oscillations by the hippocampal theta rhythm. Neuron 60:683–697. 12. Rosenberg JR, Halliday DM, Breeze P, Conway BA. (1998) Identification of patterns of neuronal connectivity – partial spectra, partial coherence, and neuronal interactions. J Neurosci Methods 83:57–72. 13. Albo Z, Di Prisco GV, Chen Y, Rangarajan G, Truccolo W, Feng J, Vertes RP, Ding M. (2004) Is partial coherence a viable technique for identifying generators of neural oscillations? Biol Cybern 90:318–326. 14. Wiener N. (1956) The theory of prediction. In: Beckenbach EF (Ed), Modern mathematics for engineers. McGraw-Hill, New York pp 125–139. 15. Granger CWJ. (1969) Investigating causal relations by econometric models and crossspectral methods. Econometrica 37: 424–438. 16. Geweke J. (1982) Measurement of linear dependence and feedback between multiple time series. J Am Stat Assoc 77:304–313. 17. Sameshima K, Baccala LA. (1999) Using partial directed coherence to describe neuronal ensemble interactions. J Neurosci Methods 94:93–103. 18. Kaminski MJ, Blinowska KJ. (1991) A new method of the description of the information flow in the brain structures. Biol Cybern 65:203–210. 19. Schelter B, Winterhalder M, Schwab K, Leistritz L, Hesse W, Bauer R, Witte H, Timmer J. (2005) Quantification of directed signal transfer between neural networks by partial directed coherence: a novel approach to infer causal time-depending influences in noisy, multivariate time series. Preprint Series DFG SPP 1114:70. 20. Freiwald WA, Valdes P, Bosch J, Biscay R, Jimenez JC, Rodriguez LM, Rodriguez V, Kreiter AK, Singer W. (1999) Testing nonlinearity and directedness of interactions between neural groups in the macaque inferotemporal cortex. J Neurosci Methods 94:105–119.

209

21. Fries P. (2005) A mechanism for cognitive dynamics: neuronal communication through neuronal coherence. Trends Cogn Sci 9:474–480. 22. Womelsdorf T, Schoffelen JM, Oostenveld R, Singer W, Desimone R, Engel AK, Fries P. (2007) Modulation of neuronal interactions through neuronal synchronization. Science 316:1609–1612. 23. Klee MR, Offenloch K, Tigges J. (1965) Cross-correlation analysis of electroencephalographic potentials and slow membrane transients. Science 147:519–521. 24. Creutzfeldt OD, Watanabe S, Lux HD. (1966) Relations between EEG phenomena and potentials of single cortical cells. I. Evoked responses after thalamic and erpicortical stimulation. Electroencephalogr Clin Neurophysiol 20:1–18. 25. Mitzdorf U. (1985) Current source-density method and application in cat cerebral cortex: investigation of evoked potentials and EEG phenomena. Physiol Rev 65:37–100. 26. Contreras D, Steriade M. (1995) Cellular basis of EEG slow rhythms: a study of dynamic corticothalamic relationships. J Neurosci 15:604–622. 27. Mahon S, Deniau JM, Charpier S. (2001) Relationship between EEG potentials and intracellular activity of striatal and corticostriatal neurons: an in vivo study under different anesthetics. Cereb Cortex 11:360–373. 28. Cheney PD. (1980) Response of rubromotoneuronal cells identified by spike-triggered averaging of EMG activity in awake monkeys. Neurosci Lett 17:137–142. 29. Fetz EE, Cheney PD, German DC. (1976) Corticomotoneuronal connections of precentral cells detected by postspike averages of EMG activity in behaving monkeys. Brain Res 114:505–510. 30. Houk JC, Dessem DA, Miller LE, Sybirska EH. (1987) Correlation and spectral analysis of relations between single unit discharge and muscle activities. J Neurosci Methods 21:201–224. 31. Sosnoff JJ, Vaillancourt DE, Larsson L, Newell KM. (2005) Coherence of EMG activity and single motor unit discharge patterns in human rhythmical force production. Behav Brain Res 158:301–310. 32. Baker SN, Pinches EM, Lemon RN. (2003) Synchronization in monkey motor cortex during a precision grip task. II. Effect of oscillatory activity on corticospinal output. J Neurophysiol 89:1941–1953.

210

Albo, Viana Di Prisco, and Vertes

33. Witham CL, Wang M, Baker SN. (2007) Cells in somatosensory areas show synchrony with beta oscillations in monkey motor cortex. Eur J Neurosci 26:2677–2686. 34. Soteropoulos DS, Baker SN. (2006) Corticocerebellar coherence during a precision grip task in the monkey. J Neurophysiol 95:1194–1206. 35. Ramos A, Schwartz E, John ER. (1976) Evoked potential-unit relationships in behaving cats. Brain Res Bull 1:69–75. 36. Gray CM, Konig P, Engel AK, Singer W. (1989) Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which reflects global stimulus properties. Nature 338:334–337. 37. Pesaran B, Pezaris JS, Sahani M, Mitra PP, Andersen RA. (2002) Temporal structure in neuronal activity during working memory in macaque parietal cortex. Nat Neurosci 5:805–811. 38. Wu W, Wheeler DW, Staedtler ES, Munk MH, Pipa G. (2008) Behavioral performance modulates spike field coherence in monkey prefrontal cortex. Neuroreport 19:235–238. 39. Hopfield JJ. (1995) Pattern recognition computation using action potential timing for stimulus representation. Nature 376: 33–36. 40. Hopfield JJ. (2004) Encoding for computation: recognizing brief dynamical patterns by exploiting effects of weak rhythms on actionpotential timing. Proc Natl Acad Sci USA 101:6255–6260. 41. O’Keefe J, Recce ML. (1993) Phase relationship between hippocampal place units and the EEG theta rhythm. Hippocampus 3:317–330. 42. Fox SE, Wolfson S, Ranck JB, Jr. (1986) Hippocampal theta rhythm and the firing of neurons in walking and urethane anesthetized rats. Exp Brain Res 62:495–508. 43. Tsodyks MV, Skaggs WE, Sejnowski TJ, McNaughton BL. (1996) Population dynamics and theta rhythm phase precession of hippocampal place cell firing: a spiking neuron model. Hippocampus 6:271–280. 44. Skaggs WE, McNaughton BL, Wilson MA, Barnes CA. (1996) Theta phase precession in hippocampal neuronal populations and the compression of temporal sequences. Hippocampus 6:149–172. 45. King C, Recce M, O’Keefe J. (1998) The rhythmicity of cells of the medial septum/diagonal band of Broca in the awake freely moving rat: relationships with behaviour and hippocampal theta. Eur J Neurosci 10:464–477.

46. Solstad T, Boccara CN, Kropff E, Moser MB, Moser EI. (2008) Representation of geometric borders in the entorhinal cortex. Science 322:1865–1868. 47. Aravamuthan BR, Bergstrom DA, French RA, Taylor JJ, Parr-Brownlie LC, Walters JR. (2008) Altered neuronal activity relationships between the pedunculopontine nucleus and motor cortex in a rodent model of Parkinson’s disease. Exp Neurol 213:268–280. 48. Castellanos NP, Malmierca E, Nunez A, Makarov VA. (2007) Corticofugal modulation of the tactile response coherence of projecting neurons in the gracilis nucleus. J Neurophysiol 98:2537–2549. 49. Scherberger H, Jarvis MR, Andersen RA. (2005) Cortical local field potential encodes movement intentions in the posterior parietal cortex. Neuron 46:347–354. 50. Percival DB, Walden AT. (1993) Spectral analysis for physical applications: multitaper and conventional univariate techniques. Cambridge University Press, Cambridge, MA. 51. Zeitler M, Fries P, Gielen S. (2006) Assessing neuronal coherence with single-unit, multiunit, and local field potentials. Neural Comput 18:2256–2281. 52. Lee H, Simpson GV, Logothetis NK, Rainer G. (2005) Phase locking of single neuron activity to theta oscillations during working memory in monkey extrastriate visual cortex. Neuron 45:147–156. 53. Bichot NP, Rossi AF, Desimone R. (2005) Parallel and serial neural mechanisms for visual search in macaque area V4. Science 308:529–534. 54. Fries P, Womelsdorf T, Oostenveld R, Desimone R. (2008) The effects of visual stimulation and selective visual attention on rhythmic neuronal synchronization in macaque area V4. J Neurosci 28: 4823–4835. 55. Tiesinga PH, Fellous JM, Salinas E, Jose JV, Sejnowski TJ. (2004) Inhibitory synchrony as a mechanism for attentional gain modulation. J Physiol Paris 98:296–314. 56. Budd JM. (2005) Theta oscillations by synaptic excitation in a neocortical circuit model. Proc Biol Sci 272:101–109. 57. Masuda N, Doiron B. (2007) Gamma oscillations of spiking neural populations enhance signal discrimination. PLoS Comput Biol 3:e236. 58. Jacobs J, Kahana MJ, Ekstrom AD, Fried I. (2007) Brain oscillations control timing of single-neuron activity in humans. J Neurosci 27:3839–3844.

Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow 59. Freeman WJ. (1999) Consciousness, intentionality, and causality. J Conscious Stud 6:143–172. 60. Ding M, Chen Y, Bressler SL. (2006) Granger causality: basic theory and application to neuroscience. In: Handbook of time series analysis: recent theoretical developments and applications. Schelter B, Winterhalder M, Timmer J (Eds), WileyVCH, Weinheim, pp 451–474. 61. Cadotte AJ, DeMarse TB, He P, Ding M. (2008) Causal measures of structure and plasticity in simulated and living neural networks. PLoS ONE 3:e3355. 62. Govindan RB, Raethjen J, Arning K, Kopper F, Deuschl G. (2006) Time delay and partial coherence analyses to identify cortical connectivities. Biol Cybern 94:262–275. 63. Baccala LA, Sameshima K. (2006) Comments on ‘Is partial coherence a viable technique for identifying generators of neural oscillations?’: Why the term ‘Gersch Causality’ is inappropriate: common neural structure inference pitfalls. Biol Cybern 95:135–141. 64. Pereda E, Quiroga RQ, Bhattacharya J. (2005) Nonlinear multivariate analysis of neurophysiological signals. Prog Neurobiol 77:1–37. 65. Winterhalder M, Schelter B, Hesse W, Schwab K, Leistritz L, Klan D, Bauer R, Timmer J, Witte H. (2005) Comparison of linear signal processing techniques to infer directed interactions in multivariate neural systems. Signal Proc 85:2137–2160. 66. Gourevitch B, Bouquin-Jeannes RL, Faucon G. (2006) Linear and nonlinear causality between signals: methods, examples and neurophysiological applications. Biol Cybern 95:349–369. 67. Schelter B, Winterhalder M, Eichler M, Peifer M, Hellwig B, Guschlbauer B, Lucking CH, Dahlhaus R, Timmer J. (2006a) Testing for directed influences among neural signals using partial directed coherence. J Neurosci Methods 152:210–219. 68. Schelter B, Winterhalder M, Hellwig B, Guschlbauer B, Lucking CH, Timmer J. (2006b) Direct or indirect? Graphical models for neural oscillators. J Physiol Paris 99:37–46. 69. Smirnov D, Schelter B, Winterhalder M, Timmer J. (2007) Revealing direction of coupling between neuronal oscillators from time series: phase dynamics modeling versus partial directed coherence. Chaos 17 (1):013111. 70. Timmer J. (1998) Modeling noisy time series: physiological tremor. Int J Bifurcat Chaos 8:1505–1516.

211

71. Bland BH, Whishaw IQ. (1976) Generators and topography of hippocampal theta (RSA) in the anaesthetized and freely moving rat. Brain Res 118:259–280. 72. Bland BH, Colom LV. (1993) Extrinsic and intrinsic properties underlying oscillation and synchrony in limbic cortex. Prog Neurobiol 41:157–208. 73. Bland BH, Oddie SD, Colom LV. (1999) Mechanisms of neural synchrony in the septohippocampal pathways underlying hippocampal theta generation. J Neurosci 19:3223–3237. 74. Vinogradova OS. (1995) Expression, control, and probable functional significance of the neuronal theta-rhythm. Prog Neurobiol 45:523–583. 75. Hasselmo ME. (2007) Arc length coding by interference of theta frequency oscillations may underlie context-dependent hippocampal unit data and episodic memory function. Learn Mem 14:782–794. 76. Tsujimoto T, Shimazu H, Isomura Y. (2006) Direct recording of theta oscillations in primate prefrontal and anterior cingulate cortices. J Neurophysiol 95:2987–3000. 77. Rizzuto DS, Madsen JR, Bromfield EB, Schulze-Bonhage A, Kahana MJ. (2006) Human neocortical oscillations exhibit theta phase differences between encoding and retrieval. Neuroimage 31:1352–1358. 78. Raghavachari S, Lisman JE, Tully M, Madsen JR, Bromfield EB, Kahana MJ. (2006) Theta oscillations in human cortex during a working-memory task: evidence for local generators. J Neurophysiol 95:1630–1638. 79. Neves G, Cooke SF, Bliss TV. (2008) Synaptic plasticity, memory and the hippocampus: a neural network approach to causality. Nat Rev Neurosci 9:65–75. 80. Jensen O, Lisman JE. (2000) Position reconstruction from an ensemble of hippocampal place cells: contribution of theta phase coding. J Neurophysiol 83:2602–2609. 81. Moser EI, Kropff E, Moser MB. (2008) Place cells, grid cells, and the brain’s spatial representation system. Annu Rev Neurosci 31:69–89. 82. Hasselmo ME, Giocomo LM, Zilli EA. (2007) Grid cell firing may arise from interference of theta frequency membrane potential oscillations in single neurons. Hippocampus 17:1252–1271. 83. Burgess N. (2008) Grid cells and theta as oscillatory interference: theory and predictions. Hippocampus 18:1157–1174.

212

Albo, Viana Di Prisco, and Vertes

84. Blair HT, Gupta K, Zhang K. (2008) Conversion of a phase- to a rate-coded position signal by a three-stage model of theta cells, grid cells, and place cells. Hippocampus 18:1239–1255. 85. Brazhnik ES, Fox SE. (1997) Intracellular recordings from medial septal neurons during hippocampal theta rhythm. Exp Brain Res 114:442–453. 86. Vertes RP, Kocsis B. (1997) Brainstemdiencephalo-septohippocampal systems controlling the theta rhythm of the hippocampus. Neuroscience 81:893–926. 87. Vertes RP. (1981) An analysis of ascending brain stem systems involved in hippocampal synchronization and desynchronization. J Neurophysiol 46:1140–1159. 88. Kirk IJ, Oddie SD, Konopacki J, Bland BH. (1996) Evidence for differential control of posterior hypothalamic, supramammillary, and medial mammillary theta-related cellular discharge by ascending and descending pathways. J Neurosci 16:5547–5554. 89. Gaztelu JM, Buno W, Jr. (1982) Septohippocampal relationships during EEG theta rhythm. Electroencephalogr Clin Neurophysiol 54:375–387. 90. Colom LV, Bland BH. (1987) Statedependent spike train dynamics of hippocampal formation neurons: evidence for theta-on and theta-off cells. Brain Res 422:277–286. 91. Li S, Arbuthnott GW, Jutras MJ, Goldberg JA, Jaeger D. (2007) Resonant antidromic cortical circuit activation as a consequence of high-frequency subthalamic deep-brain stimulation. J Neurophysiol 98:3525–3537. 92. Kocsis B, Vertes RP. (1994) Characterization of neurons of the supramammillary nucleus and mammillary body that discharge rhythmically with the hippocampal theta rhythm in the rat. J Neurosci 14:7040–7052. 93. Alonso A, Llinas RR. (1992) Electrophysiology of the mammillary complex in vitro. II. Medial mammillary neurons. J Neurophysiol 68:1321–1331. 94. Pan WX, McNaughton N. (2004) The supramammillary area: its organization, functions and relationship to the hippocampus. Prog Neurobiol 74:127–166. 95. Bland BH, Konopacki J, Kirk IJ, Oddie SD, Dickson CT. (1995) Discharge patterns of hippocampal theta-related cells in the caudal diencephalon of the urethan-anesthetized rat. J Neurophysiol 74:322–333. 96. Viana Di Prisco G, Albo Z, Vertes RP, Kocsis B. (2002) Discharge properties of neurons of the median raphe nucleus during hippocam-

pal theta rhythm in the rat. Exp Brain Res 145:383–394. 97. Kocsis B, Viana Di Prisco G, Vertes RP. (2001) Theta synchronization in the limbic system: the role of Gudden’s tegmental nuclei. Eur J Neurosci 13:381–388. 98. Feenstra BW, Holsheimer J. (1979) Dipolelike neuronal sources of theta rhythm in dorsal hippocampus, dentate gyrus and cingulate cortex of the urethane-anesthetized rat. Electroencephalogr Clin Neurophysiol 47:532–538. 99. Leung LW, Borst JG. (1987) Electrical activity of the cingulate cortex. I. Generating mechanisms and relations to behavior. Brain Res 407:68–80. 100. Borst JG, Leung LW, MacFabe DF. (1987) Electrical activity of the cingulate cortex. II. Cholinergic modulation. Brain Res 407: 81–93. 101. Colom LV, Christie BR, Bland BH. (1988) Cingulate cell discharge patterns related to hippocampal EEG and their modulation by muscarinic and nicotinic agents. Brain Res 460:329–338. 102. Talk A, Kang E, Gabriel M. (2004) Independent generation of theta rhythm in the hippocampus and posterior cingulate cortex. Brain Res 1015:15–24. 103. Dickson CT, Kirk IJ, Oddie SD, Bland BH. (1995) Classification of theta-related cells in the entorhinal cortex: cell discharges are controlled by the ascending brainstem synchronizing pathway in parallel with hippocampal theta-related cells. Hippocampus 5: 306–319. 104. Alonso A, Garcia-Austt E. (1987a) Neuronal sources of theta rhythm in the entorhinal cortex of the rat. I. Laminar distribution of theta field potentials. Exp Brain Res 67:493–501. 105. Alonso A, Garcia-Austt E. (1987b) Neuronal sources of theta rhythm in the entorhinal cortex of the rat. II. Phase relations between unit discharges and theta field potentials. Exp Brain Res 67:502–509. 106. Natsume K, Hallworth NE, Szgatti TL, Bland BH. (1999) Hippocampal thetarelated cellular activity in the superior colliculus of the urethane-anesthetized rat. Hippocampus 9:500–509. 107. Pare D, Gaudreau H. (1996) Projection cells and interneurons of the lateral and basolateral amygdala: distinct firing patterns and differential relation to theta and delta rhythms in conscious cats. J Neurosci 16: 3334–3350.

Multisite Spike-Field Coherence, Theta Rhythmicity, and Information Flow 108. Chilingarian L, Bogdanov N, Mats V. (1996) The similarity and difference between dogs assessed by the indices of the summary electrical activity of the frontal cortex, hippocampus, amygdala and hypothalamus. Zh Vyssh Nerv Deiat Im I P Pavlova 46: 1018–1031. 109. Pape HC, Pare D, Driesang RB. (1998) Two types of intrinsic oscillations in neurons of the lateral and basolateral nuclei of the amygdala. J Neurophysiol 79:205–216. 110. Pape HC, Narayanan RT, Smid J, Stork O, Seidenbecher T. (2005) Theta activity in neurons and networks of the amygdala related to long-term fear memory. Hippocampus 15:874–880. 111. Chapman CA, Lacaille JC. (1999a) Cholinergic induction of theta-frequency oscillations in hippocampal inhibitory interneurons and pacing of pyramidal cell firing. J Neurosci 19:8637–8645. 112. Chapman CA, Lacaille JC. (1999b) Intrinsic theta-frequency membrane potential oscillations in hippocampal CA1 interneurons of stratum lacunosum-moleculare. J Neuro physiol 81:1296–1307. 113. Bilkey DK, Heinemann U. (1999) Intrinsic theta-frequency membrane potential oscillations in layer III/V perirhinal cortex neurons of the rat. Hippocampus 9:510–518. 114. Glasgow SD, Chapman CA. (2007) Local generation of theta-frequency EEG activity in the parasubiculum. J Neurophysiol 97:3868–3879. 115. Wang WT, Wan YH, Zhu JL, Lei GS, Wang YY, Zhang P, Hu SJ. (2006) Theta-frequency membrane resonance and its ionic mechanisms in rat subicular pyramidal neurons. Neuroscience 140:45–55. 116. Eccles JC. (1980) The emotional brain. Bull Mem Acad R Med Belg 135:697–713. 117. Vertes RP, Albo Z, Viana Di Prisco G. (2001) Theta-rhythmically firing neurons in the anterior thalamus: implications for mnemonic functions of Papez’s circuit. Neuroscience 104:619–625. 118. Albo Z, Viana Di Prisco G, Vertes RP. (2003) Anterior thalamic unit discharge profiles and coherence with hippocampal theta rhythm. Thalamus Relat Syst 2:133–144. 119. Sporns O, Gally JA, Reeke GN, Jr., Edelman GM. (1989) Reentrant signaling among sim-

213

ulated neuronal groups leads to coherency in their oscillatory activity. Proc Natl Acad Sci USA 86:7265–7269. 120. Hutcheon B, Yarom Y. (2000) Resonance, oscillation and the intrinsic frequency preferences of neurons. Trends Neurosci 23:216–222. 121. Doheny HC, Faulkner HJ, Gruzelier JH, Baldeweg T, Whittington MA. (2000) Pathway-specific habituation of induced gamma oscillations in the hippocampal slice. Neuroreport 11:2629–2633. 122. Lisman JE, Idiart MA. (1995) Storage of 7 +/- 2 short-term memories in oscillatory subcycles. Science 267:1512–1515. 123. von Stein A, Sarnthein J. (2000) Different frequencies for different scales of cortical integration: from local gamma to long range alpha/theta synchronization. Int J Psychophysiol 38:301–313. 124. Kocsis B, Kaminski M. (2006) Dynamic changes in the direction of the theta rhythmic drive between supramammillary nucleus and the septohippocampal system. Hippocampus 16:531–540. 125. Axmacher N, Mormann F, Fernandez G, Elger CE, Fell J. (2006) Memory formation by neuronal synchronization. Brain Res Rev 52:170–182. 126. Manns ID, Alonso A, Jones BE. (2000) Discharge properties of juxtacellularly labeled and immunohistochemically identified cholinergic basal forebrain neurons recorded in association with the electroencephalogram in anesthetized rats. J Neurosci 20:1505–1518. 127. Siapas AG, Lubenov EV, Wilson MA. (2005) Prefrontal phase locking to hippocampal theta oscillations. Neuron 46:141–151. 128. Kahana MJ, Sekuler R, Caplan JB, Kirschen M, Madsen JR. (1999) Human theta oscillations exhibit task dependence during virtual maze navigation. Nature 399: 781–784. 129. Gersch W. (1972) Causality or driving in electrophysiological signal analysis. Math Biosci 14:177–196. 130. Saito Y, Harashima H (1981) Tracking of information within multichannel EEG record. In: Yamaguchi N, Fujisawa K (eds) Recent advances in EEG and EMG data processing. Elsevier, Amsterdam, pp 133–146.

Chapter 9 Cognitively Relevant Recoding in Hippocampus: Beneficial Feedback of Ensemble Codes in a Closed Loop Paradigm Robert E. Hampson, John D. Simeral, Theodore W. Berger, Dong Song, Rosa H.M. Chan, Vasilis Z. Marmarelis, and Sam A. Deadwyler Abstract The use of population codes derived from ensembles of rat hippocampal neurons to control performance of a delayed-nonmatch-to-sample (DNMS) memory task illustrates the important functional and organizational specificity of simultaneously active neurons in this important brain region. In this chapter, we show that online population analyses of firing patterns of 15–35 hippocampal neurons in a single trial provides an ensemble representation (i.e. code) of Sample response information sufficient for utilization, after an imposed (1–30 s) delay interval, to make the required Nonmatch decision on the same trial. This was conclusively demonstrated using a Closed Loop feedback procedure in which the ensemble code for information presented in the Sample phase of the task was assessed and input to a paradigm that either shortened or extended the temporal delay between Sample and Nonmatch phases of the task as a function of the “strength” or efficacy of the Sample (ensemble) code. The Closed Loop paradigm facilitated task performance in two separate ways: (1) by decreasing the number of weak less distinct, codes that were “at-risk” for errors on long delay (>10 s) trials and (2) extending the capacity to perform correctly on longer delay trials when strong ensemble codes for Sample information were present on the same trial. In addition, two models – linear and nonlinear – for assessing ensemble codes were tested in the Closed Loop paradigm with the nonlinear model showing greater efficiency. The successful application of the Closed Loop feedback in this context makes it apparent that differential hippocampal ensemble coding is a key factor underlying short-term memory, while errors, when they occur, result from neural codes of insufficient representational efficacy to be retained over long delay intervals, thereby causing lack of retrieval. Key words: Ensemble, Population vectors, Multivariate analysis, Neural encoding, Memory, Behavior, Feedback

Robert P. Vertes and Robert W. Stackman Jr. (eds.), Electrophysiological Recording Techniques, Neuromethods, vol. 54, DOI 10.1007/978-1-60327-202-5_9, © Springer Science+Business Media, LLC 2011

215

216

Hampson et al.

1. Introduction Analysis techniques designed to extract neural coding of behaviorally relevant events have been proposed for nearly 40 years (1–3). In the case of cortical motor function, early work by Georgopoulos et al. (4–6) and Schwartz and co-workers (7–9) derived “population vectors” in which specific arm movements in monkeys were predicted by precise firing in populations of cortical neurons. As proof of the functional relevance of these neural codes, the cited studies have evolved in recent years into a basis for controlling artificial limbs (10–15). However, with some exceptions (16), it has proven more difficult to identify neural patterns that predict behaviors dependent on cognition (17–21). In rat hippocampus, patterns of neural firing have been shown to register a number of different tasks and environmental features (17, 22–32). To date, however, these findings have not been extensively employed to predict, facilitate, or otherwise control the behaviors that require such processing.

2. Characterization of Hippocampal Ensemble Activity and Cognitive Performance

For a number of years, we have characterized neural activity in rat hippocampus correlated with behavioral performance in a delayednonmatch-to-sample (DNMS) task. Multivariate statistical analyses (33–39) of ensembles of neurons recorded from the CA1 and CA3 subfields have identified task-specific firing correlates such as: (1) Sample and Nonmatch phase responses in the task, (2) spatial position (left or right) of the task-relevant responses, (3) trial-type, and (4) correct or error outcome (40, 41). The basis for this ensemble representation of task-specific information was determined to be differential firing of individual hippocampal neurons, each of which exhibited firing correlated with one or more of the above task events existing on single trials and can be derived from ensembles of 15–30 hippocampal neurons (34, 41–43). We demonstrate here that online derivation of hippocampal neural ensemble codes during single DNMS trials can significantly regulate performance of the task in a manner that results in (1) decreased occurrence of errors and (2) improvement in performance beyond levels of prior training. To accomplish this, we employ a “Closed Loop” paradigm in which single trial ensemble activity is extracted and analyzed online during the trial in order to alter task parameters and contingencies on the same trial. These results show that the deciphering of neural activity in the hippocampus in real time can be employed to alter task parameters to successfully facilitate cognitive performance on the same trial.

Cognitively Relevant Recoding in Hippocampus

3. DelayedNonmatch-toSample Task 3.1. Subjects and Apparatus

3.2. Behavioral Training Procedure

217

Male Long-Evans rats ranging in age from 200–250 days were trained to a criterion of 70% in 100 trial sessions with trial delays of 1–30 s, implanted with hippocampal recording arrays (see below), and retrained to the same behavioral standard after recovery from surgery. Only animals meeting stringent neural recording conditions, requiring detectable activity from a minimal number (n = 15) of neurons, were retained for experimental purposes. All procedures conformed to National Institutes of Health (NIH) and Society for Neuroscience guidelines for care and use of experimental animals. The testing apparatus was the same as employed in other studies from this laboratory (40, 41, 44) and consists of a 43 × 43 × 53 cm Plexiglas behavioral testing chamber with two retractable levers mounted 3.5 cm above the floor on one wall separated by 14.0 cm center-to-center, positioned to either side of a water trough on the same wall. A nosepoke sensor device was mounted at the same height as the levers in the center of the opposite wall below a cue light. The apparatus was housed in a commercially built sound-attenuated cubicle (Industrial Acoustics Co.) with speakers for “white noise” (75 db) and house lights to illuminate the chamber for video monitoring. The DNMS task was also the same as described previously (44) and consists of three main behavioral phases: Sample, Delay, and Nonmatch. A trial is initiated at the Sample phase via extension of either the left or right lever, chosen at random. When the “sample lever” is pressed (Sample response, SmR), it retracts, initiating the Delay phase of the task which can be 1–30 s in duration (also selected at random). In the Delay phase, the cue light above the nosepoke device on the opposite wall is illuminated, and the animal is required to nosepoke at least once after the delay interval times out. This extinguishes the cue light and extends both levers, signaling onset of the Nonmatch phase and the animal is then required to press the lever opposite to the position of the lever pressed in the Sample phase. Animals received a drop of water (0.04 ml) immediately following a correct response and both levers were retracted for a 10-s intertrial interval (ITI), after the next trial initiated by extension of the Sample lever. An error, in which the same lever was pressed in both phases (“Match” response), Resulted in no reward, the levers were retracted and the house lights turned off for 5 s of darkness after which the illuminated ITI continued for 5 s until the start of the next trial. Criterion DNMS performance consisted of: (1) 90% correct responding on trials with delays of 1–5 s in sessions of 100–150 trials with delays ranging from 1–30 s and (2) an associated linear, delay-dependent,

218

Hampson et al.

decrease in accuracy from 85–90% correct to 50–60% (chance) across trials ranging in delays of 6–30 s, r espectively (40). 3.3. Hippocampal Ensemble Recording

Extracellular single-cell action potential waveforms were isolated and selected for analysis from each of the 32 different recording locations sampled by two bilaterally placed hippocampal array electrodes (40, 42, 45). The recording arrays were configured as two rows of eight 25 mM wires positioned along the longitudinal axis of the hippocampus at 200 mM intervals such that one row had lengths sufficient to sample neurons in CA3 with the other in the CA1 cell layers, providing recordings from eight pairs of CA3/ CA1 located electrodes in the same hippocampus. One or at most two distinct action potential waveforms recorded from the tip of each wire were isolated by time–amplitude window discrimination as well as computer-assisted template matching of individual waveform characteristics via a Plexon Multineuron Acquisition Processor (MAP, Plexon, Inc., Dallas, TX). Activity from separately identified neurons was time stamped along with the behavioral events within each DNMS trial and stored in arrays for online computer processing (see below). Only CA1/CA3 pyramidal neurons with firing rates (mean background firing rate 0.25–5.0 Hz) consistent with those previously reported (40, 41, 44), and consistently exhibited the same DNMS behavioral correlate across all test sessions, were included in the analyses. To determine DNMS behavioral correlates, perievent histograms of mean firing rates for each neuron in the ensemble were computed from 1.5 s before, to 1.5 s after, each DNMS behavioral event (lever press or nosepoke). Background firing rates were computed over the latter portion (6.5–9.5 s) of the ITI and firing rate changes within ±1.5 s of a behavioral event compared to background (i.e. ITI) firing rate using standard scores (z = [peak rate – background rate] ÷ std. dev. of background rate). Utilizing the above stringent statistical criteria for identification of single neuron activity (41, 44), we have determined that as few as 15 and as many as 35 separate neurons are sufficient to provide evidence of hippocampal ensemble encoding of behavioral events within the DNMS task. Perievent histograms calculated (±1.5 s) relative to the Sample response (SmR), last nosepoke (LNP), and NR behavioral events were further subdivided with respect to lever position (left or right) and trial outcome (correct or error), yielding a total of 16 possible behavioral event classifications for neurons in the DNMS task (40, 41, 46).

3.4. Task-Related Functional Cell Types

As indicated below, extraction by multivariate analyses of sources of variance in hippocampal ensembles associated with specific DNMS events relies on the presence of previously described DNMS task-related Functional Cell Types (FCTs) which display exclusive event-specific firing patterns (40, 41, 44). In this regard, a recorded ensemble may possess any combination of 10 previously

Cognitively Relevant Recoding in Hippocampus

219

identified hippocampal FCTs that fire differentially to: (1) Task Phase: SmR vs. NR; (2) Lever Position: Left vs. Right; (3) “Conjunctions” of Phase and Position (e.g. Left SmR vs. Right SmR); or even higher combinations such as (4) Trial Types; i.e. activity on both the Left SmR and Right NR of one trial type, but no increase in firing on the opposite trial type (i.e. Right SmR and Left NR). Historically, we have determined that an average of 1.31 ± 0.16 neurons is recorded per electrode and the frequency of FCTs approximately 0.65 ± 0.07 per electrode. Therefore, the probability of recording an FCT on each electrode during the task is 0.49 ± 0.08% and recording 15 FCTs from a 32-electrode array, 0.92 ± 0.05%. As indicated below, a minimum of 15 FCTs are required for multivariate analysis to account for the variability in firing associated with all behavioral events in the task. In the demonstration below comprising data from 15 selected animals (ensembles), an average of 21.0 ± 1.3 identified FCT neurons (range 15–30) was recorded per ensemble, which occurs on average in about 20–25% of animals utilized. 3.5. Derivation of Neural Codes

There are many ways in which ensemble activity can be analyzed to reveal information content (42, 45, 47). Techniques include both linear and nonlinear methods (42, 48) and both are discussed here. For linear analyses, we employ a multivariate canonical discriminant analysis (CDA) of ensemble neural firing during perievent histograms for SmR, LNP, NR in DNMS trials. The CDA extracts common sources of variance or Discriminant Functions (DFs) from a covariance matrix (49) of ensemble firing rates segregated into 0.25 s bins providing 12 separate time points of registered firing rate per neuron for each of the behavioral events within the trial, which are then summed over at least five consecutive daily DNMS sessions (500 total trials). CDAs for each ensemble (animal) are calculated from three-dimensional data matrices of ensemble firing rates consisting of 12 time bins per neuron (15–30 neurons) for each behavioral event (Data Matrices, Fig. 1). This data matrix of ensemble firing rates is then reduced, converted to single dimension population vectors (Xp) by collapsing the neuron (n) and time (t) bins (X1,1, X1,2 in Fig. 1) to a single dimension (p), preserving the sequential order of neurons and time bins in the matrix. Behavioral events classified for each neuron remain as the other dimension in the data matrix. The CDA program (PROC CANDISC, SAS Institute, Cary, NC) then constructs canonical covariance matrices from the matrix of ensemble firing rates for each of the 16 possible behavioral event classifications. Eigenvector decomposition is performed to produce normalized, orthogonal eigenvectors such that: X i , p = W j , p ·Di , j + ε i , p ,

(1)

220

Hampson et al.

Ensemble Firing Pattern 22

Time

Left S ample

Neuron

19

Neuron

16 13

X1,1 X2,2 … Xn,1

Firing Rate > 10 Hz

10

Data Matrices X1,1 X1,2 … X1,t X2,t X1,1 X1,2 …X1,t XX1,21,2 … X1,t X2,t Xn,t

7

X2,2 … X2,t Xn,t … Xn,2 … Xn,t

Event 1

Covariance Matrix

1 Hz

4

V1*1

V2*1 V3*1 … Vp*1

V1*2

V2*2 V3*2 … Vp*2 … V2*p V3*p … Vp*p

V1*p

1 −1.5 −1.0 −0.5 0.0

0.5

0.1

1.5

Eigenvalues

Time (s)

E1 E2 E3 E4 E5

Electrode Array CA3 CA1

22

DG CA3

Variance in Ensemble Firing Factor and compute eigenvectors from covariance

Eigenvectors W1,1 W2,1 W3,1 W4,1 W5,1

W1,2 W2,2 W3,2 W4,2 W5,2

W1,3 W2,3 W3,3 W4,3 W5,3

… W1,p … W2,p … W3,p … W4,p … W5,p

Derived Discriminant Functions

Left Sample - DF 5

19

x WDF5,p

13 10 7

SmR Delay (s) LNP DR Re

XL.SmR,p

16 Neuron

DNMS Trial Sum: (XL.SmR,p x W5,p)

Left S ample

Condense Neurons x Time (n x t ) to population vector, length p and compute covariance

Adjusted Firing Rates

CA1

ITI

Event ? Event 2

4 1 −1.5 −1.0 −0.5 0.0 0.5 0.1 1.5 Time (s)

Task

DF5 SmR L vs. R DF4: NR L vs. R

Trial

DF3: ITI vs. Trial DF2: SmR vs. NR DF1: LP vs. non-LP

Partitioned Variance in Matrix

Fig. 1. Schematic showing extraction of neural codes by canonical discriminant analysis (CDA) and application to DNMS trials. Middle Left: Diagram of recording array shows electrode arrangement along longitudinal axis of hippocampus and location of recording tips in CA3 and CA1 cell layers. Upper Left: Neural firing patterns recorded from hippocampal ensembles consist of 3 s duration perievent histograms (±1.5 s) as color contour plot featuring neurons on vertical axis and time on horizontal axis surrounding behavioral events (0 s) within DNMS trial (Left Sample shown). Middle: Data Matrices used in CDA. Firing rate (X) across neurons (n) is binned in 0.25 s increments (t) to yield data matrices Xn,t for each behavioral event (Event 1, 2, etc.). The data matrices are converted to population vectors by condensing neurons × time (n × t) to a single dimension (p) for further computation. The covariance matrix is then constructed from all data matrices and represents the total variance in ensemble firing across all neurons, time bins and behavioral events in the task (see Methods). Eigenvector deconvolution and orthogonal rotation of the covariance matrix yields eigenvalues (Ee1, E2, etc.) and eigenvectors (weighting coefficients, W 1,p, W 2,p, etc.) representing the different Discriminant Functions (DFs) as described in Methods. Lower Right: Relation between event-related discriminant functions (DFs 1–5) portrayed with respect to proportion of variance in ensemble firing (indicated level in pyramid) in terms of hierarchy of trial-relevant DFs extracted by the CDA. LP lever press events (i.e. Sample or Nonmatch); Non-LP non-lever press events (i.e. Nosepoke and ITI); SmR Sample Response; NR Nonmatch Response; ITI Intertrial Interval; L Left (position of lever press); R Right (position of lever press). Lower Middle: Perievent histogram of DF-adjusted neuron firing rates for same ensemble as in upper left display showing firing specific to Left Sample (SmR) event. Mean firing rate over trials (XL.SmR,p) was multiplied by the respective coefficient (W5,p) for DF5 for each neuron and time bin (p) to reveal the trial-unique firing pattern for Left SmR code trials shown in diagram at lower left. DNMS Trial Diagram: ITI intertrial interval; SmR Sample response; Delay Delay interval; LNP last nosepoke during Delay; DR Decision response; Re Reinforcement.

where Xi,p is the matrix of ensemble firing rates per 0.25 s time bin, with dimensions: i = total number of behavioral events in the dataset, and p = the neuron × time dimension of the population vector. Wj,p is the matrix of eigenvector coefficients, with dimensions:

Cognitively Relevant Recoding in Hippocampus

221

p (from the population vectors) and j (representing the unique eigenvectors or discriminant functions extracted by the CDA). Di,j is the matrix of discriminant scores, with dimensions: i (behavioral events) and j (discriminant functions). Finally, ei,p is the matrix of residual error, with dimensions i and p identical to the data matrix X. The symbol · indicates the matrix “dot-product” multiplication operation summed over dimension j, for every possible value of i and p. In the final step of the CDA, the eigenvectors were orthogonally rotated to distribute the resulting scores with mean = 0 and standard deviation = 1, while maximizing the Mahalonobis distances between event classifications. The variance contributions for each eigenvector produced an additional matrix of eigenvalues (Ej), one for each row (j) of the eigenvector coefficient matrix Wj,p (Fig. 1). The sum of all eigenvalues represents the total variance extracted from the dataset by the CDA (excluding the residual error, e, see (1)). The ratio of any given eigenvalue i to the sum of all eigenvalues represents the proportion of total variance accounted for by that eigenvector (proportion of variance = Ei/ ∑ E) in the CDA (50). For the data shown in Fig. 1 (upper left), the first five eigenvectors accounted for >85% of the total variance and were significantly different from all other eigenvectors as well as the residual error in the covariance matrix. The most important aspect of the analysis with respect to deriving ensemble encoding of task-related events is the fact that each eigenvector provides a Discriminant Function (DF) that identifies a single source of variance in firing calculated across all neurons in the ensemble (50). In Fig. 1, each row (E1, E2, etc.) of the matrix of eigenvectors (W1,p, W2,p, etc.) represents a set of DF coefficients with the same dimensions i and p as the perievent histograms, which can be applied to raw firing rates of single neurons to determine the degree of firing across all neurons in the ensemble associated with a particular behavioral event (see below). 3.6. Derivation of DF-Associated Ensemble Firing Patterns

Using the above parameters, it is possible to examine the associated ensemble firing pattern for all behavioral events by adjusting neuron firing rates using coefficients for a particular DF (Figs. 1 and 2). Since each DF extracted by the CDA represents a trialunique behavioral event (Fig. 1, lower right), multiplying the mean firing rates of each neuron (per 0.25 s bin) in the ensemble, summed within each trial, by the respective DF coefficients provides the adjusted ensemble “code” for that event. The example in Fig. 1 (lower middle, Adjusted Firing Rates) shows this relationship is: Ap = X (Left SmR),p × W(DF5),p

(2)

where A is the matrix of adjusted firing rates for DF5 over all Left SmR events, p is the single dimension for 0.25 s binned neuron

222

Hampson et al. 15 Neurons Rat 1

7

Left Sample

1 −1.5 −1 −.5 0 .5

1 1.5

Neuron

7 4 1 −1.5 −1 −.5 0 .5

Neuron

13 10 7

1 −1.5 −1 −.5 0

.5

Neuron

13 10 7 4 1 −1.5 −1 −.5 0 .5 1 1.5 Tim e (sec)

19 16 13 10 7 4 1 −1.5 −1 −.5 0 .5 Tim e (sec)

1 1.5

Firing Rate > 10 Hz 1 1.5

1 1.5

25 22 19 16 13 10 7 4 1 −1.5 −1 −.5 0

.5

1 1.5

1 1.5

25 22 19 16 13 10 7 4 1 −1.5 −1 −.5 0 .5

1 1.5

1 Hz

Correct

Right Sample

4

.5

1 1.5

Trial Outcome

1 1.5

19 16 13 10 7 4 1 −1.5 −1 −.5 0 .5

10

25 22 19 16 13 10 7 4 1 −1.5 −1 −.5 0 .5

Error

1 1.5

19 16 13 10 7 4 1 −1.5 −1 −.5 0

13

1 1.5

Trial Outcome

4

Error

Sample Lever Position

10

19 16 13 10 7 4 1 −1.5 −1 −.5 0 .5

25 Neurons Rat 3 25 22 19 16 13 10 7 4 1 −1.5 −1 −.5 0 .5

Correct

Neuron

13

20 Neurons Rat 2

Tim e (sec)

Fig. 2. Illustration of DF5-adjusted ensemble firing patterns for Sample response (SmR) from three different animals (Rats 1, 2, 3) with hippocampal ensembles of 15, 20, and 25 neurons, respectively. Each color contour pattern was generated as shown in Fig. 1 via by multiplying mean neuron firing rates by the respective DF5 coefficients for Left and Right SmRs. The DF5 ensemble patterns illustrate firing specific to successful SmRs on different trials (Correct Left, top row; Correct Right, bottom row). This can be contrasted with the firing for the same ensembles for Left or Right SmRs associated with error trials (second and third rows).

firing rates (as in (1)), X(Left SmR),p is the mean ensemble firing rate over all Left SmR events, and W(DF5),p is the set of coefficients for DF5. 3.7. Identification of Ensemble Codes for Behavioral Events

The individual CDA-extracted DFs account for sources of variance of ensemble neural firing patterns for specific behavioral events as depicted in the perievent histograms. The DFs are ranked in terms of proportion of variance relative to the overall ensemble firing rate, contributed across all neurons, time points, and events. Successively ranked DFs identify firing of different hippocampal cell types (FCTs) that encode less (higher proportion of overall variance) or more (less percentage) complex

Cognitively Relevant Recoding in Hippocampus

223

circumstances within the DNMS task. This is shown in Fig. 1 (lower right), where the largest source of variance is accounted for by DF1 which discriminates very general factors like ensemble firing associated with any lever press (e.g. SmRs “and” NRs) as distinct from all other DNMS trial events (i.e. nosepokes (delays), reward delivery, ITI, etc). In contrast, DF2 the next largest source of variance typically discriminates SmRs vs. NRs, and DF3 (the third highest percentage of variance) usually distinguishes firing during the ITI from all other within-trial events (i.e. SmRs, LNP, DRs). While it is clear that these three eigenvectors (DFs 1–3) satisfy a classification scheme which distinguishes task-related events from other more general factors within the testing environment, they do not discriminate events or conditions specific to performance on individual trials. However, once this larger source of variance extracted by DFs 1–3 is identified, smaller but more trial-unique sources of variance in ensemble firing are determined as shown by the blue line distinguishing Task (DFs 1–3) vs. Trial (DFs 4–5) in DF “Pyramid” in Fig. 1 (lower right). DF4 and DF5 were consistently identified with SmR and NR, respectively, as significant sources of variance by the CDA. Since DF4 and DF5 can also be classified with respect to association with behavioral success or failure (i.e. correct vs. error) in terms of magnitude of mean score (or “strength”) across trials, each reflects performance of the task. Finally, we have determined that only DF5 which reflects the ensemble code of the Sample lever press (SmRs) covaries with trial performance in terms of success or failure on the same trial (45, 47, 51, 52). DF4 (NR) covaries with trial outcome but is more influenced by the prior trials (52). However, because the SmR occurs before either the Delay or Nonmatch decision phases of the task, the magnitude of the DF5 score can “predict” performance on the same trial since it reflects the “strength” of encoding of the Sample information that must be used to make the Nonmatch decision after the variable duration of the interposed delay interval. 3.8. Computation of Ensemble Single-Trial DF Scores

It is also possible to derive an ensemble “code” (i.e. DF score) corresponding to each DF and associated DNMS event(s) on a single trial (49, 50). To calculate DF scores with dimensions i and j corresponding to DF and events, respectively, it is first necessary to transpose the eigenvector matrix Wj,p, by exchanging the rows for the columns, producing matrix W′p,j. Individual trial DF scores (D) are then computed using the formula: X i , p ·W ′ p , j = Di , j ,

(3)

where X is the raw data matrix as in (9.1), W′ is the transpose of the eigenvector matrix from (9.1), D is the resulting matrix of discriminant scores, and dimensions i, p, and j are the same as in

224

Hampson et al.

(9.1 and 9.2). Since each DF score Di,j has only the dimensions i and j for behavioral event and DF, it incorporates all of the adjusted neuron firing rates of all time bins and neurons within the ensemble. 3.9. Online Detection of Neural Codes

The next step in the implementation of ensemble codes to control behavior was to detect the code during the trial and utilize that information to change the parameters of the same trial in accordance with the “strength” of the detected code. DF5 scores were calculated online within 3 s following the SmR on each trial. The coefficients identified from prior sessions were entered into a customized MATLAB (The Mathworks, Natick, MA) program which computed DF scores “online” using the “live” ensemble firing recorded during the trial (47). DF5 adjusted ensemble firing rates were monitored continuously along with DF5 scores calculated every 0.25 s (see “waterfall” and stripchart displays in Fig. 2a). The following two exceptions from prior descriptions occur in online calculations: (1) DF5 coefficients (W) for each ensemble are derived from analysis of ≥5 previous DNMS sessions, but the neural firing dataset (X) is obtained online from the current “real-time” ensemble activity on individual trials and (2) online DF5 scores are computed continuously (with (9.3) above) using a revolving 3 s buffer on real-time activity from the same neurons and time bins as in prior constructed perievent histograms (see “Computation” above). Mean DF5 scores for Left vs. Right SmR events on correct vs. error trials computed from previous DNMS sessions are used as criteria to determine “weak” and/or “strong” SmR codes online. Weak SmR codes are designated as DF5 scores £1.0 standard deviation below the mean for Left or Right SmRs on correct trials for a given ensemble (Fig. 3a). Strong SmR codes include only DF5 scores >1.0 standard deviations above the mean of all prior Left and Right SmR correct trials for the ensemble (Fig. 3a).

3.10. Use of Hippocampal Ensemble Codes for “Closed Loop” Control of DNMS Performance

The utility of the single trial ensemble codes derived above is that they can be employed in a “Closed Loop” feedback paradigm in which the duration of the DNMS delay interval (in seconds) is altered on the basis of the strength of the SmR code (magnitude of DF5 score) on the same trial. This procedure is unique because it requires extracting in real-time, indications of how precisely the information presented (i.e. lever position) is encoded by the neural ensemble at the time of the SmR. The Closed Loop paradigm tests this directly by altering on the same trial a task parameter (the duration of the delay interval) that is dependent on the strength of such encoding. In the Closed Loop procedure, the delay interval is shortened to 10 s for weak SmR codes because most trained animals perform this type of trial correctly on ≥95% of occurrences (Fig. 4). However, when a strong SmR code is

Cognitively Relevant Recoding in Hippocampus

a

225

Single Trial Ensemble DF5 Scores Strong

Normal

Weak

Normal

Strong

DF2 Sample Score

3

Corr. Left SmR

2

Corr. Right SmR Err. Left SmR

1

0

Err. Right SmR −3

−1

−2 Right

1

0

2 Left

3

SmR Lever Position Score

b

Sample Response Code Strength and Trial Delay

Probability of Correct Response

1.0

0.9

0.8

0.7 Left Strong Code " Normal Code " Weak Code Right Strong Code Normal Code " Weak Code "

0.6

0.5 1.5

6.10

11.15

16-20

21-25

26-30

Delay (s)

Fig. 3. DF5 ensemble scores and DNMS performance. (a) Scatter plot distribution of single trial DF5 ensemble scores for Left and Right SmRs plotted conjointly with the associated CDA-extracted DF2 score for the same trial (Fig. 1). Each symbol (circles and triangles) represents both DF scores on a single trial computed for 500 trials recorded from a single animal. Firing rates for each neuron and time bin on a single trial were multiplied by the corresponding DF coefficients and summed to produce single ensemble scores for DF2 and DF5 for each trial. Single trials were plotted to reveal the ranges of DF5 scores encoded for Left and Right SmRs as clusters of filled symbols for Left vs. Right correct (Corr.) trials vs. clusters of unfilled symbols for Left and Right SmRs error (Err.) trials. Superimposed “centroids” indicate mean ± S.D. for each of the four displayed clusters of single trial DF2/DF5 scores. Centroids and clusters for Left and Right SmR error trials (arrows on horizontal axis) are closer together and show more overlap than the respective centroids and clusters for correct trials. Brackets indicate classification of scores as Strong, Normal, and Weak for Closed Loop. (b) Relationship between strength of Sample response (SmR) code and duration of DNMS delay interval. Trials were sorted according to the value of DF5 score to indicate Strong (greater than mean of correct SmR + 1 S.D.), Normal (mean of correct SmR ± 1 S.D.), or Weak (less than mean of correct SmR – 1 S.D.) SmR code strength (as indicated in A) as well as by length of delay (1–30 s) on the same trial. The mean relative frequency (probability) of a correct trial for each combination of code strength shows the likelihood of successful behavioral outcomes for each category of SmR strength at each delay interval. Trials were divided into six groups consisting of Left and Right SmR trials with Strong, Normal, or Weak SmR codes and for different trial delays. Probability of a correct response (mean ± S.E.M.) was calculated as the ratio of correct-to-error trials summed over at least 1,000 DNMS trials each, for 15 different animals.

226

a

Hampson et al.

b Operation of Closed Loop

Online Monitoring of Hippocampal Ensemble Activity Real-time Neural Activity

Record Ensemble Activity in Sample Phase

Extracted DF Scores 1s

10s

↑ DF5: Sample Left vs. Right ↑

DF4: Nonmatch Left vs. Right

Right Sample Pattern 3

5

7

9

11

13

15

Identified Single Trial Code Barpress

Sample

Right

Strong

Extract DF5 Code

Strong Code

Normal Code

↑ DF3: ITI vs. Trial 1

Weak Code

↑

Shorten Delay

Extend Delay

DF2: Sample vs. Nonmatch

↑ DF1: Leverpress vs. Nonleverpress

No Change

Closed Loop Criteria

Fig. 4. Implementation of Closed Loop paradigm to control DNMS performance. (a) Presentation shows real-time computer monitoring of hippocampal ensemble activity utilized during implementation of CDA for online extraction of single trial SmR codes utilized in Closed Loop paradigm (b). Downward scrolling “waterfall” display (Real-Time Neural Activity) provides updated moving columnar, color-coded graphic representation of neuron firing rate (red >10 Hz, blue <1 Hz) as trial proceeds in time from top to bottom of screen. DF scores are calculated online from real-time ensemble data at occurrence of SmR (framed by red rectangle on waterfall display) using CDA coefficients determined from prior sessions (see Methods). DF scores (DF5 – bold red trace) are displayed in continuously updated stripcharts that scroll from right to left in time. Colored blocks below waterfall are illuminated readouts of CDA identification of within-trial events and code strength (lever press occurrence – orange box; phase of task – green; lever position – red; code strength – blue). (b) Schematic of Closed Loop paradigm applied to single DNMS trials. Ensemble activity is recorded online and the SmR code calculated during the Sample phase and early portion of the delay. The DF5 score representing the SmR code for lever position on the same trial is computed online (a) and classified as Strong or Weak (see Fig. 3) based on data from the five previous DNMS sessions.

detected, the delay interval is extended beyond the range to which the animal was trained (1–30 s), for one of three randomly selected durations: 40, 50, or 60 s. The strong SmR code condition is implemented because in trained animals performance on Strong SmR code trials with delays ≤30 s was near 100% (Fig. 4); therefore, extending the delay to >30 s provided a behavioral measure of code strength related directly to retention of trial specific information under conditions never before experienced by the animal. 3.11. Specificity of Closed Loop Control

It is important to employ direct control measures to verify the dependence of the altered DNMS performance on ensemble activity detected within the Closed Loop paradigm. Two procedures were employed for this purpose, the first controlled for strength of SmR code by merely reversing the Closed Loop contingency such that delays were extended when a weak SmR code was detected, and shortened when a strong SmR code occurred (“Reversed” condition), while the second tested the specificity of the ensemble firing pattern by utilizing a randomized set of DF5 coefficients to adjust neuron firing rates (“Randomized” condition). The latter procedure controls for the possibility that the strong and weak SmR codes merely reflect synchronous changes in firing

Cognitively Relevant Recoding in Hippocampus

227

rates across all neurons in the ensemble rather than specifically different patterns of firing to the same behavioral event (SmR).

4. Demonstration of Closed Loop Model

4.1. Multivariate Analyses of Ensemble Firing

The following section provides a demonstration of the application of the closed loop DNMS model in each of 15 male LongEvans rats implanted with array electrodes to record bilaterally from ensembles of 15–30 CA3 and CA1 pyramidal cells and trained to criterion in the two-lever DNMS task with 1–30 s delays (40, 44). Experiments were conducted in which neural ensemble activity was recorded for a minimum of five consecutive daily sessions in each animal then subjected to multivariate CDA analysis to extract sources of variance, i.e. DFs, in ensemble neural firing associated with the encoding of specific DNMS events (see above). The sources of ensemble firing associated with Left or Right SmRs was determined and correlated with correct and error performance on individual DNMS trials to define the “SmR code,” the strength of which “predicted” behavioral outcome across trials (Fig. 4). Animals were then tested in a “Closed Loop” version of the DNMS task in which the delay interval was either extended when strong SmR codes occurred or shortened on weak SmR code trials (Fig. 5). Perievent histograms (±1.5 s in 0.25 s time bins) were constructed as synchronized to the SmR, NR, LNP, and ITI events in the task. The four events were further classified into left vs. right lever presses and correct vs. error trials, making a total of 16 possible behavioral event classifications. CDA analyses of hippocampal ensemble firing patterns were applied to firing rates of all neurons within an ensemble, and a covariance matrix was used to compute total variance in firing rate across all neurons and time bins in ensemble-based perievent histograms (Fig. 1). Eigenvector decomposition of the covariance matrix extracted five significant (F(1,1978) > 7.14, p < 0.01) sources of variance, or discriminant functions (DFs), each of which represented a proportion of the total variance associated with a specified behavioral event or event classification (Fig. 1). The coefficients of the DFs were used to “weight” single neuronal firing rates which revealed specific “adjusted” firing patterns across each ensemble (animal) for a particular source of variance in the covariance matrix (e.g. Left Sample DF5 histogram in Fig. 1). The sum of the weighted firing rates across all neurons provided a single DF score or “ensemble code,” the value of which could be used to quickly evaluate the level of ensemble activity, or code strength, for a given event. As indicated previously, DF4 and DF5 were the only sources of

Mean % Correct

a

Hampson et al. Weak Code Trials

b

100 *

90

** **

80 **

70 60 50

Control Weak Codes Control - Weak (Closed Loop) 1-5

6-10

**

16-20

Strong Code Trials 100

*

21-25

**

**

70

Control Strong Codes (Closed Loop) Remaining Codes 1-5

80 70

50

Control Weak Codes Control - Weak Strong Codes 1-5

6-10 11-15 16-20 21-25 26-30

Delay (s)

40

50

60

50

60

** **

**

Control Closed Loop Randomized Reversed

90 **

80 70

*

60

*p<0.01 **p<0.001 vs. Control

50 40

6-10 11-15 16-20 21-25 26-30 40

*

Specificity of Strong Code 100

Mean % Correct

Mean % Correct

d

90

60

**

Delay (s)

Single Animal

100

**

**

Delay (s)

c

**

80

50

26-30

**

**

90

60 **

11-15

Mean % Correct

228

40 s

50 “Strong Codes”

60

10 s “Weak Codes”

Fig. 5. Effect of implementing the Closed Loop Paradigm on DNMS performance. Mean % correct (± S.E.M.) responses comparing performance on Closed Loop and Normal (Control) trials summed over 3–5 different 100-trial sessions for the 15 different animals. (a) Closed Loop feedback applied to identified weak SmR code trials. Normal performance curve (mean ± S.E.M., circles) summed across all trials with 1–30 s delays (Control). Performance on only weak SmR code trials where the Closed Loop was not implemented (triangles). The top curve (squares) reflects performance when weak SmR codes were detected and shortened to delays of 10 s by Closed Loop manipulation. (b) Closed Loop feedback applied on strong SmR code trials shows facilitated performance on extended to 40, 50, or 60 s (diamonds) trials compared to performance on an equal number of trials extended to the same delays (circles) where SmR codes were not employed in Closed Loop paradigm (Control). The triangles represent performance on remaining trials in the same session with weak SmR code trials shortened by the same Closed Loop paradigm as in (a). Asterisks in (a and b) (*p < 0.01, **p < 0.001) indicate differences between Closed Loop and Normal (Control) means. (c) DNMS performance curves for a single animal under same conditions as in (a and b). Normal curve (circles) shows mean % correct (± S.E.M.) over all trials with delays from 1 to 60 s (including weak SmR codes) where Closed Loop procedure was not implemented. Performance on weak SmR code trials without Closed Loop indicated by triangles; weak SmR code trials shortened by Closed Loop procedure indicated by squares (as in a). Performance (mean % correct) on strong SmR code trials with Closed Loop (delays ≤30 s), and on Closed Loop trials with extended (40, 50, 60 s) delays (diamonds) compared to Control curve (as in b). (d) Tests for specificity of SmR code Closed Loop paradigm. Mean DNMS performance (% correct ± S.E.M.) summed across all animals (n = 15) for Control (black bars) and standard Closed Loop (white bars) sessions on Strong and Weak SmR code trials. Randomized Closed Loop condition (dark gray bars) shows mean performance when DF5 coefficients from the CDA were randomly assigned (“shuffled”) between neurons within the ensemble and resulting single trial DF5 scores computed. Reversed Closed Loop condition (light gray bars) depicts mean % correct performance when Closed Loop contingencies were reversed, such that weak SmR code trials were extended to 40, 50, or 60 s and strong SmR code trials shortened to 10 s. Bars indicate mean % Correct (± S.E.M.). DNMS performance was summed across 3–5 sessions (n = 15 animals) for each of the four conditions. Asterisks (*p < 0.01, **p < 0.001) indicate significant differences compared to Control performance (black bars ).

Cognitively Relevant Recoding in Hippocampus

229

variance that were associated with individual trials and could serve as a basis for the variability in performance associated with duration of delay in the DNMS task. 4.2. Ensemble Codes for Sample Response (SmR) Information

Figure 2 shows DF5-adjusted perievent firing patterns recorded from three different ensembles (animals) of 15, 20, and 25 neurons which corresponded to Left (top) or Right (bottom) SmRs on both correct and error trials. The single DF5 ensemble scores (derived by summing the DF5-adjusted firing rates across neurons and bins, Fig. 1) for the correct Left SmR pattern in Fig. 2 for Rats 1–3 were: +2.17, +1.81, and +2.11, respectively. For correct Right SmR patterns, DF5 scores were: -2.14, -1.95, and -1.66, respectively. Similarly, single DF5 ensemble scores for the error trial patterns shown in Fig. 2 were: Left: SmR, +0.76, +0.71, +0.51; Right SmR, -0.63, -0.40, -0.60 for Rats 1–3, respectively. The similarity in single DF5 scores with respect to both sign and magnitude for differential trial outcomes across the above three ensembles (animals) indicates the robustness of the DF5 measure for detecting the specificity and strength with which events are encoded within the trial across animals.

4.3. Detection of Single Trial Codes for SmR Position

Figure 3a illustrates the above analysis via a two-dimensional display of individual trial DF5 scores (horizontal axis) plotted with the magnitude (vertical axis) of the associated DF2 score (Sample vs. Nonmatch phase in Fig. 1) which does not vary on a trial-totrial basis. The two main clusters of scores (outer centroids in Fig. 3a) indicate Left SmRs (filled circles, mean DF5 score: 1.61 ± 0.02) discriminated by the CDA from Right SmRs (filled triangles, mean DF5 score: -1.62 ± 0.03) on correct trials (F(1,2493) = 21.04, p < 0.001) during the task. In addition, Fig. 3a shows that DF5 scores for the same position SmR on correct trials were significantly different from those for error trials (inner centroids, Fig. 3a; mean Left SmR: error vs. correct F(1,2493) = 9.32, p < 0.001; mean Right SmR: error vs. correct F(1,2492) = 13.4, p < 0.001). The arrows in Fig. 3a denote the overlap in the clusters of DF5 scores for right (unfilled circles) and left (unfilled triangles) error trials. Since correct and error trial outcomes are accurately discriminated by the sign and magnitude (i.e. “strength”) of the DF5 score for SmR on single DNMS trials (horizontal axis, Fig. 3a), such a SmR code could reflect performance level on the same trial. However, like performance, the SmR code should also be susceptible to the demands of the task; in this case, the duration of the DNMS delay interval. Figure 3b confirms this by showing the probability of a correct response as a function of delay interval for trials sorted according to different “strengths” or magnitudes of ensemble SmR codes (DF5 score). It is clear that the likelihood of a correct trial in the task decreased linearly with respect to each level of SmR code strength as delay duration increased. For weak

230

Hampson et al.

SmR codes (i.e. DF5 scores ≤1.0), probability of success was significantly reduced at delays >10 s (F(1,752) = 27.6, p < 0.001) compared to trials with intermediate (Medium) SmR codes, while the likelihood of success at delays >20 s was significantly elevated (F(1,752) = 16.2, p < 0.001) for strong SmR codes (i.e. DF5 scores ≥2.0). The relationship between strong SmR codes and performance accuracy remained constant across all delays ≤25 s (F(1,752) = 3.4, N.S.), whereas weak SmR codes were associated with significantly reduced accuracy as trial delays exceeded 15 s (F(1,752) = 21.9, p < 0.001). 4.4. Online Monitoring of SmR Code Strength

The SmR code was next examined for accuracy of “prediction” of trial outcome by the use of a “reclassification” scheme comparing DF5 score with trial outcome applied to the original dataset (49). Utilizing this scheme, correct identification occurred on 95.8 ± 2.1% of the occasions for all error trials, indicating SmR codes (DF5 scores) extracted online could provide similar predictability via their strength (i.e. magnitude). The CDA computed online SmR code strengths (magnitude of DF5 score) during at least four 100-trial sessions for each of the 15 animals effectively identified 89.2 ± 2.7% of all error trials in those sessions. Since the online computation of DF5 scores after the occurrence of the SmR required only 5 s (Fig. 4a), SmR code strength was used to determine the duration of the delay on the same trial, constituting a “Closed Loop” feedback procedure (Fig. 4b).

4.5. Closed Loop Feedback Control of DNMS Behavior

In the “Closed Loop” procedure, SmR codes (DF5 scores) were classified as: (1) weak SmR codes if ≤1.0 standard deviation below the mean of DF5 scores on all correct trials in the five prior sessions and (2) strong SmR codes if ≥1.0 standard deviations above the mean for correct trials in same prior sessions. Implementation of the Closed Loop required that delays on weak SmR code trials be shortened to 10 s and for strong SmR code trials extended to one of the three (randomly selected) intervals: 40, 50, or 60 s; beyond the maximum 30 s delay duration the animals were trained under (Fig. 3b). Figure 4a illustrates the real-time display of all digitized neural activity in the ensemble and the DF scores (stripcharts) computed online used to implement the Closed Loop paradigm (Fig. 4b). In order to prevent detection of the Closed Loop contingency, no more than 25 trials (50% strong or weak SmR code trials) were “adjusted” by the procedure within any one session. The number of Closed Loop sessions conducted per animal (n = 15 animals) ranged from 12 to 49. Figure 5 shows the results of applying the Closed Loop procedure on performance during the session. In Fig. 5a, the triangles show performance on trials identified online as weak SmR code within Closed Loop sessions but without shortening of delay duration. In marked contrast, however, when weak SmR

Cognitively Relevant Recoding in Hippocampus

231

code trials within the same sessions were identified and shortened to 10 s via the Closed Loop paradigm, performance for short delays (1–10 s) did not change but performance was significantly enhanced (squares) on trials with delays >15 s (F(1,401) = 8.14, p < 0.01) because of the elimination of trials that were “at risk” for error with delays >10 s (triangles) because of weak SmR codes which produced an elevation in overall performance at all delays because of weak SmR codes (squares). Figure 5b illustrates performance in sessions where both strong SmR and weak SmR codes trials were manipulated by the Closed Loop paradigm. The circles show performance on non-Closed Loop trials over the same delay intervals. For delays of 1–30 s performance on strong SmR code trials was well above Normal levels (F(1,401) = 18.39, p < 0.001), as shown also in Fig. 3b. The Closed Loop contingency was employed to extend delays to 40, 50, or 60 s in duration on strong SmR code trials. For comparison, a similar number of trials on which SmR codes were not detected were randomly extended to the same delays in the same sessions. When strong SmR codes were identified and the delay extended by the Closed Loop procedure, performance was significantly higher at all three increased delays (dashed curve, Fig. 5b) than on trials where SmR codes were not monitored to set delay duration (delay = 40 s: F(1,401) = 24.33, p < 0.001; delay = 50 s: F(1,401) = 16.23, p < 0.001; delay = 60 s: F(1,401) = 6.51, p < 0.02). The triangles in Fig. 5b show mean performance on all other trials in the session, which are elevated above normal session performance (circles) due to elimination of trials with weak SmR codes as in Fig. 5a. Figure 5b demonstrates the capacity of strong SmR codes to sustain behavioral performance well beyond the delay intervals to which animals had been previously trained (1–30 s). However, similar to normal (1–30 s) delay conditions, performance on strong SmR code trials exhibited a significant delay-dependent decline (F(5,401) = 11.75, p < 0.001) across those same Closed Loop trials with extended delays (Fig. 5b, red curve) documenting continued vulnerability to the capacity to retain trial-specific information over time. Figure 5c shows results of both types of Closed Loop manipulations from a single animal to verify that the procedure: (1) eliminated “at-risk” weak SmR code trials and (2) revealed enhanced performance on strong SmR code trials, in the same animal. 4.6. Validation of Closed Loop Control of Performance

To validate the relationship between single trial SmR code strength and Closed Loop facilitation of DNMS performance, the two tests mentioned above were employed in the same set of animals. The first test involved reversing the contingency of the Closed Loop procedure such that trials with weak SmR codes were extended to 40–60 s and strong SmR code trials shortened to 10 s (“Reverse” test). In the second test, individual neuron DF5 coefficients derived from the CDA were randomly rearranged, or

232

Hampson et al.

“shuffled,” between neurons in the ensemble and the Closed Loop operated as before (“Shuffle” test). The results of the two control tests are shown in Fig. 5d in terms of mean performance over all trials, compared with the results of Normal (black bars) and facilitated Closed Loop sessions (white bars). Performance on extended trials in the “Reversed” condition (Fig. 5d, light gray bars) did not differ significantly from performance on Normal trials (F(1,1752) = 0.5, N.S.), because extending weak SmR code trials to ≥40 s delays significantly (F(1,1752) = 5.32, p < 0.02) reduced performance relative to Normal trials (Fig. 5b), and was not different from chance levels at 50 and 60 s delays (F(1,1752) = 2.41, N.S.). However, strong SmR codes, significantly improved performance relative to non-Closed Loop conditions (F(1,1752) = 7.49, p < 0.01) when the trial was shortened to 10 s, presumably because under Normal conditions some weak SmR code trials are at-risk for error even at 10 s delays (Fig. 3). In the second control procedure (Shuffle test), calculations of online SmR code strength were made with the coefficients of the CDA “Randomized” between neurons in the ensemble and Fig. 5d (dark gray bars) shows that the Closed Loop procedure did not change performance on extended trials (F(1,1752) = 0.7, N.S.; n = 15 animals, 500 trials each), which was at chance levels for both “Shuffle” and Normal conditions. The two results (Fig. 5d) clearly establish: (1) the validity of the calculated SmR code strengths on single trials and (2) dependence of DNMS performance on specific patterns of firing (DF5) within hippocampal neural ensembles. 4.7. Reliability of the Closed Loop Paradigm Across Animals

Figure 6 shows the results of implementing the Closed Loop paradigm in 15 different animals over an average of 29.6 ± 3.1 (range 12–49) sessions per animal, each with hippocampal ensembles consisting of 15–30 neurons (mean 21.0 ± 1.3 neurons per animal). Overall, the mean number of strong SmR code trials was 13.3 ± 2.4 and weak SmR code trials was 21.6 ± 2.7 per 100-trial session (n = 15 animals). The mean (± S.E.M.) strong SmR code values (DF5 scores) across animals were: Left 2.3 ± 0.2, range 2.1–2.8; Right -2.5 ± 0.3, range -2.0 to -2.7; and for weak SmR codes: Left 0.4 ± 0.3, range 0–0.8; Right -0.5 ± 0.3, range 0 to -0.9. Figure 6 shows that performance on Closed Loop trials was significantly improved in each animal relative to the mean for Normal trials (dotted red line) at extended delays on strong SmR code trials (40 s: 90.4 ± 3.3%; 50 s: 78.1 ± 4.6%; 60 s: 61.2 ± 2.4%; all F(1,1752) > 11.6, p < 0.001). Performance across all animals on weak SmR code trials was also significantly elevated when the delay was shortened by the Closed Loop paradigm (mean 94.3 ± 1.6%), comp a r e d t o Normal trials (dotted red line, 63.7 ± 2.1%, F(1,1752) > 16.6, p < 0.001).

Cognitively Relevant Recoding in Hippocampus

233

Inter-animal Reliability of "Closed Loop" Control Animal:

100

Mean % Correct

90 80 70 60 50 40

40 s

50

60

Extended

10 s

937 947 975 929 941 938 952 961 970 973 987 988 989 994 999

Shortened

Delay (sec) Fig. 6. Reliability of Closed Loop improvement of performance across individual animals. DNMS mean % correct performance (± S.E.M.) shown separately for each of 15 animals (colored bars) summed over at least ten Closed Loop sessions on with Extended (strong SmR code) and Shortened (weak SmR code) delays. Individual means are shown for all 15 animals and compared to the overall mean % correct (dashed red lines) computed across the same animals in Normal (non-Closed Loop) control sessions with trials of the same delay. There were no significant differences across animals for performance within any of the extended or shortened trial delays.

4.8. Application of Nonlinear Ensemble Analyses to the Closed Loop

We have recently shown that hippocampal CA1 neural firing can be predicted from CA3 firing by means of a nonlinear, multiinput/multi-output (MIMO) model (53–56) utilizing the same hippocampal ensemble activity reported above. The nonlinear MIMO model is adapted from a Laguerre-Volterra network (57), whereby the fine temporal relationship between one or more input series (spike trains) and a single output series (MISO) can be described by a combination of first, second and third-order Volterra kernels (56). The resulting Volterra models with multiple inputs and a single output (MISO, Fig. 7a) are coupled with other MISO models using the same inputs to produce the MIMO model. Figure 7a shows the coupling of MISO models into a single MIMO model and also shows the decomposition of a MISO model into Volterra kernels. The MIMO model thus uses nonlinear, nonparametric estimation to predict the probability of CA1 neuron firing in terms of output trains of spike occurrences (i.e. spike trains) based on CA3 spike trains as input, which is logically consistent with the strong synaptic connection of the CA3-CA1 Shaffer collateral projections (58). Online implementation of the MIMO model allowed prediction of CA1 output

234

Hampson et al.

a

Multi-input, Single-output (MISO)

c Predicted CA1 Firing Pattern

Multi-input, Multi-output (MIMO)

b

d

Nonlinear Prediction: Mean Spike Probability on Correct Trials

Continuous Nonlinear Prediction of Ensemble Firing

Are they the same pattern? Actual CA1 Firing Pattern

NO: Shorten Delay

YES: Extend Delay

Fig. 7. Nonlinear MIMO model applied to the Closed Loop paradigm. (a) Multi-input, single-output (MISO) model shows nonlinear estimation (K) consisting of first-, second-, and third-order Volterra kernels (k0 - k2) predicting the probability of CA1 spike firing (u) from the temporal relationships between input spikes (x) recorded from multiple CA3 neurons (see Electrode Array in Fig. 1). Single CA1 spike train (y) over time is predicted by a stochastic model using threshold (q), feedback kernel (h), and noise (s) terms. (b) Expanded multi-input, multi-output (MIMO) nonlinear model is a concatenation of MISO models that produces the same inputs to all elements, but independent spike train outputs so that CA1 output spikes trains (yi), are continuously predicted from CA3 input spike trains (xi). (c) Validity of ensemble CA1 spike train estimation by MIMO model illustrated by color stripchart comparisons of predicted vs. actual CA1 firing probabilities (red = maximum probability, >10 Hz firing; blue = minimum firing, <1 Hz firing). Top trace: probability of firing of 7 CA1 neurons predicted via MIMO model with 15 CA3 spike train inputs. Bottom trace: actual mean firing rate of 7 CA1 neurons recorded simultaneously over 50 Left Sample trials recorded in 2.0 s increments (SmR occurs at 2.0 s on the horizontal axis). (d) Application of MIMO model in the same Closed Loop paradigm. The predicted CA1 spike trains on long-delay correct trials are averaged to provide a firing probability template (see text) of CA1 activity in Sample phase through SmR. For Closed Loop modifications of trial delay the CA1 spike train is compared to the previously determined template of CA1 firing on correct vs. error trials. Trials with >85% match to the “Correct” CA1 pattern are extended, while trials with <15% match are shortened.

spike firing approximately 100 ms prior to the actual occurrences, thus permitting a Closed Loop paradigm to be constructed substituting the nonlinear MIMO model, for the linear CDA analysis, to compute the likelihood of correct DNMS performance online. The MIMO model (Fig. 7b) for a given ensemble was developed for each animal using data from three successive DNMS sessions, and predicted the probability of firing for CA1 neurons as a function of the CA3 spike occurrences at any given point in time, independent of behavioral events in the task (Fig. 7c). However,

235

Cognitively Relevant Recoding in Hippocampus

MIMO-predicted spike trains during the Sample phase were sorted according to (1) correct vs. error trials and (2) duration of delay, to derive the mean probability of spike firing within 100 ms bins on correct, long (>20 s) delay trials. This MIMO-derived output probability was then used as a template for identifying both correct and incorrect codes for DNMS trials. The MIMO model was computed online during the Sample phase of DNMS trials and the derived CA1 spike trains compared to the template for correct encoding on long delay trials. Trials with >85% match to the correct trial template were extended during the delay, while trials with <15% match to the same template were shortened, in the same manner as the CDA-derived Closed Loop paradigm (Fig. 7d). The results of the Closed Loop application of the MIMO model were compared to the linear CDA Closed Loop procedure based on data from the same three prior DNMS test sessions in the same animals. Figure 8a shows performance across delays within control sessions (n = 5), CDA (n = 5), and MIMO Closed a 100

*

**

Left Sample Right Sample 100

80

**

70 Control CDA Closed Loop MIMO Closed Loop

50

% of Trials Identified by Code

*

60

Prediction Accuracy

**

90 Mean % Correct

b

Closed Loop Performance

**

80 60 40 20 0

1-5

6-10 11-15 16-20 21-25 26-30

Delay (sec)

40

50

60

Correct Error

Correct Error

CDA

MIMO

*p<0.01, **p<0.001 vs. CDA Closed Loop

Fig. 8. Comparison of linear (CDA) and nonlinear (MIMO) models for application of Closed Loop paradigm. (a) Mean (± S.E.M.) DNMS performance sessions (n = 4 animals, five sessions) for Control (circles), CDA Closed Loop (triangles) and MIMO Closed Loop (squares) trials. Control: Mean performance on non-Closed Loop trials. CDA Closed Loop: Mean performance on CDA strong SmR code trials with extended delays. MIMO Closed Loop: Mean performance on trials with MIMO model prediction of probability of CA1 spike trains during Sample phase of DNMS task. To avoid maximization of performance on Closed Loop trials, the CDA was computed using only three prior DNMS sessions, and not five sessions as in Figs. 1–7. (b) Accuracy of trial prediction by the CDA and MIMO models for different types of trials (Left or Right SmRs). Bargraph shows the percentage of trials accurately identified as Correct or Error trials by post hoc reclassification using the CDA or MIMO analyses to the SmR related firing on non-Closed Loop trials in control sessions to test for whether the CDA or MIMO analysis appropriately identified the Correct vs. Error trials. Bars indicate mean (±S.E.M.) percentage of correctly identified trials per session (n = 4 animals, 2–4 sessions each).

236

Hampson et al.

Loop (n = 5) sessions and clearly demonstrates the superiority of the MIMO model over the CDA for enhancing performance at longer delays of 21–60 s (F(8,491) = 4.26, p < 0.001). The basis for this difference in comparison to Fig. 5 was reduced number of prior DNMS sessions to derive the CDA, resulting in reduced estimation of DF5 variance and less accuracy in classifying trials compared to the MIMO model. The reduced accuracy of the CDA Closed Loop compared to the MIMO Closed Loop technique is further revealed in Fig. 8b as a significant decrease (F(2,491) = 16.39, p < 0.001) in identification of Error trials by CDA compared to MIMO Closed Loop for the same number of sessions.

5. Summary One of the most important things that can be learned regarding the relationship of neural activity to behavior is the manner in which neuronal firing patterns optimize performance under different conditions (17, 59). The manner in which hippocampal activity was analyzed here has been described in several prior publications (40, 41, 44, 59, 60) and demonstrated to accurately reflect characteristics of ensembles of individual hippocampal neurons. We have shown here that a linear CDA employed to extract these features online in a Closed Loop procedure, not only identified when the SmR code was appropriate to the DNMS trial type but also determined the reliability of that code (in terms of magnitude) for predicting success on the same trial. Notably, such predictive accuracy involves detection of the appropriate ensemble firing pattern for the current trial and direct evidence of such specificity was provided by control procedures that showed a lack of Closed Loop facilitation when the coefficients of the CDA were “shuffled” between neurons in the ensemble (Fig. 5d). This indicates that the SmR codes reflects firing within hippocampal ensembles (Figs. 1 and 2) whose strength (i.e. distinctiveness) under normal testing conditions apparently vacillates from trial-to-trial between extremes (strong and weak) during the session (Fig. 3) which can be dramatically remedied by implementing the Closed Loop feedback procedure demonstrated here. In addition, we also show that there are further advantages to implementing a nonlinear MIMO vs. linear CDA-derived Closed Loop control. One basis is that the Volterra kernels can be computed from as few as 100 trials (one session), although best results are provided using three prior DNMS sessions (55, 56). The number of trials required to compute CDAs with the same predictability is at least five times larger due to the number of variables in the population vector (49). Second, the CDA requires

Cognitively Relevant Recoding in Hippocampus

237

detection of a 3-s pattern of spike firing around each event it predicts (i.e. SmR), whereas the MIMO model predicts CA1 firing up to 100 ms in advance of firing under all conditions selected. Thus, for the MIMO model, predictions can be determined at any time after 900 ms in the Sample phase, conferring up to a 2 s advantage over the CDA. However, regardless of the method utilized to monitor ensemble-related encoding, the successful implementation of the Closed Loop paradigm in the current context unequivocally confirms the fact that hippocampal ensemble activity is a reliable predictor of cognitive performance in a task requiring short-term retention and utilization of trial-specific information.

Acknowledgments The authors thank the following for assistance with the project: David B. King, Vernell Collins, Rodrigo A. España, and Lihong Shi. This work was supported by NIH grants MH613972 and DA08549 to R.E.H., NSF BMES-ERC and NIH/NIBIB-BMSR to T.W.B. and DA07625 and DARPA contract (SPAWAR) N66001-09-C-2080 to S.A.D. References 1. Gerstein GL, Perkel DH (1969) Simultan‑ eously recorded trains of action potentials: analysis and functional interpretation. Science 164:828–830. 2. Abeles M, Gerstein GL (1988) Detecting spatiotemporal firing patterns among simultaneously recorded single neurons. J Neurosci 60: 909–924. 3. Gochin PM, Colombo M, Dorfman GA, Gerstein GL, Gross CG (1994) Neural ensemble coding in inferior temporal cortex. J Neurophysiol 71:2325–2337. 4. Georgopoulos AP, Schwartz AB, Kettner RE (1986) Neuronal population encoding of movement direction. Science 233:1416–1419. 5. Lee D, Port NP, Kruse W, Georgopoulos AP (1998) Neuronal population coding: multielectrode recording in primate cerebral cortex. In: Neuronal ensembles: strategies for recording and decoding. Eichenbaum H, Davis J (Eds), Wiley, New York. 6. Georgopoulos AP (2000) Neural aspects of cognitive motor control. Curr Opin Neurobiol 10:238–241. 7. Moran DW, Schwartz AB (1999) Motor cortical activity during drawing movements:

8.

9. 10.

11.

12.

13.

opulation representation during spiral p tracing. J Neurophysiol 82:2693–2704. Schwartz AB, Moran DW (1999) Motor cortical activity during drawing movements: population representation during lemniscate tracing. J Neurophysiol 82:2705–2718. van Hemmen JL, Schwartz AB (2008) Population vector code: a geometric universal as actuator. Biol Cybern 98:509–518. Donoghue JP, Nurmikko A, Friehs G, Black M (2004) Development of neuromotor prostheses for humans. Suppl Clin Neurophysiol 57:592–606. Friehs GM, Zerris VA, Ojakangas CL, Fellows MR, Donoghue JP (2004) Brain-machine and brain-computer interfaces. Stroke 35: 2702–2705. Chapin JK, Moxon KA, Markowitz RS, Nicolelis MA (1999) Real-time control of a robot arm using simultaneously recorded neurons in the motor cortex. Nature Neurosci 2:664–670. Wessberg J, Nicolelis MA (2004) Optimizing a linear algorithm for real-time robotic control using chronic cortical ensemble recordings in monkeys. J Cogn Neurosci 16: 1022–1035.

238

Hampson et al.

14. Velliste M, Perel S, Spalding MC, Whitford AS, Schwartz AB (2008) Cortical control of a prosthetic arm for self-feeding. Nature 453: 1098–1101. 15. Schwartz AB, Cui XT, Weber DJ, Moran DW (2006) Brain-controlled interfaces: movement restoration with neural prosthetics. Neuron 52:205–220. 16. Talwar SK, Xu S, Hawley ES, Weiss SA, Moxon KA, Chapin JK (2002) Rat navigation guided by remote control. Nature 417:37–38. 17. Ergorul C, Eichenbaum H (2004) The hippocampus and memory for “what,” “where,” and “when”. Learn Mem 11:397–405. 18. Stepniewska I, Fang PC, Kaas JH (2005) Microstimulation reveals specialized subregions for different complex movements in posterior parietal cortex of prosimian galagos. Proc Natl Acad Sci USA 102:4878–4883. 19. Freedman DJ, Riesenhuber M, Poggio T, Miller EK (2002) Visual categorization and the primate prefrontal cortex: neurophysiology and behavior. J Neurophysiol 88:929–941. 20. Wallis JD, Miller EK (2003) Neuronal activity in primate dorsolateral and orbital prefrontal cortex during performance of a reward preference task. Eur J Neurosci 18:2069–2081. 21. Eldawlatly S, Jin R, Oweiss KG (2008) Identifying functional connectivity in largescale neural ensemble recordings: a multiscale data mining approach. Neural Comput 21:450–477. 22. Wilson IA, Ikonen S, Gurevicius K, McMahan RW, Gallagher M, Eichenbaum H, Tanila H (2005) Place cells of aged rats in two visually identical compartments. Neurobiol Aging 26:1099–1106. 23. Bilkey DK, Clearwater JM (2005) The dynamic nature of spatial encoding in the hippocampus. Behav Neurosci 119:1533–1545. 24. Touretzky DS, Weisman WE, Fuhs MC, Skaggs WE, Fenton AA, Muller RU (2005) Deforming the hippocampal map. Hippocampus 15:41–55. 25. Louie K, Wilson MA (2001) Temporally structured replay of awake hippocampal ensemble activity during rapid eye movement sleep. Neuron 29:145–156. 26. Ribeiro S, Nicolelis MA (2004) Reverberation, storage, and postsynaptic propagation of memories during sleep. Learn Mem 11:686–696. 27. Leutgeb S, Leutgeb JK, Barnes CA, Moser EI, McNaughton BL, Moser MB (2005) Independent codes for spatial and episodic memory in hippocampal neuronal ensembles. Science 309:619–623. 28. Buzsaki G (2005) Theta rhythm of navigation: link between path integration and

29.

30. 31.

32.

33. 34.

35.

36.

37.

38.

39.

40.

41.

42.

landmark navigation, episodic and semantic memory. Hippocampus 15:827–840. Battaglia FP, Sutherland GR, McNaughton BL (2004) Local sensory cues and place cell directionality: additional evidence of prospective coding in the hippocampus. J Neurosci 24:4541–4550. Huxter J, Burgess N, O’Keefe J (2003) Independent rate and temporal coding in hippocampal pyramidal cells. Nature 425:828–832. Knierim JJ, Rao G (2003) Distal landmarks and hippocampal place cells: effects of relative translation versus rotation. Hippocampus 13: 604–617. de Hoz L, Martin SJ, Morris RG (2004) Forgetting, reminding, and remembering: the retrieval of lost spatial memory. PLoS Biol 2:E225. Donchin E (1966) A multivariate approach to the analysis of average evoked potentials. IEEE Trans Bio-Med Eng 19:457–463. Hampson RE, Simeral JD, Deadwyler SA (2001) What ensemble recordings reveal about functional hippocampal cell encoding. Prog Brain Res 130:345–357. Barbieri R, Frank LM, Nguyen DP, Quirk MC, Solo V, Wilson MA, Brown EN (2004) Dynamic analyses of information encoding in neural ensembles. Neural Comput 16:277–307. Truccolo W, Eden UT, Fellows MR, Donoghue JP, Brown EN (2005) A point process framework for relating neural spiking activity to spiking history, neural ensemble, and extrinsic covariate effects. J Neurophysiol 93:1074–1089. Lebedev MA, O’Doherty JE, Nicolelis MA (2008) Decoding of temporal intervals from cortical ensemble activity. J Neurophysiol 99: 166–186. Bonifazi P, Ruaro ME, Torre V (2005) Statistical properties of information processing in neuronal networks. Eur J Neurosci 22:2953–2964. Brown SL, Joseph J, Stopfer M (2005) Encoding a temporally structured stimulus with a temporally structured neural representation. Nat Neurosci 8:1568–1576. Deadwyler SA, Bunn T, Hampson RE (1996) Hippocampal ensemble activity during spatial delayed-nonmatch-to-sample performance in rats. J Neurosci 16:354–372. Hampson RE, Simeral JD, Deadwyler SA (1999) Distribution of spatial and nonspatial information in dorsal hippocampus. Nature 402:610–614. Hampson RE, Simeral JD, Deadwyler SA (2005) Cognitive processes in replacement

Cognitively Relevant Recoding in Hippocampus

43.

44.

45.

46.

47.

48.

49. 50. 51.

brain parts: a code for all reasons. In: Toward replacement parts for the brain: implantable biomimetic electronics as neural prosthesis. Berger TW, Glanzman DL (Eds), MIT Press, Cambridge, MA, pp 111–128. Wood ER, Dudchenko PA, Robitsek RJ, Eichenbaum H (2000) Hippocampal neurons encode information about different types of memory episodes occurring in the same location. Neuron 27:623–633. Deadwyler SA, Hampson RE (2004) Differential but complementary mnemonic functions of the hippocampus and subiculum. Neuron 42:465–476. Hampson RE, Simeral JD, Deadwyler SA (2008) Neural population recording in behaving animals: constituents of the neural code for behavior. In: Neural population encoding Holscher C, Munk MH (Eds), Cambridge University Press, Cambridge, UK. Hampson RE, Deadwyler SA (1996) Ensemble codes involving hippocampal neurons are at risk during delayed performance tests. Proc Natl Acad Sci USA 93:13487–13493. Simeral JD, Hampson RE, Deadwyler SA (2006) Behaviorally relevant neural codes in hippocampal ensembles: detection on single trials. In: Synaptic plasticity: from basic mechanisms to clinical applications. Baudry M, Bi X, Schreiber S (Eds), MIT Press, Camridge, MA. Berger TW, Ahuja A, Courellis SH, Deadwyler SA, Erinjippurath G, Gerhardt GA, Gholmieh G, Granacki JJ, Hampson R, Hsaio MC, LaCoss J, Marmarelis VZ, Nasiatka P, Srinivasan V, Song D, Tanguay AR, Wills J (2005) Restoring lost cognitive function. IEEE Eng Med Biol Mag 24:30–44. Stevens J (2002) Applied multivariate statistics for the social sciences. Lawrence Erlbaum Associates, Hillsdale. Rao CR (2002) Linear statistical inference and its applications. Wiley, New York. Deadwyler SA, Goonawardena AV, Hampson RE (2007) Short-term memory is modulated by the spontaneous release of endocannabi-

52.

53.

54.

55.

56.

57.

58. 59.

60.

239

noids: evidence from hippocampal population codes. Behav Pharmacol 18:571–580. Deadwyler SA, Hampson RE (2008) Endocannabinoids modulate encoding of sequential memory in the rat hippocampus. Psychopharmacology (Berl) 198:577–586. Song D, Chan RM, Marmarelis VZ, Hampson RE, Deadwyler SA, Berger TW (2006) Physiologically plausible stochastic nonlinear kernel models of spike train to spike train transformation. Conf Proc IEEE Eng Med Biol Soc 1:6129–6132. Zanos TP, Courellis SH, Hampson RE, Deadwyler SA, Marmarelis VZ, Berger TW (2006) A multi-input modeling approach to quantify hippocampal nonlinear dynamic transformations. Conf Proc IEEE Eng Med Biol Soc 1:4967–4970. Song D, Chan RH, Marmarelis VZ, Hampson RE, Deadwyler SA, Berger TW (2007) Nonlinear dynamic modeling of spike train transformations for hippocampal-cortical prostheses. IEEE Trans Biomed Eng 54:1053–1066. Zanos TP, Courellis SH, Berger TW, Hampson RE, Deadwyler SA, Marmarelis VZ (2008) Nonlinear modeling of causal interrelationships in neuronal ensembles. IEEE Trans Neural Syst Rehabil Eng 16:336–352. Marmarelis VZ, Berger TW (2005) General methodology for nonlinear modeling of neural systems with Poisson point-process inputs. Math Biosci 196:1–13. Andersen P, Soleng AF, Raastad M (2000) The hippocampal lamella hypothesis revisited. Brain Res 886:165–171. Laubach M, Wessberg J, Nicolelis MA (2000) Cortical ensemble activity increasingly predicts behaviour outcomes during learning of a motor task. Nature 405:567–571. Hampson RE, Deadwyler SA (1999) Pitfalls and problems in the analysis of neuronal ensemble recordings during performance of a behavioral task. In: Methods for simultaneous neuronal ensemble recordings, Nicolelis M (Ed), Academic Press, New York, pp 229–248.

Chapter 10 An Intact Septo-Hippocampal Preparation for Investigating the Mechanisms of Hippocampal Oscillation Romain Goutagny, Jesse Jackson, and Sylvain Williams Abstract The medial septum (MS) is a structure best known for its role in theta rhythm generation in the hippocampus. Theta rhythm is a relatively slow brain oscillation of 3–12 Hz that is important for memory formation and synaptic plasticity. Despite that the role of the MS in theta generation has been investigated for more than half a century, very little is known about how MS neurons contribute to this phenomenon. In addition, exactly how hippocampal neurons participate in rhythmic activity in the MS remains relatively unexplored. Our poor understanding of theta generation is due in part to the challenging conditions offered by in vivo experimentation. To circumvent these obstacles, this chapter describes how to obtain a complete septo-hippocampal preparation in vitro to investigate theta generation. This method allows the study of theta rhythm with whole-cell and multiple field recordings in combination with precise pharmacology in vitro. We show that the preparation displays rhythmic oscillations in the hippocampus upon activation of the septum and, conversely, exhibits rhythmic activity in the septum following hippocampal stimulation. These results suggest that the septo-hippocampal preparation is a potentially powerful tool to investigate the cellular mechanism implicated in theta generation in vitro. Key words: Theta rhythm, Hippocampus, Medial septum, In vitro preparation, Oscillation generation, Pharmacology, Field recordings, Whole-cell recordings

1. Introduction Theta rhythm in the hippocampus has generated substantial interest since the 1950s because of its association with arousal and attention (12, 47). Theta oscillations occur at frequencies in the range of 3–12 Hz and are the most prominent extracellular signal recorded in mammalian brain (52). This rhythmic activity present during the exploration of a novel environment is elicited by sensory stimuli and present during REM sleep (5, 28, 50). Because the hippocampus is widely known for its critical role in learning and memory, many studies have examined how theta contributes Robert P. Vertes and Robert W. Stackman Jr. (eds.), Electrophysiological Recording Techniques, Neuromethods, vol. 54, DOI 10.1007/978-1-60327-202-5_10, © Springer Science+Business Media, LLC 2011

241

242

Goutagny, Jackson, and Williams

to these cognitive phenomena. Two seminal studies published in the late 1970s demonstrated that theta prevalence was significantly correlated with learning, specifically eliminating hippocampal theta abolished spatial memory (53). Since then, many have shown how theta activity participates in memory formation. An early connection between theta activity and long-term potentiation (LTP), a well-known cellular substrate for learning and memory, was the observation that trains of stimuli given at intervals equivalent to theta frequency-induced LTP more readily than stimu lation at other frequencies (13, 26). Several reports have demonstrated that theta oscillations may provide a framework favoring synaptic plasticity by showing first showed that pairing perforant pathway input to the peak of dentate gyrus theta cycle in anesthetized rats induced robust LTP (40). Similar results were also obtained for the Schaffer-collateral (SC)-CA1 pathway (18). More recently, it was demonstrated that a single SC stimulation paired to the peak of CA1 theta elicited LTP but that a single stimulation paired at the trough of theta resulted in LTD (20). These experiments point to the important function of theta in terms of setting, not only the right conditions of excitability for plasticity, but also in its capacity to compartmentalize the direction of plasticity. There is also convincing evidence that theta plays an important role in the timing of place cell firing (39, 45). The so-called “place cells” are pyramidal neurons that increase firing (usually from a silent firing background) when an animal enters a particular place or area of a given environment. Hippocampal pyramidal cells undergo a phenomena known as “phase precession” when an animal enters the cell’s place field. During phase precession, the cell fires at a progressively earlier phase of the network theta rhythm while simultaneously increasing its firing rate (coded in the number of spikes per burst) (39). Therefore, place cells exploit both rate and phase coding mechanisms to increase the spatial information contained in a given spike train (19). Moreover, phase precession may help bind place cells together into assemblies, a characteristic that would be critical for episodic memory (45). Taken together, theta rhythm may serve not only to facilitate synaptic plasticity, but may also be an important mechanism that binds together place cell assemblies which may function as the basis for spatial navigation and episodic memory. 1.1. Role of the Septum

Theta activity in the hippocampus was first described in the seminal paper by Jung and Kornmuller (55). Later, two separate groups showed that the septum was critical for hippocampal theta generation (12, 47). Indeed, they showed that neurons of the medial septum and diagonal band (MS–DB) not only discharged single or bursts of spikes phase-locked to hippocampal theta, but that lesion of the MS–DB also abolished hippocampal theta (12) suggesting that the MS–DB was a

An Intact Septo-Hippocampal Preparation

243

pacemaker for hippocampal theta. Hippocampal theta is not one single entity but rather is made up of at least two different underlying components that can be distinguished pharmacologically. Indeed, hippocampal theta can be differentiated into immobile theta (also named Type II) and voluntary movement or waking theta (Type I) (51). Both types of theta have different pharmacological profiles since Type II theta is acetylcholine-dependent because it is blocked by muscarinic receptor antagonists, such as atropine or scopolamine, whereas Type I theta is atropine-insensitive and may depend on other systems, such as the glutamatergic input from entorhinal cortex (5). However, this dichotomy in hippocampal theta may not be that clear cut since lesion of the MS–DB can abolish both types of theta suggesting that the septum likely contributes to both atropine and atropine-insensitive type II and type I, respectively (54). Nevertheless, the MS–DB is critical for memory since lesions of the septum that abolished or reduces theta significantly perturbs spatial (53) or working memory (29). One striking feature of MS–DB neurons is that a significant proportion of neurons in vivo fire rhythmic bursts of action potentials phase-locked (i.e., in synchrony) with hippocampal theta (3, 5, 12). It is well documented that the MS–DB contains an important number of GABAergic and cholinergic (ACh) neurons that send projections via the fornix to the hippocampus (6, 8, 24, 30, 43). Cholinergic and GABAergic neurons may also send projections locally within the MS–DB which could contribute to its pacemaker capacity. It is widely believed that ACh neurons fire in rhythmic bursts and project to pyramidal cells and interneurons in the hippocampus (5). Although their capacity to fire in bursts has been questioned recently (10, 44), these cells are considered to induce a sustained slow muscarinic receptor-dependent depolarization on hippocampal pyramidal and interneurons (6, 33), setting perhaps the excitatory tone that favors hippocampal theta activity (5, 27). In accordance with this idea, specifically eliminating MS–DB cholinergic neurons reduces the amplitude of theta oscillations (1, 27) but does not significantly modulate its frequency. Much attention has been given to the GABAergic part of the septohippocampal projection which is proposed to be important for its pacemaking capacity (8). Anatomically, these MS/DB GABAergic neurons are unique since they selectively project to hippocampal GABAergic interneurons but not to pyramidal cells (8). Since each hippocampal GABAergic interneuron can in turn contact 500–1,200 pyramidal cells (36, 49), MS–DB GABAergic neurons may have a powerful role in controlling hippocampal activity. In agreement with this idea, electrically stimulating the fornix (the fiber bundle containing axons of septohippocampal neurons) in septo-hippocampal slices at a frequency relevant to theta can actively disinhibit pyramidal neurons

244

Goutagny, Jackson, and Williams

and pace them in a GABA-dependent manner (49). These results suggest the potential importance of MS–DB GABAergic neurons in pacing the hippocampus. More recently, several studies have reported that the medial septum not only contains ACh and GABAergic neurons, but also a third population that can release glutamate (44). This population makes up about 25% of the septal cell population and projects both locally to all MS–DB populations (34) (including glutamate neurons) and to the hippocampus on both pyramidal cells and GABAergic interneurons. It remains to be established what is the exact role of medial septal glutamate neurons in hippocampal theta. 1.2. Theta In Vivo

Despite more than 50 years of research, our understanding of the mechanisms underlying hippocampal theta is largely incomplete and probably stems from a number of reasons. First, all the data regarding hippocampal theta comes from in vivo studies which is a challenging preparation in terms of performing precise pharmacology and single-cell recordings, such as patch-clamp electrophysiology. Second, many of the in vivo theta studies utilized anesthetized animals which present many obvious shortcomings. The most common anesthetics used in theta investigations are urethane, ketamine, or a mixture of both (23) which can significantly affect synaptic transmission. For example, anesthetics such as urethane may have complex actions such as increasing GABAA receptor channel function, decreasing glutamate release and reducing presynaptic Na+ channel currents at synaptic terminals (16). Hence, anesthetics could profoundly alter the synaptic mechanism underlying theta. For example, theta elicited in urethane anesthetized animals is completely blocked by muscarinic receptor blocker (type-II theta), whereas in the absence of anesthetics only a slight reduction by atropine is observed (17, 51). Other anesthetics such as ketamine also probably modify theta since it is known to be a powerful NMDA receptor blocker (16). Hence, it is reasonable to suspect that theta elicited in the presence of anesthetics is likely generated by different mechanisms than those elicited without anesthetics. Additionally, hippocampal theta generation is probably influenced by many interacting pacemakers converging on the hippocampus. For example, in addition to the MS–DB, the supramammillary nucleus, the anterior thalamus, the mammillary body, and the entorhinal cortex probably all contribute (52) to theta generation in hippocampus. Hence, it is difficult to tease out the exact role of the MS–DB in an in vivo situation with interacting pacemakers. Taking all these pitfalls into consideration, it is not surprising that the accepted model regarding the MS–DB in theta generation was constructed mostly from indirect evidence. The evidence used to support the MS–DB pacemaker hypothesis was gathered from neuroanatomical studies looking at possible connections

An Intact Septo-Hippocampal Preparation

245

between these two structures, correlative in vivo evidence obtained from cross-correlation of septohippocampal activity in anesthetized animals, computational modeling, lesion studies, and from in vitro slice experiments looking only at the septum or hippocampus in isolation. Therefore, there is no direct evidence showing how the MS–DB and its ACh, GABA and/or glutamatergic components synaptically modulate hippocampal principal cells and interneurons to generate theta. Surprisingly, more than 50 years after the proposal that the septum is critical to hippocampal theta, we are still wondering if the septum is a real pacemaker, and if so, how it contributes hippocampal theta. One important challenge for investigating the role of the MS–DB in hippocampal theta has been to find an appropriate in vitro preparation that would offer the opportunity to use powerful electrophysiological techniques and adequate pharmacology. The first method available to study theta in vitro was the carbachol-elicited theta frequency-like oscillations in transverse hippocampal slices (25, 32). However, the mode of firing of CA3 pyramidal neurons characterized by rhythmic bursts and their insensitivity to GABAA receptor blockade suggests that these theta-like oscillations are different than those observed in vivo. Others have obtained slices containing both the septum and hippocampus (by cutting the brain at a special angle that includes a small portion of MS/DB and hippocampus) to maintain the necessary fibers between these structures in an attempt to have a more intact preparation and possibly preserve hippocampal theta (49). However, these slices do not display theta activity although the authors showed that appropriate stimulation of the septum can drive hippocampal neurons at theta frequencies through GABAergic transmission.

2. Procedure and Technique 2.1. The SeptoHippocampal Preparation

Our idea to develop a complete septo-hippocampal preparation was inspired from both the septo-hippocampal slice (49) and the intact septo-hippocampal preparation developed by (22) from immature rats. We hypothesized that the more mature septohippocampal complex could display theta rhythm in vitro. The main questions we asked was whether the MS–DB can generate hippocampal theta when it is the sole structure attached to the hippocampus, and second, if hippocampal oscillations could significantly influence neuronal septal activity. We used a freehand method to dissect out the complete septum and hippocampus, including the fornix that contains the fibers interconnecting both regions (see below). The preparation is dissected in an ice-cold sucrose-based solution (34) from juvenile Sprague–Dawley rats.

246

Goutagny, Jackson, and Williams

Septo-hippocampal preparations are then placed in a custom made bath, and perfused with regular ACSF. In our hands, the preparation is very healthy since whole-cell patch recordings can easily be obtained from neurons of either the MS–DB or the hippocampus. The majority of recorded neurons have overshooting action potentials and resting membrane potentials similar to those in slices. Extracellular field recordings of rhythmic activity can be obtained from the hippocampus and the septum. The preparation can be kept alive from up to 6–8 h. Here are some advantages in using the septohippocampal preparation over previously used approaches: 1. In contrast to brain slices, this intact preparation possesses all the synaptic connections between the septum and the hippocampus, including all connections septo-temporally and transversely within the entire hippocampus. 2. One can easily simultaneously perform extracellular and patch recordings from both MS–DB and hippocampus and directly analyze cell–cell synaptic interactions between these two structures. 3. Since this is an in vitro preparation, the experiments are performed free of anesthetics that can severely affect synaptic responses (2). 4. In-depth pharmacological investigations can be conducted by applying agonists and antagonists independently on the septal or the hippocampal side using a special recording chamber we have devised (see below). 2.2. Detailed Description of the Dissection Method

Dissection: The brain is rapidly removed from the skull and placed in ice-cold high-sucrose ASCF solution (in mM: 252 sucrose, 3 KCl, 2 MgSO4, 24 NaHCO3, 1.25 NaH2PO4, 1.2 CaCl2, and 10 glucose) and bubbled with carbogen (95% O2 and 5% CO2). A glass plate (inverted petri dish) covered with lens paper is used as the stage for the majority of the dissection. The lens paper provides friction and assists in the handling of the brain during the dissection. Both the ACSF/sucrose solution and glass plate are placed on crushed ice for the entire dissection. The glass plate can be rotated on the crushed ice throughout the dissection to gain access to various parts of the septo-hippocampus without contact with the brain. The cerebellum and frontal cortex are removed with a razor blade and the two hemispheres separated. The two hemispheres are then allowed to recover for 2–3 min in the oxygenated sucrose solution. The single septo-hippocampal isolate is then removed from the remaining hemisection in the following manner. A narrow flat blade spatula is first inserted into the lateral ventricle from the caudal direction (almost parallel with the inner surface of the hemisphere) to lift the medial and lateral septum away from the surrounding striatum (Fig. 1a). Immediately

An Intact Septo-Hippocampal Preparation

247

Fig. 1. (a–e) Sequence of steps taken to isolate the septo-hippocampal complex. See text for details. AC anterior commisure, CA1/CA3 cornu ammonis layers 1 and 3, DG dentate gyrus, HPC hippocampus, MSDBB medial septum and the diagonal band of broca, TH thalamus. The arrows in A and E indicate the caudal (C) and rostral (R) direction.

following, the spatula is inserted near the septal end of the dorsal hippocampus on top of the fornix. The spatula is then moved in an antero-ventral direction to isolate the septum from anterior portions of the brain (Fig. 1b); these two steps isolate the septum while leaving the fornix attached to the hippocampus. In order to circumvent any movement of the brain while the septum is being isolated, another spatula is placed on the top of cortex to support the brain and prevent unwanted movements (see Fig. 1). For the isolation of the hippocampus, one spatula supports the inner portion of the cortical hemisphere, and the other is used to gently pull away the brainstem and thalamus to expose the hippocampal

248

Goutagny, Jackson, and Williams

artery and underlying CA3 and dentate gyrus (DG) (Fig. 1c). To separate the hippocampus from the cortex the spatula is again placed between the cortex and extreme dorsal end of the hippocampus and moved smoothly through the caudal portion of the cortex. During this process, the spatula should separate the CA1/subicular tissue from the overlying cortex, without damaging the intact fornix (Fig. 1d). The septo-hippocampal complex can then be easily removed from the surrounding brain tissue by placing one spatula on the CA3/DG region of the dorsal hippocampus and pulling the medial septum and dorsal hippocampus toward the caudal portion of the brain (see the arrow in Fig. 1d). Blood vessels may impede prompt removal and should be cut away but not ripped out to avoid unnecessary tissue damage. The entire septo-hippocampal isolation procedure should take no longer than 1 min. Any remaining cortex can be removed using microscissors when the preparation is returned to the oxygenated sucrose solution. Following dissection, the septo-hippocampal complex stays at room temperature in ACSF bubbled with carbogen for 45–180 min. For recording, the preparation is transferred quickly to the custom submerged recording chamber.

3. Recording in the Septohippocampal Preparation

Investigating the interactions of the septum and hippocampus requires that both sides of the preparation be independently activated and/or inhibited pharmacologically. A chamber was therefore designed where both sides of the complex could be isolated hermetically while keeping the septum and hippocampus physically connected (Fig. 2). This configuration offers independent addition or removal of the ACSF-containing pharmacological agents from either side and also allows easy placement of pipettes for whole cell patch and field recordings. Practically, for dual compartment septo-hippocampal recordings (11, 35), a Teflon wall is placed between the septum and hippocampus with a hole allowing fornix fibers to pass through. The wall is sealed to the bottom of the bath with inert grease. Care must be taken not to damage the fornix during the placement of the wall. Our custom made plexiglass bath was designed to fit a Siskiyou corporation PC-A chamber adapter mounted on a microscope platform. We have extensively verified that this bath configuration provides completely separated chambers. One way to test this is by using the GABAA receptor antagonist bicuculline as a control during field recordings in septum and hippocampus (Fig. 2b). In the experiment shown, we demonstrate that long-lasting perfusion of bicuculline on the septal side of the bath (10 min) does not cause interictal activity in hippocampus as expected if bicuculline

An Intact Septo-Hippocampal Preparation

249

Fig. 2. (a) The septo-hippocampal preparation in the custom-made two-compartment recording chamber. The preparation is weighted down and held down on a nylon mesh. ACSF is perfused on top and at the bottom of the preparation. A wall separates both sides of the chamber resulting in two independently-sealed compartments. A.C. anterior commisure. (b) The traces on the left show the extracellular field recordings in CA1 and medial septum following a 10-min perfusion of 10 mM bicuculline on the septal side of the chamber. No responses can be observed in hippocampus. In contrast, the traces on the right show a pronounced response in both hippocampus and medial septum to a 2 min perfusion of bicuculline on the hippocampal side.

remains on the septal side and does not permeate in hippocampus. In contrast, direct perfusion of bicuculline on the hippocampal side led to rapid and dramatic interictal-like activity. Alternatively, to confirm that both sides of the bath are hermetically separated, it can be shown in every experiment that the level of ACSF rises only in the compartment where the suction is arrested, or conversely, that the ACSF level is reduced only on the side that the ACSF is removed (by suction). Also, electrical isolation of the two compartments can be checked by determining that the recording electrode is not operative when it is connected to the ground of the opposite chamber.

250

Goutagny, Jackson, and Williams

In these conditions, it is therefore possible to combine field and patch–clamp recordings in the septum and the hippocampus. For field recordings, glass electrodes with a resistance ranging from 1 to 2.5 MW filled with ACSF are gently lowered into CA1 and CA3 regions of the hippocampus as well in the septum. In the CA1 area of the hippocampus, the location of the electrode can be approximately determined by monitoring the recorded electrophysiological signal with an audio monitor (there is usually an increase in spiking activity around the pyramidal layer). However, for the exact location of the electrode, it is necessary to fill the electrode with pontamine sky blue (PSB, 2% in sodium acetate 0.5 M, pH 7.5). At the end of the experiments, the PSB is ejected by passing current of 10 µA (7 s on–7 s off). As a result, a blue point will be visible in the area where the recording was taken. Whole-cell recordings are performed using borosilicate glass pipettes with a resistance of 3–7 MW when filled with (in mM): 144 K-gluconate, 3 MgCl2, 0.2 EGTA, 10 HEPES, 2 ATP, and 0.3 GTP (pH 7.2, 285–295 mOsm). Due to the thickness of the preparation, patching in the whole-cell configuration in the septohippocampal preparation requires the use of the “blind” patch technique (4). Briefly, the recording pipette is slowly lowered into the tissue while monitoring the electrode resistance until a rapid drop is noticed when contact is made with the surface of a cell. For this technique to work well, strong positive pressure has to be applied to the back of the recording pipette. 3.1. Activating the Septum to Trigger Rhythmic Activity in Hippocampus

Can activating the septum induce theta activity in hippocampus and what are the neurotransmitters and receptors involved? As stated earlier, hippocampal theta is made up of at least two different components: an atropine-insensitive type-I and an atropinesensitive type-II theta. It has been suggested that the type-I theta may not directly involve the septum but rather the serotonin system (51) and/or the glutamatergic entorhinal cortex (5) while type-II theta directly involves the cholinergic input from septum. Hence, does the activation of the MS–DB induce theta activity in hippocampus and is this theta atropine-resistant, atropine-sensitive, or both? We used two agonists that were previously shown to activate septal neurons, namely, NMDA and carbachol. When either of the agonists was perfused in the septum in the dual chamber setup, rhythmic activity in the 2–6 Hz frequency was evoked in the hippocampus (11). The rhythmic activity was long-lasting (up to 6 h), was present in both CA3 and CA1, and the frequency was age-dependent with faster frequencies being more prominent at later ages. Figure 3a shows an example of carbachol perfusion in the septum initiating a 4–5 Hz oscillation recorded extracellularly in the hippocampus. Perfusion of the AMPA-kainate receptor antagonist DNQX (Fig. 3b) completely abolished theta suggest-

An Intact Septo-Hippocampal Preparation

251

Fig. 3. Pharmacology of hippocampal oscillations following the application of carbachol to septum. (a) Extracellular recordings in CA1 showing 4 Hz oscillations in response to the application of 10 mM carbachol in septum. (b) These carbachol elicited hippocampal oscillations were reversibly abolished by the AMPA-kainate receptor antagonist DNQX. (c) Such theta oscillations were reversibly reduced by the muscarinic receptor antagonist atropine.

ing that septal triggered hippocampal theta is dependent on AMPA/kainate receptors in the hippocampus. To test whether this hippocampal theta was atropine-sensitive type I, atropine was perfused in the hippocampus (Fig. 3c). Such applications reduced the power of theta by approximately 35%. These results demonstrate

252

Goutagny, Jackson, and Williams

that the theta recorded in the hippocampus following septum activation is both atropine-sensitive and atropine insensitive typeII and type-I, respectively. Together, these results suggest that the septum alone is sufficient to initiate hippocampal theta activity in the absence of other subcortical regions and the entorhinal cortex. Moreover, they suggest that the septum contributes to both type-I and type-II theta in hippocampus. These experiments demonstrate the significant potential of this preparation for investigating theta in hippocampus. It will be possible to analyze the mechanisms underlying theta at the cellular level by combining pharmacology, patch recording of pyramidal cells and GABAergic interneurons, together with the extracellularly recorded field rhythm. 3.2. Recording in Septum During Hippocampal Activation

The septo-hippocampal preparation can also be used to better understand how the hippocampus modulates septal activity. This feedback may be critical for generating and/or modulating septal activity. Anatomically, neurons of the septum send massive projections to the hippocampus, but it is also well known that the hippocampus sends projections to the septum. Previous reports showed that eliminating the hippocampal feedback to the MS–DB in vivo resulted in a dramatic decrease in the number of rhythmic theta MS/DB neurons suggesting paradoxically that the hippocampus may be responsible for pacing the MS–DB (37, 38). In contrast, more recent evidence showed that reversibly cooling axonal fibers of the fimbria-fornix in vivo (that abolishes hippocampal theta) do not arrest the rhythmic bursting of most MS–DB neurons, but produce a profound decrease (66%) in the number of action potentials per bursts (46). Together, these studies suggest that the hippocampal input to the MS–DB may partly provide a rhythmic excitatory input enhancing the strength of rhythmic bursting MS–DB neurons. It remains unknown precisely how hippocamposeptal regulation is important for cognition, but modeling studies suggest that hippocamposeptal feedback plays a vitally important role in novelty-driven learning as well as in acquisition and retention (15, 42). Most of what is known about the hippocamposeptal input is based on neuroanatomical data. The only projection back from the hippocampus to the MS/DB originates from two types of hippocampal GABAergic interneurons (5, 7, 14, 21, 48). It has not been determined if this GABAergic input to the MS–DB is functional but indirect evidence suggest that the hippocampus may influence the septum in a state dependent manner (7). In contrast to these MS–DB projecting hippocampal GABAergic neurons, hippocampal pyramidal and granule neurons project selectively to the lateral septum (31), a region that does not seem to innervate the MS–DB (although this has not been definitely determined; see (9, 41)). Taken together, although the hippocampal input to the septum may be important, a direct demonstration of the nature and role of this

An Intact Septo-Hippocampal Preparation

253

Fig. 4. Role of hippocampus in initiating septal rhythmic activity. The bottom traces shows the Carbachol (30 mM)-elicited rhythmic activity in the CA1 and CA3 of the hippocampus. Concomitantly, extracellularly recorded rhythmic activity was also present in the medial septum (MS) and the diagonal bands (DB). Traces indicated by asterisk on top are enlarged to the right to show that the each event in hippocampus preceded that recorded in the MS–DB suggesting that the hippocampus was driving the activity in septum.

input remains to be shown. We therefore initiated a number of experiments in the septo-hippocampal preparation to investigate how hippocampal rhythmic activity influenced septal neurons (35). Selective carbachol application to the hippocampus was used to elicit rhythmic activity in the hippocampus in the dual chamber. The preparation was first used to examine hippocamposeptal interactions by recording extracellular rhythmic activity elicited in the hippocampus along with field activity in the septum. Figure 4 shows that CA1 and CA3 rhythmic activity triggered by carbachol precedes the rhythmic activity recorded in two regions of the septum (see enlargement on right) suggesting that CA1 and/or CA3 initiates the rhythmic activity in septum. Figure 5 shows an experiment combining pharmacology, extracellular and patch recordings using the dual chamber setup. Figure 5a shows a sketch with the two recording pipettes on each side of the wall of the preparation. Figure 5b displays a typical recording of an MS–DB neuron in a whole cell. Activation of the hippocampus with carbachol elicited a 1–5 Hz rhythm as observed for the CA1 and CA3. Interestingly, phase-locked oscillations were recorded in the septum (enlarged in Fig. 5d,) and this activity were in synchrony with rhythmic IPSPs in the MS/DB neuron (Fig. 5c, last trace). Taken together, these results show that this is a powerful preparation offering possibilities not previously available to understand septo-hippocampal interplay. The hippocampus is a multidimensional structure possessing important circuitry organized along both tangential and longitudinal axis. Furthermore, afferent hippocampal inputs play an important role in determining spike timing during learning and memory processes. Although traditional techniques using slice

254

Goutagny, Jackson, and Williams

Fig. 5. Whole-cell recording of MS–DB neuron concomitantly to field recordings in septum and hippocampus. (a) Schematic of the pipette arrangement in the dual chamber set-up. (b) Whole cell recording of an MS–DB neuron. (c) Top two traces are the extracellular field recording of the carbachol response in CA1 and CA3. Bottom two traces are the extracellular recording from the medial septum (MS; third trace) and the bottom from a whole-cell recorded MS–DB (MS) neuron at resting membrane potential. Trace shows a patch recording of an MS neuron displaying phase-locked rhythmic inhibitory postsynaptic potentials at rest. (d) Enlargement shows that the activity in CA1 precedes that in the MS–DB suggesting that it is the hippocampus that drives the septum in these experiments.

preparations are useful for examining local cell–cell interactions and the molecular processes underlying synaptic transmission, such an approach cannot be of help to investigate complex brain rhythm phenomenon that necessitates intact short- and longrange connectivity of neuronal networks. To completely comprehend and appreciate the complex computational properties of the hippocampal networks, and appropriately study the system dynamics, the complete neural circuit must be intact. The septohippocampal preparation is of invaluable help for the understanding of the cellular mechanisms underlying brain oscillations. References 1. Bassant MH, Jouvenceau A, Apartis E, Poindessous-Jazat F, Dutar P, Billard JM (1998) Immunolesion of the cholinergic basal forebrain: effects on functional properties of hippocampal and septal neurons. Int J Dev Neurosci 16:613–632.

2. Belelli I, Pistis I, Peters JA, Lambert JJ (1999) General anaesthetic action at transmittergated inhibitory amino acid receptors. Trends Pharmacol Sci 20:496–502. 3. Bland BH, Colom LV (1993) Extrinsic and intrinsic properties underlying oscillation and

An Intact Septo-Hippocampal Preparation

4.

5. 6. 7.

8.

9.

10.

11.

12. 13.

14.

15.

16.

synchrony in limbic cortex. Prog Neurobiol 41:157–208. Blanton MG, Lo Turco JJ, Kriegstein AR (1989) Whole cell recording from neurons in slices of reptilian and mammalian cerebral cortex. J Neurosci Meth 30:203–210. Buzsaki G (2002) Theta oscillations in the hippocampus. Neuron 33:325–340. Cole AE, Nicoll RA (1983) Acetylcholine mediates a slow synaptic potential in hippocampal pyramidal cells. Science 221:1299–1301. Dragoi G, Carpi D, Recce M, Csicsvari J, Buzsaki G (1999) Interactions between hippocampus and medial septum during sharp waves and theta oscillation in the behaving rat. J Neurosci 19:6191–6199. Freund TF, Antal M (1988) GABA-containing neurons in the septum control inhibitory interneurons in the hippocampus. Nature 336:170–173. Gallagher JP, Zheng F, Hasuo H, ShinnickGallagher P (1995) Activities of neurons within the rat dorsolateral septal nucleus (DLSN). Prog Neurobiol 45:373–395. Garrido-Sanabria ER, Perez MG, Banuelos C, Reyna T, Hernandez S, Castaneda MT, Colom LV (2007) Electrophysiological and morphological heterogeneity of slow firing neurons in medial septal/diagonal band complex as revealed by cluster analysis. Neuroscience 146:931–945. Goutagny R, Manseau F, Jackson J, Danik M, Williams S (2008) In vitro activation of the medial septum-diagonal band complex generates atropine-sensitive and atropine-resistant hippocampal theta rhythm: an investigation using a complete septohippocampal preparation. Hippocampus 18(6):531–535. Green JD, Arduini AA (1954) Hippocampal electrical activity in arousal. J Neurophysiol 17:533–557. Greenstein YJ, Pavlides C, Winson J (1988) Long-term potentiation in the dentate gyrus is preferentially induced at theta rhythm periodicity. Brain Res 438:331–334. Gulyas AI, Hajos N, Katona I, Freund TF (2003) Interneurons are the local targets of hippocampal inhibitory cells which project to the medial septum. Eur J Neurosci 17:1861–1872. Hasselmo ME, Schnell E (1994) Laminar selectivity of the cholinergic suppression of synaptic transmission in rat hippocampal region CA1: computational modeling and brain slice physiology. J Neurosci 14:3898–3914. Hemmings HC Jr, Akabas MH, Goldstein PA, Trudell JR, Orser BA, Harrison NL (2005) Emerging molecular mechanisms of

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

255

general anesthetic action. Trends Pharmacol Sci 26:503–510. Hentschke H, Perkins MG, Pearce RA, Banks MI (2007) Muscarinic blockade weakens interaction of gamma with theta rhythms in mouse hippocampus. Eur J Neurosci 26: 1642–1656. Holscher C, Anwyl R, Rowan MJ (1997) Stimulation on the positive phase of hippocampal theta rhythm induces long-term potentiation that can be depotentiated by stimulation on the negative phase in area CA1 in vivo. J Neurosci 17:6470–6477. Huxter J, Burgess N, O’Keefe J (2003) Independent rate and temporal coding in hippocampal pyramidal cells. Nature 425: 828–832. Hyman JM, Wyble BP, Goyal V, Rossi CA, Hasselmo ME (2003) Stimulation in hippocampal region CA1 in behaving rats yields long-term potentiation when delivered to the peak of theta and long-term depression when delivered to the trough. J Neurosci 23: 11725–11731. Jinno S, Kosaka T (2002) Immunocytochemical characterization of hippocamposeptal projecting GABAergic nonprincipal neurons in the mouse brain: a retrograde labeling study. Brain Res 945:219–231. Khalilov I, Esclapez M, Medina I, Aggoun D, Lamsa K, Leinekugel X, Khazipov R, Ben Ari Y (1997) A novel in vitro preparation: the intact hippocampal formation. Neuron 19: 743–749. Klausberger T, Magill PJ, Marton LF, Roberts JD, Cobden PM, Buzsaki G, Somogyi P (2003) Brain-state- and cell-type-specific firing of hippocampal interneurons in vivo. Nature 421:844–848. Kohler C, Chan-Palay V, Wu JY (1984) Septal neurons containing glutamic acid decarboxylase immunoreactivity project to the hippocampal region in the rat brain. Anat Embryol (Berl) 169:41–44. Konopacki J, MacIver MB, Bland BH, Roth SH (1987) Carbachol-induced EEG ‘theta’ activity in hippocampal brain slices. Brain Res 405:196–198. Larson J, Lynch G (1986) Induction of synaptic potentiation in hippocampus by patterned stimulation involves two events. Science 232:985–988. Lee MG, Chrobak JJ, Sik A, Wiley RG, Buzsaki G (1994) Hippocampal theta activity following selective lesion of the septal cholinergic system. Neuroscience 62:1033–1047. Leung LS, Shen B, Rajakumar N, Ma J (2003) Cholinergic activity enhances hippocampal

256

29.

30.

31. 32.

33. 34.

35.

36. 37. 38.

39. 40.

41. 42.

43.

Goutagny, Jackson, and Williams long-term potentiation in CA1 during walking in rats. J Neurosci 23:9297–9304. Leutgeb S, Mizumori SJ (1999) Excitotoxic septal lesions result in spatial memory deficits and altered flexibility of hippocampal single-unit representations. J Neurosci 19:6661–6672. Lewis PR, Shute CC, Silver A (1967) Confirmation from choline acetylase analyses of a massive cholinergic innervation to the rat hippocampus. J Physiol 191:215–224. Linke R, Pabst T, Frotscher M (1995) Development of the hippocamposeptal projection in the rat. J Comp Neurol 351:602–616. MacVicar BA, Tse FW (1989) Local neuronal circuitry underlying cholinergic rhythmical slow activity in CA3 area of rat hippocampal slices. J Physiol 417:197–212. Madison DV, Lancaster B, Nicoll RA (1987) Voltage clamp analysis of cholinergic action in the hippocampus. J Neurosci 7:733–741. Manseau F, Danik M, Williams S (2005) A functional glutamatergic neuron network in the medial septum and diagonal band area. J Physiol 566:865–884. Manseau F, Goutagny R, Danik M, Williams S (2008) The hippocamposeptal pathway generates rhythmic firing of GABAergic neurons in the medial septum and diagonal bands: an investigation using a complete septohippocampal preparation in vitro. J Neurosci 28:4096–4107. McBain CJ, Fisahn A (2001) Interneurons unbound. Nat Rev Neurosci 2:11–23. McLennan H, Miller JJ (1974) The hippocampal control of neuronal discharges in the septum of the rat. J Physiol 237:607–624. McLennan H, Miller JJ (1976) Frequencyrelated inhibitory mechanisms controlling rhythmical activity in the septal area. J Physiol 254:827–841. O’Keefe J, Recce ML (1993) Phase relationship between hippocampal place units and the EEG theta rhythm. Hippocampus 3:317–330. Pavlides C, Greenstein YJ, Grudman M, Winson J (1988) Long-term potentiation in the dentate gyrus is induced preferentially on the positive phase of theta-rhythm. Brain Res 439:383–387. Risold PY, Swanson LW (1997) Connections of the rat lateral septal complex. Brain Res Rev 24:115–195. Rokers B, Mercado E III, Allen MT, Myers CE, Gluck MA (2002) A connectionist model of septohippocampal dynamics during conditioning: closing the loop. Behav Neurosci 116:48–62. Semba K (2000) Multiple output pathways of the basal forebrain: organization, chemical

heterogeneity, and roles in vigilance. Behav Brain Res 115:117–141. 44. Simon AP, Poindessous-Jazat F, Dutar P, Epelbaum J, Bassant MH (2006) Firing properties of anatomically identified neurons in the medial septum of anesthetized and unanesthetized restrained rats. J Neurosci 26: 9038–9046. 45. Skaggs WE, McNaughton BL, Wilson MA, Barnes CA (1996) Theta phase precession in hippocampal neuronal populations and the compression of temporal sequences. Hippocampus 6:149–172. 46. Stewart M, Fox SE (1989) Two populations of rhythmically bursting neurons in rat medial septum are revealed by atropine. J Neurophysiol 61:982–993. 47. Stumpf C, Petsche H, Gogolak G (1962) The significance of the rabbit’s septum as a relay station between the midbrain and the hippocampus. II. The differential influence of drugs upon both the septal cell firing pattern and the hippocampus theta activity. Electroencephalogr Clin Neurophysiol 14:212–219. 48. Toth K, Borhegyi Z, Freund TF (1993) Postsynaptic targets of GABAergic hippocampal neurons in the medial septum-diagonal band of broca complex. J Neurosci 13: 3712–3724. 49. Toth K, Freund TF, Miles R (1997) Disinhibition of rat hippocampal pyramidal cells by GABAergic afferents from the septum. J Physiol 500:463–474. 50. Vanderwolf CH (1969) Hippocampal electrical activity and voluntary movement in the rat. Electroencephalogr Clin Neurophysiol 26:407–418. 51. Vanderwolf CH (1988) Cerebral activity and behavior: control by central cholinergic and serotonergic systems. Int Rev Neurobiol 30:225–340. 52. Vertes RP, Hoover WB, Viana Di Prisco G (2004) Theta rhythm of the hippocampus: subcortical control and functional significance. Behav Cogn Neurosci Rev 3:173–200. 53. Winson J (1978) Loss of hippocampal theta rhythm results in spatial memory deficit in the rat. Science 201:160–163. 54. Yoder RM, Pang KC (2005) Involvement of GABAergic and cholinergic medial septal neurons in hippocampal theta rhythm. Hippocampus 15:381–92. 55. Jung R, Kornmüller AE (1938) “Eine Methodik der ablitung lokalisierter Potentialschwankungen aus subcorticalen Hirngebieten”. Arch Psychiat Nervenkr 109:1–30.

Chapter 11 Targeted Modulation of Neural Circuits: A New Treatment Strategy for Neuropsychiatric Disease Helen S. Mayberg and Paul E. Holtzheimer Abstract The last 20 years of clinical neuroscience research has witnessed a fundamental shift in the conceptualization of neuropsychiatric disorders, with the dominant psychological and neurochemical theories of the past now complemented by a growing emphasis on developmental, genetic, molecular, and anatomically based, system-level models. Facilitating this evolving paradigm shift has been the growing contribution of in vivo functional and structural brain imaging techniques that provide an integrative platform to characterize brain circuit dysfunction underlying specific syndromes as well as changes associated with their successful treatment. The impact of this approach is demonstrated by the recent testing of a targeted neuromodulation strategy, deep brain stimulation (DBS), for treatment-resistant major depression. This intervention leverages the system-level models and targeted stimulation techniques pioneered for the treatment of Parkinson’s disease to this and potentially other intractable neuropsychiatric disorders. The theoretical and data-driven foundation for one such imaging-derived network illness model and the initial proof-of-principle testing of subcallosal cingulate white matter DBS for depression is used to illustrate the potential of this evolving treatment strategy. Key words: Functional imaging, Affective disorders, Depression, Neuromodulation, Deep brain stimulation, Glucose metabolism, Cerebral blood flow, Cingulate, Frontal cortex

1. Circuit Model of Depression 1.1. Clinical Context

Critical to development of deep brain stimulation (DBS) as a novel therapy for intractable depression has been the evolving understanding of brain systems mediating normal and abnormal mood states and the ongoing systematic characterization of regional changes mediating successful and unsuccessful response to treatment (21, 27, 33, 95, 126, 152). The specific selection of the subcallosal cingulate region (SCC, Brodmann Area 25) as a first DBS target for testing in patients with treatment-resistant

Robert P. Vertes and Robert W. Stackman Jr. (eds.), Electrophysiological Recording Techniques, Neuromethods, vol. 54, DOI 10.1007/978-1-60327-202-5_11, © Springer Science+Business Media, LLC 2011

257

258

Mayberg and Holtzheimer

depression (TRD) was driven directly by findings from a series of complementary imaging experiments demonstrating consistent changes in this region with clinical recovery with various, wellestablished, evidenced-based antidepressant interventions (96). It was hypothesized that direct stimulation of the SCC25 and adjacent white matter would produce modulatory changes within this putative “depression network” with resulting antidepressant effects (99). As encouraging first clinical results of SCC25 DBS were reported and additional studies initiated (86, 100), continued refinement of this data-driven, system-level depression model has become more than a theoretical exercise. Rather, such a model provides a working template for improving the precision of surgical targeting (46, 49, 63, 119) as well as a translational framework to evaluate mechanisms mediating DBS effects (1, 5, 15, 57, 58, 61, 78, 102, 163), and eventually, a platform to develop evidencebased biomarkers that can effectively guide selection of the most appropriate patients. 1.2. Theoretical Framework

The development and evolution of such an imaging-based depression circuit model is in keeping with the classical neurological tradition of symptom localization. While initially grounded in standard lesion-deficit correlational studies, construction of a working model first required an initial theoretical assumption that depression was unlikely a disease of a single gene, brain region or neurotransmitter system, but rather, a systems disorder where interruption at specific nodes within a defined functional network could result in stereotypic syndromal symptoms (94, 95). With time, this was expanded to further consider a major depressive episode as the net result of failed regulation of this integrated system under circumstances of cognitive, emotional, or somatic stress (96). In this state of sustained allostatic load (103, 127), it was hypothesized that illness remission would require modulation and eventual normalization of these dysregulated pathways, regardless of treatment modality. From this perspective, a given scan pattern and the associated behavior would directly reflect both the inciting “functional lesion” and ongoing compensatory processes influenced by various contributing factors including heredity, temperament, earlylife experiences, current psychosocial stress, previous depressive episodes, as well as prior antidepressant treatment (17, 53, 71). Furthermore, these “brain states,” might additionally define distinct depression phenotypes with varying, but predictable responses to antidepressant treatment (Fig. 1). While this premise is purely speculative, published findings will be reviewed with this hypothesis in mind, highlighting disease, state, and treatmentspecific effects facilitated by pharmacotherapy, cognitive behavioral therapy (CBT), and DBS.

Targeted Modulation of Neural Circuits

259

Fig. 1. Theoretical time course of network changes during a depressive episode. Functional neuroimaging abnormalities are viewed as the net effect of a triggering event and subsequent intrinsic adaptive or maladaptive responses; in other words, failure to self-correct. The nature of these compensatory changes is considered critical for understanding clinical symptom heterogeneity and syndrome subtypes, providing a potential future framework for development of brain-based algorithms for treatment selection. CBT cognitive behavioral therapy; ECT electroconvulsive therapy; DBS deep brain stimulation. (Adapted from Mayberg (96)).

1.3. Current Model

A multi-node, putative network model of depression meeting these characteristics has been constructed from converging findings reported in various functional imaging studies including resting-state patterns in patients with unipolar and bipolar depression and depression associated with specific neurological diseases (94, 97), changes with various antidepressant treatments (42, 98, 99), and studies of acute mood change and emotional processing in healthy volunteers and both depressed patient and at-risk populations (69, 82, 99) (Fig. 2). Regions within this evolving model are clustered into four main functional compartments, based on reproducible patterns across various experiments, and informed by comparative anatomical, electrophysiological, and tract-tracing experiments in non-human primates and other animal model systems (6, 16, 20, 36, 41, 48, 54, 56, 88, 117, 120, 150, 151, 154). Such compartmental groupings aim to accommodate the major defining symptom clusters of major depression (sustained mood, motor, cognitive, circadian dysfunction) and changes in these behaviors accompanying depression treatment and recovery. The model organization further attempts to capture basic cognitive (exteroceptive) and visceral-motor (interoceptive) processes mediating normal responses to novel overt and covert emotional stimuli (13, 20, 32, 34, 50, 52, 64, 75, 108, 112, 114, 125, 137, 143, 147, 155, 160, 164), recognizing that the complexities and nuances of brain mechanisms mediating these behaviors are likely oversimplified by this model structure. With this caveat, consistent functional interactions and region–region correlations are emphasized in the assignment of a particular region to a specific

260

Mayberg and Holtzheimer

Fig. 2. Neural circuit model of depression. Regions with known anatomical interconnections that show consistent changes across converging imaging experiments form the basis of this model. Regions are grouped into four main compartments reflecting general behavioral dimensions of depression and regional targets of various antidepressant treatments. PF prefrontal; PM premotor; Par Parietal; MCC mid-cingulate; dpHc dorsal-posterior hippocampus; PCC posterior cingulate; mF medial frontal; pACC pregenual anterior cingulate; mOF medial orbital frontal; SCC subcallosal cingulate; va-hc ventral-anterior hippocampus; a-ins anterior insula; hth hypothamus; bstem brainstem; vst-cd ventral striatum-caudate; dm-thal dorsomedial thalamus; amyg amygdala; mb-vta midbrain-ventral tegmental area. Numbers Brodmann areas. Small arrows relevant anatomical connections. Large colored arrow putative connections between compartments mediating a specific treatment. CBT cognitive behavioral therapy; meds SSRI pharmacotherapy; DBS subcallosal cingulate white matter deep brain stimulation. (Adapted from Mayberg (96)).

compartment, with explicit anatomical connections linking individual regions both within and across compartments critical to the core model construct. While earlier versions of this model implicated a more limited set of regions (94, 95), what has remained consistent across models is the reciprocal interactions between ventral limbic regions (in dark grey) and dorsal cortical decreases (in black) as a hallmark of a negative mood state (limbic increases, cortical decreases), occurring either transiently, such as with an acute emotional provocation, or when sustained, as seen during a major depressive episode. Similarly, reversal of this pattern (limbic decreases, cortical increases) characterizes mood improvement with depression remission (99, 100). Medial frontal regions (in light grey) have been assigned a distinct compartment in the model, as these regions appear most critical to active cognitive control and overt regulation of emotional and affective state (32, 34, 64, 75, 114, 155, 160). Similarly, a set of subcortical regions including the amydala, ventral striatum, and thalamus (in white) consistently implicated in primary and often covert processing of novel emotional and non-emotional stimuli have been grouped together to

Targeted Modulation of Neural Circuits

261

emphasize their more general role in evaluating salience (164) and in mediating reinforcement, learning, habit, and extinction (50, 74, 108, 137, 143, 147). Ventral rostral and dorsal subregions of the anterior cingulate cortex have been similarly segregated in keeping with their differential anatomical connections within and between compartments (6, 16, 41, 123, 154). Within this framework, synchronized changes within and across compartments are considered critical for illness remission, regardless of treatment modality, accommodating pharmacotherapy as well as cognitive and surgical interventions. Strategies to formally test and distinguish disease-specific and response-specific functional interactions among regions in this depression network are illustrated in the following sections.

2. Defining Model Constituents 2.1. Structural Imaging Studies

Structure–function correlations performed in patients developing a depressive syndrome in the context of either acquired brain lesions (76, 132) or neurodegenerative disorders (94) provided an early anatomical perspective, consistently identifying clear abnormalities in both the frontal cortex and basal ganglia. Studies of structural abnormalities in primary depression on the other hand have proven more variable, generally necessitating more advanced image acquisition and analytic approaches. Such studies have identified volume changes in the amygdala, hippocampus, anterior cingulate and both ventromedial and prefrontal cortex, but with considerable variability (27, 28, 140); Post-mortem studies more consistently identify glial loss, but the findings are widespread (10, 51, 116, 131). Unlike depression following a selective brain lesion where causal inferences can be reasonably asserted, volume changes reported in major depressive disorder (MDD) appear to be more complex, particularly since acute lesions of the amygdale, hippocampus, cingulate and ventromedial frontal cortex do not reliably precipitate depressive symptoms or syndromes. Furthermore, genetic risk factors and environmental stress may further contribute to some of these findings (104, 124, 130). Nonetheless, comparative cytoarchitectural and connectivity studies generally confirm the critical involvement of these regions in animal models of depression and associated emotional behaviors (1, 5, 6, 24, 78) supporting the hypothesis that even subtle disruption of pathways linking these regions in humans can result in disturbances in emotion regulation characterized by negative mood coupled with sustained changes in motivation, motor performance, cognition, and circadian functions. Studies of regional brain dysfunction further support this hypothesis.

262

Mayberg and Holtzheimer

2.2. Functional Imaging Studies

There are now a variety of imaging methods [positron emission tomography (PET), single photon emission computed tomography (SPECT), functional magnetic resonance imaging (fMRI), MR spectroscopy, electroencephalography (EEG), magnetoencephalography (MEG), optical imaging] capable of quantifying a wide range of physiological parameters relevant to the study of major depression. In this overview, resting-state blood flow and glucose metabolism measures using PET are highlighted. Functional imaging studies of primary depression (27, 33, 96, 152) commonly report frontal and cingulate abnormalities, a pattern also seen in neurological depressions (94). Other limbic-paralimbic (amygdala, anterior temporal, insula), and subcortical (basal ganglia, thalamus) abnormalities have also been identified, but the findings are more variable. Across studies, the most robust and best-replicated finding is that of decreased prefrontal function, although normal frontal as well as frontal hyperactivity have also been reported (7, 29). Localization of abnormalities within the frontal lobe includes dorsolateral and ventral lateral prefrontal cortex (Brodmann Areas BA 9, 46, 10, 47), as well as orbital frontal and ventral medial frontal cortices (BA 11, 32, 10). Findings are generally bilateral, although asymmetries are described. Cingulate changes are also commonly seen and consistently involve anterior sectors (BA 24).

2.3. Potential Sources of Variability

While there are clearly a highly reproducible set of functional and structural findings across studies, not all patients show the same pattern. Differences among patient subgroups (familial, bipolar, unipolar, neurological, early trauma), as well as heterogeneous expression of clinical symptoms such as illness severity, cognitive impairment, anxiety, anhedonia, mood reactivity, and psychomotor slowing, are thought to contribute to the described variance, but there is not yet a consensus. The best-replicated behavioral correlate of a resting-state abnormality in depression is that of an inverse relationship between prefrontal activity and depression severity. Prefrontal activity has also been linked to psychomotor speed and executive functions; parietal and parahippocampus with anxiety; medial frontal and cingulate with cognitive performance, ventral striatum/nucleus accumbens with anhedonia; and amygdala with cortisol status (11, 26, 30, 118, 147, 153, 155). A more complex ventral–dorsal segregation of frontal lobe functions has also been described with anxiety/tension positively correlated with ventral prefrontal activity and psychomotor and cognitive slowing negatively correlated with dorsolateral activity (11). The prefrontal cortex and amygdale over-activity seen in patients with a more ruminative/ anxious clinical presentation is also consistent with findings described in primary anxiety and obsessional disorders (118), memory-evoked anxiety and fear in healthy subjects, response to

Targeted Modulation of Neural Circuits

263

the testing environment due to novelty or state anxiety, and even gene-mediated variability in emotional reactivity (50). Even with these considerations, presence of clinical symptom variability within a given patient cohort does not appear to fully explain the “consistent” inconsistencies in the published imaging literature (53, 96). One can alternatively consider variable patterns from a systems perspective, as outlined in Fig. 1, where dysregulated “network” activity identified in the baseline depressed state is seen to reflect both foci of primary dysfunction as well as sites of attempted (or failed) adaptation. Such a model would theoretically accommodate the reported variability among published depression cohorts, the recognized heterogeneity of depressive symptoms, and purported etiologic risk factors (17, 71) and is also in keeping with conceptual models of sustained allostatic load (103). Hypothetically, in the setting of sustained over-activity of the regulatory “network” (whatever the cause), an exaggerated or hypersensitive compensatory response may result in an agitated, mood-reactive, ruminative depressive state in one patient, whereas failure to initiate or maintain an adequate compensatory response may lead to anergia, psychomotor retardation, apathy, and mood non-reactivity in an equally severe second patient. In this context, a sustained but partially compensated state (Fig. 1, brain type B) would likely respond equally well to either pharmacological or psychological treatments, consistent with empirical clinical experience as well as randomized-controlled studies (23). On the other hand, more extreme states of adaptation (either over-active or underactive) would require more specific treatments (i.e., CBT or interpersonal psychotherapy (IPT) for Brain State A (109); medication augmentation or ECT, VNS, rTMS for Brain State C) (70, 79, 110, 134, 156), with sustained failure in state D defining treatment-resistant patients and a need for more aggressive interventions, such as DBS (62, 86, 90, 100, 138). Such hypotheses lay the foundation for a related goal, namely that a specific neural signature may eventually provide a therapeutic road map for optimal treatment selection in individual depressed patients, if baseline variability and associated change patterns with different treatment interventions can be fully characterized (18, 96). While important insights have been made using group-based analyses, as described in the following sections, practically speaking, fMRI may prove a more agile technology to test these hypotheses, since such neural interactions, commonly termed “functional connectivity,” are testable in individual subjects. Such strategies emphasize not merely the absolute state of regional activity but rather the way in which activity in different locales influence one another as indexed using region–region correlations or covariances (2, 4, 38, 45, 55, 122, 149).

264

Mayberg and Holtzheimer

2.4. Scan Variability as a Biomarker of Response Likelihood

3. Brain Targets of Antidepressant Treatments

3.1. Medication

Toward this eventual goal, several groups have already identified pre-treatment scan patterns that differentiate response-specific subtypes to various treatments (18, 25, 77, 83, 97, 128, 136, 139, 142, 159). Retrospective analyses of resting state PET studies and more recently fMRI challenge studies have consistently reported that increased pretreatment activity in the anterior cingulate (Cg24) distinguishes responders from non-responders to several different antidepressant interventions (18, 83, 97, 128, 136). It has not yet been prospectively tested if this pattern differentially predicts response to a specific treatment class (83). Using similar methods, subcallosal cingulate hyperactivity has also been identified as a potential response predictor in previously drug non-responsive patients treated with sleep deprivation (159), cingulotomy (25), and DBS (100). Systematic characterization of these potential predictive patterns alone or in combination with pharmacogenetic markers (31, 80) has important therapeutic implications in light of increasing evidence that the presence of residual symptoms places patients at increased risk for future relapse or recurrence (134).

As seen in studies of the baseline depressed state, PET measures of regional glucose metabolism and regional cerebral blood flow and, more recently, resting state and task functional MRI (fMRI) have also proven to be sensitive indices of changing brain function following various treatments. Changes in cortical, limbicparalimbic and subcortical regions have been described following diverse treatments such as medication, psychotherapy, sleep deprivation, electroconvulsive therapy (ECT), repetitive transcranial magnetic stimulation (rTMS), vagus nerve stimulation (VNS) ablative surgery, and DBS (3, 19, 33, 39, 42, 72, 73, 86, 96, 98–100, 111, 121, 141). While normalization of frontal abnormalities is the best-replicated finding, other regional effects are reported with variable patterns with different treatments. Contributing to variability is the nature of the specific imaging strategy and behavior provocation or task used to test the effect of treatment. Despite this caveat, modality-specific effects are consistent with the hypothesis that different interventions modulate specific regional targets, resulting in a variety of complementary, adaptive chemical, and molecular changes sufficient to re-establish a euthymic, remitted state (Figs. 1 and 2). Across studies of chronic antidepressant treatment using commonly prescribed medications, prefrontal cortical changes are the most consistently reported, with normalization of frontal

Targeted Modulation of Neural Circuits

265

over-activity and underactivity both described (42, 72, 73, 98). Additionally, changes are also seen in limbic and subcortical regions, including subgenual cingulate, amygdale, hippocampus, posterior cingulate and insula, with decreases in activity most commonly observed (3, 39, 42, 98, 141). The time course of these medication effects and differences between responders and non-responders have provided additional localizing clues as to critical brain changes mediating depression remission. In one such experiment (98), responders and non-responders to fluoxetine were differentiated by their 6-week metabolic change pattern with clinical improvement associated with limbic-paralimbic and striatal decreases and dorsal cortical increases. Failed response was associated with persistence of the 1-week pattern seen in both groups, and absence of either subcallosal cingulate (SCC25) or prefrontal changes. This combination of reciprocal dorsal cortical and ventral limbic changes appears to be a common pattern with response to serotonin reuptake inhibitors (SSRIs) (72), placebo medication (101) as well as combination serotonin-norepinephrine reuptake inhibitors (SNRIs) (73). The reversal of the 1-week pattern at 6 weeks in only those patients who showed clinical improvement suggests a process of neural plasticity or adaptation in specific brain regions with chronic treatment. These responder– non-responder differences are also consistent with the time course and location of changes identified in animal studies of antidepressant medications which emphasize early brainstem and hippocampal changes and late cortical effects involving presynaptic autoregulatory desensitization, up- and down-regulation of multiple post-synaptic receptor sites, and receptor-mediated second messenger, and neurotrophic intracellular signaling effects (35, 37). 3.2. Psychotherapy

In contrast to pharmacological treatments, theoretical models of CBT for the treatment of depression implicate “top–down” mechanisms, as the intervention focuses on modifying attention and memory functions involved in the mediation of depressionrelevant explicit cognitions, affective bias, and maladaptive information processing (133, 157), putatively localized to orbital, medial frontal, prefrontal, and anterior cingulate regions (64, 112, 113). Imaging studies examining brain changes followingIPT and CBT report significant regional effects—most prominently decreases in prefrontal cortex (12, 42, 73, 92), but with differential non-frontal changes depending on the specific therapeutic intervention employed. For example, using CBT, remission was associated with not only prefrontal changes but also decreases in posterior cingulate, dorsomedial frontal, and orbital frontal cortex as well as increases in anterior mid-cingulate and parahippocampal regions (42). This CBT-specific change pattern was generally replicated in a follow-up, randomized study

266

Mayberg and Holtzheimer

c omparing CBT to venlafaxine (73). Interestingly, this second study identified both a common decrease in dorsomedial frontal cortex with both treatments as well as reciprocal changes in subgenual cingulate activity (increases with CBT, decreases as seen previously with medication) further suggesting a critical role for this region in mediating depression remission across treatments.

4. Critical Role for the Subcallosal Cingulate

4.1. Comparative Anatomy Studies

Among the series of treatment studies surveyed, the involvement of the subcallosal cingulate region (SCC) is especially prominent (Fig. 3 a–f). Not only do changes in this region appear critical for antidepressant response to active and placebo pharmacotherapy, ECT, and CBT (73, 99, 101, 111), but functional hyperactivity of this region best characterizes treatment-resistant patients (Fig. 3, j–n) (25, 45, 77, 100, 139). Furthermore, anatomical changes on structural MRI scans as well as post-mortem identification of glial abnormalities (28, 116) are reported in depressed patient samples (Fig. 3, i). In addition, structural and functional variability in this region has been linked to a normal polymorphism in the serotonin transporter, an emerging depression risk factor (Fig. 3,h) (124). These converging anatomical findings complement a large functional imaging literature linking the SCC to the regulation of negative emotional states. Increases in activity in this region are seen with acute provocation of sad mood using either autobiographical memory or a pharmacological challenge such as tryptophan depletion (Fig. 4, f, g) (99, 146) as well as with passive exposure to sad, negative or unpleasant pictures, words, and music (44, 142, 161). Foundation for a role of the SCC in the autonomic and circadian aspects of depression including alterations in sleep, appetite, libido, and endocrine functioning is also suggested by this region’s afferent and efferent connections to the insula, brainstem, and hypothalamus (6, 20, 36, 41, 56, 115, 150). Furthermore, stimulation SCC25 grey matter in monkeys, rats, and shrews evokes a reproducible vocalization, the so-called isolation call, exhibited in the wild with acute separation (65, 68, 87, 89, 158). These studies provide an interesting bridge between animal and human behaviors associated with maternal separation and loss (84, 85, 145). Studies of regional glucose metabolism during such vocalizations is associated with prominent activity in subgenual regions (65–67, 107) consistent with the findings with blood flow PET during transient negative mood induction (99). Reciprocal pathways linking SCC25 to orbitofrontal, medial frontal, and dorsal prefrontal cortices, anterior and posterior

Targeted Modulation of Neural Circuits

267

Fig. 3. Converging evidence implicating the subcallosal cingulate region in major depression. Top row common pattern of glucose metabolic or blood flow decreases (black arrows) in SCC25 with antidepressant response to various interventions. Images demonstrate group change patterns relative to the baseline depressed state for each treatment. Exception is with cognitive behavioral therapy, where successful response is associated with increases in this region. (a) Response to a serotonin reuptake inhibitor, fluoxetine (SSRI) (8); (b) placebo pill (124); (c) serotonin and norepinephrine reuptake inhibitor, venlafaxine (SNRI) (117); (d) electroconvulsive therapy (ECT) (118); (e) cognitive behavioral therapy (CBT) (117). Middle row Common pattern of SCC25 blood flow increases (white arrows) with induction of transient sadness induced by (f): recollection of a personal sad memory (8); (g) tryptophan depletion (132); (h) anatomical differences in SCC25 distinguish healthy subjects homozygous for the S allele of the serotonin transporter promotor gene (a putative risk factor for depression) relative to L/L carriers (78); (i) area of decreased glial number in post-mortem studies of depressed patients relative to non-depressed subjects (73). Bottom row emerging baseline pattern of elevated resting-state SCC25 activity (white arrows) in more treatment-resistant depressed patients. (j) glucose metabolism increases in CBT and venlafaxine non-responders relative to both healthy subjects and similarly depressed patients who responded to either treatment (109); (k) resting BOLD fMRI increases relative to healthy controls (105); (l) glucose metabolic increases in treatment-resistant patients who later responded to cingulotomy relative to those that failed to respond (112); (m) blood flow increases in treatment-resistant patients enrolled in DBS treatment trial relative to healthy controls (9); N: structural equation model demonstrating abnormal functional connectivity of SCC25 in medication-resistant patients relative to CBT and medication-responsive patients (111). Images courtesy of Helen Mayberg (a–c, e, f, j, m, n), Mitch Nobler (d), Peter Talbot (g), Dan Weinberger (h), Dost Ongur (i), Michael Greicius (k), Darin Dougherty (l).

c ingulate, as well as to the amygdale, hippocampus, and nucleus accumbens further identify plausible pathways by which interceptive and homeostatic processes might influence aspects of learning, memory, reward, and reinforcement (16, 48, 117, 123, 150) core behaviors impaired in depressed patients. Interestingly, these various connections show considerable overlap with the pattern of regional changes seen with both CBT and pharmacotherapy

268

Mayberg and Holtzheimer

Fig. 4. Selectively targeting the depression network with deep brain stimulation in treatment-resistant depression. Images left to right: Pre-op MRI demonstrating anatomical location of DBS electrode within the SCC white matter; the four contacts on the electrode are numbered. Pre-op blood flow PET scan demonstrating baseline hyperactivity of the SCC in the TRD study group relative to healthy controls. The post-op MRI with the electrodes in place within the SCC white matter. Six-month change relative to pre-op baseline associated with chronic DBS to the SCC. White increases in blood flow; black decreases in blood flow. SCC subcallosal cingulate, oF orbital frontal, mF medial frontal, MCC mid-cingulate; hth hypothalamus, vst ventral striatum, bs brainstem; sn substantia nigra. Numbers are Brodmann designations (adapted from Mayberg et al. (100)).

treatment described above (Fig. 2), providing strong evidence to ursue strategies that might effectively alter SCC connectivity in p treatment-resistant depression.

5. Testing the Model: Deep Brain Stimulation Targeting the SCC

The repeated observation of significant increased activity in SCC25 in studies of acute negative affective states, cellular abnormalities in this region in depressed patients post-mortem, and predictable decreases in activity with a variety of pharmacological and somatic antidepressant treatments provided the critical foundation to test the use of direct modulation of SCC25 using highfrequency DBS as a novel treatment strategy for otherwise treatment-resistant major depression. It was hypothesized that focal stimulation of the SCC and adjacent white matter would reduce chronically elevated SCC25 activity with resulting clinical benefit by impacting not just the SCC but also those brain regions within the “network” directly connected to the SCC via the targeted white matter tracts (16, 47, 117). In a proof-of-principle study, six patients with refractory depression were implanted and chronic bilateral stimulation of white matter tracts adjacent to SCC25 produced a sustained remission of depressive symptoms in four of the six patients (100). Clinical antidepressant effects were further associated with a marked reduction in SCC blood flow, as well as changes in downstream limbic and cortical sites (decreases in hypothalamus, ventral striatum, orbital

Targeted Modulation of Neural Circuits

269

and medial frontal cortex; increases in dorsolateral prefrontal, parietal and mid-cingulate, and posterior cingulate cortex), measured using PET consistent with the primary hypothesis (Fig. 4). The pattern of changes associated with antidepressant response suggested that electrical modulation of this seemingly critical hub within a putative “depression network” was a reasonable tactic and worthy of further study. Studies in an additional 14 patients confirmed these first observations, with an overall 6-month response rate of 60% and with sustained response exceeding 1 year (86). Placebo-controlled studies extending these first studies are now underway examining both clinical efficacy and mechanisms. 5.1. Emergent Questions

Equally tantalizing to the apparent sustained rate of clinical recovery in this first experimental group was the observation that the precise target of stimulation was extremely critical, with variable clinical effects seen with stimulation of adjacent contacts separated by mere millimeters along the same electrode (Fig. 4, left image). Similarly, with patients awake in the operating room, positive behavioral effects were often seen with acute stimulation of some but not all of the individual contacts. Such acute effects often predicted future sustained antidepressant response with chronic stimulation at the same contact, but not always. Spontaneous patient reports of “sudden calmness or lightness,” sense of heightened awareness, increased interest, and “connectedness,” as well as objective increases in motor speed, volume and rate of spontaneous speech and improved prosody were observed (100). These provocative findings have not been systematically characterized either clinically or with appropriate imaging strategies but are the focus of current experiments.

5.2. Methodological Considerations

As refinement of DBS methods for intractable depression evolve, and as the technology is tested in other neuropsychiatric disorders, a number of additional methodological issues must be considered.

5.2.1. Patient Selection and Assessment

Standardized diagnostic and assessment tools (such as the Structured Clinical Interview for DSM-IV Diagnoses, Hamilton Depression Rating Scale, and Beck Depression Inventory) are well established for use in depression research. However, it must be recognized that these instruments have largely been developed and validated within a broadly defined population of depressed patients who are not treatment resistant. As the TRD patients appropriate for SCC DBS studies at this time comprise a small minority of the depressed population at large (probably less than 5%), it may be that currently accepted clinical instruments may not be as applicable. It is likely that the chronicity of the illness, as well as behavioral adaptations necessary to survive, may alter the clinical presentation of the illness. It is also possible that patients

270

Mayberg and Holtzheimer

responsive to SCC DBS represent a distinct biological subtype of depression associated with a phenomenology at least somewhat distinct from “typical” depression. As research in this area progresses, it will be important for methodological work to focus on carefully characterizing the signs and symptoms of this TRD population, especially clinical features that may distinguish those patients who do and do not respond to SCC DBS. Additionally, as highlighted in the previous section, careful attention should be paid to which elements of the illness may change acutely vs. chronically with stimulation. Such investigation may help delineate which specific aspects of the neural network involved are associated with specific symptoms of the disorder; this would have broad implications for treatment development beyond DBS for depression. 5.2.2. Target Identification

Currently, targeting for DBS procedures is accomplished using frame or frameless stereotaxic systems that incorporate high-resolution structural neuroimaging (often MRI) data. As neuroimaging techniques continue to develop, it will be possible to incorporate additional information to help identify the optimal target for DBS lead placement. For example, diffusion tensor imaging data, combined with novel probabilistic tractography methods, have been used to characterize the structural connectivity of the SCC region (46, 63). It is likely that such additional information may be helpful in fine-tuning target selection for DBS procedures, particularly when combined with detailed models of the electrode–tissue interface (14). Functional neuroimaging (e.g., resting-state and task-activated fMRI) may also be useful in identifying the specific nature of network dysfunction within a single patient: it is currently being tested whether specific patterns of functional connectivity are associated with TRD and clinical response to SCC DBS. If these patterns are identified and validated, these subject-level data could be used to help select patients for this procedure and potentially optimize target selection.

5.2.3. Stimulation Characteristics

Current DBS systems include (1) a thin insulated, coiled wire lead containing several electrodes; (2) an implanted pulse generator (IPG)/battery pack surgically placed within the patient’s body (often subcutaneously in the chest wall); and (3) an extension wire connecting the first two components. Typically, stimulating electrodes range from 1 to 3 mm in length and ≤1 mm in diameter and are spaced 1–3 mm apart; most DBS systems available or being studied for neuropsychiatric conditions contain four electrodes per lead, though fewer or more electrodes could theoretically be incorporated. Stimulation parameters typically fall within the falling ranges: 100–130 Hz, 60–200 µs pulse width, 4–8 mA. It should be noted, however, that most available systems allow modulation of each of these parameters within a wide range

Targeted Modulation of Neural Circuits

271

(e.g., frequency could be decreased to <5 Hz and pulse width increased to >400 µs). Monopolar stimulation is common with the IPG serving as the anode; however, bipolar stimulation is possible (with two electrodes within the lead serving as anode and cathode). Stimulation is usually constant (i.e., occurring at all times 24 h a day, 7 days a week), though incorporating a duty cycle (e.g., 5 min on, 20 min off) is possible with some DBS systems. To date, the initial stimulation parameters used for neuropsychiatric conditions have derived largely from those used in movement disorders, with subsequent adjustment based on clinical response. However, it is yet to be determined whether these parameters are optimal. From the above, it is obvious that the number of potential stimulation parameter combinations approaches infinity, complicating systematic investigation. Going forward, innovative approaches to testing various stimulation parameter combinations will be needed. Related to this, critical computer modeling work has focused on changes in the charge density field produced by DBS within neural tissues (105); this line of investigation may be especially helpful in determining how various parameter modulations affect the shape and intensity of charge delivery within the targeted brain region. Additionally, animal studies may be especially helpful in evaluating the behavioral and neurobiological effects of varying stimulation parameters more efficiently and in a more controlled way than in individual patients. Such studies will provide new models that inform on systemlevel dynamics mediating the transition from acute to sustained antidepressant effects at the cellular, local circuit, and network level.

6. Future Directions Multi-center, randomized, placebo-controlled trials will ultimately be necessary to determine the clinical efficacy of DBS of the SCC (86, 100) and the other targets currently under study for the treatment of TRD (62, 90, 138). That said, complementary research investigations additionally provide a unique opportunity to examine depression, its pathophysiology, and mechanisms of treatment from a new perspective. Specific studies of DBS will optimally benefit from a flexible research infrastructure that can take best advantage of ongoing advances in multimodal imaging and computational modeling of complex circuits and new animal models (15, 43, 60, 91, 119, 135, 163). As has proven to be the case with DBS for Parkinson’s disease (9, 22, 81, 106, 129, 148), such platforms will facilitate the necessary translational studies

272

Mayberg and Holtzheimer

needed to fully characterize DBS effects at the cellular, molecular, and network levels (1, 8, 58–60, 102, 144). It is further envisioned that continued refinement of brain-based illness models for depression and other neuropsychiatric disorders will remain an important strategy for novel treatment development more generally (5, 40, 93, 162) as well as in redefining a new depression nosology, with additional opportunities for developing medical algorithms that will ultimately guide treatment selection in individual patients.

Acknowledgments I thank my many colleagues who contributed to these studies. Work supported by grants from National Institutes of Health (NIH), NARSAD, the Dana Foundation, Canadian Institutes for Health Research (CIHR), Stanley Medical Research Foundation, and the Woodruff Fund. References 1. Airan RD, Meltzer LA, Roy M, Gong Y, Chen H, Deisseroth K (2007) High-speed imaging reveals neurophysiological links to behavior in an animal model of depression. Science 317:819–823. 2. Anand A, Li Y, Wang Y, Wu J, Gao S, Bukhari L, Mathews VP, Kalnin A, Lowe MJ (2005) Activity and connectivity of brain mood regulating circuit in depression: a functional magnetic resonance study. Biol Psychiatry 57:1079–1088. 3. Anand A, Li Y, Wang Y, Wu J, Gao S, Bukhari L, Mathews VP, Kalnin A, Lowe MJ (2005) Antidepressant effect on connectivity of the mood-regulating circuit: an FMRI study. Neuropsychopharmacology 30:1334–1344. 4. Arfanakis K, Cordes D, Haughton VM, Moritz CH, Quigley MA, Meyerand ME (2000) Combining independent component analysis and correlation analysis to probe interregional connectivity in fMRI task activation datasets. Magn Reson Imaging 18:921–930. 5. Banasr M, Duman RS (2007) Regulation of neurogenesis and gliogenesis by stress and antidepressant treatment. CNS Neurol Disord Drug Targets 6:311–320. 6. Barbas H, Saha S, Rempel-Clower N, Ghashghaei T (2003) Serial pathways from primate prefrontal cortex to autonomic areas may influence emotional expression. BMC Neurosci 4:25.

7. Baxter LR Jr, Phelps ME, Mazziotta JC, Schwartz JM, Gerner RH, Selin CE, Sumida RM (1985) Cerebral metabolic rates for glucose in mood disorders. Studies with positron emission tomography and fluorodeoxyglucose F 18. Arch Gen Psychiatry 42:441–447. 8. Belujon P, Grace AA (2008) Critical role of the prefrontal cortex in the regulation of hippocampus-accumbens information flow. J Neurosci 28:9797–9805. 9. Benabid AL, Pollak P, Louveau A, Henry S, de Rougemont J (1987) Combined (thalamotomy and stimulation) stereotactic surgery of the VIM thalamic nucleus for bilateral Parkinson disease. Appl Neurophysiol 50:344–346. 10. Bowley MP, Drevets WC, Ongur D, Price JL (2002) Low glial numbers in the amygdala in major depressive disorder. Biol Psychiatry 52:404–412. 11. Brody AL, Saxena S, Mandelkern MA, Fairbanks LA, Ho ML, Baxter LR (2001) Brain metabolic changes associated with symptom factor improvement in major depressive disorder. Biol Psychiatry 50:171–178. 12. Brody AL, Saxena S, Stoessel P, Gillies LA, Fairbanks LA, Alborzian S, Phelps ME, Huang SC, Wu HM, Ho ML, Ho MK, Au SC, Maidment K, Baxter LR Jr (2001) Regional brain metabolic changes in patients with major depression treated with either paroxetine or interpersonal

Targeted Modulation of Neural Circuits therapy: preliminary findings. Arch Gen Psychiatry 58:631–640. 13. Bush G, Luu P, Posner MI (2000) Cognitive and emotional influences in anterior cingulate cortex. Trends Cogn Sci 4:215–222. 14. Butson CR, Cooper SE, Henderson JM, McIntyre CC (2007) Patient-specific analysis of the volume of tissue activated during deep brain stimulation. Neuroimage 34:661–670. 15. Buzsaki G, Draguhn A (2004) Neuronal oscillations in cortical networks. Science 304:1926–1929. 16. Carmichael ST, Price JL (1996) Connectional networks within the orbital and medial prefrontal cortex of macaque monkeys. J Comp Neurol 371:179–207. 17. Caspi A, Sugden K, Moffitt TE, Taylor A, Craig IW, Harrington H, McClay J, Mill J, Martin J, Braithwaite A, Poulton R (2003) Influence of life stress on depression: moderation by a polymorphism in the 5-HTT gene. Science 301:386–389. 18. Chen CH, Ridler K, Suckling J, Williams S, Fu CH, Merlo-Pich E, Bullmore E (2007) Brain imaging correlates of depressive symptom severity and predictors of symptom improvement after antidepressant treatment. Biol Psychiat 62:407–414. 19. Chen CH, Suckling J, Ooi C, Fu CH, Williams SC, Walsh ND, Mitterschiffthaler MT, Pich EM, Bullmore E (2008) Functional coupling of the amygdala in depressed patients treated with antidepressant medication. Neuropsychopharmacology 33:1909–1918. 20. Craig AD (2002) How do you feel? Interoception: the sense of the physiological condition of the body. Nat Rev Neurosci 3:655–666. 21. Davidson RJ, Pizzagalli D, Nitschke JB, Putnam K (2002) Depression: perspectives from affective neuroscience. Annu Rev Psychol 53:545–574. 22. DeLong MR, Wichmann T (2007) Circuits and circuit disorders of the basal ganglia. Arch Neurol 64:20–24. 23. DeRubeis RJ, Hollon SD, Amsterdam JD, Shelton RC, Young PR, Salomon RM, O’Reardon JP, Lovett ML, Gladis MM, Brown LL, Gallop R (2005) Cognitive therapy vs medications in the treatment of moderate to severe depression. Arch Gen Psychiatry 62:409–416. 24. Diorio D, Viau V, Meaney MJ (1993) The role of the medial prefrontal cortex (cingulate gyrus) in the regulation of hypothalamic-pituitary-adrenal responses to stress. J Neurosci 13:3839–3847.

273

25. Dougherty DD, Weiss AP, Cosgrove GR, Alpert NM, Cassem EH, Nierenberg AA, Price BH, Mayberg HS, Fischman AJ, Rauch SL (2003) Cerebral metabolic correlates as potential predictors of response to anterior cingulotomy for treatment of major depression. J Neurosurg 99:1010–1017. 26. Drevets WC, Price JL, Bardgett ME, Reich T, Todd RD, Raichle ME (2002) Glucose metabolism in the amygdala in depression: relationship to diagnostic subtype and plasma cortisol levels. Pharmacol Biochem Behav 71:431–447. 27. Drevets WC, Price JL, Furey ML (2008) Brain structural and functional abnormalities in mood disorders: implications for neurocircuitry models of depression. Brain Struct Funct 213:93–118. 28. Drevets WC, Price JL, Simpson JR Jr., Todd RD, Reich T, Vannier M, Raichle ME (1997) Subgenual prefrontal cortex abnormalities in mood disorders. Nature 386:824–827. 29. Drevets WC, Videen TO, Price JL, Preskorn SH, Carmichael ST, Raichle ME (1992) A functional anatomical study of unipolar depression. J Neurosci 12:3628–3641. 30. Dunn RT, Kimbrell TA, Ketter TA, Frye MA, Willis MW, Luckenbaugh DA, Post RM (2002) Principal components of the Beck Depression Inventory and regional cerebral metabolism in unipolar and bipolar depression. Biol Psychiatry 51:387–399. 31. Eichelbaum M, Ingelman-Sundberg M, Evans WE (2006) Pharmacogenomics and individualized drug therapy. Annu Rev Med 57:119–137. 32. Etkin A, Egner T, Peraza DM, Kandel ER, Hirsch J (2006) Resolving emotional conflict: a role for the rostral anterior cingulate cortex in modulating activity in the amygdala. Neuron 51:871–882. 33. Fitzgerald PB, Laird AR, Maller J, Daskalakis ZJ (2008) A meta-analytic study of changes in brain activation in depression. Hum Brain Mapp 29:683–695. 34. Fossati P, Hevenor SJ, Graham SJ, Grady C, Keightley ML, Craik F, Mayberg H (2003) In search of the emotional self: an FMRI study using positive and negative emotional words. Am J Psychiatry 160:1938–1945. 35. Frechilla D, Otano A, Del Rio J (1998) Effect of chronic antidepressant treatment on transcription factor binding activity in rat hippocampus and frontal cortex. Prog Neuropsychopharmacol Biol Psychiatry 22:787–802. 36. Freedman LJ, Insel TR, Smith Y (2000) Subcortical projections of area 25 (subgenual

274

Mayberg and Holtzheimer

cortex) of the macaque monkey. J Comp Neurol 421:172–188. 37. Freo U, Ori C, Dam M, Merico A, Pizzolato G (2000) Effects of acute and chronic treatment with fluoxetine on regional glucose cerebral metabolism in rats: implications for clinical therapies. Brain Res 854:35–41. 38. Friston K, Phillips J, Chawla D, Buchel C (1999) Revealing interactions among brain systems with nonlinear PCA. Hum Brain Mapp 8:92–97. 39. Fu CH, Williams SC, Cleare AJ, Brammer MJ, Walsh ND, Kim J, Andrew CM, Pich EM, Williams PM, Reed LJ, Mitterschiffthaler MT, Suckling J, Bullmore ET (2004) Attenuation of the neural response to sad faces in major depression by antidepressant treatment: a prospective, event-related functional magnetic resonance imaging study. Arch Gen Psychiatry 61:877–889. 40. Garcia LS, Comim CM, Valvassori SS, Reus GZ, Andreazza AC, Stertz L, Fries GR, Gavioli EC, Kapczinski F, Quevedo J (2008) Chronic administration of ketamine elicits antidepressant-like effects in rats without affecting hippocampal brain-derived neurotrophic factor protein levels. Basic Clin Pharmacol Toxicol 103:502–506. 41. Ghashghaei HT, Hilgetag CC, Barbas H (2007) Sequence of information processing for emotions based on the anatomic dialogue between prefrontal cortex and amygdala. Neuroimage 34:905–923. 42. Goldapple K, Segal Z, Garson C, Lau M, Bieling P, Kennedy S, Mayberg H (2004) Modulation of cortical-limbic pathways in major depression: treatment-specific effects of cognitive behavior therapy. Arch Gen Psychiatry 61:34–41. 43. Grace AA, Floresco SB, Goto Y, Lodge DJ (2007) Regulation of firing of dopaminergic neurons and control of goal-directed behaviors. Trends Neurosci 30:220–227. 44. Green AC, Baerentsen KB, StodkildeJorgensen H, Wallentin M, Roepstorff A, Vuust P (2008) Music in minor activates limbic structures: a relationship with dissonance? Neuroreport 19:711–715. 45. Greicius MD, Flores BH, Menon V, Glover GH, Solvason HB, Kenna H, Reiss AL, Schatzberg AF (2007) Resting-state functional connectivity in major depression: abnormally increased contributions from subgenual cingulate cortex and thalamus. Biol Psychiatry 62:429–437. 46. Gutman DA, Holtzheimer PE, Behrens TE, Johansen-Berg H, Mayberg HS (2009) A

tractography analysis of two deep brain stimulation white matter targets for depression. Biol Psychiatry 65:276–282. 47. Haber SN (2003) The primate basal ganglia: parallel and integrative networks. J Chem Neuroanat 26:317–330. 48. Haber SN, Kim KS, Mailly P, Calzavara R (2006) Reward-related cortical inputs define a large striatal region in primates that interface with associative cortical connections, providing a substrate for incentive-based learning. J Neurosci 26:8368–8376. 49. Hamani C, Mayberg, H., Snyder, B., Giacobbe, P., Kennedy, S., Lozano, A.M (2008) Subcallosal cingulate gyrus deep brain stimulation for depression: anatomical location of the active contacts in clinical responders and a suggested guidline for targeting. J Neurosurgery 111(6):1209–1215. 50. Hariri AR, Drabant EM, Munoz KE, Kolachana BS, Mattay VS, Egan MF, Weinberger DR (2005) A susceptibility gene for affective disorders and the response of the human amygdala. Arch Gen Psychiatry 62:146–152. 51. Harrison PJ (2002) The neuropathology of primary mood disorder. Brain 125: 1428–1449. 52. Harvey PO, Fossati P, Pochon JB, Levy R, Lebastard G, Lehericy S, Allilaire JF, Dubois B (2005) Cognitive control and brain resources in major depression: an fMRI study using the n-back task. Neuroimage 26:860–869. 53. Hasler G, Drevets WC, Manji HK, Charney DS (2004) Discovering endophenotypes for major depression. Neuropsychopharmacology 29:1765–1781. 54. Hoover WB, Vertes RP (2007) Anatomical analysis of afferent projections to the medial prefrontal cortex in the rat. Brain Struct Funct 212:149–179. 55. Horwitz B (2004) Relating fMRI and PET signals to neural activity by means of large-scale neural models. Neuroinformatics 2:251–266. 56. Hsu DT, Price JL (2007) Midline and intralaminar thalamic connections with the orbital and medial prefrontal networks in macaque monkeys. J Comp Neurol 504:89–111. 57. Hyman JM, Wyble BP, Goyal V, Rossi CA, Hasselmo ME (2003) Stimulation in hippocampal region CA1 in behaving rats yields long-term potentiation when delivered to the peak of theta and long-term depression when delivered to the trough. J Neurosci 23:11725–11731. 58. Hyman JM, Zilli EA, Paley AM, Hasselmo ME (2005) Medial prefrontal cortex cells

Targeted Modulation of Neural Circuits show dynamic modulation with the hippocampal theta rhythm dependent on behavior. Hippocampus 15:739–749. 59. Iremonger KJ, Anderson TR, Hu B, Kiss ZH (2006) Cellular mechanisms preventing sustained activation of cortex during subcortical high-frequency stimulation. J Neurophysiol 96:613–621. 60. Izhikevich EM, Desai NS, Walcott EC, Hoppensteadt FC (2003) Bursts as a unit of neural information: selective communication via resonance. Trends Neurosci 26:161–167. 61. Izhikevich EM, Gally JA, Edelman GM (2004) Spike-timing dynamics of neuronal groups. Cereb Cortex 14:933–944. 62. Jimenez F, Velasco F, Salin-Pascual R, Hernandez JA, Velasco M, Criales JL, Nicolini H (2005) A patient with a resistant major depression disorder treated with deep brain stimulation in the inferior thalamic peduncle. Neurosurgery 57:585–593. 63. Johansen-Berg H, Gutman DA, Behrens TE, Matthews PM, Rushworth MF, Katz E, Lozano AM, Mayberg HS (2008) Anatomical connectivity of the subgenual cingulate region targeted with deep brain stimulation for treatment-resistant depression. Cereb Cortex 18:1374–1383. 64. Johnstone T, van Reekum CM, Urry HL, Kalin NH, Davidson RJ (2007) Failure to regulate: counterproductive recruitment of top-down prefrontal-subcortical circuitry in major depression. J Neurosci 27:8877–8884. 65. Jürgens U (2009) The neural control of vocalization in mammals: a review. J. Voice 23:1–10. 66. Jurgens U, Ehrenreich L, De Lanerolle NC (2002) 2-Deoxyglucose uptake during vocalization in the squirrel monkey brain. Behav Brain Res 136:605–610. 67. Jurgens U, Muller-Preuss P (1977) Convergent projections of different limbic vocalization areas in the squirrel monkey. Exp Brain Res 29:75–83. 68. Kalin NH, Carnes M (1984) Biological correlates of attachment bond disruption in humans and nonhuman primates. Prog Neuropsychopharmacol Biol Psychiatry 8:459–469. 69. Keightley ML, Seminowicz DA, Bagby RM, Costa PT, Fossati P, Mayberg HS (2003) Personality influences limbic-cortical interactions during sad mood induction. Neuroimage 20:2031–2039. 70. Kellner CH, Knapp RG, Petrides G, Rummans TA, Husain MM, Rasmussen K, Mueller M, Bernstein HJ, O’Connor K,

275

Smith G, Biggs M, Bailine SH, Malur C, Yim E, McClintock S, Sampson S, Fink M (2006) Continuation electroconvulsive therapy vs pharmacotherapy for relapse prevention in major depression: a multisite study from the Consortium for Research in Electroconvulsive Therapy (CORE). Arch Gen Psychiatry 63:1337–1344. 71. Kendler KS, Thornton LM, Gardner CO (2001) Genetic risk, number of previous depressive episodes, and stressful life events in predicting onset of major depression. Am J Psychiatry 158:582–586. 72. Kennedy SH, Evans KR, Kruger S, Mayberg HS, Meyer JH, McCann S, Arifuzzman AI, Houle S, Vaccarino FJ (2001) Changes in regional brain glucose metabolism measured with positron emission tomography after paroxetine treatment of major depression. Am J Psychiatry 158:899–905. 73. Kennedy SH, Konarski JZ, Segal ZV, Lau MA, Bieling PJ, McIntyre RS, Mayberg HS (2007) Differences in brain glucose metabolism between responders to CBT and venlafaxine in a 16-week randomized controlled trial. Am J Psychiatry 164:778–788. 74. Knutson B, Greer SM (2008) Anticipatory affect: neural correlates and consequences for choice. Philos Trans R Soc Lond B Biol Sci 363:3771–3786. 75. Koechlin E, Ody C, Kouneiher F (2003) The architecture of cognitive control in the human prefrontal cortex. Science 302:1181–1185. 76. Koenigs M, Huey ED, Calamia M, Raymont V, Tranel D, Grafman J (2008) Distinct regions of prefrontal cortex mediate resistance and vulnerability to depression. J Neurosci 28:12341–12348. 77. Konarski JZ, Kennedy SH, Segal ZV, Lau MA, Bieling PJ, McIntyre RS, Mayberg HS (2009) Predictors of non-response to cognitive behavioral therapy or venlafaxine using glucose metabolism in major depressive disorder. J Psychiatry Neurosci 34:175–180. 78. Krishnan V, Nestler EJ (2008) The molecular neurobiology of depression. Nature 455:894–902. 79. Lam RW, Chan P, Wilkins-Ho M, Yatham LN (2008) Repetitive transcranial magnetic stimulation for treatment-resistant depression: a systematic review and metaanalysis. Can J Psychiatry 53:621–631. 80. Lekman M, Paddock S, McMahon FJ (2008) Pharmacogenetics of major depression: insights from level 1 of the Sequenced Treatment Alternatives to Relieve Depression (STAR*D) trial. Mol Diagn Ther 12:321–330.

276

Mayberg and Holtzheimer

81. Li S, Arbuthnott GW, Jutras MJ, Goldberg JA, Jaeger D (2007) Resonant antidromic cortical circuit activation as a consequence of high-frequency subthalamic deep-brain stimulation. J Neurophysiol 98:3525–3537. 82. Liotti M, Mayberg HS, McGinnis S, Brannan SL, Jerabek P (2002) Unmasking diseasespecific cerebral blood flow abnormalities: mood challenge in patients with remitted unipolar depression. Am J Psychiatry 159:1830–1840. 83. Little JT, Ketter TA, Kimbrell TA, Danielson A, Benson B, Willis MW, Post RM (1996) Venlafaxine or bupropion responders but not nonresponders show baseline prefrontal and paralimbic hypometabolism compared with controls. Psychopharmacol Bull 32:629–635. 84. Lorberbaum JP, Newman JD, Dubno JR, Horwitz AR, Nahas Z, Teneback CC, Bloomer CW, Bohning DE, Vincent D, Johnson MR, Emmanuel N, BrawmanMintzer O, Book SW, Lydiard RB, Ballenger JC, George MS (1999) Feasibility of using fMRI to study mothers responding to infant cries. Depress Anxiety 10:99–104. 85. Lorberbaum JP, Newman JD, Horwitz AR, Dubno JR, Lydiard RB, Hamner MB, Bohning DE, George MS (2002) A potential role for thalamocingulate circuitry in human maternal behavior. Biol Psychiatry 51:431–445. 86. Lozano AM, Mayberg HS, Giacobbe P, Hamani C, Craddock RC, Kennedy SH (2008) Subcallosal cingulate gyrus deep brain stimulation for treatment-resistant depression. Biol Psychiatry 64:461–467. 87. Lyons DM, Wang OJ, Lindley SE, Levine S, Kalin NH, Schatzberg AF (1999) Separation induced changes in squirrel monkey hypothalamic-pituitary-adrenal physiology resemble aspects of hypercortisolism in humans. Psychoneuroendocrinology 24:131–142. 88. MacLean PD (1990) The triune brain in evolution: role in paleocerebral functions. Plenum, New York. 89. MacLean PD, Newman JD (1988) Role of midline frontolimbic cortex in production of the isolation call of squirrel monkeys. Brain Res 450:111–123. 90. Malone DA Jr, Dougherty DD, Rezai AR, Carpenter LL, Friehs GM, Eskandar EN, Rauch SL, Rasmussen SA, Machado AG, Kubu CS, Tyrka AR, Price LH, Stypulkowski PH, Giftakis JE, Rise MT, Malloy PF, Salloway SP, Greenberg BD (2009) Deep brain stimulation of the ventral capsule/ventral striatum for treatment-resistant depression. Biol Psychiatry 65:267–275.

91. Margulies DS, Kelly AM, Uddin LQ, Biswal BB, Castellanos FX, Milham MP (2007) Mapping the functional connectivity of anterior cingulate cortex. Neuroimage 37: 579–588. 92. Martin SD, Martin E, Rai SS, Richardson MA, Royall R (2001) Brain blood flow changes in depressed patients treated with interpersonal psychotherapy or venlafaxine hydrochloride: preliminary findings. Arch Gen Psychiatry 58:641–648. 93. Mathew SJ, Manji HK, Charney DS (2008) Novel drugs and therapeutic targets for severe mood disorders. Neuropsycho pharmacology 33:2080–2092. 94. Mayberg HS (1994) Frontal lobe dysfunction in secondary depression. J Neuro psychiatry Clin Neurosci 6:428–442. 95. Mayberg HS (1997) Limbic-cortical dysregulation: a proposed model of depression. J Neuropsychiatry Clin Neurosci 9:471–481. 96. Mayberg HS (2009) Targeted electrode – based modulation of neural circuits for depression. Journal Clin Invest 119: 717–725. 97. Mayberg HS, Brannan SK, Mahurin RK, Jerabek PA, Brickman JS, Tekell JL, Silva JA, McGinnis S, Glass TG, Martin CC, Fox PT (1997) Cingulate function in depression: a potential predictor of treatment response. Neuroreport 8:1057–1061. 98. Mayberg HS, Brannan SK, Tekell JL, Silva JA, Mahurin RK, McGinnis S, Jerabek PA (2000) Regional metabolic effects of fluoxetine in major depression: serial changes and relationship to clinical response. Biol Psychiatry 48:830–843. 99. Mayberg HS, Liotti M, Brannan SK, McGinnis S, Mahurin RK, Jerabek PA, Silva JA, Tekell JL, Martin CC, Lancaster JL, Fox PT (1999) Reciprocal limbic-cortical function and negative mood: converging PET findings in depression and normal sadness. Am J Psychiatry 156:675–682. 100. Mayberg HS, Lozano AM, Voon V, McNeely HE, Seminowicz D, Hamani C, Schwalb JM, Kennedy SH (2005) Deep brain stimulation for treatment-resistant depression. Neuron 45:651–660. 101. Mayberg HS, Silva JA, Brannan SK, Tekell JL, Mahurin RK, McGinnis S, Jerabek PA (2002) The functional neuroanatomy of the placebo effect. Am J Psychiatry 159:728–737. 102. McCracken CB, Grace AA (2007) Highfrequency deep brain stimulation of the nucleus accumbens region suppresses neuronal activity and selectively modulates

Targeted Modulation of Neural Circuits afferent drive in rat orbitofrontal cortex in vivo. J Neurosci 27:12601–12610. 103. McEwen BS (1998) Protective and damaging effects of stress mediators. N Engl J Med 338:171–179. 104. McEwen BS, Magarinos AM (1997) Stress effects on morphology and function of the hippocampus. Ann NY Acad Sci 821: 271–284. 105. McIntyre CC, Butson CR, Maks CB, Noecker AM (2006) Optimizing deep brain stimulation parameter selection with detailed models of the electrode-tissue interface. Conf Proc IEEE Eng Med Biol Soc 1:893–895. 106. McIntyre CC, Savasta M, Walter BL, Vitek JL (2004) How does deep brain stimulation work? Present understanding and future questions. J Clin Neurophysiol 21:40–50. 107. Muller-Preuss P, Jurgens U (1976) Projections from the ‘cingular’ vocalization area in the squirrel monkey. Brain Res 103:29–43. 108. Myers KM, Davis M (2007) Mechanisms of fear extinction. Mol Psychiatry 12:120–150. 109. Nemeroff CB, Heim CM, Thase ME, Klein DN, Rush AJ, Schatzberg AF, Ninan PT, McCullough JP Jr, Weiss PM, Dunner DL, Rothbaum BO, Kornstein S, Keitner G, Keller MB (2003) Differential responses to psychotherapy versus pharmacotherapy in patients with chronic forms of major depression and childhood trauma. Proc Natl Acad Sci USA 100:14293–14296. 110. Nierenberg AA, Alpert JE, Gardner-Schuster EE, Seay S, Mischoulon D (2008) Vagus nerve stimulation: 2-year outcomes for bipolar versus unipolar treatment-resistant depression. Biol Psychiatry 64:455–460. 111. Nobler MS, Oquendo MA, Kegeles LS, Malone KM, Campbell CC, Sackeim HA, Mann JJ (2001) Decreased regional brain metabolism after ect. Am J Psychiatry 158:305–308. 112. Northoff G, Heinzel A, de Greck M, Bermpohl F, Dobrowolny H, Panksepp J (2006) Self-referential processing in our brain–a meta-analysis of imaging studies on the self. Neuroimage 31:440–457. 113. Ochsner KN, Gross JJ (2005) The cognitive control of emotion. Trends Cogn Sci 9:242–249. 114. Ochsner KN, Ray RD, Cooper JC, Robertson ER, Chopra S, Gabrieli JD, Gross JJ (2004) For better or for worse: neural systems supporting the cognitive down- and up-regulation of negative emotion. Neuroimage 23:483–499.

277

115. Ongur D, An X, Price JL (1998) Prefrontal cortical projections to the hypothalamus in macaque monkeys. J Comp Neurol 401: 480–505. 116. Ongur D, Drevets WC, Price JL (1998) Glial reduction in the subgenual prefrontal cortex in mood disorders. Proc Natl Acad Sci USA 95:13290–13295. 117. Ongur D, Price JL (2000) The organization of networks within the orbital and medial prefrontal cortex of rats, monkeys and humans. Cereb Cortex 10:206–219. 118. Osuch EA, Ketter TA, Kimbrell TA, George MS, Benson BE, Willis MW, Herscovitch P, Post RM (2000) Regional cerebral metabolism associated with anxiety symptoms in affective disorder patients. Biol Psychiatry 48:1020–1023. 119. Palomero-Gallagher N, Vogt BA, Schleicher A, Mayberg HS, Zilles K (2008) Receptor architecture of human cingulate cortex: evaluation of the four-region neurobiological model. Hum Brain Mapp 30(8):2336–2355. 120. Papez JW (1937) A proposed mechanism of emotion. Arch Neurol Psychiatry 38: 725–733. 121. Pardo JV, Sheikh SA, Schwindt GC, Lee JT, Kuskowski MA, Surerus C, Lewis SM, Abuzzahab FS, Adson DE, Rittberg BR (2008) Chronic vagus nerve stimulation for treatment-resistant depression decreases resting ventromedial prefrontal glucose metabolism. Neuroimage 42:879–889. 122. Peltier SJ, Polk TA, Noll DC (2003) Detecting low-frequency functional connectivity in fMRI using a self-organizing map (SOM) algorithm. Hum Brain Mapp 20:220–226. 123. Petrides M, Pandya DN (2007) Efferent association pathways from the rostral prefrontal cortex in the macaque monkey. J Neurosci 27:11573–11586. 124. Pezawas L, Meyer-Lindenberg A, Drabant EM, Verchinski BA, Munoz KE, Kolachana BS, Egan MF, Mattay VS, Hariri AR, Weinberger DR (2005) 5-HTTLPR polymorphism impacts human cingulateamygdala interactions: a genetic susceptibility mechanism for depression. Nat Neurosci 8:828–834. 125. Phan KL, Wager T, Taylor SF, Liberzon I (2002) Functional neuroanatomy of emotion: a meta-analysis of emotion activation studies in PET and fMRI. Neuroimage 16:331–348. 126. Phillips ML, Drevets WC, Rauch SL, Lane R (2003) Neurobiology of emotion perception

278

Mayberg and Holtzheimer

II: implications for major psychiatric disorders. Biol Psychiatry 54:515–528. 127. Pittenger C, Duman RS (2008) Stress, depression, and neuroplasticity: a convergence of mechanisms. Neuropsychopharmacology 33: 88–109. 128. Pizzagalli D, Pascual-Marqui RD, Nitschke JB, Oakes TR, Larson CL, Abercrombie HC, Schaefer SM, Koger JV, Benca RM, Davidson RJ (2001) Anterior cingulate activity as a predictor of degree of treatment response in major depression: evidence from brain electrical tomography analysis. Am J Psychiatry 158:405–415. 129. Poston KL, Eidelberg D (2008) Network biomarkers for the diagnosis and treatment of movement disorders. Neurobiol Dis. 130. Radley JJ, Rocher AB, Rodriguez A, Ehlenberger DB, Dammann M, McEwen BS, Morrison JH, Wearne SL, Hof PR (2008) Repeated stress alters dendritic spine morphology in the rat medial prefrontal cortex. J Comp Neurol 507:1141–1150. 131. Rajkowska G, Miguel-Hidalgo JJ (2007) Gliogenesis and glial pathology in depression. CNS Neurol Disord Drug Targets 6:219–233. 132. Robinson RG, Kubos KL, Starr LB, Rao K, Price TR (1984) Mood disorders in stroke patients. Importance of location of lesion. Brain 107:81–93. 133. Rush AJ, Kovacs M, Beck AT, Weissenburger J, Hollon SD (1981) Differential effects of cognitive therapy and pharmacotherapy on depressive symptoms. J Affect Disord 3:221–229. 134. Rush AJ, Trivedi MH, Wisniewski SR, Nierenberg AA, Stewart JW, Warden D, Niederehe G, Thase ME, Lavori PW, Lebowitz BD, McGrath PJ, Rosenbaum JF, Sackeim HA, Kupfer DJ, Luther J, Fava M (2006) Acute and longer-term outcomes in depressed outpatients requiring one or several treatment steps: a STAR*D report. Am J Psychiatry 163:1905–1917. 135. Sakas DE, Panourias IG, Simpson BA (2007) An introduction to neural networks surgery, a field of neuromodulation which is based on advances in neural networks science and digitised brain imaging. Acta Neurochir Suppl 97:3–13. 136. Salvadore G, Cornwell BR, Colon-Rosario V, Coppola R, Grillon C, Zarate CA Jr, Manji HK (2009) Increased anterior cingulate cortical activity in response to fearful faces: a neurophysiological biomarker that predicts rapid antidepressant response to ketamine. Biol Psychiatry 65:289–295.

137. Santini E, Quirk GJ, Porter JT (2008) Fear conditioning and extinction differentially modify the intrinsic excitability of infralimbic neurons. J Neurosci 28:4028–4036. 138. Schlaepfer TE, Cohen MX, Frick C, Kosel M, Brodesser D, Axmacher N, Joe AY, Kreft M, Lenartz D, Sturm V (2008) Deep brain stimulation to reward circuitry alleviates anhedonia in refractory major depression. Neuropsychopharmacology 33:368–377. 139. Seminowicz DA, Mayberg HS, McIntosh AR, Goldapple K, Kennedy S, Segal Z, RafiTari S (2004) Limbic-frontal circuitry in major depression: a path modeling metanalysis. Neuroimage 22:409–418. 140. Sheline YI (2003) Neuroimaging studies of mood disorder effects on the brain. Biol Psychiatry 54:338–352. 141. Sheline YI, Barch DM, Donnelly JM, Ollinger JM, Snyder AZ, Mintun MA (2001) Increased amygdala response to masked emotional faces in depressed subjects resolves with antidepressant treatment: an fMRI study. Biol Psychiatry 50:651–658. 142. Siegle GJ, Carter CS, Thase ME (2006) Use of FMRI to predict recovery from unipolar depression with cognitive behavior therapy. Am J Psychiatry 163:735–738. 143. Siegle GJ, Steinhauer SR, Thase ME, Stenger VA, Carter CS (2002) Can’t shake that feeling: event-related fMRI assessment of sustained amygdala activity in response to emotional information in depressed individuals. Biol Psychiatry 51:693–707. 144. Sirota A, Montgomery S, Fujisawa S, Isomura Y, Zugaro M, Buzsaki G (2008) Entrainment of neocortical neurons and gamma oscillations by the hippocampal theta rhythm. Neuron 60:683–697. 145. Swain JE, Lorberbaum JP, Kose S, Strathearn L (2007) Brain basis of early parent-infant interactions: psychology, physiology, and in vivo functional neuroimaging studies. J Child Psychol Psychiatry 48:262–287. 146. Talbot PS, Cooper SJ (2006) Anterior cingulate and subgenual prefrontal blood flow changes following tryptophan depletion in healthy males. Neuropsychopharmacology 31:1757–1767. 147. Tremblay LK, Naranjo CA, Graham SJ, Herrmann N, Mayberg HS, Hevenor S, Busto UE (2005) Functional neuroanatomical substrates of altered reward processing in major depressive disorder revealed by a dopaminergic probe. Arch Gen Psychiatry 62:1228–1236. 148. Trost M, Su S, Su P, Yen RF, Tseng HM, Barnes A, Ma Y, Eidelberg D (2006) Network

Targeted Modulation of Neural Circuits modulation by the subthalamic nucleus in the treatment of Parkinson’s disease. Neuroimage 31:301–307. 149. Uddin LQ, Kelly AM, Biswal BB, Xavier Castellanos F, Milham MP (2009) Functional connectivity of default mode network components: correlation, anticorrelation, and causality. Hum Brain Mapp 30:625–637. 150. Vertes RP (2004) Differential projections of the infralimbic and prelimbic cortex in the rat. Synapse 51:32–58. 151. Vertes RP, Hoover WB, Szigeti-Buck K, Leranth C (2007) Nucleus reuniens of the midline thalamus: link between the medial prefrontal cortex and the hippocampus. Brain Res Bull 71:601–609. 152. Videbech P (2000) PET measurements of brain glucose metabolism and blood flow in major depressive disorder: a critical review. Acta Psychiatr Scand 101:11–20. 153. Videbech P, Ravnkilde B, Pedersen TH, Hartvig H, Egander A, Clemmensen K, Rasmussen NA, Andersen F, Gjedde A, Rosenberg R (2002) The Danish PET/ depression project: clinical symptoms and cerebral blood flow. A regions-of-interest analysis. Acta Psychiatr Scand 106:35–44. 154. Vogt BA, Nimchinsky EA, Vogt LJ, Hof PR (1995) Human cingulate cortex: surface features, flat maps, and cytoarchitecture. J Comp Neurol 359:490–506. 155. Wang L, LaBar KS, Smoski M, Rosenthal MZ, Dolcos F, Lynch TR, Krishnan RR, McCarthy G (2008) Prefrontal mechanisms for executive control over emotional distraction are altered in major depression. Psychiatry Res 163:143–155. 156. Warden D, Rush AJ, Trivedi MH, Fava M, Wisniewski SR (2007) The STAR*D Project results: a comprehensive review of findings. Curr Psychiatry Rep 9:449–459.

279

157. Watkins E, Teasdale JD (2004) Adaptive and maladaptive self-focus in depression. J Affect Disord 82:1–8. 158. Weerts EM, Miczek KA (1996) Primate vocalizations during social separation and aggression: effects of alcohol and benzodiazepines. Psychopharmacology (Berl) 127:255–264. 159. Wu J, Buchsbaum MS, Gillin JC, Tang C, Cadwell S, Wiegand M, Najafi A, Klein E, Hazen K, Bunney WE Jr, Fallon JH, Keator D (1999) Prediction of antidepressant effects of sleep deprivation by metabolic rates in the ventral anterior cingulate and medial prefrontal cortex. Am J Psychiatry 156:1149–1158. 160. Yoshimura S, Ueda K, Suzuki S, Onoda K, Okamoto Y, Yamawaki S (2009) Selfreferential processing of negative stimuli within the ventral anterior cingulate gyrus and right amygdala. Brain Cogn 69:218–225. 161. Zald DH, Mattson DL, Pardo JV (2002) Brain activity in ventromedial prefrontal cortex correlates with individual differences in negative affect. Proc Natl Acad Sci USA 99:2450–2454. 162. Zarate CA Jr, Singh JB, Carlson PJ, Brutsche NE, Ameli R, Luckenbaugh DA, Charney DS, Manji HK (2006) A randomized trial of an N-methyl-D-aspartate antagonist in treatment-resistant major depression. Arch Gen Psychiatry 63:856–864. 163. Zhang F, Aravanis AM, Adamantidis A, de Lecea L, Deisseroth K (2007) Circuitbreakers: optical technologies for probing neural signals and systems. Nat Rev Neurosci 8:577–581. 164. Zink CF, Pagnoni G, Martin ME, Dhamala M, Berns GS (2003) Human striatal response to salient nonrewarding stimuli. J Neurosci 23:8092–8097.

Index A Acetylcholine/cholinergic............................................... 243 Action potential antidromic.................................................................. 22 orthodromic................................................................ 20 Adrian, E............................................................................ 2 Affective state......................................................... 260, 268 Allostatic load......................................................... 258, 263 Alpha rhythm current generator.................................................. 34, 35 pacemaker............................................................. 35, 36 Alveus......................................................................... 17–20 AMPA receptors/AMPA-kainate receptor..................................................38, 250, 251 Amplifier........................................................12, 13, 15, 45, 49, 81–84, 87, 91, 98, 99, 194 Amygdala........................................................204, 260–262 Analysis spatial....................................................................... 185 time domain..................................................... 183–184 time frequency...........................................184–186, 192 time series................................................................. 199 Anergia........................................................................... 263 Anhedonia...................................................................... 262 Antidepressant treatment........................258–260, 264–266 Anxiety................................................................... 262, 263 Apathy............................................................................ 263 Attention........................................................28, 34, 36, 38, 128, 150, 180, 197, 198, 241, 243, 265, 270 Autonomic...................................................................... 266 Axon.................................... 3, 4, 6, 7, 10, 20, 22, 23, 42, 44, 46, 48, 50, 53, 58, 60, 62–64, 68, 70, 170–173, 243

B Basal ganglia......................................................60, 261, 262 Beck depression inventory.............................................. 269 Berger, H............................................................................ 2 Bicuculline.............................................................. 248, 249 Binomial spike statistics................................................. 198

Bipolar disorder.............................................................. 259 Bootstrap technique....................................................... 197 Brain lesions of........................................................... 195, 261 slices..............................................................70, 80, 246 Brainstem........................................... 60, 70, 148, 204, 205, 247, 260, 265, 266, 268

C Canonical discriminant analysis (CDA)................. 219, 220 Carbachol........................................ 245, 250, 251, 253, 254 Cartesian coordinates....................................................... 12 Cerebellum...............................14, 15, 21, 60, 195, 198, 246 Closed loop current flow in.......................................................... 5, 9 facilitation......................................................... 231, 236 paradigm........................................................... 215–237 Cluster cutting tools Klustakwik........................................................ 110, 111 MClust............................................................. 110, 111 Offline Sorter (Plexon)......................108–110, 139, 218 Cognitive processes................... 28, 170, 174, 178, 186, 187 Cognitive behavioral therapy (CBT)...............258–260, 267 Coherence partial.................................................193, 194, 200, 201 partial directed...........................................194, 201, 202 spectral.......................................................192, 201, 206 Consonance.................................................................... 180 Contingent negative variation (CNV)............................ 178 Correlation coefficient...................................................... 38 Cortex anterior cingulate...................................................... 261 entorhinal.....................................................36, 98, 128, 148, 150, 151, 203–206, 243, 244, 250, 252 frontal/lobe........................ 246, 261, 262, 265, 266, 269 inferotemporal................................................ 28, 34–36 orbital................................................262, 265, 268, 269 parahippocampal...................................................... 265 parietal.............................................................. 176, 197 prefrontal................................... 180, 195, 261, 262, 265 retrosplenial...............................................148, 200, 205

Robert P. Vertes and Robert W. Stackman Jr. (eds.), Electrophysiological Recording Techniques, Neuromethods, vol. 54, DOI 10.1007/978-1-60327-202-5, © Springer Science+Business Media, LLC 2011

281

Electrophysiological Recording Techniques 282 Index

Cortex (Continued) sensorimotor..................................................... 179, 186 supragranular layers of................................................ 36 temporal..................................................28, 34–36, 197 Covariance matrix...................................184, 219–221, 227 Current capacitive.................................................................... 80 field....................................................................4, 5, 174 intracellular................................................................... 4 loop........................................................4, 7, 8, 171–173 sink.................................................. 4–7, 22, 23, 28, 173 source........................................... 2, 4–9, 28, 30, 31, 173 transmembrane.............................. 27, 29, 31, 32, 36–38 Current source density (CSD) analysis............. 1–24, 27–39

D Deep brain stimulation (DBS).................................. 257, 259, 260, 268–270 Delayed-nonmatch-to-sample task (DNMS)..................................................... 217–227 Dendrite apical..........................7, 8, 16–20, 22, 23, 171, 173, 174 basal................................................... 7, 8, 18–20, 22, 23 Dendritic current loop.......................................................... 8, 171 trees.............................................................................. 3 Dentate gyrus (DG)...........................................16, 18, 128, 151, 203, 242, 247, 248 Depression...............................................257–263, 265–272 Desynchronization......................................................... 178 Diencephalon/diencephalic............................................ 204 Diffusion tensor imaging................................................ 270 Dipole dipolar source-sink................................................... 173 potentials...................................................................... 9 Directed transfer function (DTF).................................. 194 Directional behavior.....................................................158, 161, 162 firing (see Head direction cell) Discriminant functions (DFs).........................219–221, 227 DNQX................................................................... 250, 251 Domain, spatial................................................................ 10

E Eigenvalue.......................................................184, 220, 221 Eigenvector matrix........................................................220, 221, 223 orthogonal........................................................ 219, 221 Electrocorticogram (ECoG)........................................... 172 Electroencephalogram/electroencephalography (EEG).................... 2, 8, 9, 20, 31, 33, 65, 78, 83–85, 96, 104, 175, 180, 182, 183, 194, 195, 202, 206, 262 Electromagnetic activity......................................... 170–174 Emotion/emotional reactivity......................................... 263

Ensemble activity averaging......................................28, 176–179, 181–183 codes................................................................. 215–237 decoding................................................................... 117 firing rates..........................................136, 219, 220, 224 recording.....................................................78, 104–107, 109, 122, 123, 130, 135, 153, 160–161, 218 Epileptiform activity......................................................... 11 Epilepsy Monitoring Unit (EMU)................................... 95 Euthymic........................................................................ 264 Event-related potential (ERP)............................... 169–187 Evoked potential average........................................................................ 12 sensory.............................................................. 176, 177 visual......................................................................... 177

F Field potential local (LFP)........................................... 2, 27, 79, 82–84, 87, 97, 135, 172, 192 population spikes.................................13, 16–20, 22, 27 theory..................................................................... 2–10 Filter analog......................................................................... 85 digital.......................................................................... 84 Kalman..................................................................... 203 Wiener...................................................................... 183 Fluoxetine............................................................... 265, 267 Fornix.............................................. 243, 245, 247, 248, 252 Fourier analysis, transform............................................. 184 Frequency domain analysis..................................... 183–185 Functional magnetic resonance imaging (fMRI)......... 78, 262

G Gamma activity modulation............................................................... 198 rhythm.................................................................... 2, 20 Gamma amino butyric acid (GABA)/ GABAA receptors.................................................. 8 Gaussian noise........................................................................... 31 potentials.................................................................. 182 Gene.......................................... 63, 104, 129, 258, 263, 267 Gersh causality....................................................... 200, 202 Glia abnormalities of...................................................... 266 Glutamate receptors......................................................... 38 Granger causality............................. 186, 194, 199, 201, 202 Grid cell.................................... 98, 128, 129, 148, 150, 151

H Hamilton depression rating scale.................................... 269 Head direction cell firing characteristics.................................................. 147 relation to spatial behavior................................ 149–163

Electrophysiological Recording Techniques 283 Index

Hippocampus CA1 region of............................................104, 130, 162 CA3 region of........................................................... 250 parasubiculum of...................................................... 204 postsubiculum of...............................................138, 146, 147, 149, 153, 154, 158, 160–162 stratum lacunosum-moleculare of............................. 204 stratum radiatum of............................................ 17, 204 subiculum of......................................128, 146, 151, 204 Histogram, peri-event......................................67, 113, 115, 136, 218–222, 224, 227 Hodgkin-Huxley simulations......................................... 198 Homeostatic processes.................................................... 267 Hypothalamus...........................................60, 204, 266, 268

I Interdependency analysis........................................ 185–186 Interictal activity............................................................. 248 Interpersonal psychotherapy........................................... 263 Isopotential.................................................. 7, 8, 16, 19, 174 Isotropic medium............................................................. 21

L Laguerre–Volterra network............................................. 233 Libido............................................................................. 266 Limbic system, cortex............................................. 128, 151 Linear array electrode................................................. 29, 34 Linear, nonlinear autoregressive models (LNLAR)........................................................... 194 Long-term potentiation (LTP)....................................... 242

M Magnetoencephalogram/magnetoencephalography (MEG)............................................................... 262 Major depressive disorder (MDD)................................. 261 Mammillary body................................................... 203, 244 MATLAB...................................................................... 224 Maxwell’s equations.......................................................... 28 MDA. See Multiple-discriminant analysis (MDA) Memory episodic.............................. 109, 113, 135, 145, 206, 242 nonspatial................................................................. 163 retrieval..................................................................... 103 short term................................................................. 196 spatial.........................128, 135, 154, 158, 160, 203, 242 working.................28, 156, 157, 160, 180, 185, 197, 243 Microdrive..................................................12, 13, 104–110, 122, 123, 130–132, 134–135, 137 Microelectrodes.................................. 11–13, 49, 56, 80, 83, 109, 129–135, 137, 146, 171, 172, 175, 192–197 Micropipette, glass..................13, 41, 42, 45, 51, 54, 65, 250 Mismatch negativity (MMN)........................................ 180 Monkey, macaque................................ 28, 34, 176, 186, 197 Mood.......................................................257, 259–263, 266

Mouse C57BL/6J.................................. 132, 135, 140, 162, 163 genetically modified.......................................... 129, 164 Multi-input/multi-output model (MIMO)............ 233, 234 Multiple-discriminant analysis (MDA)...........116–118, 120 Multitaper spectral method............................................ 197 Multivariate linear autoregressive model (MVAR).................................................. 193 Muscarinic...............................................204, 243, 244, 251

N Neurodegenerative disorders.......................................... 261 Neuromodulation........................................................... 257 Neuron labeling Golgi impregnation.................................................... 42 iontophoretic tracer application.................................. 42 juxtacellular........................................................... 41–71 Neuropsychiatric..................................................... 257–272 Neurotransmitter............................... 38, 171, 172, 250, 258 Neurotrophic.................................................................. 265 NMDA receptors........................................38, 39, 104, 244 Noise cephalic..................................................................... 176 cranial....................................................................... 176 electronic.................................................................. 176 environmental........................................................... 176 thermal..................................................................... 176 Nonlinear methods..........................................201, 202, 219 Nosepoke.................................................217, 218, 220, 223 Nosology........................................................................ 272 Nucleus accumbens........................................................ 262, 267 diagonal band........................................................... 204 gracilis....................................................................... 195 pedunculopontine (PPN).................................. 195, 196 supramammillary (SUM)................................. 204, 244 ventral tegmental (VTG).................................. 204, 207

O Obsessional disorders..................................................... 262 Oddball detection task...................................................... 34 Oscillations oscillatory.................................................2, 28, 33, 170, 179, 182–184, 186, 195, 198, 199, 204, 205, 207 slow........................................................................... 199

P P300............................................................................... 178 Papez’s circuit......................................................... 191–208 Parametric spectral estimation........................................ 185 Parkinson’s disease.................................................. 195, 271 Patch clamp recordings................................................... 250 Pathophysiology............................................................. 271 Phase precession............................................................. 242

Electrophysiological Recording Techniques 284 Index

Phase realigned averaging technique (PRAT)........................................................... 28–30 Place cell......................................... 105, 109, 118, 128, 129, 137, 144, 146, 150, 151, 153, 154, 156, 161–163, 242 Plasticity...................................... 11, 23, 104, 123, 242, 265 Poisson’s equation............................................................... 6 Pontamine sky blue......................................................... 250 Population vectors...................................216, 219–221, 236 Position data, tools to acquire......................................... 137 Positron emission tomography (PET)............................ 262 Potentials postsynaptic....................7, 8, 17–19, 171, 172, 194, 254 reversal.......................................................................... 8 Power spectrum................................................................ 34 Principal component analysis (PCA)...................... 110, 117 Probe multishank.................................................................. 11 silicon............................................ 10–14, 17, 20, 23, 83 Prosody........................................................................... 269 Psychology/psychological............................................... 263 Psychotherapy......................................................... 263–266 Pyramidal cells of cortex.................................................................... 173 of hippocampus................................................ 242–244

R Readiness potential (RP)................................................ 179 Reentrant neural circuits................................................. 207 Resonant signaling......................................................... 205 Ripples, 200 Hz................................................................ 20 Roessler system............................................................... 201

S Second order differencing........................................... 14–15 Second spatial derivative............................27, 29–31, 35, 37 Seizure. See Epileptiform activity Septo-hippocampal complex.....................................................245, 247, 248 isolation.................................................................... 248 preparation........................................................ 241–254 recordings................................................................. 248 slices......................................................................... 243 Septum lateral................................................................ 246, 252 medial (MS).............................................203, 242, 244, 247–249, 253, 254 Serotonin reuptake inhibitors (SSRIs)............................ 265 Shaffer collaterals............................................................. 23 Sharp waves................................................................ 20, 84 Single photon emission computed tomography (SPECT)............................................................ 262 Sleep deprivation............................................................ 264 Smoothing, two step differencing............................... 15, 16

Spatial behavior......................... 97, 142, 146, 149–160, 163 See also Memory, spatial Spatial scales....................................................175, 185, 195 Spike field coherence.............................................. 191–208 Spikes complex.................................................................... 112 population............................................13, 16–20, 22, 27 trains.....................99, 192, 194, 195, 197, 198, 233–235 Spike-triggered averages..................................194, 197, 206 Stereotrode.....................................................106–112, 122, 123, 129–131, 133 Striatum.................................. 156, 160, 246, 260, 262, 268 Subcallosal cingulate cortex.............................257, 264–268 Superior colliculus.......................................................... 204 Synchronization/synchronously active neurons.......................................2, 65–67, 70, 178, 192, 194, 197, 198, 204, 207, 208

T Tetrode............................................106–112, 122, 129–133 Thalamus (or thalamic), anterior............................128, 129, 147, 148, 151, 153, 158, 160, 161, 164, 193, 200, 203, 205, 207, 244 Theta rhythm atropine-resistant...................................................... 250 atropine-sensitive.............................................. 250–252 generator................................................................... 205 oscillations......................... 197, 199, 207, 241–243, 251 rhythmic circuit........................................................ 207 semi-autonomous resonator...................................... 207 type I.........................................................243, 250–252 type II................................................243, 244, 250, 252 Tracer. See Neuron labeling Tractography.................................................................. 270 Treatment resistant depression (TRD)........................... 268 Tryptophan............................................................. 266, 267

U Unit recordings, multiunit activity............................ 34, 195

V Venlafaxine............................................................. 266, 267 Video tracking........................................................ 137–140 Volterra kernels................................................233, 234, 236 Volume conduction................................ 6, 10, 16, 18, 20, 22

W Wavelet time-frequency analysis.................................... 192 Wavelet transforms......................................................... 184 Wireless telemetry analog......................................................................... 89 digital.................................................................. 76–100 power sources...................................................92, 93, 99