The Functional Role of Neural Oscillatory Activity in Perception and Action

Material Information

The Functional Role of Neural Oscillatory Activity in Perception and Action
Zhang, Yan
Place of Publication:
[Gainesville, Fla.]
University of Florida
Publication Date:
Physical Description:
1 online resource (106 p.)

Thesis/Dissertation Information

Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Biomedical Engineering
Committee Chair:
Ding, Mingzhou
Committee Members:
Mareci, Thomas H.
Ditto, William L.
Perlstein, William
Hermer-Vazquez, Linda
Graduation Date:


Subjects / Keywords:
Causality ( jstor )
Electrodes ( jstor )
Electroencephalography ( jstor )
Mental stimulation ( jstor )
Monkeys ( jstor )
Neural oscillations ( jstor )
Prefrontal cortex ( jstor )
Scalp ( jstor )
Signals ( jstor )
Somatosensory cortex ( jstor )
Biomedical Engineering -- Dissertations, Academic -- UF
causality, eeg, erp, lfp, n1, oscillations, perception
bibliography ( marcgt )
theses ( marcgt )
government publication (state, provincial, terriorial, dependent) ( marcgt )
born-digital ( sobekcm )
Electronic Thesis or Dissertation
Biomedical Engineering thesis, Ph.D.


Spontaneous ongoing neural oscillations are traditionally considered to reflect an idling or deactivated brain state. In contrast, recent slice studies have demonstrated that the moderate amount of membrane potential oscillations can bring the local neuron populations closer to the firing threshold, making them more sensitive to a weak sensory stimulus. Other reports showed that very high levels of spontaneous oscillations inhibit sensory evoked responses. Hence, to distinguish these competing views, we investigated the relationship between ongoing neural oscillations preceding stimulus presentation and subsequent cognitive processing and behavior. First, we studied the relationship between reaction times and prestimulus neural oscillations in a broad frequency band over multiple cortical regions. Second, we explored the relationship between middle-latency somatosensory evoked potentials and prestimulus mu oscillations. Third, we investigated the large-scale synchronization and directionality of prestimulus neural activity in sensory perception and sensorimotor processing by applying adaptive multivariate autoregressive coherence and Granger causality analysis. In conclusion, our main findings suggest that cortical neural oscillations, especially spontaneous ongoing neural oscillations, are crucial in subsequent cognitive perception and behavioral performance. Also, the prestimulus neural synchronization over large-scale cortical areas, play a key role in sensory processing and sensorimotor integration. For perception and action, these neural synchronizations help the top-down signals intrinsically generated from higher-order cortical regions (e.g., the prefrontal cortex) to interact with the bottom-up signals from lower-order cortical regions to facilitate the processing of forthcoming stimuli. These top-down attentional influences among multiple cortical regions are expected to play a functional role in selective processing of sensory input and motor output. Hence, this study suggests that spontaneous ongoing activity prior to stimulus onset may reflect an active not an idling brain state. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis (Ph.D.)--University of Florida, 2008.
Adviser: Ding, Mingzhou.
Electronic Access:
Statement of Responsibility:
by Yan Zhang.

Record Information

Source Institution:
Rights Management:
Copyright Zhang, Yan. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
LD1780 2008 ( lcc )


This item has the following downloads:

Full Text




2008 Yan Zhang 2


To my Dad, Mom and my husba nd, for their love and support 3


ACKNOWLEDGMENTS It has been a great journey for me to spe nd four years in J. Crayton Pruitt Family Department of Biomedical Engineering at Univer sity of Florida. I would like to thank my advisor, Dr. Mingzhou Ding for hi s mentoring, motivation and enc ouragement for the past five years. I thank my doctoral committee members, Dr. William Ditto, Dr. Linda Hermer-Vazquez, Dr. William M. Perlstein, and Dr. Thomas Mareci for their support and encouragement. I also would like to thank Dr. Steven Br essler in the Center for Comp lex Systems and Brain Sciences at Florida Atlantic University for his support and kindness. I thank my classmates, postdoctors, a nd research scientists working in the Neuroinfomatics labs for interest ing discussion and collaboration. I thank the staff at J. Crayton Pruitt Family Department of Biomedical Engineer ing for their generous assistance. I also thank all the participants in my experime nt for their honest participation. I thank my parents for their unconditional l ove and encouragement. I want to thank my younger brother for his support and understanding. Particularly, I would like to thank my husband Youhua Wei for his three-year encouragement, support and love. 4


TABLE OF CONTENTS page ACKNOWLEDGMENTS ...............................................................................................................4 LIST OF FIGURES .........................................................................................................................7 LIST OF ABBREVIATIONS ..........................................................................................................9 ABSTRACT ...................................................................................................................................11 CHAPTER 1 INTRODUCTION................................................................................................................. .13 1.1 Background .......................................................................................................................13 1.2 Objectives .........................................................................................................................15 2 GENERAL METHODS.........................................................................................................18 2.1 Adaptive Multivariate Autore gressive Spectral Analysis .................................................18 2.2 Granger Causality Analysis ..............................................................................................20 2.2.1 Pairwise Analysis ...................................................................................................20 2.2.2 Conditional Granger Causality Analysis ................................................................22 3 ONGOING OSCILLATIONS AND VISUOMOTOR PROCESSING SPEED....................24 3.1 Introduction .......................................................................................................................24 3.2 Materials and Methods.....................................................................................................26 3.2.1 Paradigm and Data Acquisition ..............................................................................26 3.2.2 Data Analysis ..........................................................................................................27 3.3 Results ...............................................................................................................................29 3.4 Discussion .........................................................................................................................31 4 CORTICAL SENSORIMOTOR RHYTHMS AND THEIR FUNCTIONAL ROLES.........39 4.1 Introduction .......................................................................................................................39 4.2 Methods and Materials .....................................................................................................40 4.2.1 LFP Experiment and Data Preprocessing ...............................................................40 4.2.2 EEG Recording and Paradigm ................................................................................42 4.3 Results ...............................................................................................................................43 4.3.1 LFP Results ............................................................................................................43 4.3.2 Scalp EEG Results ..................................................................................................44 4.4 Discussion .........................................................................................................................45 5 NEURAL CORRELATES OF SOMATOSENSORY PROCESSING OF A WEAK STIMULUS....................................................................................................................... .....53 5


5.1 Introduction .......................................................................................................................53 5.1.1 Somatosensory Evoked Potentials ..........................................................................53 5.1.2 Correlation of N1 and Sensory Attention ...............................................................54 5.1.3 Spontaneous Mu Rhythm .......................................................................................55 5.1.4 N1 and Ongoing Activity .......................................................................................56 5.1.5 N1 and Top-down Attentional Control ...................................................................58 5.2 Materials and Methods.....................................................................................................59 5.2.1 EEG Experiment and Design ..................................................................................59 5.2.2 Data Preprocessing .................................................................................................61 5.2.3 Correlation of Prestimulus Mu Power and N1 .......................................................62 5.2.4 Granger Causality Spectral Analysis ......................................................................62 5.3 Results ...............................................................................................................................63 5.3.1 Behavioral Results ..................................................................................................63 5.3.2 Evoked Potentials ...................................................................................................64 5.3.3 Correlation of Prestimulus Mu Power and N1 Amplitude .....................................65 5.3.4 Granger Causality Influences between PFC and S1 ...............................................66 5.4 Discussion .........................................................................................................................68 5.4.1 Spontaneous Membrane Potent ials and Sensory Processing ..................................69 5.4.2 Somatosensory Evoked Poten tials and Sensory Processing ...................................71 5.4.3 Top-down Attentional Cont rol on Sensory Processing ..........................................72 6 CONCLUSIONS AND FURTURE RESEARCH..................................................................87 6.1 Conclusion........................................................................................................................87 6.2 Future Research ................................................................................................................88 APPENDIX A MULTITAPER SPECTRAL ANALYSIS.............................................................................90 B SECOND ORDER BLI ND IDENTIFICATION....................................................................91 LIST OF REFERENCES ...............................................................................................................93 BIOGRAPHAICAL SKETCH ....................................................................................................106 6


LIST OF FIGURES Figure page 3-1 Approximate electrode placements as mark ed visually during surgery in the three macaque monkeys, shown on schematic drawings of the lateral cortical hemispheres .....34 3-2 Prestimulus power spectra (8 Hz) and co rrelation plots for tw o recording sites in monkey TI ..........................................................................................................................35 3-3 Spearmans rank correlation coefficients between prestimulus group spectral power and group mean RT for monkey TI. The horizontal axis indicates frequency in the range of 8 to 40 Hz. The vertical axis disp lays the correlations from -1 to 1. Gray shaded areas signify correlati ons that are significant at p < 0.05. .....................................36 3-4 Spearmans rank correlation coefficients between prestimulus group spectral power and group mean RT for monkey LU ..................................................................................37 3-5 Spearmans rank correlation coefficients between prestimulus group spectral power and group mean RT for monkey GE ..................................................................................38 4-1 A schematic of approximate electrode placement in monkey GE and LU ........................47 4-2 Time-frequency plots of averaged power spectra computed over all sites in GE and LU ......................................................................................................................................48 4-3 Averaged power spectra over three selected sites, averaged coherence spectra, and averaged Granger causality spectra computed over all pairwise combinations in the recurrence windows for LU and GE. .................................................................................49 4-4 Schematic Granger causality graphs dur ing the recurrence window in GE and LU.........50 4-5 Granger causality spectra from left S1 cort ical areas to left motor cortical areas for the prestimulus holding period and the baseline condition ................................................51 4-6 Granger causality Graphs for the prestimulus time holding period and the baseline condition ............................................................................................................................52 5-1 Characteristics of spontaneous mu oscillations .................................................................76 5-2 Behavioral performance in a somatosens ory perception task. A) The histogram bars show the probability of detecting a near -threshold electrical stimulus for each individual. The grand average probability over all thirteen subj ects is 0.51. B) The probability of detecting a stimulus was es timated in each block and averaged across all thirteen subjects. Error bars are standard error of mean. ..............................................77 5-3 Behavioral performances in a somatosensory perception task ..........................................78 7


5-4 Averaged somatosensory evoked potentials for ten power groups ....................................79 5-5 Regression data fit fo r all thirteen subjects ........................................................................80 5-6 Prestimulus mu oscillations facilitate somatosensory perception ......................................81 5-7 An example of Granger causality spectr a between prefrontal and contralateral S1 cortical activities ................................................................................................................82 5-8 Granger causality influences between pref rontal and contralatera l S1 cortical areas. ......83 5-9 Relationship between the Granger causal influences from prefrontal to S1 cortical areas and N1 amplitud. ......................................................................................................84 5-10 Relationships between prestimulus mu pow er and the Granger causality influences from prefrontal to contra lateral S1 cortical areas ..............................................................85 5-11 Granger causal influences be tween different cortical areas. ..............................................86 8


LIST OF ABBREVIATIONS ACC Anterior cingulate cortex AMVAR Adaptive multivariate autoregressive ANOVA Analysis of variance BESA Brain electrical source analysis CSD Current source density EEG Electroencephalogram ECoG Electrocorticographm ERP Event-related potential EMG Electromyogram EPSP Excitatory postsynaptic potentials fMRI Functional magnetic resonance imaging ICA Independent component analysis IPSP Inhibitory postsynaptic potentials ISI Interstimulus interval LFP Local field potential LORETA Low resolution brain electromagnetic tomography M1 Primary motor cortex MEG Magnetoencephalograpm MPOs Membrane potential oscillations MUA Multiple unit activity N1 Middle-latency (~140 ms) evoked potential PCA Principle component analysis PET Positron emission tomography 9


PFC Prefrontal cortex PPC Posterior parietal cortex ROI Region of interest RT Reaction time S1 Primary somatosensory cortex S2 Secondary somatosensory cortex SCD Scalp current density SD Standard deviation SEM Standard error of mean SEP Somatosensory evoked potential sLORETA Standardized low reso lution brain electromagnetic tomography SOBI Second order blind identification 10


Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy THE FUNCTIONAL ROLE OF NEURAL OSCILLATORY ACTIVITY IN PERCEPTION AND ACTION By Yan Zhang August 2008 Chair: Mingzhou Ding Major: Biomedical Engineering Spontaneous ongoing neural oscillations are trad itionally considered to reflect an idling or deactivated brain state. In contrast, recent s lice studies have demonstr ated that the moderate amount of membrane potential osc illations can bring the local ne uron populations closer to the firing threshold, making them more sensitive to a weak sensory stimulus Other reports showed that very high levels of spontaneous oscillat ions inhibit sensory evoked responses. Hence, to distinguish these competing views, we investigated the relationship between ongoing neural oscillations preceding stimulus presentation a nd subsequent cognitive processing and behavior. First, we studied the relationship betw een reaction times and prestimulus neural oscillations in a broad frequency band over multiple cortical regions. Second, we explored the relationship between middle-latency somatose nsory evoked potentials and prestimulus mu oscillations. Third, we investig ated the large-scale synchroni zation and directionality of prestimulus neural activity in sensory perception and sens orimotor processing by applying adaptive multivariate autoregressive coherence and Granger causality analysis. In conclusion, our main findings suggest that cortical neural osci llations, especially spontaneous ongoing neural oscill ations, are crucial in subseq uent cognitive perception and behavioral performance. Also, the prestimulus ne ural synchronization over large-scale cortical 11


areas, play a key role in sens ory processing and sensorimotor integration. For perception and action, these neural synchronizations help the top-down signals intrinsically generated from higher-order cortical regions (e.g., the prefrontal cortex ) to interact with the bottom-up signals from lower-order cortical regions to facilitate the processing of forthcoming stimuli. These topdown attentional influences among multiple cortical regions are expected to play a functional role in selective processing of sensory input and motor output. Hence, this study suggests that spontaneous ongoing activity prior to stimulus ons et may reflect an active not an idling brain state. 12


CHAPTER 1 INTRODUCTION 1.1 Background Oscillatory neural activitie s have been widely observed in humans, nonhuman primates, and other mammalian species for several decad es. These oscillations are conventionally described in terms of frequency bands, delta ( 1 Hz), theta (3 Hz), alpha (8 Hz), beta (14 Hz) and gamma (30 Hz). In recent years, new recording techniques have been widely applied to measure these neural oscillatory activities at high te mporal and spatial resolutions. Examples include invasively-recorded local field potential (LFP) a nd electrocorticography (ECoG), as well as non-invasively recorded electroencephalogram ( EEG), magnetoencephalogram ( MEG), positron emission tomogra phy (PET), and functional magnetic resonance imaging (fMRI). Particularly, LFP is an electrical signal of oscillatory neural activity from a population of cortical neuron s located around electrodes, while scalp EEG or MEG measure a larger scale elec trical neural activity, mostly like a superposition of LFPs. They have high temporal resolution (~ 1 ms), but lo w spatial resolution, while PET and fMRI have excellent spatial resolution but poor temporal resolution. Spontaneous EEG signals were first robustly recorded from the human scalp in the mid1920s by a psychiatrist (Berger, 1929). The firs t observed brain wave was termed the alpha ( ) rhythm, mostly dominating in pos terior regions. In healthy humans, the alpha rhythm has highamplitude quasi-sinusoid waveforms during eye-clos ed rest and diminishes when the eyes are opened or under cognitive tasks. In addition, there are two more functionally distinct alpha-like rhythms. One occurring in preand post-central gyrus is called the mu rhythm (), sensitive to somatosensory stimuli and motor movement. The ot her generated above aud itory cortical region is termed the tau ( ) rhythm, significantly blocked by acous tic stimulation. The alpha-like rhythm 13


is also found in prefrontal and other frontal cortices of healthy humans or during coma or anesthesia states (Nunez et al ., 2001). The faster brain rhythm observed in sensorimotor and other frontal cortices is called the beta ( ) rhythm, effectively associated with high-level stimulus processing and sensorimotor integration. The smallest-amplitude brain waveform, called the gamma ( ) rhythm, is often observed in pr imary sensory cortical areas. These brain waves can be categorized as ongoing, evoked and induced oscillations. Ongoing oscillations indicate the spontaneous brain waves occurring before a stimulus or without any sensory stimulation. Evoked and induced oscillations both are referred to poststimulus events: e voked oscillations are phase-locked to stimulus onset and can be observed in average event related potentials, whereas induc ed oscillations can be cancelled out completely in this average approach. These alpha-like rhythms and other band r hythms are believed to represent neural oscillations of postsynaptic potentials in ne ocortex (Nunez et al., 2001) At cellular levels, neurons in neocortex and other cor tical structures have intrinsic membrane potential oscillations (MPOs), influencing the precise timing and inte gration of sensory input (Desmaisons et al., 1999; Bland et al., 2002; Vertes, 2005; Schaefer et al., 2006). These oscillatory activities are also determined by the intrinsic network organization and synchronization of the underlying neurons (da Silva, 1991). These intrinsic MPOs were found to synchronize over several centimeters and were also correlated with LFP (Nunez et al., 198 7; Bland et al., 2002). Thus, these brain rhythms to some extent influence subsequent sensory and motor responses. The brain rhythms are extens ively explored in many resear ch fields, including brain computer interface and cognitive neuroscience. Speci ally, scalp EEG signals are widely used in various clinical applicat ions. Analyses of these rhythms have helped with clinical diagnosis of 14


various diseases, such as epilepsy, developmen tal disorders, movement disorder, autism, dementia and psychiatric diseases (da Silva, 1990). An increasing number of cutting-edge techniques help evaluate a nd analyze EEG signals, includ ing power spectral analysis, independent component analysis (ICA), principle component analys is (PCA), second-order blind identification (SOBI), low re solution brain electromagne tic tomography (LORETA), and minimum-norm and beamforming algorithms. 1.2 Objectives In cognitive neuroscience, an increasing number of studies support the idea that preor post-stimulus neural oscillations play signifi cant roles in attention (Engel et al., 2001), anticipation (Brunia, 1999), associative learni ng (Miltner et al., 1999), movement maintenance (Conway et al., 1995; Brown, 2000; Brovelli et al., 2004), cognition and perception (Rodriguez et al., 1999). However, the underlying neural basis and functions of these ne ural oscillations are still in active debate. In a classic view, ongoing s pontaneous activity is consid ered background noise and unrelated to subsequent stimul us processing. Specifically, the fact that spontaneous neural oscillations around 10-Hz and 20Hz occur in the absence of stimuli or movement and are suppressed after stimulus appearance or during movement execution is taken as evidence that they reflect passive or idling brain states (Pfurtscheller et al., 1996b). Furthermore, these oscillations, specially alpha or alpha-like rhythms, are thought to reflect the top-down inhibitory control of cortical stimulus processing becau se they strengthen during memory retention, implying the block of item retrieval, and dimini sh during memory retrieval (Klimesch et al., 2007). However, these two hypotheses have been challenged by recent studies (Arieli et al., 1996; Destexhe and Pare, 1999; McCormick et al ., 2003; Shu et al., 2003; Linkenkaer-Hansen et 15


al., 2004), which suggest that spontaneous oscillatory activity may reflect a depolarizing drive to local neuron populations and bri ng them closer to firing thres hold, thereby facilitating the sensory detection of weak sensory input. Thus, the absence or lack of spontaneous activity results in small sensory responses. However, the extreme high level of spontaneous ongoing oscillations has been also found to inhibit not enhance cortical sensor y processing by competing with sensory evoked responses (Petersen et al., 2003; Dehaene and Changeux, 2005). Classical theories postulated that the brain is a passive system, determined by external stimuli. However, recent studies (Engel et al ., 2001) suggest that the brain is an active and adaptive system and the internal brain states influence stimulus information processing. Topdown signals, which reflect states of attention and anticipation, intrinsically generated from higher-order cortical regions, inte ract with bottom-up signals genera ted from lower-order cortical regions to create an optimal brain state a nd thus facilitate se nsory responses. The highest level of this corti cal hierarchy is the prefrontal cortex, located in the anterior part of the frontal lobe. Anatomical evidence and numerous physiological studies (Fuster, 2001) suggest that the prefrontal cort ex mediates top-down control by sending excitatory and inhibitory bias signals to lower-order sensorimotor circ uits. Conversely, the primary sensory cortices, including visual, auditory, and somatosensory cortex, are believed to be the low order cortical regions. The top-down control is found to be mediated by corticocortical synchronization in both ongoing and evoked neural activit y. A great deal of research focuses on the functional significance of neural synchronization during stimulus proce ssing. However, ongoing neural activity preceding the presentation of stimul us, reflecting anticipa tion and prediction of forthcoming sensory and motor events, attracts in creasing interest (Arieli et al., 1996; Kastner et 16


al., 1999; Tsodyks et al., 1999; Engel et al., 2001; Dehaene and Cha ngeux, 2005). It is suggested that the intrinsic bias signals fr om high-order cortical areas might create a facilitatory brain state conducive to subsequent stimulus processing. Hence, this research investigates the relationship between spontaneous ongoing oscillations and sensorimotor proc essing along three specific aims. Investigate the relationship between prestimulus neural oscillations over multiple cortical regions and the speed of visuomotor processing. Investigate time-frequency feat ures and network connection ch aracteristics of cortical sensorimotor oscillations in a vi suomotor discrimination task. Investigate the relationship between prestimu lus mu oscillations in the sensorimotor cortex and the N1 component, which is a middle-latency somatosensory evoked potential component, in a somatosensory perception experiment and test the hypothesis that the top-down attentional control facilitates subsequent sensory perception. 17


CHAPTER 2 GENERAL METHODS The main analysis techniques applied in the present study are: adaptive multivariate autoregressive (AMVAR) power and coherence spectral analysis, Granger causality spectral analysis and conditional Gra nger causality analysis. 2.1 Adaptive Multivariate Autoregressive Spectral Analysis Adaptive multivariate autoregressive modeling is a parametric spectral analysis method in which time series models are adaptively extracted from the dataset having a number of realizations (Ding et al., 2000). Th e fundamental assumption of this algorithm is that the shortwindow time series can be treated as realizations of an underlying stationary stochastic process. As cognitive information processing changes rapidly, AMVAR can be used to investigate the time series in a short window size (< 100 ms). The following is the procedure of AMVAR. Let ( ) X t be a p-dimensional stationary random process[ (1,),(2,),,(,)]T X tXtXpt () X t can be modeled by the autoregressive equations (Equation 2-1). Here () A i are the unknown coefficient matrices where pp 1,2,,i m () E t is the zero-mean uncorrelated noise term with covariance matrix and m is the model order. ()(1)(1)(2)(2)()()() X tAXtAXtAmXtmEt (2-1) We multiply ()T X tk where 1,2,,k m to Equation 2-1 and take expectation on both sides to acquire the Yule-Walker equations. () R j are the covariance matrices of ()()T X tXtj with lag j Coefficient matrices () A i and covariance matrix of noise term () E t are obtained by solving Equation 2-2 thr ough the Levinson, Wiggins and Robinson (LWR) algorithm (Morf et al., 1978). 18


()(1)(1)()()0RkARkAmRkm (2-2) Spectral features are derived fr om AMVAR models after acquiring () A i and estimates. Equation 2-1 can be written in a spec tral domain equation (Equ ation 2-3). The spectral transform () X f can be obtained by dividing () A f on both sides of Equation 2-3 and define () H f as a transfer matrix. ()()() A fXfEf (2-3) ()()() X fHfEf (2-4) 1 2 0()()m ijf iHfAie (2-5) Thus, spectral power can be obtained through Equation 2-6. The symbol means both transpose and complex conjugate. Spectral power is contained in the diagonal terms of the spectral matrix. Coherence spec tra between two random process ()Sf (,) X it and (,) X jt is defined as If the coherence value is equal to 1 or 0, the two processes are maximally interdependent or independent, re spectively. Here the autoregressi ve model order is determined by minimizing the Akaike In formation Criterion (AIC). ()Cf *1 ()lim[()()]()()NSfEXfXfHfHf N (2-6) 2() () ()()ij ij iijjSf Cf SfSf (2-7) 22 ()2log[det()] mp AICm N (2-8) 19


2.2 Granger Causality Analysis Granger causality is a key technique to assess the di rectionality of neural communications. Multivariate auto regressive modeling provides a natural and useful framework for incorporating the computa tion of Granger causality. The fundamental idea of Granger causality is that if the better prediction of one time series () X t could be achieved by including the past measurement of time series is said to have a causal influence on ()Yt ()Yt () X t 2.2.1 Pairwise Analysis Given two stochastic processes t X and and assuming they are jointly stationary, tY t X can be represented by linear autoregressive models. Here n is a gauge of the prediction accuracy and n is a gauge of the prediction accuracy of the new expanded predictor. The covariance matrix of the noise term is 11 11 11 nnmnmn nnknknknkXaXaX XbXbXcYcY n (2-9) (2-10) 1112 2122 Granger (Granger, 1969) formul ated that, if variance of n is less than that of n namely () 1 ()n nVar Var is said to have a causal influence on ()Yt () X t Geweke (Geweke, 1982; Ding et al., 2006) showed that, the to tal linear in terdependence between two time series XYF () X t and can be decomposed into three components and Here and are the causal influences due to their intrinsic interaction patterns. is a measure of the linear directional influence from ()Yt XYF YXF XYF XYF YXF XYF () X t to Similarly, is a measure of the return ()Yt YXF 20


influence from to ()Yt () X t is the instantaneous causality due to factors possibly exogenous to the (, XYF ) X Y system (e.g., a common driving input) XYXYYXXYFFFF (2-11) Geweke demonstrated that the measures of Granger causality in time domain could be decomposed into their frequency content. is defined as Granger causality in time domain and F () I f is defined as Granger causality spectrum. 1 () 2YXYX F Ifd f (2-12) Each of the above time domain quantitie s is defined in the following equations. C( ) is the coherence function and 2 f ,1 ln(1()) 2XY F Cd (2-13) 1 () 2XY XY F I d (2-14) 1 () 2YX YX F I d (2-15) ..1 () 2XY XY F Id (2-16) 2 12 1122() () ()() S C SS (2-17) After some algebraic deriva tion, the Granger causality spec trum from one time series 1() X t to the other one 2() X t is expressed as: 2 2 12 11 21 22 12 22()( ()ln(1 ) () Hf If Sf ) (2-18) Similarly, the Granger causality spectrum from 2() X t to 1() X t is: 21


2 2 12 22 12 11 21 11()( ()ln(1 ) () Hf If Sf ) (2-19) Here is the power spectra of ()iiSf ()i X t 2.2.2 Conditional Granger Causality Analysis For multiple time series, the aforementioned pair wise analysis is unable to reflect the true casual patterns due to mediated causal influences. Conditional Granger ca usality analysis (Chen et al., 2006; Ding et al., 2006) is defined to resolve whether th e interaction between two time series is direct or is me diated by a third process. Given three stochastic processes t X and tY t Z their joint represen tation is given by the following linear models: 11 111 111 njtjjtjjtj jjj njtjjtjjtj jjj njtjjtjjtj jjjXdXeYgZ YkXmYnZ ZuZvYwZ t t t (2-20) The covariance matrix of the noise terms is: x xxyxz yxyyyz zxzyzz (2-21) The Granger causality from to tY t X conditional on t Z is defined to be: 11 |lnYXZ x xF (2-22) 22


If the causal influence from to tY t X is completely mediated by t Z 11 x x and The spectral representation of conditional Granger causality has also been developed and will be used in this study. |0YXZF 23


CHAPTER 3 ONGOING OSCILLATIONS AND VISUOMOTOR PROCESSING SPEED 3.1 Introduction Oscillatory neural activity has been widely observed in nonhuman primates and humans. Alpha and gamma oscillations ar e characteristics of the primat e visual systems (Vanessen and Gallant, 1994), while Beta (15 Hz) oscillations were prevalent in local field potentials of the primate sensorimotor and parietal cortices (Sanes and Donoghue, 1993; MacKay and Mendonca, 1995). These neural oscillations are believed to be associated with neural communications and to some extent be correlated with the human brain rhythms. However, the underlying neural mechanism of these neural oscill ations is still unclear. The classical notion of neural activities in the alpha and low beta (8 Hz) bands is that they represent an idling state of the brain (Pfurtscheller et al., 1996b). The main evidence is that the alpha and low beta activ ity is diminished with attentio nal demand whereas the high beta band activity is enhanced in cognitive processing (Ray and Cole, 1985). Al so, alpha oscillations have been proposed to reflect the spatially-sp ecific disengagement of visual anticipatory attention (Foxe et al., 1998). On the other hand, alpha band rhythm ha s been reported to be a sign of good cognitive and memory performance (Klimes ch, 1999). Likewise, cortical beta activity has been reported to be associated with states of vigilance, attention to sensorimotor integration (Murthy and Fetz, 1992) or maintenance of a pr ecise motor force or limb position (MacKay and Mendonca, 1995; Brovelli et al., 2004). These conf licting views about these alpha and beta oscillations may stem from diffe rent experimental paradigms and different analysis strategies employed in various studies. A uniform design in conjunction with standa rdized analysis holds the key to resolve these conflicts. 24


Reaction time (RT) measures the overall efficacy of cognitive and sensorimotor processing. RT is known to exhibit c onsiderable variability from tria l to trial, which is thought to reflect the spontaneous fluctuation of the subjects state of expect ancy and attention, with faster RT arising from a more alert and attentive state. Theoretically, the fluctuation of RT is often modeled from the perspective of stimulus-drive n information processing. Recent work has begun to examine how cortical activity prior to stimul us presentation impacts subsequent visuomotor performance. In light of the fact that various oscillatory population activities have been shown to mediate important cortical functions, such as anticipation of forthcoming stimuli and motor preparation (Brunia, 1999; Enge l et al., 2001; Varela et al., 20 01), it raises the question of whether and how spontaneous ongoing oscill atory cortical activity affects RT. Previous studies (Donchin and Lindsley, 1966; Morrell and Morrell, 1966; Haig and Gordon, 1998; Gonzalez Andino et al., 2005) have de monstrated that specific components of EEG and MEG, for example, ERP amplitude an d spectral power, are correlated with RT. Specially, it has been suggested that prestimulus neural activity is significantly correlated with motor performance (Winterer et al., 1999; Liang et al., 2002; Linkenkaer-Hansen et al., 2004; Gonzalez Andino et al., 2005). Liang et al. (2002) showed that pr estimulus beta power in the prefrontal cortex was negatively correlated with RT. Winterer et al. (1999) demonstrated that increased prestimulus delta-band EEG signals in frontal cortical areas reflecting increased cortical activation, were significan tly correlated with faster RT. However, many previous reports of prestimu lus activity have been basically limited in two key aspects. First, due to common refe rence and volume conduction, studies based on human scalp EEG/MEG have methodological diffic ulties to localize the underlying neural generators. The relatively poor spa tial resolution results in inabil ity to study brain functions in 25


specific brain systems. Second, data analysis ha s been often confined to a single predefined frequency band, which prevents the investigation of oscillatory interaction between different frequencies. In the present study, to overcome these probl ems, we analyzed simultaneous LFPs from up to sixteen cortical recording sites distributed broadly over one hemisphere in three macaque monkeys trained to perform a visuomotor patte rn discrimination task. The relation between prestimulus LFP activity and RT was examined across a broad range of frequencies and cortical regions. 3.2 Materials and Methods 3.2.1 Paradigm and Data Acquisition Experiments were performed at the Laborator y of Neuropsychology at National Institute of Mental Health (NIMH). Animal care was in accordance with the institutional guidelines at that time. The LFPs data, called the Nakamura dataset, were simultane ously recorded from multiple surface-to-depth bipolar teflon-coated pl atinum electrodes chronically implanted in one cerebral hemisphere of three highly trained maca que monkeys performing a visual-motor pattern discrimination task. The data were analog filter ed (-6 dB at 1 and 100 Hz, 6 dB per octave falloff) and digitized at 200 samples/s. The monkeys initiated each trial by depressi ng and holding steadily a lever with their preferred hand. Visual stimuli randomly appeared after the beginning of eac h trial and stayed on a display screen for 100 ms. The visual stimulus ons et was defined as 0 ms. Each visual stimulus consisted of four dots arranged as a (leftor right-slanted) diamond or line. Monkeys responded (GO conditions) to one visual pattern type ( line or diamond), and wit hheld responding (NO-GO conditions) to the other. 26


Data collection began about 90 ms preceding the presentation of visual stimuli and continued until 505 ms post-stimulus. On GO trials, the monkey received a water reward at 505 ms post-stimulus if the hand was lifted within 500 ms. On NO-GO trials, the lever was kept depressed for 500 ms post-stimulus, and released thereafter. RT was defined as the time when the monkey released the lever. GO and NO-GO trials were presented with equal probability in each session. The approximate electrode placemen ts as marked visually during surgery are shown on schematic drawings of the lateral view of three macaque brains (Figure 3-1). The locations of all the elect rodes were designated by arbitrary letters. 3.2.2 Data Analysis In the present study, due to two reasons, th e prestimulus time window was defined as (90 ms, 35 ms). First, previous work (Ledberg et al., 2007) showed that the earliest evoked sensory responses occurred about 50 ms. This fact suggests 35 ms poststimulus activity doesnt interfere with the evoked activit y. Second, the interval of the de fined prestimulus period is 125 ms, including the lowest (8 Hz) frequenc y of alpha (8 Hz) oscillations. Three LFP data sets, each including several sessions, were selected from three monkeys (TI, LU, and GE). Even in the same monkey, RT showed sizeable differences on the order of 150 ms or more. Since this study inve stigated the relationships betw een RT and prestimulus neural oscillations, only trials with correct GO responses were used. We preprocessed the LFP data by the following procedure. First, for each monke y, the trials having random responses (< 200 ms) or delayed responses (> 450 ms) were removed. Second, due to an insufficient number of trials of a single session for subsequent spectral analys is, it was essential to combine trials across sessions in each monkey. Third, to reduce session-to-session vari ability in each monkey, sessions having the similar statistical characteristics, including consistent av eraged ERP waveforms, spectral power, and RT distributi on, were selected. After rejecti ng outliers, bad channels and 27


sessions, three sessions in TI, five in LU, and three in GE respec tively were used for subsequent analysis. Totally, 1000-1500 trials for each monkey were used in this study. After preprocessing, trials were sorted by RT into 100-trial groups, starting with the fastest RT and proceeding to the slowest. To yield more groups, each group overlapped 50 trials with the previous one. Twenty one, twenty seven and eighteen groups for TI, LU, and GE respectively resulted. The RTs within each grou p were averaged to yield group mean RT. Adaptive multivariate autoregressive power spectral analysis (Ding et al., 2000) was performed on each subensemble group during the prestimulus period. The subensemble mean at each site during the prestimulus period was subtracted from the individual prestimulus time series on each trial to ensure that the time seri es could be treated as coming from a zero-mean stochastic process, which is essential for the au toregressive spectral anal ysis. A model order of 8 was chosen by locating the minimum of the Akai ke Information Criterion (AIC) (Akaike, 1974). For each recording site, the Spearman-rank correlation coefficients ( ) were computed between the power at each integer frequency in the range of 8to 40-Hz and group mean RT. Thus, a multiple comparison correction was performed (14 sites and 33 frequencies in each monkey) by a permutation procedure (Nichols an d Holmes, 2002). The order of group mean RT was randomly permutated for 500 times. At each permutation step, a maximum value of correlations over 14 recording sites and 33 frequency points was saved. All these maximum values were used to create a null-hypothesis di stribution. By comparing the original correlation value to this permutation distri bution we made sure that the pr obability for one correlation of having a significant effect at one frequency point occurring by chan ce was significant at p < 0.05. 28


3.3 Results The relationship between prestimulus spect ral power (8 Hz) and group mean RT was examined here. The goal was to determine whethe r prestimulus spectral power was significantly correlated with RT, and if so, at which locations and which frequencies and this occurred. The correlations coefficients between spectral power and RT were found to significantly positive or negative, sometimes even non-significant, which depend on recording site and frequency. Figure 3-2 shows two examples from monkey TI. The powe r spectra from a prefr ontal site (M) have a prominent peak at 16 Hz, whereas those from a striate site (B) don t show peaks at any frequency. On the bottom row, the scatter plot s between group power at 16 Hz and group mean RT show a significant nega tive correlation (Spearman, = -0.91, p < 0.0001) at site M and a significant positive correlation (Spearman, = .73, p < 0.001) at site B. Here the p-values were calculated on a single recording site and single integer frequency basis. Figure 3-3, 3-4, and 3-5 show the summarized re sults of the correlation analysis for all frequencies and sites in monkeys TI, LU, and GE, respectively. The horizontal axis displays frequencies from 8 to 40 Hz, and the vertical displays Spearmans rank correlation coefficients from -1 to 1. Gray shaded regions indicate the frequencies having spectral power that was correlated with group mean RT at p < 0.05. Four main results were found. First, beta-band spectral power in the pr efrontal cortex was negatively correlated with group mean RT in monk eys TI and GE. In monkey TI, of the four prefrontal sites (L, M, N, and O), three sites (L, M, and O) had signifi cant negative correlation coefficients ( < -0.70, p < 0.05) in the frequency ranges 16 Hz, 12 Hz, and 17 Hz, respectively. In the other prefr ontal site (N), the correlations in the frequency range 16 Hz were also negative, although these correlations di d not reach statistical significance. In monkey GE, site O among the two prefrontal sites (N a nd O) also had significan t negative correlations ( 29


< -0.62, p < 0.05) in the frequency range of 15 Hz. On ly one prefrontal site (N) was recorded from monkey LU and had insignificant correlations with group mean RT. The second findings concern sites in the se nsorimotor cortical areas from all three monkeys. In LU, two (K and L) of three sites had significant positive correlations ( > 0.61, p < 0.05) over a broad frequency range from 8 to 33 Hz. The remaining site (M) had also positive correlations in the frequency range of 8 Hz al though they did not reach significance. In GE, two (J and K) of four sensorimotor s ites showed significant positive correlations ( > 0.62, p < 0.05) in the frequency range of 8 Hz. Power at th e two sensorimotor sites (J and K) in TI was not significantly correlated with grou p mean RT in the range of 8 Hz. Third, the correlations found in the visual cortex were consistent across all three monkeys. Here the visual cortical areas include st riate, prestriate, and in ferior-temporal recording sites. Figure 3-3 shows that spectral power at all six visual cortical recording sites (B, C, D, E, F, and G) in monkey TI had significant positive correlations ( > 0.65, p < 0.05) with group mean RT, mostly in the range of 8 Hz. In LU, three (A, B, and D) of th e six visual cor tical sites had significant positive correlations ( >0.61, p <0.05), primarily in the range of 8 Hz, whereas the positive correlations found in the remaining three visual sites (C, E, and F) didnt reach statistical significance. In GE, power at three (B, E, and F) of six visual cortical site s (A, B, C, D, E, and F) had significant positive correlations ( > 0.65, p < 0.05) with group mean RT, and this was mostly in the frequency band of 8 Hz. Fourth, the superior-temporal recording sites in two monke ys (TI and LU) had positive correlations with group mean RT. In TI, one su perior-temporal site (H) showed significant positive correlation ( > 0.73, p < 0.05) at 25 Hz. In LU, one of the two superior-temporal sites (H) had significant positive correlations ( > 0.62, p < 0.05) in the range of 12 Hz. The 30


other site (G) had also positive correlations in the range of 8 Hz, which didnt reach significance. Monkey GE had no superior-temporal sites. 3.4 Discussion In this study, we investigated the statistic al relationship between prestimulus spectral power in the range of 8 Hz and RT over broa d cortical regions. Our findings showed that prestimulus oscillatory activity was either positively or negatively correlated with RT in different cortical regions, which would indicate different underlying neural functions. Our main findings are summarized as follows: (1) spectral power in th e prefrontal cortical si tes, in the beta range (14 Hz), showed significant nega tive correlations with RT in two monkeys; (2) power at the visual cortical sites, in th e alpha/low beta (8 Hz) ranges, was significantly positively correlated with RT in two monkeys ; (3) power in the sensorimotor cortical sites, in alpha and beta ranges (8 Hz), showed significant positiv e correlations with RT in two monkeys. The prefrontal cortex is highly interconnect ed with all cortical sensory and motor systems, and is believed to be situated at the hi ghest level of the cortical executive hierarchy that controls attention, movement execution, and othe r cognitive functions (Br unia, 1999; Knight et al., 1999; Engel et al., 2001; Fu ster, 2001; Miller and Cohen, 2001). In our study, the observed negative correlations of prestimulu s neural oscillations in prefront al cortex are consistent with a hypothesized role of the prefrontal cortex in mediating top-down c ontrol in anticipa tory attention by sending modulatory bias signals to lower-order sensorimotor circuits and thus enhance subsequent stimulus processing.. Li ang et al. (2002) found that pr estimulus beta oscillations in the prefrontal sites in one monkey TI was s ynchronized in a prefr ontal network and was negatively correlated with RT. In a human EEG study, Silberstein et al. (2004) found that the increased phase synchronization over frontal recording electrodes prior to stimulus onset was correlated with faster RT, these results similar to our findings for these pr efrontal sites. All the 31


findings suggest that before stimulus presentati on the more enhanced attentional processing is mediated by the prefrontal cortex the faster is the behavioral response. LFP activity within alpha and lower beta fr equency band has been widely observed in the visual cortical areas of cats and monkeys (E ckhorn et al., 1988; Gray and Singer, 1989). Different hypotheses have been proposed to acc ount for the role of these synchronized neural oscillations: cortical idling (Pfu rtscheller et al., 1996b), inhibitory control and timing of cortical processing (Klimesch et al., 2007), and visual attentional disengagement (Foxe et al., 1998). Pfurtscheller et al. (1996) hypothesized that neural oscill ations in the range of 8 Hz are characteristic of cortical areas that pass into an idling stat e, in which no sensory and motor information is received or processed. Furthermor e, Klimesch et al. (2007) proposed that alpha oscillations may reflect the inhibition control of conflicting or unnecessary sensory cortical processes to the tasks in hand by increasing si gnal to noise ratios within cortex. Foxe et al. (1998) reported that elevated alpha activity during the prestimulus time period at parietooccipital regions indicated visual attentional disengagement duri ng selective auditory attention when subjects were cued to the auditory rather than to the visual modality. These three hypotheses are not exclusive, and all of them ag ree with the idea that alpha and lower beta activity is inversely re lated to cognitive information pro cessing. Our findings that alpha and lower beta activity at visual cortical sites was positively correlated with RT imply that higher levels of alpha and lower beta act ivity reflect lower levels of vi sual processing, hence leading to longer RT, which is concordant with the aforementioned hypotheses. Oscillatory LFPs within the frequency range of 15 Hz have frequently been observed in the sensory and motor cortical circuit of nonhuman primates (Murthy and Fetz, 1992; Sanes and Donoghue, 1993; Baker et al., 1999). However, th e functions of these os cillations are still 32


elusive. Some authors have proposed that thes e high-frequency LFP oscillations may play an active role for cortical binding between different motor areas and differe nt parts of the motor system (Donoghue et al., 1998; Baker et al., 1999). Brovelli et al (2004) reported that a largescale sensorimotor cortical ne twork including the primary soma tosensory cortex and primary motor cortex was bound together by prestimulus be ta oscillations in supporting prestimulus motor maintenance of a fixed hand position. Our obs ervation of positive correlations with RT in sensorimotor cortical sites further supports th e hypothesized role of these oscillations in supporting motor maintenance behavior. That hypothe sis is also line with the neural competition theory that a greater allocation of neural resources to the main tenance of hand position would be associated with larger high-frequency sensorimotor oscillations, and consequently longer RT to the visual stimulus would result. In summary, we found that the spectral power of prestimulus oscillatory activities was correlated with response time in the range of 8 Hz, in prefrontal, sensorimotor, temporal, and visual cortical areas. Our findings suggest the po sitive or negative correlation is a useful index for sensory and motor processing over broad corti cal areas. Hence, this approach investigating the relationship between ongoing activity and behavior al responses would enab le us to tease out the functional role of these ongoing oscillatory activities. 33


Figure3-1. Approximate electrode placements as marked visually during surgery in the three macaque monkeys, shown on schematic drawings of the lateral cortical hemispheres. The locations of all the electrodes are de signated by arbitrary uppercase letters. The Ant. indicates the anterior part of the hemisphere and the Post. means the posterior part. 34


Figure 3-2. Prestimulus power spect ra (8 Hz) and correlation plot s for two recording sites in monkey TI. Top row: prestimulus power spectra for a prefrontal s ite (M) and a striate site (B). Bottom row: scatter plots sh owing significantly positive or negative correlations between prestimulus group spectral power and group mean RT at 16 Hz. The solid lines shown in c) and d) are s uperimposed linear least-square fits. 35


Figure 3-3. Spearmans rank correlation coefficients between prestimulus group spectral power and group mean RT for monkey TI. The horizontal axis indicates frequency in the range of 8 to 40 Hz. The vertical axis disp lays the correlations from -1 to 1. Gray shaded areas signify correlati ons that are significant at p < 0.05. 36


Figure 3-4. Spearmans rank correlation coefficients between prestimulus group spectral power and group mean RT for monkey LU. The horiz ontal axis indicates frequency in the range of 8 to 40 Hz. The vertical axis disp lays the correlations from -1 to 1. Gray shaded area signifies correlati ons that are significant at p < 0.05. 37


Figure 3-5. Spearmans rank correlation coefficients between prestimulus group spectral power and group mean RT for monkey GE. The horiz ontal axis indicates frequency in the range of 8 to 40 Hz. The vertical axis disp lays the correlations from -1 to 1. Gray shaded area signifies correlati ons that are significant at p < 0.05. 38


CHAPTER 4 CORTICAL SENSORIMOTOR RHYTHMS AND THEIR FUNCTIONAL ROLES 4.1 Introduction Cortical sensorimotor rhythms, occurring in preand post-central co rtical areas, usually include two frequency bands: 8 Hz and 15 Hz. These oscillator y neural activities have been widely observed in rode nts, primates, and humans (M urthy and Fetz, 1992; Sanes and Donoghue, 1993; MacKay and Mendonca, 1995; Bake r et al., 1997; Baker et al., 1999; Wiest and Nicolelis, 2003). A human MEG study (Salme lin and Hari, 1994) demonstrated that the lower frequency band (8 Hz) oscillations, called the mu rhythm, were most likely dominant in posterior somatosensory cortical areas whereas the higher frequency band (15 Hz) oscillations, called the beta rhythm, were orig inated in frontal motor cortical areas. These neural oscillatory activ ities have been found to be synchronized between preand postcentral cortical s ites (Murthy and Fetz, 1992; 1996), and between sensorimotor activity and muscle activity EMG (Conway et al., 1995; Halliday et al., 1998; Gross et al., 2000; Mima et al., 2000; Kilner et al., 2003). However, the functional significance of these synchronous neural oscillations in sensorimot or cortex is enigmatic. It is well known that mu or beta oscillations occur befo re stimulus presentation and voluntary movement and suppress during movement execution. In a visuomotor GO/NO-GO task, the sensorimotor neural oscillations decr ease markedly shortly af ter stimulus onset and rebound after a brief suppression in the NO-GO tr ials (Kuhn et al., 2004). In the Nakamura dataset described in Chapter 3, a strong beta rhyt hmic component was observed in the preand post-central cortical sites prior to stimulus presentation. A previous study (Brovelli et al., 2004) has characterized the prestimulus state of the motor system by applying power, coherence and Granger causality spectral analysis to local fiel d potentials in a visuomotor discrimination task. 39


These authors showed that the pre-and post-cen tral cortical sites wa s bounded by prestimulus beta oscillations and posterior somatosensory cortical sites ex ert causal influences on frontal motor cortical sites. Thus, due to consistent moto r behavior in preand post-stimulus periods, the first goal of the present study wa s to test the hypothesis that the poststimulus rebounded neural beta oscillatory network has the same cortical dynamical organization as the prestimulus state. This was accomplished by characterizing the rebounded beta network with the same coherence and Granger causality analysis em ployed by Brovelli et al. (2004). To explore the robustness of the first hypothesis and its feasibility on other measurement, the second goal is to test the hypothesis that during prestimulus time intervals the similar sensorimotor dynamical organization occurs in the similar scalp EEG experiment. Thus, we conducted the same visual pattern discrimina tion task, performed by normal human subjects. Adaptive multivariate autoregressive modeling tec hniques were applied to the prestimulus state to extract temporal functions of spectral power, coherence and Granger causality (Kaminski et al., 2001; Brovelli et al., 2004). 4.2 Methods and Materials 4.2.1 LFP Experiment and Data Preprocessing Recording sites located in preand post-centr al cortical areas of two monkeys (4 in LU and 6 in GE) were selected for subsequent data analysis. A schematic of approximate electrode placement is shown in Figure 4-1. Trials having inco rrect behavioral responses and artifacts were excluded. For each monkey, the experimental sessions having similar RT histograms were combined to yield approximately 1000 GO trials and the matching numbers of NO-GO trials were selected from the same sessions. Time-frequency power spectral analysis was applied to determine the beta rebound onset in the NO-GO condition. Each sliding window wa s 100 ms in length and slid 5 ms forward. A 40


total of 101 windows was resulted for the entire trial time period (-90 ms, 505 ms). Power in the 15 Hz range averaged over all preand post-cen tral sites was plotted as a function of time (Figure 4). The following procedures were used to detect the onset latency of beta rebound: identify the minimum value of post-stimulus beta power and determine the time (defined as the onset latency of beta rebound) at which this be ta power increased by at least 10% above this minimum. Adaptive MultiVariate AutoRegressive mode ling method (Ding et al., 2000; Ding et al., 2006) was applied to the poststimulus window. To treat the residuals as coming from a zeromean stochastic process which is essential for AMVAR data modeling (Ding et al., 2000), the ensemble mean was removed from each trial fo r each recording site. A model order of 10 was determined by the Akaike Inform ation Criterion (Akaike, 1974). Spectral power, coherence, and Granger causality was calculated in each analysis window. Spectral power reflects the degree of neural synchronizati on in a local neural ensemble. Coherence estimates the degree of neural sync hrony between two distant neural ensembles. Granger causality further gauges whether one neural ensemble exerts a causal influence on another. Significance testing for coherence a nd Granger causality was accomplished by a random permutation procedure (Nichols and Holmes, 2002). Details have been described in Brovelli et al. (2004). In Brovelli et al. (2004), the time interv al (-90 ms, 20 ms) was defined as the wait window for both GE and LU. In the present study, to investigate the dynamical organization of the rebound beta sensorimotor network after its brief suppression in the NO-GO condition, based on the onset latency of beta rebound in each monkey, a post-rebound window was selected and referred to as the recurrence window, (395 ms, 505 ms) for GE and (300 ms, 505 ms) for LU. 41


Coherence and Granger causal ity analysis was performed for the recurrence window and compared with the known results obtained from th e same data sets during the prestimulus time period (Brovelli et al., 2004). 4.2.2 EEG Recording and Paradigm We applied the same AMVAR modeling an alyses to scalp EEG recordings and investigated characteristics of the sensorimot or dynamical organization in two conditions: hold and rest. Human participants (3 males, average 28 years old) were trained to perform the same visual-motor pattern discrimination task in an acoustically and electri cally shielded EEG booth. The study was approved by the Institutional Revi ew Board of University of Florida. At the beginning of experiment, baseline eyeclosed rest EEG signals were recorded for 5 min. Before stimulus presentation, participants were given a visual cue to press and hold steadily a response key on a response panel with the righ t index finger for at least 800 ms. In the hold condition, the prestimulus time interval was define d as (-300 ms, -10 ms). In the rest condition, the continuous baseline EEG data were epoc hed to the matching number of trials and the matching length of each trial. Granger causality spectral analysis was applied to measure directional influences among the appropria te EEG electrodes representing primary somatosensory, frontal motor, and posterior pari etal areas. According to the international 10-10 system, the EEG electrodes contralateral or ip silateral to the perf orming hand about 2 cm posterior to C3/C4 were used fo r left or right S1 activity (LS1 or RS1), electrodes at FC3/FC4 were selected to reflect activity of the frontal motor areas, and the site at Pz was chosen to represent the posterior parietal activity. 42


4.3 Results 4.3.1 LFP Results For each monkey, power spectra were computed for each of the sliding analysis windows and averaged across all recording sites over th e entire trial time peri od. Figure 4-2 shows the power result for GE (A, left column) and LU (B right column) where the GO condition is in the top row, the NO-GO condition in the middle row, and the bottom row is the averaged beta band power as a function of time. In the NO-GO condition, beta activity re bounded significantly at approximately 300 ms for GE and approximate ly 260 ms for LU. The rebounded beta activity was sustained until the end of the trial. Note th at the rebounded beta oscillations in LU were stronger than that existing prio r to stimulus presentation wher eas those in GE were weaker compared with the prestimulus state. Coherence spectra for all site pairs and Granger causality spectra in both directions for all site pairs were computed over the recurrence win dow. Figure 4-3 shows that the beta activity in power, coherence, and Granger causality spectra was prevalent in the recurrence window. Three recording sites common to both monkeys were se lected: primary somatosensory (S1), primary motor (M1) and inferior posterior parietal area 7b. The coheren ce and Granger causality spectra were averaged in the beta frequency range The threshold for significant coherence at p < 0.005 was determined to be 0.020 for GE and 0.016 for LU and that for significant Granger causality was 0.012 for GE and 0.011 for LU. In the wait window, the previous study (Br ovelli et al., 2004) showed that the anterior postcentral (primary somatosensory cortex) sites (S1) played a clear role as a driver to precentral (motor cortex) site (M1) and to post-central (inferior posterior pari etal cortex) sites (7b), with the latter one exerted causal direction to motor cort ex. In the recurrence win dow, these main causal influences (S1-M1, S1-7b, and 7b-M1) were sim ilar to those in the wa it window, except that 43


motor cortex also sent feedback causal influences back to 7b in LU. Figur e 4-4 shows that both monkeys GE and LU had the similar dynamical or ganization in the recurrence window to those in the wait window. Only signifi cant Granger causality values gr eater than the threshold were shown. Our findings from the Nakamura dataset were that the large-scale sensorimotor cortical networks were bound by beta oscillations during wait window and recurrence window and dissolved during movement execution. In both premovement and poststimulus hold periods, Granger causal influences were carried from S1 to both M1 and 7b, and 7b also exerted Granger causal influences on M1, in agreement with the id ea that somatosensory feedback is essential for the sensorimotor system to maintain a constant motor output. 4.3.2 Scalp EEG Results Figure 4-5 shows the Granger causality spectra fr om the left S1 electrode (LS1) to the left frontal motor electrode (FC3) for the presti mulus period condition and the baseline EEG condition. The directional influences in the ra nge of 8 Hz, peaking at 11 Hz, were largely enhanced for the hold condition whereas those in higher (20 Hz) fr equency range are not different between the two conditions. The pairwise directional influences from five electrodes were summarized in Figures 4 A and 4 B. Th e lines connecting different cortical areas correspond to averaged Granger cau sality values between 8 Hz and only the values greater than 0.05 were shown. The arrowheads indicate th e direction of Granger causal influence. For the hold condition, the cortical areas in the left hemisphere are more actively interacted than those in the righ t hemisphere. The left S1 cortical area drives left frontal motor cortical areas and bidirectionally interact with right frontal motor and posterior parietal cortical areas, whereas the right S1 cortical areas do not receive or exert any causal influences. For the baseline condition, the main driver is posterior parietal cortex and left or ri ght S1 cortical areas 44


are not engaged in any sensorimot or networks. All the re sults are in line with the functional role of the primary somatosensory cortex in the ma intenance of a steady motor output. Hence, the main sensorimotor dynamical organization in hu mans and nonhuman primates performing the same tasks are remarkablely consistent. 4.4 Discussion For voluntary movement, the post-movement mu or beta rebound is t hought to reflect an idling state of the brain (Pfurt scheller et al., 1997a) or activ e immobilization (Salmelin et al., 1995), and is thought to be indepe ndent of sensory input (Salmelin et al., 1995; Pfurtscheller et al., 1996a). In the present study, the mu or beta oscillations took place in the conditions when there was no overt movement, prestimulus hold periods and poststimulus rebound periods. Our previous work analysis for the prestimul us periods found that beta activity mediated the causal influences from primary somatosensory cortex (S1) to posterior parietal area 7b, and from both S1 and 7b to primary motor cortex M1 (Brovelli et al., 2004; Chen et al., 2006). Based on these previous results, it was hypo thesized that the functional role of the sensorimotor beta network is to facilitate the sensorimotor integration of sensory feedback information for the maintenance of the depressed hand lever. As th e lever pressure was maintained, it is thus reasonable to suggest that beta rebound signaled the resumption of the same network in support of sensorimotor integration. Similarly, the same pattern of these dyna mical organizations during hold period in EEG recording was also found, even though these sensor imotor networks are mostly active in mu not beta frequency range. Thus, no matter prestimulus or poststimulus hold periods, the functional connectivity plays a key role servi ng to maintain steady motor output. Our network analysis results are thus in line with recent studies that postulate an important role for sensory feedback in coherent mu or beta oscillatory activity (Conway et al., 45


1995; Baker et al., 1997; Baker et al ., 1999; Mima et al., 1999; Kiln er et al., 2000; Kilner et al., 2003; Brovelli et al., 2004; Houdaye r et al., 2006). Baker et al. (1997, 1999) showed that, during a precision grip task, sensorimotor beta oscilla tions were coherent with contralateral hand EMG. Cassim et al. (2001) reported that beta synchronization disappeared after subjects were sensorydeafferented. In summary, cortical sensorim otor rhythms (mu or beta) play a crucial role in binding a large-scale sensorimotor oscillat ory network and thus facilitate sensory integration of sensory feedback information to the motor system, which is part of the cortical portion of a control loop serving to maintain steady motor output. 46


GE LU Figure 4-1. A schematic of approximate electro de placement in monkey GE (left hemisphere) and LU (right hemisphere). Black do ts indicate electrode locations. 47


A B Figure 4-2. Time-frequency plots of averaged power spectra comput ed over all sites in GE (A) and LU (B). Average beta power as function of time for both GO and NO-GO conditions are shown in the bottom row. The color scale indicates the values of power spectra. The vertical lines s hown in pictures were stimulus onset (0 ms) and average response time, respectively. 48


A B C Figure 4-3. Averaged power spectra (5 Hz) over three selected sites (S1, M1, and 7b) (A), and averaged coherence spectra (B) and averaged Granger cau sality spectra (C) computed over all pairwise combinations in the recurrence windows for LU (gray curves) and GE (black curves), respectively. 49


GE LU Figure 4-4. Schematic Granger causality graphs duri ng the recurrence window in GE (top) and LU (bottom). The thickness of the line s between three recording sites and the numbers near the lines indicate the averag ed Granger causality values in beta frequency range. The arrowheads indicate the direction of Granger causality influences. S1: primary somatosensory cortex, M1: primary motor cortex and 7b: posterior parietal cortex. 50


Figure 4-5. Granger causality spectra from left S1 co rtical areas (LS1) to left motor cortical areas (FC3) for the prestimulus holding period (so lid line) and the eye-close rest period (dashed line). 51


FC3 FC4 0.15 Pz LS1 0.1 0.2 RS1 0.25 0.15 FC3 FC4 0.1 Pz LS1 0.1 RS1 0.1 0.1A B Figure 4-6. Granger causality Graphs for (A) the prestimulus time holding period and (B) the baseline condition. LS1 (RS1): left (r ight) primary somatosensory cortex. 52


CHAPTER 5 NEURAL CORRELATES OF SOMATOSENSORY PROCESSING OF A WEAK STIMULUS 5.1 Introduction 5.1.1 Somatosensory Evoked Potentials Somatosensory evoked potentials (SEPs) are time-locked electr ical signals in response to tactile, electrical and other somatosensory stimuli. Since scalp-recorded SE Ps were first found by George Dawson in 1940s, analyses of SEP compon ents have been widely used in clinical diagnosis of various brain, brainstem and spinal cord disorders, such as multiple sclerosis, spinal cord trauma, coma and peripheral nerve lesions (Kraft et al., 1998). In addition, SEPs are extensively studied in vari ous cognitive tasks. In human scalp EEG recording, SEPs (< 200 ms) have several components N20 (20 ms), P60 (50 ms), N80 (70 ms), P 100 (90 ms), and N140 (110 ms) (P, positivity; N, negativity; numbers, msec from st imulus onset). The N20 component, reflecting the initial activation of peripheral somatosensory input in cortical areas, stems from area 3b of S1 (Allison et al., 1989) and varies solely as a func tion of stimulus intensity (Lesser et al., 1979). P60 (or P50) also originates in contralateral S1 (Allison et al., 1992) and mainly encodes the parameters of sensory stimuli even though it is slightly affected by cognitive factors (Desmedt and Tomberg, 1989; Tomberg and Desmedt, 1996). N 80 is also recorded in contralateral S1 (Allison et al., 1989) and independent of sensory perception and attention (Schubert et al., 2006). The P100 component is generated from SII areas (A llison et al., 1992) and is sensitive to target stimuli (Desmedt and Robertson, 1977; Desmedt et al., 1983). Due to different experimental conditions and recording methods, the late SEP components (>200 ms), including N250 and P300, are also reported. In the present study, on ly the middle-latency somatosensory evoked potentialN140, highly correlated with somatosens ory perception and atte ntion, is discussed. 53


The somatosensory N140 component (110 ms, termed here N1 component), being functionally analogous to the visual or auditory N1 component, is the large-amplitude slow SEP component, appearing first in the contralateral S1 and S2 cortical ar eas and then in ipsilateral S1 and bilateral frontal areas. The amplitude of th e N1 component is signi ficantly affected by both exogenous (e.g., stimulus intensity) (Naka jima and Imamura, 2000) and endogenous (e.g., attention) factors (Desmedt and Robertson, 1977 ; Allison et al., 1992; Garcia-Larrea et al., 1995; Nakajima and Imamura, 2000; Giaquinto and Fraio li, 2003; Schubert et al., 2006). For example, N1 amplitude is enhanced with increase in s timulus intensity and attention. N1 amplitude is vulnerable to short ISI (< 400 ms) (Tomberg et al., 1989; Kekoni et al., 1997; Kida et al., 2004). Additionally, the latency of the N1 component is hardly a ffected by the intensity of stimulus (Nakajima and Imamura, 2000) and cognitive tasks (Desmedt and Robertson, 1977; Desmedt and Tomberg, 1989). 5.1.2 Correlation of N1 and Sensory Attention When a near-threshold somatosensory stimul us is presented, it may or may not be perceived. Such perceptual outcomes are belie ved to be reflected by somatosensory evoked responses. In a pioneer study, Libet and his co-workers (Libet et al., 1967) demonstrated that the early somatosensory evoked responses, recorded subdurally in patients, had no significant differences between conscious and unconscious experiences wh ereas the late responses was deficient in unconscious experiences. A similar finding was reported in awake monkeys (Kulics and Cauller, 1986; Cauller and Kulics, 1991) th at the early surface-positive component (P1, 13 ms) was associated with multiple unit activity (MUA) in the middle layers and signaled an earliest thalamocortical activati on of S1 whereas the late su rface-negative component (N1, 45 ms), a sensitive index of behavior performance, wa s associated with current sinks in superficial layers I/II and elevated MUA in layers III-V. 54


Hence, based on the behavioral responses, human scalp N1 component, correlated with sensory attention, is functionally similar to t hose of the aforementioned surface-negative evoked responses. In a scalp EEG study, according to the scalp topography and behavioral performance, Garcia-Larrea et al. (1995) suggested that soma tosensory N1 component could be split in two components: N120 (110-130 ms) and N140 (130-160 ms). The N120 component originates in contralateral S2, occurs in both neutral and attend ed conditions, and therefore is suggested to be associated with somatosensory awareness, wh ereas N140 component, obser ved only in attended conditions, occurs first in contralateral hemisphe re and then is bilaterally distributed. Hence, N140 component is linked to spa tial attention of stimuli and re flects activation and reciprocal interactions among several areas, such as, prefro ntal cortex, supplementa ry motor cortex, and other subcortical structures (Allison et al., 1989; Desmedt and Tomber g, 1989; Allison et al., 1992). Here our current research mostly focuses on the attention effects of the N1 component. It is known that attention causes stronger neur al responses to weak stimuli than those to strong stimuli (Reynolds et al ., 2000). When the effect of the exogenous factors on N1 component becomes smaller, the effect of endog enous factors on N1 becomes bigger. Hence, the neural basis of N1 component to a near-threshold stimulus, are explored in the present study. 5.1.3 Spontaneous Mu Rhythm Mu rhythms (7 Hz), also called rolandic, wicket or center rhythms, gradually gain more and more attention since 1950s (Gastaut and Bert, 1954; Chatri an et al., 1959). The research on mu oscillations is carried out in many broad areas, including brain computer interface (Pineda et al., 2000), movement disorder s (Oberman et al., 2005; Bernier et al., 2007), and smoking addiction (Pineda and Oberman, 2006). In some studies, mu rhythms sometimes also include sensorimotor signa ls in 15 Hz range, which are pr evalent in pre-central gyrus 55


(Salmelin and Hari, 1994). In the present study, mu rhythms refer to the neural oscillations in the 7 Hz frequency range in the post-central gyrus. Spontaneous mu rhythms have been observed in scalp EEG/MEG r ecordings from many healthy humans (Makeig et al., 2002). Figure 5-1 A shows the time traces of spontaneous mu rhythms during eye-closed rest. Mu rhythms have rhythmic waxing and waning patterns and the length of each mu segm ent is around 0.5-2 s. Spontaneous mu oscillations were found to attenuate significantly by movement onset (Pfurtscheller et al., 1997b), movement observa tion and imagination (Pineda et al., 2000), and somatosensory stimulation (Salenius et al., 1997 ; Wiest and Nicolelis, 200 3), and to increase during visual sensory processing (P furtscheller et al., 1996b). Theref ore, these facts are taken as evidence that spontaneous mu oscillations reflect idling or deactivated somatosensory cortical areas (Kuhlman, 1978; Pfurtscheller et al., 1996b; Pfurtscheller et al., 1997b), which implies that spontaneous mu oscillations are uncorre lated with somatosensory processing. Furthermore, according to the phenomenon that the neural synchronization in the alpha or alpha-like rhythms was observed in retention period and the desynchronization was associated with memory retrieval (Klimesch et al., 1999; Jensen et al., 2002), an inhibition-timing hypothesis (Klimesch et al., 1999; Klimesch et al., 2007) is proposed that the synchronization in alpha range blocks the retrieval of previous items, avoids the encoding of new items, and thus reflects the top-down inhibitory control over task-irrelevant brai n areas. On the other hand, an alternative interpretation is that these neural act ivities are involved in coupling broad cortical areas and sustaining memorized items (Jen sen et al., 2002; Palva and Palva, 2007). 5.1.4 N1 and Ongoing Activity Past research has reported conflicting results concerning how N1 amplitude is associated with ongoing mu or alpha power. The positive (B randt et al., 1991; Arie li et al., 1996; Nikouline 56


et al., 2000), negative (Rahn and Basar, 1993; Chen et al., 1999; Ploner et al., 2006), or nonsignificant (Simoes et al., 2004) correlations were found. These conflicting findings may be partly due to experimental conditions and methodological differences, such as, recording techniques, stimulus intensity, and ERP measurement. Nevertheless, these conflicting findings also imply that the relationship between ongoing mu (or alpha) activity and N1 amplitude may be not monotonic. In a recent MEG study, Linkenkaer-Hansen et al. (2004) found that the in termediate level of prestimulus mu power immediately preceding a weak electrical stimulus was correlated with the highest probability of sensory detection, which suggested that pres timulus mu oscillations optimize stimulus processing through intrinsic stochastic resonanc e (Ho and Destexhe, 2000; Stocks and Mannella, 2001). In addition, N1 amplitude is also signifi cantly affected by the phase of spontaneous ongoing mu or alpha oscillations before stimulus presentation Jansen and Brandt (1991) demonstrated that the trials having the phase of positive-going zero crossing at stimulus onset had maximal N1 amplitudes, whereas the trials having negative-going zero crossing has minimal N1 amplitudes. Since the latency of N1 compon ent is in the alpha frequency range and N1 amplitude is affected by prestimulus alpha power, they thought N1 component might be a part of entrained alpha activity. Furthermore, Makeig and his coworkers proposed the phase-resetting hypothesis (Makeig et al., 2002), which suggested th at ERPs are partly generated by the phaseresetting of multiple ongoing EEG processes. In the present study, we only considered the effect of prestimulus mu power on N1 and thus the ra ndom distributions of ISI were used here. Hence, given the preponderance of eviden ce linking higher N1 amplitude with increased sensory processing, we predict that an inverted-U relationship also exists between prestimulus 57


mu oscillations and N1 amplitude. It means that the intermediate level of prestimulus mu power prior to stimulus onset facilita tes cortical sensory processing wh ereas too much or too little mu power may reduce the efficiency of perception. 5.1.5 N1 and Top-down Attentional Control Cauller and Kulics (1986, 1991) postulated that somatosensory N1 component was generated by the interaction between feedforward projections from ascending signals and feedback projections from othe r higher order cortices, includ ing S2, frontal, and posterior parietal cortical areas (Cauller and Kulics, 1991; Cauller, 1995; Cauller et al., 1998; Jackson and Cauller, 1998; Staines et al., 2002; Golmayo et al., 2003). In other sensory modalities, the similar results that the re-entrant processing is essential for sensory perception were found (Lamme and Roelfsema, 2000). These feedforward and feedback projections are believed to be implemented by neural oscillations (Engel et al ., 2001; Buschman and Miller, 2007). An increasing number of studies suggeste d that alpha or alph a-like rhythms were sensitive to various cognitive and attentional tasks and likely played significant roles in stimulus sensory processing (Halgren et al., 2002; Jensen et al., 2002; Se rrien et al., 2004; Pineda, 2005; Pollok et al., 2005; Sauseng et al ., 2005a; Sauseng et al., 2005b; K limesch et al., 2007; Palva and Palva, 2007). Sauseng et al. (2005) reported that the increased pr efrontal alpha synchronization was associated with occipital al pha suppression in a working memo ry task and prefrontal alpha oscillations modulated occipital alpha oscillations by investigating alpha latency shifts. It supports previous studies that the corticocortical synchronizati on in the alpha frequency range may play a significant role in top-down processing (von Stein et al., 2000). Furthermore, the global binding hypothesis (Nunez et al., 2001; Pineda, 2005) was propos ed that all the alpha-like rhythms, locally independent, become coupled an d translate all sensory inputs into perception 58


and action. In addition, a recent simulation study (V anrotterdam et al., 1982) showed that alpha oscillations were available for a large-scale cerebr al integration. Most of these aforementioned findings focu s on the functional signi ficance of neural synchronization during stimulus processing. However, ongoing neural activity prior to the presentation of stimulus, reflecting anticipation and prediction to forthcoming sensory and motor events, attracts increasing inte rest (Arieli et al., 1996; Kastne r et al., 1999; Tsodyks et al., 1999; Engel et al., 2001; Dehaene and Changeux, 2005). It is suggested th at the intrinsic bias signals from high-order cortical areas might create a facilitatory br ain state conducive to sensory perception. To achieve the better understanding the relationship between ongoing mu activity and subsequent somatosensory processing, we invest igated 1) the relations hip between ongoing mu oscillations and stimulus evoked N1 amplitude; 2) the functional role of top-down signals from prefrontal cortex and other fr ontal areas in sensory perceptio n and in the modulation of N1 amplitude. 5.2 Materials and Methods 5.2.1 EEG Experiment and Design A somatosensory perception EEG experiment was performed in an acoustically and electrically shielded booth. EEG signals were recorded from thir teen right-handed participants (24-38 years of age, 3 females), who were fr ee of any movement disorders and neurological diseases. Participants were given informed cons ent before participation and were paid $30 for their participation. The study was approved by the Institutional Review Boar d of University of Florida. Before the experiment, baseline EEG was recorded during eye-closed rest for 5 min. To avoid visual modulation of somatosensory processing (TaylorClarke et al., 2002) and minimize 59


eye blinks and eye movement, participants were asked to close their eyes during the experiment. Both hands of participants were resting on a ta ble in front of them. A pair of stimulation electrodes, 2 cm apart, were placed on the dist al (anode) and middle (cathode) phalanges of the right index finger. The left hand was resting on the ERTS re sponse panel. A biphasic nearthreshold electrical stimulus (0.3 ms duration) was delivered to the right index finger tip. Participants were asked to pre ss a button by the left index finge r as quickly as possible when they felt the electrical stimulus. The maximal inte nsity of electrical stimuli was less than 5 mA. Electrical stimulus onset was defined as 0 ms. The interstimulus intervals varied randomly between 4500 ms to 5000 ms in 6 steps. The intens ity of the electrical st imulus was adjusted to each individual subjects threshold using a staircase method (Leek, 2001). At the threshold intensity the subject was able to perceive half the stimuli. The stimulus intensity is kept the same for the entire experiment. The experiment for each subject includes at least 10 blocks each having 40 stimulation trials. To minimize learning and adaptation effects, before the experiment started, participants were given 40 practice trials. A few breaks were given between blocks. The EEG data were acquired by a 12 8-channel BioSemi ActiveTwo system ( Four extra electrodes placed around the eyes were used to record horizontal and vertical eye movement. Two electr odes were placed in the left and right mastoid. EEG signals were sampled at 2048 Hz online, down-sampled to 250 Hz offline, and bandpass filtered (0.3 Hz). EEG electrode locations were obtained by an electromagnetic 3 dimensional digitizer (Polhemus Corp). Electrical stimuli were delivered by a Grass stimulator (model S48). Response time was recorded by the EXKEY micropr ocessor logic of BeriSoft Experimental Run Time System (ERTS) system 60


5.2.2 Data Preprocessing Trials having incorrect responses (partici pants responded automatical ly without stimulus presentation) or contaminated by EOG or other artifacts (> 100 V) were excluded from further analysis. In one or two specific subjects, second-order blind identification algorithm (Belouchrani et al., 1997) was applied to remove big EOG artifacts and outliers. The trials having response time occurring from 0.15 s to 2.5 s after stimulus onset were accepted for further analysis. Skin conducted stimulation artifact s occurred between -10 ms to 10 ms. Average reference was applied for subsequent analysis Here we preferred not to apply Laplacian algorithms to remove common reference and volume conduction, due to underestimation of coherence and Granger causality by spatial filter ing (Nunez et al., 1999). All data analysis was performed with BESA 5.0 (Brain Electrical Source Analysis, MEGIS software GmbH, Munich, Germany), SPSS (Statistical Packag e for the Social Sciences), MATLAB (Mathworks Inc.), and custom Matlab software. SEP amplitudes were estimated with baseline -to-peak. Baseline correction was performed using the time interval (-100 ms, -20 ms). For each subject, the representation areas of the finger in contralateral S1 cortex was determined by the maximum positive deflection of the SEP components from 50 ms to 70 ms and also by scal p current density (SCD) maps (Perrin et al., 1987) (Figure 5-3 B). The SCD map, generated in BESA 5.0, is an es timate of the second spatial derivative of the voltage potential It shows the scalp areas where the current either emerges or sinks between the scalp and the brain. Several EEG electrodes were selected to represent activities in cortical regions of interest (ROI). The SCD map at 60 ms shows the two EEG electrodes (usually 2 cm posterior to C3) are included in the appr oximately same current contour. Therefore, in the present study, these two electrodes were selected to represent contralateral S1 recording sites. Three electrodes (Fpz, Fp1, a nd Fp2), according to the International 10-10 61


System, were selected to repres ent prefrontal sites (Homan et al., 1987). An electrode (Oz) was selected to represent an occipital cortical site. The other electrode (Pz) wa s selected to represent a posterior parietal cortical site. Statistical significance level was defined as p < 0.05. 5.2.3 Correlation of Prestimulus Mu Power and N1 For each individual, prestimulus mu power of a single trial was calculated by multitaper spectral analysis (Thomson, 1982) in a given pres timulus window, and then averaged over the two selected S1 electrodes. To avoid the erro renous power estimation caused by a short window size, the prestimulus window was defined as (-500 ms, -20 ms). Here the number of tapers was three and the frequency resolu tion (zero-padding used) was 1 Hz. To normalize the interindividual differences of prestimulus mu pow er, the trials were rank ordered by averaged prestimulus band power in the mu frequency ra nge and then sorted into groups whose size corresponded to 10 % of the tota l trials, starting with the sma llest prestimulus mu power and proceeding to the largest, and each group had 5 % of trials overlapped with the previous one. For each power group, the probability of behavior detection was calculated and somatosensory N1 amplitude was averaged w ithin 40 ms window cent ering on the maximum negative deflection of averaged SEP in 110 ms for each individual. Similarly, detection rates and N1 amplitudes for each power group were normalized as the percen tage change of the mean values for each subject. Normalized detection rates and N1 amplitude were averaged across all subjects. The relati onship between power group, norma lized detection rates and N1 amplitudes were calculated. Statistic al tests were based on ANOVA. 5.2.4 Granger Causality Spectral Analysis Granger causality spectral analysis (Geweke, 1982; Ding et al., 2006) was applied to measure the directional influences among several cortical regions. First, we investigated the effect of the directional influe nces between frontal and S1 elec trodes on behavioral performance. 62


An autoregressive model order was chosen for each individual by locating the minimum of the Akaike Information Criterion (Akaike, 1974). Fo r each individual, Gran ger causality spectra were calculated in the prestimulus window for perceived and unperceived conditions, respectively. Second, we investigated the relationship between N1 amplitude and the Granger causal influences between frontal and S1 cortical areas For each individual, trials were grouped by the aforementioned approach. Granger causality spectra and N1 amplitude was calculated for each group subensemble, respectively. All the causality results were averaged over six site pairs between two electrodes in contralateral S1 areas and three in prefrontal areas. Third, the relationship between the Granger causal influences from frontal to contralateral S1 cortical areas and prestimulus mu power was i nvestigated. For each indi vidual, all trials were sorted by prestimulus mu power and ten subensemble groups were resulted. Granger causality spectral analysis was applied in each subens emble group. Statistical tests were based on ANOVA. 5.3 Results 5.3.1 Behavioral Results Trials having incorrect responses or inco rrect response time (<0.15 s or >2.5 s) were less than 1%. Median response time over all thirteen subjects was 612 35 ms (Mean SEM). After data preprocessing, each subj ect had at least 280 trials. The histogram in Figure 5-2 A shows the pr obability of detecting a near-threshold stimulus for each individual. The grand average probability of detecting a stimulus over all subjects was 0.51 0.01 (mean SEM) and it was not significantly different from 0.50 (student t test, t = 1.33, p = 0.21). Figure 5-2 B shows the probability of detecting stimuli as a function of time (around 5 min per block) and it was also not significantly different from 0.50 (student t test, 63


t = -0.02, p = 0.98) even though it was more variable in the last two or three blocks. These behavioral results suggest that, in the present EEG experiment, the probability of detecting a near-threshold electrical stim ulus was kept around 0.50 and did not change significantly across subjects or over time. 5.3.2 Evoked Potentials Figure 5-3 A shows the grand average SE P waveforms (0 ms) for perceived and unperceived conditions. In the present study, the scalp-recorded SEPs for perceived conditions have three main components: N20 (20 ms ), P60 (50 ms), and N140 (120 ms). N20 component was not prominent due to the small intensity of the stimulus used in this EEG experiment and it was also distorted by the f iltering of stimulation artifacts. The mean amplitudes of the positive deflection P60 didnt show significant differences (paired t test, t = 1.81, p > 0.05) between perceived and unperceived conditions. In contrast, N140 component was significantly enhanced (paired t test, t = -8.25, p < 0.00002) for perceived conditions, compared with unperceived conditions. Thes e ERP results were consistent with previous ERP findings (Libet et al., 1967; Ray et al ., 1999; Meador et al., 2002; Schubert et al., 2006), which suggested that SEPs (< 100 ms) were not significantly different between conscious and unconscious experiences. The two SEP componentsN80 and P100 only appeared in three subjects and thus they were not discussed here. Figure 5-3 B depicts 3D scalp current densit y maps of grand average SEPs at 60 ms and 140 ms. Current sources or sinks ar e indicated in red or blue, resp ectively. A dipolar pattern was shown at 60 ms, a current source in post-gyrus area and a current sink in pre-gyrus area. N1 component at 140 msa current sink, was displaye d in broad cortical ar eas, including pre-gyrus areas, post-gyrus areas, and sec ondary somatosensory cortex. 64


5.3.3 Correlation of Prestimulus Mu Power and N1 Amplitude Figure 5-1 shows that ongoing neural activity in contralate ral S1 was dominated by mu rhythm whereas the higher frequency neural os cillations (> 20 Hz) were not salient in contralateral S1, a finding in agreement with pr evious studies (Salmelin and Hari, 1994). All thirteen subjects had prominent mu power peaks over contralateral S1, with peak frequencies from 9 to 11 Hz (mean: 10.5 0.66 Hz). The relationship between prestimulus mu power and N1 amplitude was calculated for each subject. Figure 5-4 shows an example of th e relationship between prestimulus mu power and N1 amplitude for one subjects data set. Averaged SEPs (-500 ms, 200 ms) for ten power groups were shown in ascending order from t op to bottom. Each group was averaged over 30 trials and two contralateral S1 electrodes. Aver aged prestimulus waveforms have prominent mu oscillations in group 9 and 10. It is clear that the middle power groups (e.g., Group 6 or 7) have biggest N1 amplitude, whereas the largest or sma llest power groups have smallest N1 amplitude. In nine of thirteen subj ects, the relationships between prestimulus mu power and N1 amplitude were well described by a quadratic regression model (, p < 0.05) and were poorly described by a lin ear regression model ( p > 0.17) (Figure 5-5 A). Figure 5-5 B shows an example of the inverted -U relationship (quadratic fit, p < 0.006) in one subject. By contrast, the relationships in three subjects were significantly positive or negative (Figure 5-5 C-D). No significant relationships were found in subject 2. 20.56 r 20.22 r 20.76 r The correlations between prestimulus mu power and N1 amplitudes were averaged across all thirteen subjects. Figure 5-6 A shows that the level of prestimulus mu power was correlated with the probability of detecting stimuli in a parabolic way ( p < 0.0006), which was in line with the findings in Linkenkaer-Hansen s paper (2004). The unfilled dots indicate the 20.88 r 65


change of grand average detecti on rate across all thirteen subj ects for each group and the solid curve indicates a quadratic regression fit fo r the data. The horizontal axis indicates the percentage of prestimulus mu power. The vertical bar is the standard error of mean. The unequal relationship was poorly described by a linear regression model ( p = 0.1). 20.32 r Figure 5-6 B shows that the relationship between prestimulus mu power and N1 amplitude from only perceived conditions was better described in a parabolic way ( p < 0.0006) and was poorly descri bed by a linear regression ( p = 0.78). Due to the lack of N1 amplitude in unperceived conditions, only perceived conditions were used in the following. N1 amplitudes in the intermediate power group increased 32% 17% (mean SEM) or 44% 19% (mean SEM) respectively, compared with those in the lo west and highest power groups. 20.88 r 20.01 r 5.3.4 Granger Causality Influences between PFC and S1 The Granger causality spectra were calculated between prefrontal and contralateral S1 electrodes for perceived and unperceived conditions respectively. Figure 5-7 shows the Granger causality spectra between prefrontal and contralateral S1 cortical ar eas for one subjects data set. The Granger causality spectra from PFC to S1 were much enhanced in mu range (peak at 10 Hz, 0.17) for perceived conditions, compared with those (peak at 10 Hz, 0.05) for unperceived conditions. In constrast, the spect ra from S1 to PFC had smaller va lues (< 0.05) and didnt show differences between perceived and unperceived conditions. In addition, the Granger causality values in the beta and gamma bands (20 Hz) were small. Hence, to achieve better comp arisons, the Granger causality sp ectra were averaged in mu band for perceived and unperceived conditions, re spectively. To remove the variability from different subjects, the averaged Granger causality values were normalized to (0, 1), namely, the 66


value (x or y) estimated from perceived or unperc eived trials divided by the summed value (x + y), respectively. The Granger causal influences for perceived conditi ons were significantly stronger (paired t -test, t = 3.89, p < 0.002) than those for unperceived trials (Figure 5-8 A). However, the directional influences from S1 to PFC cortical areas were found to be not significantly different (paired t -test, t = 0.62, p = 0.55) between perceived and unperceived conditions and were also inconsistent across subjec ts (Figure 5-8 B). Six of thirteen subjects had higher causal influences from S1 to PFC fo r perceived conditions whereas the other seven subjects had higher causal influences fr om S1 to PFC for unperceived conditions. Given the evidence of the effect of top-dow n attentional control on the N1 amplitude, the Granger causal influences from PFC to S1 were pr edicted to be positively correlated with the N1 amplitude. Due to the lack of N1 in unperceived tr ials, only perceived trials were used here. The Granger causality analysis was applied in each aforementioned subensemble group. To remove the variability of each individual, Granger causality and N1 amplitude both were normalized to (0, 1). Figure 5-9 shows that the relationship between the directiona l influences from PFC to S1 cortical areas and N1 amplitude was significantly positive (Spearman, = 0.79, p < 0.005). The nine subjects having the inverted-U re lationship were selected. Trials, including perceived and unperceived trials, were sorted by prestimulus mu power. The Granger causality analysis was applied in each subensemble group. Figure 5-10 A show s that the highest directional influences from PFC to contralate ral S1 were significantly (quadratic fit, p < 0.003) correlated with the inte rmediate level of prestimulus mu power. Figure 5-10 B shows an example of the inverted-U relationship (quadratic fit, p < 0.01) estimated from subject 10. In constrast, subject 4 or 13 have positive (linear regression fit, p < 0.001) or negative (linear regression fit, p < 0.03) relationships, re spectively (Figure 5-10 C) 20.81 r 20.73 r 20.75 r 20.48 r 67


and D). These relationships are in line w ith the aforementioned relationships between prestimulus mu power and N1 amplitude. We did several extra studies. First, we investigated the Granger causal influences in higher frequency range (> 16 Hz and < 45 Hz) an d no significant results were found. In the present study, only the consistent results abou t Granger causal influences in mu band from PFC to contralateral S1 cortical areas were found. Second, we changed the prestimulus window to (600 ms, -20 ms) or (-400 ms, -20 ms) and then re calculated the aforementi oned correlations and Granger causality influences. The results for different prestimulus time windows didnt show significant differences. Third, to explore the effect of other cortical areas on contralateral S1, we selected the posterior parietal cortex (PPC), which is important in sensory integration, and investigated the directional infl uences from posterior parietal cortex to contralateral S1. No significant differences ( p = 0.12) between perceived and unperceived conditions were found (Figure 5-11 A). Fourth, we al so investigated the Granger causal influences from PFC to occipital cortex. The results didnt show significant differences ( p = 0.17) between perceived and unperceived conditions (Figure 5-11 B). 5.4 Discussion In our study, we investigat ed the effects of ongoing mu oscillations on behavioral detection of a weak stimulus and stimulus e voked N1 amplitude. An intermediate amount of prestimulus mu power was signifi cantly correlated with enhanced sensory processing to a weak stimulus. We also investigated the impacts of the top-down atte ntional influences on sensory processing and N1 amplitude. We found that the top-down influences in mu band from prefrontal and other frontal structures to S1 were stronge r for perceived trials than those for unperceived trials and was positively correlated with N1 am plitude. These higher orde r frontal structures 68


exert the top-down excitatory or inhibitory influences on posterior sensory cortices and mediate them in the optimal state to sensory input and th us facilitate subsequent sensory processing. 5.4.1 Spontaneous Membrane Potentials and Sensory Processing Scalp EEG signals reflect the oscillations of postsynaptic po tentials of neocortex, which are synchronized across several centimeters (Nunez et al., 2001). Postsynaptic potentials are membrane potential changes of postsynaptic terminals of a synapse. The frequency characteristics of scalp EEG depend on the membrane properties and th eir intrinsic network organization (da Silva, 1991). At a cellular level, both the phase and am plitude of membrane potential oscillations (MPOs) are reported to significantly a ffect sensory processing. Numerous in vitro or in vivo studies demonstrated that spontaneous membrane potential fluctuations, especially the large amplitude oscillations in 5 Hz, carried significant temporal in formation and thus influenced the timing of action potential firing and integrat ion of sensory input (Lampl and Yarom, 1993; Desmaisons et al., 1999; Schaefer et al., 2006). Schaefer et al. ( 2006) showed that sensory input could be robustly perceived when external stim ulus input arrived at the trough and early rising phase of MPOs, which was remarkably in line wi th the scalp EEG findings of Jansen and Brandt (1991). However, the effect of the amplitude of MP Os on sensory processi ng is still in active debate. There are three different hypotheses about the interaction between spontaneous membrane potential fluctuations and sensory input. First, spon taneous MPOs were thought to reflect a depolarizing drive on pr inciple cells (Azouz and Gray 1999; Destexhe and Pare, 1999; McCormick et al., 2003; Shu et al ., 2003; Vertes, 2005). Input-related release of glutamate, when coupled with episodes of spontaneous MPOs, le ads to more vigorous activation. Thus, the 69


absence or very low levels of spontaneous activ ity fails to bring local neuron populations closer to firing threshold and thus results in w eak subsequent sensory evoked responses. In constrast, the excessive level of spont aneous MPOs has also been found to impair sensory processing by competing with sensory evoke d responses (Destexhe et al., 2003; Petersen et al., 2003; Dehaene and Changeux, 2005; Schaef er et al., 2006). The high level of prolonged spontaneous membrane potential fluctuations increases the membrane conductance and thus causes spontaneous firing of action potentials pr ior to synaptic input. The prolonged periods of spontaneous firing preceding syna ptic input result in short-te rm depression of excitatory synapses by depletion of synapt ic vesicles (Abbott et al., 1997 ; Galarreta and Hestrin, 1998; Chung et al., 2002; Petersen, 2002; Zucker and Regehr, 2002). In addition, the activation of a subset of GABAergic interneurons surrounding the excitatory pyramidal neuronal populations enhances the inhibitory synaptic influences on sensory responses (Galarreta and Hestrin, 1998; Markram et al., 1998; Reyes et al., 1998). Furthermore, recent slice and simulation studi es have demonstrated that the moderate amount of spontaneous membrane pot ential fluctuations, not too small or too big, can make local neuron populations more sensitive to synaptic input, especially the small-amplitude input (Destexhe and Pare, 1999; Ho and Destexhe 2000; Destexhe et al., 2001; Rudolph and Destexhe, 2001). The facilitating effect of these spontaneous b ackground activities on stimulus processing is called intrinsic stochastic resonance, which is functionally similar to the stochastic resonance shown to enhance sensory detec tion by adding noise in a nonlinear system (Wiesenfeld and Moss, 1995; Ho and Destexhe, 2000; Stocks and Mannella, 2001). Thus, the relationship between ongoing neur al activity and sensory evoked responses may be not monotonically positive or negative. The intermediate level of the alpha or alpha-like 70


brain rhythms generated in sensor y cortical areas is predicted to optimize stimulus processing to a weak stimulus. It also may explain the afor ementioned inconsistent findings by other EEG research groups. 5.4.2 Somatosensory Evoked Potentials and Sensory Processing Three SEP components-N20, P60, and N140 ar e known to have distinct neural origins and physiological characteristics and interact differently with prestimulus mu oscillations. N20 reflects a population EPSP generated from excitati on of the basal dendrite s of pyramidal neurons in area 3b (Wikstrom et al., 1996). P60 represents an IPSP near the somata of pyramidal cells in area 3b (Wikstrom et al., 1996; Huttunen et al ., 2006). N20 and P60 components were reported to be not significantly correlated with prestimu lus mu power (Nikouline et al., 2000) but varied with stimulus parameters (Lesser et al., 1979; Allison et al., 1992). Thus, N20 and P60 components are mainly exogenously determined components. Somatosensory N1 component, accompanied by high firing rate of action potentials in layers IV-V (Kulics and Cauller, 1986; Cauller and Kulics, 1991), is generated in layer I/II of S1 by the interaction between the f eedforward projections from S1 and the feedback projections from other higher order cortices, including S2, frontal areas, and posteri or parietal cortex (Cauller and Kulics, 1991; Cauller, 1995; Cauller et al., 1998; Jackson and Cauller, 1998; Staines et al., 2002; Golmayo et al., 2003). N1 component is enhanced by both exogenous (e.g., stimulus intensity) and endogenous (e.g., atte ntion) factors (Nakajima and Imamura, 2000). In the present study, N1 amplitude was mainly affected by en dogenous not exogenous factors due to the small intensity of electrical stimuli. Recently, a novel model linking ERP components (> 100 ms) to the N-methyl-Daspartate (NMDA) receptors (Mccarley et al., 1991; da Rocha et al., 2001; Giaquinto and Fraioli, 2003) is supported by the fact that NMDA receptors mediate prolonged sensory evoked 71


responses (Armstrongjames et al., 1993; Salin and Bullier, 1995), whereas non-NMDA receptors (e.g., AMPA) modulate fast responses. NMDA recepto rs, denser in supragranular layer, have stronger influences on cortical f eedback projections (Fox et al., 1989). In addition, it has been demonstrated that spontaneous neural oscill ations in the 8 Hz range were NMDA receptordependent (Silva et al., 1991; F lint and Connors, 1996). Thus, both prestimulus mu oscillations and N1 component are NMDA receptor-dependent, whereas, as aforementioned, N20 and P60 are mostly generated in infragrandular layers. Altogether, it is not surprising that prestimulus mu oscillations significantly affect N1 component, not early evoked responses N20 or P60. 5.4.3 Top-down Attentional Control on Sensory Processing Anatomical evidence and numerous physiologi cal studies (Fuster, 2001) suggest that the prefrontal cortex is the highest level of the sensorimotor hierarchy, which mediates the top-down control by sending modulatory bias signals to lo wer-order circuits and priming sensorimotor information processing. Engel et al. (2001) considered that synchronous neural oscillations, carrying these modulatory bias signals, were im portant in the top-down processing. Summarizing previous experimental and m odeling studies, a recent hypothesi s postulates that the lower frequency synchronization, due to long conduction delays, are more likely acclimated to the large-scale top-down processing (Kopell et al., 2000; von Stein et al., 20 00; Engel et al., 2001; Buschman and Miller, 2007). Recently, a growing number of studies demons trated that mu (or alpha) band neural synchronization in frontal-parietal networks plays a significant role in cognitive attentional tasks by calculating coherence, Granger causality and, phase synchronization (Kaminski et al., 2001; Halgren et al., 2002; Jensen et al., 2002; Hesse et al., 2003; Yamagishi et al., 2003; Serrien et al., 2004; Palva et al., 2005; Sauseng et al., 2005a; Sauseng et al., 2005b). The top-down control can act either before or after a stimulus. In part icular, during anticipation or expectancy, the top72


down factors can significantly enhance subsequent sensory pr ocessing in posterior sensory cortices (Engel and Singer, 2001; Fr ies et al., 2001; Super et al., 2003; Palva et al., 2005). In our present study, the top-down influences, indicated by the Granger causality spectra from frontal to posterior sensory cortical areas, imply that the pref rontal cortex sets up a brain state in posterior sensory areas that facilitates the detection of weak stimuli, further supporting these findings. Somatosensory N1 component is historical ly known to be enhanced by the top-down signals exerted from the anterior attention syst em, such as, prefrontal cortex and anterior cingulate cortex (Posner and Pe tersen, 1990; Waberski et al., 2002). Conversely, neurological patients with prefrontal damage were observed to have attention deficits, which markedly reduce N1 amplitude (Knight, 1997; Chao and Knight, 1998; Knight et al., 1999). Hence, the robust positive relationship between N1 amplitude and the directional influences from frontal areas to S1 suggests that prefrontal a nd other higher order areas send bi as signals to S1 and then modulate spontaneous mu oscillations to the opti mal intermediate level, which is conducive to sensory processing to a weak stimulus. In addition, during ongoing pe riods, the top-down attentional influences have been shown to significantly affect the levels of spontane ous neuronal activation at an attended or nonattended location (Kastner et al ., 1999; Smith et al., 2000). However, there is a discrepancy about the relationship between atten tion and spontaneous neural act ivation (Luck et al., 1997; McAdams and Maunsell, 1999). Luck et al. (1997) reported that when attention was shifted to a particular stimulus location the s pontaneous firing rate increased >30% in the same area. On the other hand, McAdams and Maunsell (1999) report ed the baseline activity didnt change with attention shift. 73


Furthermore, our study examined the rela tionship between the top-down attentional influences and spontaneous neural activation levels in a more de tailed quantitative analysis. Our findings raise the possibility that preceding sens ory input the facilitato ry top-down modulation from prefrontal cortex to primary somatosensory cortex is associated with the optimal level of prestimulus mu oscillations. Figure 5-10 A shows that the intermediate level of ongoing mu oscillations is correlated with the strongest top-down causal influences, whereas the highest or lowest oscillations are correla ted with the weakest top-down influences. This physiological phenomenon supports the hypothesis postulated by Knight and his colleagues (Knight et al., 1999) that prefrontal cortex exerts parallel excitatory and inhi bitory modulations of neural activity over posterior sensory and association cortices. Cortical oscillatory activity is thought to reflect the excitation level of cortical areas (Fries, 2005). The highest level of prestimulus mu oscillations in primary somatosensory cortex suggests the highest excitation level, accompanie d by the most vigorously firing, which is the result of a lack of inhibitory regulation from pre frontal cortex and other fr ontal cortical areas. On the other hand, the lowest level implies that there is a lack of excitatory regulation in S1. Hence, the weak top-down influences result in too much or too small excitati on in posterior sensory cortices and thus reduce the efficacy of stim ulus processing, whereas through the balanced excitatory and inhibitory regul ation from prefrontal cortex the strong top-down influences modulate posterior sensory areas to the intermediate excitation level and thus facilitate stimulus processing. Two remarks are in order. During ongoing period s, the fact that the prefrontal and other frontal structures, not posterior parietal cortex, effectively exert the top-down influences on S1 suggests the prefrontal and other frontal structures are crucial in subsequent sensory processing. 74


Second, the fact that the prefront al cortex has little influences on the occipital cortex suggests that somatosensory perception is enhanced by th e higher top-down influe nces on S1, not simply affected by the increased level of arousal. In summary, our findings suggest that ther e exists an optimal intermediate level of spontaneous neural oscillations (~10 Hz) in poste rior somatosensory cortex for late enhanced sensory responses to a weak stimulus. Prefrontal cortex and other higher-order structures, via ongoing neural oscillations, exert cognitive atte ntional control over posterior somatosensory cortical areas, modulate spontaneous mu oscillati ons in the optimal level, and thus improve behavioral performance and somatosensory N1 amplitude. 75


Mu (7-13 Hz) A B Figure 5-1. Characteristics of spontaneous mu oscillations. A) Examples of spontaneous mu oscillations recorded from two EEG electrodes over contra lateral S1 cortical areas. The length of EEG traces is 8 sec. B) Gra nd average prestimulus power spectra (3-40 Hz) over contralateral S1. 76


A B Figure 5-2. Behavioral performance in a somatosensor y perception task. A) The histogram bars show the probability of detecting a near -threshold electrical stimulus for each individual. The grand average probability ove r all thirteen subjects is 0.51. B) The probability of detecting a stimulus was es timated in each block and averaged across all thirteen subjects. Error bars are standard error of mean. 77


N1 P60 A B 60 ms1.2 v 0 -1.5 v 140 ms Figure 5-3. Characteristics of somatosensory evoked potentials in a somatosensory perception task. A) Grand average SEPs for perceive d (solid curve) an d unperceived (dotted curve) trials in contralateral S1. Positivity is represented as an upward deflection. N1 component was significantly enhanced for perceived trials. B) Scalp current source density maps of grand average SEPs at 60 ms and 140 ms in left hemisphere. Locations of 128-channel EEG recording electrodes are shown by red dots. 78


Figure 5-4. Averaged somatosensory evoked potentia ls for ten power groups. The averaged SEPs (-500 ms, 200 ms) were calculated from one subjects data set. Prestimulus mu power increases from top to bottom. The arrow indicates the peak latencies of N1 component. 79


A B C D Figure 5-5. Regression data fit for all thirteen subjects. A) p value for all quadratic and linear regression fits. The horizontal dotted line indicates the significance level p = 0.05. The filled square indicates a quadratic regression fit and the symbol x means a linear regression fit. The relationship between prestimulus mu power and N1 amplitude is an inverted-U shape (B, quadratic fit, p < 0.006) in subject 10, positive (C, linea r regression fit, p < 0.005) in subject 4, and negative (D, linear regression fit, p < 0.02) in subject 13. 20.76 r 20.64 r 20.45 r 80


A B Figure 5-6. Prestimulus mu oscillations facilitate somatosensory perception. A) An inverted-U relationship (quadratic fit, p < 0.0006) exists be tween prestimulus mu power and the probability of detection rate s. B) The relationship between prestimulus mu power and N1 amplitude is also an inverted-U shape (quadratic fit, p < 0.0006). Only perceived trials were used here Error bars are standard error of mean. 20.88 r 20.88 r 81


B A Figure 5-7. An example of Granger causality sp ectra between prefrontal and contralateral S1 cortical activities. A) Granger causality spec tra from prefrontal to contralateral S1 cortical areas. B) Granger causality spect ra from contralateral S1 to prefrontal cortical areas. Those for perceived conditions were shown in solid curves and those for unperceived conditions in dashed curves. 82


A B Figure 5-8. Granger causality infl uences between prefrontal and c ontralateral S1 cortical areas. Granger causality spectra were averaged in mu frequency range for each individual. A) The histogram shows that the Granger cau sality influences fr om prefrontal to contralateral S1 cortical ac tivity are significantly (paired t -test, p < 0.002) stronger for perceived (black bars) conditions than those for unperceived (grey bars) conditions. b) The directional influences from contrala teral S1 to PFC cortical activity are not significant different (paired t -test, t = 0.62, p = 0.55) between two conditions. 83


Figure 5-9. Relationship between th e Granger causal influences from prefrontal to S1 cortical areas and N1 amplitude. 84


A B C D Figure 5-10. Relationships between prestimulus mu power and the Granger causality influences from prefrontal to contralateral S1 corti cal areas. A) An inverted-U relationship (quadratic fit, p < 0.003) exists between pr estimulus mu power and the causal influences from PFC to contralatera l S1. B) An example of the inverted-U relationship (quadratic fit, p < 0.01) estimated from subject 10. Subject 4 and 13 have positive (C, linear regression fit, p < 0.001) or negative (D, linear regression fit, p < 0.03) relationships, respectively. 20.81 r 20.73 r 20.75 r 20.48 r 85


A B Figure 5-11. Granger causal influences between different cortical areas. A) The Granger causal influences from posterior parietal cortex to contralateral S1 cortical areas are not significantly ( p = 0.12) different between perceived and unperceived conditions. B) Those from prefrontal to occipital cortical areas didnt show si gnificantly differences ( p = 0.17). 86


CHAPTER 6 CONCLUSIONS AND FURTURE RESEARCH 6.1 Conclusion The brain is not a passive or stimulus driven system but an active and adaptive system. According to the intrinsic cortical signals re flecting previous experi ence, anticipation, and prediction, the brain can be modul ated to the optimal state to f acilitate the sensory and motor responses to external stimuli. In the cortical hierarchy, the highe r order cortical areas, such as, prefrontal cortex and anterior cingulate cortex, are dedicated to play this functional role. They generate the intrinsic bias signals and send them to the lower order cortical areas, such as, primary sensory cortices, and further medi ate the ongoing state in primary sensory and association cortices. Such top-down processi ng has been found before and after stimulus presentation in various cognitive atte ntional tasks (Engel et al., 2001). The neural communication am ong different cortical areas is believed to interact effectively through synchronous ne ural oscillations (Engel et al., 2001; Fries, 2005). Fries (2005) proposed that coherent neural osci llations can cause a co mmunication window for sensory input and output in diffe rent cortical areas to open at the same time, whereas weak incoherent neural oscillations lose the timing for input or output. Hence, first, we investigated the effect of ne ural oscillations in different cortical areas on behavioral performance and sensory percepti on. Second, we applied adaptive multivariate autoregressive spectral analysis to investigat e the characteristics of coherence and Granger causality of neural oscillations in sensorimotor integration an d sensory perception. Our main findings are threefold: (1) in monkeys, depe nding on the brain region and frequency band, prestimulus oscillations are e ither significantly negatively corre lated with RT or positively correlated with RT; (2) in monkeys and in humans, synchronized cortical sensorimotor networks 87


are bound together by mu or beta oscillations subserving sustained motor output during hold periods; (3) in humans, during the prestimulus interval, top-down attentional influences exerted from prefrontal and other frontal cortical structures can brin g the posterior sensory neural activities to the optimal level fo r enhanced stimulus processing and behavioral performance. Together, these results demonstrate that ongoing neural oscillations pl ay crucial roles in sensorimotor behavior in both humans and nonhuman primates. 6.2 Future Research The future research has three aspects. In Chapter 5, the scalp EEG experiment was designed to explore the neural co rrelates of somatosensory proce ssing to a weak stimulus. It has a lack of attention control and may be more or less affected by the level of arousal or other cognitive factors even though it was proven that th e arousal level has insi gnificant effects on the EEG task. Thus, to extensively investigate th e top-down attention control on somatosensory processing, well-designed attention conditions where the same stimulus can be either attended to or ignored, need to be added in future EEG experiments. Another facet of my future research is s ource localization of ongoi ng neural oscillations. Nunez et al. (1997) pointed out th at the approach to calculate c oherence on raw scalp EEG data has a few pitfalls, including common reference and volume conduction. The effect of common reference or volume conduction ca n be effectively removed by applying average reference or Laplacian approach, respectively. However, th e Laplacian and other high-resolution methods may underestimate the coherence and Granger caus ality by spatial filter ing (Nunez et al., 1997; Nunez et al., 1999). Thus, to remove the impact of volume conduction effectively, it is better to localize the distinct underlying sources and appl y the aforementioned spectral analysis on the source level. Thus, several source localiz ation algorithms, including minimum-norm and beamforming techniques, will be applied in the future research. 88


Third, the current way to localize the neural activity in a distributed source model (e.g., sLORETA) has methodological difficulties, causing the mislocalization of sources, such as ghost sources and lost sources (Menendez and Andino, 2000). Additionally, some studies showed that these methods have difficulties to locate the d eep sources, such as an terior cingulate cortex (Gomez et al., 2006). One possibility to constr ain the solutions of source localization of EEG/MEG signals is the utiliz ation of fMRI (Ritter and Vill ringer, 2006). Thus, our further study will coregister EEG with fMRI to lo calize the underlying sources more accurately. 89


APPENDIX A MULTITAPER SPECTRAL ANALYSIS The multitaper algorithm is a nonparametric method designed for very short-time data segments. The multitaper spectrum is estimated by averaging multiple windowed faster Fourier transforms generated with multiple orthogonal da ta tapers, particularly discrete prolate spheroidal sequences. Let () x t be a zero mean, stationary random process, 1,2,, tN and are the orthogonal taper functions The Fourier transform of ()twk 1,2,, k K () x t is defined in Equation A-1. The Fourier transform of the data sequences can be obtained through Equation A2 and A-3. 1/2 2 1/2()()iftxtdXfe (A-1) 1/2 2 1/2()()(',)(')N ift j x fxteKffNdXf (A-2) 2(')1sin((')) (',) 2sin(('))iffNNff KffNe ff (A-3) The general multitaper spectrum estimate is: 2 11 ()()K k kSfxf K (A-4) 2 1()()N ift kttxfwkxe (A-5) 90


APPENDIX B SECOND ORDER BLIND IDENTIFICATION Second-order blind identification (SOBI) is a blind source separation method that estimates a joint diagonalizer of a set of cova riance matrix by minimizing joint diagonalization (JD) criterion over the set of covariance matrices in order to separate temporally correlated signals and reduce noise effect. SOBI enables the decomposition of n -channel continuous EEG and MEG signals into m SOBI components ( m n ). Let () x t represent the measured n -channel EEG time series and represent the m unknown underlying sources. () st () x t is an instantaneous linear mixture of source signals via an unknown mixing matrix A. To obtain the estimated source signals an unknown mixing matrix W multiplies () st () st () x t 123()[(),(),(),,()]T n x txtxtxtxt (B-1) 123()[(),(),(),,()]T mststststst (B-2) ()() x tAst (B-3) ()() s tWxt (B-4) 1AW (B-5) The SOBI algorithm has two main steps: data whitening and joint approximate diagonalization. The first step is to whiten the measured data () x t (must be centered) by a whitening matrix P so that: *[()()](0)HHH x H E PxtxtPPRPPAAPI (B-6) The second step is to comput e a set of covariance matrix ()xR and then look for a unitary matrix V to join t diagonalize the set of ()xR by minimizing the JD criterion: 91


2()T ij ijVRV (B-7) After SOBI separation, each SOBI component has one time series and one associated component scalp maps. If is an estimated n x n matrix, the columns of are the component scalp maps. The SOBI separation of EEG here was done with the codes modified from EEGLAB. ()ist A A 92


LIST OF REFERENCES 1. Abbott LF, Varela JA, Sen K, Nelson SB (1997 ) Synaptic depression and cortical gain control. Science 275:220-224. 2. Akaike H (1974) New look at statistical-m odel identification. IEEE Trans on Automat Contr Ac19:716-723. 3. Allison T, Mccarthy G, Wood CC (1992) The relationship between human long-latency somatosensory evoked-potentials recorded from the cortical surface and from the scalp. Electroencephalogr Clin Neurophysiol 84:301-314. 4. Allison T, Mccarthy G, Wood CC, Darcey TM, Spencer DD, Williamson PD (1989) Human cortical potentials-evoked by stimulatio n of the median nerve .1. cytoarchitectonic areas generating short-latency activ ity. J Neurophysiol 62:694-710. 5. Arieli A, Sterkin A, Grinvald A, Aert sen A (1996) Dynamics of ongoing activity: explanation of the large variab ility in evoked cortical re sponses. Science 273:1868-1871. 6. Armstrongjames M, Welker E, Callahan CA (1993) The contribution of NMDA and nonNMDA receptors to fast and slow transmission of sensory information in the rat Sl barrel cortex. J Neurosci 13:2149-2160. 7. Azouz R, Gray CM (1999) Cellular mechanisms contributing to response variability of cortical neurons in vivo. J Neurosci 19:2209-2223. 8. Baker SN, Olivier E, Lemon RN (1997) Cohere nt oscillations in m onkey motor cortex and hand muscle EMG show task-dependent modulation. J Physiol 501 ( Pt 1):225-241. 9. Baker SN, Kilner JM, Pinches EM, Lemon RN (1999) The role of synchrony and oscillations in the motor out put. Exp Brain Res 128:109-117. 10. Belouchrani A, AbedMeraim K, Cardoso JF, Moulines E (1997) A blind source separation technique using second-order statistics. IEEE Trans on Signal Proces 45:434-444. 11. Berger H (1929) Electroencephalogram in humans. Arch Psychiatr Nervenkr 87:527-570. 12. Bernier R, Dawson G, Webb S, Murias M (2007) EEG mu rhythm and imitation impairments in individuals with autism spectrum disorder. Brain Cogn 64:228-237. 13. Bland BH, Konopacki J, Dyck RH (2002) Relationship between membrane potential oscillations and rhythmic discharges in identified hippocampal theta-related cells. J Neurophysiol 88:3046-3066. 14. Brandt ME, Jansen BH, Carbonari JP (1991) Pre-stimulus spectral EEG patterns and the visual evoked-response. Electroence phalogr Clin Neurophysiol 80:16-20. 93


15. Brovelli A, Ding M, Ledberg A, Chen Y, Nakamura R, Bressler SL (2004) Beta oscillations in a large-scale sensorimotor cort ical network: directional influences revealed by Granger causality. Proc Natl Acad Sci U S A 101:9849-9854. 16. Brown P (2000) Cortical drives to human muscle: the Piper and related rhythms. Prog Neurobiol 60:97-108. 17. Brunia CH (1999) Neural aspects of anticip atory behavior. Acta Psychol (Amst) 101:213242. 18. Buschman TJ, Miller EK (2007) Top-down vers us bottom-up control of attention in the prefrontal and posterior parietal cortices. Science 315:1860-1862. 19. Cauller L (1995) Layer I of primary sensor y neocortex: where t op-down converges upon bottom-up. Behav Brain Res 71:163-170. 20. Cauller LJ, Kulics AT (1991) The neural basis of the behaviorally relevant N1 component of the somatosensory-evoked potential in SI cortex of awake m onkeys evidence that backward cortical projections signal consci ous touch sensation. Exp Brain Res 84:607-619. 21. Cauller LJ, Clancy B, Connors BW (1998) B ackward cortical projections to primary somatosensory cortex in rats extend long hori zontal axons in layer I. J Comp Neurol 390:297-310. 22. Chao LL, Knight RT (1998) Contribution of hu man prefrontal cortex to delay performance. J Cogn Neurosci 10:167-177. 23. Chatrian GE, Petersen MC, Lazarte JA (1959) The blocking of the rolandic wicket rhythm and some central changes related to move ment. Electroencephalogr Clin Neurophysiol 11:497-510. 24. Chen R, Corwell B, Hallett M (1999) Modulation of motor cortex excitability by median nerve and digit stimulation. Exp Brain Res 129:77-86. 25. Chen Y, Bressler SL, Ding M (2006) Frequency decomposition of conditional Granger causality and application to multivariate neural field potential data. J Neurosci Methods 150:228-237. 26. Chung S, Li XR, Nelson SB (2002) Short-te rm depression at thalamocortical synapses contributes to rapid adaptation of cortical sensory response s in vivo. Neuron 34:437-446. 27. Conway BA, Halliday DM, Farmer SF, Shahani U, Maas P, Weir AI, Rosenberg JR (1995) Synchronization between motor cortex a nd spinal motoneuronal pool during the performance of a maintained motor task in man. J Physiol (Lond) 489:917-924. 28. da Rocha AF, Pereira A, Coutinho FAB (2 001) N-methyl-D-aspartate channel and consciousness: from signal coincidence dete ction to quantum computing. Prog Neurobiol 64:555-573. 94


29. da Silva FHL (1990) A critical-review of clin -applications of topographic mapping of brain potentials. J Clin Neurophysiol 7:535-551. 30. da Silva FL (1991) Neural mechanisms unde rlying brain waves from neural membranes to networks. Electroencephalogr Clin Neurophysiol 79:81-93. 31. Dehaene S, Changeux JP (2005) Ongoing spontaneous activity controls access to consciousness: A neuronal model for inatte ntional blindness. Plos Biol 3:910-927. 32. Desmaisons D, Vincent JD, Lledo PM (1999) Control of action potential timing by intrinsic subthreshold oscillations in olfactory bulb output neurons. J Neurosci 19:10727-10737. 33. Desmedt JE, Robertson D (1977) Differential en hancement of early and late components of cerebral somatosensory evoked-potentials during forced-paced cognitive tasks in man. J Physiol (Lond) 271:761-782. 34. Desmedt JE, Tomberg C (1989) Mapping ear ly somatosensory evoked-potentials in selective attention critical-evaluation of control conditions used for titrating by difference the cognitive P-30, P40, P100 and N140. Electroencephalogr Clin Neurophysiol 74:321346. 35. Desmedt JE, Huy NT, Bourguet M (1983) The cognitive P40, N60 and P100 components of somatosensory evoked potentials and the ea rliest electrical signs of sensory processing in man. Electroencephalogr Clin Neurophysiol 56:272-282. 36. Destexhe A, Pare D (1999) Impact of netw ork activity on the integrative properties of neocortical pyramidal neurons in vivo. J Neurophysiol 81:1531-1547. 37. Destexhe A, Rudolph M, Pare D (2003) The high-conductance state of neocortical neurons in vivo. Nat Rev Neurosci 4:739-751. 38. Destexhe A, Rudolph M, Fellous JM, Sej nowski TJ (2001) Fluctuating synaptic conductances recreate in vivo-li ke activity in neocortical ne urons. Neuroscience 107:13-24. 39. Ding M, Chen Y, Bressler SL (2006) Granger causality: basic theory and application to neuroscience. In: Handbook of time series an alysis (B. Schelter MW, and J. Timmer, ed), pp 437-460. Weinheim: Wiley-VCH Verlag. 40. Ding MZ, Bressler SL, Yang WM Liang HL (2000) Short-wi ndow spectral analysis of cortical event-related potentials by adaptive multivariate autoregressive modeling: data preprocessing, model validation, and variab ility assessment. Biol Cybern 83:35-45. 41. Donchin E, Lindsley DB (1966) Average evoked potentials and reaction times to visual stimuli. Electroencephalogr Clin Neurophysiol 20:217-223. 42. Donoghue JP, Sanes JN, Hatsopoulos NG, Gaal G (1998) Neural disc harge and local field potential oscillations in prim ate motor cortex during volunt ary movements. J Neurophysiol 79:159-173. 95


43. Eckhorn R, Bauer R, Jordan W, Brosch M, Kruse W, Munk M, Reitboeck HJ (1988) Coherent oscillations: a mechanism of featur e linking in the visu al cortex? Multiple electrode and correlation analyses in the cat. Biol Cybern 60:121-130. 44. Engel AK, Singer W (2001) Temporal bindi ng and the neural correlates of sensory awareness. Trends Cogn Sci 5:16-25. 45. Engel AK, Fries P, Singer W (2001) Dynamic predictions: oscillations and synchrony in top-down processing. Nat Rev Neurosci 2:704-716. 46. Flint AC, Connors BW (1996) Two types of ne twork oscillations in neocortex mediated by distinct glutamate receptor subtypes and ne uronal populations. J Neurophysiol 75:951-956. 47. Fox K, Sato H, Daw N (1989) The location and function of NMDA receptors in cat and kitten visual-cortex. J Neurosci 9:2443-2454. 48. Foxe JJ, Simpson GV, Ahlfors SP (1998) Pari eto-occipital approximately 10 Hz activity reflects anticipatory state of visual atte ntion mechanisms. Neuroreport 9:3929-3933. 49. Fries P (2005) A mechanism for cognitive dynamics: neuronal communication through neuronal coherence. Trends Cogn Sci 9:474-480. 50. Fries P, Neuenschwander S, Engel AK, Goebel R, Singer W (2001) Rapid feature selective neuronal synchronization through correlated latency shifti ng. Nat Neurosci 4:194-200. 51. Fuster JM (2001) The prefrontal cortex--an updat e: time is of the essence. Neuron 30:319333. 52. Galarreta M, Hestrin S (1998) Frequency-depende nt synaptic depression and the balance of excitation and inhibition in the neocortex. Nat Neurosci 1:587-594. 53. Garcia-Larrea L, Lukaszewicz AC, Mauguiere F (1995) Somatosensory responses during selective spatial attention the N120-to-N140 transition. Psychophysiology 32:526-537. 54. Gastaut HJ, Bert J (1954) EEG changes duri ng cinematographic presentation (Moving picture activation of the EEG). Electroencephalogr C lin Neurophysiol 6:433-444. 55. Geweke J (1982) Measurement of linear dependence and feedback from multiple time series. J Am Stat Assoc 77:304-313. 56. Giaquinto S, Fraioli L (2003) Enhancement of the somatosensory N140 component during attentional training after stroke. Clin Neurophysiol 114:329-335. 57. Golmayo L, Nunez A, Zaborszky L (2003) Electrophysiological evidence for the existence of a posterior cortical-prefrontal-basal forebr ain circuitry in modula ting sensory responses in visual and somatosensory rat cort ical areas. Neuroscience 119:597-609. 96


58. Gomez CM, Marco-Pallares J, Grau C (20 06) Location of brain rhythms and their modulation by preparatory attention estimated by current density. Brain Res 1107:151-160. 59. Gonzalez Andino SL, Michel CM, Thut G, Landis T, Grave de Peralta R (2005) Prediction of response speed by anticipatory high-freque ncy (gamma band) osci llations in the human brain. Hum Brain Mapp 24:50-58. 60. Granger CWJ (1969) Investigating causal relations by econmetric models and crossspectral methods. Econometrica 37:424-438 61. Gray CM, Singer W (1989) Stimulus-specific ne uronal oscillations in orientation columns of cat visual cortex. Proc Na tl Acad Sci U S A 86:1698-1702. 62. Gross J, Tass PA, Salenius S, Hari R, Fr eund HJ, Schnitzler A (2000) Cortico-muscular synchronization during isometric muscle contraction in humans as revealed by magnetoencephalography. J Physiol 527 Pt 3:623-631. 63. Haig AR, Gordon E (1998) Prestimulus EEG alpha phase synchronicity influences N100 amplitude and reaction time. Psychophysiology 35:591-595. 64. Halgren E, Boujon C, Clarke J, Wang C ( 2002) Rapid distributed fronto-parieto-occipital processing stages during working memory in humans. Cereb Cortex 12:710-728. 65. Halliday DM, Conway BA, Farmer SF, Rosenberg JR (1998) Using electroencephalography to study functional coupling betw een cortical activity and electromyograms during voluntary contractio ns in humans. Neurosci Lett 241:5-8. 66. Hesse W, Moller E, Arnold M, Schack B (2003) The use of time-variant EEG Granger causality for inspecting directed interdependencies of neural assemblies. J Neurosci Methods 124:27-44. 67. Ho N, Destexhe A (2000) Synaptic background activity enhances th e responsiveness of neocortical pyramidal neurons. J Neurophysiol 84:1488-1496. 68. Homan RW, Herman J, Purdy P (1987) Cerebr al location of intern ational 10-20 system electrode placement. Electroencephalogr Clin Neurophysiol 66:376-382. 69. Houdayer E, Labyt E, Cassim F, Bourriez JL, Derambure P (2006) Relationship between event-related beta synchronization and afferent inputs: analysis of finger movement and peripheral nerve stimulations. Cl in Neurophysiol 117:628-636. 70. Huttunen J, Komssi S, Lauronen L (2006) Spat ial dynamics of popula tion activities at S1 after median and ulnar nerve stimulation revi sited: an MEG study. Neuroimage 32:10241031. 71. Jackson ME, Cauller LJ (1998) Neural activity in SII modifies sensory evoked potentials in SI in awake rats. Neuroreport 9:3379-3382. 97


72. Jensen O, Gelfand J, Kounios J, Lisman JE (2002) Oscillations in the alpha band (9 Hz) increase with memory load during retention in a short-term memory task. Cereb Cortex 12:877-882. 73. Kaminski M, Ding MZ, Truccolo WA, Bressler SL (2001) Evaluating causal relations in neural systems: Granger causality, directed tr ansfer function and statistical assessment of significance. Biol Cybern 85:145-157. 74. Kastner S, Pinsk MA, De Weerd P, Desimone R, Ungerleider LG (1999) Increased activity in human visual cortex during directed atten tion in the absence of visual stimulation. Neuron 22:751-761. 75. Kekoni J, Hamalainen H, Saarinen M, Grohn J, Reinikainen K, Lehtokoski A, Naatanen R (1997) Rate effect and mismatch responses in the somatosensory system: ERP-recordings in humans. Biol Psychol 46:125-142. 76. Kida T, Nishihira Y, Wasaka T, Nakata H, Sakamoto M (2004) Differential modulation of temporal and frontal components of the somatosensory N140 and the effect of interstimulus interval in a selective attention task. Cogn Brain Res 19:33-39. 77. Kilner JM, Baker SN, Salenius S, Hari R, Lemon RN (2000) Human cortical muscle coherence is directly related to specific motor para meters. J Neurosci 20:8838-8845. 78. Kilner JM, Salenius S, Baker SN, Jackson A, Hari R, Lemon RN (2003) Task-dependent modulations of cortical oscillatory activit y in human subjects during a bimanual precision grip task. Neuroimage 18:67-73. 79. Klimesch W (1999) EEG alpha and theta os cillations reflect cognitive and memory performance: a review and analysis. Brain Res Rev 29:169-195. 80. Klimesch W, Sauseng P, Hanslmayr S (2007) EEG alpha oscillations: the inhibition-timing hypothesis. Brain Res Rev 53:63-88. 81. Klimesch W, Doppelmayr M, Schwaiger J, Au inger P, Winkler T (1999) 'Paradoxical' alpha synchronization in a memory task. Cogn Brain Res 7:493-501. 82. Knight RT (1997) Distributed cortical networ k for visual attention. J Cogn Neurosci 9:7591. 83. Knight RT, Staines WR, Swick D, Chao LL (1 999) Prefrontal cortex regulates inhibition and excitation in distributed neural netw orks. Acta Psychol (Amst) 101:159-178. 84. Kopell N, Ermentrout GB, Whittington MA, Traub RD (2000) Gamma rhythms and beta rhythms have different synchronization pr operties. Proc Natl Acad Sci U S A 97:18671872. 85. Kraft GH, Aminoff MJ, Baran EM, Litchy WJ, Stolov WC, subcomm ASEP (1998) Somatosensory evoked potentials: Clin uses. Muscle Nerve 21:252-258. 98


86. Kuhlman WN (1978) EEG feedback training enhancement of somatosensory cortical activity. Electroencephalogr C lin Neurophysiol 45:290-294. 87. Kuhn AA, Williams D, Kupsch A, Limousin P, Hariz M, Schneider GH, Yarrow K, Brown P (2004) Event-related beta desynchronization in human subthalamic nucleus correlates with motor performance. Brain 127:735-746. 88. Kulics AT, Cauller LJ (1986) Cerebral corti cal somatosensory evoked-responses, multiple unit-activity and current source-densities th eir interrelationships and significance to somatic sensation as revealed by stimulatio n of the awake monkeys hand. Exp Brain Res 62:46-60. 89. Lamme VAF, Roelfsema PR (2000) The distinct modes of vision offered by feedforward and recurrent processing. Tr ends Neurosci 23:571-579. 90. Lampl I, Yarom Y (1993) Subthreshold os cillations of the membrane-potential a functional synchronizing and timing device. J Neurophysiol 70:2181-2186. 91. Ledberg A, Bressler SL, Ding MZ, Coppol a R, Nakamura R (2007) Large-scale visuomotor integration in the cerebr al cortex. Cereb Cortex 17:44-62. 92. Leek MR (2001) Adaptive procedures in psychophysical research. Percept & Psychophys 63:1279-1292. 93. Lesser RP, Koehle R, Lueders H (1979) Eff ect of stimulus-intensity on short latency somatosensory evoked-potentials. Electro encephalogr Clin Neurophysiol 47:377-382. 94. 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. 95. Libet B, Alberts WW, Wright EW, Feinstei.B (1967) Responses of human somatosensory cortex to stimuli below threshold for co nscious sensation. Science 158:1597-1600. 96. Linkenkaer-Hansen K, Nikulin VV, Palva S, Ilmoniemi RJ, Palva JM (2004) Prestimulus oscillations enhance psychophysical performa nce in humans. J Neurosci 24:10186-10190. 97. Luck SJ, Chelazzi L, Hillyard SA, Desimone R (1997) Neural mechanisms of spatial selective attention in areas V1, V2, and V4 of macaque visual cortex. J Neurophysiol 77:24-42. 98. MacKay WA, Mendonca AJ (1995) Field potential oscillatory bursts in parietal cortex before and during reach. Brain Res 704:167-174. 99. Makeig S, Westerfield M, J ung TP, Enghoff S, Townsend J, Courchesne E, Sejnowski TJ (2002) Dynamic brain sources of visual evoked responses. Science 295:690-694. 100. Markram H, Wang Y, Tsodyks M (1998) Di fferential signaling via the same axon of neocortical pyramidal neurons. Proc Natl Acad Sci U S A 95:5323-5328. 99

PAGE 100

101. McAdams CJ, Maunsell JHR (1999) Effects of a ttention on orientation-tuning functions of single neurons in macaque cortical area V4. J Neurosci 19:431-441. 102. Mccarley RW, Faux SF, Shenton ME, Nestor PG Adams J (1991) Even t-related potentials in schizophrenia their biolog ical and clin correlates and a new model of schizophrenic Pathophysiology. Schizophr Res 4:209-231. 103. McCormick DA, Shu YS, Hasenstaub A, Sanchez-Vives M, Badoual M, Bal T (2003) Persistent cortical activity: mechanisms of generation and effects on neuronal excitability. Cereb Cortex 13:1219-1231. 104. Meador KJ, Ray PG, Echauz JR, Loring DW, Vachtsevanos GJ (2002) Gamma coherence and conscious perception. Neurology 59:847-854. 105. Menendez RGD, Andino SLG (2000) Discussing the capabilities of Laplacian Minimization. Brain Topogr 13:97-104. 106. Miller EK, Cohen JD (2001) An integrative theory of prefront al cortex function. Annu Rev Neurosci 24:167-202. 107. Miltner WH, Braun C, Arnold M, Witte H, Taub E (1999) Coherence of gamma-band EEG activity as a basis for associativ e learning. Nature 397:434-436. 108. Mima T, Matsuoka T, Hallett M (2000) Functional coupling of human right and left cortical motor areas demonstrated with par tial coherence analysis. Neurosci Lett 287:9396. 109. Mima T, Simpkins N, Oluwatimilehin T, Ha llett M (1999) Force level modulates human cortical oscillatory activities. Neurosci Lett 275:77-80. 110. Morf M, Vieira A, Lee DTL, Kailath T (1 978) Recursive multichannel maximum entropy spectral estimation. IEEE Trans Geosci and Remote 16:85-94. 111. Morrell LK, Morrell F (1966) Evoked potentials and reaction times: a study of intraindividual variability. Electroencephalogr Clin Neurophysiol 20:567-575. 112. Murthy VN, Fetz EE (1992) Coherent 25to 35 -Hz oscillations in the sensorimotor cortex of awake behaving monkeys. Proc Natl Acad Sci U S A 89:5670-5674. 113. Murthy VN, Fetz EE (1996) Oscillatory activity in sensorimotor cortex of awake monkeys: synchronization of local field potentials an d relation to behavior. J Neurophysiol 76:39493967. 114. Nakajima Y, Imamura N (2000) Relationships between attention effects and intensity effects on the cognitive N140 and P300 components of somatosensory ERPs. Clin Neurophysiol 111:1711-1718. 100

PAGE 101

115. Nichols TE, Holmes AP (2002) Nonparam etric permutation tests for functional neuroimaging: a primer with examples. Hum Brain Mapp 15:1-25. 116. Nikouline VV, Wikstrom H, Linkenkaer-Hansen K, Kesaniemi M, Ilmoniemi RJ, Huttunen J (2000) Somatosensory evoked magnetic fields : relation to pre-stimulus mu rhythm. Clin Neurophysiol 111:1227-1233. 117. Nunez A, Garciaustt E, Buno W (1987) Intrace llular theta-rhythm generation in identified hippocampal pyramids. Brain Res 416:289-300. 118. Nunez PL, Wingeier BM, Silberstein RB (2001) Spatial-temporal structures of human alpha rhythms: theory, microcurrent sources, multiscale measurements, and global binding of local networks. Hum Brain Mapp 13:125-164. 119. Nunez PL, Srinivasan R, Westdorp AF, Wijesinghe RS, Tucker DM, Silberstein RB, Cadusch PJ (1997) EEG coherency. I: statis tics, reference electrode, volume conduction, Laplacians, cortical imaging, and interpretation at multiple scales. Electroencephalogr Clin Neurophysiol 103:499-515. 120. Nunez PL, Silberstein RB, Shi Z, Carpenter MR, Srinivasan R, Tucker DM, Doran SM, Cadusch PJ, Wijesinghe RS (1999) EEG cohe rency II: experiment al comparisons of multiple measures. Clin Neurophysiol 110:469-486. 121. Oberman LM, Hubbard EM, McCleery JP, Alts chuler EL, Ramachandran VS, Pineda JA (2005) EEG evidence for mirror neuron dysfunction in autism spectrum disorders. Cogn Brain Res 24:190-198. 122. Palva S, Palva JM (2007) New vistas for alpha-frequency band oscillations. Trends Neurosci 30:150-158. 123. Palva S, Linkenkaer-Hansen K, Naatanen R, Palva JM (2005) Early neural correlates of conscious somatosensory perceptio n. J Neurosci 25:5248-5258. 124. Perrin F, Bertrand O, Pernier J (1987) Scalp current-density mapping value and estimation from potential data. IEEE Trans Biomed Eng 34:283-288. 125. Petersen CCH (2002) Short-term dynamics of sy naptic transmission wi thin the excitatory neuronal network of rat layer 4 ba rrel cortex. J Neurophysiol 87:2904-2914. 126. Petersen CCH, Hahn TTG, Mehta M, Grin vald A, Sakmann B (2003) Interaction of sensory responses with spontaneous depolariza tion in layer 2/3 barrel cortex. Proc Natl Acad Sci U S A 100:13638-13643. 127. Pfurtscheller G, Stancak A, Jr., Neuper C ( 1996a) Post-movement beta synchronization. A correlate of an idling motor area? Electroencephalogr Clin Neurophysiol 98:281-293. 101

PAGE 102

128. Pfurtscheller G, Stancak A, Jr., Neuper C ( 1996b) Event-related synchronization (ERS) in the alpha band--an electrophysiological correla te of cortical idling: a review. Int J Psychophysiol 24:39-46. 129. Pfurtscheller G, Stancak A, Jr., Edlinger G (1997a) On the existence of different types of central beta rhythms below 30 Hz. Electroencephalogr Clin Neurophysiol 102:316-325. 130. Pfurtscheller G, Neuper C, Andrew C, Edlinge r G (1997b) Foot and hand area mu rhythms. Int J Psychophysiol 26:121-135. 131. Pineda JA (2005) The functional significan ce of mu rhythms: tr anslating "seeing" and "hearing" into "doing". Brai n Res Brain Res Rev 50:57-68. 132. Pineda JA, Allison BZ, Vankov A (2000) The effects of self-movement, observation, and imagination on mu rhythms and readiness pot entials (RP's): toward a brain-computer interface (BCI). IEEE Trans Rehabil Eng 8:219-222. 133. Pineda JOA, Oberman LM (2006) What goa ds cigarette smokers to smoke? Neural adaptation and the mirror neuron system. Brain Res 1121:128-135. 134. Ploner M, Gross J, Timmermann L, Pollok B, Schnitzler A (2006) Oscillatory activity reflects the excitability of the human so matosensory system. Neuroimage 32:1231-1236. 135. Pollok B, Gross J, Muller K, Aschersleben G, Schnitzler A (2005) The cerebral oscillatory network associated with auditorily paced finger movements. Neuroimage 24:646-655. 136. Posner MI, Petersen SE (1990) The Attent ion System of the Human Brain. Annu Rev Neurosci 13:25-42. 137. Rahn E, Basar E (1993) Enhancement of Visu al-Evoked Potentials by Stimulation during Low Prestimulus EEG Stages. Int J Neurosci 72:123-136. 138. Ray PG, Meador KJ, Smith JR, Wheless JW, Sittenfeld M, Clifton GL (1999) Physiology of perception Cortical stim ulation and recording in hum ans. Neurology 52:1044-1049. 139. Ray WJ, Cole HW (1985) EEG alpha activity reflects attentional demands, and beta activity reflects emotional and cogniti ve processes. Science 228:750-752. 140. Reyes A, Lujan R, Rozov A, Burnashev N, Somogyi P, Sakmann B (1998) Target-cellspecific facilitation and depression in ne ocortical circuits. Nat Neurosci 1:279-285. 141. Reynolds JH, Pasternak T, Desimone R (2000) Attention increases sensitivity of V4 neurons. Neuron 26:703-714. 142. Ritter P, Villringer A (2006) Simultaneous EEG-fMRI. Neurosci and Biobehav Rev 30:823-838. 102

PAGE 103

143. Rodriguez E, George N, Lachaux JP, Mart inerie J, Renault B, Varela FJ (1999) Perception's shadow: long-distance synchroni zation of human brain activity. Nature 397:430-433. 144. Rudolph M, Destexhe A (2001) Do neocorti cal pyramidal neurons display stochastic resonance? J Comput Neurosci 11:19-42. 145. Salenius S, Schnitzler A, Salmelin R, Jous maki V, Hari R (1997) Modulation of human cortical rolandic rhythms du ring natural sensorimotor ta sks. Neuroimage 5:221-228. 146. Salin PA, Bullier J (1995) Corticocortical conn ections in the visual-s ystem structure and function. Physiol Rev 75:107-154. 147. Salmelin R, Hari R (1994) Spatiotemporal characteristics of sensorimotor neuromagnetic rhythms related to thumb moveme nt. Neuroscience 60:537-550. 148. Salmelin R, Hamalainen M, Kajola M, Hari R (1995) Functional segregation of movementrelated rhythmic activity in the hu man brain. Neuroimage 2:237-243. 149. Sanes JN, Donoghue JP (1993) Oscillations in local field potentials of the primate motor cortex during voluntary movement. Pr oc Natl Acad Sci U S A 90:4470-4474. 150. Sauseng P, Klimesch W, Doppelmayr M, Pecherstorfer T, Freunberger R, Hanslmayr S (2005a) EEG alpha synchronization and functio nal coupling during top-down processing in a working memory task. Hum Brain Mapp 26:148-155. 151. Sauseng P, Klimesch W, Stadler W, Scha bus M, Doppelmayr M, Hanslmayr S, Gruber WR, Birbaumer N (2005b) A shift of visual spa tial attention is selectively associated with human EEG alpha activity. Eur J Neurosci 22:2917-2926. 152. Schaefer AT, Angelo K, Spors H, Margrie TW (2006) Neuronal oscillations enhance stimulus discrimination by ensuring action po tential precision. Pl os Biol 4:1010-1024. 153. Schubert R, Blankenburg F, Lemm S, Villrin ger A, Curio G (2006) Now you feel it now you don't: ERP correlates of somatosensor y awareness. Psychophysiology 43:31-40. 154. Serrien DJ, Pogosyan AH, Cassidy MJ, Brown P (2004) Anticipatory cortico-cortical interactions: switching the task configurati on between effectors. Exp Brain Res 154:359367. 155. Shu YS, Hasenstaub A, Badoual M, Bal T, McCormick DA (2003) Barrages of synaptic activity control the gain and se nsitivity of cortical neuron s. J Neurosci 23:10388-10401. 156. Silva LR, Amitai Y, Connors BW (1991) Intrin sic oscillations of neocortex generated by layer-5 pyramidal neurons. Science 251:432-435. 103

PAGE 104

157. Simoes C, Salenius S, Curio G (2004) Shor t-term (approximate to 600 ms) prediction of perturbation dynamics for 10-and 20-Hz ME G rhythms in human primary sensorimotor hand cortices. Neuroimage 22:387-393. 158. Smith AT, Singh KD, Greenlee MW (2000) A ttentional suppression of activity in the human visual cortex. Neuroreport 11:271-277. 159. Staines WR, Graham SJ, Black SE, McIlr oy WE (2002) Task-relevant modulation of contralateral and ipsilateral primary somatose nsory cortex and the role of a prefrontalcortical sensory gating syst em. Neuroimage 15:190-199. 160. Stocks NG, Mannella R (2001) Generic noiseenhanced coding in neuronal arrays. Phys Rev E Stat Nonlin Soft Ma tter Phys 64:030902(1 4). 161. Super H, van der Togt C, Spekreijse H, Lamme VAF (2003) Internal state of monkey primary visual cortex (V1) predicts figureground perception. J Neurosci 23:3407-3414. 162. Taylor-Clarke M, Kennett S, Haggard P (2002) Vision modulates somatosensory cortical processing. Curr Biol 12:233-236. 163. Thomson DJ (1982) Spectrum estimation and harmonic-analysis. Proc IEEE 70:1055-1096. 164. Tomberg C, Desmedt JE (1996) Non-averaged hu man brain potentials in somatic attention: the short-latency cognition-related P-40 component. J Physiol (Lond) 496:559-574. 165. Tomberg C, Desmedt JE, Ozaki I, Nguyen TH, Chalklin V (1989) Mapping somatosensory evoked-potentials to finger stimulation at intervals of 450 to 4000 msec and the issue of habituation when assessing early cognitiv e components. Electroencephalogr Clin Neurophysiol 74:347-358. 166. Tsodyks M, Kenet T, Grinvald A, Arieli A (19 99) Linking spontaneous activity of single cortical neurons and the underlying func tional architecture. Science 286:1943-1946. 167. Vanessen DC, Gallant JL (1994) Neural mechan isms of form and motion processing in the primate visual-system. Neuron 13:1-10. 168. Vanrotterdam A, Dasilva FHL, Vandenende J, Viergever MA, Hermans AJ (1982) A model of the spatial-temporal characteristics of the alpha -rhythm. Bull Math Biol 44:283305. 169. Varela F, Lachaux JP, Rodriguez E, Ma rtinerie J (2001) The brainweb: phase synchronization and large-scale integr ation. Nat Rev Neurosci 2:229-239. 170. Vertes RP (2005) Hippocampal theta rhythm : a tag for short-term memory. Hippocampus 15:923-935. 171. von Stein A, Chiang C, Konig P (2000) T op-down processing medi ated by interareal synchronization. Proc Natl Acad Sci U S A 97:14748-14753. 104

PAGE 105

172. Waberski TD, Gobbele R, Darvas F, Schmitz S, Buchner H (2002) Spatiotemporal imaging of electrical activity related to attention to somatosensory stimulation. Neuroimage 17:1347-1357. 173. Wiesenfeld K, Moss F (1995) Stochastic resona nce and the benefits of noise from ice ages to crayfish and squids. Nature 373:33-36. 174. Wiest MC, Nicolelis MAL (2003) Behavioral detection of t actile stimuli during 7 Hz cortical oscillations in awake rats. Nat Neurosci 6:913-914. 175. Wikstrom H, Huttunen J, Korvenoja A, Virtan en J, Salonen O, Aronen H, Ilmoniemi RJ (1996) Effects of interstimulus interval on somatosensory evoked magnetic fields (SEFs): a hypothesis concerning SEF generation at the primary sensorimotor cortex. Evoked Potentials-Electroencephalogr Clin Neurophysiol 100:479-487. 176. Winterer G, Ziller M, Dorn H, Frick K, Mu lert C, Dahhan N, Herrmann WM, Coppola R (1999) Cortical activation, signal-to-noise ratio and stochastic resonance during information processing in man. Clin Neurophysiol 110:1193-1203. 177. Yamagishi N, Callan DE, Goda N, Anderson SJ Yoshida Y, Kawato M (2003) Attentional modulation of oscillatory activity in huma n visual cortex. Neuroimage 20:98-113. 178. Zucker RS, Regehr WG (2002) Short-term s ynaptic plasticity. Annu Rev Physiol 64:355405. 105

PAGE 106

BIOGRAPHAICAL SKETCH Yan Zhang received her Bachelor of Engineer ing from Xian Institute of Technology in 1996. After receiving top academic records through examination, she became a graduate student in the Department of Electronics in the College of Information Science at the Beijing Normal University, majoring in nonlinear system and data processing. After completing her masters degree in 1999, she stayed at the Beijing Normal University and worked as a college teacher and a researcher. In 2003, she became a PhD student in Center for Complex Systems and Brain Sciences at the Florida Atlantic University. In 2004, she was awarded an Alumni Fellowship for graduate study in biomedical e ngineering at University of Flor ida. Her research interests are biomedical signal processing, sensory perception, attention control and human motor control. She completed a Doctorate in Philosophy in Augu st of 2008 specializing in neuroinformatics. Yan Zhang is currently a member of Society of Neuroscience and Institu te of Electrical and Electronics Engineers. 106