Citation
Network Analysis of Neural Activity

Material Information

Title:
Network Analysis of Neural Activity
Creator:
Kang, Daesung
Publisher:
University of Florida
Publication Date:
Language:
English

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Biomedical Engineering
Committee Chair:
DING,MINGZHOU
Committee Co-Chair:
YANG,LIN
Committee Members:
GUNDUZ,AYSEGUL
KEIL,ANDREAS

Subjects

Subjects / Keywords:
causality
granger

Notes

General Note:
Connectivity in the brain that can be categorized according to a pattern of anatomical connections is called anatomical connectivity, that is defined according to statistical dependencies is called functional connectivity, and that can give information on causal interactions is called effective connectivity. While anatomical connectivity refers to physical connections between neurons or neuronal elements, functional connectivity captures deviations from statistical independence between neural units. Effective connectivity can describe directional effects of one neural unit over another. Neural activities, which were generated by neural units, can be measured by various modalities such as single neuron spike trains, local field potentials (LFP), electroencephalogram (EEG), or functional magnetic resonance imaging (fMRI) etc. and those neural activities can be used to infer connectivity and to elucidate how neural units and network of neurons process information. In this dissertation I examine the analysis of functional and effective connectivity from single neuron spike trains, LFPs, and fMRI. Granger causality is a new technique to measure effective connectivity and to determine the strength and direction of relationships between bivariate or multivariate signals. Although GC between continuous time series or GC between point process data are mathematically well-defined, GC for mixed time series, such as one being local field potential and the other being spike train data, is not well-understood. For the analysis of this type of data, we introduce the non-parametric GC method. With the method, we examine the interaction patterns between medial septum (MS) and hippocampus in slow-wave sleep (SWS) and microarousal. Then we extend our interests from SWS/microarousal to other theta and non-theta states including quiet waking and REM sleep, and infer the effective connectivity and temporal timing relationship between MS and hippocampus. Functional magnetic resonance imaging (fMRI) has become a powerful imaging modality for studying human brain function. Blocked design in fMRI takes advantage of allowing the functional data collected during a task to be clearly separated from data obtained during another task or rest period. It also provides a large number of temporally contiguous data points, which makes time series based analysis possible. However, a shortcoming of the blocked design is that it is difficult to estimate the response to each individual stimulus and what proportion of the evoked BOLD response is attributable to each of the cognitive subcomponents of the task. Event-related design can overcome this drawback of the blocked design since it is possible to examine individual stimulus evoked responses and temporally distinguishable stages of a task. Although some multivariate methods for modeling functional connectivity, such as time series correlations, are not capable of evaluating inter-regional interactions within closely spaced stages of a task, beta-series correlation, a method using single trial estimated BOLD responses, can measure inter-regional correlations between brain regions during distinct stages of task. The method is implemented by using separate covariates to model the activity evoked by each stimulus and during each stage of each individual trial in the context of the general linear model (GLM). The resulting parameter estimates (beta values) are sorted according to stimulus types and the stage from which they are derived to form a set of stimulus-specific or stage-specific beta series. Regions whose beta series are correlated are inferred to be functionally interacting. Using this method we investigate how the medial prefrontal cortex (mPFC)-linked appetitive network and amygdala-linked aversive network are generated and differ from each other by analyzing fMRI data recorded from human subjects viewing emotional pictures.

Record Information

Source Institution:
UFRGP
Rights Management:
All applicable rights reserved by the source institution and holding location.
Embargo Date:
5/31/2018

Downloads

This item has the following downloads:


Full Text

PAGE 1

NETWORK ANALYSIS OF NEURAL ACTIVITY By DAESUNG KANG A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSIT Y OF FLORIDA 2016

PAGE 2

2016 Daesung Kang

PAGE 3

To my parents and family, for their love and support

PAGE 4

ACKNOWLEDGMENTS I am thankful to my advisor Dr. Mingzhou Ding for letting me and researching this neuroscience field. T hanks to his help, I could finish my researches. I also express my gratefulness to Dr. Andreas Keil, Dr. Aysegul Gunduz, and Dr. Lin Yang for their comments and discussions as committee members. Id like to express my thanks to my Korean advisor s Prof. Jooyoung Park, Wh eekuk Kim, Gangbak Park and JuNo Jung for guiding me how to study and encouraging me to study in Univ ersity of Florida. Im thankful to all the colleagues and fellow students in the Lab and BME staff s I would like to dedicate my work to my parents Sangsoo Kang, Kyungran Jung, Byungseop Noh, and Seungsoon Cho who always supportive for any of my decision and happy for any of my accomplishments. Finally, I am also grateful to my wife Minjung Noh and kids, Jeongdoo Kang and Yura Kang for cheering me up in times of depression and giving me advice in improving my weaknesses. 4

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDGMENTS ...............................................................................................................4 LIST OF TABLES ...........................................................................................................................7 LIST OF FIGURES .........................................................................................................................8 ABSTRACT .....................................................................................................................................9 CHAPTER 1 INTRODUCTION ..................................................................................................................12 2 NONPARAMETRIC GRANGER CAUSALI TY AND APPLICATION TO MIXED RECORDINGS .......................................................................................................................17 2.1 Background .......................................................................................................................17 2.2 Method ..............................................................................................................................18 2.2.1 Spectral Representation of Continuous Time Series and Discrete Point Process .........................................................................................................................18 2.2.2 Granger Causality via Spectral Matrix Factorization .............................................21 2.2.3 Summary of the Algorithm .....................................................................................22 2.3 Simulation Results ............................................................................................................23 2.3.1 Example 1: Twonode Model .................................................................................23 2.3.2 Example 2: Three node Model ...............................................................................24 2.4 Discussion .........................................................................................................................25 3 THETA RHYTHMIC DRIVE BETWEEN MEDIAL SEPTUM AND HIPPOCAMPUS IN SLOW WAVE SLEEP AND MICROAROUSAL: A GRANGER CAUSALITY ANALYSIS .............................................................................................................................29 3.1 Background .......................................................................................................................29 3.2 Methods ............................................................................................................................30 3.2.1 Experimental Procedures ........................................................................................30 3.2.2 Data Analysis ..........................................................................................................33 3.3 Results ...............................................................................................................................34 3.4 Discussion .........................................................................................................................36 4 THETA RHYTHMIC DRIVE BETWEEN MEDIAL SEPTUM AND HIPPOCAMPUS IN THETA VS. NONTHE TA STATES ...............................................................................47 4.1 Background .......................................................................................................................47 4.2 Methods ............................................................................................................................48 4.2.1 Experimental Procedures ........................................................................................48 4.2.2 Data Analysis ..........................................................................................................49 5

PAGE 6

4.2.2.1 Phase detection and Z shift method .............................................................49 4.2.2.2 Nonparametric Granger causality ...............................................................52 4.3 Results ...............................................................................................................................53 4.3.1 Phase Locking between MS Units and HIPP LFP s ................................................53 4.3.2 Nonparametric Granger Causality .........................................................................55 4.4 Discussion .........................................................................................................................60 5 A FFECTIVE SCENE PROCESSING: L ARGE SCALE FUNCTIONAL INTERACTIONS REVEALED BY BETASERIES CONNECTIVITY ANALYSIS .........73 5.1 Background .......................................................................................................................73 5.2 Methods ............................................................................................................................75 5.2.1 Experimental Procedures and Data Acquisition .....................................................75 5.2.2 Data Preprocessing .................................................................................................79 5.2.3 Seed Region Selection ............................................................................................80 5.2.4 Functional Connectivity .........................................................................................80 5.2.5 Single trial Estima tion of LPP ................................................................................81 5.2.6 Statistical Inference ................................................................................................82 5.3 Results ...............................................................................................................................83 5.3.1 MPFC seeded Correlation Maps ............................................................................83 5.3.2 Amygdala seeded Correlation Maps ......................................................................83 5.3.3 LPP defined Correlation Maps ...............................................................................84 5.4 Discussion .........................................................................................................................86 5.4.1 MPFC and Its Patterns of Functional Connectivity ................................................86 5.4.2 Amygdala and Its Patterns of Functional Connectivity ..........................................88 5.4.3 Emotional Modulation of Visual Processing ..........................................................89 5.4.4 Functional Connectivity and Intensity of Emotional Engagement .........................90 5.4.5 Summary .................................................................................................................90 6 CONCLUSION .....................................................................................................................100 REFERENCES ............................................................................................................................105 BIOGRAPHICAL SKETCH .......................................................................................................116 6

PAGE 7

LIST OF TABLES Table page 51 Regions connected with mPFC during viewing of pleasant and unpleasant pictures. .......94 52 Regions connected with amygdala during viewing of pleasant and unpleasant pictures. ..............................................................................................................................95 53 Regions connected with mPFC for large LPP trials and small LPP trials ........................98 54 Regions connected with amygdala for large LPP trials and small LPP trials ....................99 7

PAGE 8

LIST OF FIGURES Figure page 21 Simulation results for two coupled processes. ...................................................................27 22 Simulation results for three coupled processes. .................................................................28 31 Single unit recording in the MS. ........................................................................................42 32 MS and hippoca mpal activity during SWS and microarousals. ........................................43 33 MS HIPP relationship in SWS and microarousal in two representative neurons .. ............44 34 Coherence and GC of MS unit and HIPP LFP activity. ....................................................45 35 Further characterization of GC.. ........................................................................................46 41 Examples of phase locking t o the HIPP LFP for a MS neuron during SWS, QW, AE, and REM sleep. ..................................................................................................................64 42 Examples of the Rayleighs Z statistics during SWS, QW, AE, and REM sleep.. ...........65 43 Phaselocking activity between MS units and HIPP LFP during SWS, QW, AE, and REM sleep. .........................................................................................................................66 44 One way ANOVA tests and multiple comparisons for further analysis.. ..........................67 45 MS and hippocampal activity during SWS, QW, AE, and REM sleep. ............................68 46 Analysis of coherence during SWS, QW, AE, and REM sleep. ........................................69 47 Analysis of GC during SWS, QW, AE, and REM sleep.. .................................................70 48 Further analysis of GC based on GC direction during SWS, QW, AE, and REM sleep. ..................................................................................................................................71 49 Further analysis of GC. ......................................................................................................72 51 MPFC seeded betaseries correlation maps .. .....................................................................92 52 Amygdala seeded betaseries correlation maps ................................................................93 53 MPFC seeded betaseries correlation maps based on engagement intensity contrast. ......96 54 Amygdala seeded betaseries correlation maps based on engagement intensity contrast. ..............................................................................................................................97 8

PAGE 9

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy NETWORK ANALYSIS OF NEURAL ACTIVITY By Daesung Kang May 2016 Chair: Mingzhou Ding Major: Biomedical Engineering Co nnectivity in the brain that can be cate gorized according to a pattern of anatomical connections is called anatomical connectivity, that is defined according to statistical dependencies is called functional connectivity, and that can give information on causal interactions is called effective connectivity. While a natomical connectivity refers to physical connections between neurons or neuronal elements f unctional connectivity captures deviations from statistical independence between neural units. Effective connectivity can describe directiona l effects of one neural unit over another. Neural activities, which were generated by neural units, can be measured by various modalities such as single neuron spike trains, local field potentials (LFP), electroencephalogram (EEG), or functional magnetic resonance imaging (fMRI) etc. and those neural activities can be used to infer connectivity and to elucidate how neural units and network of neurons process information. In this dissertation I examine the analysis of functional and effective connectivity f rom single neuron spike trains, LFPs and fMRI Granger causality is a new technique to measure effective connectivity and to determine the strength and direction of relationships between bivariate or multivariate signals. Although GC between continuous t ime series or GC between point process data are mathematically well 9

PAGE 10

defined, GC for mixed time series such as one being local field potential and the other being spike train data is not well understood. For the analysis of this type of data we introduce the nonparametric GC method With the method, we examin e the interaction patterns between medial septum (MS) and hippocampus in slow wave sleep (SWS) and microarousal Then we extend our interests from SWS/microa rousal to other theta and non theta states including quiet waking and REM sleep and infer the effective connectivity and temporal timing relationship between MS and hippocampus. Functional magnetic resonance imaging (fMRI) has become a pow erful imaging modality for studying human brain function. Blocked design in fMRI take s advantage of allowing the functional data collected during a task to be clearly separated from data obtained during another task or rest period. It also provides a large number of temporally contiguous data points, which makes time series based analysis possible. However, a shortcoming of the blocked design is that it is d ifficult to estimate the response to each individual stimulus and what proportion of the evoked BOLD response is attributable to each of the cognitive subcomponents of the task Event related design can overcome this drawback of the blocked design since it is possible to examine individual stimulus evoked responses and temporally distinguishable stages of a task Although some multivariate methods for modeling functional connectivity, such as time series correlations, are not cap able of evaluating inter regional interactions within closely spaced stages of a task, beta series correlation a method using single trial estimated BOLD responses, can measure inter regional correlations between brain regions during distinct stages of ta sk. The method is implemented by using separate covariates to model the activity evoked by each stimulus and during each stage of each individual trial in the context of the general linear model (GLM). The resulting parameter estimates (beta values) are so rted according to stimulus types 10

PAGE 11

and the stage from which they are derived to form a set of stimulus specific or stagespecific beta series. Regions whose beta series are correlated are inferred to be functionally interacting Using this method w e investigate how the medial prefrontal cortex (mPFC) linked appetitive network and amygdala linked aversive network are generated and differ from each other by analyzing fMRI data recorded from human subjects viewing emotional pictures 11

PAGE 12

CHAPTER 1 INTRODUCTION Co nnectivity in the brain can be categoriz ed as anatomical connectivity, functional connectivity, and effective con nectivity (Friston, 1994; Sporns, 2007) While a natomical connectivity refers to physical connections between neurons or neuronal elements f unctional connectivity captures deviations from statistical independence between neural units. Effective connectivi ty can describe directional effects of one neural unit over another (Sporns, 2007) Neural activities, which were generated by neural units, can be measured by various modalities such as single neuron spike trains, local field potentials (LFP), electroenc ephalogram (EEG), or functional magnetic resonance imaging (fMRI) etc. and those neural activities can be used to infer connectivity and to elucidate how neural units and network of neurons process information (Sporns, 2007) Inferring the directionality of interactions between neur al signals with Granger causality (GC) plays an important role in analysis of multichannel recordings. However, most commonly used GC methods are parametric approaches which depends on autoregressive (AR) models of time series data. For continuous valued signals such as EEG and LFP, AR model fitting is readily obtained using MVAR (multiva riate autoregressive) modeling and t he MVAR model order m i s determined by the Akaike information criterion (Akaike, 1974). With the MVAR spectra analysis, GC has been applied to investigate directional relationship of continuous valued signals in many applications and has proven to be an effective method (Brovelli et al. 2004; Seth et al. 2005; Ding et al. 2006). For point process data such as spike trains it is difficult to apply the parametric GC since AR model of spike train data is not directly obtained. As a remedy of this problem, a nonparametric GC method was proposed and has shown reasonable results ( Dhamala et al. 2008a; Dham ala et al. 2008b; Nedungadi et al. 2009) However, there have been 12

PAGE 13

few studies of GC applied to mixed data such as bivariate data with point process data and continuous valued data. In this dissertation, w e present a novel nonparametric approach for c onstruction of directionality measures for bivariate discrete point processes and continuous valued data. The nonparametric GC method is applied to examine the effectivity connectivity between medial septum (MS) and hippocampus (HIPP) by recording MS unit activity together with HIPP LFP during SWS and its brief interruptions (<10s) called microarousals (Halasz et al. 1979; Schieber et al. 1971). M S plays a critical role in controlling the electrical activity of the HIPP. In particular, theta rhythmic burst firing of MS neurons is thought to drive lasting HIPP theta oscillations in rats during waking motor activity, and REM sleep. Less is known about MS HIPP interactions in nontheta states such as non REM sleep in which HIPP theta oscillations are absent but theta rhythmic burst firing in subsets of MS neurons is preserved. SWS is a nontheta state which cycles through several stages characterized by different patterns (varying frequency, amplitude, spindles, etc.) of slow wave activity in HIPP. Microarousal s represent disruptions of slow wave activity and are characterized by transient muscle activation and HIPP theta rhythm Microarousals are theta states and recur during these SWS cycles. To assess the directional influences between MS and HIPP, a nonparametric Granger causality (GC) i s applied to the mixed spike field data to decompose the MS HIPP relationship into their directional components MS HIPP and HIPP MS. Power, coherence and GC are then compared between the SWS and microarousals to assess the di fferences in the pattern of MS HIPP interactions. B roadly, hippocampal activities can be classified into two distinct state dependent local field potential patterns, theta and nontheta states. Theta state is selectively characterized as two behavioral states, active exploration (AE) and rapid eye movement (REM) sleep, while nontheta 13

PAGE 14

state occurs during quiet waking (QW) and slow wave sleep (SWS). Instead of dealing with microarousals as a theta state, we conducted a m ore thorough study including additional recordings of different behavioral states including AE and REM sleep. In AE, animals are engaged in exploratory behavior (locomotion, sniffing, and whisking) showing voluntary motor activity. During REM sleep, animals are immobile and atonic except for intermittent whisker and ear twitches. Those two behaviors show low amplitude local field potentials (LFPs) and high theta (5 9 Hz) and gamma (30 55 Hz) power spectrum density and are theta states In QW, animals are immobile (standing or sitting quietly) or engaged in automatic stereotyped behaviors (eating, drinking, and grooming) representing low amplitude LFPs and is considered a non theta state During SWS, animals are lyi ng immobile with eyes closed and slow regular respiratory movements. The LFPs represent highamplitude and slow waves with delta oscillations (1 4 Hz). It begins with sleep spindles (10 14 Hz) superimposed to delta oscillations As SWS deepens, delta wav es stay dominated although isolated spindles can still be observed (Gervasoni et al., 2004). The HIPP deals with information from several cortical areas as well as subcortical areas, such as medial septum (MS). Specifically, the MS and the HIPP have recipr ocal pathways (Raisman 1966). The HIPP receives both GABAergic and cholinergic fibers located in the MS via the fimbria fornix (Freund and Antal 1988; Frotscher and Leranth 1985) whereas the HIPP is terminated on the GABAergic neurons in the MS nucleus (Toth and Freund 1992). The MS plays a critical role in regulating the electrical activity of the HIPP and provides information about the behavioral states of the animal (Khakpai et al. 2012). Although the characteristics of the neurons participating in these two way projections have been studied extensively (Dragoi et al. 1999; Ford et al. 1989; King et al. 1998; Petsche et al. 1962; Sweeney et al. 1992), the mechanisms of their functional interactions are not well understood (Bland 1986; Vertes and Kocsis 1997). In an 14

PAGE 15

attempt to uncover the functional interactions between MS and HIPP, we investigated the relationship between single unit activity in MS and LFP oscillations in HIPP during theta states and nontheta states. First, we analyzed the temporal activi ty between MS units and HIPP LFP to reveal the timing relationships between MS unit firings and HIPP LFP based on behavioral states. Then, we evaluate information flows between MS and HIPP with a non parametric Granger causality to examine how the pattern has changed during theta and nontheta states Blocked design experiments in fMRI has the advantage of allowing the functional data collected during a task to be clearly separated from data obtained during another task or rest period. It also provides a large number of temporally contiguous data points, which makes time series analysis possible (Rissman et al. 2004). However, a shortcoming of the blocked design experiments is that it is difficult to estimate brain response evoked by individual stimuli and the proportion of the evoked BOLD response that is attributable to each of the cognitive subcomponents of the task (DEsposito et al. 1999). Event related design can overcome the drawback of the blocked design since it is possible to employ a task with temporally distinguishable stages and to study the effect of individual stimuli (Postle et al. 2000; Zarahn et al. 1997). Although some multivariate methods for modeling functional connectivity, such as time series correlations, are not capable of evaluating inter regional interactions within closely spaced stages of a task, betaseries correlation by measuring brain response to individual stimuli, can a ssess inter regional correlations between brain regions for different stimulus types and during distinct stages of task. The method is implemented by using separate covariates to model the activity evoked during each stage of each individual trial in the c ontext of the general linear model (GLM). The resulting parameter estimates (beta values) are sorted according to stimulus types or the stage from which they are derived to form a set of stimulus specific or stage specific 15

PAGE 16

beta series. Regions whose be ta series are correlated during a given stage or for a given stimulus types are inferred to be functionally interacting during that stage or for that stimulus type (Rissman et al. 2004). W e applied beta series correlatio n analysis to investigate how the medial prefrontal cortex (mPFC)linked appetitive network and amygdala linked aversive network are generated and differ from each other It has been hypothesized that the medial prefrontal cortex (mPFC) is a hub in the network that mediates appetitive responses whereas the amygdala plays a crucial role in mediating both aversive and a ppetitive processing. These structures facilitate adaptive actions by linking perception, attention, memory, and motor circuits. In this dissertation, w e provide an initial exploration of these hypotheses by recording simultaneous EEG fMRI in eleven participants viewing affective pictures. MPFC and amygdala seeded functional connectivity maps are generated by applying the beta series correlation method. Another important factor that influences large scale brain network responses concerns the intensity of e motional engagement. As with many other complex functions, emotional processing exhibits substantial trialby trial variability depending on the eliciting stimuli and the functional state of the brain. Numerous studies indicate that the late positive potential (LPP) reliably indexes the degree of emotional arousal and processing intensity (Cuthbert, Schupp, Bradley, Birbaumer, & Lang, 2000; Keil, et al., 2002; Schupp, et al., 2004) LPP has been shown to be associat ed with BOLD responses in distributed brain areas (Liu, Huang, McGinnis Deweese, Keil, & Ding, 2012; Sabatinelli, Keil, Frank, & Lang, 2013; Sabatinelli, Lang, Keil, & Bradley, 2007) To further assess the modulati on of these correlations by emotional intensity, we used single trial LPP from the simultaneously recorded EEG to index the varying intensity of emotional engagement within a hedonic valance and examined functional connectivity separately for large LPP tri als and small LPP trials. 16

PAGE 17

CHAPTER 2 N ONPARAMETRIC G RANGER CAUSALITY AND APPLICATION TO MIXED RECORDINGS 2.1 Background Granger causality (GC) is a statistical measure based on the con cept of time series forecasting. Specifically, if the current state of a time series is better predicted by incorporating the past knowledge of a second one, the second series is said to have a causal influence on the first (Ding et al. 2006) The estimation of GC relies on autoregressive (AR) models in parametric GC For continuous valued signals such as EEG and LFP, AR model fitting is readily obtained using MVAR (multivariate autoregressive) modeling The continuous valued signals are subjected to MVAR s pectral analysis from which power, coherence, and Granger causality spectral estimates are derived. The essential steps of MVAR can be summarized as follows. Let p channels ( p = 2 or more ) of continuous signals at time t be denoted by = , , where T stands for matrix transposition. Assume that the data over an analysis window are described by an MVAR model: = (2 1) where is a temporally uncorrelated residual error series with covariance matrix and are coefficient matrices to be estimated from data (Ding et al. 2000) The MVAR model order m was determined by the Akaike information criterion (Akaike, 1974) With the AR model, GC has been applied to investigate directional relationship of continuous valued signals in many applications and has proven to be an effective method (Brovelli et al. 2004; Seth et al. 2005; Ding et al. 2006). For discrete time series such as spike train dat a it is difficult to apply the above GC approach since AR model of spike train data is not directly obtained. To resolve this issue, 17

PAGE 18

several approaches have been attempted to apply GC to the spike train data ( Sameshima and Baccala 1999; Fanselow et al. 20 01; Kaminski et al. 2001; Zhu et al. 2003; Nedungadi et al. 2009; Kim et al. 2011). Some work s attempted to convert the spike train data into continuous time series by using a low pass filter ( Kaminski et al. 2001; Zhu et al. 2003) or a smoothing ke rnel (Sameshima and Baccala 1999; Fanselow et al. 2001) While th ese approach es ha ve been applied to both simulated and experimental data with generally acceptable results, it is cautioned that the smoothing operation violates the point process character of spike trains Furthermore, the approach es are highly kernel dependent and may introduce unwanted distortions (Truccolo et al. 2005). As another remedy to tackle discrete time series in GC application, a nonparametric GC method (Nedungadi et al. 2009) a nd likelihood based GC (Kim et al. 2011) have been proposed and showed reasonable results in both simulated and experimental data. Although GC between continuous time series or GC between point process data are mathematically welldefined, GC between local field potential and spike trains, referred to as mixed recordings or mixed signals, is not well understood. In o rder to obtain directionality between mixed signals, w e are not able to apply the AR model based parametric GC for continuous valued signals and nonparametric GC for spike train data explicitly. Instead, we extend the nonparametric GC to the mixed time s eries of LFP and spike train data by combining spectral matrix factorization of mixed time series with Geweke s spectral formulation of GC. Based on our knowledge, the GC to the mixed time series has not been addressed, so we will explain the method. 2.2 Method 2.2.1 Spectral R epresentation of C ontinuous T ime S eries and D iscrete P oint P rocess Let a realization of zero mean time series (e.g. LFP) be denoted by ( ) It is also assumed that the time series is real valued with sample values defined at equispaced intervals 18

PAGE 19

equal to the sampling interval Let the realizations of stochastic point processes (e.g. spike trains) be denoted ( ) A stochastic point process, ( ) is represented by a counting variable which indicates the number of spikes in the interval (0,t ]. It is assumed that spike events of the process ( ) do not occur simultaneously and this assumption is known as orderly condition (Conway et al. 1993; Halliday et al. 1995). To represent the number of spikes in a small interval differential increment is defined as = ( + ) ( ) and will take on the value 0 or 1 depending on the occurrence of a spike in the sampling interval To deal with mixed signals one signal being continuous data and the other point process data, two further conditions, wide sense stationarity condition (statistical properties of the process and time series are time invariant) and mixing condition (differential increments and/or sample values occurring in intervals separated wi dely in time are statistically independent) should be satisfied (Halliday et al. 1995). In this dissertation, we used the LFP as a continuous time series and spike trains as a point process data. Both LFP and spike trains are divided into nonoverlapping disjoint windows and each of window length is For each window ( = 1 , ), the finite Fourier transform of LFP time series ( ) at frequency is defined as (Brillinger, 1972) ( ) = ( ) ( 2 ) ( ) ( 2 ) ( ) ( 22) As in the above way, the finite Fourier transform of spike trains, denoted ( ) for each window is defined as ( ) = ( 2 ) ( ) ( ) 2 ( ) (2 3) where are the times of occurrence of the spikes from process The spectral matrix between LFP ( ) and spike trains ( ) is defined as ( ) = ( ) ( ) ( ) ( ) ( 24) 19

PAGE 20

with diagonal terms representing auto spectra and off diagonal term cross spectra. The spectral density matrix from LFP and spike train s of length can be estimated using the multitaper method (Thomson 1982; Jarvis and M itra 2001; Walden 2000). We start by applying data tapers { } to the ( ) and ( ) respectively. Then we take the Fourier transform of each ( ) and ( ) For the LFP, the tapered ( ) is obtained as follows ( ) = ( ) ( ) ( 2 ) = 2 ( 25) where ( ) = 1 For the spike train data, the tapered ( ) is expressed as ( ) = ( ) ( 2 ) ( ) = 2 ( 26) In the multitaper method, the set of orthogonal data tapers used (given by the discrete prolate spheroidal sequences) are optimal in that they have good leakage properties. We then obtain estimates for the matrix elements of the spectral matrix using the tapered ( ) and ( ) : the estimates of auto spectrum ( ) for LFP, ( ) for spike trains and cross spectrum ( ) between LFP and spike trains (Rosenberg et al. 1998) The cross spectrum ( ) can be estimated as ( ) = ( ) ( ), ( 27) where denotes complex conjugate. The autospectrum of the LFP, ( ) is estimated by replacing the ( ) with ( ) in equation ( 27). Similar procedures hold for the autospectrum of spike trains, ( ) In case of multiple realizations (trials), equation ( 27) gives an es timate of cross spectrum using one realization. Averaging these estimates over all trials will give the estimate of all the elements of spectral matrix. Note that in case of binned data set, in equation ( 25) and ( 26) is replaced by where is the number of spikes in the th bin. 20

PAGE 21

2.2.2 Granger C ausality via S pectral M atrix F actorization For the parametric approach an AR model can be fit to the data. From the model one computes the spectral matrix according to ( ) = ( ) ( ) ( 28) where the asterisk denotes matrix transposition and complex conjugation, ( ) = i s the transfer function which depends on the coefficients of the AR model, and is the cov ariance matrix of the error terms in the AR model (Ding et al. 2000, 2006). The three entities in spectral matrix are the basis for estimating GC in the spectral domain (Geweke 1982, 1984) The power spectrum of channel l is given by ( ) which i s the l th diagonal element of the spectral matrix ( ) The coherence spectrum between channel l and channel k is as follows: ( ) =| ( ) | | ( ) ( ) | (2 9) The value of coherence ranges from 1 to 0, with 1 indicating maximum interdependence between channel l and channel k at frequency f and 0 indicating no interdependence. For the nonparametric approach, the spectral density matrix ( ) is estimated from data using cross spectrum To compute GC we still require the other two entities in spectral density matrix These can be obtained by applying spectral density matrix factorization (Wilson 1972) which decomposes ( ) into a unique corresponding transfer function ( ) and the noise covarian ce matrix (Dhamala et al. 2008a, b). Since we obtained all three entities in spectral density matrix, we can start with the 22 spectral submatrix formed by taking the appropriate entries from the overall spectral matrix in equation ( 24). After spectral factorizing this spectral density matrix, GC from ( ) to ( ) at frequency f is given by (Ding et al. 2006): 21

PAGE 22

( ) = ( ) ( ) | ( ) | ( 210) which can be interpr eted as the proportion of ( ) s causal contribution to the power of the ( ) series at frequency f The logarithm is taken to preserve certain favorable statistical properties. Similarly, the causality spectrum from ( ) and ( ) can be ob tained by switching the indices and x in e quation (2 10) In a system of three or more mixed continuous time series and spike train data, it is often desirable to find out whether a causal influence between any pair of neurons is direct or mediated by others. The above pairwise measure of causality is not able to resolve this issue (Chen et al. 2006b). This has led to the development of the conditional GC (Geweke 1984, Chen et al. 2006, Ding et al. 2006) ). The formulations for the frequency domain c onditional GC can be readily extended see for example, Ding et al. (2006). 2.2.3 Summary of the A lgorithm A step by step algorithm for computing non parametric GC between LFP and spike train s is given below. Step 1 The spike trains gener ated by a neuron are taken as one realization of stochastic point process denoted by ( ) The spike trains are binned where the bin width was pre chosen (in this dissertation, the bin width is 1 ms) When we define smaller bin width, many of the bins will contain no spikes leading to poor estimation of the spectral density matrix. Step 2 The Fourier transform of ( ) and ( ) are estimated using equation ( 22) and ( 23) respectively Step 3 The spectral density matrix ( ) for the time series ( ) and stochastic process ( ) is obtained by using equation ( 27) 22

PAGE 23

Step 4 T he spectral density submatrix is factorized using the spectral factorization algorithm (Wilson 1972; Dhamala et al. 2008a, b), thus giving the decomposition in equation ( 28). Step 5 GC from LFP ( ) to spike train s ( ) is evaluated as a function of frequency by substituting the transfer function ( ) and the noise covariance matrix in equation ( 210). This function can be examined for frequency characteristics of causal influences or summed over all frequencies to obtain a single time domain causal influence. A similar procedure is carried out to evaluate causality from spike train s ( ) to LFP ( ) after reversing ( ) and ( ) in equation (210) 2.3 Simulation Results We consider numerical examples to illustrate the application of the nonparametric GC between continuous time series and point process data outlined earlier. 2.3.1 Example 1: Twonode Model Consider the following three simple two node models. From the first model, equation (211), t wo n odes were coupled in such a way that the output of the continuous time series ( ) was fed into the point process data ( ) ( ) = 0 8 ( 1 ) 0 7 ( 2 ) + ( ) = 1 ( ) + 1{ } ( 1 ) > 0 0 (2 11) The second model, equation (212), also describes unidirectional two nodes model, but the directional causality is from point process data ( ) to continuous time series ( ) ( ) = 0 7 ( 1 ) 0 5 ( 2 ) 0 7 ( 1 ) + ( ) = 1 ( ) > 0 0 (2 12) 23

PAGE 24

The last model illustrates bidirectional mixed signal model as following equation (213). ( ) = 0 9 ( 1 ) 0 7 ( 2 ) + 0 4 ( 1 ) + ( ) = 1 ( ) + 1{ } ( 1 ) > 0 0 (2 13) where is Gaussian white noise processes with zero mean and variance = 0 3 Poisson process ( ) has a parameter = 0 1 and is a uniform variable between 0 and 1. We define 1{ } to be 1 if an arbitrary uniform variable is less than 0.15 and 0 otherwise. We generated a data set of 1000 realizations of 1000 time points each. The sampling interval is 1ms and bin size is 1ms. The schematic diagrams of the above models are shown in the left side of Figure 2 1 and the corresponding GC results are illustrated in the right side of Figure 2 1. The result shown in Figure 2 1 clearly recovers the pattern of connectivity in e quation ( 213 ). 2.3.2 Example 2: Three node Model In this example, we consider tw o models each consisting of two continuous time series and one point process The first model simulates the case in which influences from continuous time series ( ) to point process data ( ) is indirect and completely mediated by ( ) which is continuous time series : ( ) = 0 8 ( 1 ) 0 7 ( 2 ) + ( ) = 0 7 ( 1 ) 0 4 ( 2 ) 0 7 ( 1 ) + (2 14) ( ) = 1 ( ) + 1{ } ( 1 ) > 0 0 The second model generates bidirectional interaction between continuous time series ( ) and point process ( ) with a common source ( ) ( ) = 0 9 ( 1 ) 0 6 ( 2 ) + 0 4 ( 1 ) + 0 5 ( 1 ) + ( ) = 0 8 ( 1 ) 0 4 ( 2 ) + (2 15) 24

PAGE 25

( ) = 1 ( ) + 1{ } ( 1 ) + 1{ } ( 1 ) > 0 0 For both models, and are two independent Gaussian white noise processes with zero means and variances of = 0.2, = 0.3, respectively. Poisso n process ( ) has a parameter = 0 1 and is a uniform variable between 0 and 1. We set 1{ } is 1 if an arbitrary uniform variable is less than 0.15 and 0 otherwise. Each model was simulated to generate a data set of 1000 realizations of 1000 time points each like example 1 For the analyses, pairwise nonparametric G C was performed on the simulated data set of each model. The results are shown in Figure 22. From these result s it is clear that the nonparametric GC can represent the interacti on between three time series. 2.4 Discussion Parametric GC has been applied to investigate directional relationship of continuous valued signals in many applications and has proven to be an effective method (Brovelli et al. 2004; Seth et al. 2005; Ding et al. 2006). However, concerns have been raised regarding the strong underlying assumptions and its suitability for data with complex power spectra since the parametric spectral approach requires the autoregressive (AR) models of data (Mitra & Pesaran 1999) In this dissertation, w e have presented a non parametric Granger causality in which GC is estimated directly from Fourier transforms of data without the need for AR models. The mathematical basis of this method is a combination of spectral matrix factori zation and Geweke's spectral formulation of Granger causality. Thus, the nonparametric GC method is able to measure the direct ional causal effects between mixed signals, bivariate continuous valued signal and point process data. The proposed GC method is tested on simulated data of two node and three node models. In the examples, multiple realizations of time series were generated by a two node and three 25

PAGE 26

node models. The patterns of network connectivity were correctly recovered. This provided the validatio n of this method for mixed signals. These results are important since there are few methods which can assess the interactions between mixed signals as well as can be applied to diverse application domains. 26

PAGE 27

Figure 2 1. Simulation results for two couple d processes T hree distinc t patterns of connectivity between continuous time series ( ) and point process data ( ) (A) Unidirectional coupling from continuous time series to point process ( B ) unidirectional coupling from point process to continuous time series and ( C ) bidirectio nal coupling. 27

PAGE 28

Figure 22. Simulation results for three coupled processes Two distinct patterns of connectivity among two continuous time series ( ) ( ) and point process ( ) (A) Continuous time series ( ) has an indirect influence on ( ) via ( ) ( B ) Conti nuous time series ( ) has a direct influence on ( ) and ( ) 28

PAGE 29

CHAPTER 3 THETA RHYTHMIC DRIVE BETWEEN MEDIAL SEPTUM AND HIPPOCAMPUS IN SLOW WAVE SLEEP AND MICROAROUSAL: A GRANGER CAUSALITY ANALYSIS 3.1 Background The o verwhelming majority of prior investigations of the relationship between medial septum (MS) and hippocampus (HIPP) focused on the role of MS input in controlling hippocampal activity. However, the MS and the HIPP are reciprocally connected (Raisman 1966) ; ascending cholinergic (Frotscher and Leranth 1985) and GABAergic projections from MS to HIPP (Freund and Antal 1988) are reciprocated by descending GABAergic projections from HIPP to MS (Toth and Freund 1992) Although the characteristics of the neurons participating in these two way projections have be en studied extensively (Dragoi et al. 1999; Ford et al. 1989; King et al. 1998; Petsche et al. 1962; Sweeney et al. 1992) the mechanism of their functional interactions are not well understood (Bland 1986; Vertes and Kocsis 1997) Functional MS HIPP interactions are statedependent. During theta states, including waking motor activity, REM sleep, and arousals, when HIPP activity is dominated by rhythmic field oscillations in the 410Hz range (Buzsaki 2002) MS neurons fire rhythmic bursts in synchrony with local field potential (LFP) in HIPP (Petsche et al. 1962) Lesion and pharmacological studies further suggest that theta rhythmic cells in MS are pacemakers of HIPP theta oscillations (Lawson and Bland 1993) To what extent this theta pacemaker hypothesis extends to nonREM sleep remains unclear. Although MS HIPP interactions in nontheta states have received less attention, i t was noted that theta burst firing is preserved in slow wave sleep (SWS) in a subset of MS neurons (Sweeney et al. 1992) when HIPP theta is absent. Two hypotheses may be put forth to account for this observation. First, necessary theta drive from MS does not reach HIPP in nontheta states Second, theta drive from MS reaches HIPP but low HIPP responsiveness to MS input prevents a theta rhythmic response in HIPP. Accordingly, 29

PAGE 30

switch to theta may be induced either by increased synchronization of MS neuron firing, thereby increasing the ascen ding MS drive over the threshold to force theta in HIPP, or by an increase of HIPP network responsiveness to MS input, thereby generating intrinsic theta, and activating a descending theta drive to synchronize the MS theta pacemaker, as suggested by in vit ro and modeling studies (Manseau et al. 2008; Wang 2002) In this study we examined functional MS HIPP interactions by recording MS unit activity together with HIPP LFP during SWS and its brief interruptions (<10s) called microarousals (Halasz et al. 1979; Schieber et al. 1971) SWS is a nontheta state which cycles through several stages characterized by different patterns (varying frequency, amplitude, spindles, etc.) of s low wave activity in HIPP. Microarousals, which represent disruptions of slow wave activity and are characterized by transient muscle activation and HIPP theta rhythm, are theta states recurring during these SWS cycles. To assess the directional influences between MS and HIPP, a novel method called nonparametric Granger causality (GC) was applied to the mixed spikefield data to decompose the MS HIPP relationship into their directional components MS HIPP and HIPP MS. Power, coherence and GC were then compa red between the theta and non theta states to assess the differences in the pattern of MS HIPP interactions. 3.2 Methods 3.2.1 Experimental Procedures Experiments were performed on male Sprague Dawley rats (Charles River Laboratories, MA), treated in accor dance with NIH guidelines. All experimental procedures were approved by the Institutional Animal Care and Use Committee of the Beth Israel Deaconess Medical center. Four r ats (300350 g body weight at the time of surgery), were anesthetized with a mixture of Ketamine and Xylazine (70 80 and 10 mg/kg, respectively, injected intraperitoneally) for implantation of stainless steel wires for recording hippocampal LFP, stainless steel screws for 30

PAGE 31

reference, ground, and cortical EEG recording, and multithreaded wir es to record neck muscle activity (EMG). The rats head was fixed in a stereotaxic frame and stereotaxic coordinates, antero posterior (AP), lateral (Lat) and dorso ventral (DV) relative to bregma, were measured according to Paxinos atlas (Paxinos and Watson 1986) Surface EEG was recorded over the frontal (AP: +1.0 mm, Lat: 2 mm) and parietal (AP: 6.5 mm, Lat: 2.5 mm) cortex. Hippocampal electrode was placed in the CA1 region (AP: 3.7 mm, Lat: 2.2 mm, DV: 2.5 mm) to record from the upper theta dipole (Buzsaki 2002) oscillating in phase with theta waves in the parietal cortical EEG. For MS unit recording, three tetrodes made of 13um nichrome microwires were mounted on individually movable (0.3 mm axial movement per complete turn) microdrives and lead into a guide tube placed above the MS (AP +0.5mm, Lat 0.0mm, DV 5.0mm). The tetrodes were moved another 2 mm out of the cannula at the end of the implantation surgery. All wires were led to a 16 pin omnetics microconnector. The screws, wires and the electrodemicrodrive assembly were fixed to the skull with dental acrylic. Electrophysiological recordings started after a 7 10 day recovery period. Daily recording sessions lasted 2 6 hours during daylight period, in a 26x17x17cm r ecording box. After stable LFP and EMG recordings were attained, the tetrodes were moved slowly into the MS until discriminable unit activities were found. The tetrodes moved through the MS in small steps over several consecutive days. The MS electrode location was marked at the end of the experiment by direct current to generate lesions at several dorsoventral locations which together with the damage due to the guide cannula served for verification of electrode placement in the MS ( Figure 3 1A) while the d orsoventral location of individual neurons were estimated using the number of turns of the microdrive. Theta rhythmic cells were encountered in the MS along the 31

PAGE 32

midline in 3 out of 4 rats; in 1 rat in which electrode tracks were found more lateral no theta cells were found and thus was excluded from the analysis. The electrical signals were amplified, filtered (LFP: 0.1 100Hz, EMG: 0.13kHz, units: 6003kHz) and digitally sampled (16bit, 10 kHz; Neuralynx, Inc.). MS single neurons were identified and extra cted off line based on their amplitude and wave shape using principal component analysis and K means clustering algorithms (Spike2, Cambridge Electronic Devices, UK). Units showing a refractory period of 2ms or higher were considered as single units. The number of simultaneously recorded units varied between 1 and 10 (median 5) (see example in Figure 31B). SWS periods and microarousal episodes were identified according to standard polysomnographic evaluation criteria based on visual sleep scoring aided by auxillary signals representing running averages of EMG total power, as well as EEG power in the delta range (14 Hz) over the frontal cortex and in the theta range (510 Hz) over the parietal cortex and in the hippocampus, as well as the theta/delta ratio. Specifically, SWS was characterized by concurrent low EMG activity, high delta and low theta EEG power. Microarousals were short episodes during SWS characterized by abrupt disappearance of large cortical delta waves and concurrent switch of HIPP LFP to t heta rhythm and appearance of motor activity on the background of low muscle tone. Recordings which included SWS and microarousals and had at least one theta rhythmic single unit in any of the tetrodes were considered for further analysis. All neurons enc ountered in these recording sites, a total of 79 cells in SWS and 71 cells in microarousal with the 71 cells being a subset of the 79 cells, were then included in the analysis independent of their firing properties. The spike trains of identified MS units along with HIPP LFP signals and markers of sleep stages were transferred to MATLAB for analysis. 32

PAGE 33

3.2.2 Data Analysis HIPP LFP and MS spike trains were subjected to spectral analysis. Power spectra and MS HIPP coherence spectra were estimated according to established procedures. MS HIPP interaction was further decomposed into their directional components, MS HIPP and HIPP MS, using a recently proposed non parametric Granger causality (GC) algorithm designed for point processes as well as continuous valued recordings (Dhamala et al. 2008a; b; Nedungadi et al. 2009) GC is a statistical measure based on the concept of time series forecasting. Specifically, if the current state of a time series is better predicted by i ncorporating the past knowledge of a second one, the second series is said to have a causal influence on the first. Past work has demonstrated that Granger causal influence can be interpreted in terms of synaptic transmission between neurons and neuronal e nsembles (Bollimunta et al. 2008; Ding et al. 2006; Nedungadi et al. 2009) Traditionally, GC is estimated parametrically via autoregressive models of time series data. For spike trains, autoregressive modeling is not directly applicable. A nonparametric framework for estimating GC based on Fourier transforms has been proposed to overcome this problem (Dhamala et al. 2008a; b; Nedungadi et al. 2009) For the present experim ent, it consisted of the following steps. First, the continuous recordings were divided into 2s nonoverlapping epochs which were treated as realizations of an underlying stochastic process. A KSPP test demonstrated that over 99% of the LFP epochs met the stationarity requirement (Kwiatkowski et al. 1992) Second, each epoch was further divided into 1ms bins where the bin size was chosen such that no more than 1 spike can be found in any bin. If we use narrower bin size, many of the bins will contain no spikes which lead to poor estimation of the spectral density matrix. Third, HIPP LFP and MS spike train were subject to separate Fourier transforms; through proper averaging across all the recording epochs within a behavioral state the spectral density matrix was obtained. Fourth, the spectral density matrix was 33

PAGE 34

factorized and combined with Gewekes spectral GC formalism to yield MS HIPP and HIPP MS in the spectral domain (Ding et al. 2006; Geweke 1982) For statistical analysis, a random permutation procedure was used to generate the significance thresholds for coherence and GC. Specifically, for each neuron, the epoch labels for LFP and the epoch labels for spike train were permuted randomly 1000 times. Coherence and GC were computed for each of the 1000 permuted datasets. Null hypothesis distributions were constructed based on these coherence and GC values. Thresholds corresponding to p=0.01 were determined and neurons whose coherence or GC was ab ove their respective thresholds were considered statistically significant. 3.3 Results Microarousals were identified and extracted from continuous SWS recordings. These microarousal episodes, characterized by the appearance of HIPP theta and abrupt motor activity appearing on the background of low muscle tone, were brief, usually lasting less than ~10s ( Figure 3 2A); there were no discernible transition periods between SWS and microarousal. A total of 271 microarousal episodes with average length of 7.60 0.58s extracted from 15 SWS recordings in 3 rats (data from 1 rat were excluded, see Methods) were used for analysis. Power spectra of both MS unit activity and HIPP LFP showed prominent theta peaks in microarousals, whereas delta activity dominated HIPP L FP during SWS (Bland 1986) In MS, spectral power was distributed fairly broadly during SWS, with relatively weak local peaks present in most neurons in the 412Hz range ( Figure 32B). The firing rate of the MS units were in the range reported in earlier studies (Dragoi et al. 1999; Ford et al. 1989; King et al. 1998; Sweeney et al. 1992) and was not significantly different (t[148]=0.47, p=0.6414) between SWS ( 10.73 2.70 spikes /s) and microarousal ( 9.93 1.89 spikes/s) ( Figure 32C). Functional MS HIPP interactions were investigated by spectral coherence and Granger causality; the latter offers the advantage of characterizing the strength of causal influences in 34

PAGE 35

different frequen cy components. Results for 2 neurons are demonstrated in Figure 33; group averages are summarized in Figure 34. During SWS, MS HIPP unit LFP coherence was weak but significant in 24 of 79 (30.38%) neurons (group average coherence: 0.0238 0.0094; Fig ure 34A 4C); the peak frequency varied in the range of 312Hz (average 8.25 0.46Hz, Figure 34D). In microarousals, MS HIPP coherence was significant in 56.34% of the neurons, the coherence was significantly higher ( 0.0413 0.0118, t[ 148 ]= 2.30, p= 0.0228) and t he peak frequency was concentrated in a narrow frequency band (67Hz; average 6.81 0.37Hz, Figure 34D). GC during SWS indicated a nearly unidirectional MS HIPP drive in which MS neuronal activity affected HIPP LFP (group average: 0.0162 0.0079) at frequen cies within the theta range (see examples in Figure 33) but the firing activity of these neurons was much less affected by HIPP activity (HIPP MS GC= 0.0042 0.0013; comparison of GC in the two directions shows that MS HIPP is significantly greater than HIP P MS: t[ 156]= 2.93, p= 0.0039; Fig ure 3 4E 4 G). During microarousals, the interaction became bidirectional, with the advent of a strong descending HIPP MS GC component ( 0.0162 0.0044; the increase was significant: t[ 148]= 5.40, p< 0.0001). In contrast, MS HI PP GC in microarousal ( 0.0200 0.0068) did not change significantly compared with that in SWS (t[ 148]= 0.70, p= 0.4820); moreover, in microarousal, HIPP MS and MS HIPP GC no longer differed (t[ 140]= 0.92, p= 0.3614). Furthermore, in microarousal, the peak fre quency of MS HIPP GC spectra shifted to lower frequencies around the theta peak, with maxima either at ~4 or at ~8Hz (group average: 6.46 0.50Hz; Figure 34H) whereas the shape of HIPP MS GC was similar to theta components in the power spectra and in coher ence function ( Figure 3 3) with peak at the same frequency ( 6.18 0.33Hz; Figure 34H). 35

PAGE 36

In individual neurons, if MS HIPP GC was significant in SWS it remained significant in microarousal as well, although the GC value could either increase or decrease, com pared with SWS ( Figure 35A). In microarousal, there was further recruitment of MS neurons from the pool showing insignificant MS HIPP in SWS ( Figure 35A, red dots). Recruitment was substantially stronger for MS neurons receiving the descending HIPP MS dr ive; almost half of the MS population (45.07%) showed significant HIPP MS GC in microarousal whereas very few neurons (5.06%) showed HIPP MS GC in SWS ( Figure 34E ). Furthermore, both MS HIPP ( Figure 35B) and HIPP MS ( Figure 35C ) GC positively correlated with the neurons firing rate indicating that fast firing neurons were more likely engaged in MS HIPP theta synchrony. Although identification of the type of neurons is impossible when using extracellular recordings, it was suggested earlier, based on spi ke shape and firing rate analysis (King et al. 1998; Matthews and Lee 1991) that rhythmic theta bursting activity is the property of fast firing GABAergic neurons in the MS. 3.4 Discussion The present study exam ined the causal interactions between MS and HIPP within the theta frequency band in SWS and its short, waking interruptions, called microarousals. A novel nonparametric Granger causality method was applied to decompose the MS HIPP synchrony into their dir ectional components (Dhamala et al. 2008a; b; Nedungadi et al. 2009) Because our data consisted of mixed spike field recordings, time domain GC methods developed specifically for spike trains are difficult to appl y (Kim et al. 2011) whereas our GC method, formulated in the frequency domain, overcomes this limitation The main finding is that there is a significant unidirectional MS HIPP influence over a wide band (210Hz) in SWS which switches to bidirectional theta drive during microarousals with MS HIPP and HIPP MS GC being of equal magnitude. Unidirectional MS HIPP 36

PAGE 37

influence in SWS was accompanied by significant MS HIPP coherence (max at 8.25 0.46Hz), but no theta peak in HIPP power spectra. In microarousal, a rise in HIPP MS close to the level of the MS HIPP drive appeared together with elevated, sharp theta coherence and strong theta power in both structures. These findings are in agreement with predictions of a computational model (Wang 2002) indicating that even though MS neurons possess the membrane machinery to generate rhythmic firing in the theta range (Serafin et al. 1996) robust theta synchronization only emerges with the addition of a second GABAergic population, which in the present case is the HIPP GABAergic network projecting back to the MS. Theta burst firing in single MS units cooccurring with nontheta HIPP LFP has been observed in the very first reports on MS theta activity (Petsche et al. 1962) A detailed, quantitative account of such MS cells by Sweeney (1992) documented strong burst firing, equivalent to that in theta states, in 8% of MS neurons during SWS in headrestrained rats and in 20% during non theta state under urethane anesthesia. Comparable, systematic investigations of this problem in freely moving rats have been lacking as most studies focused on theta states (Dragoi et al. 1999; Ford et al. 1989; King et al. 1998) or on spike wave ripple associations in SWS (Dragoi et al. 1999; Jinno et al. 2007; Vandecasteele et al. 2014) Weaker or transient theta rhythmic MS unit autocorrelograms have been frequently presented, however (Alonso et al. 1987; Apartis et al. 1998; Dragoi et al. 1999; Dutar et al. 1995; Macadar et al. 1970; Ranck 1976) These findings from nontheta states have been interpreted in the fr amework of the MS theta pacemaker hypothesis by contending that the lack of effect of MS theta rhythm on HIPP LFP was due either to a lack of synchrony or a weak ascending compound theta signal resulting from reduced number of burst firing MS neurons. The present investigation quantified, for the 37

PAGE 38

first time, the effect of MS input on HIPP LFP, and found detectable, i.e. significant (see Methods), MS HIPP GC in 38% of MS neurons during SWS. Critically, on group average, MS HIPP GC value during SWS was not si gnificantly below the level of MS HIPP GC during microarousals, with the latter being accompanied by strong theta LFP in the hippocampus. Thus, weak subthreshold MS to HIPP input is not sufficient to explain the lack of theta LFP during SWS. On the other hand, distribution of GC over a wide frequency range in individual neurons and substantial variability across neurons, indicate that a lack of a synchronized ascending theta signal generated by MS network may underlie the absence of HIPP theta. Striking d ifferences between SWS and microarousals were observed, however, at the level of theta influence in the opposite direction carried by the descending HIPP MS GABAergic pathway. Activation of descending HIPP MS theta drive during microarousal did not change the magnitude of MS HIPP. Rather, it led to more regular rhythmic MS bursts, sharpened the MS HIPP GC spectra, and shifted peak frequency in the MS HIPP GC spectra to ~6Hz, to synchronize with the peak frequency in HIPP MS GC spectra. These results are con sistent with the role of GABAergic input in enhancing MS neuronal synchrony (Wang 2002) They also suggest the possibility of an arousal dependent mechanism controlling MS HIPP theta coupling. Under low arousal state such as SWS, MS theta drive to HIPP is gated out, as evidenced by the significant MS HIPP drive and a lack of HIPP theta activity during SWS. Under higher arousal state such as microarousal, the HIPP theta oscillator activates and becomes responsive to MS theta drive resulting in HIPP theta activity and the subsequent HIPP MS GAB A ergic drive which in turn affects MS synchrony and firmly establishes the MS HIPP theta network. 38

PAGE 39

The specific mechanism of the HIPP gating operation is not known. Arousal involves activation of several subcortical, aminergic and cholinergic, neurotransmitter systems which are not directly involved in rhythm generation but exert important modulatory effect on oscillating neural networks in HIPP. Wake inducing histaminergic and norepinephrinergic neurons densely innervate all sectors of the hippocampus and were shown to modify theta by acting directly on their respective receptors in HIPP (Hajos et al. 2003; Masuoka and Kamei 2007) Systemic increase of histamine and norepinephrine release was also shown to enhance theta activity (Hajos et al. 2008; Kocsis et al. 2007) either by directly modifying the HIPP network or by indirectly through activation of MS cholinergic neurons (Gorelova and Reiner 1996) leading to increased acetylcholine release in HIPP (Bacciottini et al. 2002) While chol inergic transmission from MS to HIPP is too slow to directly drive theta, cholinergic tone enhances theta activity. Selective lesion of MS cholinergic neurons, sparing MS GABAergic neurons, does not affect theta frequency but dramatically reduces theta amp litude (Apartis et al. 1998; Lee et al. 1994) As shown recently using optogenetic stimulation (Dannenberg et al. 2015; Vandecasteele et al. 2014) the primary effect of th e activation of MS cholinergic neurons on HIPP LFP is a strong suppression at delta and 1025Hz frequencies surrounding the theta band which leads, indirectly, to increased theta coherence and theta/delta ratio. The effect was larger in SWS than waking and even larger under urethane anesthesia. In contrast, selective optogenetic stimulation of parvalbuminexpressing GABAergic neurons in the MS had its largest effect on HIPP LFP at ~10 Hz (Dannenberg et al. 2015) W hen recruited by cholinergic activation, this group of MS cells fire in synchrony with HIPP theta at lower frequencies. Although identification of the type of neurons is impossible when using extracellular recordings, the rate and characteristics of firing (King et al. 1998; Matthews and 39

PAGE 40

Lee 1991) suggest that the neurons engaged in theta rhythmic firing in our study were GABAergic. Thus the shift in the MS HIPP GC peak frequency from high theta in SWS to lower thet a band when HIPP theta is present (in this study) might be analogous to the 15 Hz to 4 Hz shift in their effect on HIPP LFP (Dannenberg et al. 2015) when activated under urethane either selectively or recruited by cholinergic neurons acting on both MS and HIPP networks. Several other lines of evidence further suggest that brief activation of these neurotransmitters may be responsible for microarousals. Activation of cortically projecting cholinergic neurons in the nucleus basalis via local microinfusion of histamine (Luo and Leung 2009) or norepinephrine (Pillay et al. 2011) lead to brief episodes of cortical electroencephalographic ( delta suppression) and behavioral (spontaneous head and limb movements, sporadic crawling) emergence from light anesthesia that resembled microarousals observed in natural sleep (Halasz et al. 1979; Schieber et al. 1971) Pro arousal features of a subset of slow firing, putative cholinergic, MS neurons were also demonstrated in freely behaving rats (Zhang et al. 2011) These cells showed rapid response (15 30ms) to auditory stimuli and were activated during transient (1.55s) arousal epochs in SWS associated with smallamplitude irregular HIPP activity (Jackson et al. 2008; Jarosiewicz et al. 2002) Interestingly, during theta states the firing of these neurons followed, rather than lead, HIPP theta oscillations, leading to the hypothesis that they were driven by the descending hippocamposeptal input (Toth and Freund 1992) The effect of descending theta drive on single unit burst firing was also shown under urethane anesthesia in response to brief (10s) sensory induced (tail pinch) arousal in subsets of neurons in the MS (Hangya et al. 2009) and posterior hypothalamus (Kocsis et al. 1999; Kocsis and Kaminski 2006) 40

PAGE 41

An important limitation of this study is related to the issue of the topography of projections to the HIPP arising from the MS and the fact that the phase and coherence of theta oscillations systematically change across the septotemporal axis (Long et al. 2015; Penley et al. 2012) This raises the possibility tha t cells recorded at different locations within the MS that project differentially across the septotemporal axis may not have their strongest relationship to theta recorded at the single septal HIPP recording site, but instead maintain a strong causal relat ionship with theta oscillations generated at more temporal levels. Further uncertainty may be related to the HIPP LFP recorded in one single location. LFPs are usually considered to reflect synaptic currents in the immediate vicinity of the electrode, alth ough more distant sources may also have significant contributions through volume conduction (Kajikawa and Schroeder 2015) Low frequency components have relatively larger spatial reach (Leski et al. 2013) and thus the variablity in the contribution of different hippocampal regions could certainly have contributed to the results. 41

PAGE 42

Figure 31. Single unit recording in the MS. A. Histological reconstruction of the electrode track (blue arrow) in the MS in one experiment. B. Waveform of 4 simultaneously recorded units in the MS; three units (Neuron 13, green) were separated from 2 wires of tetrode #1, the 4th (Neuron 4, blue) from a single electrode of tetrode #2 (y axis use the same scale for all units). C. Three traces of original recording from tetrode #1 (T1a and T1b) and from tetrode #2 (T2a) in the MS, spike trains of 4 separated units (N1 3 and N4), and HIPP LFP during a 5s recording. aca: anterior commissure, CPu: caudal putamen, LS: lateral septum, LV: lateral ventricle, MS: medial septum. 42

PAGE 43

Figure 32. MS and hippocampal activity during SWS and microarousals. A. Sample recording of MS spikes (top trace ), hippocampal LFP (middle), and neck muscle EMG (bottom). Note sudden switch during a short (2.4s) microarousal (shaded area) from irregular spike train to theta burst in MS, from large amplitude irregular activity to theta waves in HIPP LFP, and from sta ble, low tone to phasic motor activity in the EMG. B. Power spectra of hippocampal LFP (top panels) and MS activity (bottom) during SWS (left panels) and microarousals (right); power spectra were autoscaled to focus on the pattern. C. Firing rate of MS uni ts in SWS and microarousal; left: group averages, right: scatterplot of individual neurons. 43

PAGE 44

Figure 33. MS HIPP relationship in SWS and microarousal in two representative neurons (A and B; B is the same neuron presented in Figure 3 2A ). Top: MS HIPP c oherence, middle: MS HIPP GC, bottom: HIPP MS GC during SWS (red) and microarousal (blue). Note dramatically increased HIPP MS theta GC in both neurons during microarousals, reaching the level of MS HIPP GC; in contrast, MS HIPP GC is associated with no ch ange (A) or moderate change (B). Note also the similarity between the frequency distribution (shape of the spectra) of coherence and HIPP MS GC in microarousal, in contrast to MS HIPP during SWS which extends over a wider band. 44

PAGE 45

Figure 34. Coherence an d GC of MS unit and HIPP LFP activity. A D. Comparison of MS HIPP coherence in SWS and microarousals. A. Percentage of neurons with significant coherence, B. Group averages of coherence, C. Coherence of individual neurons, D Group averages of peak frequency of coherence in different states. E H. Comparison of MS HIPP and HIPP MS GC in SWS and microarousal. E. Percentage of neurons with significant GC, F. Group averages of GC, G. GC of individual neurons, H. Group averages of peak frequency of GC in differe nt states. '*' and '**': significance indicators representing 'p<0.05' and 'p<0.005', respectively. 45

PAGE 46

Figure 35. Further characterization of GC. A. Change in MS HIPP GC after transitioning from SWS to microarousal, relative to GC in SWS; each dot repr esents one neuron; red dots represent neurons in which MS HIPP GC was significant in microarousal, but not in SWS. B and C. Relationship between firing rate and GC during microarousal. B. MS HIPP and firing rate, C. HIPP MS and firing rate; red dots repres ent neurons with significant GC. 46

PAGE 47

CHAPTER 4 THETA RHYTHMIC DRIVE BETWEEN MEDIAL SEPTUM AND HIPPOCAMPUS IN THETA VS. NON THETA STATES 4.1 Background In rats, t he hippocampus ( HIPP ) can exhibit two distinct local field potent ial (LFP) patterns, theta and non theta states which correspond to different behavioral states. A ctive exploration (AE) and rapid eye movement (REM) sleep are theta st ates while quiet waking (QW) and slow wave sleep (SWS) are non theta states In AE, animals are engaged in exploratory behavior (locomotion, sniffing, and whisking) showing voluntary motor activity. During REM sl eep, animals are immobile and atonic except for intermittent whisker and ear twitches. Those two behaviors, which are considered theta state, show low amplitude local field potentials (LFPs) and high theta (5 9 Hz) and gamma (30 55 Hz) pow er spectrum density. In QW, animals are immobile (standing or sitting quietly) or engaged in automatic stereotyped behaviors (eating, drinking, and grooming) representing low amplitude LFPs. During SWS, animals are lying immobile with eyes closed and slow regular respiratory movements. The LFPs represent highamplitude and slow waves with delta oscillations (1 4 Hz). It begins with sleep spindles (10 14 Hz) superimposed on delta oscillations. As SWS deepens, delta waves stay dominat ing although isolat ed spindles can still be observed (Gervasoni et a l. 2004). The HIPP receives input from several cortical areas as well as subcortical areas, such as medial septum (MS). Specifically, the MS and the HIPP have reciprocal pathways (Raism an 1966). The HIPP receives both GABAergic and cholinergic fibers originating in the MS via the fimbria fornix ( Freund and Antal 1988; Frotscher and Leranth 1985) whereas the HIPP projection to MS terminate s on the GABAergic neurons in the MS nuc leus (Toth and Freund 1992). The MS plays a critical role in regulating the electrical activity of the HIPP which 47

PAGE 48

provides information about the behavioral state s of the animal (Khakpai et al. 2012). In particular, theta rhythmic burst firing of MS neur ons is thought to drive lasting HIPP theta oscillations in rats during waking motor activity and REM sleep. Although the characteristics of the neurons participating in these twoway projections have been studied extensively (Dragoi et al. 1999; Ford et al 1989; King et al. 1998; Petsche et al. 1962; Sweeney et al. 1992), the mechanisms of their functional interactions are not well understood (Bland 1986; Vertes and Kocsis 1997). In an attempt to uncover the functional interactions between MS and HIPP dur ing different behavioral states we extend the research in Chapter 3 by investigating the relationship between singleunit activity in MS and LFP oscillations in HIPP during theta states (AE, REM) and nontheta states (SWS QW ). First, we analyzed the temporal activity between MS units and HIPP LFP to reveal the timing relationships between MS unit firings and HIPP LFP based on behavioral states with the Z shift method. Then, we evaluate information flows between MS and HIPP utilizing a non parametric Gr anger causality described in Chapter 2 by recording MS unit activity together with HIPP LFP during theta and nontheta states. Furthermore, power, coherence and GC are then compared between four different behavior states to assess the characteristics in th e pattern of MS HIPP interactions. 4.2 Methods 4.2.1 Experimental Procedures Detailed descriptions of the experimental procedure and related data analysis can be found in Chapter 3. Briefly, Male SpragueDawley rats were anesthetized fo r implantation of stainless steel wires for recording HIPP LFP, stainless steel screws for reference, ground, and cortical EEG recording, and multithreaded wires to record neck muscle activity (EMG). For MS unit recording, three tetrodes were mounted on individually movable microdrives and lead into a 48

PAGE 49

guide tube placed above the MS (AP +0.5mm, Lat 0.0mm, DV 3.0mm). Electrophysiological recordings started after a 7 10 day recovery period. Daily recording sessions lasted 26 hours during daylight period, in a 26x17x17c m recording box. After stable LFP and EMG recordings were attained, the tetrodes were moved slowly into the MS until discriminable unit activity were found and t racks which had at least o ne thetarhythmic single unit were considered for further analysis. The electrical signals were amplified, filtered (LFP: 0.1 100Hz, EMG: 0.13kHz, units: 6003kHz) and sampled (16bit, 10 kHz; Neuralynx, Inc.). MS single neurons were identified and extracted off line based on their amplitude and wave shape using principal component and K means clustering algorithms (Spike2, Cambridge Electronic Devices, UK). Units showing a refractory period of 2ms or higher were considered as single units. Behavior states such as AE, REM, QW, and SWS periods were identified based on different oscillation patterns of LFPs (Vanderwolf 1969; Buzsaki et al. 1992). These behavioral states were transferred to MATLAB for analysis. All neurons encountered in these recording sites, a total of 70 cells in QW, 79 cells in SWS, 80 cells i n AE, and 57 cells in REM sleep were then included in the analysis independent of their firing properties. The spike trains of identified MS units along with HIPP LFP signals in each behavioral sta te were transferred to MATLAB for analysis. 4.2.2 D ata Analysis 4.2.2.1 Phase detection and Z shift method To characterize the timing relationship of MS neuron firing to the HIPP theta rhythm for each state, the HIPP LFP was first band pass filtered in the theta frequency band (4 10 Hz), and the filtered trace was decomposed into instantaneous amplitude ( ) and phase ( ) components using the Hilbert transform (Siapas et al. 2005; Hangya et al 2009). Phase in theta oscillation was defined by the zero crossing method. Briefly, the special points were the positive to negative 49

PAGE 50

and negative to positive ze ro crossings of the band pass filtered HIPP LFP, and they were assigned instantaneous phases of 90 and 90, respectively. A spike occurring at time was assigned a phase value ( ) To assess the phase locking characteristics of a neuron, we collected the LFP phase values that corresponded to the spike times for all spikes within a time window, across all trials. In this fashion, the spike train of each neuron for each trial was converted into a sequence of unit length vectors oriented by the phase val ues of their corresponding spikes. The mean resultant vector was computed as the average of the vector sequence, and the mean direction or mean preferred phase was computed as the orientation of the mean resultant vector. Intuitively, if the firing of a g iven neuron is independent of the theta rhythm, the distribution of its phase values will be random, i.e., uniformly distributed on [ + ], and its mean resultant vector will be short. Conversely, if the firing of a given neuron is phase locked to the theta rhythm, its phase value distributions will be unimodal, and its resultant vector will be long. The presence of phase locking was evaluated by applying the Rayleigh test for circular uniformity on distribution of the spike firing along the theta phase value. A neuron was considered significantly phase locked if the distribution of spike phase angles departed from a uniform circular distribution (Rayleigh s test for circular uniformity, p < 0.005). If p is sufficiently small, the null hypothesis of unif ormity can be rejected. The alternative hypothesis is that the data are unimodal (one mean direction). Notice that the Rayleighs test is strictly a function of R as well as of n (number of spikes included). The same value of R thus leads to different sign ificance values depending on the number of spikes that are included. In practice this leads to the problem that, given enough spikes, neurons that are not convincingly phase locked lead to statistical significance at p < 0.05. We thus used p < 0.005 to avoid false positives. For 50

PAGE 51

statistical analysis, we used circStat MATLAB scripts to calculate mean direction (preferred phase) and Rayleighs Z value (Berens, 2009). If any trial (5 sec) had <6 spikes, we could not rely on Rayleigh s test (Fisher 1993), and w e chose to remove that neuron from analysis. The value of Rayleigh s Z statistic for a neuron was used as a measure of phase consistency or the strength of spike LFP phase locking. Rayleigh s Z was calculated by = (4 1) where R denotes the mean resultant length of the given phase series ( j = 1,. . ,n): = (4 2) The probability p value for Rayleighs test for circular uniformity can b e calculated as follows: = [ 1 + ( 2 ) ( 4 ) ( 24 132 + 76 9 ) ( 288) ] (4 3) After fixing HIPP LFP in the time domain, the unit was shifted relative to the field activity by different time values ( 1 s < < 1 s). Subsequently, we calculated Rayleighs Z statistics (proportion to degree of phase locking) for all time shifts ( ). Z shift was defined as the value by which shifting unit relative to LFP results in maximum degree of phase locking (the highest val ue of the Z statistics) T he highest peak of the Z value in the function of time shift s corresponds to strongest phase coupling. It appears in the positive half plane, which implies that the action potential series have to be shifted with a positive value to get maximal phase preference; thus unit leads LFP activity. Vice versa, highest peak in the negative half plane would infer the lead of the LFP signal over unit activity. Past work has used this method to determine the time delay between corresponding events in the MS unit and HIPP LFP activity (Siapas et al., 2005; Hangya et al., 2009). 51

PAGE 52

4.2.2.2 Non parametric Granger causality HIPP LFP and MS spike trains were subjected to spectral analysis. Power spectra and MS HIPP co herence spectra were estimated according to established procedures. MS HIPP interaction was further decomposed into their directional components, MS HIPP and HIPP MS, using a recently proposed non parametric Granger causality (GC) algorithm designed for point processes as well as continuous valued recordings (Dhamala et al. 2008a; b; Nedungadi et al. 2009) Similar to Chapter 3, f or the present experiment, the procedure of data analysis is as follows. First, the continuous recordings were divided into 2s nonoverlapping epochs which were treated as realizations of an underlying stochastic process. A KSPP test demonstrated that over 99% of the LFP epochs met the stationarity requirement (Kwiatkowski et al. 1992) Second, each epoch was further divided into 1ms bins where the bin size was chosen such that no more than 1 spike can be found in any bin. Third, HIPP LFP and MS spike train were subject to separate Fourier transforms; thr ough proper averaging across all the recording epochs within a behavioral state the spectral density matrix was obtained. Fourth, the spectral density matrix was factorized and combined with Gewekes spectral GC formalism to yield MS HIPP and HIPP MS in th e spectral domain (Ding et al. 2006; Geweke 1982) For statistical analysis, a random permutation procedure was used to generate the significance thresholds for coherence and GC. Specifically, for each neuron, the epoch labels for LFP and the epoch labels for spike train were permuted randomly 1000 times. Coherence and GC were computed for each of the 1000 permuted datasets. Null hypothesis distributions were constructed based on these coherence and GC values. Thresholds corresponding to p=0.01 were determined and neurons whose coherence or GC was above their respective thresholds were considered statistically significant. 52

PAGE 53

4.3 Results 4.3.1 Phase Locking between MS Units and HIPP LFPs As a measure of interaction betw een MS units and HIPP LFP, we measured spikeLFP phase locking values. This analysis correlated spike event times from the MS with LFP phase recorded in the HIPP simultaneously. Examples of phase locking to the HIPP LFP for a MS neuron (P 107/Neuron3) were depicted in Figure 4 1 for each state. The top row of Figure 4 1 describes circular distribution of phase values of MS units relative to HIPP LFP in case phase shift is not applied. The maximal phase preference distribution of the shifted MS unit relative to HIPP LFP is depicted in the bottom row of Figure 41. The von Mises concentration parameter is increased for different four states after phase shifts which represents more peaked distribution: SWS (0.0504 0.2084), QW (0.1626 0.2330), AE (0.4766 0.6618), and REM (0.4820 0.7570). The shifted MS unit is slightly more focused than the nonshifted MS unit based on von Mises distribution concentration parameter, which means higher level of phase preference. Rayleighs Z statistic is sensitive enough to measure the differences of preferred phase () histograms for a phase locked MS unit. In Figure 42, the maximal Z value ( ) as a function of time shift ( ) and corresponding time lags ( ) of a MS neuron (P 107/Neuron3) were plotted for each state. Examples of m aximal time shift between a MS unit and HIPP LFP occurred between 90 ms and 140 ms. At the population level, the Z shift analysis was depicted in Figure 43 for each state. The pseudocolor panels indicates the Z shift method w as performed for all MS units. Rayleighs Z value as a function of time shift ( ) was normalized by its maximal Z value ( ) for each neuron and was displayed in a row. The rows were sorted by in descending order. Black and white dots indicate the location of maximum coupling ( ) between for each MS unit and HIPP LFP, and the color describes whether the maximal Z value 53

PAGE 54

( ) exceeded the significance threshold (black dots ) or not (white dots). The significance threshold was elevated to p=0.005/405 by Bonferronis correction, since a hypothesi s test can be thought to occur at each of the 405 time shift values ( ) varying between 1010ms and +1010ms in 5ms increments. Based on Bonferronis correction, we found 67.09% (53/79, SWS ), 90.00% (63/70, QW ), 87.50% (70/80, AE ), and 73.68% (42/57, REM ) of MS neurons fired spikes phase locked to the HIPP LFP oscillations in theta range for each state. Figure 44A and 44 B depict the v on Mises distribution parameter and Figure 4 4C displays the maximal coupling time, Since p hase locking value only depicts the distribution of phase values of a neuron, we need ed more analyses to characterize the locking properties of individual units by fitting von Mises densities to their corresponding phase value distributions. The von Mises distribution is the circular analog of the normal distribution and is parameterized by a preferred phase (or mean direction), and concentration parameter, with larger values corresponding to more peaked (phase locked) distributions. We compared the means of preferred phases , to test the hypothesis that the mean direction of SWS ( 1.0492 rad), QW ( 2.4514 rad), AE ( 1.9392 rad), and REM ( 1.5854 rad) wer e identical or not using one way ANOVA (or Watson Williams) test and the test of hypothesis was rejected ( F (3,282)= 7.75, p=0.0001) as shown in Figure 44 A With the Watson Williams two sample test, we compared the mean directions of preferred phases be tween state pair s : SWS vs QW ( F ( 1,147)= 20.54, p< 0.0001), SWS vs. AE ( F ( 1,157)= 9.10, p =0.0030) QW vs. AE ( F ( 1,148)= 4.43, p= 0.0369), and QW vs. R E M ( F ( 1,125)= 10.32, p = 0.0017) were significantly different while the rest of state pairs SWS vs. REM ( F ( 1,134)= 2.92, p = 0.0897) and AE vs. REM ( F ( 1,135)= 1.91, p = 0.1690) were nonsignificant. In a perspective of concentration parameter , the means of SWS (0.27420.0352), QW (0.31720.0395), AE (0.48360.0577), and REM (0.65910.0871) wer e significantly 54

PAGE 55

different in one way ANOVA test ( F (3,282)=35.86, p<0.0001) as shown in Figure 44 B For ad hoc multiple comparison of the concentration parameter between state pairs SWS vs. QW was nonsignificant ( p =0.6825) whereas SWS vs. AE ( p<0.0001) SWS vs. REM ( p<0.0001) QW vs. A E ( p=0.0001), QW vs. REM ( p<0.0001), and AE vs. REM ( p=0.0001) wer e significantly different However, in case of maximally coupled time, was not significantly different among SWS (130.2544.21), QW (105.1433.77), AE (92.7544.00), and REM (118.7740.88) based on one way ANOVA test ( F (3,282)=0.63, p =0.5931) as shown in Figure 4 4 C Especially, the ad hoc multi comparison be tween state pairs for maximal shift time show ed that they were not significantly different for all pairs of states (SWS vs. QW: p=0.8307, SWS vs. AE: p=0.5545, SWS vs. REM: p=0.9831, QW vs. AE: p=0.9750, QW vs. REM: p=0.9743, and AE vs. REM: p=0.8384). 4.3.2 Non parametric Granger Causality Group averaged relative power spectra of both MS unit activity and HIPP LFP were shown in Figure 45 A D. B oth MS unit activity and HIPP LFP power spectra showed prominent theta peaks (ranges 4 10Hz) in AE and REM sleep whereas delta activity (ranges 1 3Hz) dominated HIPP LFP during SWS (Bland 1986) The QW stage showed both delta and theta peaks in power spectra of MS unit activity and HIPP LFP The firing rate of the MS units were in the range reported in earlier studies (Dragoi et al. 1999; Ford et al. 1989; King et al. 1998; Sweeney et al. 1992) and was not significantly different ( F [ 3,282]= 1.1699, p=0. 3215) among SWS ( 10.73 2.70 spikes/s ) QW (14.13343.25 spikes/s) AE ( 14.1289 2.7 1 spikes/s), and REM (14.40914.76 spikes/s) in one way ANOVA analysis ( Figure 45E ). The densit y distributions of firing rate for each state were depicted in Figure 4 5F. 55

PAGE 56

Functional MS HIPP interactions were investigated by spectral coherence and Granger causality; the latter offers the advantage of characterizing the strength of causal influences in different frequency components. Results for group averaged coherence are demonstrated in Figure 46 A D T he means of coherence were significantly different ( F [3,282]=10.8028, p<0.0001) in one way ANOVA analysis among SWS (0.02380.0094), QW (0.04050.0138), AE (0.06960.0188), and REM (0.08920.0258). Comparing coherence of each state p air by ad hoc multiple comparison test showed that SWS vs. AE ( p=0.0005), SWS vs. REM ( p<0.0001), and QW vs. REM ( p=0.0012) were significantly different while the rest of state pairs (SWS vs. QW: p =0.5092, QW vs. AE: p =0.0735, AE vs. REM: p =0.4122) were not significantly different. Moreover MS HIPP coherence was significant in 26.58%, 50.00%, 73.75%, and 87.72% of the neurons during SWS, QW, AE, and REM sleep, respectively as shown in Figure 45G. The coherence dens it y distributions for each state were plotted in Figure 4 6F. From the figure, the densit y distributions of theta states were biased to the right and of nontheta states were biased to the left. The peak frequencies of coherence were demonstrated in Fi gure 4 6G and the means of peak frequencies (SWS: 8.02560.3251, QW: 6.85920.2593, AE: 6.84330.2118, and REM: 6.96560.2530) were significantly different ( F [3,282]=18.5737, p<0.0001) in one way ANOVA analysis. Performing ad hoc multiple comparison test, the mean peak frequency of SWS was significantly different from other states: SWS vs. QW ( p<0.0001), SWS vs. AE ( p<0.0001), and SWS vs. REM ( p<0.0001) while QW vs. AE ( p=0.9998), QW vs. REM ( p=0.9560), and AE vs. REM ( p=0.9300) were not significantly diff erent. The densit y distributions of coherence peak frequency were depicted in Figure 4 6H and peak frequency of 56

PAGE 57

SWS had two peaks around 8Hz and 10Hz and peak frequencies of other states resided between 67Hz. Group averaged GC was depicted in Figure 47A D. During SWS the percentage of neurons with significant GC was 2.53% in GC from HIPP to MS and 43.04% in GC from MS to HIPP as shown in Figure 45H. GC indicated a nearly unidirectional MS HIPP drive in which MS neuronal activity affected HIPP LFP at frequencies within the theta range (group average of HIPP MS GC spectra =0.0042 and MS HIPP GC spectra=0.016) but the firing activity of these neurons was much less affected by HIPP activity as shown in Figure 45E From the Figure 4 9 A, MS HIPP GC was significantly greater than HIPP MS GC (paired t test t [78]= 3.3202, p=0 .0014), and the results delineated that causal interaction was unidirectional In QW state, 20.00% of neurons showed significant GC from HIPP to MS and 47.14% of neurons showed signif icant GC from MS to HIPP. Comparing with SWS, the percentage of significant neurons was increased from 2.53% to 20.00% as shown in Figure 45 H. Group averaged GC was 0.0103 during HIPP MS and was 0.0231 during MS HIPP. Even though, GC was increased during HIPP MS comparing with SWS, the GC between HIPP MS and MS HIPP was significantly different in paired t test ( t [69]= 3.8291, p= 0.0003) shown in Figure 49 A. These results from nontheta states indicated that causal interaction from MS to HIPP is significa ntly greater than that in the opposite direction. During theta states the percentage of neurons with significant GC was increased for both HIPP MS (65.00% in AE and 80.70% in REM sleep) and MS HIPP (61.25% in AE and 80.70% in REM sleep) directions as depicted in Figure 4 5H. T he interaction became bidirectional, with the advent of a strong descending HIPP MS GC component ( group average: 0.0317 for AE and 0.0458 for REM sleep ) and consistent ascending MS HIPP GC component 57

PAGE 58

(group average: 0.0309 for A E and 0.0376 for REM sleep) From the Figure 48A, GC from HIPP to MS was significantly different ( F (3,282)=22.9615, p<0.0001) in one way ANOVA analysis among SWS (0.0042 0.0013), QW (0.0103 0.0034), AE (0.0317 0.0084), and REM (0.0458 0.0144). Performing multiple comparison test, nontheta states had significantly different GC means from the means of theta states: SWS vs. AE ( p<0.0001), SWS vs. REM ( p<0.0001), QW vs. AE ( p<0.0004), and QW vs. REM ( p<0.0001). The mean difference within theta states (SWS vs QW: p=0.6752) and nontheta states (AE vs. REM: p=0.0646) was not significantly different by ad hoc multiple comparison. The density distributions of HIPP MS were separated to the left during nontheta states and to the right during theta state s as shown in Figure 4 8 B. The MS HIPP GC among four different states (SWS: 0.0162 0.0079, QW: 0.0231 0.0089, AE: 0.0309 0.0118, and REM: 0.0376 0.0153) was significantly different ( F (3,282)=2.6855, p =0.0469) in one way ANOVA analysis as shown in Figure 4 8C. In multiple comparison test with 95% confidence interval, SWS showed significantly different means from REM ( p=0.0434) while other pairs of state were not significantly different (SWS vs. QW: p=0.8070, SWS vs. AE: p=0.1959, QW vs. AE: p=0.7380, QW vs REM: p=0.3083 and AE vs. REM: p=0.8462). The densit y distributions of MS HIPP GC were depicted in Figure 4 8D. M oreover, in theta states HIPP MS GC and MS HIPP GC were no longer differed in paired t test ( AE: t [ 79]= 0.1807, p= 0.8571 and REM: t [56]=1.4285, p=0.1587) as shown in Figure 49 A. Figure 4 8 E H showed the GC changes of each neuron between HIPP MS GC and MS HIPP GC for different states. The gray lines mean that there is no significant change between HIPP MS GC and MS HIPP GC for each neuron. The magenta lines explain that MS HIPP GC is significantly greater than HIPP MS GC and blue lines delineated that 58

PAGE 59

MS HIPP GC is significantly lower than HIPP MS GC. In order to construct null hypothesis that each neuron is not significantly changed with 95% confidence interval, we conducted bootstrapping with 1000 random sampling with replacement. During SWS, 8 of 79 neurons (10.13%) significantly increased and 23 neurons (29.11%) significantly decreased from HIPP MS to MS HIPP. In QW, 4 of 70 neurons (5.71%) significantly increased and 20 neurons (28.57%) significantly decreased from HIPP MS to MS HIPP. During AE, 24 of 80 neurons (30.00%) significantly increased and 7 neurons (8.75%) significantly decreased from HIPP MS to MS HIPP. 15 of 57 neurons (26.3 2%) significantly increased and 8 neurons (14.04%) significantly decreased during REM sleep. Furthermore, the peak frequencies of HIPP MS GC spectra (SWS: 6.77700.4042, QW: 6.47520.2346, AE: 6.82180.2276, and REM: 6.89200.2709) were not significantly different in one way ANOVA analysis ( F (3,282)=1.3520, p=0.2578) as shown in Figure 48 E. In ad hoc multiple comparison between pairs of each state, there were no significant difference (SWS vs. QW: p=0.4911, SWS vs. AE: p=0.9964, SWS vs. REM: p=0.9571, QW vs. AE: p=0.3631, QW vs. REM: p=0.2755, and AE vs. REM: p=0.9895). The densit y distributions of peak frequency of HIPP MS GC were depicted in Figure 4 8 F. In contrast, the peak frequency of MS HIPP GC spectra (SWS: 7.79150.3693, QW: 6.48580.3945, AE: 6.61210.4309, and REM: 6.63240.5487) showed significantly different in one way ANOVA analysis ( F (3,282)=8.2838, p <0.0001) as shown in Figure 48G. Performing ad hoc multiple comparison test, SWS had significantly different GC peak frequency means from the means of other states: SWS vs. QW ( p=0.0001), SWS vs. AE ( p=0.0004), and SWS vs. REM ( p=0.0018) whereas QW vs. AE ( p=0.9757), QW vs. REM ( p=0.9709), and AE vs. REM ( p=0.9999) were not significantly different. The densit y distributions of peak frequency of MS HIPP GC were 59

PAGE 60

shown in Figure 44 H. The peak frequency of HIPP MS GC was significantly different comparing that of MS HIPP (paired t test t [78]= 4.2617, p<0.0001) during SWS as shown in Figure 4 8 B. However, other states did not represent any s ignificant difference between GC peak frequencies of MS HIPP and HIPP MS: QW ( t [69]= 0.0424, p=0.9663), AE ( t [ 79]= 0.9410, p= 0.3495), and REM ( t [ 56]= 0.8628, p= 0.3919) when paired t test was perform ed 4.4 Discussion The MS has a critical role in generat in g theta rhythm in the HIPP because lesioning or inactivating the MS disrupts theta oscillations. Previous studies have extensively studied the role of GABAergic (Hangya et al. 2009; Bender et al., 2015), cholinergic (Vandecasteele et al., 2014), and gluta matergic (Robinson et al., 2016) neurons in the MS and shown that the se neurons contributes to generate the HIPP theta rhythm or modulate the excitability of other neurons in a way that promotes their theta rhythmic firing. These results are consistent with a classical view of the MS as a major input pathway to the hippocampus, contributing to hippocampal theta rhythms. In this chapter, we have demonstrated the relationship between MS neurons and HIPP LFP in temporal and frequency domain using Z shift method and nonparametric Granger causality for different nontheta (SWS, QW) and theta (AE, REM) states. With Z shift method, we showed that the MS neurons are phase locked to the HIPP LFP. Interestingly, this phase locking between MS units and HIPP LFP was presented in all states, even when there was little apparent theta power in HIPP LFPs. Furthermore, MS unit firings were maximally phase locked to HIPP LFP t heta rhythm del aying between 92 ms and 130ms regardless of behavioral states. This phase locking between MS units and HIPP LFP supports the idea that these two structures are i nteractively connected and t he substantia l delay may reflect the time required to recruit a significant proportion of neurons into a synchronized theta network (Hangya et al., 2009). 60

PAGE 61

Hippocamposeptal feedback as well as septo hippocampal feedback is also essential for the generation and maintena nce of hippocampal theta oscillation and ha d shown in both experimental (Toth et al., 1993) and modeling (Wang, 2002) studies. Moreover, the classical view of MS theta generation has recently been challenged by in vitro data (Manseau et al., 2008), suggest ing a hypothesis of hippocampal lead over the MS in the regulation of theta. A recent study demonstrated that HCN(hyperpolarization activated and cyclic nucleotidegated nonselective cation channel) immunonegative neurons in the MS form a septal follower group, which receiv e rhythmic input s from hippocampal and/or from GABAergic MS neurons (Hangya et al., 2009). Through the nonparametric GC, we decomposed the MS HIPP synchrony into their directional components and examine the causal interactions between them within the theta frequency band during theta (AE, REM) and nontheta (SWS, QW) states. The main finding of this analysis is that there is a significant unidirectional MS HIPP influence over a wide band (2 10Hz) in nontheta states w hich switches to bidirectional theta drive during theta states with MS HIPP and HIPP MS GC being of equal magnitude. Unidirectional MS HIPP influence in SWS was accompanied by significant MS HIPP coherence (max at 8.250.46Hz), but no theta peak in HIPP power spectra. During QW, HIPP MS GC was slightly increased comparing to SWS. However, the increase is not significantly different by adhoc multiple comparison test. In theta states, a rise in HIPP MS close to the level of the MS HIPP drive appeared togethe r with elevated, sharp theta coherence and strong theta power in both structures. These findings are in agreement with predictions of a computational model (Wang 2002), indicating that even though MS neurons possess the membrane machinery to generate rhythmic firing in the theta range (Serafin et al. 1996), robust theta synchronization only emerges with the addition of a 61

PAGE 62

second GABAergic population, which in the present case is the HIPP GABAergic network projecting back to the MS. Striking differences betwe en theta and nontheta states were observed, however, at the level of theta influence in the opposite direction carried by the descending HIPP MS GABAergic pathway. Activation of descending HIPP MS theta drive during theta states did not change the magnitude of MS HIPP. Rather, it led to more regular rhythmic MS bursts, sharpened the MS HIPP GC spectra, and shifted peak frequency in the MS HIPP GC spectra to ~6 Hz, to synchronize with the peak frequency in HIPP MS GC spectra. These results are consistent wi th the role of GABAergic input in enhancing MS neuronal synchrony (Wang 2002). Recent research (Hangya et al. 2009) proposed that only a subset of MS and HIPP neurons show theta oscillation during nontheta states. Most of the neurons are not functionally coupled; only some of the neurons are synchronized with each other. The elevation of GABAergic and/or cholinergic activity pushes the nontheta states into theta states. This transition state enhance s the coupl ing between MS and HIPP neurons. The s pread of coupling between MS and HIPP neurons leads to theta states. On the course of theta synchronization, GABAergic pacemaker neurons in MS inhibit subsets of hippocampal interneurons leading to the disinhibition of pyramidal cells and generate theta LF P (Freund and Antal, 1988; Toth et al., 1997). Backprojection from HIPP to MS would further enhance synchrony of MS GABAergic cells (Toth et al., 1993; Takacs et al., 2008). An important limitation of this study is related to the issue of the topography o f projections to the HIPP arising from the MS and the fact that the phase and coherence of theta oscillations systematically change across the septotemporal axis (Long et al. 2015; Penley et al. 2012). Furthermore, MS neurons represent different actions in the control of HIPP activity. Fast 62

PAGE 63

firing parvalbumin (PV) containing GABAergic neurons innervate hippocampal interneurons mediating disinhibition of pyramidal and granule cells (Hangya et al. 2009; Toth et al., 1997) while slow firing cholinergic neurons appear to modulate HIPP activity (Scotty et al., 2003). This raises the possibility that neurons recorded at different locations within the MS that project differentially across the septotemporal axis may not have their strongest relationship to theta recorded at the single septal HIPP recording site, but instead maintain a strong causal relationship with theta oscillations generated at more temporal levels. Further uncertainty may be related to the HIPP LFP recorded in one single location. LFPs are usuall y considered to reflect synaptic currents in the immediate vicinity of the electrode, although more distant sources may also have significant contributions through volume conduction (Kajikawa and Schroeder 2015). Low frequency components have relatively la rger spatial reach (Leski et al. 2013), and thus the variability in the contribution of different hippocampal regions could certainly have contributed to the results. 63

PAGE 64

Figure 41. Examples of phase locking to the HIPP LFP for a MS neuron during SWS, QW AE, and REM sleep The top row describes c ircular distribution of phase values of MS unit relative to HIPP LFP for two cycles with no phase shifts. The bottom row depicts histograms of preferred phase () for a phaselocked MS unit relative to HIP P LFP for two cycles 64

PAGE 65

F igure 42. Examples of the Rayleighs Z statistics during SWS, QW, AE, and REM sleep. The Z statistics are plotted against different time lags from 1 to 1 sec. Maximum location of the function ( ) describes the optimal delay to reach the highest phase locking, whereas maximal Z value ( ) is the strength of the maximal coupling between MS unit and HIPP LFP. 65

PAGE 66

Figure 43. Phase locking activity between MS units and HIPP LFP during SWS, QW, AE, and REM sleep. Pseudocolor panels of the populational Z shift analysis depict the Rayleighs Z value normalized with the maximal Z value ( ) as a function of time shift ( ) for each neuron. Z values for the dataset shifted b y 5ms for each neuron were displayed in rows and were sorted by maximal Z value in descending order. Black and white dots indicate the location of maximum coupling between for each MS unit and HIPP LFP ( ), and the color describes whether the maximal Z value ( ) exceeded the significance threshold (black dots) or not (white dots). 66

PAGE 67

Figure 4 4. One way ANOVA tests and multiple comparisons for further analysis. (A) Preferred imum coupling between MS unit and HIPP LFP ( ). 67

PAGE 68

Figure 45. MS and hippocampal activity during SWS, QW, AE, and REM sleep. A D. Group averaged relative p ower spectra of hippocampal LFP (top panels) and MS activity (bottom) E F. Comparison of MS units f iring rate for each state (E) and firing rate density distribution of each state (F). G H. Percentage of neurons with significant coherence (G) and GC (H) 68

PAGE 69

Figure 46. Analysis of coherence during SWS, QW, AE, and REM sleep. A D. Group a veraged coherence between hippocampal LFP and MS activity. E F. Comparison of MS HIPP coherence for each state (E) and coherence density distribution of each state (F). G H. Comparison of MS HIPP coherence peak frequency (G) and peak frequency density dis tribution (H). S ignificance indicators '**' represent s p<0.005. 69

PAGE 70

Figure 47. Analysis of GC during SWS, QW, AE, and REM sleep. A D. Group averaged GC between hippocampal LFP and MS activity. E H. GC changes of individual neurons for each state. 70

PAGE 71

Figure 48. Further analysis of GC based on GC direction during SWS, QW, AE, and REM sleep. A. Comparison of HIPP MS. B. Density distribution of HIPP MS. C. Comparison of MS HIPP. D. Density distribution of MS HIPP. E. Comparison of peak frequency of HIPP MS. F. Density distribution of HIPP MS peak frequency. G. Comparison of peak frequency of MS HIPP. H. Density distribution of MS HIPP peak frequency. S ignificance indicators '*' and ** represent s p<0.05 and p<0.005, respectively 71

PAGE 72

Figure 49. Furt her analysis of GC A. Comparison of HIPP MS and MS HIPP GC for each state. B. Comparison of peak frequency of HIPP MS and MS HIPP GC for each state. 72

PAGE 73

CHAPTER 5 A FFECTIVE SCENE PROCESSING: L ARGE SCALE FU NCTIONAL INTERACTIONS REVEALED BY BETA SERIES CONNECTIVITY ANALYSIS 5.1 Background In this chapter we switch to a different signal modality: fMRI and consider functional connectivity in brain networks during emotional processing. Emotions are rooted in ancient cortico limbic survival circuits that generate behavioral dispositions in response to appetitive and aversive exteroceptive signals ( Lang & Bradley, 2010; LeDoux, 2012) In humans, exposure to affectively aro using stimuli triggers activity in multiple brain regions that coalesce into large scale functional assemblies ( Lindquist, Wager, Kober, Bliss Moreau, & Barrett, 2012; Pessoa & Adolphs, 2010) The amygdala and mPFC are thought to form communication hubs within these assemblies ( Kinnison, Padmala, Choi, & Pessoa, 2012) owing to their dense interconnectivity with many regions of the brain supporting perceptual motor processin g and behavioral control ( Swanson, 2000; Young, Scannell, Burns, & Blakemore, 1994) Accordingly, understanding their patterns of functional interaction with other brain regions during emotional engagement is not only a major goal of systems neuroscience but it may also hold important clues for revealing impaired network activity in various affective disorders ( Lang & Bradley, 2010) Many brain regions related to affective processing exhibit increased hemodynamic activity during both aversive and appetitive challenge ( Bradley, et al., 2015 ; Lindquist, Wager, Kober, Bliss Moreau, & Barrett, 2012; Sabatinelli, Keil, Frank, & Lang, 2013) Abundant evidence, however, also e xists for valencespecific affective networks within the human brain. Studies have reported findings consistent with the existence of a core aversive/defensive network that responds either specifically to socially aversive inputs ( Seymour, Singer, & Dolan, 2007) or to aversive information in general ( Hayes & Northoff, 2011) The amygdala and the anterior insula are often mentioned as crucial components of this core aversive network ( Fan, et al., 73

PAGE 74

2011) On the other hand, studies using high arousing appetitive and aversive stimuli have reliably observed amygdala engagement for emotionally arousing stimuli, irrespective of hedonic valence ( Lang & B radley, 2010) For example, connectivity between the left amygdala and peri hippocampal regions measured by hemodynamic imaging predicted report accuracy in a memory task with emotional pictures (Ritchey, Dolcos, & Cabeza, 2008). While the amygdala is clearly responsive to emotional arousal (intensity) in general, an untested hypothesis is whether the specific signature of distributed network interactions between the amygdala and other structures exhibits motive system (i.e., appetitive versus defensive) s pecificity (Pessoa, 2014) The first goal of the present study is to test this hypothesis. For appetitive processing, converging evidence suggests a central role for the nucleus accumbens (NAcc) and mPFC. Enhanced hemodynamic responses in NAcc/mPFC have been observed to be specific to the viewing of appetitive stimuli (Sabatinelli, Bradley, Lang, Costa, & Versace, 2007) Interestingly, mPFC activity dropped below basel ine when viewing images with neutral or aversive content (Sabatinelli, Flaisch, Bradley, Fitzsimmons, & Lang, 2004) Extending this research, imagining pleasant scenes also led to NAcc and mPFC recruitment, and the functional connectivity between these regions and the amygdala uniquely characterized the imagery of pleasant affective content (Costa, Lang, Sabatinelli, Versace, & Bradley, 2010) Based on these findings and our preliminary data we hypothesize that mPFC functionally interacts with other brain structures, including areas in the perceptual motor system, d uring appetitive processing specifically. The second goal of this study is to test this hypothesis. Another important factor that influences large scale brain network responses concerns the intensity of emotional engagement. As with many other complex func tions, emotional processing exhibits substantial trialby trial variability depending on the eliciting stimuli and the functional 74

PAGE 75

state of the brain. Numerous studies indicate that the late positive potential (LPP) reliably indexes the degree of emotional arousal and processing intensity ( Cuthbert, Schupp, Bradley, Birbaumer, & Lang, 2000; Keil, et al., 2002; Schupp, et al., 2004) LPP has been shown to be associated with BOLD responses in distributed brain areas ( Liu, Huang, McGinnis Deweese, Keil, & Ding, 2012; Sabatinelli, Keil, Frank, & Lang, 2013; Sabatinelli, Lang, Keil, & Bradley, 2007) It is thus reasonable to expect that the strength of functional interactions betwe en hub regions such as amygdala and mPFC and other brain areas should be modulated by the relative intensity of emotional processing. The third goal of our study is to test this hypothesis by applying a multimodal neuroimaging approach. These goals were achieved by reanalyzing data from a previous study (Liu et al., 2012) in which we recorded simultaneous fMRI and EEG during free viewing of affective pictures containing pleasant, unpleasant, and neutral scenes. For hemodynamic data, amygdala and mPFC s eeded functional connectivity were assessed using the betaseries correlation method (Rissman, Gazzaley, & D'Esposito, 2004) LPP amplitude was estimated from the EEG data to provide an objective, electrophysiologi cal index of emotional engagement for each individual and each image presentation. Our analysis was focused on aversive versus neutral and appetitive versus neutral contrasts. Within each contrast, functional connectivity was further parsed by separating l ow and high LPP amplitude trials. 5.2 Methods 5.2.1 Experimental Procedures and D ata A cquisition Fifteen healthy volunteers with normal or corrected to normal vision participated in the experiment in exchange for either course credits or a financial incentive of $30. The experimental protocol was approved by the Institutional Review Board of the University of Florida. Written informed consent was obtained from all participants before the experiment. One 75

PAGE 76

participant withdrew from the experiment. In additio n, data from three participants were discarded due to artifacts generated by excessive movements inside the scanner. The remaining 11 participants (7 females; mean age, 20 yrs.; SD, 2.65) performed the task according to instructions and were included in the analysis. The stimuli consisted of 20 pleasant, 20 neutral, and 20 unpleasant pictures selected from the International Affective Picture System (IAPS) based on their content, as well as their normative hedonic valence and emotional arousal ratings (Lang, Bradley, & Cuthbert, 2008) The IAPS picture numbers of these stimuli were: Pleasant: 4311, 4599, 4610, 4624, 4626, 4641, 4658, 4680, 4694, 4695, 2057, 2332, 2345, 8186, 8250, 2655, 4597, 4668, 4693, 8030. Neutral: 2398, 2032, 2036, 2037, 2102, 2191, 2305, 2374, 2377, 2411, 2499, 2635, 2347, 5600, 5700, 5781, 5814, 5900, 8034, 2387. Unpleasant: 1114, 1120, 1205, 1220, 1271, 1300, 1302, 1931, 3030, 3051, 3150, 6230, 6550, 9008, 9181, 9253, 9420, 9571, 3000, 3069. The pictures were selected to cover a wide range of contents to avoid category specific brain activity, such as that evoked by only erotic scenes, and to enable robust appetitive and aversive engagements across observers. The pleasant pictures included sport scenes, romance, and erotic couples, whereas the unpleasant pictures included threat, attack scenes, and bodily mutilations. The neutral pictures included landscapes and neutral human bein gs. Across contents, pictures were matched for presence/absence of living/non living content, as well as for landscape/scene versus close up shots, and they were also matched for inhouse ratings of perceived complexity obtained from several hundreds of undergraduate students in a preliminary rating study. The mean pleasure (valence) rating for pleasant, neutral, and unpleasant pictures was 7.0, 6.3, and 2.8, respectively. The pleasant and unpleasant pictures had similar mean arousal levels (pleasant, 5.8; unpleasant, 5.9), both being higher than neutral pictures (4.2). Statistical tests of 76

PAGE 77

these means showed that the three picture categories differed in reports of hedonic valence, as expected (pleasant > neutral, t(38)=2.7, p=.01; pleasant > unpleasant, t( 38)=18.7, p<0.001; neutral > unpleasant, t(38)=9.9, p<0.001). Also as expected, rated emotional arousal did not differ between pleasant and unpleasant pictures, t(38)=1.1, p=0.27, but did differ between pleasant and neutral, t(38)=6.1, p<0.001; and unpleas ant and neutral pictures, t(38)=7.0, p<0.001. Despite these differences, it should be noted that the categories did not show strong mean differences in affective ratings. Hence, replication with a set of pictures showing greater spread between categories i n terms of hedonic valence and emotional arousal would be desirable. The experiment was implemented in an event related fMRI design using E Prime software. Each IAPS picture was centrally displayed on an MR compatible monitor for 3 seconds followed by a variable (2800 or 4300 ms) interstimulus interval. Participants viewed the images in the scanner via a reflective mirror system. There were five viewing blocks of 60 trials each. A break was given between blocks. In each block, the same 60 pictures were rep eated in different random orders. The order of picture presentation was further randomized across different participants. A fixation cross was displayed at the center of the screen. Before the start of the experiment, participants were instructed to fixate on the central cross and to view the pictures without moving their eyes. After the experiment participants were invited to rate 12 representative pictures (4 pictures within each category) they had not seen during the experiment on scales of hedonic valence and emotional arousal, using a paper and pencil version of the self assessment manikin ( Bradley & Lang, 1994 ) The entire experiment lasted ~40 minutes. We note that previous neuroimaging work examining effects of emotional scene repetition has demonstrated that repetition effects and emotion effects do not interact in terms of BOLD (Bradley et al., 2015). Similarly, the LPP emotion effect has b een robustly established as 77

PAGE 78

being unaffected by repetition (Codispoti, Ferrari, & Bradley, 2006). These considerations allowed us to treat the five blocks of trials on an equal footing. The rating of new pictures at the end of the experiment was to ensure that the subjects can properly perceive the emotional content of stimuli (see e.g., Codispoti, Ferrari, & Bradley, 2006). Because the entire experiment lasted approximately 40 minutes, this being in addition to the extensive EEG preparation time, a decisio n was made not to invite affective ratings during the experiment on a trialby trial basis. Doing so would have lengthened the session dramatically, reduced signal quality, and possibly induced more motion artifacts. MRI data were collected on a 3T Philips Achieva scanner (Philips Medical Systems). Two hundred and twelve volumes of functional images were acquired using a gradient echo echoplanar imaging (EPI) sequence during each session [echo time (TE), 30 ms; repetition time (TR), 1.98 s; flip angle, 80; slice number, 36; field of view, 224 mm; voxel size, 3.5x3.5x3.5 mm; matrix size, 64x64]. The slices were acquired in ascending order and oriented parallel to the plane connecting the anterior and posterior commissure. EEG data were recorded using a 32 c hannel MR compatible EEG system (Brain Products GmbH). Thirty one sintered Ag/AgCl electrodes were placed on the scalp according to the 10 20 system, and one additional electrode was placed on subjects upper back to monitor electrocardiograms (ECG). The r ecorded ECG was used to detect heartbeat events that subsequently aided in the removal of the cardioballistic artifacts. The EEG channels were referenced to the FCz electrode during recording. EEG signal was recorded with an online 0.1~250 Hz bandpass fil ter and digitized to 16 bit at a sampling rate of 5 kHz. The EEG recording system was synchronized with the scanners internal clock throughout the recording session to ensure the successful removal of the gradient artifact in subsequent analyses. 78

PAGE 79

5.2.2 Data Preprocessing The fMRI data were preprocessed using SPM5 (http://www.fil.ion.ucl.ac.uk/spm/). The first five volumes in each experimental session were discarded to eliminate transient effects. Slice timing was corrected using sinc interpolation to accou nt for differences in slice acquisition time. The images were then corrected for head movement by spatially realigning them to the sixth image of each session, normalized and registered to the Montreal Neurological Institute (MNI) template, and resampled t o a spatial resolution of 3 x 3 x 3 mm. The transformed images were smoothed by a Gaussian filter with a fullwidth at half maximum of 8 mm. The low frequency temporal drifts were removed from the functional images by applying a highpass filter with a cutoff frequency of 1/128 Hz, and the global signal was removed by dividing every voxel in a slice by the estimated global signal value. Brain Vision Analyzer 2.0 (Brain Products GmbH) was used to remove scanner artifacts in the EEG data. The gradient artifac ts were removed by using a modified version of the original algorithms proposed by Allen et al. ( Allen, Josephs, & Turner, 2000) Briefly, an artifact template w as created by segmenting and averaging the data according to the onset of each volume within a sliding window consisting of 41 consecutive volumes, and subtracted from the raw EEG data. The cardioballistic artifacts were removed by using an average artifact subtraction method ( Allen, Polizzi, Krakow, Fish, & Lemieux 1998) In this method R peaks were detected in the low pass filtered ECG signal and used to construct a delayed average artifact template over 21 consecutive heartbeat events. The average artifact template was subtracted from the original EEG signal in a sliding window approach. The EEG data after these two steps were low pass filtered with the cutoff set at 50 Hz, downsampled to 250 Hz, and re referenced to the average reference. These data were then exported to EEGLAB ( Delorme & Makeig, 2004) and SOBI (Second Order Blind Identification) ( Belouchrani, Abed Meraim, Cardoso, & Moulines, 79

PAGE 80

1993) was applied to further correct for eye blinking, residual cardioballistic, and movement related artifacts. The artifacts corrected data were then epoched from 300 ms to 2000 ms with 0 ms representing image onset. The pre stimulus baseline was defined as 300 to 0 ms for ERP analysis ( Liu, Huang, McGinnis Deweese, Keil, & Ding, 2012) 5.2.3 Seed R egion S election F MRI activation by contrasting pleasant versus neutral pictures and unpleasant versus neutral pictures employing the gen eral linear model (GLM) approach has reported on the same data ( Liu, Huang, McGinnis Deweese, Keil, & Ding, 2012) The publishe d activation maps became the basis for selecting mPFC and amygdala seed regions in the present study. For mPFC, a sphere of 5 mm in radius centered at the most activated voxel in mPFC (MNI coordinate [9, 60, 3]) under the grouplevel contrast of pleasant v ersus neutral was selected as the seed region. For amygdala, the spheres for the left and right amygdala were similarly chosen under the grouplevel contrast of unpleasant versus neutral (center coordinates for left and right amygdala: [ 21, 0, 18] and [21, 0, 18]). Because left and right amygdala yielded similar connectivity patterns they were combined into a single amygdala seed. As discussed in the Introduction of this chapter the choice of mPFC and amygdala as seed regions was well grounded, reflecti ng their distinct involvement in the processing of appetitive and aversive pictures ( Sabatinelli, et al., 2011; Sabatinelli, Lang, Keil, & Bradley, 2007) Seed regions were also based on findings with an earlier analysis of this data ( Liu, Huang, McGinnis Deweese, Keil, & Ding, 2012) in which mPFC and amygdaloid BOLD activation were found to covary with the LPP in a content specific fashion. 5.2.4 Functional C onnectivity We applied the betaseries correlation method to assess functional interactions between the seed region and other brain regions ( Rissman, Gazzaley, & D'Esposito, 2004; Rissman, 80

PAGE 81

Gazzaley, & D'Esposito, 2008) In this method beta values representing the extent of brain acti vation were estimated for each single trial using the GLM approach. For each participant 300 covariates of interest (5 sessions with 60 picture trials each) were introduced into the GLM design matrix to model the hemodynamic response elicited by each pictu re presentation ( Axmacher, Schmitz, Wagner, Elger, & Fell, 2008 ; Chadick & Gazzaley, 2011; Rissman, Gazzaley, & D'Esposito, 2004) Nuisance factors such as head movements were included in the model as additional regressors of no interest. For a given group of trials the beta series from the seed region was correlated with those from every other voxel in the brain using Pearsons correlation, with higher correlation coefficients assumed to indi cate stronger inter areal functional connectivity. The correlation coefficients were standardized into z scores using Fishers rto Z transform for further statistical analysis ( Rissman, Gazzaley, & D'Esposito, 2004) 5.2.5 Single trial E stimation of LPP EEG data from channel Pz was subject to ERP single trial estimation ( Liu, Huang, McGinnis Deweese, Keil, & Ding, 2012) using the Analy sis of Single trial ERP and Ongoing activity (ASEO) method ( Xu, et al., 2009) Briefly, single trial EEG data are modeled as comprised of tw o parts: ERP and ongoing activity. The ERP part contains multiple components whose amplitude and latency vary from trial to trial. The ongoing activity part is assumed to be an autoregressive stochastic process. Based on this model, ASEO estimates single t rial ERPs using an iterative maximum likelihood approach. From each single trial ERP the magnitude of LPP for that trial was measured. These LPP amplitude estimates from each subject were then used to sort the trials and divide them into high and low LPP t rials within either pleasant or unpleasant image categories using median split. 81

PAGE 82

We note that a median split was preferred over alternative methods such as quartiles or tertiles, as it maximized the number of trials included in the analysis, thereby enhanc ing statistical power. An additional advantage of the median split in this small sample of observers is that it enables a face (quasi splithalf) reliability check: Reliable network composition across halves of trials would be supported by the finding that networks (both appetitive and aversive) for high versus low LPP trials differ in strength and extent but not in the gross composition of regions. Using LPP single trial amplitudes to quantify the emotional intensity for each trial instead of self report ( affective ratings) affords a measure that is not bounded by an (ordinal) rating scale, can be objectively measured, can vary freely with multiple exposure to the same picture, and more directly reflects the brain processes mediating emotional engagement, c ompared to affective ratings. 5.2.6 Statistic al I nference Two classes of group level random effects analyses were conducted via paired t tests on the Z transformed correlation maps of individual subjects: (1) pleasant versus neutral scenes and unpleasant versus neutral scenes (valence contrast) and (2) high LPP group versus neutral and low LPP group versus neutral within either pleasant or unpleasant image category (intensity contrast). The statistical threshold for all secondlevel analyses was set at p < 0.05 corrected for false discovery rate (FDR). For the valence contrast clusters containing more than 10 significant contiguous voxels were shown. For the intensity contrast, because of reduced statistical power owing to smaller numbers of trials used, clusters containing more than 5 contiguous significant voxels were shown. 82

PAGE 83

5.3 Results 5.3.1 MPFC seeded C orrelation M aps Contrasting pleasant versus neutral scenes revealed increased functional connectivity between the mPFC and higher order executive structur es and motor control areas (Figure 51A), including bilateral dorsolateral prefrontal cortex, supplementary motor area, anterior cingulate cortex, and left orbitofrontal cortex. Furthermore, mPFC interacted with the left calcarine sulcus and several extras triate visual cortical areas such as the left cuneus, lingual gyrus, and superior occipital regions. Several regions often associated with attention and perception were also functionally coupled with the mPFC during pleasant scene viewing, including the superior and inferior parietal cortex, fusiform gyrus, and the temporal pole. Regions often associated with emotion related mobilization for action showed strong appetitive mPFC connectivity, including the bilateral amygdala, hippocampus, as well as the prec entral and postcentral gyri. Strikingly, there was no increase in functional connectivity between mPFC and other brain regions when unpleasant scenes were contrasted against neutral ones (Figure 51B), consistent with our hypothesis that mPFC centered netw orks exhibit specificity to appetitive content. Table 51 lists the regions exhibiting heightened mPFC linked connectivity during emotional scene viewing. 5.3.2 Amygdalaseeded C orrelation M aps For the pleasant versus neutral contrast, the amygdala exhibit ed increased connectivity with the left orbitofrontal cortex, bilateral dorsolateral prefrontal cortices, as well as areas in the extended visual cortex along the ventral stream, including left peri calcarine, right precuneus and left cuneus, as well as th e right inferior temporal gyrus (Figure 52A). Additional regions in the connectivity map included the caudate, the anterior cingulate cortex, left insula, and hippocampus. Contrasting unpleasant versus neutral viewing (Figure 52B) revealed a much broader network of areas that exhibited increased functional connectivity with the amygdala 83

PAGE 84

including striate and extrastriate visual cortices, areas in the temporal lobe (superior/middle/inferior temporal gyrus), and areas in dorsolateral prefrontal cortices, or bitofrontal cortex, and parietal cortices. Thalamus, hippocampus, parahippocampal gyrus, insula, anterior cingulate cortex, motor cortices, and somatosensory cortices also exhibited heightened interaction with the amygdala during aversive processing. From Table 52 the number of regions exhibiting increased amygdala connectivity during the viewing of unpleasant scenes is more than twice that during the viewing of pleasant pictures. These findings, consistent with our hypothesis, provide large scale network evidence supporting the notion that amygdala forms distinct profiles of functional links with motive specific brain regions during appetitive and aversive processing. 5.3.3 LPP defined C orrelation M aps To examine whether seed based functional connectivity is modulated by the intensity of emotional engagement, we used single trial LPP amplitudes estimated from simultaneously recorded EEG as an index to divide the trials within a picture category into two groups of trials: large and small LPP amplitudes. The large and small LPP trials from pleasant and unpleasant categories were then separately contrasted against the trials from the neutral category. Regarding the mPFC seeded correlation maps during the viewing of pleasant scenes, our analyses demonstrated t hat trials with large LPP amplitudes were associated with extensive interactions involving precentral/postcentral gyri which were often associated with emotion related mobilization for action (Figure 5 3A). Functional connectivity also extended to the fusi form gyrus associated with higher level visual processing and perception and the anterior cingulate cortex related to the modulation of emotional responses. Additional regions in the connectivity map included the supplementary motor area, middle/inferior t emporal cortices and right occipital cortex. By contrast, the mPFC seeded correlation maps revealed no significant 84

PAGE 85

interactions on trials with small LPP amplitudes, suggesting that the formation of large scale networks reflects the extent of emotional enga gement (Figure 5 3B). Mirroring our previous analyses, there was no increase in mPFC mediated functional connectivity during the viewing of unpleasant scenes, regardless of LPP amplitude (Figure 53C and 53D). This observation, together with Figure 51B, provides further support for the notion that mPFC forms a key component in a broadly distributed network that responds selectively to appetitive stimuli. Table 53 lists the regions with increased mPFC functional connectivity during emotional scene viewing separately for the large and small LPP groups. The amygdala seeded connectivity map during viewing of pleasant scenes included the right temporal pole and left dorsolateral prefrontal cortex for trials with large LPP amplitudes (Figure 5 4A); it contain ed no regions for trials with small LPP amplitudes (Figure 5 4B). During the viewing of unpleasant scenes the amygdalaseeded functional connectivity maps for the large LPP group (Figure 54C) contained large portions of the striate and extrastriate visual cortices, temporal lobe structures, as well as higher order executive structures including the dorsolateral prefrontal cortex, orbitofrontal cortex, and anterior cingulate cortex. Bilateral insula, left parietal cortex, left hippocampus, left parahippocam pal gyrus, anterior cingulate cortex motor cortex, and somatosensory cortex also exhibited heightened interaction with the amygdala during more intense aversive processing. Only a small subset (Figure 54D) of these areas remained in the amygdalaseeded co rrelation map when examining trials with small LPP amplitudes. Table 54 lists the regions that exhibited increased functional connectivity with the amygdala during the viewing of pleasant and unpleasant scenes, separately for the large and small LPP trials. 85

PAGE 86

5.4 Discussion Viewing emotional pictures leads to heightened hemodynamic responses in a host of brain areas ( Sabatinelli, et al., 2011) Past work has mainly focused on characterizing the coactivation of these areas. Here we aimed to map distributed interaction patterns among brain areas during exposure to natural scenes depicting pleasant, neutral, and unpleasant content. Motivated by the hypothesis that different coalitions of brain regions are assembled into large scale networks depending on the motive system being engaged (e.g., appetitive vs. aversive), we first correlated single trial beta values for each of the three picture categories. This approach allowed us to uncover distinct large scale functional connectivity profiles by revealing the amount of correlated hemodynamic activity between core emotion processing structures (such as the amygdalo id complex and mPFC) and extended brain regions involved in perceptual motor processing, memory and behavioral control. To further assess the modulation of these correlations by emotional intensity, we used single trial LPP from the simultaneously recorded EEG to index the varying intensity of emotional engagement within a hedonic valance and examined functional connectivity separately for large LPP trials and small LPP trials. 5.4.1 MPFC and I ts P atterns of F unctional C onnectivity The mPFC was hypothesized to be the core component of a large scale network sensitive to appetitive processing, important for coordinating sensory and motor areas to support amplified sensory processing and response mobilization towards appetitive cues ( Price, 1999) The present functional connectivity analyses are consistent with evidence from previous studies that mPFC networks exhibit preferential recruitment during appetitive processing ( Costa, Lang, Sabatinelli, Versace, & Bradley, 2010; Sabatinelli, Bradley, Lang, Costa, & Versace, 2007; Sabatinelli, Flaisch, Bradley, Fitzsimmons, & Lang, 2004) In a previous study with the same data (Liu et al., 2012), the LPP amplitude elicited by pleasant pictures co varied with BOLD in the mPFC, 86

PAGE 87

occipitotemporal junction, amygdala, and precuneus. In the present study, t he mPFC mediated appetitive network included wide spread cortical regions in volved in visual processing (ventral visual stream) and action selection, in addition to numerous deep brain regions such as the hippocampal, amygdaloid complexes and the ACC responsible for memory and cognitive control. By contrast, our analyses revealed no mPFC linked areas that were preferentially engaged during the processing of aversive stimuli. These findings support the existence of a unique and dedicated mPFC mediated functional neural organization for appetitive processing. The absence of mPFC rela ted connectivity in the unpleasant condition strongly suggests that the appetitive network outlined above (Figure 51 and Table 51) is not driven by stimulus salience or general arousal. Previous research has also suggested an important role for the nucle us accumbens (NAcc) in appetitive engagement both during pleasant scene viewing ( Costa, Lang, Sabatinelli, Versace, & Bradley, 2010; Sabatinelli, Lang, Bradley, Costa, & Keil, 2009) and during pleasant imagery ( Costa, Lang, Sabatinelli, Versace, & Bradley, 2010) In the present study, however, NAcc was only found to be contained in the mPFC seeded correlation maps for pleasant versus neutral contrast under relaxed threshold (p<0.07 FDR). Using the NAcc seed identified by contrasting BOLD evoked by pleasant and neutral pictures (Liu et al., 2012) we found no areas appearing in the corresponding correlation maps under p<0.05 FDR. Using the NAcc seed identified in a previous study (Heldmann, et al., 2012) (MNI coordinates: [10,12,2] and [ 10,12,2]), the corresponding correlation maps revealed very few regions under p<0.05 FDR (not shown). This suggests that mPFC is the main valence sensitive region for coordinating a network of areas involved in perceiving and mobilizing action in r esponse to pleasant motivational affordances. 87

PAGE 88

5.4.2 Amygdala and I ts P atterns of F unctional C onnectivity The amygdala was hypothesized to connect to different brain regions depending on the hedonic valence of the system being engaged. Our findings suppor t this hypothesis. Amygdala seeded coupling during appetitive processing was restricted to the visual cortex as well as areas involved in attention and executive functions (e.g., precuneus) ( Nagahama, et al., 1999) and the encoding of hedonic value such as the caudate and insula ( Naqvi & Bechara, 2010; Phan, Wager, Taylor, & Liberzon, 2002) During aversive processing, connections between the bilateral amygdala and visual cortex were stronger and more widespread. In addition, our findings suggest a unique profile of amygdala seeded functional connectivity during the processing of aversive scenes, characterized by involvement of the superior/middle temporal gyrus, superior/inferior parietal cortex, the cingulate cortex, and supplementary motor area, along with critical subcortical structures including the thalamus, hippocampus, and insula. These regions, including perceptual and motor areas as well as cortical regions involved in executive and memory processing, correspond largely with previous studies examining beta series connectivity when observers view IAPS pictures (e.g., St. Jacques, Dolcos, & Cabeza, 2010). The formation of such functiondependent flexible links between the amygdaloid body and other structures during specific motivational and behavioral conditions is in line with research in animal models ( Salzman, Paton, Belova, & Morrison, 2007) Our previous results on the same data (Liu et al., 2012) observed trial by trial correlation of LPP amplitude and BOLD for unpleasant pictures in temporal and ventrolateral prefrontal cortices, insula, and cingulate cortex. Paralleling findings for pleasant picture viewing, the beta series method (which highlights brain areas displaying systematic BOLD co variation across time) was more sensitive than the LPP BOLD coupling analysis (Liu et al., 2012), which identifies brain structures in which hemodynamic and electrophysiological reactivity co vary 88

PAGE 89

across trials. This heightened sensitivity might be related to favorable signal to noise ratios when using BOLD time series, compared to when using a mixture of BOLD and an external, noisy, variable such as singletrial estimates derived from EEG data. Future work may s ystematically compare functional networks characterized by the beta series method with activation maps reflecting coupling of BOLD and EEG derived measures. Our findings support the notion that the function of a given brain region is an emergent phenomenon that depends on its participation within broadly distributed brain networks rather than being an isolated property of neural tissue ( Pessoa, 2014; Singer, 2013 ) For example, while numerous findings suggest that ge neral emotional arousal enhances BOLD responses in the amygdaloid complex ( Lang & Bradley, 2010 ) the pattern of broad network interactions appears to exhibit motive system specificity. Indeed, a conventional fMRI activation analysis of our BOLD data via GLM showed that amygdala is activated for both unpleasant and pleasant pictures ( Liu, Huang, McGinnis Deweese, Keil, & Ding, 2012) Yet, amygdalaseeded connectivity sugge sts that this region establishes flexible interactions with other brain regions and both the anatomical specificity and breadth of these interactions differs for aversive versus appetitive processing. 5.4.3 Emotional M odulation of V isual P rocessing A consi stent finding is that both the mPFC seeded appetitive network and amygdala seeded appetitive and aversive networks included visual cortices in the current study. In particular, large portions of the ventral stream coalesced into assemblies established arou nd valence sensitive brain areas. This observation provides support for the re entry hypothesis of emotional perception. Stated generally, the re entry hypothesis predicts that emotion sensitive core structures engage sensory areas through feedback connect ions, facilitating the perceptual processing of emotionally salient stimuli ( Sabatinelli, Lang, Bradley, Costa, & Keil, 2009) 89

PAGE 90

Although the present data are limited in that they do not reflect directional information flow among brain regions, the bi directional communication associated with re entry is consistent with the coupling observed in the current study. 5.4.4 Functional C onnectivity and I ntensity of E motional E ngagement Considerable evidence exists that LPP amplitude is closely associated with the degree of emotional arousal ( Cuthbert, Schupp, Bradley, Birbaumer, & Lang, 2000; Hajcak, MacNamara, & Olvet, 2010; Keil, et al., 2002; Liu, Huang, McGinnis Deweese, Keil, & D ing, 2012) Even for neutral scenes, LPP amplitude tracks variability in relative arousal levels (e.g., LPP is higher for scenes with people than scenes without) ( Weinberg & Hajcak, 2010) For affective scenes LPP amplitudes can be used to index the intensity of emotional engagement. The mPFC seeded appetitive network and the amygdala seeded appetitive and aversive networks were generated according to the LPP amplitude estimates to identify the relation between the level of LPP amplitude estimates and seed based functional connectivity in the brain. Our findings accord with the hypothesis that strength of functional interactions between hub regions such as amygdala and mPFC and other brain areas is positively associated with the intensity of emotional processing. This in turn suggests that the LPP provides a surface electrophysiological correlate of activation in and interaction among broadly distributed brain networks ( Liu, Huang, McGinnis Deweese, Keil, & Ding, 2012; Moratti, Saugar, & Strange, 2011; Sabatinelli, Keil, Frank, & Lang, 2013) It is worth noting that Tables 53 and 54 contained fewer areas than Tables 51 and 52. Weakened statistical power due to reduced numbers of trials used to construct the entries in Tables 53 and 5 4 was a contributing factor. 5.4.5 Summary Taken together, the present data converge with animal and human findings ( Keil, et al., 2012) to demonstrate that appetitive and aversive stimulus processing are associated with 90

PAGE 91

heightened coupling between core affective structures and an array of brain areas associated with sensory and motor processing as well as executive control ( Spielberg, et al., 2012) In line with motivational and evolutionary theories of affect, emotion networks in the brain may link perceptual features reliably paired with threat or reward to support attentive behavior and ultimately mediate the preparation of the organism for goal oriented adaptive action. Although the networks emerging in response to different emotional challenges share individual components, the specific nature of motive system engagement is reflected in the distinct forms of functional coupling of var ying structures into large scale brain networks. 91

PAGE 92

Figure 51. MPFC seeded betaseries correlation maps A, An extensive network was connected to mPFC when pleasant was contrasted against neutral condition. B, No region was connected to mPFC when unpleasa nt was contrasted against neutral condition. All maps are thresholded at p<0.05 (FDR corrected); only significant clusters with more than 10 contiguous voxels are shown. HIPP, hippocampus; LNG, lingual gyrus; OCC, occipital cortex; FG, fusiform gyrus; ITG, inferior temporal gyrus; ACC, anterior cingulate cortex; PHG, parahippocampal gyrus; AMG, amygdala; TP, temporal pole; DLFPC, dorsolateral prefrontal cortex; ANG, angular gyrus. 92

PAGE 93

Figure 52. Amygdala seeded beta series correlation maps A, Regions con nected to amygdala when pleasant was contrasted against neutral condition. B, Regions connected to amygdala when unpleasant was contrasted against neutral condition. All maps are thresholded at p<0.05 (FDR corrected); only significant clusters with more than 10 contiguous voxels are shown. OFC, orbitofrontal cortex; IFG, inferior frontal gyrus ; PCu, precuneus; Cu, cuneus; STG, superior temporal gyrus; MTG, middle temporal gyrus; INS, insula; CCR, calcarine; SMA, supplementary motor area; PC, parietal cortex 93

PAGE 94

Table 51. Regions connected with mPFC during viewing of pleasant and unpleasant pictures. Anatomical Regions MNI coordinates x,y,z (Z score) Pleasant pictures Unpleasant pictures Frontal NO AREAS Dorsolateral prefrontal cortex 27, 6, 6 6 (4.65) 27, 9, 72 (4.45) Orbitofrontal cortex 27, 30, 21 (3.54) Precentral gyrus 18, 12, 69 (3.61) 63, 12, 18 (3.68) Supplementary motor area 0, 9, 66 (3.69) Temporal Middle temporal gyrus 57, 36, 12 (2.81) Inferior temporal gyrus 57, 42, 15 (3.92) 54, 45, 21 (4.33) Temporal pole 36, 15, 36 (3.75) Parietal Superior p arietal cortex 30, 69, 54 (3.70) 21, 72, 51 (3.49) Inferior parietal cortex 60, 36, 42 (3.33) Postcentral gyrus 51, 9, 39 (4.31) 48, 12, 33 (4.11) Angular gyrus 45, 63, 51 (3.00) 42, 66, 51 (3.32) Precuneus 15, 57, 51 (4.73) Supramarginal gyrus 66, 24, 24 (3.61) Occipital Superior occipital cortex 24, 66, 33 (3.09) 27, 66, 24 (3.54) Medial oc cipital cortex 24, 81, 18 (3.38) 30, 78, 24 (3.37) Inferior occipital cortex 36, 72, 9 (4.26) Lingual gyrus 18, 57, 3 (5.08) 12, 51, 6 (4.73) Calcarine 21, 72, 12 (2.78) Subcortical Caudate 18, 18, 18 (3.34) 21, 9, 24 (3.49) Anterior cingulate cortex 0, 12, 27 (4.25) Para h ippocampal gyrus /Hippocampus 27, 9, 15 ( 4.04 ) 21, 6, 21 ( 4.34 ) 94

PAGE 95

Table 52. Regions connected with amygdala during viewing of pleasant and unpleasant pictures. Anatomical Regions MNI coordinates x,y,z (Z score) Pleasant pictures Unpleasant pictures Frontal Dorsolateral prefrontal cortex 27, 33, 45 ( 4.30 ) 42, 51, 18 (4.24) 36, 24, 51 ( 4.59 ) 36, 30, 36 (4.36) Orbitofrontal cortex 27, 45, 6 ( 4.21 ) Orbitofrontal cortex 39, 27, 21 (5.33) Inferior frontal gyrus 51, 24, 12 (4.52) Inferior frontal g yrus 45, 15, 42 (4.77) Precentral gyrus 45, 6, 42 (3.43) Precentral gyrus 33, 15, 69 ( 4.30 ) 45, 3, 42 (2.80) S upplementary motor area 3, 9, 63 (3.6) Temporal Superior temporal gyrus 63, 45, 18 (5.24) 51, 9, 12 (5.04) Middle temporal gyrus 6 1 45, 1 3 (3.87) 55 13 1 8 (4.24) Inferior temporal gyrus 63, 24, 18 (4.41) Inferior temporal gyrus 42, 3, 45 ( 3.93 ) 57, 18, 18 (5.48) Temporal pole 30, 9, 27 ( 5.08 ) 48, 15, 18 (4.83) Fusiform gyrus 24, 33, 18 (4.42) 36, 30, 21 (4.11) Parietal Superior p arietal cortex 24, 48, 63 (3.83) Inferior p arietal cortex 48, 36, 51 (3.55) 42 5 5, 50 (3.85) Angular gyrus 42, 63, 45 (5.36) Postcentral gyrus 45, 24, 48 (3.60) 51, 21, 33 (3.09) Precuneus 15, 42, 3 ( 3.65 ) 15, 45, 45 (3.43) Precuneus 15, 42, 42 (4.2) Occipital Medial o ccipital cortex 30, 90, 21 (3.64) 42, 90, 0 (2.77) Inferior occipital cortex 48, 81, 3 (3.62) Lingual gyrus 12, 66, 0 (4.64) 6, 69, 0 (4.63) Calcarine 3 64 27 (3.19) 0, 99, 9 (4.34) Cuneus 6, 63, 27 ( 3.91 ) Subcortical Anter ior cingulate cortex 18, 45, 6 ( 3.27 ) 3, 33, 15 (4.01) Middle cingulate cortex 0, 15, 42 (3.72) Posterior cingulate cortex 0, 39, 30 (3.92) Parahippocampal gyrus /Hippocampus 27, 33, 3 (3.78) 33, 3, 27 (4.39) 28, 9, 2 8 (4.39) 27, 3, 24 (3.35) Insula 33, 6, 15 (4.89) 33, 21, 8 (3.93) Insula 46, 20, 8 (3.72) Caudate 12, 24, 0 ( 3.55 ) 12 15 18 ( 3. 63) Thalamus 6, 9, 0 (4.37) 6, 18, 3 (2.85) 95

PAGE 96

Figure 53. MPFC seeded betaseries correlation maps based on engagement intensity contrast. A, Regions connected to mPFC when pleasant was contrasted against neutral condition (large LPP group). B, Regions connected to mPFC when pleasant was contrasted against neutral condition (small LPP group). C D, Regions connected to mPFC when unpleasant was contrasted against neutral condition (large and small LPP groups). All maps are thresholded at p<0.05 (FDR corrected); only significant clusters with more than 5 contiguous voxels are s hown. PoCTR postcentral gyrus 96

PAGE 97

Figure 54. Amygdala seeded beta series correlation maps based on engagement intensity contrast. A, Regions connected to amygdala when pleasant was contrasted against neutral condition (large LPP group). B, Regions connect ed to amygdala when pleasant was contrasted against neutral condition (small LPP group). C, Regions connected to amygdala when unpleasant was contrasted against neutral condition (large LPP group). D, Regions connected to amygdala when unpleasant was contr asted against neutral condition (small LPP group). All maps are thresholded at p<0.05 (FDR corrected); only significant clusters with more than 5 contiguous voxels are shown. M CC, middle cingulate cortex. 97

PAGE 98

Table 53. Regions connected with mPFC for large LPP trials and small LPP trials Pleasant pictures MNI coordinates x,y,z (Z score) Large LPP Small LPP Frontal NO AREAS Precentral gyrus 18, 12, 69 (3.90) 63, 12, 21 (4.22) Temporal Middle temporal gyrus 54, 39, 12 (3.90) Inferior temporal gyrus 33, 3, 42 (3.43) Fusiform gyrus 33, 6, 39 (4.51) Parietal Postcentral gyrus 51, 9, 39 (4.88) 51, 12, 30 (4.91) Supramarginal gyrus 66, 21, 27 (3.97) Occipital Medial o ccipital cortex 36, 81, 3 (3.79) Inferior occipital cortex 39, 90, 0 (3.82) Subcortical Anterior cingulate cortex 3, 12, 27 (4.18) Unpleasant pictures Large LPP Small LPP NO AREAS NO AREAS 98

PAGE 99

Table 54. Regions connected wi th amygdala for large LPP trials and small LPP trials Pleasant pictures MNI coordinates x,y,z (Z score) Large LPP Small LPP Dorsolateral prefrontal cortex 27, 15, 51 (4.41) NO AREAS Temporal pole 27, 9, 27 (4.50) Unpleasant pictures Large LPP S mall LPP Frontal Dorsolateral prefrontal cortex 42, 51, 18 (3.79) 45, 12, 54 (4.02) Orbitofrontal cortex 3, 54, 3 (3.80) 39, 27, 21 (4.18) Inferior frontal gyrus 48, 18, 0 (3.55) Precuneus 12, 45, 48 (3.21) 15, 42, 42 (4.10) Supplementary motor area 3, 9, 63 (3.32) Temporal Superior temporal gyrus 54, 6, 6 (4.22) 57, 6, 9 (3.85) 54, 3, 9 (3.85) 60, 3, 12 (4.27) Middle temporal gyrus 51, 18, 15 (3.42) 57, 18, 18 (4.44) 54, 21, 15 (3.92) 57, 15, 15 (3.73) Inferior temporal gyrus 54, 15, 21 (4.79) 51, 18, 21 (4.30) 63, 24, 21 (4.57) 63, 24, 18 (3.98) Temporal pole 45, 15, 21 (4.46) 54, 9, 18 (3.70) 57, 6, 0 (4.10) 57, 9 3 (3.51) Fusiform gyrus 24, 33, 18 (4.01) 24, 33, 18 (4.15) Fusiform gyrus 36, 27, 21 (3.86) Parietal Superior p arietal cortex 24, 48, 63 (4.86) Angular gyrus 42, 63, 51 (4.87) 42, 63, 45 (4.85) Occi pital Lingual gyrus 24, 60, 3 (4.20) 15, 51, 9 (4.31) Lingual gyrus 18, 45, 9 (3.33) Calcarine 18, 66, 15 (4.95) 21, 69, 18 (3.90) Subcortical Anterior cingulate cortex 3, 33, 15 (3.60) 0, 30, 15 (4.79) Middle cingulate cortex 3, 15, 39 (3.23) 0, 15, 42 (3.38) Insula 39, 15, 3 (3.75) 42, 15, 3 (3.61) Insula 36, 21, 6 (4.14) Parahippocampal gyrus/Hippocampus 33, 3, 27 (4.16) 21, 39, 9 (3.17) Thala mus 6, 9, 0 (4.51) 99

PAGE 100

CHAPTER 6 CONCLUSION Neural activities generated by neural units can be measured by various modalities such as single neuron spike trains, local field potentials (LFP), electroencephalogram (EEG), or functional magnetic resonance imaging (fMRI) In order to infer connectivity between brain areas and to elucidate how neural units and network of neurons process information, we generated f unctional connectivity which captures deviations from statistical independence between neural units and we also studied e ffective connectivity which describe directional effects of one neural unit over another. A total of 4 studies are carried out in this dissertation. In the first study (Chapter 2) we introduced a non parametric Granger causality method for mixed signals, one being a continuous valued signal and the other being a point proces s, in which GC is estimated directly from Fourier transforms of data without the need for AR models. The mathematical basis o f this method is a combination of spectral density matrix factorization and Geweke's spectral formulation of Granger causality. Thus, the nonparametric GC method is able to measure the direct ional causal effects between mixed signals, namely, bivariate con tinuous valued signal and point process data. The proposed GC method is tested on simulated data of two node and three node network models. In the examples, multiple realizations of time series were generated by simulating the netw ork models. The patterns of network connectivity were shown to be correctly recovered. This provided the validation of this method for analyzing mixed signals. These results are important since there are few methods which can assess the interactions betwee n mixed signals as well as can be applied to diverse application domains. In the second study (Chapter 3) we e valuated the causal interactions between MS and HIPP within the theta frequency band in SWS and its short, waking interruptions, called microarousals. For the analysis, the nonparametric G C method tested in Chapter 2 was applied 100

PAGE 101

to decompose the MS HIPP synchrony into their directional components (Dhamala et al. 2008a; b; Nedungadi et al. 2009) The main finding is that there is a significant unidirectional MS HIPP influence over a wide band (2 10Hz) in SWS which switches to bidirectional theta drive during microarousals with MS HIPP and HIPP MS GC being of equal magnitude. Unidirectional MS HIPP influence in SWS was accompanied by significant MS HIPP coherence (max at 8.25 0.46Hz), but no theta peak in HIPP power spectra indicating a lack of theta activity in HIPP In contrast, during microarousal, a rise in HIPP MS close to the level of the MS H IPP drive appeared together with elevated, sharp theta coherence and strong theta power in both structures. Thus, GC in SWS vs. microarousal primarily differed in the level of HIPP MS. The present findings suggest a modification of our understanding of the role of MS as the theta generator in two regards. First, a MS HIPP theta drive does not necessarily induce theta field oscillations in the hippocampus, as found in SWS. Second, HIPP theta oscillations entail bidirectional theta rhythmic interactions betwe en MS and HIPP. In the third study (Chapter 4) we conducted a more thorough analysis of MS HIPP interactions which includes additional theta and nontheta states such as AE, REM sleep, and QW. In order to reveal the functional interactions betwee n MS and HIPP, we investigated the relationship between single unit activity in MS and LFP oscillations in HIPP during theta states and nontheta states. First, we analyzed the temporal activity between MS units and HIPP LFP to uncover phase locking and th e timing relationships between MS unit firings and HIPP LFP based on behavioral states. With Z shift method, we showed that the MS neurons were phase locked to HIPP LFP and this phase locking was presented in all states, even when there was little apparent theta power in HIPP LFPs. Furthermore, MS unit firings were maximally phase locked to HIPP LFP theta rhythm delaying between 92ms and 130ms regardless of behavioral states. This phase 101

PAGE 102

locking between MS unit s and HIPP LFP supports the idea that these two structures are interactively connected and th e substantial delay may reflect the time required to recruit a significant proportion of neurons into a sync hronized theta network (Hangya et al., 2009). Then, we evaluated information flo ws between MS and HIPP with a nonparametric Granger causality method. Through the non parametric GC, we decomposed the MS HIPP synchrony into their directional components and examine the causal interactions between the m within the theta frequency band during theta (AE, REM) and nontheta (SWS, QW) states. The main finding of this analysis is that there is a significant unidirectional MS HIPP influence in nontheta states which switches to bidirectional theta drive durin g theta states with MS HIPP and HIPP MS GC being of equal magnitude. Although HIPP MS GC was slightly increased during QW comparing to SWS the increase is not significantly different by ad hoc multiple comparison test. In theta states, a rise in HIPP MS c lose to the level of the MS HIPP drive appeared together with elevated, sharp theta coherence and strong theta power in both structures. Striking differences between theta and non theta states were observed, however, at the level of theta influence in the opposite direction carried by the descending HIPP MS GABAergic pathway. Activation of descending HIPP MS theta drive during theta states did not change the magnitude of MS HIPP. The phase locking and GC information flow results suggest that many neurons are not functionally coupled each other during non theta states The non theta states can be changed to theta states with t he elevation of GABAergic and/or cholinergic activity which enhances the coupl ing between MS and HIPP neurons. The spr ead of coupling between MS and HIPP neurons leads to theta states. In the middle of theta synchronization, a possible mechanism is that GABAergic pacemaker neurons in MS inhibit subsets of hippocampal interneurons leading to the disinhibition of pyramidal cells and generate theta LFP (Freund and Antal, 1988; 102

PAGE 103

Toth et al., 1997). Backprojection from HIPP to MS would further enhance synchrony of MS GABAergic cells (Toth et al., 1993; Takacs et al., 2008) For the final study (Chapter 5) we switc hed attention to fMRI data and investigated the hypotheses that mPFC is a hub in the network mediat ing appetitive responses whereas the amygdala mediat e s both aversive and appetitive processing using bet a series correlation Betaseries correlation is a novel method based on estimating brain response s to input stimuli on a trial by trial bas is Aided by this novel method w e found that t he mPFC mediated appetitive network included wide spread cortical r egions involved in visual processing (ventral visual stream) and action selection, in addition to numerous deep brain regions such as the hippocampal, amygdaloid complexes and the ACC, responsible for memory and cognitive control. By contrast, no mPFC linked areas were found to be preferentially engaged during the processing of aversive stimuli. These findings support the existence of a unique and dedicated mPFC mediated functional neural organization for appetitive processing. The absence of mPFC related connectivity in the unpleasant condition strongly suggests that the appetitive network is not driven by stimulus salience or general arousal. For the amygdalaseeded correlation map, the coupling was restricted to the visual c ortex as well as areas involved in attention and executive function and the encoding of hedonic value such as the caudate and insula during appetitive processing. During aversive processing, connections between the bilateral amygdala and visual cortex were stronger and more widespread. In addition, our findings suggest a unique profile of amygdalaseeded functional connectivity during the processing of aversive scenes, characterized by involvement of the superior/middle temporal gyrus, superior/inferior par ietal cortex, the cingulate cortex, and supplementary motor area, along with critical subcortical structures including the thalamus, hippocampus, and insula. Using the late positive potential (LPP) to index 103

PAGE 104

the intensity of emotional engagement, functional connectivity was found to be stronger in trials with larger LPP. These results demonstrate that mPFC mediated functional interactions are engaged specifically during appetitive processing, whereas the amygdala is coupled to distinct sets of brain regions during both aversive and appetitive processing. The strength of these interactions varies as a function of the intensity of emotional engagement. 104

PAGE 105

REFERENCES Akaike H (1974) A new look at the statistical model identification. IEEE Trans actions on Automat ic Cont rol, 19, 716 723. Allen, P. J., Josephs, O., & Turner, R. (2000). A Method for Removing Imaging Artifact from Continuous EEG Recorded during Functional MRI. Neuroimage 12, 230239. Allen, P. J., Polizzi, G., Krakow, K., Fish, D. R., & Lemieux, L. (1998). Identification of EEG Events in the MR Scanner: The Problem of Pulse Artifact and a Method for Its Subtraction. Neuroimage 8, 229239. Alonso, A ., Gaztelu J M ., Buno W ., Jr., & GarciaAustt E (1987) Cross correlation analysis of septohippocam pal neurons during theta rhythm. Brain Res earch, 413, 135146. Apartis E ., Poindessous Jazat F R ., Lamour Y A ., & Bassant M H (1998) Loss of rhythmically bursting neurons in rat medial septum following selective lesion of septohippocampal choliner gic system. J Neurophysiol 79, 16331642. Axmacher, N., Schmitz, D. P., Wagner, T., Elger, C. E., & Fell, J. (2008). Interactions between medial temporal lobe, prefrontal cortex, and inferior temporal regions during visual working memory: a combined intra cranial EEG and functional magnetic resonance imaging study. J Neurosci 28, 73047312. Bacciottini, L ., Passani M B ., Giovannelli L ., Cangioli I ., Mannaioni P F ., Schunack, W ., & Blandina P (2002) Endogenous histamine in the medial septum diagona l band complex increases the release of acetylcholine from the hippocampus: a dual probe microdialysis study in the freely moving rat. Eur J Neurosci 15, 16691680. Belouchrani, A., AbedMeraim, K., Cardoso, J. F., & Moulines, E. (1993). Secondorder blin d separation of temporally correlated sources. Proc Int Conf on Digital Sig Proc 346351. Bender F ., Gorbati M ., Cadavieco M C ., Denisova N ., Gao X ., Holman, C ., Korotkova T ., & Ponomarenko, A (2015) Theta oscillations regulate the speed of locomotion via a hippocampus to lateral septum pathway. Nat Commun, 6, 8521 Berens P. ( 2009) CircStat: a MATLAB toolbox for circular statistics. J Stat Software 31, 121. Bland B H (1986) The physiology and pharmacology of hippocampal formation theta rhyt hms. Prog Neurobiol 26, 154. Bollimunta A ., Chen, Y Schroeder C E ., & Ding M (2008) Neuronal mechanisms of cortical alpha oscillations in awake behaving macaques. J Neurosci 28, 99769988. Bradley, M. M., Costa, V. D., Ferrari, V., Codispoti, M., Fitzsimmons, J. R., & Lang, P. J. (2015). Imaging distributed and massed repetitions of natural scenes: spontaneous retrieval and maintenance. Hum Brain Mapp, 36, 13811392. 105

PAGE 106

Bradley, M. M., & Lang, P. J. (1994). Measuring emotion: the Self Assessment Man ikin and the Semantic Differential. J Behav Ther Exp Psychiatry, 25, 4959. Brillinger D R (1972) The spectral analysis of stationary interval functions. Proceedings of the Sixth Berkeley Symposium on Mathematical Statistics and Probability, Volume 1: Theory of Statistics 483513. Brovelli A ., Ding M ., Ledberg A ., Chen, Y ., Nakamura R ., & Bressler S L (2004) Beta oscillations in a large scale sensorimotor cortical network: directional influences revealed by Granger causality. Proc Natl Acad Sci U S A 101, 98499854. Buzsaki G ., Horvath, Z ., Urioste R ., Hetke J ., & Wise K. ( 1992) High frequency network oscillation in the hippocampus. Science, 256, 1025 1027. Buzsaki G (2002) Theta oscillations in the hippocampus. Neuron, 33, 325 340. Chad ick, J. Z., & Gazzaley, A. (2011). Differential coupling of visual cortex with default or frontal parietal network based on goals. Nat Neurosci 14, 830832. Chen Y ., Bressler S L ., & Ding M (2006) Frequency decomposition of conditional Granger causality and application to multivariate neural field potential data. J Neurosci Methods 150, 228237. Codispoti, M., Ferrari, V., & Bradley, M. M. (2006). Repetitive picture processing: Autonomic and cortical correlates. Brain Research 1068(1), 213220. Co nway B A ., Halliday D M ., & Rosenberg J R (1993) Detection of weak synaptic interactions between single Ia afferent and motor unit spike trains in decerebrate cats. The Journal of Physiology, 471, 379409. Costa, V. D., Lang, P. J., Sabatinelli, D., Versace, F., & Bradley, M. M. (2010). Emotional imagery: assessing pleasure and arousal in the brain's reward circuitry. Hum Brain Mapp, 31, 14461457. Cuthbert, B. N., Schupp, H. T., Bradley, M. M., Birbaumer, N., & Lang, P. J. (2000). Brain potentials in affective picture processing: covariation with autonomic arousal and affective report. Biol Psychol 52, 95111. Dannenberg H ., Pabst M ., Braganza O ., Schoch, S ., Niediek J ., Bayraktar M ., Mormann, F ., & Beck H (2015) Synergy of direct and indir ect cholinergic septo hippocampal pathways coordinates firing in hippocampal networks. J Neurosci 35, 8394 8410. Delorme, A., & Makeig, S. (2004). EEGLAB: an open source toolbox for analysis of single trial EEG dynamics including independent component ana lysis. J Neurosci Methods 134, 921. DEsposito, M., Zarahn, E., & Aguirre, G.K. ( 1999) Event related functional MRI: implications for cognitive psychology. Psychol. Bull. 125 (1) 155164. 106

PAGE 107

Dhamala, M ., Rangarajan G ., & Ding M (2008a) Analyzing infor mation flow in brain networks with nonparametric Granger causality. Neuroimage 41, 354362. Dhamala, M ., Rangarajan G ., & Ding M (2008b) Estimating Granger causality from fourier and wavelet transforms of time series data. Phys Rev Lett 100, 018701. Ding M ., Bressler S L ., Yang W ., & Liang H (2000) Short window spectral analysis of cortical event related potentials by adaptive multivariate autoregressive modeling: data preprocessing, model validation, and variability assessment. Biol Cybern 83, 3545. Ding M ., Chen, Y ., Bressler S L (2006) Granger causality: Basic theory and applications to neuroscience. In: Handbook of Time Series Analysis (Schelter B, Winterhalder M, Timmer J, eds), pp 437460. Weinheim: Wiley VCH. Dragoi G ., Carpi D ., Recce, M ., Csicsvari J ., & Buzsaki G (1999) Interactions between hippocampus and medial septum during sharp waves and theta oscillation in the behaving rat. J Neurosci 19, 61916199. Dutar P ., Bassant M H Senut M C ., & Lamour Y (1995) The septohippocampal pathway: structure and function of a central cholinergic system. Physiol Rev, 75, 393 427. Fan, J., Gu, X., Liu, X., Guise, K. G., Park, Y., Martin, L., de Marchena, A., Tang, C. Y., Minzenberg, M. J., & Hof, P. R. (2011). Involvement of t he anterior cingulate and frontoinsular cortices in rapid processing of salient facial emotional information. Neuroimage 54, 25392546. Fanselow E E ., Sameshima K ., Baccala, L A ., & N icolelis M A L (2001) Thalamic bursting in rats during different awake behavioral states. Proceedings of the National Academy of Sciences of the United States of America, 98, 1533015335. Fisher N. ( 1993) Statistical analysis of circular data. Cambridge (UK): Cambridge University Press. Ford R D ., Colom L V ., & Bland B H (1989) The classification of medial septumdiagonal band cells as theta on or theta off in relation to hippocampal EEG states. Brain Res 493, 269282. Freund, T F ., & Antal M (1988) GABA containing neurons in the septum control inhibitory interneurons in the hippocampus. Nature 336, 170173. Friston, K J (1994) Functional and effective connectivity in neuroimaging: A synthesis. Human Brain Mapping, 2, 5678. Frotscher M ., & Leranth C (1985) Cholinergic innervation of the rat hippocampus as revealed by choline acetyltransferase immunocytochemistry: a combined light and electron microscopic study. J Comp Neurol 239, 237246. 107

PAGE 108

Gervasoni, D., Lin, S. C., Ribeiro, S., Soares, E. S., Pantoja, J. & Nicolelis, M. A. L. (2004). Global fore brain dynamics predict rat behavioral states and their transitions. The Journal of Neuroscience, 24(49),1113711147. Geweke, J (1982) Measurement of linear dependence and feedback between multiple time series. Journal of the American Statistical Associat ion, 77, 303313. Geweke, J (1984) Measures of conditional linear dependence and feedback between time series. Journal of the American Statistical Association, 79 907915. Gorelova N ., & Reiner P B (1996) Histamine depolarizes cholinergic septal n eurons. J Neurophysiol 75, 707714. Hajcak, G., MacNamara, A., & Olvet, D. M. (2010). Event related potentials, emotion, and emotion regulation: an integrative review. Dev Neuropsychol 35, 129155. Hajos M ., Hoffmann, W E ., Robinson, D D ., Yu J H ., & Hajos Korcsok, E (2003) Norepinephrine but not serotonin reuptake inhibitors enhance theta and gamma activity of the septohippocampal system. Neuropsychopharmacology 28, 857864. Hajos M ., Siok, C J ., Hoffmann, W E ., Li S ., & Kocsis B (2008) M odulation of hippocampal theta oscillation by histamine H3 receptors. J Pharmacol Exp Ther 324, 391398. Halasz P ., Kundra O ., Rajna P ., Pal I ., & Vargha M (1979) Micro arousals during nocturnal sleep. Acta physiologica Academiae Scientiarum Hungar icae, 54, 112. Halliday D M ., Rosenberg J R ., Amjad A M ., Breeze P ., Conway B A ., & Farmer S F (1995) A framework for the analysis of mixed time series/point process data theory and application to the study of physiological tremor, single motor unit discharges and electromylograms. Progress in Biophysics and Molecular Biology 64, 237238. Hangya B ., Borhegyi Z Szilagyi N ., Freund, T F ., & Varga, V (2009) GABAergic neurons of the medial septum lead the hippocampal network during theta activity. J Neurosci 29, 80948102. Hayes, D. J., & Northoff, G. (2011). Identifying a network of brain regions involved in aversionrelated processing: a cross species tran slational investigation. Front Integr Neurosci 5, 49. Heldmann, M., Berding, G., Voges, J., Bogerts, B., Galazky, I., Muller, U., Baillot, G., Heinze, H.J., & Munte, T.F. (2012). Deep brain stimulation of nucleus accumbens region in alcoholism affects rew ard processing. PLoS One 7, e36572. Jackson J ., Dickson, C T ., & Bland B H (2008) Median raphe stimulation disrupts hippocampal theta via rapid inhibition and state dependent phase reset of theta related neural circuitry. J Neurophysiol 99, 30093026. 108

PAGE 109

Jarosiewicz B ., McNaughton, B L ., & Skaggs W E (2002) Hippocampal population activity during the small amplitude irregular activity state in the rat. J Neurosci 22, 13731384. Jarvis M R ., Mitra P P (2001) Sampling properties of the spectr um and coherency of sequences of action potentials. Neural Computation, 13, 717749. Jinno, S ., Klausberger T Marton, L F ., Dalezios Y ., Roberts J D ., Fuentealba P ., Bushong E A ., Henze D ., Buzsaki G ., & Somogyi P (2007) Neuronal diversity i n GABAergic long range projections from the hippocampus. J Neurosci 27, 87908804. Kajikawa, Y ., & Schroeder C E (2015) Generation of field potentials and modulation of their dynamics through volume integration of cortical activity. J Neurophysiol 11 3, 339351. Kaminski M ., Ding M ., Truccolo, W A ., & Bressler S L (2001) Evaluating causal relations in neural systems: Granger causality, directed transfer function and statistical assessment of significance. Biological Cybernetics, 85, 145157. Ke il, A., Bradley, M. M., Hauk, O., Rockstroh, B., Elbert, T., & Lang, P. J. (2002). Large scale neural correlates of affective picture processing. Psychophysiology 39, 641 649. Keil, A., Costa, V., Smith, J. C., Sabatinelli, D., McGinnis, E. M., Bradley, M M., & Lang, P. J. (2012). Tagging cortical networks in emotion: a topographical analysis. Hum Brain Mapp, 33, 29202931. Khakpai, F., Nasehi, M., Haeri Rohani, A., Eidi, A., & Zarrindast, M. R. (2012). Scopolamine induced memory impairment possible invol vement of NMDA receptor mechanisms of dorsal hippocampus and/or septum. Behav Brain Res, 231, 110. Kim S ., Putrino, D ., Ghosh, S ., & Brown E N (2011) A Granger Causality Measure for Point Process Models of Ensemble Neural Spiking Activity PLoS Comput ational Biol ogy 7(3), e1001110. King C ., Recce, M ., & O'Keefe J (1998) The rhythmicity of cells of the medial septum/diagonal band of Broca in the awake freely moving rat: relationships with behaviour and hippocampal theta. Eur J Neurosci 10, 464477. Kinnison, J., Padmala, S., Choi, J. M., & Pessoa, L. (2012). Network analysis reveals increased integration during emotional and motivational processing. J Neurosci 32, 83618372. Kocsis B ., Kaminski M (2006) Dynamic changes in the direction of the theta rhythmic drive between supramammillary nucleus and the septohippocampal system. Hippocampus 16, 531540. Kocsis B ., Bragin A ., & Buzsaki G (1999) Interdependence of multiple theta generators in the hippocampus: a partial coherence analysis. J Neurosci 19, 62006212. 109

PAGE 110

Kocsis B ., Li S ., & Hajos M (2007) Behavior dependent modulation of hippocampal EEG activity by the selective norepinephrine reuptake inhibitor reboxetine in rats. Hippocampus 17, 627633. Kwiatkowski D ., Phillips P C ., Sc hmidt P ., & Shin, Y (1992) Testing the null hypothesis of stationarity against the alternative of a unit root J Econ 54, 159178 Lang, P. J., & Bradley, M. M. (2010). Emotion and the motivational brain. Biol Psychol 84, 437450. Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (2008). International affective picture system (IAPS): Affective ratings of pictures and instruction manual. Technical Report A 8. University of Florida, Gainesville, FL Lawson V H ., & Bland B H (1993) The role of the septo hippocampal pathway in the regulation of hippocampal field activity and behavior: analysis by the intraseptal microinfusion of carbachol, atropine, and procaine. Exp Neurol 120, 132144. LeDoux, J. (2012). Rethinking the emotional brain. Neuron, 73, 653676. Lee M G ., Chrobak, J J ., Sik, A ., Wiley R G ., & Buzsaki G (1994) Hippocampal theta activity following selective lesion of the septal cholinergic system. Neuroscience, 62, 10331047. Leski S ., Linden, H ., Tetzlaff T ., Pettersen K H & Einev oll G T (2013) Frequency dependence of signal power and spatial reach of the local field potential. PLoS Comput Biol 9, e1003137. Lindquist, K. A., Wager, T. D., Kober, H., Bliss Moreau, E., & Barrett, L. F. (2012). The brain basis of emotion: a meta analytic review. Behav Brain Sci 35, 121143. Liu, Y., Huang, H., McGinnis Deweese, M., Keil, A., & Ding, M. (2012). Neural substrate of the late positive potential in emotional processing. J Neurosci 32, 1456314572. Long L L ., Bunce J G ., & Chrobak, J J (2015) Theta variation and spatiotemporal scaling along the septotemporal axis of the hippocampus. Front Syst Neurosci 9, 37. Luo T ., & Leung L S (2009) Basal forebrain histaminergic transmission modulates electroencephalographic activity an d emergence from isoflurane anesthesia. Anesthesiology 111, 725733. Macadar O ., Roig J A ., Monti J M ., & Budelli R (1970) The functional relationship between septal and hippocampal unit activity and hippocampal theta rhythm. Physiol Behav 5, 14431449. 110

PAGE 111

Manseau F ., Goutagny R ., Danik, M ., & Williams S (2008) The hippocamposeptal pathway generates rhythmic firing of GABAergic neurons in the medial septum and diagonal bands: an investigation using a complete septohippocampal preparation in vit ro. J Neurosci 28, 40964107. Masuoka T ., & Kamei C (2007) Role of hippocampal H1 receptors in radial maze performance and hippocampal theta activity in rats. Brain Res Bull 73, 231237. Matthews R T ., & Lee W L (1991) A comparison of extracell ular and intracellular recordings from medial septum/diagonal band neurons in vitro. Neuroscience, 42, 451 462. M itra P P ., & Pesaran B (1999) Analysis of dynamic brain imaging data. Biophys. J. 76, 691708. Moratti, S., Saugar, C., & Strange, B. A. ( 2011). Prefrontal occipitoparietal coupling underlies late latency human neuronal responses to emotion. J Neurosci 31, 1727817286. Nagahama, Y., Okada, T., Katsumi, Y., Hayashi, T., Yamauchi, H., Sawamoto, N., Toma, K., Nakamura, K., Hanakawa, T., Konishi, J., Fukuyama, H., & Shibasaki, H. (1999). Transient neural activity in the medial superior frontal gyrus and precuneus time locked with attention shift between object features. Neuroimage 10, 193199. Naqvi, N. H., & Bechara, A. (2010). The insula and drug addiction: an interoceptive view of pleasure, urges, and decisionmaking. Brain Struct Funct 214, 435450. Nedungadi A G ., Rangarajan G ., Jain N ., & Ding M (2009) Analyzing multiple spike trains with nonparametric Granger causality. J Comput Neurosci 27, 5564. Paxinos G ., & Watson C (1986) The Rat Brain in Stereotaxic Coordinates. Sydney: Academic Press. Penley S C ., Hinman, J R ., Sabolek, H R ., Escabi M A ., Markus E J ., & Chrobak, J J (2012) Theta and gamma coherence across t he septotemporal axis during distinct behavioral states. Hippocampus 22, 11641175. Pessoa, L. (2014). Understanding brain networks and brain organization. Physics of Life Reviews 11, 400435. Pessoa, L., & Adolphs, R. (2010). Emotion processing and the amygdala: from a 'low road' to 'many roads' of evaluating biological significance. Nat Rev Neurosci 11, 773783. Petsche, H ., Stumpf C ., & Gogolak G (1962) [The significance of the rabbit's septum as a relay station between the midbrain and the hippoc ampus. I. The control of hippocampus arousal activity by the septum cells]. Electroencephalogr Clin Neurophysiol 14, 202211. 111

PAGE 112

Phan, K. L., Wager, T., Taylor, S. F., & Liberzon, I. (2002). Functional neuroanatomy of emotion: a meta analysis of emotion acti vation studies in PET and fMRI. Neuroimage 16, 331348. Pillay S ., Vizuete J A ., McCallum J B ., & Hudetz A G (2011) Norepinephrine infusion into nucleus basalis elicits microarousal in desflurane anesthetized rats. Anesthesiology 115, 733742. P ostle, B.R., Zarahn, E., & DEsposito, M. ( 2000) Using event related fMRI to assess delay period activity during performance of spatial and nonspatial working memory tasks. Brain Res. Brain Res. Protoc. 5 (1) 5766. Price, J. L. (1999). Prefrontal cortica l networks related to visceral function and mood. Ann N Y Acad Sci 877, 383396. Raisman G (1966) The connexions of the septum. Brain 89, 317348. Ranck J B (1976) Behavioural correlates and firing repertoires of neurons in septal nuclei in unrest rained rats. In: The Septal Nuclei (DeFrance JF, ed), pp 423462. New York: Plenum Press. Rissman, J., Gazzaley, A., & D'Esposito, M. (2004). Measuring functional connectivity during distinct stages of a cognitive task. Neuroimage 23 752763. Rissman, J., Gazzaley, A., & D'Esposito, M. (2008). Dynamic adjustments in prefrontal, hippocampal, and inferior temporal interactions with increasing visual working memory load. Cereb Cortex, 18, 16181629. Ritchey, M., Dolcos, F., & Cabeza, R. (2008). Role of Amygdala Connectivity in the Persistence of Emotional Memories Over Time: An Event Related fMRI Investigation. Cerebral Cortex, 18(11), 24942504. Robinson, J ., Manseau F ., Ducharme G ., Amilhon, B ., Vigneault E ., Mestikawy S. E., & Williams S (2016) Opt ogenetic activation of septal glutamatergic neurons drive hippocampal theta rhythms. J Neurosci 36, 30163023. Rosenberg J R ., Halliday D M ., Breeze, P ., & Conway B A (1998) Identification of patterns of neuronal connectivity partial spectra, part ial coherence, and neuronal interactions. Journal of Neuroscience Methods, 83, 5772. Sabatinelli, D., Bradley, M. M., Lang, P. J., Costa, V. D., & Versace, F. (2007). Pleasure rather than salience activates human nucleus accumbens and medial prefrontal co rtex. J Neurophysiol 98, 13741379. Sabatinelli, D., Flaisch, T., Bradley, M. M., Fitzsimmons, J. R., & Lang, P. J. (2004). Affective picture perception: gender differences in visual cortex? Neuroreport 15, 11091112. 112

PAGE 113

Sabatinelli, D., Fortune, E. E., Li, Q., Siddiqui, A., Krafft, C., Oliver, W. T., Beck, S., & Jeffries, J. (2011). Emotional perception: meta analyses of face and natural scene processing. Neuroimage 54, 25242533. Sabatinelli, D., Keil, A., Frank, D. W., & Lang, P. J. (2013). Emotional per ception: correspondence of early and late event related potentials with cortical and subcortical functional MRI. Biol Psychol 92, 513519. Sabatinelli, D., Lang, P. J., Bradley, M. M., Costa, V. D., & Keil, A. (2009). The timing of emotional discriminatio n in human amygdala and ventral visual cortex. J Neurosci 29, 1486414868. Sabatinelli, D., Lang, P. J., Keil, A., & Bradley, M. M. (2007). Emotional perception: correlation of functional MRI and event related potentials. Cereb Cortex, 17, 10851091. Salz man, C. D., Paton, J. J., Belova, M. A., & Morrison, S. E. (2007). Flexible neural representations of value in the primate brain. Ann N Y Acad Sci 1121, 336354. Sameshima K ., & Baccala, L A (1999) Using partial directed coherence to describe neuronal ensemble interactions. Journal of Neuroscience Methods, 94, 93103. Schieber J P ., Muzet A & Ferriere P J (1971) [Phases of spontaneous transitory activation during normal sleep in humans]. Archives des sciences physiologiques 25, 443465. Schupp, H. T., Cuthbert, B. N., Bradley, M. M., Hillman, C. H., Hamm, A. O., & Lang, P. J. (2004). Brain Processes in emotional perception: Motivated attention. Cognition and Emotion, 18, 593611. Scotty F ., Danik, M ., Manseau F ., Laplante F ., Quirion, R ., & Williams S (2003) Distinct electrophysiological properties of glutamatergic, cholinergic and GABAergic rat septohippocampual neurons: novel implications for hippocampal rhythmicity. J Physiol 551, 927943. Serafin M ., Williams S Khateb A ., Fort P ., & Muhlethaler M (1996) Rhythmic firing of medial septum noncholinergic neurons. Neuroscience, 75, 671675. Seth A K (2005) Causal connectivity of evolved neural networks during behavior. Network Computation in Neural Systems, 16, 3554. Seymour B., Singer, T., & Dolan, R. (2007). The neurobiology of punishment. Nat Rev Neurosci 8, 300311. Siapas A G ., Lubenov, E V ., & Wilson M A (2005) Prefrontal phase locking to hippocampal theta oscillations. Neuron 46, 141151. Singer, W. (2013). C ortical dynamics revisited. Trends Cogn Sci 17, 616626. Sporns O. (2007) Brain connectivity, Scholarpedia, 2(10), 4695. 113

PAGE 114

Spielberg, J. M., Miller, G. A., Warren, S. L., Engels, A. S., Crocker, L. D., Banich, M. T., Sutton, B. P., & Heller, W. (2012). A brain network instantiating approach and avoidance motivation. Psychophysiology 49, 12001214. St Jacques, P., Dolcos, F., & Cabeza, R. (2010). Effects of aging on functional connectivity of the amygdala during negative evaluation: a network analysis of f MRI data. Neurobiol Aging, 31(2), 31527. Swanson, L. W. (2000). Cerebral hemisphere regulation of motivated behavior. Brain Res 886, 113164. Sweeney J E ., Lamour Y ., & Bassant M H (1992) Arousal dependent properties of medial septal neurons in the unanesthetized rat. Neuroscience, 48, 353362. Takacs ,V T ., Freund, T F ., & Gulyas A I (2008) Types and synaptic connections of hippocampal inhibitory neurons reciprocally connected with the medial septum. Eur J Neurosci 28, 148164. Thomson, D J (1982) Spectrum estimation and harmonic analysis. Proceedings of the IEEE, 70, 10551096. Toth K ., Freund, T F (1992) Calbindin D28kcontaining nonpyramidal cells in the rat hippocampus: their immunoreactivity for GABA and projection to the medial septum. Neuroscience, 49, 793 805. Toth K ., Borhegyi Z ., & Freund, T F. ( 1993 ) Postsynaptic targets of GABAergic hippocampal neurons in the medial septum diagonal band of Broca compl ex. J Neurosci 13,371224. Toth K ., Freund, T F ., & Miles R (1997) Free in PMC disinhibition of rat hippocampal pyramidal cells by GABAergic afferents from the septum. J Physiol 500,463474. Truccolo W ., Eden, U T ., Fellows M R ., Donoghue J P ., & Brown E N (2005) A point process framework for relating neu ral spiking activity to spiking history, neural ensemble, and extrinsic covariate effects. Journal of Neurophysiology, 93, 10741089. Vandecasteele, M ., Varga V ., Berenyi A ., Papp, E ., Bartho, P ., Venance L ., Freund, T F ., & Buzsaki G (2014) Optogen etic activation of septal cholinergic neurons suppresses sharp wave ripples and enhances theta oscillations in the hippocampus. Proc Natl Acad Sci U S A 111, 1353513540. Vanderwolf C H. ( 1969 ) Hippocampal electrical activity and voluntary movement in the rat. Electroencephalogr Clin Neurophysiol 26,407418. Vertes R P ., & Kocsis B (1997) Brainstem diencephaloseptohippocampal systems controlling the theta rhythm of the hippocampus. Neuroscience, 81, 893 926. Walden A T (2000) A unified view of multitaper multivariate spectral estimation. Biometrika, 87, 767788. 114

PAGE 115

Wang X J (2002) Pacemaker neurons for the theta rhythm and their synchronization in the septohippocampal reciprocal loop. J Neurophysiol 87, 889900. Weinberg, A., & Hajcak, G. (2010). Beyond good and evil: the time course of neural activity elicited by specific picture content. Emotion, 10, 767782. Wilson G .T (1972) The factorization of matricial spectral densities. SIAM Journal on Applied Mathematics, 23, 420426. Xu, L., St oica, P., Li, J., Bressler, S. L., Shao, X., & Ding, M. (2009). ASEO: a method for the simultaneous estimation of single trial event related potentials and ongoing brain activities. IEEE Trans Biomed Eng 56, 111121. Young, M. P., Scannell, J. W., Burns, G. A., & Blakemore, C. (1994). Analysis of connectivity: neural systems in the cerebral cortex. Rev Neurosci 5, 227250. Zarahn, E., Aguirre, G., & DEsposito, M. (1997). A trial based experimental design for fMRI. NeuroImage 6 (2) 122138. Zhang H ., Li n, S C ., & Nicolelis M A (2011) A distinctive subpopulation of medial septal slow firing neurons promote hippocampal activation and theta oscillations. J Neurophysiol 106, 27492763. Zhu L ., Lai Y C ., Hoppensteadt F C ., & He J (2003) Probing changes in neural interaction during adaptation. Neural Computation, 15, 23592377. 115

PAGE 116

BIOGRAPHICAL SKETCH Daesung Kang was born in 1978 in the Sachoen si South Korea H e earned the B.E. degree in c ontrol and i nstrumentation e ngineering at Korea Univ ersity in 2004 and earned the M.E degree at the same department He enrolled in the graduate program of e lectrical and c omputer e ngin e ering at the University of Florida in 2009 and obtained M.E. degree of e lectrical and c omputer e ngineering in 2011 and tr ansferred to biomedical e ngineering in 2012 to pursue the Doctor of Philosophy degree. H e conducted h is doctoral studies in biomedical engineering specializing in neuroimaging, signal processing and cognitive neuroscience under Dr. Mingzhou Ding. He intends to pursue his research interests in machine learning for neuroscience and continue to explore other potential applications of biosignal analysis in the clinical and commercial fields. 116