E

Vftp

BTI

a

b

c

d

e

f

g

RA

H

h0

pilepsy Research (2014) 108, 1896—1903

j ourna l h om epa ge: www.elsev ier .com/ locate /ep i lepsyres

alproate treatment normalizes EEGunctional connectivity in successfullyreated idiopathic generalized epilepsyatients

. Clemensa, S. Puskásb,∗, M. Besenyeic, N.Zs. Kovácsa,. Spisákd, S.A. Kisd, M. Emrid, K. Hollódye, A. Fogarasi f,. Kondákorg, I. Feketeb

Kenézy Gyula County Hospital, Department of Neurology, Debrecen, HungaryUniversity of Debrecen, Medical Center, Department of Neurology, Debrecen, HungaryUniversity of Debrecen, Medical Center, Department of Pediatrics, Debrecen, HungaryUniversity of Debrecen, Department of Nuclear Medicine, Debrecen, HungaryUniversity of Pécs, Department of Pediatrics, Pécs, HungaryEpilepsy Center, Bethesda Children’s Hospital, Budapest, HungaryBalassa János Hospital, Department of Neurology, Szekszárd, Hungary

eceived 25 June 2014; received in revised form 13 September 2014; accepted 29 September 2014vailable online 13 October 2014

KEYWORDSIdiopathicgeneralized epilepsy;Valproate;EEG functionalconnectivity

SummaryAim: To investigate the effect of chronic VPA treatment of EEG functional connectivity insuccessfully treated idiopathic generalized epilepsy (IGE) patients.Patients and methods: 19-channel waking, resting-state EEG records of 26 IGE patients wereanalyzed before treatment (IGE) and after the 90th day of treatment (VPA), in seizure-free con-dition. Three minutes of artifact-free EEG background activity (without epileptiform potentials)was analyzed for each patient in both conditions. A group of 26 age-matched healthy norma-tive control persons (NC) was analyzed in the same way. All the EEG samples were processed to

LORETA (Low Resolution Electromagnetic Tomography) to localize multiple distributed sources of EEG activity. Current source density time series were generated for 33 regions of interest (ROI)in each hemisphere for four frequency bands. Pearson correlation coefficients (R) were com-puted between all ROIs in each hemisphere, for four bands across the investigated samples. Rvalues corresponded to intrahemispheric, cortico-cortical functional EEG connectivity (EEGfC).Group and condition differences were analyzed by statistical parametric network method.

∗ Corresponding author at: University of Debrecen, Medical Center, Department of Neurology, Móricz Zsigmond krt. 22, 4032 Debrecen,ungary. Tel.: +36 52 255255; fax: +36 52 255590.

E-mail address: szilvia.puskas@yahoo.com (S. Puskás).

ttp://dx.doi.org/10.1016/j.eplepsyres.2014.09.032920-1211/© 2014 Elsevier B.V. All rights reserved.

Valproate normalizes EEG functional connectivity 1897

Main results: p < 0.05, corrected for multiple comparisons: (1) The untreated IGE group showedincreased EEGfC in the delta and theta bands, and decreased EEGfC in the alpha band (as com-pared to the NC group); (2) VPA treatment normalized EEGfC in the delta, theta and alpha bands;and (3) degree of normalization depended on frequency band and cortical region.Conclusions: VPA treatment normalizes EEGfC in IGE patients.© 2014 Elsevier B.V. All rights reserved.

pocwTsorofhwtw

oicjitcbaittJ1pI

aTTottntn

gca

EEG recording and sample selection

Introduction

Valproate (VPA) is an effective antiepileptic drug foridiopathic generalized epilepsy (IGE) syndromes. Its anti-convulsive mechanisms were thoroughly investigated atthe molecular and cellular levels. Classic pharmacologicalresearch demonstrated that VPA binds to several neuronaland glial targets and modifies electric activity of these cells(Cotariu et al., 1990; Löscher, 2002; Capek and Esplin, 1990).However, the results do not explain the full therapeuticeffect of VPA. The best argument for this is the so-calleddelayed or carryover effect of VPA (Löscher, 2002) that con-tributes to its superior therapeutic efficacy in IGE (Nicolsonet al., 2004). The delayed effect suggests that VPA causes re-arrangement of cerebral structure and/or function, whichprotects against seizures and is independent of the momen-tary serum level of the drug (Burr et al., 1984; Stefan et al.,1984). Recently, a few other studies supported the enduringeffect of VPA on cerebral connectivity, which may give riseto beneficial or adverse effects. VPA inhibits effective con-nectivity among motor areas in healthy volunteers (Li et al.,2011). VPA alters expression of multiple genes in the CNS inrats (Fukuchi et al., 2009) and epilepsy patients as well (Tanget al., 2004). VPA modifies cortical excitability by modify-ing neuron—glia relationship (Wang et al., 2012), influencesmyelin production, repairs and alters neuronal connectivity(Rosenzweig et al., 2012). IGEs are increasingly realized asnetwork disorders, so it is reasonable to suppose that VPAmodifies abnormal cerebral connectivity.

The aim of this study was to investigate the effect ofchronic VPA administration on functional EEG connectivity(EEGfC) in IGE patients. The key-lock principle of pharmaco-EEG (Saletu et al., 2002) suggests that if epilepsy resultsfrom abnormal cortical function, normalization of that func-tion results in clinical improvement. We have tested thehypothesis that VPA reverses abnormal resting-state intra-hemispheric, cortico-cortical EEGfC in successfully treatedIGE patients. Targeting this part of cerebral connectivityis justified because abnormal intrahemispheric connectivityis the neurophysiological basis of seizure-prone state andictogenesis in experimental models (Timofeev and Steriade,2004) and IGE syndromes as well (Holmes et al., 2004, 2010;Clemens et al., 2013).

Patients and methods

Patients and control persons

The study design was approved by the Local ResearchEthics Committee of Kenézy Gyula County Hospital, Debre-cen, Hungary. All unmedicated IGE patients, who visitedone of the collaborating epilepsy outpatient services, were

Eotc

otential candidates for the study. Patients with recent-nset IGE were diagnosed according to generally acceptedriteria (Panayiotopoulos, 2005). No diagnostic procedureas indicated, missed, or postponed for study purposes only.herapeutic decision has done after correct diagnosis wastated. The patients were informed about risks and benefitsf the drug treatment, with special reference to VPA-relatedisk for females of childbearing age. VPA was not the drugf first choice for those who desired to be pregnant within aew years. Also patients who had a well-documented, longeristory of IGE but did not take medication for any reasonere potentially eligible for the study. Risk-benefit estima-

ion was exposed to them as well and individual experienceith prior drug treatment was taken into consideration.

Inclusion criteria were: first seizure after the fifth yearf life; clinically and electrographically unequivocal find-ngs indicating one of the common IGE syndromes: idiopathichildhood absence or juvenile absence epilepsy (ABS),uvenile myoclonic epilepsy (JME), epilepsy with general-zed tonic—clonic seizures exclusively (GTC); the decisiono treat the patient with VPA monotherapy. Exclusionriteria were: significant neurological or psychiatric comor-idity, metabolic disorders, alcohol or substance abusend any other medical condition that is known to signif-cantly modify EEG activity. Patients having a generalizedonic—clonic seizure in the five days prior EEG investiga-ion were excluded. 26 patients entered the study (ABS = 14,ME = 7, GTC = 5). Age-and sex distribution were: 11 males,5 females, 7—54 years of age, mean age: 17.2 years. 17atients had recent-onset IGE, 4 patients had long-lastingGE, disease onset was uncertain in 5 patients.

Baseline clinical and EEG evaluations were carried outt entry visit. Initial daily dose of VPA was 300 to 500 mg.he dose was increased until seizure freedom was reached.he final daily dose of VPA was 300 to 1500 mg. The sec-nd EEG was recorded 3 month after the first one becausehe initial, transient EEG effects of the drug disappear byhis time (Sannita et al., 1989). The patients’ general andeurological condition did not change from initial to con-rol evaluation. No patient reported complaints indicatingeurotoxicity.

To evaluate the baseline EEGfC abnormality in the patientroup, a group of healthy, normal control persons (NC) wasreated. Each patient was matched to a NC person of samege and sex. Mean age for the NC group was 16.9 years.

EG recordings were carried out in the morning, after a nightf sufficient sleep, in a semi-isolated room, with the sameype of digital equipment, by trained personnel. Silver-silverhloride electrodes were placed according to the 19 sites of

1

ta1rAbmS1ctpstwoeoscwctw1tprnra

Q

Qtws

LLaEeacbitrsetp2ltLa‘L

bLsefni2caa(ldgt2

AC(cesS(oaotelacicGoosciEi

sasaaA(nRMbt

898

he 10—20 system and the earlobes, fixed by appropriatedhesive and conductive gel. Impedances did not exceed0 k�. EEG was recorded against Fpz sampling reference andecomputed against a mathematical linked ears reference.dditional bipolar derivations were used to differentiateetween EEG and eye movement potentials and to detectyogenic activity. EEG filters were set at 0.1 and 33.6 Hz.

ampling rate was 256 per second, on-line digitization was2 bit. 30 min EEG was recorded in the waking-relaxed, eyes-losed condition. The state of vigilance was controlled byhe EEG technician who gently aroused the patient when theosterior alpha rhythm disappeared. Thirty 2-s epochs wereelected for analysis, according to our standard epoch selec-ion protocol: (1) the presence of continuous alpha activityith voltage maximum in posterior regions, (2) the absencef artifacts, epileptiform potentials and other nonstationarylements, (3) the absence of patterns indicating drowsinessr arousal. As to prevent unwanted selection bias, epochelection was carried out by one of us who was blinded to theondition of the patient (untreated vs. treated) in which EEGas recorded. Final visual control of the selected epochs wasarried out by the senior author as to ensure the quality ofhe sample. This electrographic definition of the relaxed-aking state refers to a narrow range of vigilance (Bente,979). Two reproducibility measures were used to minimizehe effect of short- and long-term variability within the sam-les. Only samples with at least 95% of average split-halfeliability and test—retest reliability (across the 19 chan-els) entered further analysis. This level of reliability waseached in all samples, so no records were excluded fromnalysis.

uantitative EEG analyses

uantitative EEG analyses were carried out by means ofhe NeuroGuide Deluxe (Version 2.7.4) software (Thatcher,ww.appliedneuroscience.com) and joined software as

pecified below.

ocalization of the sources of EEG activityow Resolution Electromagnetic Tomography (LORETA) is

method for localizing multiple distributed sources ofEG activity in the three-dimensional space (Pascual-Marquit al., 1994). LORETA demonstrates the synchronouslyctivated neuronal populations underlying EEG activity byomputing their cortical localization from the scalp distri-ution of the electric field. The LORETA inverse solutions based on the ‘‘smoothness’’ assumption, which meanshat neighboring EEG generators produce maximally cor-elated activity in terms of orientation and strength. Themoothness assumption is based on neuroanatomical andlectrophysiological constraints. The brain compartment ofhis model is restricted to the cortical gray matter and hip-ocampus. The gray matter compartment is divided into394 voxels, which allows a spatial resolution of 7 mil-imeters. Localization of voxels is based on coordinates ofhe Talairach Brain Atlas (Talairach and Tournoux, 1988).

ORETA computes current source density (CSD; expresseds Amperes/meters squared) for each voxel, briefly called‘LORETA activity’’ in the literature. The consistency ofORETA with physiology and localization has been validated

SSi

B. Clemens et al.

y several authors (Pascual-Marqui, 2002). Importantly,ORETA based on 16—28 channel EEG recordings localizes theources of maximum EEG activity concordantly to the refer-nce localization methods (positron emission tomography,unctional MRI, MRI diffusion spectral imaging, intracra-ial EEG recordings) when the electrical source distributions ‘‘neurophysiologically smooth’’ (Pascual-Marqui et al.,002; Oakes et al., 2004). This is the case in the epilepticortex where abnormally synchronized activity is distributedcross a lot of cortical columns and even greater corticalreas via the dense network of intracortical connectionsChagnac-Amitai and Connors, 1989). Furthermore, LORETAocalization with this number of electrodes is reliable whenelta, theta, alpha and lower beta frequencies that subservelobal integration of higher cerebral functions and penetratehe entire cerebral volume are investigated (Babiloni et al.,006).

nalysis of resting-state EEG functional connectivityorrelating the activity of LORETA-localized sourcesCSD = current source density) is a useful alternative toorrelate quantitative EEG variables measured at scalplectrodes and offers a deeper understanding of intrahemi-pheric cortico-cortical connectivity (Thatcher et al., 2007;choffelen and Gross, 2009). LORETA Source CorrelationLSC) analysis means computing the temporal covariancer correlation of LORETA-defined CSD between two corticalreas (region of interest, ROI), across successive 2-s epochsver the investigated sample. The Pearson product correla-ion coefficient (r) is a valid measure of oscillator coupling,specially when a relatively long interval of time is ana-yzed, as in this study. Authors who compared the sensitivitynd reliability of several methods concluded that Pearsonorrelation is a robust method being sensitive to all thenvestigated coupling parameters and does not require spe-ific assumptions about the model (Wendling et al., 2009).iven the 19 scalp electrodes, the effect of the point spreadn CSD estimates was minimized by clustering hundredsf nearby voxels into 33 ROIs in each hemisphere. Fig. 1hows the flowchart of computing asymmetric EEGfC matri-es (Thatcher et al., 2007; Clemens et al., 2013). This figurendicates that two correlation coefficients characterize theEGfC between two ROIs, producing asymmetric connectiv-ty matrices.

In order to avoid the asymmetry we have generated aet of symmetric source correlation matrices from the aver-ge of the two correlation coefficients between ROIs: Rgthbs,tand for group (g ∈ {vpa, nc}), treatment (t ∈ {before,fter}), hemisphere (h ∈ {LH, RH}), band (b ∈ {delta, theta,lpha, beta}) and subject indices (S = 1 . . . Ng), respectively.

single element of an Rgthbs matrix was denoted by regthbs

where e represents a connection between two regions). Theumber of rows and columns are equal with the number ofOIs (N = 33) and with the number of correlation coefficients

= N(N − 1)/2 (M = 528). All analyses were based on broad-and results of four frequency bands (delta: 0.5—3.0 Hz,heta: 3.5—7.0 Hz; alpha: 7.5—12.0 Hz; beta: 12.5—25.0 Hz).

tatistical inference of connectionstatistical parametric network (SPN) terminology has beenntroduced (Ginestet et al., 2011). In our study we

Valproate normalizes EEG functional connectivity 1899

Figure 1 The flowchart describes the steps of the calculation of the LORETA source correlation matrices and the evaluation ofd righperm

ILqapivswl

R

U

SImsgestrEp

T

CsEt

the connectivity matrices used in the SPN analysis (left (LH) anThe LSC flowchart was drawn after Thatcher et al. (2007) with

generated population and treatment-related differentialSPNs which provide a statistical method to investigate dif-ferences of connections. SPNs were calculated from Rgthbs

matrices, using M mixed-effect models:

regthbs = Xe

gthbsˇe + Ze

S beS + εe

gthbs

where r are the correlation coefficients of interest, ˇ is avector of fixed effect (group, treatment, hemisphere andband) which does not vary over subjects, b is the subject-specific random effects (subject, age-group) and ε are theresiduals.

In this study four age-groups were defined: child (lessthan 10 years), teenager (between 10 and 20 years), youngadult (between 20 and 40 years) and middle aged (morethan 40 years). The matrices X and Z contain the fixed-effectand the random-effect components of the linear model. Thegroup and the treatment effects for all bands were evalu-ated by a post hoc Tukey test which produced te

hb Studentt-values for all edges, hemispheres and bands. These t-values were stored in N × N SPN matrices for visualization.Statistically significant differences of EEGfC at correctedp < 0.05 level were thresholded by local false discovery rate(FDR) for multiple comparison (Efron, 2004, 2007).

SoftwareFor EEG sample selection and quantitative EEG anal-yses, we used NeuroGuide Deluxe (Version 2.7.7;www.appliedneuroscience.com), LORETA (Pascual-Marquiet al., 1994) and LORETA Source Correlation (Thatcheret al., 2007). SPN were evaluated by home-developed

BrainNetTools software. For the visualization of dif-ferences between significant connections BrainCON(www.minipetct.com/braincon) was used (Spisák et al.,2013).

dsla

t (RH) hemispheres for four broad bands).ission of the author.

nterpretation of the results in the topographic domainORETA localization is the key feature of all the subse-uent analyses. Because its localization accuracy is limited,ctivity of small ROIs may be falsely localized. However, arior study demonstrated that VPA alters LORETA activityn widespread cortical regions (Clemens et al., 2007), soery precise localization was presumably not essential in thistudy. We discuss EEGfC results at sub-lobar, lobar and net-ork levels, which is a usual approach in the neuroimaging

iterature (Anderson and Hamandi, 2011).

esults

ntreated IGE group vs. NC group

tatistically significant differences between the untreatedGE and the NC groups demonstrate baseline EEGfC abnor-ality in IGE (Fig. 2, top row). The untreated IGE group

howed overall greater delta and theta EEGfC than the NCroup. No striking asymmetry and no opposite sense differ-nces emerged in these bands. The untreated IGE grouphowed greater alpha EEGfC among frontal ROIs includinghe uncus (a temporal ROI) and less alpha EEGfC across theest of the cortex as compared to the NC group. Greater betaEGfC emerged among right frontal ROIs in the untreatedatients than in the NC group.

reated vs. untreated conditions of the IGE group

omparison of the treated and untreated IGE conditionshows drug effects (Fig. 2, middle row). VPA decreased deltaEGfC between most ROIs except the left and right paracen-ral lobules, cingulate gyri, and the right parietal lobe. VPA

ecreased theta EEGfC in the anterior parts of the hemi-pheres and increased theta EEGfC between the paracentralobules and the parietal ROIs bilaterally. VPA increasedlpha EEGfC between most ROIs but did not affect EEGfC

1900 B. Clemens et al.

Figure 2 EEGfC group differences in four frequency bands. Top row: untreated IGE group (NAE) vs. healthy controls (NC). Middlerow: treated condition (VPA) vs. untreated condition (NAE) of the IGE group. Bottom row: Treated condition of the IGE group( nly sd text

btlrv

T

Ch(wpfaitaoi

V

Vipsv‘abta

btcgltcg

dfeiVrtbcwEfttt

D

Ae

VPA) vs. healthy controls (NC). Color scale indicates t-values. Oemonstrated. (For interpretation of the references to color in

etween frontal ROIs (where increased EEGfC was found inhe untreated patients) and even decreased EEGfC in theeft lateral frontal area. VPA increased beta EEGfC betweenight frontal and temporal ROIs, and three increased EEGfCalues were found in the left hemisphere.

reated IGE patients vs. NC group

omparison of the treated IGE group and the NC groupsighlights the degree of the normalization in the four bandsFig. 2, bottom row). Greater overall delta and theta EEGfCere found in the treated condition of the IGE group as com-ared to the NC group. Greater alpha and beta EEGfC wereound between right frontal ROIs in the treated IGE groups compared to the NC group. Less alpha EEGfC was foundn a few, fronto-parietal and fronto-occipital connections inhe treated IGE group as compared to the NC group. Greaterlpha and beta EEGfC in the right frontal lobe (and in a fewther connections) were found in the treated IGE group thann the NC group.

PA-related EEGfC normalization

PA treatment decreased abnormally high EEGfC andncreased abnormally low EEGfC between most, but not all,airs of ROIs in the delta, theta and alpha bands. VPA-relatedhift of a single abnormal EEGfC value toward the normativealue means EEGfC normalization and is defined as follows.‘Full normalization’’ of an abnormal connection means:

bnormal baseline value + statistically significant differenceetween the untreated and treated conditions + no statis-ically significant difference between the treated conditionnd the NC group. ‘‘Partial normalization’’ means: abnormal

prcm

tatistically significant (p < 0.05, FDR-corrected) differences are, the reader is referred to the web version of this article.)

aseline value + statistically significant difference betweenhe untreated and treated conditions + statistically signifi-ant difference between the treated condition and the NCroup. ‘‘No normalization’’ is defined as: abnormal base-ine value + no statistically significant difference betweenhe untreated and treated conditions + statistically signifi-ant difference between the treated condition and the NCroup.

With this logic in mind, VPA treatment fully normalizedelta EEGfC between several frontal ROIs bilaterally (medialrontal gyrus, orbital gyrus, straight gyrus, subcallosal andxtra-nuclear areas) and the uncus. Lack of abnormal EEGfCn this area is better demonstrable in sagittal view (Fig. 3).PA partially normalized delta EEGfC between ROIs in theest of the cortex except the right parietal area. VPA par-ially normalized EEGfC between anterior ROIs in the thetaand. Full EEGfC normalization occurred in most long-rangeonnections in the alpha band. On the other hand, thereas no normalization of abnormally increased baseline alphaEGfC between frontal ROIs in the right hemisphere, whileull or partial normalization occurred in the connections ofhe left frontal area. No EEGfC normalization occurred inhe beta band. More increased EEGfC values were found inhe treated than in the baseline condition.

iscussion

s far as is known, this is the first study to investigate theffect of therapeutic VPA administration on EEGfC in IGE

atients. The findings confirmed the hypothesis that VPAeverses abnormal resting-state intrahemispheric, cortico-ortical EEGfC in successfully treated IGE patients. Thiseans that VPA shifted most baseline EEGfC abnormalities

Valproate normalizes EEG functional connectivity 1901

Figure 3 EEGfC differences in delta band between the treated IGE group and healthy controls. Middle plot of this figure (topview) shows the same results as the left plot in the bottom row of Fig. 2. Lateral views demonstrate better the lack of statistically

t hem

tetotpaccspE

C

Vmammksa

fldtOfs

csup

L

significant differences between frontal ROIs in the left and righ

toward normative values. However, the degree of normal-ization depended on frequency bands, cortical regions andthe magnitude of baseline EEGfC abnormality.

Changes in the delta and theta bands

Delta, theta, alpha and beta bands reflect natural classesof brain oscillations (Buzsáki and Draguhn, 2004) generatedby ‘‘selectively distributed neuronal systems’’ (Basar andSchürmann, 1999). Thus, it not surprising that VPA causesband-specific changes in electrical activity of the cortex.In this study we demonstrated decrease and normalizationof EEGfC in the delta and theta bands. This finding is rem-iniscent of activity-dependent, normalizing effect of VPAon delta and theta spectral power in IGE patients. Theeffect was independent of the daily dose and serum levelof VPA, so we concluded that it reflects enduring changeswithin the CNS (Clemens, 2008). Local EEG synchronization(expressed as voltage of the signal or CSD) depends on thedense intracortical circuitry modified by subcortical inputs(Nunez, 1995). Remote EEG synchronization (EEGfC, definedin Section ‘Analysis of resting-state EEG functional connec-tivity’) means synchronized fluctuations of CSD in two greatneuronal ensembles, in other words, two cortical areas.This effect is mediated by short- and long distance asso-ciation fibers (Thatcher et al., 2007) and other, yet not fullyknown mechanisms (Margineanu, 2010). Estimates of localand remote EEG synchronization may change independently.However, VPA decreases and normalizes delta and thetaactivity at both spatial scales, which is in accord with thecomplexity of drug effects on the CNS (Margineanu, 2012).

Little is known about the relationship of band-relatedEEGfC and epilepsy. Epilepsy patients show increased EEGfCin the delta band in the interictal state (Horstmann et al.,2010). Recently, increased theta band synchronization wasreported in epilepsy (Douw et al., 2010). Increased delta andtheta EEGfC were found in untreated JME patients in theinterictal state (Clemens et al., 2013). These findings sug-gest that increased remote EEG synchronization in the slow

frequency bands (delta and/or theta) is somehow relatedto epilepsy and is independent of the syndrome. An experi-mental study demonstrated that increased phase stability(another measure of remote EEG synchronization) in the

Bt(i

ispheres.

heta band is a neurophysiological corollary of epileptogen-sis and persists in the epileptic state (Ge et al., 2013). Withhis evidence in mind we suggest that VPA-related decreasef delta and/or theta EEGfC may be related to successfulreatment. This argumentation is in accord with the key-lockrinciple of pharmaco-EEG (Saletu et al., 2002). However,lternative solutions cannot be excluded. VPA may causeonnectivity changes in the CNS that do not modify seizureontrol (Rosenzweig et al., 2012). Demonstrating the lack ofignificant EEGfC drug effects in VPA-treated nonresponderatients might support the relationship between delta-thetaEGfC normalization and therapeutic efficacy.

hanges in the alpha and beta bands

PA increased and normalized alpha band EEGfC betweenost ROIs that belong to the ‘‘classic’’ alpha system (Basar

nd Schürmann, 1999). Cortical generators of this system areostly localized in posterior areas but some alpha activityay be recorded over the rest of the cortex. As far as is

nown, epileptogenesis and the epileptic condition do notignificantly affect the alpha system, so the significance oflpha EEGfC normalization remains hidden.

VPA did not normalize baseline alpha EEGfC among rightrontal ROIs and partially normalized alpha EEGfC in theeft frontal area. This unexpected finding demonstrated thatrug effect may differentiate between neuronal systemshat operate at the same frequency band at the same time.ur results are consistent with the opinion that the medial

rontal alpha rhythm is not generated by the ‘‘classic’’ alphaystem (Röschke et al., 1997; Feshchenko et al., 2001).

VPA did not affect beta band EEGfC in the majority ofonnections. Treated condition of the IGE group showedomewhat more abnormal beta connectivity values than thentreated condition. This drug effect has no explanation atresent.

imitations of the broad-band approach

esides the band-specific trends of EEGfC change a few,opographically sporadic VPA effects occurred in this studyFig. 2, middle row). These findings may be explained bynter-individual variability of electrophysiological results.

1

IfiEEEne1btstttd

R

PmbPf2

cadshdV

braapfta

R

A

B

B

B

B

B

C

C

C

C

C

C

D

E

E

F

F

G

G

H

H

H

L

902

ndividual variability may cause blurring of the broad bandndings in group studies. Decreased left fronto-lateral alphaEGfC actually might belong to strongly decreased thetaEGfC in the same region. Increased cingular—parietal thetaEGfC is topographically similar to increased EEGfC in theeighboring alpha band. Our unpublished personal experi-nce with parallel broad band and very narrow band (VNB,

Hz bandwidth) analyses shows that strong topographicalroad band patterns may intrude into neighboring VNB ofhe next broad band. So, there is no obligatory need toearch after neurophysiological basis and clinical impor-ance of findings that seem to intrude from one broad bando the neighboring one. If the finding seems to be impor-ant, narrow band analysis can highlight its exact spectralistribution (Szava et al., 1994).

egional differences of the VPA effect

artial normalization of delta EEGfC was found betweenost ROIs but full normalization occurred in the medial-asal frontal area and the uncus in the temporal lobe.otential importance of this finding is that the medial—basalrontal cortex is the seizure onset zone in ABS (Holmes et al.,004) and JME (Holmes et al., 2010).

Lack of VPA effect in the delta band and rightingular—parietal area was a peculiar finding. Major clinicalnd EEG manifestations are usually symmetrical in IGE syn-romes. Whether the reported asymmetrical drug effects ispecific to epilepsy or IGE or reflect general response of theuman brain to repeated VPA administration remains hid-en. We are not aware of any topographical EEG analysis ofPA effects in healthy persons and patient groups.

VPA significantly decreased theta EEGfC between anteriorut not posterior ROIs. This finding harmonizes with VPA-elated decrease of local CSD in frontal but not posteriorreas in IGE (Clemens et al., 2007). Baseline theta EEGfCbnormality was greater between anterior than betweenosterior ROIs. Furthermore, also neuroimaging studiesound the main structural and functional abnormalities inhe anterior parts of the brain in IGE syndromes (Andersonnd Hamandi, 2011).

eferences

nderson, J., Hamandi, K., 2011. Understanding juvenile myoclonicepilepsy: contributions from neuroimaging. Epilepsy Res. 94 (3),127—137.

abiloni, B., Binetti, G., Cassarino, A., Dal Forno, G., Del Percio,C., Ferreri, F., Frisoni, G., Galderisi, S., Hirata, K., Lanuzza, B.,Miniussi, C., Mucci, A., Nobili, F., Rodriguez, G., Romai, G.L.,Rossini, P.M., 2006. Sources of cortical rhythms in adults duringphysiological aging: a multicentric EEG study. Hum. Brain Mapp.27, 162—172.

asar, E., Schürmann, M., 1999. Brain functioning: integrativemodels. In: Basar, E. (Ed.), Brain Function and Oscillations:Integrative Brain Function. Neurophysiology and Cognitive Pro-cesses, vol. II. Springer, pp. 393—406.

ente, D., 1979. Vigilance and evaluation of psychotropic drugeffect on EEG. Pharmacopsychiatry 12, 137—147.

urr, W., Fröscher, W., Hoffmann, F., Stefan, H., 1984. Lack of sig-nificant correlation between circadian profiles of valproic acid

B. Clemens et al.

serum levels and epileptiform electroencephalographic activity.Ther. Drug Monit. 6, 179—181.

uzsáki, G., Draguhn, A., 2004. Neuronal oscillations in corticalnetworks. Science 304, 1926—1929.

apek, R., Esplin, B., 1990. Mechanisms of anticonvulsantaction of valproate: an overview and perspective. In: Avoli,É.M., Gloor, P., Kostopoulos, G., Naquet, R. (Eds.), Gener-alised Epilepsy: Neurobiological Approaches. Birkhauser, Boston,pp. 436—459.

hagnac-Amitai, Y., Connors, B.W., 1989. Horizontal spread ofsynchronized activity in neocortex and its control by GABA-mediated inhibition. J. Neurophysiol. 61, 747—758.

lemens, B., 2008. Valproate decreases EEG synchronization ina use-dependent manner in idiopathic generalized epilepsy.Seizure 17, 224—233.

lemens, B., Bessenyei, M., Tóth, M., Kondákor, I., 2007. Valproateselectively reduces EEG activity in anterior parts of the cortex inpatients with idiopathic generalized epilepsy: a low resolutionelectromagnetic tomography (LORETA) study. Epilepsy Res. 75,186—191.

lemens, B., Puskás, S., Besenyei, M., Spisák, T., Kis, S.A., Hollódy,K., Fogarasi, A., Fekete, I., Emri, M., 2013. Neurophysiology ofjuvenile myoclonic epilepsy: EEG-based network and graph anal-ysis of the interictal and immediate preictal states. Epilepsy Res.106 (3), 357—369.

otariu, D., Zaidman, J.L., Evans, S., 1990. Neurophysiological andbiochemical changes evoked by valproic acid in the central ner-vous system. Prog. Neurobiol. 34, 343—354.

ouw, L., de Groot, M., van Dellen, E., Heimans, J.J., Ronner, H.E.,Stam, C.J., Reijneveld, J.C., 2010. ‘Functional connectivity’ isa sensitive predictor of epilepsy diagnosis after the first seizure.PLoS ONE 5, e10839.

fron, B., 2004. Large-scale simultaneous hypothesis testing: thechoice of a null hypothesis. J. Am. Stat. Assoc. 99, 96—104.

fron, B., 2007. Correlation and large-scale simultaneous signifi-cance testing. J. Am. Stat. Assoc. 102, 93—103.

eshchenko, V.A., Reinsel, R.A., Veselis, R.A., 2001. Multiplicity ofthe alpha rhythm in normal humans. J. Clin. Neurophysiol. 18(4), 331—344.

ukuchi, M., Nii, T., Ishimaru, N., Minamino, A., Hara, D., Takasaki,I., Tabuchi, A., Tsuda, M., 2009. Valproic acid indices up- ordown-regulation of gene expression responsible for the neu-ronal excitation and inhibition in rat cortical neurons throughits epigenetic action. Neurosci. Res. 65, 35—43.

e, M., Wang, D., Dong, G., Guo, B., Cao, R., Sun, W., Zhang, J., Liu,H., 2013. Transient impact of spike on theta rhythm in temporallobe epilepsy. Exp. Neurol. 250, 136—142.

inestet, C.E., Nichols, T.E., Bullmore, E.T., Simmons, A., 2011.Brain network analysis: separating cost from topology using cost-integration. PLoS ONE 6, 7.

olmes, M.D., Brown, M., Tucker, D.M., 2004. Are generalizedseizures truly generalized? Evidence of localized mesial frontaland frontopolar discharges in absence. Epilepsia 45, 1568—1579.

olmes, M.D., Quiring, J., Tucker, D.M., 2010. Evidence thatjuvenile myoclonic epilepsy is a disorder of frontotemporal cor-ticothalamic networks. Neuroimage 49, 80—93.

orstmann, M.T., Bialonski, S., Noennig, N., Mai, H., Prus-seit, J., Wellmer, J., Hinrichs, H., Lehnertz, K., 2010. Statedependent properties of epileptic brain networks: comparativegraph—theoretical analyses of simultaneously recorded EEG andMEG. Clin. Neurophysiol. 121, 172—185.

i, X., Large, C.H., Ricc, R., Taylor, J.J., Nahas, Z., Bohnin, D.E.,Motgan, P., George, M.S., 2011. Using interleaved magneticstimulation/functional magnetic resonance imaging (fMRI) anddynamic causal modeling to understand the discrete circuit

specific changes of medication: lamotrigine and valproic acidchanges in motor and prefrontal effective connectivity. Psychi-atry Res. 194, 141—148.

S

S

S

S

S

T

T

T

T

W

Wendling, F., Ansari-Asl, K., Bartolomei, F., Senhadji, L., 2009.

Valproate normalizes EEG functional connectivity

Löscher, W., 2002. Basic pharmacology of valproate: a review after35 years of clinical use for the treatment of epilepsy. CNS Drugs16, 669—694.

Margineanu, D.G., 2010. Epileptic hypersynchrony revisited. Neu-roreport 210 (21), 963—967.

Margineanu, D.G., 2012. Systems biology impact on drug discovery.Epilepsy Res. 98 (2-3), 104—115.

Nicolson, A., Appleton, R.E., Chadwick, D.W., Smith, D.F., 2004.The relationship between treatment with valproate, lamotrig-ine, topiramate and the prognosis of the idiopathic generalizedepilepsies. J. Neurol. Neurosurg. Psychiatry 75, 75—79.

Nunez, P.L., 1995. Quantitative states of neocortex. In: Nunez, P.L.(Ed.), Neocortical Dynamics and Human EEG Rhythms. OxfordUniversity Press, pp. 1—18.

Oakes, T.R., Pizzagalli, D.A., Hendrick, A.M., Horras, K.A., Larson,C.L., Abercrombie, H.C., Schaefer, S.M., Koger, J.V., Davidson,R.J., 2004. Functional coupling of simultaneous electrical andmetabolic activity in the human brain. Hum. Brain Mapp. 21,257—270.

Panayiotopoulos, C., 2005. Idiopathic generalized epilepsies. In:Panayiotopoulos, C. (Ed.), The Epilepsies: Seizures, Syndromes,and Management. Springer, pp. 271—348.

Pascual-Marqui, R.D., 2002. Functional imaging with low-resolutionbrain electromagnetic tomography (LORETA): a review. MethodsFind. Exp. Clin. Pharmacol. 24, 91—96.

Pascual-Marqui, R.D., Michel, C.M., Lehmann, D., 1994. Low resolu-tion electromagnetic tomography: a new method for localizingelectrical activity in the brain. Int. J. Psychophysiol. 18, 49—65.

Pascual-Marqui, R.D., Esslen, M., Kochi, K., Lehmann, D., 2002.Functional imaging with low-resolution brain electromagnetictomography (LORETA): review, new comparisons, and new vali-dation. Jpn. J. Clin. Neurophysiol. 30, 81—94.

Röschke, J., Fell, J., Mann, K., 1997. Non-linear dynamics ofalpha and theta rhythm: correlation dimension and Lyapunovexponents from healthy subject’s spontaneous EEG. Int. J. Psy-chophysiol. 26, 251—261.

Rosenzweig, I., Vukadinocic, Z., Turner, A.J., Catani, M., 2012.

Neuroconnectivity and valproic acid: the myelin hypothesis.Neurosci. Behav. Rev. 36, 1848—1856.

Saletu, B., Anderer, P., Saletu-Zyhlarz, G.M., Pascual-Marqui,R.D., 2002. EEG topography and tomography in diagnosis and

1903

treatment of mental disorders: evidence for a key-lock principle.Methods Find. Exp. Clin. Pharmacol. 24, 97—106.

annita, W.G., Gervasio, L., Zagoni, P., 1989. Quantitative EEGeffects and plasma concentration of sodium valproate: acuteand long-term administration to epileptic patients. Pharmaco-electroencephalography 22, 231—235.

choffelen, J.M., Gross, J., 2009. Source connectivity analysis withMEG and EEG. Hum Brain Mapp. 30, 1857—1865.

pisák, T., Opposits, G., Kis, S.A., Pohubi, L., Jakab, A., Puskás, S.,Clemens, B., Emri, M., 2013. BrainCON: Graph theory based mul-timodal brain connectivity analysis and visualization software.In: Electronic Presentation Online System: ECR Congress 2013.European Society of Radiology, pp. C-2588.

tefan, H., Burr, W., Fischel, H., Fröscher, W., Penin, H., 1984.Intensive follow-up monitoring in patients with once dailyevening administration of sodium valproate. Epilepsia 25,152—160.

zava, S., Alan, L., Bosch, J., Clark, I., Jimenez, J.C., 1994.High resolution quantitative EEG analysis. Brain Topogr. 6 (3),211—219.

alairach, J., Tournoux, P., 1988. Co-planar Stereotaxic Atlas ofthe Human Brain: Three-Dimensional Proportional System. G.Thieme, Stuttgart.

ang, Y., Glauser, T.A., Gilbert, D.L., Hershey, A.D., Privitera, M.D.,Ficker, D.M., Szaflarski, J.P., Sharp, F.R., 2004. Valproic acidblood genomic expression patterns in children with epilepsy apilot study. Acta. Neurol. Scand. 109, 159—168.

hatcher, R.W., Biver, C.J., North, D., 2007. Spatial—temporal cur-rent source correlations and cortical connectivity. Clin. EEGNeurosci. 38, 35—48.

imofeev, I., Steriade, M., 2004. Neocortical seizures: initiation,development and cessation. Neuroscience 123, 299—336.

ang, C.C., Chen, P.S., Hsu, S.J., Lin, C.T., Gean, P.W., 2012. Val-proic acid mediates the synaptic excitatory—inhibitory balancethrough astrocytes — a preliminary study. Prog. Neuropsy-chopharmacol. Biol. Psychiatry 37, 111—120.

From EEG signals to brain connectivity: a model-based evalu-ation of interdependence measures. J. Neurosci. Methods 183,9—18.