ResponsiveNeurostimulationSuppressesSynchronized CorticalRhythms in Patientswith Epilepsy

Vikaas S. Sohal, MD, PhDa,b,c,*, Felice T. Sun, PhDd

KEYWORDS

� Synchrony � Gamma rhythms � Phase locking � Stimulation� Neocortex � Hippocampus

Well established as a modality for the treatment ofParkinson disease, deep brain stimulation and otherdirect stimulations of neural tissues are increasinglybeing investigated to treat several neurologic andpsychiatric conditions, including epilepsy (for reviewsee1), depression,2 obsessive-compulsive dis-order,3,4 and Tourette syndrome.5 Despite this in-creasing potential clinical utility, the mechanismsby which deep brain stimulation and other formsof neurostimulation, including electroconvulsivetherapy, transcranial magnetic stimulation, and va-gus nerve stimulation, modulate neuronal activityremain unknown. Deep brain stimulation and otherforms of intracranial neurostimulation requiremech-anistic explanation at multiple levels: (1) how doesneurostimulation affect directly stimulated neuronsand their processes, (2) how do these direct localeffects of stimulation acutely affect patterns oflarge-scale activity in populations of neurons, and(3)howdotheseeffectsultimatelyaltermacroscopic

Disclosure: V.S. Sohal was a paid consultant for NeuroPacNeuroPace, Inc.V.S. Sohal is supported by the Staglin Family and the(IMHRO).a Department of Psychiatry, University of California, SCA 94127, USAb Keck Center for Integrative Neuroscience, UniversitySan Francisco, CA 94127, USAc Sloan-Swartz Center for Theoretical Neurobiology, UnCA 94127, USAd NeuroPace, Inc, 1375 Shorebird Way, Mountain View,* Corresponding author.E-mail address: vikaas.sohal@ucsf.edu

Neurosurg Clin N Am 22 (2011) 481–488doi:10.1016/j.nec.2011.07.0071042-3680/11/$ – see front matter � 2011 Elsevier Inc. All

activity in brain regions over the long term. Recentwork suggests that deep brain stimulation in thesubthalamic nucleus may alleviate the symptomsof Parkinsondiseaseby exciting axons fromdistant,possibly cortical, structures.6 Imaging worksuggests that in depression, deep brain stimulationin axons near the subgenual cingulate (Cg25) mayultimately lead to downregulation of activity in theCg25 and consequent changes in other intercon-nected regions.7 In this article, we focus on howintracranial neurostimulation affects synchronousrhythmic activity in populations of neurons. Disrupt-ing synchronous activitymay be an important thera-peutic mechanism for deep brain stimulation inParkinson disease8 and is also likely to be criticallyimportant for the treatment of epilepsy, which is,almost by definition, characterized by abnormalneural synchrony. We describe the effects of neuro-stimulation on rhythmic activity recorded intracrani-ally fromtheneocortexandhippocampus inpatients

e, Inc from 2006 until 2010. F.T. Sun is an employee of

International Mental Health Research Organization

an Francisco, 401 Parnassus Avenue, San Francisco,

of California, San Francisco, 401 Parnassus Avenue,

iversity of California, San Francisco, San Francisco,

CA 94043, USA

rights reserved. neurosurgery.th

eclinics.com

Sohal & Sun482

with epilepsy participating in a clinical investigationof an implantable responsive neurostimulation(RNS) system (RNSSystem,NeuroPace, Inc,Moun-tain View, CA, USA).We found that responsive stim-ulation acutely suppresses phase locking betweengamma-frequency rhythmic activities recorded atdifferent locations.

METHODSElectrocorticographic Recordings

Electrocorticographic (ECOG) signals were re-corded from 65 patients participating in a feasibilityclinical trial of the responsive stimulation System.This was a multicenter trial conducted between2004 and 2007 designed to demonstrate adequatesafety and provide preliminary evidence for effec-tiveness. The trial was approved by the Food andDrug Administration and investigational reviewboards of each center, and informed consentwas obtained from all patients.The responsive stimulation System provides

responsive cortical stimulation via a cranially im-planted programmable neurostimulator con-nected to 1 or 2 recording and stimulating depthand/or subdural cortical strip leads that are surgi-cally placed in the brain according to the seizurefocus. The neurostimulator continually sensesECOG activity and is programmed by the physi-cian to detect abnormal ECOG activity and thenprovide stimulation. Forty-one subjects had leadsimplanted in the neocortex, 19 had leads im-planted in the hippocampus, and 5 had leadsimplanted in the hippocampus and neocortex. Intotal, 95 patients had leads located in theneocortex, and 29 had leads in the hippocampus.Two channels of bipolar recordings were availablefor each lead. Whenever recording sites or therecording montage changed, data from that patientwere treated as a new data set, resulting in 146 datasets.The cranially implanted neurostimulator pro-

cesses the signals in real time, using detectionparameters that were unique to each patient andadjusted over time to detect epileptiform activity.When stimulation was enabled, detection eventswere followed after a short latency (typically 60–300 microseconds) by electrical stimulation(typically high frequency; >90% of stimulationoccurred at frequencies between 100 and 333Hz, delivered pulses that were 120–200 microsec-onds wide, and lasted 100–200 milliseconds).ECOG records (sampled at 250 Hz and typically60–180 seconds in duration) were stored inresponse to preprogrammed events (such as anindividual detection event or multiple consecutivedetection events).

Selection of ECOG Data

Some ECOG records contained multiple detectionevents, each of which would be followed by stimu-lation if stimulation was enabled. To minimizeeffects from one stimulation that might affect theanalysis of later stimuli, we only analyzed data cor-responding to the first stimulation event in eachECOG record. Thewavelet analysesdescribed lateroccurred at discrete time points, which weredefined relative to the detection event. In ECOGrecords containing responsive stimulation, thesetime points were always measured relative to thedetection event immediately preceding the stimula-tion. This measurement allowed us to compareactivity at corresponding time points in ECOGrecords that contained stimulation (after detectionevents) and those containing detection eventsonly. During the period immediately following stim-ulation, signals couldbesaturatedor blanked (ie, nosignal available for analysis) or contain transientlow-frequency (<1 Hz) artifacts. To ensure that ourresults were not affected by these sorts of stimula-tion artifacts, we verified that ECOG signals fromevery recording channel were nonzero and nonsa-turated throughout a window surrounding eachtimepoint. Thiswindowwas largeenough to includeall time points that might contribute to the analysisvia temporal filtering and the wavelet transforma-tion. Any ECOG records that did not meet thesecriteria were excluded from the analysis. We onlyanalyzed data sets in which a minimum of 10ECOG records with and without stimulations wereavailable to calculate the phase-locking statistic.

Wavelet Analysis

To study rhythmic activity, we first computed thediscretewavelet transformof activity in eachchannelat each timepoint.Webased our analysis on a previ-ously described approach.9 Specifically, for eachfrequency, f, we first band-pass filtered recordingsbetween f � 2.5 Hz (Fig. 1B) and then convolvedthe filtered signal with a wavelet of frequency fto obtain an amplitude and a phase given by

e�t2=2s2

e2pif (1)

where s 5 4/3f and f is the frequency.

Measuring Phase Locking

Following the approach described by Lachaux andcolleagues,9 we used phases obtained from thediscrete wavelet transformation described earlierto calculate a measure of phase locking. Foreach pair of recordings from each data set, wecomputed phase differences, converted these tounit vectors in the complex plane (see Fig. 1C),

DetectionStimulation

1 second

Channel 1Channel 2

*

Filtered 37.5 - 42.5 Hz

Filter

Convolve w/ wavelet

Convert phase differences tounit vectors in complex plane

vectoraverage

measure of phase locking= synchrony of oscillations

100 msec

Pre-detection(t - 0.2 sec)

Post-detection(t + 1 sec)

Compute phase difference

amplitude =strength of oscillations

A

C

B

D

Fig. 1. Overview of responsive stimula-tion and data analysis. (A) ECOG signalswere recorded using chronically im-planted intracranial electrodes. Signalswere processed in real time to detectepileptiform activity. When epilepti-form activity was detected and the stim-ulation was enabled, detections werefollowed by high-frequency electricalstimulation, as shown (of note, thesensing amplifiers were blanked duringthe brief period of stimulation). Todetermine acute effects of stimulation,we compared signals 1.0, 0.5, and 0.2seconds before and 1.0, 1.2, and 2.0seconds after detection events withstimulation with signals before andafter detection events in the absenceof stimulation. (B) To analyze activity ata particular frequency, we band-passfiltered each signal around thatfrequency, before convolving witha wavelet to obtain the amplitude andphase at that frequency. (C, D) Tocompute phase locking between a pairof signals, for each ECOG record, wecomputed the phase difference for those2 signals and converted all the phasedifferences (corresponding to differentECOG records) to unit vectors in thecomplex plane. The amplitude of thevector average of all these vectorsmeasured phase locking.

Neurostimulation Suppresses Phase-Locking 483

and used the amplitude of the average of theseunit vectors to measure the degree of phase lock-ing or synchrony at frequency f. We refer to thismeasure as the phase-locking statistic.

Determination of Statistically SignificantPhase Locking

We determined whether a particular value of thephase-locking statistic was statistically significantusing bootstrapping. Specifically, we computed1000 random values of the phase-lockingstatistic. Each of these was the sum of N unitvectors in the complex plane, each of which had

a pseudorandom phase. N is the number of phases(or unit vectors) that was used to calculate theoriginal phase-locking statistic. Then, we obtainedaP value by determining howoften the actual calcu-lated phase-locking statistic exceeded these pseu-dorandom values of the phase-locking statistic.

RESULTS

We studied how responsive stimulation affectedrhythmic activity in the neocortex and hippo-campus. As described in the “Methods” section,ECOG signals were recorded from chronically im-planted intracranial electrodes in patients with

0.12

0.16

0.2

0.24

-1 0 1 2

Stimulation

Detection-only

P<.05

Time (sec relative to event detection)

Phas

e lo

ckin

g st

atis

tic

Fig. 2. Responsive stimulation acutely suppresses phaselocking. The phase-locking statistic is plotted as a func-tion of time, averaged over all cases (ie, a particularfrequency and electrode pair from a particular dataset),which showedstatistically significantphase lockinginbothECOGrecords containing stimulationsand thosecontaining detections only. We calculated the phase-locking statistic at 6 timepoints: 1.0 second, 0.5 seconds,and 0.2 seconds before the detection event and 1.0second, 1.2 seconds, and 2.0 seconds after thedetectionevent. In the events with responsive stimulation, stimu-lations were delivered shortly after the detection event(ie, within 300 milliseconds). There is a sharp rise in thephase-locking statistic between 0.5 and 0.2 secondsbefore the detection event. After stimulation, thephase-locking statistic drops, whereas in ECOG recordscontaining detections only, the phase-locking statisticcontinues to rise, such that the phase-locking statisticis significantly different in these 2 cases (P<.05, 2-tailedt test). Error bars5 �1 standard error of the mean.

Sohal & Sun484

epilepsy. Briefly, a cranially implanted neurostimu-lator continually senses ECOG activity and detectsabnormal ECOG activity. If stimulation wasenabled, detection events were followed aftera short latency by electrical stimulation. BriefECOG records were stored containing detectionevents (with and without stimulation). Fig. 1Ashows ECOG data from 2 distinct channels re-corded simultaneously during a detection eventfollowed by stimulation.To study rhythmic activity, wemeasuredboth the

amplitude of rhythmic activity and the degree ofphase locking, or synchrony, between rhythmicactivities recorded simultaneously from differentchannels, based on a previously describedapproach.9 The detailed method for computingthis measure of phase locking, referred to as thephase-locking statistic, is described in the“Methods” section. This measure of phase lockingis similar to coherence, but does not requirestationary signals, and measures the phase rela-tionship independent of amplitude. We comparedthe strength and synchrony of oscillations 1.0,0.5, and 0.2 seconds before detection events and1.0, 1.2, and 2.0 seconds after these events. Asdescribed in the “Methods” section, we omitteddata containing stimulation artifacts. Wemeasuredthe statistical significance of phase locking bybootstrapping as described in the “Methods”section. We analyzed frequencies between 10and 100 Hz in 5-Hz intervals, although in somecases we then grouped frequencies into 4 bands:10 to 30 Hz representing alpha and beta bands,35 to 50 Hz representing low-gamma band, 55 to80 Hz representing the mid-gamma band, and 85to 100 Hz representing the high-gamma band.Low frequencies (<10 Hz) were excluded becauseof the potential amplifier-sensing artifact, whereashigh frequencies (>100Hz)were excluded becauserecordings included a low-pass analog filter ofapproximately 80 to 90 Hz to minimize aliasing forsignals lower than 125 Hz.To study the effects of responsive stimulation on

phase locking, we focused on those ECOG recordscontaining statistically significant phase locking(P<.05) either before or after detection events(with and without stimulation). The presence ofphase locking was similar in ECOG records con-taining stimulations (16.8% of all electrode pairsexhibited phase locking) and those containingdetection events that were not followed by stimula-tions (16.3% of all electrode pairs exhibited phaselocking). This fraction was much higher than wouldbe expected by chance (9%). To compare changesin phase locking that occur after rhythmic stimula-tion with those occurring after detection events inthe absence of stimulation, we looked at data

collected from the same patients, using the sameelectrode configurations, in which both the stimu-lation and detection-only data sets containedstatistically significant phase locking at the samefrequency. There were 239 such cases, and thesewere drawn from 60 data sets (out of 129 totaldata sets).Fig. 2 shows phase locking, as a function of time

relative to the detection event, for these cases(stimulation present vs detection events only).Note that the phase-locking statistic can varybetween 0 (no phase locking) and 1 (perfect phaselocking). Both the data containing stimulations anddata containing detection events only show a sharprise in phase locking from 500 to 200 millisecondspreceding the detection event and an eventual re-turn to thebaseline level of phase locking.However,after stimulations, the phase-locking 1 second afterthe detection event decreases by an average of�8%� 4%. By contrast, in the absence of stimula-tion, the phase locking continues to rise, by anaverage of 3%� 4%. As a result, the phase locking

10-30 Hz

0.17

0.22

0.27

Stimulation Detectiononly

Phas

e-lo

ckin

g st

atis

tic

pre post pre post0.13

0.18

0.23

Stimulation Detectiononly

Phas

e-lo

ckin

g st

atis

tic

pre post pre post

0.16

0.21

0.26

Phas

e-lo

ckin

g st

atis

tic

Stimulation Detectiononly

pre post pre post 0.13

0.18

0.23

Phas

e-lo

ckin

g st

atis

tic

Stimulation Detectiononly

pre post pre post

35-55 Hz

60-80 Hz 85-100 Hz

Neocortex, 35-100 Hz Hippocampus, 35-100 Hz

0.13

0.18

0.23

Phas

e-lo

ckin

g st

atis

tic

Stimulation Detectiononly

pre post pre post

p < 0.05

Phas

e-lo

ckin

g st

atis

tic

0.13

0.18

0.23

Stimulation Detectiononly

pre post pre post

p < 0.05

0.16

0.2

0.24

0.28

-1 0 1 2

Stimulation

0.11

0.15

0.19

0.23

-1 0 1 2

Detection only Stimulation Detection only

Phas

e-lo

ckin

g st

atis

tic

Phas

e-lo

ckin

g st

atis

tic

Time (sec relative to detection event) Time (sec relative to detection event)

p < 0.01p = 0.74

10-30 Hz 35-100 Hz

A

C

B

ig. 3. The effects of responsive stimulation are frequency dependent. (A) Comparison of the phase-lockingatistic 0.2 seconds before and 1.0 second after detection events (pre and post, respectively) for variousequency bands and for ECOG records that contain stimulation events or contain detection events only. Therea significant interaction between the frequency band (10–30 vs 35–100 Hz) and the condition (stimulation vsetection-only) on the change in phase locking (post-pre) (P<.05 by analysis of variance), that is, stimulationppresses the phase-locking statistic for gamma-frequency (35–100 Hz) activity but not lower-frequency0–30 Hz) activity. (B) The phase-locking statistic over time for activity in the alpha and beta frequency bands0–30 Hz, left) and the gamma-frequency band (35–100 Hz, right). As in Fig. 2, data are averaged over all casese, a particular frequency and electrode pair from a particular data set), which showed statistically significanthase locking in both ECOG records containing stimulations and those containing detections only. (C) Compar-on of the phase-locking statistic 0.2 seconds before and 1.0 second after detection events for gamma-equency activity recorded from the neocortex (left) or hippocampus (right). In both regions, stimulationppresses gamma-frequency phase locking. Error bars 5 �1 standard error of the mean.

485

Fstfrisdsu(1(1(ipisfrsu

486 Sohal & Sun

is significantly lower after stimulations than in theabsence of stimulation (P<.05; 2-tailed t test).We then looked at the change in phase locking

after stimulation versus detection only as a functionof frequency (Fig. 3A). The effects of stimulation onphase locking seem to occur throughout thegamma-frequency range (35–100 Hz) but not atlower frequencies (10–30 Hz). There was a statisti-cally significant interaction between frequencyband (10–30 vs 35–100 Hz) and the presence ofstimulation on the change in phase locking (seeFig. 3B; P<.05 by 2-way analysis of variance):stimulation reduces gamma-frequency (35–100Hz) phase locking measured 1 second after thedetection event by 15% � 7%, whereas in theabsence of stimulation, gamma-frequency phaselocking actually increases by 20% � 5% (P<.01for a difference in the change in phase locking afterstimulation vs after detection only; 2-tailed t test).By contrast, the change in lower-frequency (10–30 Hz) phase locking was similar whether or notstimulation was present (change in phase lockingafter stimulation: �1% � 7%, change in phaselocking after detection event only: �4% � 6%;P 5 .74 by 2-tailed t test). Moreover, the markedlydifferent changes in gamma-frequency phaselocking after stimulation versus after detectiononly were also found when we analyzed datafrom neocortical and hippocampal leads sepa-rately (see Fig. 3C; P<.05 for a difference in thechange in phase locking after stimulation vs afterdetection only, for both hippocampal and neocor-tical data sets; 2-tailed t test).We also studied how stimulation affects rhyth-

mic amplitudes measured using our waveletanalysis (see Fig. 1). The mean baseline amplitude

40

50

60

70

-1 0 1 2

Stimulation

Detection-only

Time (sec, relative to detection)

Ampl

itude

Amplitudes of 10-100 Hz activityfrom all electrodes

A

Fig. 4. Responsive stimulation does not produce clear effeobtained via the wavelet transformation, over time, averatrode from a particular data set), which contained at least 1vs detection only). (B) Rhythmic amplitudes over time, nocant phase locking and frequencies in the gamma range (

(1 second before detection events) was verydifferent in ECOG records containing stimulationsand those containing detections only (74 � 2 vs120 � 4 arbitrary units, respectively). Therefore,we restricted our analysis to cases in which, fora particular patient, frequency, and electrode, themean amplitudes for ECOG records containingstimulations and those containing detection eventsonly differed by less than 25% (N 5 2464/4500cases). Aswith phase locking,weobserved a sharpincrease in rhythmic amplitudes between 500and 200 milliseconds before detection events(Fig. 4A). One second after the detection event,amplitudes had decreased sharply regardless ofwhether or not stimulation had occurred (seeFig. 4A). We observed a similar pattern even afterrestricting analysis to rhythmic activity in thegamma-frequency range from electrodes thatexhibit statistically significant phase locking (seeFig. 4B). We also observed a similar pattern whenwe normalized amplitudes, relative to all ampli-tudes recorded from the same channel. Thus,although detection events seem to be precededand followed by sharp changes in the amplitudesof rhythmic activity, we did not find evidence thatresponsive stimulation affects these changes.

DISCUSSION

We found that responsive stimulation acutely(ie, within 1 second) suppresses the long-rangesynchrony of gamma-frequency rhythmic activity inintracranial ECOG data. Long range refers tosynchronization between distinct leads, as opposedto local synchrony of activity recorded at a singlelocation. Notably, this long-range synchrony,

15

20

25

-1 0 1 2

Stimulation

Detection-only

Time (sec, relative to detection)

Ampl

itude

Amplitudes of 35-100 Hz activityfrom electrodes that exhibit phase-locking

B

cts on rhythmic amplitudes. (A) Rhythmic amplitudesged over all cases (ie, a particular frequency and elec-0 ECOG records in each condition (stimulation presentw restricted to cases that exhibited statistically signifi-35–100 Hz).

Neurostimulation Suppresses Phase-Locking 487

measured using the phase-locking statistic, risessharply during the second preceding detectionevents and continues to rise in the absence of stimu-lation.Thephase-lockingstatistic remainsdepressed2 seconds after the detection event, suggestingthat the effects of responsive stimulation are nottransient. Rather, responsive stimulation seems toaccelerate the return of the phase-locking statisticto baseline after the detection of epileptiform activity.

We initially focused on phase locking, quantifiedin a way that is independent of the amplitude of 2signals, because there is the possibility that ampli-fier gain could be compromised immediately afterstimulation because of effects such as saturation.However, thewavelet analysisweuseddidmeasurerhythmic amplitudes, and, although we observedsharp changes in rhythmic amplitudes before andafter the detection of epileptiform activity, we didnot find consistent effects of stimulation on theseamplitude changes.

Relationship with Possible TherapeuticMechanisms of Deep Brain Stimulation

Although we did not directly measure the clinicalconsequences of these effects, a decrease inlong-range rhythmic synchronyalmost bydefinitioncorresponds to a decrease in epileptiform activity.These effects represent clinical evidence for thehypothesis that deep brain stimulation reducespathologic cortical synchrony10 andmaybe relatedto the observation that responsive stimulation ismost effective during particular phases of an after-discharge.11 These effects may also explain howdeep brain stimulation could potentially disconnectthe site of stimulation from downstream targets7

because synchronized oscillations may mediatecommunication between brain regions. Changesin phase locking were confined to the gamma-frequency band, which is thought to be particularlyimportant for communication between corticalregions.12–14

LIMITATIONS

As described in the “Methods” and “Results”sections, we have tried to control for possibleconfounders in several ways. First, we focused onphases, rather than amplitudes, which could beaffected by amplifier saturation. Second, westudied the effects of stimulation 1 second aftera detection event in order to minimize electronicartifacts related to stimulation. Third, we band-pass filtered our signals to eliminate low-frequency artifacts of stimulation. In particular,the observed changes in phase locking at high,but not low, frequencies suggest that our resultsdo not reflect such artifacts. Fourth, we compare

changes in phase locking after stimulation withthose that occur after detection events in theabsence of stimulation in order to focus on trueeffects of stimulation rather than regression to themean and other factors related to the intrinsicevolution of neural signals after a detection event.

Despite these efforts, several possible con-founders remain. ECOG records containing detec-tion events, but not stimulations, are not perfectcontrols for ECOG records containing stimulationsbecause in many cases these were recorded fromthe same patients with the same electrode andlead configurations but at different times. Further-more, even though we only analyzed effects ofthe first stimulation in each ECOG record, ECOGrecords containing stimulations are likely to haveoccurred in the context of other stimulations. Asa result, the acute effects of stimulation that wedescribe should be interpreted in the context ofpossible subacute and chronic effects of stimula-tion.We studied the effects of stimulation on phaselocking by restricting our analysis to the subset ofcases (activity at a particular frequency, thatoccurred in a particular electrode pair, froma particular patient) that exhibited statisticallysignificant phase locking. As described earlier,these cases represent a small fraction of the totaldata collected. Of course, our data are limited inthat we could only sample from the locationswhereelectrodes were implanted and leads were placed.Thus, it is possible that stimulation had an effect onmany phase-locked oscillations in the brain that wecould not detect. In addition, as mentioned earlier,in order to avoid impact from stimulation artifacts,we did not analyze signals for 1 second after thedetection event. It is possible that by doing so, wemissed some acute effects of stimulation or under-estimated themagnitude of the stimulation-inducedsuppression of phase locking, which seems todecay over time (see Fig. 2). An inherent limitationis that electrodes were implanted in epileptogenicregions. Thus, our results are limited to analysis ofsuch regions andmay not necessarily reflect effectsof neurostimulation in or on other regions.

SUMMARY

We found that responsive stimulation acutely (eg,within 1 second) suppresses phase-locked oscilla-tions in the neocortex and hippocampus ofpatients with epilepsy. These results suggesta specific mechanism by which responsive stimu-lation could suppress epileptiform activity anddisconnect stimulated regions from downstreamtargets, possibly contributing to the therapeuticeffects of neurostimulation in epilepsy and otherneuropsychiatric conditions.

Sohal & Sun488

ACKNOWLEDGMENTS

We thank the editors of Neurosurgery Clinicsand the guest editors of this special issue forinviting this submission.

REFERENCES

1. Saillet S, Langlois M, Feddersen B, et al. Manipu-

lating the epileptic brain using stimulation: a review

of experimental and clinical studies. Epileptic Disord

2009;11(2):100–12.

2. Mayberg HS, Lozano AM, Voon V, et al. Deep brain

stimulation for treatment-resistant depression.

Neuron 2005;45(5):651–60.

3. Greenberg BD, Malone DA, Friehs GM, et al. Three-

year outcomes in deep brain stimulation for highly

resistant obsessive-compulsive disorder. Neuropsy-

chopharmacology 2006;31(11):2384–93.

4. Nuttin B, Gybels J, Cosyns P, et al. Deep brain stim-

ulation for psychiatric disorders. Neurosurg Clin N

Am 2003;14(2):xv–xvi.

5. Visser-Vandewalle V, Temel Y, Boon P, et al. Chronic

bilateral thalamic stimulation: a new therapeutic

approach in intractable Tourette syndrome. Report

of three cases. J Neurosurg 2003;99(6):1094–100.

6. Gradinaru V, Mogri M, Thompson KR, et al. Optical

deconstruction of parkinsonian neural circuitry.

Science 2009;324(5925):354–9.

7. Johansen-Berg H, Gutman DA, Behrens TE, et al.

Anatomical connectivity of the subgenual cingulate

region targeted with deep brain stimulation for

treatment-resistant depression. Cereb Cortex 2008;

18(6):1374–83.

8. Hauptmann C, Tass PA. Therapeutic rewiring by

means of desynchronizing brain stimulation. Biosys-

tems 2007;89(1–3):173–81.

9. Lachaux JP, Rodriguez E, Martinerie J, et al.

Measuring phase synchrony in brain signals. Hum

Brain Mapp 1999;8(4):194–208.

10. Popovych OV, Hauptmann C, Tass PA. Effective de-

synchronization by nonlinear delayed feedback.

Phys Rev Lett 2005;94(16):164102.

11. Motamedi GK, Lesser RP, Miglioretti DL, et al. Opti-

mizing parameters for terminating cortical afterdi-

scharges with pulse stimulation. Epilepsia 2002;

43(8):836–46.

12. Rodriguez E, George N, Lachaux JP, et al. Percep-

tion’s shadow: long-distance synchronization of

human brain activity. Nature 1999;397(6718):

430–3.

13. Womelsdorf T, Schoffelen JM, Oostenveld R, et al.

Modulation of neuronal interactions through

neuronal synchronization. Science 2007;316(5831):

1609–12.

14. Sohal VS, Zhang F, Yizhar O, et al. Parvalbumin

neurons and gamma rhythms enhance cortical circuit

performance. Nature 2009;459(7247):698–702.