Neuronal avalanches organize as nested theta-and beta/gamma-oscillations duringdevelopment of cortical layer 2/3Elakkat D. Gireesh and Dietmar Plenz*

Laboratory of Systems Neuroscience, National Institute of Mental Health, 9000 Rockville Pike, Bethesda, MD 20892

Edited by Nancy J. Kopell, Boston University, Boston, MA, and approved March 27, 2008 (received for review January 18, 2008)

Maturation of the cerebral cortex involves the spontaneous emer-gence of distinct patterns of neuronal synchronization, whichregulate neuronal differentiation, synapse formation, and serve asa substrate for information processing. The intrinsic activity pat-terns that characterize the maturation of cortical layer 2/3 arepoorly understood. By using microelectrode array recordings invivo and in vitro, we show that this development is marked by theemergence of nested �- and �/�-oscillations that require NMDA-and GABAA-mediated synaptic transmission. The oscillations orga-nized as neuronal avalanches, i.e., they were synchronized acrosscortical sites forming diverse and millisecond-precise spatiotem-poral patterns that distributed in sizes according to a power lawwith a slope of �1.5. The correspondence between nested oscil-lations and neuronal avalanches required activation of the dopa-mine D1 receptor. We suggest that the repetitive formation ofneuronal avalanches provides an intrinsic template for the selec-tive linking of external inputs to developing superficial layers.

dopamine � in vivo � multielectrode array � organotypic culture � rat

The development of the cortex is marked by distinct patternsof synchronized neuronal activity that emerge spontane-

ously, are preserved across different species, and regulate manycellular processes (1–4). In the early developing, deep corticallayers (5), the emergent activity is characterized by spindle-oscillations (�15 Hz) (2, 6, 7). In contrast, the patterns ofneuronal synchronization for the late-developing, superficiallayer 2/3 are poorly understood. For the mature layer 2/3, at leasttwo varieties of spontaneous synchronization have been re-ported. Fast oscillations such as �-oscillations (30–100 Hz) havebeen described in vivo (8, 9) and in vitro (10, 11) and reflect rapidneuronal synchronization that is thought to be important forsensory processing, motor activity, and cognitive functions (forthe latest reviews, see refs. 12 and 13). Although �-oscillationsare easily recognizable in the local population activity, theirtransient nature in the form of coherent bursts of variableduration, their amplitude modulation, and cooccurrence with�-oscillations (15–30 Hz) (14) has made it difficult to identify theorganization of spatiotemporal patterns that emerges from thisactivity. We recently described another type of spatiotemporaldynamics in mature layer 2/3 of rat cortex in vitro (15–18), whichwe termed ‘‘neuronal avalanches.’’ Neuronal avalanches arecharacterized by a scaling property in which neuronal synchro-nization, as measured in the local field potential (LFP) atdifferent cortical sites, emerges in the form of highly diversespatiotemporal patterns that distribute in size s according to apower law P(s) � s� and � � �1.5. The relationship in patternorganization between coherent, oscillatory bursts and neuronalavalanches and how both activities emerge in the developinglayer 2/3 is not clear.

Here, we show in vivo and in vitro that cortical layer 2/3, at thetime of their maturation, establish a dynamics characterized bynested �- and �/�-oscillations. Importantly, the oscillations arecoherent between cortical sites such that the resulting diversespatiotemporal patterns follow the power law statistics described

for neuronal avalanches. Our findings unify two seeminglydifferent views of neuronal synchronization—oscillations andavalanches—and suggest that the superficial layers form a‘‘critical state’’ (15, 19, 20) during cortical maturation that allowsfor the spontaneous, repetitive formation of diverse synchro-nized neuronal groups.

ResultsWe measured spontaneous neuronal activity in rat cortical layer2/3 at the beginning and the end of the second week postnatal,the time of superficial layer development (8 � 1 postnatal days,P8, n � 5; P13 � 2, P13, n � 7). An 8 � 4 microelectrode array(MEA) was inserted 1 mm deep into the somatosensory cortexof urethane-anesthetized rats (Fig. 1A) to record spontaneousLFP activity (1–200 Hz; 4-kHz sampling per electrode).

At P8, the LFP revealed oscillatory periods lasting �200 msthat occurred at a rate of 0.86 � 0.05 s�1 on the MEA (Fig. 1CLeft). Wavelet analysis demonstrated the presence of �- (4–15Hz), �- (15–30 Hz), and �-oscillations (30–100 Hz) during bursts(Fig. 1E Left). At P13, bursts appeared slightly more often(1.15 � 0.1 s�1; P � 0.04) with significantly increased amplitude(P � 0.005) and durations (P � 0.004) (Fig. 1 G and H).Importantly, between P8 and P13, the power at correspondingpeak frequencies increased 10- to 50-fold for each frequencyrange and the bursts now clearly revealed a temporal organiza-tion of �- and/or �-oscillations nested into �-oscillations (Fig. 1C Right, E Right, and I), which we will refer to as ‘‘nested �- and�/�-oscillations.’’

During the second week postnatal, the nested oscillations alsobecame more synchronized between cortical sites (Fig. 2). Basedon wavelet analysis, we quantified the neuronal synchronizationbetween sites for each frequency, i.e., coherence, and foundsignificant coherence at distinct frequency peaks at P8 and P13(Fig. 2 A and B) (�peak � 6.24 � 0.34 Hz; �peak � 22.63 � 0.21Hz; �peak � 37.97 � 0.07 Hz). The coherence was low earlyduring development, however, significantly increased at P13 (�,100%, P � 0.001; �, 50%, P � 0.001; �, 32%, P � 0.001) showinga clear distance dependence for each of the three frequencybands (Fig. 2C).

The emergence of synchronized nested �- and �/�-oscillationsduring the second week postnatal could reflect a change incortical organization or, alternatively, a change in the sensitivityto urethane anesthesia. It is well established that layer 2/3

Author contributions: E.D.G. and D.P. designed research; E.D.G. performed research; E.D.G.analyzed data; and E.D.G. and D.P. wrote the paper.

Conflict of interest statement: Provisional patent filing PCT/US2006/03 1884 ‘‘NeuronalAvalanche Assay.’’

This article is a PNAS Direct Submission.

*To whom correspondence should be addressed at: Section on Critical Brain Dynamics/Laboratory of Systems Neuroscience, Porter Neuroscience Research Center, NationalInstitute of Mental Health, Room 3A-100, 35 Convent Drive, Bethesda, MD 20892. E-mail:plenzd@mail.nih.gov.

This article contains supporting information online at www.pnas.org/cgi/content/full/0800537105/DCSupplemental.

7576–7581 � PNAS � May 27, 2008 � vol. 105 � no. 21 www.pnas.org�cgi�doi�10.1073�pnas.0800537105

Dow

nloa

ded

by g

uest

on

June

20,

202

0

develops in organotypic cortex cultures prepared from tissuetaken at early postnatal age (10, 21, 22). Therefore, slices ofsomatosensory cortex and the ventral tegmental area (VTA),which provides dopaminergic input to the cortex that facilitatesthe maturation of layer 2/3 fast-spiking interneurons (23), weretaken at P1–P2 and cocultured on planar 8 � 8 MEAs (Fig. 1B).Spontaneous extracellular activity in the cortex culture wasrecorded for up to 17 days in vitro (DIV) (Fig. 1 D and F) (n �15 cultures).

Activity before 6 DIV was low and mainly composed of sparsemultiunit activity; however, oscillatory bursts in the LFPemerged at 6–8 DIV corresponding to the beginning of thesecond week postnatal in vivo. As found in vivo, the bursts lastedseveral hundreds of milliseconds and increased in amplitude andduration during the second week (Fig. 1 G and H). The fre-quency composition of bursts in vitro also paralleled our findingsin vivo. The bursts at 6 DIV were composed of nested �- and�/�-oscillations that rapidly increased in power by an order ofmagnitude within the first half of the second week in culture (Fig.1 F and I). Two-dimensional current-source density analysis (see

Materials and Methods) [supporting information (SI) Fig. S1]demonstrated that the nested oscillations were located in theupper region of the cortical culture, which corresponds tocortical layer 2/3 (21). These similarities in our in vivo and in vitroexperiments support the idea that the emergence of nested �-and �/�-oscillations reflects the development of layer 2/3.

Nested �- and �/�-Oscillations Carry the Signature of NeuronalAvalanches. The transient nature and amplitude modulation ofthe coherent, nested oscillations suggest an underlying diversityin neuronal activity. We therefore recorded simultaneouslymultiunit activity (MU) (0.3–3 kHz) and the LFP at eachelectrode and calculated spike-triggered LFP averages. Weconsistently found in vivo and in vitro that the preferred time ofneuronal spiking correlated with negative LFP deflections(nLFP) (Fig. 2D and Fig. S2) (n � 3 rats; n � 4 cultures), in linewith findings in vivo during �-activity (9, 24–26). Correspond-ingly, the density of extracellular unit activity peaked clearlywithin �2 ms of the nLFP both in vivo and in vitro (Fig. 2D Rightand Fig. S2). These results suggest that characterizing the

A B

II/III

II/III

I

PIA

1

1

32

200 μmIED

V/VIV/VI

IVII

VTA

WM

WMWM

V/VI

VTA

Cortex

MEAin vivo in vitro

00 0

50

75

100

600 600

25

Freq

uenc

y(H

z)

E 2mV

60

0 600 time (ms)time (ms) time (ms)

0.1

10

80R

elat

ive

Pow

erI

F

G H

DIVDIV

in vitro in vitro in vivoin vivo in vivo

6 618 1812 12

800

200

00 0

12 50

θ β γ8 8 13 13 P P

nLFP

(μV)

Dur

atio

n(m

s)

* * * **

1

0

50

75

100

25

DIV

in vitro

6 18

4 - 15 Hz15 - 30 Hz

30 - 100 Hz

12

8

0 0

16

1mm

400 µm

2mV

DC P8 14 DIVP13

20 μV 1 s

40 μV 200 ms

12 DIV

Fig. 1. Nested �- and �/�-oscillations emerge during maturation of cortical layer 2/3. (A) Sketch of the 8 � 4 MEA placement in somatosensory cortex in vivo.(Left) PIA, pia mater; WM, white matter; IED, interelectrode distance of 200 �m. (Right) The 60-�m-thick Nissl-stained section with tracks from four electrodeshanks (arrows). (B) (Left) Sketch of the 8 � 8 MEA placement in the cortex–VTA coculture. (Right) Light microscopic image of the living coculture (12 DIV). IED,200 �m. (C and D) Spontaneous bursts at P8 and P13 in vivo (C) and at DIV 14 in vitro (D). Brackets, Periods in E and F. (E) Nested �- and �/�-oscillations emergeduring the second week postnatal in vivo. Time–frequency plot for burst periods in C at P8 (Left) and at P13 (Right) in vivo. (F) Nested �- and �/�-oscillations invitro. Time–frequency plot for burst period in D. (G and H) Burst amplitude (G) and duration (H) increase significantly during the second week postnatal. (I) �-,�-, and �-frequency power increases significantly during second week postnatal (normalized to power at P8 in vivo and DIV 6 in vitro).

Gireesh and Plenz PNAS � May 27, 2008 � vol. 105 � no. 21 � 7577

NEU

ROSC

IEN

CE

Dow

nloa

ded

by g

uest

on

June

20,

202

0

spatiotemporal organization of nLFPs might provide insight intothe diverse spatiotemporal organization of synchronized neuro-nal groups in the network. Furthermore, if the size distributionof spatiotemporal nLFP clusters follows the power law P(s) � s�,� � �1.5, the diversity would be similar to that described forneuronal avalanches (16–18). We therefore first identifiednLFPs at each electrode by negative threshold crossing andcombined all nLFPs on the array (Fig. 3A Top) into a raster ofnLFP times with corresponding amplitudes (Fig. 3A Middle). Wethen binned the raster at the temporal resolution �tavg, which isthe average time between successive nLFPs on the array(�tavg � 3.70 � 0.33 ms, P8; 2.01 � 0.21 ms, P13; P � 0.01) (seeMaterials and Methods) (16). Finally, we concatenated nLFPs onsuccessive time bins of width �tavg into nLFP clusters until a timebin with no nLFP was encountered (Fig. 3A Bottom). The size sof an nLFP cluster, either expressed as number of nLFPs, or theabsolute sum of nLFP amplitudes, was calculated and the densityof cluster sizes, P(s), was further examined.

At P8 in vivo, the nLFP clusters occurred at a rate of 2.4 � 0.3s�1, and P(s), when plotted in log–log coordinates, followed astraight line for sizes ranging from 6 to 60 �V, indicating theearly presence of a power law with a slope of � � �1.72 � 0.05(Fig. 3C, blue). At P13, the cluster rate had increased 10-fold(19 � 3 s�1; P � 0.005; P8 vs. P13) and the power law relationshipbecame more robust with a slope of � � �1.54 � 0.03 over the

range of 8–300 �V in cluster size (Fig. 3C, black). This slope isclose to the exponent of � � �1.5 reported previously forneuronal avalanches in mature cortex tissue (16, 18).

The power law did not simply arise from widely varying, noisyneuronal activities within and across experiments, but insteadindicated the presence of long-range spatiotemporal correlationsthat crucially depended on the precise phase relationships of thenested oscillations across sites. First, the power law was readilyobtained for single experiments; thus, it did not result fromaveraging across widely different experimental out comes (Fig.3 E and G). Second, phase-shuffling of the oscillations in thefrequency domain destroyed the power law with a correspondingloss in large nLFP clusters (Fig. 3 D–G, open red circles). Third,

Fig. 2. Coherence between cortical sites peaks at �-, �-, and �-frequencies.(A) Time course and corresponding time–coherence plot for two simultaneousbursts separated by 600 �m. Note transient period of high coherence at �- and�-frequencies at �500 ms and a longer lasting coherence at �-frequency. (B)Time-averaged coherence spectrum (black) from A reveals peaks in the �-, �-,and �-frequency band. (Red) Expected coherence from corresponding time-shifted traces. (C) Change in coherence during the second week and distance(d) dependence of the coherence for �-, �-, and �-frequencies (normalized tocorresponding coherence from time-shift traces). (D) Summary of size andrelative timing of the average nLFP peak for all electrodes with MU activity (invivo; n � 3 experiments; colors). (Left) Average MU-triggered LFP waveform atone electrode. (Center) Average nLFP amplitude normalized to the SD of theaverage nLFP from randomly chosen time points within a burst. (Right)MU-activity peaks around the time of the nLFP. Average unit density functionsaround the time of nLFP occurrences.

A

20 μV500 ms

32

Elec

.(n)

B

τmax

G HP(s)

D E

Single

1

20 μV100 ms

0

30μV

n n+1 n+2 n+3 n+4 n+5

P(s) P(s) Original

Shuffled

-510-510

1 40 40

Single F

Size s (n) Size s (n)

Size s ( )μV

Average

Average

1

C

8-10-510

-310

010

-310 -310

-310

010010

-410

-110

-110

CC

τ()

00 0.2-0.2

0.015 P 8P 13

Time (s)

in vitro

Size s (μV)10 10000

τmaxCC

(τ)

-0.2 0 0.20

0.03

Time (s)

Size s (μV)-510

0.3

P(s)

2 600

in vivoAverage

P13P8

Original

Shuffled

I

I

II

II

P(s)

2 600

Fig. 3. Nested �- and �/�-oscillations organize in the form of neuronalavalanches. (A) Definition of neuronal avalanches formed by the nested �- and�/�-oscillations. (Top) Threshold detection (broken line) of nLFPs (filled circles)at a single electrode. (Middle) Corresponding time–amplitude raster plot ofnLFPs on the MEA (color: nLFP amplitude). (Bottom) Spatiotemporal nLFPclusters occupy successive bins of width �tavg (dotted rectangles). (B) Averagecross-correlation function for nLFPs in vivo at P8 (red) and P13 (black; singleexperiments). (C) nLFP clusters from nested �- and �/�-oscillations organize inthe form of neuronal avalanches, i.e., distribute in sizes according to a powerlaw with slope close to � � �1.5 (broken line). Average cluster size distributionin vivo plotted in log–log coordinates for P8 (red open circles; n � 5) and P13(black; n � 7). (D) Example of two simultaneous burst periods before (black)and after (red) phase-shuffling. (E) The power law in cluster sizes is establishedfor cluster area and cluster intensity (G) in single in vivo experiments and in theaverage (n � 7; F; cp. also C; all P13), but is destroyed on phase-shuffling of theLFP (open red). (H) Average cluster size distribution in vitro follows a powerlaw with slope � � �1.5 (broken line; n � 15; �10 DIV). (Inset) Average nLFPcross-correlation function for single experiment.

7578 � www.pnas.org�cgi�doi�10.1073�pnas.0800537105 Gireesh and Plenz

Dow

nloa

ded

by g

uest

on

June

20,

202

0

similar results were obtained when cluster sizes were calculatedin the form of participating electrodes per cluster, which mea-sures the spatial extend of a cluster, i.e., cluster area (Fig. 3 Eand F).

The correspondence between the nested oscillations andneuronal avalanches was demonstrated also in vitro in thecortex–VTA cocultures. For cultures older than 9 DIV, a powerlaw ranging from 10 to �10,000 �V described the distribution ofnLFP cluster size s with a corresponding slope value of � ��1.53 � 0.02, which, again, is close to �1.5 (Fig. 3H) (n � 15cultures) (see also Fig. S3). Finally, the lifetime distribution forneuronal avalanches arising from nested oscillations in vivo andin vitro was also in accordance with the lifetimes of neuronalavalanches in mature cortex (Fig. S4). These results establisheda clear correspondence between neuronal avalanches as de-scribed for the mature cortex and coherent nested �/�-oscillations during the development of superficial layers.

Nested �- and �/�-Oscillations Require GABAA and Glutamate NMDAReceptor Activation. We then identified the pharmacologicalprofile of the nested oscillations to better understand theirdependence on the properties of the cortical network. Blockadeof fast inhibition by using the GABAA receptor antago-nist picrotoxin (10 �M) changed the sinusoidal-like nestedoscillations into prolonged periods of ictal-like, sharp waves (Fig.4A). Correspondingly, the relatively narrow and well defined�-frequency peak in the power spectrum broadened substantiallywith a reduction in peak frequency (P � 0.03, n � 3) and power(Fig. 4 A–D) (P � 0.02), whereas the power for the peak�-frequency increased (Fig. 4D) (P � 0.008). As reported formature cortex (18), picrotoxin also destroyed the power law inavalanche size distribution and supported a preferential increasein large avalanches (Fig. 4E).

The nested oscillations also depended on glutamatergic syn-aptic transmission. The glutamate NMDA receptor antagonistDL-2-amino-5-phosphonopentanoic acid (AP5) (50 �M) abol-ished �-oscillations by reducing the power at the peak �- and�-frequency by 80 and 50%, respectively (Fig. 4D) (P � 0.01, P �0.02, n � 3) without changing the peak frequencies (Fig. 4C).Correspondingly, blockade of the NMDA receptor abolished thepower law (Fig. 4E). In contrast, the AMPA receptor antagonist6,7-dinitroquinoxaline-2,3(1H,4H)-dione (DNQX) (10 �M),whereas reducing the peak �-frequency by �10 Hz, did notchange the �-power (P � 0.01, n � 3) and maintained the powerlaw in avalanche size with a slope � near �1.5 (Fig. 4 B and E)(�1.59 � 0.02, pre; �1.66 � 0.05, drug; �1.63 � 0.04, post) (P �0.6). Blockade of fast glutamatergic and GABAergic inhibitiondid not induce significant changes in the �-frequency range.

The maturation of layer 2/3 in rat cortex involves a shift fromgap junction (27, 28) to chemical synapse-mediated communi-cation (29). In line with this transition, the gap junction blockercarbenoxolone, which reduces fast oscillations in entorhinalcortex at 0.1 mM (11), did not affect the nested oscillations at thisconcentration (n � 3) (Fig. S5). Finally, because our mesence-phalic VTA slice might have included cholinergic subcorticaltissue, we also tested for the effect of muscarinic (M1) receptors,which are reported to induce cortical �- and �-oscillations duringthe first week postnatal (30). However, the cholinergic M1receptor antagonist pirenzepine did not affect the nested oscil-lations in cortical layer 2/3 (10 �M; n � 3) (Fig. S5).

To summarize, the pharmacological profile of the nestedoscillations and neuronal avalanches demonstrate that this dy-namics arises from an immature excitatory–inhibitory networkwith a dominance of GABAA and glutamate NMDA receptor-mediated synaptic transmission that is independent of gap-junction coupling.

Dopamine Regulates the Correspondence Between Nested Oscilla-tions and Avalanches. In the adult cortex, dopamine D1 receptorstimulation regulates neuronal avalanche formation (16) andincreases pyramidal cell to fast-spiking interneuron coupling(31) that facilitates the formation of �-oscillations (32). Wetherefore tested whether the neuromodulator dopamine regu-lates the correspondence between nested oscillations and neu-ronal avalanches early during development. Indeed, we found invivo and in vitro that the dopamine D1 receptor antagonistR()-7-chloro-8-hydroxy-3-methyl-1-phenyl-2,3,4,5-tetrahydro-1H-3-benzazepine hydrochloride (SCH23390) (20 �M) selec-tively reduced the �-oscillation power (Fig. 5 A, D, and E) (�:�33%, P � 0.01, �61%, P � 0.03; n � 5 in vivo; n � 6 in vitro).Importantly, the antagonist made the avalanche size distributionsteeper (Fig. 5 B and C) (� � �1.74 � 0.02, P � 0.003, in vivo;

Fig. 4. Nested �- and �/�-oscillations in vitro predominantly depend on theGABAA and glutamatergic NMDA receptor. (A) Example bursts before (Pre),during (Drug), and after drug application (Wash). Picrotoxin (PTX) (10 �M)(Left) induces prolonged, ictal-like bursts, whereas DNQX (10 �M) has noeffect (Center). In contrast, AP5 (50 �M) (Right) selectively abolishes high-frequency oscillations. (B) Time-averaged wavelet spectrum of the color-coded traces in A. Note the shift toward �-oscillations caused by PTX and theblock of high-frequency oscillations by AP5. (C) Change in peak frequencywithin each frequency band for all three antagonists. (D) Change in peakpower within each frequency band (normalized to the pre condition). (E)Disinhibition by PTX increases the number of large avalanche sizes (Left;arrow), whereas AP5 blocks most avalanches (Right; arrow; single cultures).For better visualization, each distribution was normalized to the maximumvalue of P and transformed by P 3 P � s�� (� taken from precondition). Thistransformation changes the power law into a horizontal distribution.

Gireesh and Plenz PNAS � May 27, 2008 � vol. 105 � no. 21 � 7579

NEU

ROSC

IEN

CE

Dow

nloa

ded

by g

uest

on

June

20,

202

0

� � �1.60 � 0.03, P � 0.006, in vitro), indicating a relativereduction of large avalanches, and significantly reduced thecoherence of �-oscillations for all distances on the array (Fig. 5F)(in vivo: mean � �32%, P � 0.001; in vitro: mean � �40%, P �0.005). The antagonist did not change oscillatory burst ampli-tudes, nor burst durations in vivo or in vitro (P � 0.88, P � 0.43).Whereas a small reduction in power was also observed in the�-frequency band (in vivo: �17%, P � 0.03; in vitro: �30%, P �0.3), no reduction was observed in the �-frequency band (in vivo:P � 0.3; in vitro: P � 0.16). In contrast, the dopamine D2 receptorantagonist sulpiride did not significantly alter the avalanche sizedistribution or oscillatory bursts (Fig. S6). These findings estab-lish a developmental role for dopamine for the emergence ofnested oscillations and neuronal avalanches.

DiscussionWe demonstrated that coherent �- and �/�-oscillations sponta-neously emerged during the formation of superficial corticallayers and that the resulting spatiotemporal patterns of neuronalsynchronization were equivalent to neuronal avalanches. Thus,the transient nature of the coherent, oscillatory bursts results indiverse spatiotemporal activity patterns that follow a precisescaling relationship given by the power law in pattern sizes withan exponent of �1.5 (15).

Earlier studies have shown coherent oscillations in the form ofspindle bursts or delta-brushes during cortex development invivo and in vitro (2, 6, 7). Spindle bursts occur during the firstweek postnatal, are located in deep cortical layers, show a lowfrequency range (5–25 Hz), and only weakly depend on theGABAA and NMDA receptor. In contrast, the nested �- and�/�-oscillations reported in the present study emerge during thesecond week postnatal in superficial layers, cover a frequencyrange from 4 to 100 Hz, and depend on the GABAA andglutamate NMDA receptor. Correspondingly, the fast timecourse of the nested �- and �/�-oscillations requires a matureGABAergic system (33) that develops between P10 and P15 inrat cortex (34). The differential effect of GABAA blockade on�- and �-oscillations in the current study has also been found forkainate-induced cortical �/�-activity in vitro for later develop-mental stages (35). The reduction in �-oscillations is in line withan inhibitory interneuron microcircuit in layer 2/3 that is re-sponsible for �-activity (32). The enhancement of �- power couldresult from a change in local microcircuit dynamics (36), an‘‘uncovering’’ of �-generation from deeper layers (35), or simplyreflect a frequency component arising from the ictal spikes. Thesensitivity of the dynamics to the NMDA and not the AMPAreceptor antagonists early during development suggests an im-mature glutamatergic network (4), whose temporal dynamics ismainly determined by the fast inhibition. Our results alsoestablished that this immature network requires dopamine D1receptor activation to maintain the correspondence betweennested oscillations and neuronal avalanches.

The linking of the coherence of the fast �/�-oscillations to anabsolute criterion, i.e., the power law exponent � � �1.5, mightallow for a simple and direct evaluation of the amount ofoscillation synchrony necessary to maintain the default networkstate of cortical layer 2/3. Conversely, a deviation from � � �1.5could identify pathological conditions related to �-oscillations,NMDA receptor functioning, and/or dopamine during develop-ment, as, e.g., suggested for schizophrenia (37, 38).

The neuronal avalanches capture a large variety of spatiotem-poral patterns. In the temporal domain these patterns rangefrom ‘‘instantaneous synchronization,’’ i.e., when all nLFPs arein one time bin, to ‘‘sequential synchronization’’ (39) when eachnLFP occupies exactly one time bin. In the spatial domain, thepatterns can include just one or all sites in the network. Incontrast, coherent oscillations in a homogenous medium pro-duce much simpler patterns, for example, stationary or travelingwaves. Such waves would represent synchronized or successiveactivation of neurons at all sites in the network at every cycle ofthe oscillation: a relatively simple pattern. Our findings suggestthat coherent oscillations in superficial layers coexist withcomplex spatiotemporal patterns whose diversity is quantified bythe power law in pattern size distribution. Thus, the networkdynamics produces avalanches that approximately repeat at peak�/�-oscillation frequency where spatial patterns for all sizes n �s are n�1.5 less likely to occur than patterns of size s (n is aconstant; see also ref. 15).

Although power laws in event size distributions can result fromvarious mechanisms (40), such distributions will arise when asystem is in a ‘‘critical state’’ as pioneered by complex systemsresearch in statistical physics (19, 41). Critical state dynamicsallows local events to percolate through the system: although themost likely event in the system is local, i.e., of relatively small size,large events that engage many sites cannot be ignored. Such anorganization of cortical activity, which is regulated homeostati-cally (42), differs profoundly from disinhibited activity, wheremost events are either small and local or very large and engagemost of the network resulting in bimodal event size distributions(18). Neuronal modeling suggests that networks in a critical stateoptimize information transfer and the dynamical range of input–output processing (18, 43, 44), increase speed and complexity of

F

A

SCH SCH

Pre Pre

WashWash

40 μV100 ms

D E

60006

in vitro

Size s (μV)

B C1

1

1

2 600

0.03

0.03

0.03

in vivo

Size s (μV)

0 0

1

2γ–

Coh

.(r e

l.)

in vivo in vivo in vivoin vitro in vitro in vitro

*

*

*

PreSCHWash

γ–Po

wer

(rel

.)

80

1

3

*

γ–Fr

eq.(

Hz)

P -αs

P -αs

P -αs.

.

.

Fig. 5. Tonic dopamine D1 receptor activation organizes nested �- and�/�-oscillations into neuronal avalanches. (A) In vitro example bursts before(pre), during SCH23390 (SCH) (10 �M), and after (wash). (B) SCH23390 reducesthe probability of large avalanches in vivo. The size distribution of avalanchesbefore (Top), during (Middle), and after (Bottom) drug application are over-plotted for each experiment (n � 5; P12–P15, color-coded). Each distributionwas normalized and transformed by P 3 P � s��. A downward tilt from thehorizontal precondition (upper arrow) and early cutoff in the distribution(bottom arrow) indicates a reduction in large avalanches, i.e., a steeper slope� in the distribution (slope: P � 0.003; pre vs. drug). (C) Change in slope � inresponse to 10 �M SCH23390 for the in vitro networks (DIV 10–15). As shownin B for in vivo, SCH23390 reduces the probability of larger avalanches (n � 6;DIV 10, color-coded; slope, P � 0.003; pre vs. drug). (D) The power at peak�-frequency is reduced by SCH23390 (�33%, P � 0.01 in vivo; �61%, P � 0.03in vitro). (E) The peak �-frequency is unaffected by SCH23390 (P 0.2 in vivo;P 0.1 in vitro). (F) The coherence across electrodes at peak �-frequency isreduced by SCH23390 (in vivo: mean, �32%, P � 0.001; in vitro: mean, �40%,P � 0.005).

7580 � www.pnas.org�cgi�doi�10.1073�pnas.0800537105 Gireesh and Plenz

Dow

nloa

ded

by g

uest

on

June

20,

202

0

neuronal pattern formation (45, 46), and allow for diverse andprecise network synchronization in the absence of epilepsy (17,18). These aspects could be of importance as the emergence ofthe neuronal avalanches and nested �- and �/�-oscillationsduring the late second week postnatal coincides with the devel-opment of sensory maps in superficial layers (47). Importantly,the transient nature of the nested oscillations results in arepetitive formation of diverse avalanches within a short period.Such a repetition might efficiently link external inputs to spatiallydiverse, but synchronized neuronal groups in superficial layers.

Materials and MethodsRats were anesthetized with urethane (1.25–1.75 g/kg body weight, i.p.). TheMEA (Neuronexus Technologies) was inserted in the coronal plane into thesomatosensory cortex (�3 mm lateral, 1 mm caudal to bregma) (Fig. 1A). Theanimal was hydrated frequently (Ringer’s lactate solution; 0.5–1 ml/h, i.p.) andwas monitored for respiratory rate (80–120/min), tail color, and tail pinchreflex. Anesthesia was maintained with supplemental doses of urethane(�0.25 g/kg). After the recordings, the brains were fixed in paraformaldehyde,sectioned, and Nissl-stained to reconstruct the electrode locations. Coronalslices from rat somatosensory cortex (350 �m thick) and the VTA (500 �m thick)were taken from pups (P0–P2; Sprague–Dawley), placed on a poly-D-lysine-

coated 8 � 8 planar MEA (MultiChannelSystems), cocultured, and recorded instandard culture medium (42).

Drugs were either applied topically to the cortex surface (in vivo) orbath-applied (in vitro). One hour/4 h of predrug condition were followed by15 min/4 h of drug application, followed by up to 1 h/4 h of wash (in vivo/invitro). SCH23390 (20 �M; Sigma-Aldrich) was dissolved in PBS and sulpiride (20�M; RBI) was dissolved in DMSO and further diluted in PBS (�0.01% finalconcentration of DMSO).

A Morlet-wavelet (12 suboctaves) was used for frequency and coherenceanalysis. From the time-averaged wavelet spectrum, the corresponding peakpower and frequency was calculated within the �-, �-, and �-frequency bands.Coherence was estimated as the average coherence between the cortical siteswith the largest nLFP in relation to all other active sites per nested oscillationperiod.

For neuronal avalanche analysis (16, 18), nLFPs were identified by thresholdcrossing (�4SD at P8 � 1, �3SD at P13 in vivo, and �3SD in vitro) (Fig. 3) andwere grouped into clusters for successive time bins of duration �tavg. The slope� was obtained by linear regression from the cluster size distributions. Forfurther details, see SI Text.

ACKNOWLEDGMENTS. We thank Craig Stewart for help with the preparationof the cultures and Drs. T. Petermann, T. Thiagarajan, W. Shew, Y. LeFranc, T.Bellay, H. Parthasarathy, and R. D. McKay for valuable comments during theconduct of this project. This work was supported by the Intramural ResearchProgram of the National Institutes of Health and National Institute of MentalHealth.

1. Spitzer NC (2006) Electrical activity in early neuronal development. Nature 444:707–712.

2. Khazipov R, Luhmann HJ (2006) Early patterns of electrical activity in the developingcerebral cortex of humans and rodents. Trends Neurosci 29:414–418.

3. O’Donovan MJ (1999) The origin of spontaneous activity in developing networks of thevertebrate nervous system. Curr Opin Neurobiol 9:94–104.

4. Ben-Ari Y (2001) Developing networks play a similar melody. Trends Neurosci 24:353–360.

5. Ignacio MP, Kimm EJ, Kageyama GH, Yu J, Robertson RT (1995) Postnatal migration ofneurons and formation of laminae in rat cerebral cortex. Anat Embryol (Berlin)191:89–100.

6. Khazipov R, et al. (2004) Early motor activity drives spindle bursts in the developingsomatosensory cortex. Nature 432:758–761.

7. Minlebaev M, Ben Ari Y, Khazipov R (2007) Network mechanisms of spindle-burstoscillations in the neonatal rat barrel cortex in vivo. J Neurophysiol 97:692–700.

8. Murthy VN, Fetz EE (1992) Coherent 25- to 35-Hz oscillations in the sensorimotor cortexof awake behaving monkeys. Proc Natl Acad Sci USA 89:5670–5674.

9. Chrobak JJ, Buzsaki G (1998) Gamma oscillations in the entorhinal cortex of the freelybehaving rat. J Neurosci 18:388–398.

10. Plenz D, Kitai ST (1996) Generation of high-frequency oscillations in local circuits of ratsomatosensory cortex cultures. J Neurophysiol 76:4180–4184.

11. Cunningham MO, et al. (2004) A role for fast rhythmic bursting neurons in corticalgamma oscillations in vitro. Proc Natl Acad Sci USA 101:7152–7157.

12. Fries P, Nikolic D, Singer W (2007) The gamma cycle. Trends Neurosci 30:309–316.13. Jensen O, Kaiser J, Lachaux JP (2007) Human gamma-frequency oscillations associated

with attention and memory. Trends Neurosci 30:317–324.14. Steriade M (2006) Grouping of brain rhythms in corticothalamic systems. Neuroscience

137:1087–1106.15. Plenz D, Thiagarajan TC (2007) The organizing principles of neuronal avalanches: Cell

assemblies in the cortex? Trends Neurosci 30:101–110.16. Stewart CV, Plenz D (2006) Inverted-U profile of dopamine-NMDA-mediated sponta-

neous avalanche recurrence in superficial layers of rat prefrontal cortex. J Neurosci26:8148–8159.

17. Beggs JM, Plenz D (2004) Neuronal avalanches are diverse and precise activity patternsthat are stable for many hours in cortical slice cultures. J Neurosci 24:5216–5229.

18. Beggs JM, Plenz D (2003) Neuronal avalanches in neocortical circuits. J Neurosci23:11167–11177.

19. Stanley HE (1971) Introduction to Phase Transitions and Critical Phenomena (OxfordUniv Press, New York).

20. Bak P (1996) How Nature Works: The Science of Self-Organized Criticality (CopernicusBooks, New York).

21. Gotz M, Bolz J (1992) Formation and preservation of cortical layers in slice cultures.J Neurobiol 23:783–802.

22. Plenz D, Aertsen A (1996) Neural dynamics in cortex-striatum co-cultures. I. Anatomyand electrophysiology of neuronal cell types. Neuroscience 70:861–891.

23. Porter LL, Rizzo E, Hornung JP (1999) Dopamine affects parvalbumin expression duringcortical development in vitro. J Neurosci 19:8990–9003.

24. Gray CM, Singer W (1989) Stimulus-specific neuronal oscillations in orientation col-umns of cat visual cortex. Proc Natl Acad Sci USA 86:1698–1702.

25. Destexhe A, Contreras D, Steriade M (1999) Spatiotemporal analysis of local fieldpotentials and unit discharges in cat cerebral cortex during natural wake and sleepstates. J Neurosci 19:4595–4608.

26. Hasenstaub A, et al. (2005) Inhibitory postsynaptic potentials carry synchronizedfrequency information in active cortical networks. Neuron 47:423–435.

27. Yuste R, Peinado A, Katz LC (1992) Neuronal domains in developing neocortex. Science257:665–669.

28. Kandler K, Katz LC (1998) Coordination of neuronal activity in developing visual cortexby gap junction-mediated biochemical communication. J Neurosci 18:1419–1427.

29. Kandler K, Katz LC (1998) Relationship between dye coupling and spontaneous activityin developing ferret visual cortex. Dev Neurosci 20:59–64.

30. Buhl EH, Tamas G, Fisahn A (1998) Cholinergic activation and tonic excitation inducepersistent gamma oscillations in mouse somatosensory cortex in vitro. J Physiol 513(Pt1):117–126.

31. Seamans JK, Yang CR (2004) The principal features and mechanisms of dopaminemodulation in the prefrontal cortex. Prog Neurobiol 74:1–58.

32. Bartos M, Vida I, Jonas P (2007) Synaptic mechanisms of synchronized gamma oscilla-tions in inhibitory interneuron networks. Nat Rev Neurosci 8:45–56.

33. White JA, Banks MI, Pearce RA, Kopell NJ (2000) Networks of interneurons with fastand slow �-aminobutyric acid type A (GABAA) kinetics provide substrate for mixed �-�rhythm. Proc Natl Acad Sci USA 97:8128–8133.

34. Micheva KD, Beaulieu C (1996) Quantitative aspects of synaptogenesis in the rat barrelfield cortex with special reference to GABA circuitry. J Comp Neurol 373:340–354.

35. Roopun AK, et al. (2006) A beta2-frequency (20–30 Hz) oscillation in nonsynapticnetworks of somatosensory cortex. Proc Natl Acad Sci USA 103:15646–15650.

36. Kopell N, Ermentrout GB, Whittington MA, Traub RD (2000) � rhythms and � rhythmshave different synchronization properties. Proc Natl Acad Sci USA 97:1867–1872.

37. Lewis DA, Levitt P (2002) Schizophrenia as a disorder of neurodevelopment. Annu RevNeurosci 25:409–432.

38. Winterer G, Weinberger DR (2004) Genes, dopamine and cortical signal-to-noise ratioin schizophrenia. Trends Neurosci 27:683–690.

39. Abeles M (1992) Corticonics (Cambridge Univ Press, New York).40. Newman MEJ (2005) Power laws, Pareto distributions and Zipf’s law. Contemp Physics

46:323–351.41. Bak P, Paczuski M (1995) Complexity, contingency, and criticality. Proc Natl Acad Sci

USA 92:6689–6696.42. Stewart CV, Plenz D (2008) Homeostasis of neuronal avalanches during postnatal

cortex development in vitro. J Neurosci Methods 169:405–416.43. Kinouchi O, Copelli M (2006) Optimal dynamical range of excitable networks at

criticality. Nat Phys 2:348–351.44. Eurich CW, Herrmann JM, Ernst UA (2002) Finite-size effects of avalanche dynamics.

Phys Rev E Stat Nonlin Soft Matter Phys 66:066137.45. Herz AV, Hopfield JJ (1995) Earthquake cycles and neural reverberations: Collective

oscillations in systems with pulse-coupled threshold elements. Phys Rev Lett 75:1222–1225.

46. Legenstein R, Maass W (2007) Edge of chaos and prediction of computational perfor-mance for neural circuit models. Neural Netw 20:323–334.

47. Stern EA, Maravall M, Svoboda K (2001) Rapid development and plasticity of layer 2/3maps in rat barrel cortex in vivo. Neuron 31:305–315.

Gireesh and Plenz PNAS � May 27, 2008 � vol. 105 � no. 21 � 7581

NEU

ROSC

IEN

CE

Dow

nloa

ded

by g

uest

on

June

20,

202

0