Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Look up keyword
Like this
0Activity
0 of .
Results for:
No results containing your search query
P. 1
Buzsaki and Draguhn 2004

Buzsaki and Draguhn 2004

Ratings: (0)|Views: 25|Likes:
Published by Andre Watson
Cortical oscillations in neuronal networks
Cortical oscillations in neuronal networks

More info:

Published by: Andre Watson on Jan 15, 2014
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

01/15/2014

pdf

text

original

 
Neuronal Oscillations in Cortical Networks
Gyo¨rgy Buzsa´ki
1
* and Andreas Draguhn
2
Clocks tick, bridges and skyscrapers vibrate, neuronal networks oscillate. Are neuronaloscillations an inevitable by-product, similar to bridge vibrations, or an essential part of the brain’s design? Mammalian cortical neurons form behavior-dependent oscillatingnetworks of various sizes, which span five orders of magnitude in frequency. Theseoscillations are phylogenetically preserved, suggesting that they are functionally rele-vant. Recent findings indicate that network oscillations bias input selection, temporallylink neurons into assemblies, and facilitate synaptic plasticity, mechanisms that coop-eratively support temporal representation and long-term consolidation of information.
T
he first human electroencephalographic(EEG) pattern described was an 8 to 12Hz rhythm, the alpha waves of Berger (
1
), followed by a barrage of intensive clinicaland basic research. From scalp recordings, in-vestigators identified various other oscillatory patterns that were particularly obvious duringrest and sleep. However, the scalp EEG duringconscious, waking behavior demonstrated low-amplitude, “desynchronizedpatterns. Thisapparent inverse relation between cognitiveactivity and brain rhythms was further em- phasized by the dominance of oscillations inanesthesia and epilepsy, states associated with loss of consciousness (
2
). Therefore, themotivation to relate these “idling” or evenharmful rhythms to complex cognitive brainoperations was diminished.The recent resurgence of interest in neu-ronal oscillations is a result of several paralleldevelopments. Whereas in the past we simplywatched oscillations, we have recently beguncreating them under controlled situations (
3
 – 
8
). Detailed biophysical studies revealed thateven single neurons are endowed with com- plex dynamics, including their intrinsic abil-ities to resonate and oscillate at multiple fre-quencies (
9
,
 10
), which suggests that precisetiming of their activity within neuronal net-works could represent information. At thesame time, the neuronal assembly structuresof the oscillatory patterns found during sleepwere related to the experiences of the previ-ous awake period (
11
,
 12
). These results led to the tantalizing conjecture that perception,memory, and even consciousness could resultfrom synchronized networks (
13
 – 
17 
). Thesynchronous activity of oscillating networksis now viewed as the critical “middle ground”linking single-neuron activity to behavior (
2
 – 
,
 15
). This emerging new field, “neuronaloscillations,” has created an interdisciplinary platform that cuts across psychophysics, cog-nitive psychology, neuroscience, biophysics,computational modeling, physics, mathemat-ics, and philosophy (
2
 – 
11
,
 13
 – 
22
).
A System of Brain Oscillators
 Neuronal networks in the mammalian fore- brain demonstrate several oscillatory bandscovering frequencies from approximately0.05 Hz to 500 Hz (Fig. 1). The mean fre-quencies of the experimentally observed os-cillator categories form a linear progressionon a natural logarithmic scale (
23
) with aconstant ratio between neighboring frequen-cies, leading to the separation of frequency bands. Neighboring frequency bands withinthe same neuronal network are typically as-sociated with different brain states and com- pete with each other (
15
,
 24
 – 
26 
). On theother hand, several rhythms can temporallycoexist in the same or different structures and interact with each other (
2
,
 25
).The power density of EEG or local field  potential is inversely proportional to frequency(
 f  
 ) in the mammalian cortex (
27 
) (Fig. 1C).This
 1/f  
 power relationship implies that pertur- bations occurring at slow frequencies can causea cascade of energy dissipation at higher fre-quencies (
28
) and that widespread slow oscil-lations modulate faster local events (
2
,
 25
,
 29
).These properties of neuronal oscillators are theresult of the physical architecture of neuronalnetworks and the limited speed of neuronalcommunication due to axon conduction and synaptic delays (
30
). Because most neuronalconnections are local (
31
), the period of oscil-lation is constrained by the size of the neuronal pool engaged in a given cycle. Higher frequen-cy oscillations are confined to a small neuro-nal space, whereas very large networks arerecruited during slow oscillations (
2
,
 25
).These relations between anatomical architec-ture and oscillatory patterns allow brain op-erations to be carried out simultaneously atmultiple temporal and spatial scales (
32
).
Assembly Synchronization byOscillation
Integrationofinformationrequires“synchronyof the convergent inputs. Synchrony is defined 
1
Center for Molecular and Behavioral Neuroscience,Rutgers, The State University of New Jersey, Newark,NJ 07102, USA.
 2
Institut fu¨r Physiologie und Patho-physiologie, University of Heidelberg, INF 326, 69120Heidelberg, Germany.*To whom correspondence should be addressed. E-mail: buzsaki@axon.rutgers.edu
Fig. 1.
 An interacting system of brain oscilla-tors. (
A
) Power spectrum of hippocampal EEGin the mouse during sleep and waking periods.The spectrum was “whitened” by removing logslope [as shown in (C)]. Note four peaks closeto
 ln
 integers. Color-code band peaks as in (B).(
B
) Oscillatory classes in the rat cortex. Notethe linear progression of the frequency classeson the
 ln
 scale. For each band, the range of frequencies is shown, together with its com-monly used term (
23
). (
C
) Power spectrum of EEG from the right temporal lobe in a sleepinghuman subject. Subdural recording. Note thenear-linear decrease of log power with increas-ing log frequency from 0.5 to 100 Hz. [Adaptedwith permission from
 23
 (A and B) and
 27
 (C)]
R
EVIEW
25 JUNE 2004 VOL 304 SCIENCE www.sciencemag.org
1926
 
 by the temporal window within which sometrace of an earlier event is retained, which thenalters the response to a subsequent event. Incontrast, successive events that evoke identicalresponses are deemed nonsynchronous. Al-though synchronous assemblies can also be brought about by strong common inputs thatoccur irregularly, oscillation-based synchronyis the most energy-efficient physical mecha-nism for temporal coordination (
21
,
 22
). Brainoscillators, like most biological rhythms, be-long to limit-cycle and weakly chaotic oscilla-tors and share features of both harmonic and relaxation oscillators (
22
,
 33
). The macroscopicappearance of several brainrhythms, such as the 5- to10-Hzthetaoscillationsinthehippocampus, resembles thesinusoid pattern of harmonicoscillators. A major advan-tageofharmonicoscillatorsisthat their long-term behavior can be predicted from short-term observations of their  phase angle (
22
,
 33
). Howev-er, groups of harmonic oscil-lators poorly synchronizetheir phases (
5
). On the other hand, these macroscopic os-cillations are generated byneurons, whose spiking pat-terns share characteristicswith relaxation oscillators.Because of its phase-dependent excitability, the re-laxation oscillator separatesthe information transfer (
duty cycle
) phase from thereceiving phase. Relaxationoscillators synchronize ro- bustly and with great stability(
5
). These combined featuresofbrainoscillatorsmaketheir time course predictable and their phase easy to reset.
Scaling of OscillatoryNetworks in GrowingBrains
Because the different class-es of oscillations (Fig. 1)and their behavioral correlates are largely pre-served throughout the mammalian evolution (
,
13
,
 14
,
 17 
,
 20
,
 34
), it is reasonable to assumethat they are supported by universal mecha-nisms in brains of various sizes. Oscillationsemerge from the dynamic interplay betweenintrinsic cellular and circuit properties (
2
 – 
5
,
 – 
10
). Whereas the spiking of single cortical principal neurons typically displays Poissonstatistics (
35
), their assembly behavior is oftencharacterized by oscillatory properties (
,
 8
,
13
). Complex brains have developed special-ized mechanisms for the grouping of principalcells into temporal coalitions: inhibitory inter-neuron
 “
clocking
 networks (
19
,
 32
). In manysystems, electrical coupling by gap junctionsassists chemical synaptic signaling in oscillato-ry synchronization (
3
,
 36 
 – 
38
). However, localconnections alone place major constraints onglobal synchrony in growing brains (
21
,
 31
). Inthe cortex, the densely connected local neuronnetworks are supplemented by a small fractionof long-range connections (
31
), which ef-fectively reduces the synaptic path lengths between distant cell assemblies (
32
). This ar-chitectural design, reminiscent of the mathe-matically defined 
 “
small-world 
 networks (
39
),keeps the synaptic path lengths short and main-tains fundamental functions in growing brainswithout excessive wiring. Despite the progres-sively decreasing fraction of long-range con-nections in larger brains, synchronization of local and distant networks can be readily ac-complished by oscillators because of the lowenergy costs involved in coupling rhythms.
Functions for Brain Oscillators
The specific benefits of a particular oscilla-tion depend on the function of the brainsystem that supports it. Nevertheless, thereare a few general principles, some of whichare independent of physical substrate.
 Input selection and plasticity
. Single neu-rons and networks respond with transient os-cillations to a strong input. The natural fre-quency, or eigenfrequency, of the damped oscillation is a result of two opposing effects.The leak conductance and capacitance of theneuronal membrane are mainly responsiblefor the low-pass filtering property of neurons.On the other hand, several voltage-gated cur-rents, whose activation range is close to theresting membrane potential, act as high-passfilters, making the neuron responsive to fasttrains of spikes (
9
,
 10
). The appropriate com- bination of high-pass (voltage-dependent) and low-pass (time-dependent)filtering properties of neu-rons can be exploited for theconstruction of resonators(band-pass filters),
 
notch
orband-stopfilters,andsub-threshold oscillators (
40
 – 
43
) (Fig. 2). These resonant-oscillatory features allowneurons to select inputs based on their frequencycharacteristics. Input fre-quency preference can bedynamically tuned by bias-ing the membrane conduc-tance and potential. Corticalinterneuron classes have awide range of preferred fre-quencies (
19
,
 40
,
 43
), and their diverse frequency-tuning properties are impor-tant for setting networdynamics (
3
). For example,the high-frequency dis-charge of a pyramidal cell inits receptive field 
 “
enslaves
its basket cells throughresonance tuning (
42
),which, in turn, suppress theactivity of the surrounding pyramidal neurons.Subthreshold oscilla-tions in single neurons canoccur at a different frequen-cy from that of the network.The augmenting propertiesof resonators-oscillators arealso at work at the network level, and coherentsummation of oscillators is an effective mech-anism for the detection and amplification of weak signals (
44
,
 45
). For example, rhythmiccortical feedback to the thalamus is a major factor in the amplification of thalamocorticaloscillations (
2
,
 
,
 45
). With their increasingcommitment to an oscillatory network, theresponsiveness of neurons to external inputs progressively decreases. As a result, thalamo-cortical spindle oscillations effectively re-duce environmental influences on neocorticalactivity, thereby actively shifting sleep intodeeper stages (
46 
).
Fig. 2.
 Resonance and oscillation are affected by both synaptic and intrinsic mecha-nisms. (
A
) Depression (low-pass filtering) and facilitation (high-pass filtering) of inhibitory postsynaptic potentials by two interneurons (circles) converging onto thesame neocortical pyramidal cell (triangle). (
B
) Band-stop (notch) filtering of excitatorypostsynaptic potentials at gamma frequency (15 to 30 ms, arrowhead) between layer 5 pyramidal cells. (
C
) Band-pass filtering of spike transmission. (Top) Monosynapticdischarge of a CA1 interneuron by an intracellularly driven pyramidal cell. (Bottom)Spike transmission probability is frequency dependent and peaks at
 
10 Hz. (
D
)Voltage-dependent, subthreshold oscillation in layer 2 entorhinal cortical neuron at
8Hz. [Adapted with permission from
 40
 (A),
 43
 (B),
 42
 (C), and
 41
 (D)]
R
 E V I E W
www.sciencemag.org SCIENCE VOL 304 25 JUNE 2004
 1927
 
A further refinement of input selectioncan be achieved by phase biasing. The ongo-ing phase of a centrally organized oscillatorynetwork is independent of the temporal fluc-tuation of sensory signals. The oscillation-related fluctuation of the membrane poten-tials in the participating neurons continuouslyand predictably biases the open-time proba- bility of a multitude of voltage-gated chan-nels (
9
). This design is an energy-efficientsolution for periodically elevating the mem- brane potential close to threshold, providingdiscrete windows of opportunities for theneuron to respond. If the input is not appro- priately timed, however, it is ignored alto-gether or the response is delayed (
10
). For example, proper coordination of afferent ac-tivity with the phase of an intrinsic oscillationcan amplify the somadendritic backpropaga-tion of the action potential (
47 
) and bias themagnitude and direction of spike-timing
 – 
dependent plasticity (
48
). In the hippocam- pus, brief pulse trains delivered at the peak of the theta oscillation result in long-term poten-tiation, whereas the same trains applied out-of-phase weaken the previously strengthened inputs (
49
). Conversely, rhythmic but out-of- phase influences can selectively suppress os-cillations in the target network, as exempli-fied by the suppression of gamma-frequencyrhythm in the hippocampus by the dentategyrus input (
25
). The systems level implica-tion of these oscillation-gating functions isthat perception is not a continuous event butis subject to the cyclic changes of the net-works processing the input (
15
 – 
16 
).
 Binding cell assemblies
. Information inthe brain has been hypothesized to be pro-cessed, transferred, and stored by flexible cellassemblies, defined as distributed networksof neuronal groups that are transiently syn-chronized by dynamic connections (
15
,
 16 
).The mechanisms by which such ephemeralneuronal coalitions are brought about are notknown. One possible mechanism supportingsynchrony is a dynamic change in synapticstrengths across the assemblies, a processthat would require energy-demanding bio-chemical steps. An alternative mechanism isoscillatory synchrony (
13
,
 15
). Transient as-sembly synchronization by oscillation is costeffective. The ability of neuronal assembliesto synchronize depends on the couplingstrength and the distribution of natural fre-quencies (
21
,
 22
). As long as the frequenciesof the coupled oscillators remain similar, syn-chrony can be sustained even with very weak synaptic links (
21
,
 32
). This inherent featureof oscillations allows activated neuronalgroups in distant cortical regions with sparseinterconnections to become temporally linked and then activate unique sets of downstreamassemblies. For example, the various at-tributes of a visual image might be processed separately in distributed neuronal assembliesacross widespread cortical regions and linked  by a common gamma-frequency oscillation.In turn, the phase-locked discharges of thesedistributed groups may be responsible for the
 binding
 of the various features into a co-herent cognitive percept (
13
). Numerous ex- periments support and expand the
 
 binding- by-gamma
 hypothesis (
15
) (but see
 50
,
 51
).The time span required for bringing together transient cell assemblies (
52
) closely fits thegamma cycle, and the induced oscillation islong enough to establish an elementary cog-nitive act (
14
 – 
17 
).
Consolidation and combination of learned information
. Global oscillation is an inherent behavior of balanced systems, and the fre-quency is determined by the time constants of its constituents (
8
,
 53
). Networks built fromnonoscillating pyramidal neurons of similar types inevitably gave rise to a self-sustained oscillation (
54
). The pattern of neuronal ac-tivity depends not only on the precise neuro-nal architecture but also, importantly, on itsinitial conditions (
22
,
 33
,
 53
). Unless theoscillator is perturbed, the sequences of neu-ronal activity will repeat infinitely in a noise-free system (
53
,
 55
). In other words, theconditions that gave rise to a rhythm are
frozen
 into the deterministic nature of theoscillatory dynamics.The
 
default
 state of the unperturbed,sleeping brain is a complex system of numerous self-governed oscillations, par-ticularly in the thalamocortical system(
2
,
 45
,
 46 
). Thecontent of theseoscillations reflectsspike sequence pat-terns created by prior waking expe-rience (
2
,
 7 
,
 11
,
 12
).Synaptic modifica-tions brought about by learning are thusfrozen into the var-ious time windowsof self-organized oscillatory networksof sleep to be turned ultimately into long-term memory bymeans of functionaland structural syn-aptic modifications(
11
,
 12
). This self-sustained replay of learned informationallows for the dissem-ination and combina-tion of temporallydiscontiguous pat-terns of activityacquired during previous waking behaviors. This
 “
off line,
 assembly-
Fig. 3.
 Representation by oscillation phase. (
A
) EEG thetarhythm and place-cell firing (red ticks) on a single run andfalse-color firing field created from multiple runs. Peak firingoccurs on the trough of the theta cycle. (
B
) Spike-triggeredaverage of theta waves and autocorrelogram of spikes, initi-ated by spike occurrence at place-field peak. Groups of spikesoccur at higher than theta frequency, causing each successiveburst to move to an earlier phase of the theta cycle. Note theprogressive forward shift of the preferred phase. (
C
) Place fields for two neurons (a and b) with overlapping place fields andtime cross-correlation between them. Note the theta-frequency modulation of the cross-correlogram. (
D
) Relation betweendistance of place-field peaks and temporal peaks of cross-correlograms [stars in (C)] for a population of neurons. Dot cloudscorrespond to three theta cycles and reflect
 
30-cm distance representation by the cell assembly in each theta cycle. Duringsubsequentcycles,representationshiftsby
6cm,sothatoverlappingportionsoftheenvironmentarescannedrepeatedly.(
E
)Thecurrent position of the rat is identified by the most strongly discharging neuronal assembly (a) at the trough of the theta cycle thatforces the trailing oscillator(s) (e.g., b) to advance its phase (
21
,
 58
). [Adapted with permission from
 58
 (A) and
 60
 (C and D)]
R
 E V I E W
25 JUNE 2004 VOL 304 SCIENCE www.sciencemag.org
1928

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->