Gamma oscillations dynamically couple hippocampalCA3 and CA1 regions during memorytask performanceSean M. Montgomery and Gyorgy Buzsaki*

Center for Molecular and Behavioral Neuroscience, Rutgers University, 197 University Avenue, Newark, NJ 07102

Edited by Fred H. Gage, The Salk Institute for Biological Sciences, San Diego, CA, and approved July 20, 2007 (received for review February 27, 2007)

The hippocampal formation is believed to be critical for theencoding, consolidation, and retrieval of episodic memories. Yet,how these processes are supported by the anatomically diversehippocampal networks is still unknown. To examine this issue, wetested rats in a hippocampus-dependent delayed spatial alterna-tion task on a modified T maze while simultaneously recordinglocal field potentials from dendritic and somatic layers of thedentate gyrus, CA3, and CA1 regions by using high-density, 96-sitesilicon probes. Both the power and coherence of gamma oscilla-tions exhibited layer-specific changes during task performance.Peak increases in the gamma power and coherence were found inthe CA3–CA1 interface on the maze segment approaching the Tjunction, independent of motor aspects of task performance. Theseresults show that hippocampal networks can be dynamically cou-pled by gamma oscillations according to specific behavioral de-mands. Based on these findings, we propose that gamma oscilla-tions may serve as a physiological mechanism by which CA3 outputcan coordinate CA1 activity to support retrieval of hippocampus-dependent memories.

hippocampus � local field potential � retrieval � synchrony

Based primarily on lesion studies, a consensus has emergedthat the hippocampal formation is critical for episodic

memories (1–3). In accord with lesion studies, the firing patternsof hippocampal and entorhinal neurons exhibit prospective andretrospective coding of episodic information (4–7). However,the specific roles of the various hippocampal subregions (8)responsible for encoding, consolidation, and retrieval of memorytraces have been long debated. Computational models (9–12)have postulated the autoassociative recurrent network of theCA3 region as a suitable substrate for storing memories, whichcould be subsequently recalled via replay from CA3 to CA1.Circumscribed lesions in animals (13, 14) and genetic manipu-lations (15) also support the view that the integrity of the CA3region is crucial in memory retrieval, but little is known about thephysiological mechanisms of CA3–CA1 coordination that mightsupport this process.

Although recordings from multiple single neurons can assessthe output representations of a network, the currently availablelarge-scale unit recording methods do not have the ability toeffectively monitor how information transfer is coordinatedacross several neuronal networks (16). On the other hand,although local field potentials (LFPs) lack single-neuron reso-lution, they reflect the temporal synchrony of local afferentactivity and can effectively detect the changing modes of oper-ation in local circuits (17–19). For example, synchronization ofneuronal activity into coherent gamma-frequency oscillationscan serve to bind representations (20, 21) and couple hippocam-pal and rhinal cortices during successful formation of declarativememories (22). To examine how regional networks within thehippocampal formation are dynamically coordinated during theretrieval process, we recorded LFPs from rats with high-densitysilicon probe arrays (23) during performance of a hippocampus-dependent delayed spatial alternation task on a modified T maze

(24). Because gamma oscillations in the hippocampus show clearregional and laminar patterns and the CA3 region is known togenerate self-organized gamma oscillatory patterns (25–28), weexamined the laminar-specific power and coherence of gammaoscillations. Our results show that gamma-frequency synchroni-zation at the CA3–CA1 interface selectively increases on thecenter arm maze segment immediately preceding the T junction,independent of the motor aspects of task performance. Thesefindings signify that gamma oscillations can dynamically coor-dinate hippocampal networks according to behavioral demands.Based on these results, we hypothesize that gamma oscillationsmay provide a physiological mechanism by which CA3–CA1activity is coordinated to support retrieval of hippocampus-dependent memories.

ResultsLFPs were recorded simultaneously in the dentate, CA3, andCA1 regions of the dorsal hippocampus by using a two-dimensional silicon probe array with 96 monitoring sites. Inagreement with previous observations, the in situ recording sitesin the various regions and layers could be determined with highspatial resolution (�30 �m) by using a combination of sponta-neous LFP patterns, multiple unit activity, evoked potentials inresponse to perforant path and commissural stimulation andposthoc histological identification of the anatomical position ofeach recording shank [Fig. 1; see supporting information (SI)Fig. 6A for electrode positions in all animals] (23, 29).

To engage hippocampal networks, rats (n � 4) were trained ona hippocampus-dependent, delayed spatial alternation task (Fig.2A) (24). This task requires rats to encode their spatial responseon each trial and, after a 10-second delay, retrieve this infor-mation to appropriately choose the opposite arm from the oneentered on the previous trial. Although retrieval of previous trialinformation could occur in this task at any time before crossingthe T junction, rats typically ran smoothly through the center armand the T junction in one swift trajectory, suggesting thatretrieval processes likely occur before reaching the T junction.All rats performed the task at high levels of proficiency (�85%correct).

Layer-Specific Gamma Power Increase on the Center Arm of the TMaze. We investigated the involvement of different hippocampalnetworks during performance of the spatial alternation task by

Author contributions: S.M.M. and G.B. designed research; S.M.M. performed research;S.M.M. analyzed data; and S.M.M. and G.B. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Abbreviations: LFP, local field potential; str., stratum; GLM, general linear model; CSD,current source density.

*To whom correspondence should be addressed. E-mail: buzsaki@rutgers.edu.

This article contains supporting information online at www.pnas.org/cgi/content/full/0701826104/DC1.

© 2007 by The National Academy of Sciences of the USA

www.pnas.org�cgi�doi�10.1073�pnas.0701826104 PNAS � September 4, 2007 � vol. 104 � no. 36 � 14495–14500

NEU

ROSC

IEN

CE

measuring the power (amplitude) of gamma oscillations simul-taneously in various layers, across different portions of the task.In agreement with previous findings, maximum LFP gammapower was recorded in the hilar region of the dentate gyrus (25,30). However, the relative power changes in the separate mazeregions varied differentially across layers. As illustrated in Fig.2 B and C, the amplitude of the gamma oscillation (40–120 Hz;SI Fig. 7), recorded from the middle of CA1 stratum (str.)radiatum, showed a selective increase on the center arm that was

not reliably observed in other hippocampal layers. Relativechanges in gamma power were also observed in areas outside theCA1 str. radiatum in other segments of the maze, suggesting thatthe various hippocampal regions and afferents are involved invarious task components (see SI Movie 1), but the presentbehavioral paradigm did not reliably isolate these events.

Because the running pattern of the rat was not identical in thedifferent segments of the maze, and because previous researchhas shown speed and acceleration-dependent changes of hip-pocampal unit and field activity (31, 32), we first addressed theissue of whether overt motor behavior could account for thefluctuations in gamma power. Gamma power measurementswere fit separately for each recording site with a general linearmodel (GLM), including as explanatory variables, runningspeed, acceleration, and maze region (see Methods). Similar toan analysis of covariance, GLM analysis dissects the trial-by-trialvariance in gamma-power measurements that can be accountedfor by the variables of interest (33, 34). GLM analysis showedthat gamma power only weakly correlated with running speedand acceleration, as indicated by the consistently low r2 values(‘‘explained variance,’’ �10%) across all recording sites (Fig.3A). Maze region, on the other hand, could explain �30% of theobserved variance in gamma power at a number of sites.Especially striking was the degree to which the effect of mazeregion respected the laminar anatomy of the hippocampus, withthe highest r2 values confined to the CA1 str. radiatum. � valuesof the GLM analysis, showing the direction and magnitude of thechange in gamma power for each maze region (Fig. 3B), revealedincreased gamma power in CA1 str. radiatum on the center armof the maze. Because the dominant projection to CA1 str.radiatum derives from the CA3 region (8), the increase ofgamma power in this layer suggests an enhancement of CA3output to CA1 on the center arm of the T maze.

To statistically compare task-related changes in the anatom-ical distribution of gamma power across animals, recording siteswere assigned to specific layers (as shown in SI Fig. 6B), and thedistribution of � values from each layer across all four rats wastested for reliable behavior-related changes (two-tailed t tests).Fig. 3C shows group statistics revealing differential increases ingamma power across hippocampal layers on the center armversus other arms of the T maze. Although significant increasesin gamma power were observed in several hippocampal layers onthe center arm of the maze, the gamma power increase in CA1str. radiatum was larger than in all other layers except for theCA3 pyramidal layer (ANOVA, Tukey post hoc test, P � 0.05).Further statistical analysis, including the effect of running speed,acceleration and all maze regions is shown in SI Fig. 8. We alsotested whether gamma patterns were different on ‘‘error’’ versus‘‘correct’’ trials (data not shown) but found no reliable differ-ence, possibly because of the low number of error trials, highbehavioral variability on error trials, and/or the foiled engage-ment of cognitive processes, e.g., retrieval mechanisms thatreturn the wrong information.

Even though the center-arm increase in LFP gamma powershown in Fig. 3C was concentrated in specific layers, the LFP maybe contaminated by volume conduction of activity in otherlayers. To reduce the contribution of possible volume-conductedfields, we first performed a one-dimensional current sourcedensity (CSD) analysis on the LFP data and repeated the GLMcalculations on the derived data (Fig. 3D). This analysis revealedthat the increased gamma oscillations in the CA3 pyramidal layerand CA1 str. radiatum were generated by local currents ratherthan volume conducted from the dentate gyrus or direct ento-rhinal input to the CA1 region (i.e., str. lacunosum-moleculare).CSD gamma power showed significant increases in the dentatemolecular layer, possibly reflecting a change in entorhinal input,local circuitry, or CA3 back-projections to the inner molecularlayer of the dentate gyrus (35). Gamma power also increased in

SourceSink

A

B

C

2.4mV160ms

45mV19ms

Fig. 1. On-line calibration of recording sites. A 96-site silicon probe withrecording pads spaced regularly over a 1.5 mm � 1.5 mm area is shown. Theprobe sites have a vertical spacing of 100 �m and a horizontal spacing of 300�m. The anatomical position of recording electrodes in the different layers inthe dentate gyrus and CA1 and -3 regions of the hippocampus was determinedby using 1,1�-dioctadecyl-3,3,3�,3�-tetramethylindocarbocyanine (DiI) label-ing of electrode tracks (A) matched with known electrophysiological andanatomical characteristics of the hippocampus for each recording session. (Band C) Event-triggered averages of the LFP (traces) and CSD (color) centeredon the estimated position from which the activity was recorded. (B) Ripple-triggered average responses (centered on ripple event, n � 179). Note thelarge sink (blue) associated with a sharp wave in CA1 str. radiatum. (C) Averageevoked activity in response to perforant path stimulation (left aligned tostimulation, n � 3). Note the large sinks in the molecular layer of the dentategyrus.

14496 � www.pnas.org�cgi�doi�10.1073�pnas.0701826104 Montgomery and Buzsaki

the CA1 pyramidal layer, possibly reflecting the entrainment ofCA1 gamma oscillations by CA3 input (26, 36) or str. radiatumreturn currents unmasked from the volume conduction biascaused by anatomical curvature.

Enhanced CA3–CA1 Gamma Coherence on the Center Arm. Becausepower analysis suggested the engagement of specific intrahip-pocampal networks on the center arm of the T maze, we furtherexamined the functional connectivity by analyzing the coherenceof gamma oscillations between hippocampal regions duringalternation task performance. The mean gamma coherence washighest between sites within the same layer (see ‘‘Constant’’term, SI Fig. 8C), revealing that within-layer processing is highlycoordinated even over large physical distances (�1.5 mm). Fig.4A shows the changes in CSD gamma coherence between areference site in the CA3 pyramidal layer and two example sitesin CA1 str. radiatum and the dentate molecular layer averagedwith respect to spatial position on the T maze (see also SI Fig.9). The anatomical distribution of the center arm-relatedchanges in gamma coherence between the CA3 pyramidal layerand all other hippocampal recording sites are shown in Fig. 4B(see also SI Fig. 8 and SI Movie 2). Gamma coherence betweensite pairs was separately fit with GLM statistics for all rats. Thematrices in Fig. 4 C and D show the group statistics for gammacoherence between all hippocampal layer pairs. Extending thegamma power results, the coherence analysis showed increasedcoordination of intrahippocampal circuits, reflected by the max-imal gamma coherence increase between the CA3 and CA1regions on the center arm of the maze.

Task-Specific Enhancement of CA3–CA1 Gamma Synchrony. Althoughthe GLM statistics are specifically designed to dissect individualexplanatory variables (33), we performed additional controlexperiments to isolate the hypothesized mnemonic contributionto the observed changes in gamma oscillations. In addition to thealternation task, each rat was trained and tested on one of threecontrol tasks (SI Fig. 10) possessing no memory requirement andno dependence on the hippocampus to control for motor andnonmnemonic cognitive aspects of task performance. Similar tothe alternation task, in each control task the rat was held in asmall delay area and then released to run for water reward. Theinitial segment of these control tasks showed similar speed andacceleration profiles to the center arm of the alternation task

(SI Fig. 10). Thus, comparison of these initial maze segments ofcontrol tasks to the center arm of the alternation task controlsfor basic route ‘‘planning’’ operations that may accompany thestart of a journey and further controls for the effects of speed andacceleration. As evidence against these alternatives, we foundsignificantly higher gamma power in the CA1 str. radiatum LFPin the alternation task relative to tasks with no mnemonicdemand (Fig. 5A; also see SI Fig. 11). Separate analysis of CSDtraces also revealed increased gamma power in the CA1 pyra-midal layer and confirmed that the str. radiatum gamma oscil-lation was locally generated, indicating that the LFP gammapower effects in CA1 str. lacunosum-moleculare were likely dueto volume-conducted contamination from str. radiatum (Fig.5B). Analysis of gamma coherence revealed enhanced CA3–CA1coordination in the alternation task compared with controls.Smaller increases in coherence between CA1 and other layerswere also observed, but, notably, effects were weak with the CA1str. lacunosum-moleculare and between the CA3 pyramidal layerand dentate layers, indicating that the other effects may havebeen mediated multisynaptically via the dominant changes inCA3–CA1 synchrony. These results provide evidence thatgamma coordination of CA3–CA1 networks is enhanced by thespecific cognitive demands of the delayed alternation task,independent of motor performance.

DiscussionThe major finding of these experiments is that gamma oscilla-tions are differentially modulated by aspects of behavior in thevarious hippocampal networks. Both the power and coherence ofgamma oscillations at the CA3–CA1 interface were selectivelyenhanced by the cognitive demands of the spontaneous alter-nation task.

Gamma oscillations have been suggested to assist variousoperations, including the binding of sensory attributes (20),synaptic plasticity (25), working memory (37, 38), sensory-motorcoordination (21), attention (39), formation of cell assemblies(40), memory coding (22, 41, 42), and even consciousness (43,44). These diverse putative functions indicate that gamma os-cillations per se cannot be explicitly assigned to specific behav-ioral events. Instead, changes in the gamma oscillations recordedfrom a particular brain structure may be taken as an indicatorthat the supporting circuitry is engaged in a particular mode ofcomputation.

0.5 sec

-3

0

3

-3

0

3

Fre

quen

cy (

Hz)

40

80

120

DG

mol

Fre

quen

cy (

Hz)

40

80

120

CA

1 ra

d

(time)/ ~ /

Pow

er (

z-sc

ore)

Pow

er (

z-sc

ore)

-3

0

3

-3

0

3

Pow

er (

z-sc

ore)

Pow

er (

z-sc

ore)

Dela

y

RightReward

Port

LeftRewardPort

Delay (10sec)

Trial n,Trial n,Trial n+1,Trial n+1,

. . .. . .

A B C

0.1 mV

0.2 mV

Fig. 2. Changes in gamma power during performance of a delayed spatial alternation task. Rats were trained to perform a hippocampus-dependent, delayedspatial alternation task on a modified T maze (shown in A). (B) Single-trial example of the task-related fluctuations in LFP gamma oscillations from recordingsites positioned in CA1 str. radiatum (CA1 rad) and the dentate molecular layer (DG mol). The filtered (40–120 Hz) gamma trace (gray) is overlaid on thenormalized spectrogram (color). Note the burst of gamma power in CA1 str. radiatum but not the dentate molecular layer on the center arm of the T maze. (C)Normalized LFP gamma power averaged over one recording session (20 trials) from the same layers with respect to spatial position on the maze.

Montgomery and Buzsaki PNAS � September 4, 2007 � vol. 104 � no. 36 � 14497

NEU

ROSC

IEN

CE

CA3 and CA1 Networks Are Dynamically Coordinated by GammaOscillations. Gamma oscillations have diverse expression through-out the hippocampal networks. They are generated by a consortiumof mechanisms and depend primarily on the activity of fast spikingbasket neurons and GABAA receptor-mediated inhibition and�-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA)receptor-mediated excitation (19, 45–50). Gamma oscillations inthe CA1 str. lacunosum-moleculare and the dentate gyrus aremainly under the control of entorhinal input (25, 51). However,evidence from in vitro and in vivo studies suggests that the CA3region can generate gamma oscillations from the interactionsbetween CA3 pyramidal cells and basket interneurons (25–27). Thegamma phased-locked firing of CA3 projection neurons can in turnentrain the CA1 network (26, 36). We found dynamic changes in thepower and coherence of CA3 and CA1 gamma oscillations. Someanalyses also implicated specific layers in the dentate gyrus, al-though future experiments with dual entorhinal-hippocampal re-cordings will be needed to precisely determine to the extent towhich these effects were related to direct entorhinal involvement,local activity, or CA3-dentate projections. Because our recordingsprincipally sampled the CA3b and -c subregions, we cannot spec-ulate about the role of the CA3a subregion from these data, giventhe differential wiring patterns of these subregions (35). Neverthe-less, our results provide further support for a role of the CA3 regionin generating self-organized gamma activity, independent of the

gamma oscillations in other hippocampal regions. Additionally, ourresults provide evidence that the temporal coordination of CA3 andCA1 networks is dynamically regulated via gamma synchronization,according to behavioral demands.

Mnemonic Function of CA3–CA1 Gamma Coupling. Numerous lesionstudies in humans and other animals link hippocampal networksto memory (2, 3). Previous computational models (9–11, 52),lesion studies (13, 14), and genetic knockout studies (15)have implicated CA3–CA1 communication in retrieval ofhippocampus-dependent memories. We found a consistent pat-tern of changes in the power and coherence of gamma oscilla-tions, showing increased synchrony in the CA3–CA1 interface onthe center arm of the T maze. Using statistics and analysis ofcontrol tasks to rule out the contribution of overt behavior andother nonmnemonic variables, we found that the increasedCA3–CA1 gamma coordination was specifically enhanced on thecenter arm of the T maze by cognitive aspects of alternation taskperformance. Performance of the delayed spatial alternationtask requires, on each trial, that rats retrieve information aboutthe previous journey to choose the opposite arm upon reachingthe T junction. Although the exact spatial position(s) at whichretrieval processes occur likely varies from trial to trial (as doesthe CA3–CA1 gamma synchrony increase seen in SI Movies 1and 2), the fact that rats typically run through the center arm andthe T junction in one swift trajectory suggests that retrieval and

-2 0 2

Spe

edA

ccel

Maz

e R

egio

n

R-Square Beta

0 0.5 1Center Arm

∆ Gamma Power (dB)Center Arm

∆ Gamma Power (dB)

A

D

B

pyr hil

gran mol

lm radpyr or

DG

CA1

CA30 0.5 1

C LFP CSD

0 0.1 0.2 0.3

Fig. 3. Gamma power on the center arm is maximally expressed in theCA3–CA1 associational system. Fit of gamma power with a GLM was used toassess contribution of various overt and covert variables. (A) r2 of runningspeed and acceleration (Accel) and maze region. Note that the maze regionexplains substantial variance for CA1 str. radiatum recording sites, whereasspeed and acceleration explains very little. (B) � values showing changes in LFPgamma power (in decibels) across the different maze regions for a singlerecording session. (C and D) Group statistics of gamma power changes on thecenter arm versus other arms across different anatomical layers (decibels, �standard error; P � 0.001, Bonferroni corrected t tests) for LFP (C) and CSD (D).Also see SI Fig. 8.

-0.04

0

0.04

-10

-8

-6

-4

-2

0

pyr hil

gran mol

lm radpyr or

DG

CA1

CA3

pyr hil

gran mol

lm radpyr or

DG

CA1

CA3

pyr hil

gran m

ol lm

radpyr or

DG

CA

1

CA

3

∆ G

amm

a C

oher

ence

∆ G

amm

a C

oher

ence

p-va

lue

(log 1

0)

A

D

C

B

-0.05

0

0.05Center Arm

Fig. 4. CA3–CA1 gamma coherence is enhanced on the center arm. (A)Spatial distribution of gamma coherence between the CA3 pyramidal layerand example sites in CA1 str. radiatum (Upper; shown in B) and the dentatemolecular layer (Lower). (B) Center-arm change in gamma coherence betweena CA3 pyramidal reference site (circle) and all other hippocampal recordingsites averaged over an entire recording session (20 trials). (C and D) Groupdata. (C) Change in gamma coherence between all hippocampal layer pairs onthe center arm of the maze. Note that the peak increases in coherence are atthe CA3–CA1 interface. (D) Bonferroni-corrected P values. Note that the mostsignificant values are between the CA3 and CA1 layers. Also see SI Fig. 8.

14498 � www.pnas.org�cgi�doi�10.1073�pnas.0701826104 Montgomery and Buzsaki

decision-making usually occur on the maze before the T junc-tion. This observation is corroborated by previous studies findingthat turn-selective neuronal firing emerges in the CA1 networkduring traversal of the center arm in similar tasks (refs. 4–7; alsosee refs. 24 and 53). Based on our results and convergentevidence pointing to the mnemonic role of the hippocampus andits subregions, we hypothesize that the increased CA3–CA1gamma coordination on the center arm of the T maze may reflectretrieval of previous trial information. Overall, our findingssuggest that gamma oscillations may provide a mechanism bywhich hippocampal networks can dynamically coordinate toperform specific mnemonic operations.

Our results also posit specific predictions for future experi-ments investigating the changes in CA3 and CA1 unit firing thataccompany retrieval processes. The observed increase in CA1str. radiatum gamma power on the center arm of the T maze issuggestive of accompanying changes in bursting and/or gammasynchrony of CA3 units. Increased bursting leads to a nonlinearincrease in neurotransmitter release (54), thereby increasingpostsynaptic currents and the amplitude of LFP oscillations.Similarly, increased synchrony of unit firing increases the am-plitude of LFP oscillations through greater temporal summationof postsynaptic currents (55, 56). Local modulation of specificCA1 layers could also facilitate CA3 inputs, but this must occurat fast (ostensibly ionotropic) receptors to mediate the observedgamma fluctuations. The observed increase in gamma coherencebetween CA3 and CA1 suggests that the synchrony between CA3and CA1 unit firing increases at short (monosynaptic) andgamma time scales during the retrieval process. Because hip-pocampal principal neurons fire in restricted spatial locations,

and because the �-amino-3-hydroxy-5-methyl-4-isoxazolepropi-onic acid (AMPA) receptor-mediated effects between pyrami-dal cells are weak (57), assessment of the CA3–CA1 connectionsby monitoring unit activity would require recording very largenumbers of neurons simultaneously. In contrast, by using LFPrecording, the gamma power and coherence provide informationabout the collective behavior of neurons and the mode ofoperation of a given network.

In summary, our findings show that gamma oscillations dy-namically coordinate the activity of hippocampal networks dur-ing performance of a hippocampus-dependent memory task.The most robust change in network activity was increasedgamma coordination between CA3 and CA1 regions during theportion of the task associated with the retrieval of prior expe-rience. Although the specific content of the retrieved memorymay not be accessible without large-scale single-cell recordings,our results show that monitoring gamma oscillations with suffi-cient spatial resolution and proper behavioral control is aneffective tool for identifying specific circuit dynamics involved incognitive operations.

MethodsAnimals and Behavior. Four Long–Evans rats (male, 300–400 g)were water-deprived and trained to run in a hippocampus-dependent continuous delayed nonmatch to place (‘‘alterna-tion’’) task (24) and a control task. All rats performed thealternation task well (�85% correct). The control tasks includeda C-shaped (n � 2 rats) and a Z-shaped (n � 1 rat) linear track,requiring the rats to run back and forth for water reward, and acue task (n � 1 rat), requiring the rat to run in a path similar tothe alternation task; but, on each trial, the rat’s trajectory wasrandomly cued left or right by a large block that was visible afterthe delay period. To constrain the behavioral and cognitivevariability, trials in which the rat engaged in rearing, excessivesniffing, grooming, or immobility were eliminated from furtheranalysis.

Spectral and GLM Analysis. Gamma-band (40–120 Hz) poweranalysis was performed by using the filter–rectify–smoothmethod, and spectral power and coherence analysis was per-formed by using Morlet wavelet analysis (courtesy of AslakGrinsted, Arctic Centre, University of Lapland, Rovaniemi,Finland) (see SI Detailed Methods). Fitting data from correcttrials in the alternation task (errors were rare and often asso-ciated with exploratory behavior), the distribution of spectralestimates (either power or coherence) for each electrode site wasfit with a linear model (ANOVAN, Type 3 SS; MatLab; TheMathworks, Natick, MA): G � �constant � �running speed ��running acceleration � �maze region (4 arms) � �.

To perform group statistics across animals, resulting � valuesfrom one electrode site per shank per animal were sampled fromeach layer (SI Fig. 6B). The resulting sampled distribution of �values from each anatomical layer was subsequently tested for astatistical difference from zero (t tests, P � 0.001, Bonferronicorrected). The normality assumption of the t test and assump-tions of the general linear model, including normality of resid-uals, absence of interaction effects, and uncorrelated residuals,were assessed. Similar analyses were performed to comparephysiological parameters on the center arm of the alternationtask to those in the corresponding region of a control task (SIFig. 10).

We thank Martha Betancur for assistance with behavioral training;Anton Sirota, Ken Harris, Asohan Amarasingham, Kamran Diba, JoshJacobs, and Michael Kahana for advice and support; and Aslak Grinstedfor providing wavelet coherence algorithms. This work was supported bythe National Institutes of Health Grants NS34994, NS43157, andMH54671.

pyr hil

gran m

ol lm

radpyr or

DG

CA

1

CA

3

pyr hil

gran m

ol lm

radpyr or

DG

CA

1

CA

3

pyr hil

gran mol

lm radpyr or

DG

CA1

CA3

C D

0 0.5 1

CSDB

-0.04 0 0.04∆ Gamma Coherence

-10 -8 -6 -4 -2 0p-value (log10)

0 0.5 1

pyr hil

gran mol

lm radpyr or

DG

CA1

CA3

LFPA∆ Gamma Power (dB) ∆ Gamma Power (dB)

Fig. 5. Enhanced CA3–CA1 gamma coordination on the center arm cannotbe explained by motor performance. Gamma power and coherence on thecenter arm of the alternation task was compared with the correspondingregion of a control task (see SI Fig. 10). (A and B) Group data show increasesin gamma power (decibels, � standard error) of the LFP (A) and CSD (B) in thealternation versus control tasks (P � 0.001, Bonferroni-corrected t tests). (Cand D) Group data for gamma coherence difference between the alternationtask and control tasks (C) and corresponding Bonferroni-corrected P values (D)from t test comparisons. Also see SI Fig. 11.

Montgomery and Buzsaki PNAS � September 4, 2007 � vol. 104 � no. 36 � 14499

NEU

ROSC

IEN

CE

1. Squire LR (1992) Psychol Rev 99:195–231.2. Milner B, Squire LR, Kandel ER (1998) Neuron 20:445–468.3. Eichenbaum H, Cohen NJ (2001) From Conditioning to Conscious Recollection:

Memory Systems of the Brain (Oxford Univ Press, Oxford).4. Wood ER, Dudchenko PA, Robitsek RJ, Eichenbaum H (2000) Neuron

27:623–633.5. Frank LM, Brown EN, Wilson M (2000) Neuron 27:169–178.6. Ferbinteanu J, Shapiro ML (2003) Neuron 40:1227–1239.7. Shapiro ML, Ferbinteanu J (2006) Proc Natl Acad Sci USA 103:4287–4292.8. Amaral DG, Witter MP (1995) in The Rat Nervous System, ed Paxinos G

(Academic, New York), pp 443–490.9. Marr D (1971) Philos Trans R Soc London B 262:23–81.

10. Treves A (1995) J Comput Neurosci 2:259–272.11. Rolls ET (1996) Hippocampus 6:601–620.12. Wallenstein GV, Eichenbaum H, Hasselmo ME (1998) Trends Neurosci

21:317–323.13. Brun VH, Otnass MK, Molden S, Steffenach HA, Witter MP, Moser MB,

Moser EI (2002) Science 296:2243–2246.14. Steffenach HA, Sloviter RS, Moser EI, Moser MB (2002) Proc Natl Acad Sci

USA 99:3194–3198.15. Nakazawa K, Quirk MC, Chitwood RA, Watanabe M, Yeckel MF, Sun LD,

Kato A, Carr CA, Johnston D, Wilson MA, et al. (2002) Science 297:211–218.16. Buzsaki G, Draguhn A (2004) Science 304:1926–1929.17. Bressler SL, Kelso JA (2001) Trends Cognit Sci 5:26–36.18. Varela F, Lachaux JP, Rodriguez E, Martinerie J (2001) Nat Rev Neurosci

2:229–239.19. Bartos M, Vida I, Jonas P (2007) Nat Rev Neurosci 8:45–56.20. Gray CM, Konig P, Engel AK, Singer W (1989) Nature 338:334–337.21. Engel AK, Fries P, Singer W (2001) Nat Rev Neurosci 2:704–716.22. Fell J, Klaver P, Lehnertz K, Grunwald T, Schaller C, Elger CE, Fernandez G

(2001) Nat Neurosci 4:1259–1264.23. Csicsvari J, Henze DA, Jamieson B, Harris KD, Sirota A, Bartho P, Wise KD,

Buzsaki G (2003) J Neurophysiol 90:1314–1323.24. Ainge JA, van der Meer MA, Langston RF, Wood ER (June 6, 2007)

Hippocampus, 10.1002/hipo.20301.25. Bragin A, Jando G, Nadasdy Z, Hetke J, Wise K, Buzsaki G (1995) J Neurosci

15:47–60.26. Csicsvari J, Jamieson B, Wise KD, Buzsaki G (2003) Neuron 37:311–322.27. Mann EO, Paulsen O (2005) Prog Biophys Mol Biol 87:67–76.28. Isomura Y, Sirota A, Ozen S, Montgomery S, Mizuseki K, Henze DA, Buzsaki

G (2006) Neuron 52:871–882.29. Bragin A, Jando G, Nadasdy Z, van Landeghem M, Buzsaki G (1995)

J Neurophysiol 73:1691–1705.

30. Stumpf C (1965) Electroencephalogr Clin Neurophysiol 18:477–486.31. Whishaw IQ, Vanderwolf CH (1973) Behav Biol 8:461–484.32. McNaughton BL, Barnes CA, O’Keefe J (1983) Exp Brain Res 52:41–49.33. Rencher AC (2002) Methods of Multivariate Analysis (Wiley, New York).34. Jacobs J, Hwang G, Curran T, Kahana MJ (2006) NeuroImage 32:978–987.35. Li XG, Somogyi P, Ylinen A, Buzsaki G (1994) J Comp Neurol 339:181–208.36. Fisahn A, Pike FG, Buhl EH, Paulsen O (1998) Nature 394:186–189.37. Lisman JE, Idiart MA (1995) Science 267:1512–1515.38. Howard MW, Rizzuto DS, Caplan JB, Madsen JR, Lisman J, Aschenbrenner-

Scheibe R, Schulze-Bonhage A, Kahana MJ (2003) Cereb Cortex 13:1369–1374.

39. Fell J, Fernandez G, Klaver P, Elger CE, Fries P (2003) Brain Res Brain ResRev 42:265–272.

40. Harris KD, Csicsvari J, Hirase H, Dragoi G, Buzsaki G (2003) Nature424:552–556.

41. Sederberg PB, Kahana MJ, Howard MW, Donner EJ, Madsen JR (2003)J Neurosci 23:10809–10814.

42. Sederberg PB, Schulze-Bonhage A, Madsen JR, Bromfield EB, McCarthy DC,Brandt A, Tully MS, Kahana MJ (2007) Cereb Cortex 17:1190–1196.

43. Llinas R, Ribary U, Contreras D, Pedroarena C (1998) Philos Trans R SocLondon B 353:1841–1849.

44. Sauve K (1999) Conscious Cognit 8:213–224.45. Buzsaki G, Leung LW, Vanderwolf CH (1983) Brain Res 287:139–171.46. Whittington MA, Traub RD, Jefferys JG (1995) Nature 373:612–615.47. Whittington MA, Traub RD, Kopell N, Ermentrout B, Buhl EH (2000) Int

J Psychophysiol 38:315–336.48. Wang XJ, Buzsaki G (1996) J Neurosci 16:6402–6413.49. Traub RD, Whittington MA, Colling SB, Buzsaki G, Jefferys JG (1996)

J Physiol (London) 493:471–484.50. Traub RD, Bibbig A, LeBeau FE, Buhl EH, Whittington MA (2004) Annu Rev

Neurosci 27:247–278.51. Charpak S, Pare D, Llinas R (1995) Eur J Neurosci 7:1548–1557.52. Hasselmo ME, Bodelon C, Wyble BP (2002) Neural Comput 14:793–817.53. Bower MR, Euston DR, McNaughton BL (2005) J Neurosci 25:1313–1323.54. Lisman JE (1997) Trends Neurosci 20:38–43.55. Buzsaki G, Traub RD, Pedley TA (2003) in Current Practice of Clinical

Encephalography, eds Ebersole JS, Pedley TA (Lippincott Williams & Wilkins,Philadelphia), pp 1–11.

56. Robbe D, Montgomery SM, Thome A, Rueda-Orozco PE, McNaughton BL,Buzsaki G (2006) Nat Neurosci 9:1526–1533.

57. Sayer RJ, Friedlander MJ, Redman SJ (1990) J Neurosci 10:826–836.

14500 � www.pnas.org�cgi�doi�10.1073�pnas.0701826104 Montgomery and Buzsaki