Ketamine alters oscillatory coupling in the hippocampus

Recent studies show that higher order oscillatory interactions such as cross-frequency coupling are important for brain functions that are impaired in schizophrenia, including perception, attention and memory. Here we investigated the dynamics of oscillatory coupling in the hippocampus of awake rats upon NMDA receptor blockade by ketamine, a pharmacological model of schizophrenia. Ketamine (25, 50 and 75 mg/kg i.p.) increased gamma and high-frequency oscillations (HFO) in all depths of the CA1-dentate axis, while theta power changes depended on anatomical location and were independent of a transient increase of delta oscillations. Phase coherence of gamma and HFO increased across hippocampal layers. Phase-amplitude coupling between theta and fast oscillations was markedly altered in a dose-dependent manner: ketamine increased hippocampal theta-HFO coupling at all doses, while theta-gamma coupling increased at the lowest dose and was disrupted at the highest dose. Our results demonstrate that ketamine alters network interactions that underlie cognitively relevant theta-gamma coupling.

were associated with greater latency to peak locomotor activity (16,36 and 76 minutes, respectively; Fig. 1d), mainly due to transitory ataxia.
In Fig. 2a we show the gamma band power averaged across all electrodes in a representative animal treated with ketamine. Peak gamma power occurred during the first hour post-ketamine injection, and approached baseline levels three hours afterwards. In Fig. 2b we show the time-course of gamma power variations in each of the 8 electrodes in the same animal ( Fig. 2b inset); notice that gamma power increases along the CA1-dentate gyrus axis, as previously reported 27 . The increase in gamma power induced by ketamine was apparent in all recording sites; in fact, baseline normalised gamma power provided similar time-courses for all electrodes in the bundle (data not shown). At the group level, in contrast to the time-course of locomotor activity (Fig. 1a), gamma power peaked within 5-10 minutes after ketamine injection at all doses studied (Fig. 2c). These results show that the time-course of locomotor and electrophysiological alterations caused by NMDAR blockade may be dissociated (see cross-correlations in Fig. 2c insets and ref. 14).
We found that ketamine also increased hippocampal HFO power, with a similar time-course to the increase in gamma oscillations (see Fig. 3a for a representative electrode and Fig. 3b for group results). Incidentally, it has recently been shown that ketamine increases HFO activity in motor cortex, nucleus accumbens, and other basal ganglia nuclei 28,29 , suggesting that abnormally high levels of HFO may be a widespread effect of NMDAR blockade.
NMDAR blockade modulates hippocampal theta oscillations in a layer-dependent manner. Along with high-frequency alterations, we found that low-frequency LFP signals were also modulated by NMDAR blockade. In Fig. 4a we show the power spectral density in the theta range of three electrodes recorded simultaneously from an animal during pre-and post-injection of 50 mg/kg ketamine IP. Notice in this example that while theta band power decreased after ketamine injection in stratum pyramidale, it did not change in stratum radiatum, and was markedly increased at the hippocampal fissure. Notice further that theta peak frequency was shifted in all recording sites from 6-8 Hz during drug-free locomotion periods to 7-10 Hz following ketamine injection, probably due to increases in locomotion speed. Fig. 4b shows group results for theta band power modulation during peak locomotion induced by NMDAR blockade in a subset of electrodes located in stratum oriens-alveus and pyramidale (Electrodes #1-3) and in another subset located in stratum lacunosum-moleculare, hippocampal fissure and dentate gyrus (Electrodes #6-8). The mean power spectral density is displayed in Supplementary Fig. S1 online. Ketamine differentially affected theta band power in the two subsets of electrodes at all doses studied, and this effect occurred specifically during hyperlocomotion (Fig. 4c).
A recent study showed that ketamine IP at the dose of 50 mg/kgbut not 20 mg/kg -increases the power of hippocampal delta (1)(2)(3)(4) oscillations during a period of ,15 minutes following the injection 17 (see also ref. 19). Since a greater level of delta power can potentially lead to a greater area under the curve of the power spectral density in the theta range (spectral leakage 30 ) even in the absence of a theta peak, we next investigated whether putative changes in delta power could account for the changes in theta power reported above. As shown in Fig. 4d, delta power was highly modulated by locomotion speed; in particular, delta power was high during periods of low locomotion preceding saline and ketamine injections, and also after the hyperlocomotion episode had ceased (Fig. 4d). Thus, the apparent high levels of theta band power in these periods (Fig. 4c) are actually due to spectral leakage from delta power and do not correspond to a genuine theta activity. Consistent with a previous report 17 , the doses of 50 and 75 mg/kg transiently increased delta power, which returned to basal levels before the peak of ketamineinduced hyperlocomotion activity (Fig. 4d), i.e., when the layerdependent variations in theta power were most striking (Fig. 4c). Moreover, contrary to theta, delta power time-course was qualitatively similar in all electrodes (Fig. 4d). Therefore, ketamine-induced alterations in delta power cannot account for the dichotomy in theta modulation across the CA1-dentate gyrus axis observed during hyperlocomotion.
NMDAR blockade leads to increased phase synchrony in multiple high-frequency bands. We next investigated the levels of phase coherence across recording sites, and found that ketamine induced transient hypersynchrony in a wide range of fast LFP oscillations from 30 to 200 Hz (Fig. 5a). Phase coherence spectra were typically multimodal, exhibiting peak values in the traditional gamma range, as well as in frequencies above 100 Hz. The changes in coherence induced by ketamine were observed among electrode pairs located in multiple hippocampal depths (Fig. 5a). Interestingly, electrode pairs located at stratum lacunosum-moleculare and dentate gyrus presented coherence peaks in a faster gamma frequency than electrode pairs located at stratum oriens-alveus and pyramidale. Peaks in HFO phase coherence were particularly prominent for electrodes pairs across hippocampal layers.
In Fig. 5b we show time-courses of mean phase coherence changes in the gamma and HFO bands across all electrode pairs. Gamma and HFO phase coherence increased immediately after ketamine injection and only returned to baseline values after the hyperlocomotion episode ended. Interestingly, the relative increase in phase coherence induced by ketamine was much more prominent for HFO than gamma oscillations. In all, these results show that ketamine alters inter-site synchrony of multiple frequency bands in the hippocampus. NMDAR blockade alters cross-frequency coupling. We next examined the effects of ketamine on the coupling between low-and 75 mg/kg -n 5 5) at 60 and 120 min after the beginning of the recording session, respectively, and were monitored for additional 180 min. In this and all other figures, grey and black arrows denote saline and ketamine injections, respectively. Locomotion speed was monitored at 30 frames per second and averaged over 5-min blocks among all animals from each group. Notice that ketamine induces an increase in locomotion speed at all doses. (b-d) Total locomotion (b), peak speed (c), and latency to peak locomotion speed (d) after ketamine injection are shown for each dose. Data are shown as mean 6 SEM over animals. high-frequency LFP oscillations. Typically, low frequency phase modulates the amplitude of higher frequency oscillations 22 . This type of oscillaory interaction is deemed to be involved in cognitive processing (see ref. 21 for a review). Consistent with previous reports 23,24,31 , we found prominent CFC in most CA1 electrodes; theta phase strongly modulated the amplitude of HFO in electrodes located above the pyramidal layer (i.e., stratum oriens-alveus), while the amplitude modulation of high-gamma (HG; 60-100 Hz) was maximal in electrodes located in stratum lacunosum-moleculare and hippocampal fissure 23 . Finally, in spite of the low-gamma (30-60 Hz) power increase depicted in Fig. 2a, we did not find prominent coupling between theta and low-gamma in CA1, as reported previously 23 .
To illustrate the effect of acute NMDAR blockade on CFC, for each example in Fig. 6 we show six comodulation maps computed for 5min time blocks before and after ketamine injection (as indicated by These results are representative for recording sites with theta-HG (Fig. 6a, c) and theta-HFO coupling (Fig. 6b, d) for the lowest (Fig. 6a, b) and highest (Fig. 6c, d) ketamine dose. Surprisingly, we found that ketamine had a differential effect on theta-HG coupling depending on dose: while the lowest dose increased theta-HG coupling (Fig. 6a), the highest dose disrupted this oscillatory interaction ( Since CFC strength typically varies with theta power 23,24,31 , we next investigated whether the results above could be related to ketamine effect on theta oscillations. To that end, we plotted mean CFC strength as a function of the theta/delta power ratio (Fig. 7a, b). We note that due to the spectral leakage of delta power into the theta rage that occurs during periods of immobility (c.f. section above), the theta/delta ratio is a better measure of genuine theta activity in the LFP than the mean power in the theta range. Moreover, the theta/ delta ratio is a spectral measure highly correlated with locomotion speed ( Supplementary Fig. S2 online), and thus also serves to investigate whether changes in CFC strength are explained by changes in locomotion. We found that ketamine altered CFC in a similar way as described above even after controlling for this confounding factor (see Fig. 7c for multiple regression analyses). These results therefore show that acute NMDAR blockade alters CFC in a frequency-specific and dose-dependent way.
NMDAR blockade does not alter the distribution of electrical dipoles in the hippocampus. Finally, we performed current source density (CSD) analysis in one additional animal. Baseline CSD plots (Fig. 8, top row) for the different frequency ranges were similar to those previously described 27,30 . We found that ketamine did not alter the spatial distribution of sinks and sources pairs (Fig. 8, bottom row). These results indicate that NMDAR blockade alters preexisting hippocampal oscillations but does not generate new dipoles. Further, these analyses indicate that the oscillations investigated in this work are generated in the hippocampus, and not volume conducted from other brain regions.

Discussion
In this study we showed that acute sub-anaesthetic doses of ketamine alter the cross-frequency interaction between theta phase and the amplitude of two higher frequency rhythms in the hippocampus: high-gamma and HFO. In addition, we also found that ketamine increases gamma and HFO power, alters oscillatory phase synchrony, and differentially modulates theta power in a layer-specific manner.
Consistent with previous studies 12,14,15 , we found altered behaviour and increased hippocampal gamma power during acute blockade of NMDAR. While low doses of ketamine cause correlated increases in locomotor activity and total gamma power, higher doses can induce different time-courses of behavioural and electrophysiological alterations. In fact, ketamine also increases gamma oscillations in sedated and anesthetised animals 14 . These observations indicate that alterations in gamma power and hyperlocomotion are two independent effects of NMDAR blockade. While hyperlocomotion induced by acute NMDAR blockade is currently considered a predictive model of positive schizophrenic symptoms 16 , the dissociation between gamma activity and hyperlocomotion suggests that altered gamma oscillations may have additional translational significance 32 .
Theta oscillations are believed to serve as a temporal organizer for a variety of functions, such as sensorimotor integration 33 and coordination of cell assemblies by means of phase modulating gamma oscillations 2 . Consistent with recent findings 15 , here we found that acute NMDAR blockade differently alters theta power depending on hippocampal layer. These findings show that different theta dipoles  have different sensitivities to NMDAR blockade. Entorhinal cortex inputs give rise to the theta dipole in stratum lacunosum-moleculare 34 . A greater theta activity in this layer following NMDAR blockade may thus be associated with an overflow of sensory information from the entorhinal cortex to the hippocampus. Neuronal synchrony has been proposed to play a role in dynamically selecting and routing information within and across brain structures 1 . This hypothesis gave rise to the idea that abnormal synchrony would underlie symptoms of cognitive disorders such as autism and schizophrenia 4 . However, whether schizophrenia is associated with increased or decreased neuronal synchrony remains an open question 5 . Previous studies found reduced inter-trial phase coherence (ITC) in schizophrenic patients 6,35 , typically accompanying a decrease in stimulus-evoked gamma power 35 . It should be noted that ITC measures the level of phase resetting following a sensory stimulus within a recording site, and not the level of phase locking between LFP oscillations recorded from different sites, as studied here. Regarding the latter, positive schizophrenia symptoms may be associated with increased connectivity 7,36 . Recent evidence suggests that although schizophrenic patients have reduced evoked gamma power, they could have abnormally high levels of basal gamma power 36,37 . If confirmed, these findings would solve current inconsistencies between animal models (which show increased levels of gamma power and synchrony) and human studies (which point to reduced evoked gamma power and synchrony; for discussion, see ref. 37). Altogether, our and other results suggest that psychotic symptoms caused by NMDAR hypofunction are associated with an overprocessing of information through functionally hyper-connected structures. Therefore, like in other brain disorders such as Parkinson disease and epilepsy 4 , pathological hypersynchrony could also play a role in schizophrenia.
Theta-gamma coupling has been hypothesised to form a neural coding system that allows the representation of multiple items in a sequential order 38 . Abnormalities in theta-gamma coupling have been thus suggested as a possible electrophysiological substrate of disordered thoughts and impaired working memory 8,38 . Also, it should be noted that recent CFC studies have demonstrated that theta modulates multiple higher frequency bands, which occur within (30-100 Hz) and beyond (.100 Hz) the traditional gamma band [22][23][24]39 . For instance, theta preferentially modulates high-gamma (60-100 Hz) in CA1 and low-gamma (30-60 Hz) in CA3 22,24,40 . Additionally, theta phase also modulates higher frequency activity circumscribed into the 110-160 Hz band in stratum oriens-alveus 23,39 , which we refer to as HFO. Interestingly, ketamine leads to a significant increase in HFO activity in the nucleus accumbens 28 , a limbic region implicated in schizophrenia that receives massive connections from the hippocampus 41 . This suggests that higher frequency oscillations above the gamma range may also be altered in schizophrenia.  Here we showed that ketamine increases theta-HFO coupling during peak locomotion at all doses studied, while its effect on theta-gamma coupling was dose dependent. Importantly, none of these effects can be explained by changes in theta power occurring during hyperlocomotion. The cognitive implications of increased theta-HFO coupling remain to be better understood, as well as the functional role of HFO per se 23,28,39 . A recent study has shown that physiological theta-HFO coupling significantly increases during REM sleep 31 . REM sleep is a brain state associated with incongruous thoughts and dreams. Many similarities have been pointed out between REM sleep and psychosis 42 , leading some to suggest that psychotic symptoms would be associated with intrusion of a dreaming state into an awake mind 43,44 . While these suggestions remain to be appropriately tested, the finding of enhanced theta-HFO coupling during REM sleep 31 and following NMDAR blockade (present results) supports such a view.
Theta-gamma coupling, on the other hand, increased with the lowest dose of ketamine but was disrupted with the highest dose. Current theories on the combined function of theta and gamma oscillations suggest that disrupting their coupling would lead to deficits in brain functions such as working memory 38 . However, the functional implications of increased theta-gamma coupling upon lower levels of NMDAR blockade are harder to interpret. It may be that increased oscillatory power, synchrony and cross-frequency coupling are all correlates of an aberrant state of brain hyperexcitability and altered information flow, which could underlie dysfunctions such as hallucinations and flight of ideas.
Recent findings suggest that dysfunction of GABAergic interneurons are likely to underlie the electrophysiological alterations reported here 45 . Accordingly, ablation of NMDAR in parvalbumin (PV) positive interneurons in mice leads to enhancements of basal gamma activity 46,47 . Moreover, ketamine does not induce hyperlocomotion in these knockout mice 47 , suggesting a critical involvement of the blockade of NMDAR in PV interneurons for the manifestation of positive schizophrenic symptoms. NMDAR ablation in corticolimbic interneurons has also been associated with the negative symptoms of the disease 48 . These findings help link together the NMDA hypofunction Theta/delta     hypothesis of schizophrenia with the alterations of GABAergic interneurons seen in post-mortem studies of schizophrenic subjects 49 . In addition to PV1 interneurons, hypofunction of NMDAR in oriens lacunosum-moleculare (OLM) interneurons could also be involved in the pathophysiology of schizophrenia 50,51 . These cells synapse on distal portions of the apical dendrites of pyramidal cells, where projections from the entorhinal cortex arrive 50 . A hypofunction of OLM cells would thus favour entorhinal cortex inputs 52 , which could then lead to increased theta oscillations in stratum lacunosum-moleculare, as observed here. GABAergic interneurons are also likely to underlie coupling between theta and gamma rhythms 50,53 , and would thus further mediate aberrant CFC patterns following NMDAR blockade. In all, a preferential action of NMDA antagonists on inhibitory cells is compatible with increased levels of excitation 54,55 and altered neuronal oscillations. Building a bridge between electrophysiological findings in animal models and schizophrenic patients has proven to be a challenge 5 . A large part of the alterations described here, particularly at the HFO band, would not have been noticed by scalp EEG because of its frequency band limitations. Data obtained by invasive techniques, such as electrocorticograms, are unfortunately scarce in schizophrenia. In addition, antipsychotic drugs by themselves cause oscillatory changes 32 , and are therefore important confounding factors in clinical studies. Thus, while at variance with some previous human studies 6,35 , our results add to others 7,36,37 in the suggestion that some symptoms of the schizophrenia syndrome are mediated by an aberrant state of brain hyperactivity, including increases in the activity of fast oscillations, phase synchrony and cross-frequency coupling. Experimental procedures. After recovering for 7-10 days, animals were individually habituated to the recording room for 3 days. Experiments consisted of video and electrophysiological recordings of freely moving rats in a rectangular arena (50 3 4 3 40 cm) placed in a dimly lighted room. Recordings consisted of 3 stages: animals were first allowed to explore the arena for one hour (basal); then were injected with saline and recorded for another hour (saline); finally, animals received a single ketamine injection (Ketamina AgenerH, 100 mg/ml, Agener União, Embu-Guaçu, SP) of either 25 (n 5 6 rats), 50 (n 5 7 rats) or 75 mg/kg (n 5 5 rats) and were recorded for additional three hours (ketamine). All injections were intraperitoneal (IP). Depending on the stability of the recordings, each rat received up to 3 different doses separated by at least 3 days.

Methods
It should be noted that ketamine effects vary widely across species 56 : while 1-4 mg/ kg of intravenous (IV) ketamine induces deep anaesthesia in humans, 20 mg/kg of IV ketamine induces only hypnosis in rats 57 . Used in isolation, the reported anaesthetic dose of ketamine in rats is 200 mg/kg IP 55 . Also, IP administration is far less effective than subcutaneous (SC) injections 15 .
Electrophysiological recordings. Continuous recordings were performed using a multi-channel acquisition processor (MAP, Plexon Inc). Local field potentials (LFPs) were pre-amplified (10003), filtered (0.7-300 Hz), and digitised at 1000 Hz. Electrode placement in CA1 was confirmed by inspecting coronal brain sections stained with cresyl violet, and by assessing responses evoked by perforant path stimulation (single pulse, 500 mA) along with other standard electrophysiological parameters such as presence of ripple oscillations and multi-unit activity at the pyramidal cell layer, theta phase reversal across stratum radiatum, and maximal theta power at the hippocampal fissure 27 .
Behavioural analysis. Animals were video-recorded at 30 frames/second. Tracking of the animals position was made using MouseLabTracker (http://www.neuro.ufrn.br/ incerebro/mouselabtracker.php), an open-source MATLAB version of a previously described software 58 . In order to avoid measuring small movements such as head and tail movements only displacements $2.0 mm/frame were considered. Locomotor activity was binned into 5-min blocks.
Data analysis. Analyses of electrophysiological data were performed in MATLAB (MathWorks).
Filter settings and extraction of the instantaneous phase and amplitude. Filtering was obtained using a linear finite impulse response filter by means of the eegfilt function from the EEGLAB toolbox (http://sccn.ucsd.edu/eeglab/), which applies the filter forward and then again backwards to ensure that phase delays are nullified. The instantaneous amplitude and phase time series of a filtered signal were computed from the analytical representation of the signal based on the Hilbert transform (hilbert function, Signal Processing Toolbox).
Spectral analyses. Power spectra estimation was done by means of the Welch periodogram method using the pwelch function from the Signal Processing Toolbox (50% overlapping 4-s Hamming windows). The mean power over frequency ranges of interest was calculated for each electrode individually, then averaged across electrodes and animals. Phase coherence was calculated using the multitaper method by means of the coherencysegc function from the Chronux toolbox 59 (http://chronux.org/) with parameters TW 5 3 and K 5 5 tapers, and window length of 4 seconds. Phase coherence was averaged from all electrode pairs in all animals. Power and phase coherence time-courses were obtained by averaging their values in 5-min blocks.
Estimation of phase-amplitude coupling and comodulation maps. To assess phase-amplitude CFC, we used the Modulation Index (MI) recently described 22,24 . This index measures coupling strength between two frequency ranges of interest: a phase-modulating (f p ) and an amplitude-modulated (f A ) frequency. The comodulation map is obtained by expressing the MI for several frequency band pairs (4-Hz bin width with 2-Hz steps for f p , and 10-Hz bin width with 5-Hz steps for f A ) in a bi-dimensional pseudocolour plot (see Supplementary Fig. S3 online for an illustrative example). Comodulation maps were computed using 5-min long LFPs recorded from single electrodes; only time windows associated with robust theta oscillations were used. Mean CFC strength between two frequency ranges was obtained by averaging the corresponding MI values; for example, mean theta-HG coupling corresponds to the average of MI values in the (4-10 Hz) 3 (60-100 Hz) region of the comodulation map, and similarly for theta-HFO coupling. We only computed theta-HG and theta-HFO coupling strength for electrodes that had theta-HG and theta-HFO coupling in the comodulation map, respectively (see Supplementary Fig. S4 online). Recording sites that did not present clear CFC in the comodulation map, or which the comodulation map revealed spike contamination 30,60 (common in recordings from the CA1 pyramidal layer 30 and dentate gyrus 23 ), were not taken into account in further analyses (see Supplementary  Fig. S4 online for representative examples of discarded electrodes). Consistent with recent reports 23,31 , theta-HFO coupling was mainly present in electrodes in stratum oriens-alveus , and theta-HG coupling from the CA1 pyramidal layer to stratum lacunosum-moleculare/hippocampal fissure.
Triggered LFP averages and current source density (CSD). Probe signals were amplified (2003), filtered (1 Hz-7.5 kHz), and digitised at 25 kHz (RHA2116, Intan Technologies). LFP averages were obtained by first filtering the LFP signal into the frequency ranges of interest; the amplitude peaks of each band were then identified and used for averaging 500-ms epochs centred at these timestamps. CSD analysis was obtained by -A12B-C for adjacent sites. We used 60-s periods of prominent theta oscillations in these analyses.
Statistics. Group means were compared by t-test for independent samples or by repeated measures ANOVA followed by Bonferroni post-hoc test. Multiple regression was performed to study changes in CFC level corrected for changes in locomotion and theta activity (as assessed by the theta/delta ratio; see Supplementary