Frequency- and state-dependent effects of hippocampal neural disinhibition on hippocampal local field potential oscillations in anaesthetised rats

Reduced inhibitory GABA function, so-called neural disinhibition, has been implicated in cognitive disorders, including schizophrenia and age-related cognitive decline. We previously showed in rats that hippocampal disinhibition by local microinfusion of the GABA-A antagonist picrotoxin disrupted memory and attention and enhanced hippocampal multi-unit burst firing recorded around the infusion site under isoflurane anaesthesia. Here, we analysed the hippocampal LFP recorded alongside the multi-unit data. We predicted frequency-specific LFP changes, based on previous studies implicating GABA in hippocampal oscillations, with the weight of evidence suggesting that disinhibition would facilitate theta and disrupt gamma oscillations. Using a new semi-automated method based on the kurtosis of the LFP peak-amplitude distribution as well as on amplitude envelope thresholding, we separated three distinct hippocampal LFP states under isoflurane anaesthesia: ‘burst’ and ‘suppression’ states – high-amplitude LFP-spike bursts and the interspersed low-amplitude periods – and a medium-amplitude ‘continuous’ state. The burst state showed greater overall power than suppression and continuous states and higher relative delta/theta power, but lower relative beta/gamma power. The burst state also showed reduced functional connectivity across the hippocampal recording area, especially around theta and beta frequencies. Overall neuronal firing was higher in the burst than the other two states, whereas the proportion of burst firing was higher in burst and continuous states than the suppression state. Disinhibition caused state- and frequency-dependent LFP changes, tending to increase power at lower frequencies (<20Hz), but to decrease power and connectivity at higher frequencies (>20Hz) in burst and suppression states. The disinhibition-induced enhancement of multi-unit bursting was also state-dependent, tending to be more pronounced in burst and suppression states than the continuous state. Overall, we characterized three distinct hippocampal LFP states in isoflurane-anaesthetized rats. Disinhibition changed hippocampal LFP oscillations in a state- and frequency-dependent way. Moreover, the disinhibition-induced enhancement of multi-unit bursting was also LFP-state dependent.


INTRODUCTION
Subconvulsive neural disinhibition, i.e. reduced inhibitory GABA function, within the hippocampus has been implicated in important neuropsychiatric disorders, including schizophrenia and age-related cognitive decline, and in the cognitive impairments characterizing these disorders (Bast, Pezze, & McGarrity, 2017;Benes & Berretta, 2001;Heckers & Konradi, 2015;McGarrity, Mason, Fone, Pezze, & Bast, 2017;Nava-Mesa, Jiménez-Díaz, Yajeya, & Navarro-Lopez, 2014;Palop & Mucke, 2016;Stanley, Fadel, & Mott, 2012;Thomé, Gray, Erickson, Lipa, & Barnes, 2016). In a recent study (McGarrity et al., 2017), we pharmacologically disinhibited the temporal (also known as ventral) to intermediate hippocampus in rats by acute local microinfusion of the GABA-A antagonist picrotoxin. Such hippocampal disinhibition caused clinically-relevant cognitive impairments, including in hippocampal memory function and in attentional performance that relies on prefrontal-striatal mechanisms, consistent with the idea that hippocampal disinhibition disrupts both hippocampal processing and processing in hippocampal projection sites (Bast et al., 2017). In addition, electrophysiological recordings around the infusion site under isoflurane anaesthesia showed that disinhibition enhanced burst firing of hippocampal neurons, as reflected by multi-unit data. The purpose of the present paper is to report the analysis of the impact of disinhibition on the hippocampal local field potential (LFP), which we recorded alongside the multi-unit data.
Brain rhythms or oscillations, i.e. synchronized changes in the activity of many neurons, as intrinsic but severed extrinsic connectivity, suggesting that GABAergic inhibition by parvalbumin neurons supports intrinsically generated theta (Amilhon et al., 2015). Finally, genetic knock-out of the GABA-A receptor subunit β3 decreased power, frequency and regularity of hippocampal theta and theta cross-correlation between different hippocampal subfields in freely moving mice (Hentschke et al., 2009).
Overall, the reviewed evidence on the GABAergic mechanisms of hippocampal theta and gamma oscillations suggests that pharmacological hippocampal disinhibition may alter both oscillations, with the weight of evidence suggesting that gamma amplitude may be reduced, whereas theta amplitude may be enhanced. However, there is also evidence suggesting that some features of the theta rhythm, including its coherence across the hippocampus, may be disrupted.
[ INSERT FIGURE 1] When analysing LFP recordings from anaesthetized rodents, it is important to consider that, under anaesthesia, many brain regions, including the hippocampus, show different LFP states characterized by distinct LFP patterns which can alternate (Clement et al., 2008;Kenny, Westover, Ching, Brown, & Solt, 2014;Land, Engler, Kral, & Engel, 2012;Lustig, Wang, & Pastalkova, 2016;Wolansky, Clement, Peters, Palczak, & Dickson, 2006) (for an example of our hippocampal recordings, see Figure 1A). One key LFP state observed with many LFP effects of hippocampal disinhibition anaesthetics, including isoflurane, is the burst-suppression state. The burst-suppression state refers to periods of LFP 'bursts' with high amplitude deflections interspersed between quiet periods of low amplitude LFP signal (= 'suppression' phases). Another type of LFP state is characterised by a more continuous LFP pattern with continuous LFP activity of lower amplitude than during LFP 'bursts'. The balance between the burst-suppression and continuous states changes with anaesthetic depth, with burst-suppression tending to become more dominant the deeper the anaesthesia (Land et al., 2012;Lustig et al., 2016). To analyse the impact of neuropharmacological manipulations, including hippocampal disinhibition, on LFP oscillations under anaesthesia it is important to separate these different LFP states for two main reasons. First, the large-amplitude LFP signals characterizing the burst state reflect major non-stationarities in the otherwise low-amplitude signals and may thus dominate the power spectral analysis, occluding frequency components that may be more characteristic of the other LFP states. Second, given the distinct characteristics of LFP patterns in different LFP states, neuropharmacological mechanisms underlying oscillatory activity may be LFPstate dependent.
In the present study, we aimed to characterise the effect of hippocampal disinhibition, by picrotoxin infusions into the temporal to intermediate hippocampus, on hippocampal neural oscillations around the infusion site using the LFP data recorded by McGarrity et al. (2017).
First, we developed an objective, semi-automated method to separate the three distinct LFP states described above (the continuous LFP state from the burst-suppression state, and, within the latter, the burst from the suppression state; Figure 1A) and compared several properties of these states (frequency-specific power and connectivity, and multi-unit parameters).
Second, we examined the impact of picrotoxin on hippocampal LFP properties (frequencyspecific power and connectivity) in the three distinct LFP states. Third, we examined if the disinhibition-induced enhancement of multi-unit burst firing reported previously (McGarrity et al., 2017) is dependent on the LFP state.

LFP and multi-unit data recording and pre-processing
Analysis was carried out on the hippocampal LFP and multi-unit data collected in our previous study (McGarrity et al. 2017). In brief, LFP and multi-unit data were recorded simultaneously in isoflurane anesthetized male adult (2-3 months) Lister-hooded rats, using a custom-made assembly of a 33-gauge stainless steel infusion cannula and an 8-electrode LFP effects of hippocampal disinhibition (microwire) recording array that was implanted into the hippocampus (Figure 2, top left).
The cannula tip, through which saline or picrotoxin solution could be injected, touched the electrodes and was positioned about 0.5 mm above the tips of the central electrodes. The assembly was stereotactically implanted into the right hippocampus, such that the 8-electrode array was arranged perpendicular to the brain midline and anterior to the infusion cannula, with the cannula tip aimed at coordinates in the right temporal (also known as ventral) to intermediate hippocampus (5.2 mm posterior to bregma, 4.8 mm lateral from midline, and 6.5 mm ventral from dura).The 8-electrode array spread approximately 2 mm, in the mediolateral direction (Figure 2, top right). The extracellular signal recorded by the electrodes was band-pass filtered into LFP (0.7 and 170 Hz) and multi-unit (250 Hz to 8kHz) data, which were recorded for a 30-min baseline period and a 60-min period following infusion (over about 1 min) of either 0.5ul saline (n = 7 rats) or 150ng/0.5ul picrotoxin (n = 6 rats) in a between-subjects design. All data were collected continuously and saved in 5 minutes bins, i.e. six pre-infusion (baseline) and 12 post-infusion 5-min bins. Although the picrotoxin group in the original study included eight rats (McGarrity et al. 2017), only six of these rats could be included in the present analysis, because one of the datasets could not be read into MATLAB for analysis and another dataset was recorded using a bundle array and, therefore, could not be combined with the rest of the data for LFP analysis.
[ INSERT FIGURE 2] All data were processed in MATLAB (R2016a) (The MathWorks, Inc., USA) using the FieldTrip toolbox (Oostenveld, Fries, Maris, & Schoffelen, 2011). To reduce noise, far-field effects and shared signal from all the electrode recordings, a montage of seven bipolar channels was digitally created using a pairwise bipolar subtraction (Land et al., 2012) of the eight original electrodes. Data was then pre-processed in fieldtrip by re-referencing with a common reference over all bipolar channels (except for the functional connectivity analysis) to remove any remaining linear gradient affecting all bipolar pairs, DC offset removal for each episode, and a band-stop filter between 49.5 and 50.5 Hz to remove the electric linenoise. The most medial and/or the most lateral electrodes of the 2-mm 8-electrode recording array were located outside of the hippocampus (typically 1-3 electrodes per rat) (Figure 2, bottom) and our previous multi-unit analysis revealed that hippocampal picrotoxin infusion did not affect any of the multi-unit parameters analysed from these electrodes (consistent with the densely packed fibre bundles surrounding the hippocampus) (McGarrity et al., 2017). If both of the electrodes used to calculate a synthetic bipolar electrode were outside of LFP effects of hippocampal disinhibition the hippocampus, that particular bipolar channel was removed, otherwise it was included in the analysis. As bipolar subtraction was not appropriate for multi-unit activity, in order to match the 7 bipolar channels used for the LFP analysis, electrodes 2 to 8 were selected and matched to their corresponding bipolar channel (i.e., electrode 2 to bipolar 1, electrode 3 to bipolar 2 etc.). Electrodes 2 to 8 were selected in place of 1 to 7 because in most cases electrode 1 (the most medial) sat outside the hippocampus and had to be excluded anyway.

Semi-automated separation of LFP states
The hippocampal LFP recordings under isoflurane anaesthesia clearly showed the three distinct LFP states outlined in the Introduction, including a burst, suppression and continuous state ( Figure 1A). The simultaneously recorded multi-unit activity also appeared to vary depending on the LFP state, with the burst state tending to show a higher multi-unit firing than the suppression state, and the continuous state characterized by continuous firing at a similar rate to that in the burst state. Our semi-automated separation strategy was to start by separating the combined burst-suppression state from the continuous state and then to separate the burst from the suppression state.
Simple thresholding based on continuous amplitude/power measures did, with our data, not result reliably in state separation that corresponded to the states that would be assigned by visual inspection (although thresholding worked for the example LFP traces in Figure 1A, it did not reliably separate the LFP states when these showed less pronounced differences; for examples, see Figure 1B, right). Therefore, we used a higher order moment (kurtosis) of the distribution of peak amplitude values (rather than the power of the continuous signal). To separate the three states, for each rat the channel with the most marked difference between LFP states based on visual inspection of the raw LFP traces, the 'LFP state-defining' channel, was selected (see Figure 1B for example). For this LFP state-defining channel, the peaks and troughs of the LFP trace were identified over baseline and post-infusion periods, by identifying the sign change of the first derivative. The distribution of these peaks and troughs identified from the sign-change turned out to be unimodal (not shown), and thus not enabling a clear separation into discrete states. In contrast, a measure of the kurtosis ('tailedness') of the amplitude distribution of those peaks and troughs separated the burst-suppression and continuous states (Figure 3). The kurtosis was calculated for a 10-s sliding window with a 1s step size (Figure 4A,B). A histogram of kurtosis values (a total of 4391) was plotted in 100 LFP effects of hippocampal disinhibition bins ( Figure 4C), and the most distinct local minimum (the first or second) separating peaks in the multimodal distribution, was selected as threshold to separate the combined burstsuppression states from the continuous state ( Figure 4D). If kurtosis values exceeded this threshold, the corresponding time points were marked as being in the 'burst-suppression' state, whereas time points with subthreshold values were marked as in the 'continuous' state ( Figure 4E). These labels were then applied to the samples of every channel from the same rat.

[INSERT FIGURE 3 AND 4 ABOUT HERE]
In a final step, we further separated the 'burst-suppression' state into the burst and suppression components, using a semiautomatic LFP-amplitude thresholding routine in Fieldtrip. To this end, a distribution of the amplitude envelope of the analytic signal was formed. Samples that exceeded a z-value of 5 and those falling within a symmetrical 0.25s window around the threshold crossing were consecutively labelled as 'Burst' state. This thresholding was run on each individual channel to prevent summation of z-values, which could occur from synchrony across channels and convolute the routine. The output of this routine gave the boundaries between low amplitude activity, corresponding to the suppression state, and the burst state.
Not all rats were spending time in each of the LFP states for every 5-min bin of the recording period, and four rats were not spending any time in the continuous state during baseline (therefore not allowing baseline normalisation of the post-infusion data). Therefore, the number of rats contributing to the different analyses of state-separated data could differ from the overall number of rats from which recordings were made ( Figure 5). The reduced sample size during the continuous state implies that the statistical power of any analysis in the continuous state is reduced compared to the analysis of the other two states.

Power-spectral density
Power at different frequencies was analysed separately for the different LFP states in FieldTrip, using Fast Fourier Transform with a Hanning window from 0.5 to 40 Hz. These power-spectra were subsequently averaged across channels and compared across the different LFP effects of hippocampal disinhibition states. Given the substantial raw-power differences between the different states (already separated in this analysis) and in order to investigate potential differences in relative spectral density, for each individual channel we additionally normalized the spectra to the total area under the curve of the power spectrum before comparison across states.

Connectivity
Phase-Lock Values (PLV) were calculated to indicate the phase-coherence (a measure of neuronal synchronization that is independent of amplitude correlations) of two signals originating from a pair of electrodes (measuring local activity due to their bipolar montage) (Aydore, Pantazis, & Leahy, 2013;Bastos & Schoffelen, 2015;Srinivasan, Winter, Ding, & Nunez, 2007). PLVs were calculated for every possible pair of hippocampal electrodes for every 0.5-Hz frequency bin. Because the individual electrodes between rats were not at consistent homological locations within the temporal to intermediate hippocampus, instead of averaging PLVs of electrode-pairs across subjects (as is common in EEG/MEG analysis where a standard mapping of electrodes does exist e.g. (Oostenveld & Praamstra, 2001), we used a different method to extract the communality of connectivity between all electrode pairs and to then calculate a statistic of this across rats (Cohen, 2015). To this end, we used singular value decomposition (SVD) over the imaginary part of the PLV matrix of all states and times. We used the imaginary part of the complex phase-locking-value (complex crossspectrum normalized on power in individual trials) as this is not corrupted by volumeconduction problems (Nolte et al., 2004). For each rat and frequency bin, we extracted the singular vectors U and V that satisfy the equation PLV = U * S * V', with S being the matrix of singular values and PLV the matrix of imaginary PLVs. The individual imaginary PLV matrices for each state and time combination were then projected into this space using the respective singular vectors and the first two "singular values" were extracted. These two values represent the connectivity strength within the two dominant 'neural networks' of the rat in the area of the temporal to intermediate hippocampus sampled by the bipolar 7electrode montage.

Multi-unit activity (MUA)
We compared the following key MUA parameters between LFP states: overall firing rat, bursts per minute, percentage of spikes in burst and burst duration. The original MUA LFP effects of hippocampal disinhibition analysis in our previous study (McGarrity et al., 2017) had been completed using NeuroExplorer (version 4). However, in order to examine the state-dependency of MUA parameters (and, subsequently, also of the previously reported disinhibition-induced enhancement of MUA burst parameters, see below), we replicated the original analysis using MATLAB, where we could then separate the different snippets of MUA by their associated LFP state. The MUA data was separated according to the different LFP states based on their time stamps in a way that preserved the original MUA bursts, with each burst assigned to the state occupied longest during the burst duration.

Statistical comparison of the three distinct hippocampal LFP states during baseline
To analyse the effect of the different states on the overall LFP and MUA we compared the baseline data, removing the factor of drug. For the baseline frequency specific effects (powerspectra and PLV eigenvalues) we compared each state to both other states in a separate analyses, after running a moving average over the frequency values, with a kernel of 1.5Hz sliding over 0.5Hz. The statistical analysis of these effects were carried out using pairwise comparisons at each frequency bin. In order to address the multiple comparisons problem arising from the simultaneous analysis of 80 different frequency bins, we conducted a cluster permutation analysis (Hayasaka, Phan, Liberzon, Worsley, & Nichols, 2004;Maris & Oostenveld, 2007). This method runs a mass-univariate independent samples t-test for each frequency bin, for both real data pairs (of the conditions to be compared) and randomly assigned pairs (combining data of both conditions to obtain a reference distribution).
Adjacent frequency bins that pass the univariate significance threshold form a cluster and a summary statistic of these clusters (e.g. summed t-value) is calculated for real data and reference distribution data. Only those clusters from the comparisons of the real data pairs which cut-off less than 5% of the maximum cluster-statistic from the reference distribution (random pairs) were considered as statistically significantly different.
For the state-separated analysis of MUA, a problem of missing data points over time arose due to the limited time each rat spent in each state (see Figure 5 for the missing data in different 5-min bins). Therefore, it was not possible to conduct a repeated-measures ANOVA and we used a pooling strategy. More specifically, we used the 5-min bins in the different states as the units of observation, such that data from all rats and time bins were pooled according to stateafter averaging across channels (for each rat). This was then subjected to LFP effects of hippocampal disinhibition a between-subjects one-way ANOVA, with LFP state as the between-subjects factor; this reflects a fixed-effect analysis.

Power-spectral Density
For the comparison of frequency-specific effects of drug infusion, the difference between pre-and post-infusion power was expressed as log of the ratio between post-infusion (powerspectra over 60 min) and pre-infusion (baseline; power spectra over 30 min) values for each channel. These difference values were then averaged per rat for statistical analysis, then for each drug group for plotting. The statistical analysis of the frequency specific effects of picrotoxin compared to saline infusions on LFP power were carried out using cluster permutation where, separately for each LFP-state, the log-ratio of both drug groups was compared.

Connectivity
The first two PLV 'Eigenvalues' for each rat (see State-separated analysis of hippocampal LFP properties, Connectivity) were averaged to give a single measure of aggregate connectivity. To normalise post-infusion values to pre-infusion (baseline) values, the preinfusion values were subtracted from the post-infusion values. As was done for the spectral power analysis, the cluster permutation statistic was used to compare the effects of picrotoxin vs. saline for each of the LFP-states separately.

MUA
Our previous analysis, which did not separate MUA data according to LFP state, showed that hippocampal disinhibition by picrotoxin causes enhanced burst firing, as reflected by increased bursts per minute, percentage of spikes in burst, and burst duration (McGarrity et al., 2017). Therefore, our present analysis focused on these burst parameters.
To assess the effect of picrotoxin on the state-separated MUA analysis, we could not apply the mixed-model ANOVA of MUA parameters with drug as between-subjects and time (5min bin) as repeated-measures factors, which we used in our original MUA data analysis LFP effects of hippocampal disinhibition (which did not separate data by LFP state) (McGarrity et al., 2017). This is because of the many missing data points, following separation of data according to LFP state (see Figure 5).
Therefore, we used the same pooled analysis as for the comparison of MUA parameters between LFP states during baseline, however, here with the additional factor of infusion group. Again, data were pooled over time and rats and separated according to LFP-state and infusion group. Prior to pooling, data for each rat, in each state, were normalised by subtracting the average of the pre-infusion values for each channel from the individual postinfusion values for the corresponding channel; subsequent to this, an average was taken across the channels, for each rat and time point separately. The normalised and pooled MUA data were subsequently analysed using a two-way ANOVA, with LFP state and infusion condition as between-subjects factors (and time-bins and rats as units of observation). A significant infusion condition X state interaction would statistically support state-dependent drug effects.

Differences in frequency-dependent power and functional connectivity
To compare quantitatively the three distinct hippocampal LFP states (Figure 1A), we combined the baseline data (i.e., data recorded before any drug infusion) from all infusion groups. During the baseline period, all rats showed the burst and suppression states, but not all rats showed a continuous state, resulting in a sample size of n=13 rats for the burst and suppression state and of n=9 rats for the continuous state ( Figure 5).
The three LFP states substantially differed with respect to overall raw power and with respect to relative power at different frequencies (Figure 6). The overall raw power in the burst state was significantly and substantially higher than in the suppression (cluster level p = 0.000999, corrected for multiple comparisons; cluster spanning across all frequencies, i.e. 1-39 Hz) and continuous states (p = 0.005, cluster including 1-22.5Hz). Suppression and continuous states did not differ significantly (p>0.05) indicating similar overall power, although at lower frequencies power tended to be somewhat higher in the continuous compared to the suppression state ( Figure 6A). With respect to relative power at different frequencies, the burst state showed a higher proportion of power within delta/low theta than the other two states and tended to show a lower proportion of power in high frequencies from about 20 to LFP effects of hippocampal disinhibition 40 Hz than both the suppression and continuous states, which were both characterized by conspicuous gamma 'bumps' in the normalised power spectra in this frequency range ( Figure   6B). For the comparison between burst and continuous states, a significant cluster was found at 3.5-10 Hz (p = 0.024) and between burst and suppression states a significant cluster was found at 4-14 Hz (p = 0.009). The numerical differences at higher frequencies did not reach statistical significance; neither for the comparisons of continuous and burst states (p = 0.16) nor suppression and burst states (p = 0.10).
[ INSERT FIGURE 6] Regarding connectivity, we computed an aggregate measure of phase-locking between all electrode pairs spread across the temporal to intermediate hippocampus (see Methods). To exclude any contamination by volume conduction, only the imaginary part of the PLVs was used and submitted to a SVD, of which we report the first two diagonal (Eigen-)values.
Across the frequency range examined (0.7 to 40 Hz), the burst state showed generally lower connectivity than both suppression and continuous states, which showed very similar connectivity ( Figure 6C). One significant cluster was found for the comparison between burst and continuous states at 1.5-9 Hz (p=0.018), and two significant clusters for the comparison between burst and suppression states: one cluster at 2.5-13 Hz (p =0.019) and another cluster at 17-38 Hz (p = 0.005). In all these clusters connectivity was reduced in the burst state, in comparison to both the continuous and the suppression state.
Overall, in terms of total power, relative power at different frequencies and functional connectivity, the burst state of the hippocampal LFP markedly differs from the suppression and continuous states, whereas the latter two states show similar characteristics in these parameters.

Differences in associated MUA
At baseline, some of the MUA parameters differed between states ( Table 1). The pooled analysis explained in the methods revealed that states significantly differed with respect to overall firing rate (spikes per second) (F(2,180) = 3.37, p = 0.037), with the firing rate in the burst state being higher or tending to be higher, respectively, than in the suppression state (p = 0.02) and continuous state (p = 0.058), which did not differ (p = 0.904). States also differed significantly with respect to percentage spikes in burst (F(2,177) = 9.44, p < 0.0001), with LFP effects of hippocampal disinhibition both burst and continuous states, which did not differ significantly (p = 0.108), showing a higher percentage of spikes in bursts than the suppression state (p = 0.002 and p<0.0001, respectively). States did not differ significantly with respect to bursts per minute (F(2,177) = 0.95, p = 0.383) or burst duration (F(2,172) = 1.19, p = 0.307). Overall, this quantitative comparison of MUA parameters is consistent with the impression based on visual inspection of the MUA data in the three states ( Figure 1A) that MUA activity is 'suppressed' in the suppression state compared to the burst and continuous state.

Hippocampal neural disinhibition did not affect the expression of the three hippocampal LFP states
During the post-infusion period, the cumulative time spent in the different LFP states was similar in rats infused with picrotoxin and saline (Figure 7). Independent sample t-tests revealed that there was no significant difference between the picrotoxin and saline groups in the total time spent in any of the three LFP states (continuous state, t(7)<1; suppression and burst state: t(11)<1).

Hippocampal neural disinhibition caused state-and frequency-dependent effects on hippocampal LFP power and functional connectivity
Hippocampal picrotoxin infusions, compared to saline, tended to increase power in the lower frequencies (<20 Hz) and to decrease power in the higher frequencies (>20 Hz), in both burst and suppression states, while having minimal effect in the continuous state ( Figure 8A,C,E).
However, only the picrotoxin-induced increase in power at 6-18Hz in the burst state attained statistical significance (p = 0.017). In the continuous state, power was numerically higher during the post-infusion period, as compared to the pre-infusion baseline; this was regardless of infusion, as indicated by a positive difference in log ratio power, especially at lower frequencies, which indicates 'baseline drift'. There was no difference between the drug groups in this state (no clusters) ( Figure 8A).
Hippocampal picrotoxin tended to decrease functional connectivity (as reflected by phaselock Eigenvalues) at higher frequencies in the burst and suppression states and at lower frequencies in the continuous state ( Figure 8B,D,F). However, the picrotoxin-induced LFP effects of hippocampal disinhibition decreases in connectivity reached statistical significance only in the burst state, in a highfrequency cluster ranging from 29.5-33 Hz (p = 0.049), but there was no significant picrotoxin-induced decrease in connectivity during the suppression state (no clusters). The visually apparent decrease in the continuous state (at lower frequencies) also failed to reach statistical significance, although there was a trend in a cluster from 6-7 Hz (p = 0.065).
[  (Figure 9M-O). ANOVA of these values, using LFP state and drug infusion as independent variable (see Methods 2.4.3.), revealed a significant main effect of drug infusion for all three burst parameters (outcomes of statistical analysis not shown), reflecting that, overall, picrotoxin infusion increased these values as compared saline infusion. Importantly, the ANOVA of burst duration ( Figure 9O) revealed a significant interaction of drug infusion X LFP state (F(2,316)=3.86, p=0.022), reflecting that picrotoxin increased burst duration in the burst and suppression states, but not the continuous state. Although, numerically, the picrotoxin-induced increases in bursts per min and LFP effects of hippocampal disinhibition percentage of spikes in bursts also tended to be somewhat weaker in the continuous, compared to the other two states (Figure 9M,N), ANOVA did not support significant interactions of drug infusion X LFP state for these two parameters (burst per min: F(2,343) = 1.118, p = 0.328; percentage spikes in bursts: F(2,343) <1). Overall, the increased multi-unit burst duration following hippocampal neural disinhibition was markedly less pronounced in the continuous state of the hippocampal LFP, as compared to the other two states. The other two burst parameters showed numerical tendencies in the same direction, but these were not statistically significant.

Discussion
We, first, developed a semi-automated method based on the kurtosis of the LFP peakamplitude distribution and amplitude envelope thresholding to separate three distinct hippocampal LFP states in the isoflurane anaesthetized rat: burst, suppression and continuous states (Figure 1), and we described some of the properties of these three LFP states. The burst state shows substantially larger overall raw power across the frequency range examined, compared to the suppression and continuous states (Figure 6A), as well as higher relative power in the delta/theta range (<15 Hz), but lower relative power in the high beta/low gamma range (>20 Hz) ( Figure 6B). Moreover, compared to suppression and continuous states, the burst state showed lower frequency specific 'functional connectivity' across the recording area in the temporal to intermediate hippocampus, as reflected by reduced phase-lock eigenvalues of the LFP signals recorded from the different electrodes ( Figure 6C). In terms of MUA, overall neuronal firing was higher in the burst state than in the two other states, whereas the percentage of spikes fired in bursts was higher in both the burst and continuous states as compared to the suppression state (Table 1). Finally, the expression of the hippocampal LFP states, as reflected by the cumulative duration of these states across the recording period, was not affected by hippocampal neural disinhibition (Figure 7).
Second, and importantly, we then showed that hippocampal neural disinhibition by picrotoxin changed the hippocampal LFP in a state-and frequency-dependent way (Figure 8). In both burst and suppression states, but not the continuous state, hippocampal neural disinhibition tended to increase LFP power at lower frequencies (<20 Hz) and to decrease power and functional connectivity at higher frequencies (>20 or >15 Hz, respectively). There was also a LFP effects of hippocampal disinhibition trend towards picrotoxin decreasing functional connectivity in the theta range (6-7 Hz) during the continuous state.
Third, we also found that the increase in multi-unit burst parameters by hippocampal neural disinhibition, which we reported in a previous study (McGarrity et al., 2017), was partially dependent on the hippocampal LFP state: disinhibition-induced increases in burst duration were limited to the burst and suppression state and not apparent in the continuous state (Figure 9).

Visual inspection of the LFP recorded from the temporal to intermediate hippocampus in
isoflurane-anaesthetised rats indicates three distinct LFP states, including burst, suppression and continuous states ( Figure 1A). We found that the kurtosis of the LFP peak and trough amplitude distribution was the best metric to separate the periods including burst and suppression states from the continuous state. Using this kurtosis measure, we thus separated the continuous state from the combined burst and suppression states, which could then be separated by amplitude envelope thresholding. Power analysis corroborated the impression based on visual inspection of the LFP traces, that raw power during the burst state is substantially higher than in the other two states, owing to large amplitude LFP bursts, and that the continuous state tends to show higher power than the suppression state (although the latter difference failed to reach statistical significance). Interestingly, analysis of relative power revealed a marked beta/gamma shoulder (24 to 34 Hz) in the power spectra of However, if the burst state is not separated from the other LFP states, hippocampal LFP power spectra recorded under anaesthesia do typically not show a beta/gamma shoulder (Clement et al., 2008;Lustig et al., 2016;Wolansky et al., 2006;and our unpublished observations). This likely reflects that the power spectra are dominated by the largeamplitude burst state (which lacks a beta/gamma shoulder). Overall, in the frequency range examined, continuous and suppression states tended to show higher connectivity, as indicated by phase-lock Eigenvalues, than the burst state across the recording area in the temporal to intermediate hippocampus. In the continuous state, a peak in connectivity was apparent at LFP effects of hippocampal disinhibition lower frequencies (<10 Hz, Figure 4C), which included the theta range, although statistically connectivity in this range did not significantly differ between continuous and suppression states. The numerically increased theta-range connectivity may correspond to findings of characteristic activity in the theta range in related hippocampal LFP states identified by others. More specifically, several studies have described hippocampal LFP states under anaesthesia, distinct from periods characterized by burst-suppression patterns and referred to as 'continuous'/ 'activated' (Clement et al., 2008;Wolansky et al., 2006) or 'light anaesthesia' (Land et al., 2012;Lustig et al., 2016) states, with distinct activity in the theta range (3-12Hz); although these studies did not examine connectivity within the hippocampal recording area at the frequencies we investigated.
The three hippocampal LFP states differ with respect to the associated multi-unit activity.
Our quantitative analysis of multi-unit parameters ( Table 1) corroborated the impression based on visual inspection of hippocampal recordings (Figure 1) that, overall, neuronal firing, in particular burst firing, is suppressed in the suppression as compared to the burst and continuous states. More specifically, the percentage of spikes fired in burst was significantly higher and overall firing rate was, or strongly tended to be, higher, respectively, in the burst and continuous state compared to the suppression state. It should be noted that, although these differences were suggested by visual inspection of most recordings, they were not always apparent even in recordings from different electrodes in the same rat. For example, in Figure 1 burst and suppression states clearly show higher MUA on channel 1 whereas on channel 2, which shows a higher overall firing rate, the states display similar MUA, possibly reflecting a ceiling effect. Increased neuronal firing in the burst compared to the suppression state has been reported previously in neocortical regions (Ferron, Kroeger, Chever, & Amzica, 2009;Kroeger & Amzica, 2007;Steriade, Amzica, & Contreras, 1994) and in the subiculum (Land et al., 2012). To explain this difference, it was proposed that anaesthesia induces a hyperexcitable state with an increased extracellular Ca2+ concentration. During the burst state the high Ca2+ would facilitate synaptic transmission causing the Ca2+ to move into the neuron, which would then lead to the suppression state, a refractory period during which Ca2+ concentration is relatively high within (and relatively low outside) of the neurons, therefore decreasing the likelihood of neuronal activity until the Ca2+ is transported back into the extracellular space (Ferron et al., 2009;Kroeger & Amzica, 2007). Taken together, our semi-automated separation method based on the kurtosis of the LFP amplitude distribution, revealed three distinct hippocampal LFP statesburst, suppression and LFP effects of hippocampal disinhibition continuous statewhose properties correspond to those of similar hippocampal LFP states identified previously based on visual inspection or other methods ( Table 2). The cumulative duration of each of the three hippocampal LFP states was not affected by local infusion of the GABA-A receptor antagonist picrotoxin, indicating that the expression of the states does not depend on GABA-A receptor-mediated inhibition in the hippocampus. However, as discussed in the next section, hippocampal neural disinhibition by picrotoxin changed the LFP properties within the three hippocampal LFP states.

State-and frequency-dependent changes in the hippocampal LFP power by local picrotoxin disinhibition
In both burst and suppression states picrotoxin, as compared to saline infusions, tended to increase hippocampal LFP power at lower frequencies (<20 Hz) and to decrease power at higher frequencies (>20 Hz) although only the increase at 6-18 Hz in the burst state reached statistical significance. In contrast, there was no evidence of picrotoxin-induced power changes in the continuous LFP state. Our findings are consistent with several previous in vivo and in vitro studies reporting changes in the hippocampal LFP following manipulations of GABAergic inhibition. Our finding that hippocampal GABA antagonism by picrotoxin facilitates LFP oscillations at lower frequencies, including the theta range, is consistent with many previous in vitro and vivo findings indicating an inverse relationship between GABAergic inhibition and theta oscillations in the hippocampus; which is consistent with the idea that hippocampal disinhibition by the medial septum, from where long-range GABAergic projections synapse onto parvalbumin-positive GABAergic inhibitory hippocampal interneurons (Freund & Antal, 1988), is a key extrinsic mechanism of hippocampal theta generation (Buzsaki, 2002;Colgin, 2016;Kowalczyk et al., 2013)see Introduction). However, some aspects of hippocampal theta and other slow oscillations may be facilitated by GABAergic inhibition. In rat hippocampal slices, GABA-A antagonists decreased the amplitude of electrically induced theta oscillations (Heynen et al., 1993), and abolished 'regular ventral hippocampus spontaneous synchronous activity', a 1-2 Hz oscillation (Papatheodoropoulos & Kostopoulos, 2002), possibly reflecting specific properties of these in vitro hippocampal slow oscillations. Moreover, optogenetic activation of parvalbumin positive GABAergic hippocampal interneurons strengthened, whereas silencing LFP effects of hippocampal disinhibition of these neurons disrupted, theta oscillations in an in vitro hippocampal preparation with intact intrinsic but severed extrinsic connectivity, showing that GABAergic inhibition by parvalbumin interneurons supports intrinsically generated hippocampal theta (Amilhon et al., 2015).
Although we did not find any statistically significant effects of picrotoxin on hippocampal LFP power in the beta/gamma frequencies (specifically 20-30Hz), picrotoxin tended to decrease power in this frequency band within the burst and suppression states. These numerical decreases are consistent with previous findings that pharmacological disinhibition disrupts hippocampal in vitro gamma oscillations and that GABAergic inhibition within the hippocampus plays a key role in generating these oscillations (Bartos et al., 2007;Buzsaki & Wang, 2012;Traub et al., 2004;Whittington et al., 2000). There is a wealth of in vitro studies showing that the GABA-A antagonist bicuculline applied to hippocampal slices decreases not only the power of 20-30Hz oscillations, but also the frequency of these oscillations (Arai & Natsume, 2006;Boddeke, Best, & Boeijinga, 1997;Shimono, Brucher, Granger, Lynch, & Taketani, 2000;Trevino, Vivar, & Gutierrez, 2007).

Reduced functional connectivity within the temporal to intermediate hippocampus after picrotoxin disinhibition
In the suppression and burst states, functional connectivity within the temporal to intermediate hippocampus, as reflected by phase locking of the LFP signal across the different electrodes of our recording array, was decreased in the gamma range (> 20 Hz), this decrease was statistically significant in the burst state from 29.5 -33Hz. This is consistent with previous in vitro findings that disinhibition by morphine disrupted synchrony of gamma oscillations across a hippocampal slice preparation (Faulkner, Traub, & Whittington, 1998;Whittington, Traub, Faulkner, Jefferys, & Chettiar, 1998). Synchrony was disrupted between areas that were 1.5-2.5 mm apart, whereas effects were not observed at shorter distances (Whittington et al., 1998). It is therefore possible that, in the present study, this effect was weakened, because we assessed the collective connectivity over our 2-mm electrode array, including short distances between neighbouring electrodes. Faulkner and colleagues (1998) also reported that a GABA-A agonist applied to hippocampal slices causes a similar decrease in gamma-frequency synchrony, suggesting that such synchrony depends on a balanced level of hippocampal GABA inhibition. Furthermore, neuro-computational studies suggest that, LFP effects of hippocampal disinhibition through multiple mechanisms, GABAergic interneurons are important for the synchronisation of gamma oscillations (Bartos et al., 2007;Whittington et al., 2000). For example, in a neurocomputational model of hippocampal pyramidal cells and GABAergic interneurons, blocking GABA-A receptors on the pyramidal cells abolished coherence at gamma frequencies (Traub et al., 2000). Our findings are the first in vivo findings to support these in vitro and modelling findings that the disruption of GABAergic inhibition, through blocking of hippocampal GABA-A receptors, disrupts synchronization of gamma oscillations across the hippocampus.
Finally, we found a numerical trend for picrotoxin to reduce functional connectivity in the theta frequency in the continuous hippocampal LFP state. This is consistent with previous findings that blockade of GABA-A receptors decreases enthorinal-subicular theta coherence in vitro (Levesque et al., 2017) and that genetic knock-out of the GABA-A receptor subunit β3 decreased theta cross-correlation between different hippocampal subfields in freely moving mice (Hentschke et al., 2009).

State-dependence of enhanced multi-unit burst firing caused by hippocampal neural disinhibition
Similar to the state-dependence of LFP changes, the picrotoxin-induced enhancement of hippocampal burst firing (McGarrity et al., 2017) tended to be more pronounced in the burst and suppression states than in the continuous state, although only the analysis of burst duration revealed a statistically significant interaction of the LFP state with the infusion effect. Previous findings suggest that in an isoflurane-induced burst-suppression state, the cortex is in a hyperexcitable state due to a decrease of excitation leading to a decrease in inhibition, which overall results in the excitation/inhibition balance being shifted towards excitation (Ferron et al., 2009). Neuronal hyperexcitability is also supported by the finding that, during the burst-suppression state, subliminal sensory stimuli activate cortical areas, including the subiculum, that are not activated by the same stimuli during other anaesthetized or the awake states (Kroeger & Amzica, 2007;Land et al., 2012). The more pronounced effects of picrotoxin during the burst and suppression states, compared to the continuous state, may reflect neuronal hyperexcitability during these states.

Hippocampus
Morphine disrupts long-range synchrony of gamma oscillations in hippocampal slices.

No equivalent state reported
State referred to as: Burst-Suppression LFP pattern: 'The EEG pattern of burst-suppression that followed administration of various anesthetics generally consisted of sharp waves with high amplitudes, at frequen-cies usually ranging between 2 and 7 Hz, separated by progressively longer periods of complete flatness. The electrical silence lasted for 1-2 sec during the initial stage of this pattern (Fig. 1A3)  The direct quotations highlight state properties corresponding to the 'continuous' and 'burst-suppression' states shown in the present study. Additional comments regarding the authors' interpretation of these states, and any comments on the relation to the states described in the present studies, are shown in the additional comments column.   posterior to the center of the array. In the bottom, the most medial (black dots) and most lateral (grey dots) electrode placements are indicated for all rats included in the analysis on drawings of coronal brain sections taken from the atlas by (Paxinos & Watson, 1998). Note that the most medial and/or the most lateral electrodes of the 2-mm 8-electrode recording array were located outside of the hippocampus (typically 1-3 electrodes per rat) and data from these electrodes was excluded from the analysis. Figure         state; these plots also indicate that the picrotoxin-induced increase in burst parameters is, at least numerically, less pronounced in the continuous than in the other two states.

Figure legends
LFP effects of hippocampal disinhibition Figure 1 Time (