(PDF) Fast network oscillations

European Journal of Neuroscience European Journal of Neuroscience, Vol. 34, pp. 394–403, 2011 doi:10.1111/j.1460-9568.2011.07748.x NEUROSYSTEMS Fast network oscillations in vitro exhibit a slow decay of temporal auto-correlations Simon-Shlomo Poil,1 Rick Jansen,1,2 Karlijn van Aerde,1,* Jaap Timmerman,1 Arjen B. Brussaard,1 Huibert D. Mansvelder1,  and Klaus Linkenkaer-Hansen1,  1 Department of Integrative Neurophysiology, Center for Neurogenomics and Cognitive Research (CNCR), Neuroscience Campus Amsterdam (NCA), VU University Amsterdam, De Boelelaan 1085, 1081 HV Amsterdam, The Netherlands 2 Department of Mathematics, VU University Amsterdam, De Boelelaan 1085, 1081 HV Amsterdam, The Netherlands Keywords: acetylcholine, memory, mouse, ongoing oscillations, temporal auto-correlations Abstract Ongoing neuronal oscillations in vivo exhibit non-random amplitude fluctuations as reflected in a slow decay of temporal auto- correlations that persist for tens of seconds. Interestingly, the decay of auto-correlations is altered in several brain-related disorders, including epilepsy, depression and Alzheimer’s disease, suggesting that the temporal structure of oscillations depends on intact neuronal networks in the brain. Whether structured amplitude modulation occurs only in the intact brain or whether isolated neuronal networks can also give rise to amplitude modulation with a slow decay is not known. Here, we examined the temporal structure of cholinergic fast network oscillations in acute hippocampal slices. For the first time, we show that a slow decay of temporal correlations can emerge from synchronized activity in isolated hippocampal networks from mice, and is maximal at intermediate concentrations of the cholinergic agonist carbachol. Using zolpidem, a positive allosteric modulator of GABAA receptor function, we found that increased inhibition leads to longer oscillation bursts and more persistent temporal correlations. In addition, we asked if these findings were unique for mouse hippocampus, and we therefore analysed cholinergic fast network oscillations in rat prefrontal cortex slices. We observed significant temporal correlations, which were similar in strength to those found in mouse hippocampus and human cortex. Taken together, our data indicate that fast network oscillations with temporal correlations can be induced in isolated networks in vitro in different species and brain areas, and therefore may serve as model systems to investigate how altered temporal correlations in disease may be rescued with pharmacology. Introduction Neuronal oscillations are thought to play a critical role in the encoding correlations (1–25 s) and less stable alpha oscillations on short-to- and retention of information (Singer, 1999; Buzsaki, 2006; Michels intermediate time scales (< 1 s) compared with healthy subjects et al., 2008; Palva et al., 2010). In humans, working memory studies (Montez et al., 2009). Altered temporal correlations without changes have pointed to a functional connection between memory and the to time-averaged oscillation power have also been observed in other amplitude modulation of oscillations in different frequency bands disorders and frequency bands (Linkenkaer-Hansen et al., 2005; (Jensen et al., 2002; Howard et al., 2003; Jokisch & Jensen, 2007; Monto et al., 2007). Thus, identifying factors influencing amplitude van Vugt et al., 2010). Notably, parietal oscillations exhibit system- fluctuations of oscillations could provide a better understanding of atically longer periods of elevated amplitude with longer encoding and pathophysiological states and, possibly, yield new targets for retention intervals in a Sternberg task (Raghavachari et al., 2001). treatment. Thus, the temporal stability of oscillations may be important for The hippocampus is a key neuronal structure for memory and mnemonic operations. cognition (Axmacher et al., 2006; Montgomery & Buzsaki, 2007), and Preclinical studies have also pointed to a functional importance of it has been argued that gamma oscillations in the hippocampus play an the temporal structure of oscillations. For example, patients with important role in memory formation, because they serve the function of early-stage Alzheimer’s disease had less persistent temporal (auto-) binding functional regions (Freund & Buzsaki, 1996; Fell et al., 2001; Lisman et al., 2005; Axmacher et al., 2006; Colgin et al., 2009). Correspondence: Klaus Linkenkaer-Hansen, as above. Hippocampal network oscillations can be induced in vitro by E-mail:

muscarinic acetylcholine receptor activation (Fisahn et al., 1998). These oscillations fall in the beta-frequency range (15–30 Hz) when *Present address: Forschungszentrum Ju¨lich, Institute of Neuroscience and Medicine, INM-2, Leo-Brandt-Straße, D-52425, Ju¨lich, Germany. measured below 30 C, but in the gamma range (30–100 Hz) if   recorded at physiological temperatures (Dickinson et al., 2003). Here, H.D.M. and K.L.-H. contributed equally to this work. we will refer to these oscillations as ‘fast network oscillations’ (Mann Received 8 March 2011, revised 12 April 2011, accepted 3 May 2011 et al., 2005). ª 2011 The Authors. European Journal of Neuroscience ª 2011 Federation of European Neuroscience Societies and Blackwell Publishing Ltd Stability of fast network oscillations 395 Fast network oscillations induced in vitro share many characteris- et al., 2009 for more details). After recovery, slices were mounted on tics with gamma oscillations generated in CA3–CA1 in vivo (Csics- 8 · 8 arrays of planar Indium tin oxid microelectrodes (electrode size vari et al., 2003; Lu et al., 2010) and have been used as a model to 50 · 50 lm, electrode impedance < 22 kX, interpolar distance, identify cellular and synaptic mechanisms that shape the power, 150 lm; Panasonic MED-P5155, Tensor Biosciences, Irvine, CA, frequency or coherence of these oscillations (Bragin et al., 1995; USA). Multi-electrode probes were coated with 0.1% polyethyleni- Fisahn et al., 1998; Traub et al., 2000; Tiesinga et al., 2001; Csicsvari mine (Sigma-Aldrich) in 10 mm borate buffer (pH 8.4) for at least 6 h et al., 2003; Mann & Paulsen, 2005). before use. The multi-electrode probe was then placed in a chamber It is not known whether temporal (auto-)correlations are an saturated with humidified carbogen gas for at least 1 h. For recordings, exclusive feature of extended neuronal networks in vivo, or whether slices were maintained in submerged conditions at 25 C and reduced neuronal networks in vitro may also exhibit these correla- superfused with ACSF, bubbled with carbogen, at 4–5 mL ⁄ min. tions, and what factors may shape their amplitude structure. To address Spontaneous field potentials from all 64 recording electrodes were these questions, we studied cholinergic fast network oscillations acquired simultaneously at 20 kHz, using the Panasonic MED64 in vitro. We modulated cholinergic and GABAergic signalling to test system (Tensor Biosciences), and down-sampled off-line to 200 Hz. how these neurotransmitter systems affect the amplitude structure of Slices were recorded for a minimum of 10 min with 25 lm carbachol. fast network oscillations. We found that temporal auto-correlations can emerge in acutely isolated hippocampal and prefrontal cortex brain slices. Interestingly, the temporal correlations were maximal at Hippocampal concentration curves physiologically relevant levels of cholinergic drive, and could be During each recording session, four slices were recorded simulta- enhanced by increasing inhibition. neously. After placing the slices in the recording units with ACSF, spontaneous activity was recorded for 15–20 min. Subsequently, two different experimental protocols were followed. For the first protocol, Materials and methods the following concentrations of carbachol were washed onto the slice: Slice preparation and local field potential recordings 1, 1, 5, 10, 15, 20 and 25 lm. For each concentration, the activity was recorded for 20 min. For the second protocol, to measure the effect of Experiments were performed in accordance with the guidelines and GABAergic modulation, we used the following concentrations of with the approval of the Animal Welfare Committee of the VU carbachol and zolpidem, and the activity was recorded for the University Amsterdam, which operates in accordance with Dutch and specified duration: 5 lm carbachol, 0 lm zolpidem, 45 min; 5 lm European law. Unanaesthetized DBA ⁄ 2J mice (19 slices, 10 mice, 5 carbachol, 0 lm zolpidem, 25 min; 5 lm carbachol, 100 nm zolpi- males, Jackson Laboratories) and Wistar rats (n = 10, Harlan, The dem, 15 min; and 5 lm carbachol, 1 lm zolpidem, 20 min. The time Netherlands) were decapitated at postnatal day 13–15 and postnatal lines of these experiments are shown in Fig. 1F. day 14–28, respectively. Their brains were quickly removed and placed in ice-cold artificial cerebrospinal fluid (ACSF) containing (in mm): 125 NaCl, 25 NaHCO3, 3 KCl, 1.2 NaH2PO4, 1 CaCl2, Hippocampal slice selection and preanalysis 3 MgSO4 and 10 D(+)-glucose (carboxygenated with 5% CO2 ⁄ 95% O2) (for rats: 1.25 NaH2PO4 and 26 NaHCO3). For mice, horizontal For each experiment, a photograph was taken of the slice in the slices (400 lm thick) from the ventral hippocampus were cut by a recording unit in order to identify electrode locations (Fig. 1A). The microtome (Microm, Waldorf, Germany). Slices were stored in an hippocampus consists of three main regions: CA1, CA3 and the interface storage chamber at room temperature (22 C) and placed in dentate gyrus. We divided CA3 and CA1 into the sub-regions stratum ACSF containing 2 mm CaCl2 and 2 mm MgSO4. After 1 h, slices oriens (basal dendrites), stratum pyramidale (cell bodies) and stratum were placed on 8 · 8 planar multi-electrode arrays (Titanium nitrade radiatum ⁄ lacunosum–moleculare (apical dendrites), and the dentate electrode grids, electrode diameter 30 lm, contact impedance 30– gyrus into stratum moleculare (basal dendrites), stratum granulosum 50 kX, 200 lm distance, 60 recording electrodes; Multi Channel (cell bodies) and hilus (apical dendrites) (Fig. 1B). We developed an Systems GmbH, Reutlingen, Germany) (see Fig. 1A), with polyethy- interactive MATLAB procedure to classify each electrode in one of lenimine coating (Sigma-Aldrich, St Louis, MO, USA), and left for these nine hippocampal sub-regions, based on the picture of the 1 h in a chamber with humidified carbogen gas before their placement electrode grid on the hippocampus. Using Fourier analysis (see into the recording unit. The setup allowed for simultaneous measure- below), we determined for each channel whether oscillatory activity ments of four slices. The flow rate during recordings was 4–5 mL ⁄ min was present. Channels with no clearly detectable peak and, thus, a and the temperature was kept low (30 ± 0.3 C) to preserve slice very low signal-to-noise ratio were excluded from further analysis stability. The amplitude of oscillations at higher temperatures was also (< 5% of the data). markedly lower than at 30 C, resulting in an unfavourable signal-to- noise ratio. Therefore, all experiments were performed at 30 C (Van Aerde et al., 2009). Zolpidem was purchased from Duchefa (Haarlem, Peak frequency fitting and preprocessing of data The Netherlands) and carbachol from Sigma (St Louis, MO, USA). We developed an algorithm to objectively determine the peak Local field potentials were recorded (Fig. 1E) at 1 kHz. The frequency of the dominant oscillation (Fig. 1C). First, we computed recordings were down-sampled off-line to 200 Hz and converted to the power spectrum using the Welch method (Fig. 1C, thick blue MATLAB (The Mathworks, USA) files. Analysis was performed in line), and determined the approximate frequency interval at which MATLAB. the peak occurred by visual inspection in the 10–25 Hz range, e.g. For rats, coronal sections (350–400 lm) of the prefrontal cortex from 14 to 18 Hz in Fig. 1C. The median selected frequency range were cut using a Leica VT1000S vibratome slicer. Slices were then was 15–23 Hz. If no clear peak was seen, a standard interval of transferred to holding chambers in which they were left to recover at 10–25 Hz was chosen. A 1 ⁄ f line was fitted to the power spectrum in the room temperature for 1 h in ACSF containing (in mm): 1.25 interval from 2 to 43 Hz, excluding the peak interval (Fig. 1C, thin red NaH2PO4, 2 MgSO4, 2 CaCl2 and 26 mm NaHCO3 (see Van Aerde line). The 1 ⁄ f power spectrum was subtracted from the original ª 2011 The Authors. European Journal of Neuroscience ª 2011 Federation of European Neuroscience Societies and Blackwell Publishing Ltd European Journal of Neuroscience, 34, 394–403 396 S.-S. Poil et al. A B CA1 stratum oriens CA3 stratum radiatum/ CA1 stratum pyramidale lacunosum moleculare CA1 stratum radiatum/ DG stratum moleculare lacunosum moleculare DG stratum granulosum CA3 stratum oriens DG hilus region 200 μm CA3 stratum pyramidale 0 μM carbachol C D E Power (arb.) 15 μM carbachol 1 0.1 10 –2 0 10 20 10 30 50 Frequency (Hz) 2 mV 0.5 s F 40 Power (Arb.) 40 Freq. (Hz) 30 30 20 20 10 10 0 1200 2400 3600 4800 6000 7200 8400 9600 Time (s) ACSF 1 μM 1 μM 5 μM 10 μM 15 μM 20 μM 25 μM carbachol 40 Power (Arb.) 40 Freq. (Hz) 30 30 20 20 10 10 0 900 3600 5100 6000 7200 Time (s) ACSF 5 μM 5 μM 5 μM 5 μM carbachol 0.1 μM 1 μM zolpidem Fig. 1. Carbachol (carbamylcholine chloride) (CCH) induces oscillations in hippocampal slices. (A) A multi-electrode array covering an entire hippocampal slice. Black dots are electrodes, which have a separation of 200 lm. This picture was made for every experiment, and used to classify the electrodes into the nine hippocampal sub-regions shown in B. (B) Schematic overview of the hippocampal slice divided into sub-regions of dentate gyrus (DG); stratum molecular, stratum granulosum and hilus region, CA1 ⁄ CA3; stratum radiatum ⁄ lacunosum–moleculare, stratum pyramidale and stratum oriens. (C) Peak frequency fitting method. A 1 ⁄ f spectrum (thin red line) is fitted to the power spectrum (Welch method) (thick blue line), and the confidence interval is determined (dashed green line). If the peak is above the confidence interval, the peak frequency, power and width are estimated by fitting a Gaussian to the power spectrum subtracted with the 1 ⁄ f spectrum (see Materials and methods). (D) Power spectrum (Welch method). We studied oscillations in the dominant frequency range. (E) Visual inspection of local field potential (LFP) signals reveals that CCH induces oscillations. Examples of filtered LFP signals (10–40 Hz) at different concentrations from one channel in CA3 stratum radiatum ⁄ lacunosum–moleculare. (F) Grand-average multi-taper time–frequency plot shows the main frequency of the oscillations. Moving window size 1024 points (3.7–47.9 Hz). (Top) CCH experiment. Median across 14 DBA ⁄ 2J mice in CA3 stratum radiatum ⁄ lacunosum–moleculare, one channel per slice. (Bottom) Zolpidem experiment. Median across 11 DBA ⁄ 2J mice in CA3 stratum radiatum ⁄ lacunosum–moleculare, one channel per slice. For interpretation of color references in figure legend, please refer to the Web version of this article. spectrum, and a Gaussian was fitted in the interval defined by the visual As reported previously (Mann & Paulsen, 2005; Jansen et al., inspection to find the peak frequency and peak power. We calculated the 2009), the dominant frequency component of carbachol-induced power as the median of the peak power minus the 1 ⁄ f-fitted baseline oscillations is in the interval 10–25 Hz (Fig. 1D). Here, we only power (Fig. 1C), across channels in each sub-region. Peaks were analysed the dominant frequency component, because higher frequen- excluded from peak frequency ⁄ power analysis if: (i) the peak power cies were either harmonics of this dominant frequency or of very low was below the 95th confidence interval (Fig. 1C, dashed green line) of Signal-to-noise-ratio, which does not allow for the detection of the 1 ⁄ f fit, (ii) the peak was outside the 10–25 Hz interval, or (iii) the temporal correlations (Linkenkaer-Hansen et al., 2007). peak width was < 0.5 Hz. This method is similar to the one used in Jansen et al. (2009). Note that in vitro experiments performed below physiological temperatures lead to prolonged GABAergic inhibitory Amplitude envelope postsynaptic currents and we therefore observed peak frequencies at The amplitude envelope of the oscillations was extracted using band- lower frequencies than have typically been reported at higher temper- pass filters (finite impulse response filters with a Hamming window) atures (see Supporting Information Fig. S1) (Bragin et al., 1995; and the Hilbert transform (Fig. 2B). The band-pass was centred at the Dickinson et al., 2003). Importantly, the frequencies found in our study peak frequency and had a width of 0.5*(2)*SD [filter order 60, SD is are similar to those reported in similar studies (Fisahn et al., 2008; the width of the Gaussian fit to the peak (Fig. 1C); a 10–25 Hz band- Jansen et al., 2009; Van Aerde et al., 2009). pass was used if no peak frequency was found]. The median frequency ª 2011 The Authors. European Journal of Neuroscience ª 2011 Federation of European Neuroscience Societies and Blackwell Publishing Ltd European Journal of Neuroscience, 34, 394–403 Stability of fast network oscillations 397 Auto-correlations A B Life-time C D Median 0.5*median 95 Norm. Life-time Area 'X' P(life-time < t)(%) 1 6 4 4 50 ACSF 2 mV Cross-correlations 5 μM CCH 1 10 μM CCH 1 M 0.5 ed 5 15 25 ia Area 'Y' 0 0 1 –1 0.5 1.5 2.5 n Carbachol (μM) 1s Life-time (s) 0 2.5 5 7.5 Time (s) E DFA: 0.52; life-time: 445 ms F G 0.16 1 μM Auto-correlation 1.8 0.63 0.12 Log10 F(time) DFA: 0.63; life-time: 2560 ms 1.6 0.08 1.4 15 μM 1.2 0.04 1.0 0.50 DFA: 0.55; life-time: 1635 ms 0 3.2 10 32 3.2 10 32 20 μM Time (s) Time (s) 1 mV 5s Fig. 2. Temporal correlation and oscillation burst life-time analyses characterize the non-random temporal structure of amplitude fluctuations in oscillations. (A) The study of spatial and temporal dimensions of neuronal processing requires different correlation analyses. Coordination of anatomically distributed activity (parallel processing) may be studied by computing correlations between neuronal signals from different hippocampal regions (Cross-correlations). In contrast, coordination of brain activity over time (serial processing) may be studied by computing temporal auto-correlations in neuronal signals within a single hippocampal region (Auto-correlations). Serial processing requires a sequence of causally related neuronal activities, which is likely to give rise to correlations over time (temporal correlations), e.g. persistent oscillatory activity as reflected in a slow amplitude modulation as studied here. Thus, by studying auto-correlation properties we may learn about novel mechanisms of attention and memory. Figure from Montez et al. (2009). ª2009 National Academy of Sciences, USA. (B) The raw signal was band-pass filtered around its peak frequency (thick blue line), and the amplitude envelope (thin red line) was extracted using the Hilbert transform. To quantify differences in oscillation burst dynamics on short to intermediate time scales (< 1 s), we introduced a threshold at multiples of the median amplitude envelope (exemplified by the median threshold, top horizontal dashed line, black areas) and 0.5*median amplitude (lower horizontal dashed line, black and blue areas), and defined the start and end of an oscillation burst as the time points of crossing this threshold. The oscillation burst structure is largely independent of which threshold is used. (C) Cumulative probability distribution plot of life-times calculated using 0.5*median as the threshold at different concentrations. (D) Three-dimensional plot showing the 95th percentile life-time vs. carbachol (carbamylcholine chloride) (CCH) concentration, normalized with the life-time at the first period with 1 lm CCH, for different thresholds. We tested thresholds in the range of 0.1–1.1 times the median amplitude envelope. We normalized the life-time to avoid displaying the trivial effects of increased life-time with lower threshold. We observed that the life-time is largely independent of which threshold is used. (E) Amplitude envelopes of a signal with a low DFA exponent (0.52; 1 lm CCH, blue trace), a high DFA exponent (0.65; 10 lm CCH, red trace), and an intermediate exponent (0.55; 20 lm CCH, green trace). Visual inspection of the amplitude modulation at different CCH concentrations reveals a more stable modulation (high DFA exponent, long life- time) at physiologically relevant concentrations (15 lm) compared with higher CCH concentrations. Example segments of the amplitude envelope from the same slice at different CCH concentration. (F) Visualization of DFA for a signal with a low DFA exponent (0.52; blue circles), an intermediate exponent (0.55; green squares) and a high DFA exponent (0.63; red pluses). When the signal with high DFA is shuffled in oscillation cycle wide windows, the correlations disappear (0.50, black dots). The DFA exponents are the slopes of the lines obtained with linear regression in log–log coordinates. The DFA plots correspond to the amplitude envelopes partially shown in E. (G) Auto-correlations of the amplitude envelopes in E [1 lm CCH (blue circles), 20 lm CCH (green squares) and 10 lm CCH (red pluses)]. When the amplitude envelope of the 10 lm CCH is shuffled in oscillation cycle wide windows, the correlations disappear (black dots, Supporting Information Fig. S5). For interpretation of color references in figure legend, please refer to the Web version of this article. range was 10–25 Hz. For the rat data only a 10–25 Hz band-pass was series (Chen et al., 2002). Our artefact-rejection algorithm examined used, because analyses were also performed on channels not having a the amplitude envelope twice. In the first run, a window of 2 s was clear frequency peak. removed around amplitude values larger than 6 standard deviations from the median of the entire amplitude envelope and replaced with a zero. In the second run, the algorithm removed a window of 1 s around Rejection of high-amplitude artefacts large-amplitude artefacts that were larger than 6 standard deviations. We performed a rejection of high-amplitude artefacts in the amplitude envelope before oscillation burst analysis and detrended fluctuation analysis (DFA) (see below). This is recommended, because high- Oscillation burst structure analysis amplitude artefacts can corrupt the temporal structure and estimates of The life-times of oscillation bursts were based on the amplitude temporal correlations, whereas the removal of small segments of the envelope. We defined an oscillation burst to begin when the amplitude time series has a negligible influence on the DFA exponent of a time envelope passed above a predefined threshold (determined in time ª 2011 The Authors. European Journal of Neuroscience ª 2011 Federation of European Neuroscience Societies and Blackwell Publishing Ltd European Journal of Neuroscience, 34, 394–403 398 S.-S. Poil et al. windows of 1 min, to compensate for any slow transients in the level A B of activity), and to end when the amplitude envelope passed below this 18 Norm. life-time threshold (Fig. 2B) (Poil et al., 2008). The life-time measure for a 14 0.7 signal was then determined as the 95th percentile in the cumulative 10 DFA probability distribution of the life-times (Fig. 2C). We systematically 0.6 mapped the influence of different thresholds (Fig. 2D). The results 6 were largely independent of threshold and therefore we only report 2 0.5 statistics for the 0.5*median threshold. 0 1 1 5 10 15 20 25 0 1 1 5 10 15 20 25 Carbachol (μM) Carbachol (μM) Detrended fluctuation analysis of temporal auto-correlation C D decay 25 Peak Freq. (Hz) 6 Power (mV2) The DFA is a method used to quantify the decay of temporal (auto-) 20 correlations in time series with less strict assumptions about the 4 stationarity of the signal than the classical auto-correlation function or 2 15 power spectral density (Peng et al., 1994). An additional advantage of DFA is the greater accuracy in estimating the decay of auto- 0 correlations from a limited amount of data (Gao et al., 2006; 10 0 1 1 5 10 15 20 25 0 1 1 5 10 15 20 25 Linkenkaer-Hansen et al., 2007). Here, we used the DFA to quantify Carbachol (μM) Carbachol (μM) the complex temporal structure of amplitude fluctuations in ongoing oscillations in the time range of 3–20 s. We note that previous studies Fig. 3. Hippocampal oscillations in vitro exhibit a slow decay of temporal have applied DFA on similar time scales and used the DFA exponents correlations, and the oscillation burst life-time has a bell-shape dependence on to show that the temporal structure of ongoing oscillations differs carbachol concentration. (A) Box plots showing the normalized life-time at the 0.5*median threshold in CA3 stratum radiatum ⁄ lacunosum–moleculare (yel- between males and females (Nikulin & Brismar, 2005), is heritable low region on the hippocampus diagram). The boxes have lines at the lower (Linkenkaer-Hansen et al., 2007) and is sensitive to disease (Montez quartile, median (red line) and upper quartile values. The whiskers are lines et al., 2009). Together, these findings in humans underscore the extending from each end of the boxes to show the extent of the rest of the data relevance of quantifying auto-correlations in ongoing oscillations on up to the 1.5 interquartile range of the sample. The hippocampus diagram time scales up to a few tens of seconds using DFA. shows areas with a significant effect of carbachol across the carbachol concentrations of 10, 15, 20 and 25 lm (black dots) (Friedman test, P < 0.05, The main steps from the broadband signal to the quantification of n > 17, binomial corrected). (B) Box plot showing the DFA exponent fitted in temporal correlations using DFA have been explained in detail the range 3–20 s for different carbachol concentrations in CA3 stratum elsewhere (Linkenkaer-Hansen et al., 2001). In brief, the DFA radiatum ⁄ lacunosum–moleculare. The hippocampus diagram shows the areas provides a measure of how the root-mean-square fluctuation of the with a significant effect of carbachol across the carbachol concentrations of 10, 15, 20 and 25 lm (black dots) (Friedman test, P < 0.05, n = 19, bionomial integrated and linearly detrended signals, F(t), scales as a function of corrected). (C) Box plot showing the dependence of the power on the carbachol time window size, t (Fig. 2F). We computed the fluctuation function, concentration in CA3 stratum radiatum ⁄ lacunosum–moleculare (median across F(t), with an overlap of 50% between windows. The DFA exponent is channels). We did not observe an effect of carbachol on power across the the slope of the fluctuation function. A DFA exponent in the interval carbachol concentrations of 10, 15, 20 and 25 lm (Friedman test, not significant, n > 9, binomial corrected). (D) Same as in C, for peak frequency. of 0.5–1.0 indicates the presence of temporal (auto-)correlations, Black dot: Friedman test, P < 0.05, n > 9, Holm corrected. Note that two whereas an uncorrelated signal is characterized by an exponent of 0.5. recordings were made at 1 lm. For interpretation of color references in figure The temporal correlations are not computed between areas, but within legend, please refer to the Web version of this article. the signal (auto-correlations) (Fig. 2A). In Fig. 2E and F it is shown that the high DFA exponent at 15 lm in Fig. 2F is qualitatively example, in an interval from 10 to 25 lm, it would test for paired reflected in amplitude modulation on long time scales compared with differences in the median between the observations at 10, 15, 20 and lower or higher concentrations of carbachol (Fig. 2E). The correspon- 25 lm. For the Friedman test, a Greenhouse–Geisser correction was dence between the DFA and the classical auto-correlation function is used to correct for sphericity when the Greenhouse–Geisser epsilon shown in Fig. 2F and G using the representative signals from Fig. 2E. was below 0.75, and a Huynh-Feldt correction was used otherwise As reported earlier, however, we found that the auto-correlation (Greenhouse & Geisser, 1959; Feldt, 1976). When comparing the function was too noisy for estimating the decay of correlations (Gao medians of two groups, such as two concentrations, we used et al., 2006; Linkenkaer-Hansen et al., 2007). permutation tests on the median, with more than 10 000 permutations (Box & Andersen, 1955; Ernst, 2004). Figures 3A, 4B and 5A show life-times normalized with the life-times from the first 1 lm carbachol Statistical analysis measurement; nevertheless, all tests were performed on non-normal- Biomarkers obtained from the analysis of the oscillations were tested ized data. Box plots were used to display the distribution of data; the for normality using Lilliefors’ composite goodness-of-fit test and were boxes have lines at the lower quartile, median (red line) and upper often observed to not follow a Gaussian distribution. As an example of quartile values. The whiskers are lines extending from each end of the this we show distributions in Supporting Information Fig. S4. boxes to show the extent of the rest of the data (up to 1.5· interquartile Therefore, we used non-parametric tests. To test for the effect of the range of the sample) (McGill et al., 1978). Confidence intervals were carbachol or zolpidem concentration on frequency, power, life-time or found using non-parametric bias corrected and accelerated bootstrap DFA exponents, we used the non-parametric Friedman test (with (BCa) (n = 5000) (DiCiccio & Efron, 1996). Note that we used paired F statistics) (Friedman, 1937; Conover & Iman, 1981). In our case, the data for the statistical tests on the effect of concentrations, and Friedman test tests for paired differences in the median between therefore the confidence intervals cannot be used to infer significance. observations at different concentrations of carbachol or zolpidem. For We used Holm’s stepwise Bonferroni correction for multiple compar- ª 2011 The Authors. European Journal of Neuroscience ª 2011 Federation of European Neuroscience Societies and Blackwell Publishing Ltd European Journal of Neuroscience, 34, 394–403 Stability of fast network oscillations 399 isons within single measures when appropriate (Holm, 1979), and 2001). Whether the temporal structure of oscillations in neuronal binomial correction when the power of Holm correction was too low networks in vitro also shows temporal auto-correlations with a slow (defined as when more than one sub-region had P-values below 0.05, decay is not known. We used the DFA exponent as an index of the but all sub-regions were rejected by Holm correction). Binomial correlations from 3 to 20 s (see Materials and methods), and observed correction only controls the Type 1 error at the number of significant a significant variation with carbachol concentration in all sub-regions sub-regions. Significance was defined as P < 0.05 for all tests. (0–25 lm, n = 18–19, Friedman test, P < 10)5, Fig. 3B). The DFA exponents in CA3 stratum radiatum ⁄ lacunosum–moleculare at 15 lm were 0.62 ± 0.03 (n = 18). These DFA exponents clearly indicate that Results the correlations were significantly higher than uncorrelated noise (i.e. The frequency and power of cholinergically-induced fast network higher than 0.5, Supporting Information Fig. S5), and they were also oscillations in hippocampal slices in vitro strongly depend on the higher than those obtained with ACSF (0 lm carbachol) (0.53 ± 0.01; activity of pyramidal cells and interneurons in the CA3 area (Traub P < 0.002, paired permutation test). We conclude that hippocampal et al., 2000; Mann & Paulsen, 2005; Heistek et al., 2010). Typically, oscillations, even in isolated networks in vitro, can exhibit temporal cholinergic receptor activation induces oscillations in the CA3 and CA1 correlations with a slow decay, and that these are modulated by area with the strongest power in CA3 but with similar frequency. It is not cholinergic activation. This suggests that the correlations can also be known whether amplitude fluctuations of these oscillations have a non- generated in vivo in localized networks without external modulation. random temporal structure. Therefore, we recorded local field potentials from hippocampal slices using 8 · 8 multi-electrode grids with 200 lm spacing that covered the transverse section (Fig. 1A and B). Fast Carbachol concentration differentially influences peak frequency network oscillations were induced by application of the muscarinic and power acetylcholine receptor agonist carbachol, and field potential oscillations The classic oscillation properties of power and frequency may also be in the different hippocampal areas were analysed for peak frequency sensitive to changing cholinergic drive. The power of oscillations, (Fig. 1C), power, burst duration and temporal correlations using DFA. however, exhibited similar values at carbachol concentrations from 10 We quantified the duration or ‘life-time’ of oscillation bursts by the to 25 lm (Friedman test; not significant, n = 10–14, Fig. 3C). It heavy tail of their probability distribution using the 95th percentile should be noted, however, that the power is highly variable across (Fig. 2B–D). In Figs 3 and 5, all sub-regions in which significant effects slices. In contrast, the peak frequency decreased in stratum radia- were observed are marked with black dots. In the following we will for tum ⁄ lacunosum–moleculare of CA3 from 18.0 ± 1.0 to 16.0 ± 1.5 Hz simplicity only comment on observations from sub-regions in CA3. with increasing concentration (n = 14, Fig. 3D). Overall, we conclude that the oscillation power does not show the bell-shaped dependence on carbachol concentration in the range from 10 to 25 lm as was Oscillation burst life-time has a bell-shaped dependence on observed for the temporal correlations on short and long time scales. carbachol concentration However, both life-times and DFA exponents were correlated with To test whether the temporal structure of oscillations is dependent on the power (Supporting Information Fig. S3). level of cholinergic activation, we applied increasing concentrations of carbachol (Fig. 1F). First, spontaneous activity was recorded for 15– 20 min in the absence of carbachol. Subsequently, carbachol was bath Fast network oscillations in rat prefrontal cortex also exhibit a applied in concentrations of 1, 5, 10, 15, 20 and 25 lm. For each slow decay of temporal auto-correlations concentration, the activity was recorded for 20 min. The life-time of To investigate whether temporal correlations are a unique phenomenon oscillation bursts increased with increasing carbachol concentration (1– for mouse hippocampal slices, we analysed the amplitude modulation 10 lm) and, interestingly, decreased when the carbachol concentration of carbachol-induced oscillations (25 lm) at 10–25 Hz in rat prefrontal was further increased (10–25 lm) in all sub-regions of CA3 (Fig. 3A), cortex slices. We found that cholinergically-induced oscillations in the e.g. from 1600 ± 680 ms (median ± 95% BCa confidence interval prelimbic area of the medial prefrontal cortex have significantly higher halfwidth) to 1220 ± 340 ms in stratum radiatum ⁄ lacunosum–molecu- DFA exponents (0.58 ± 0.02) and life-times (470 ± 60 ms) in a lare of CA3 (Fig. 3A). To test the significance of the apparent bell-shaped channel with high signal-to-noise ratio (prelimbic layer 3 ⁄ 5, n = 10, dependence of life-time on carbachol concentration, we tested for a Fig. 4B and C) compared with the channel with the lowest mean difference in the median with one-tailed paired permutation tests between amplitude (0.51 ± 0.02 and 430 ± 20 ms, respectively; n = 10) (one- 5 and 15 lm, and 15 and 20 lm, assuming that 15 lm was maximal. tail paired permutation test), thus showing that these correlations were Indeed, the life-time was significantly longer at 15 lm (1820 ± 460 ms) not due to correlated background noise in the 10–25 Hz range. The compared with 5 lm (1160 ± 150 ms) and 20 lm (1260 ± 280 ms) in DFA exponents were not different from exponents found for hippo- all sub-regions of the hippocampus except CA3 stratum oriens and campal oscillations (CA3) at similar carbachol concentrations, but life- dentate gyrus stratum granulosum. The tendency for oscillations to have times were markedly shorter in the rat prefrontal cortex (Fig. 4B and long-lasting bursts at intermediate concentrations of carbachol was clear C). We conclude that temporal correlations may be a robust across many amplitude thresholds (Fig. 2D) and hippocampal regions phenomenon of neuronal oscillations in vitro. (Fig. 3A, inset and Supporting Information Fig. S2). Temporal correlations are enhanced by GABAergic potentiation Hippocampal oscillations exhibit a slow decay of temporal To investigate whether the temporal correlations could be mani- auto-correlations pulated, we applied the GABAA receptor allosteric modulator The temporal structure of brain oscillations recorded with electro- zolpidem in concentrations of 0, 0.1 and 1 lm in the presence of encephalography in vivo shows substantial temporal correlations on 5 lm carbachol. Zolpidem is known to increase the inhibitory time scales up to several tens of seconds (Linkenkaer-Hansen et al., postsynaptic current decay time in both pyramidal cells (Goldstein ª 2011 The Authors. European Journal of Neuroscience ª 2011 Federation of European Neuroscience Societies and Blackwell Publishing Ltd European Journal of Neuroscience, 34, 394–403 400 S.-S. Poil et al. A A B 18 Norm. life-time 14 0.7 10 DFA 1 μV 6 0.6 0.5 s 2 B C * 0.5 490 * 0 5 5 5 5 0 5 5 5 5 0.60 470 Carbachol (μM) Carbachol (μM) Lifetime (ms) 0.56 0.1 1 0.1 1 450 Zolpidem (μM) Zolpidem (μM) DFA 430 0.52 C D 25 410 0.48 6 Peak Freq. (Hz) High S/N Min amp. High S/N Min amp. 20 Power (mV2) Fig. 4. Prelimbic prefrontal cortex oscillations in vitro exhibit a slow decay of 4 temporal correlations. (A) Band-pass-filtered local field potential oscillations (10–25 Hz) from the prelimbic region of a prefrontal cortex slice from a 15 channel with high signal-to-noise ratio (S ⁄ N) (top), and from the channel with 2 minimum amplitude (bottom). (B) Oscillations in channels with high S ⁄ N have longer life-times. Box plot showing life-times (473 ± 12 ms, n = 12) from high S ⁄ N channels across slices, and life-times (435 ± 17 ms) from channels with 0 10 minimum amplitude. (C) Oscillations in channels with high S ⁄ N have temporal 0 5 5 5 5 0 5 5 5 5 correlations. Box plot showing DFA exponents (0.58 ± 0.03, n = 12) from high Carbachol (μM) Carbachol (μM) S ⁄ N channels across slices, and DFA exponents (0.52 ± 0.02) from channels with the minimum amplitude. *P < 0.05 (one-tail paired permutation test) 0.1 1 0.1 1 Zolpidem (μM) Zolpidem (μM) et al., 2002; Cope et al., 2005) and fast-spiking interneurons (Bacci Fig. 5. Temporal correlations are increased by GABAergic potentiation. (A) et al., 2003), and has been shown to decrease the frequency of Box plots of normalized life-times measured using 0.5 median as the threshold. carbachol-induced oscillations in the hippocampus (Palhalmi et al., Friedman tests were performed using zolpidem concentrations of 0, 0.1 and 2004; Cope et al., 2005; Heistek et al., 2010). We observed that the 1 lm (all with 5 lm carbachol). Black dots in hippocampus diagrams represent burst structure of carbachol-induced oscillations was influenced by the areas with a significant effect of zolpidem application (Friedman test, P < 0.05, n = 11, binomial corrected). (B) Same as in A, for DFA. (C) Same as in A, for potentiation of GABAergic signalling as reflected in life-times power. We observed no significant effects on power of zolpidem concentration increasing from 1040 ± 1100 to 2450 ± 1910 ms with the application (Friedman test, P > 0.05, n = 10, Holm corrected). (D) Same as in A, for peak of 1 lm zolpidem in CA3 stratum radiatum ⁄ lacunosum–moleculare frequency. Note that two recordings were made at 1 lm. and stratum pyramidale (n = 11, Friedman test, Fig. 5A). We also observed an increase in temporal correlations in these regions, with oscillations in mouse hippocampal and rat prefrontal cortex slices. We DFA exponents in CA3 stratum radiatum ⁄ lacunosum–moleculare at found that the amplitude modulation of mouse hippocampal and rat 1 lm zolpidem of 0.61 ± 0.03 compared with 0.59 ± 0.04 before the prefrontal cortex fast network oscillations in vitro exhibits a slow zolpidem application, and 0.53 ± 0.02 during ACSF (n = 11, Fried- decay of temporal auto-correlations, suggesting that these correlations man test, Fig. 5B). may emerge due to intrinsic properties of local neuronal circuits. Modulation of fast network oscillations with zolpidem at a low Interestingly, we found maximal life-times and temporal correlations concentration of carbachol (5 lm) did not affect power (n = 10, around 10–15 lm carbachol, suggesting that there is an optimal Friedman test, Fig. 5C). The peak frequency, however, decreased in all dynamic range of amplitude dynamics around the physiologically CA3 sub-regions from 18 ± 2 to 14 ± 1 Hz (Fig. 5D). We conclude relevant levels of cholinergic drive (Menschik & Finkel, 1998). Our that temporal correlations on both short and long time scales depend results may help to explain why patients with Alzheimer’s disease, on the level of inhibitory activity. known for their cholinergic deficits, have decreasing life-times and temporal correlations as observed for neuronal oscillations in temporo- parietal regions (Montez et al., 2009). Discussion Ongoing oscillations in humans exhibit complex fluctuations in amplitude, which are rich in information about the functional state of Life-time and temporal correlation analyses point to an optimal the underlying networks and thought to play an important role in range of cholinergic drive for meta-stable dynamics memory and attention (Raghavachari et al., 2001; Linkenkaer-Hansen The DFA pointed to a maximum strength of temporal correlations at et al., 2007; Montez et al., 2009). It is not known, however, whether concentrations of 10–15 lm, and with magnitudes (DFA around 0.6) oscillations in vitro have a similarly rich temporal structure. To test that were similar to those observed in human electroencephalogra- this, we investigated the temporal structure of carbachol-induced phy ⁄ magnetoencephalography recordings in the beta-frequency band ª 2011 The Authors. European Journal of Neuroscience ª 2011 Federation of European Neuroscience Societies and Blackwell Publishing Ltd European Journal of Neuroscience, 34, 394–403 Stability of fast network oscillations 401 (Linkenkaer-Hansen et al., 2001; Nikulin & Brismar, 2005; Monto that a decreased level of acetylcholine signalling is a key factor et al., 2007). The temporal correlations of fast network oscillations causing impaired cognitive function in Alzheimer’s disease (Menschik had a different dependence on cholinergic drive compared with the & Finkel, 1998; Francis et al., 1999; Ikonomovic et al., 2003; Moretti peak frequency and power, which is in line with previous reports et al., 2004). Interestingly, it was recently reported that the amplitude showing that temporal correlations are largely independent of the stability of oscillations is already impaired in the early stages of time-averaged power (Linkenkaer-Hansen et al., 2007; Monto et al., Alzheimer’s disease (Montez et al., 2009), which is in agreement with 2007). It has been suggested that a slow decay of temporal correlations the present data showing a high sensitivity of oscillation amplitude may arise from sub-cortical modulation or other mechanisms affecting stability to changing levels of cholinergic drive. We suggest that cortical excitability on longer time scales than the duration of biomarkers of amplitude stability based on the analytic techniques individual bursts (Poil et al., 2008). The present findings suggest that used in this study may prove valuable, e.g. in memory-related drug temporal correlations can also emerge in local networks without research. For example, future studies should test whether cholinergic external modulation on long time scales in both the hippocampus and or GABAergic manipulations in vivo lead to changes in the amplitude prefrontal cortex. This observation, however, does not exclude a stability of oscillations as observed in the present study, and whether possible additional external influence in vivo. The application of these changes co-vary with performance in cognitive tasks associated zolpidem led to enhanced temporal correlations, which is in agreement with sustained oscillatory activity. with the effect observed in the beta-frequency band in non-epileptic cortical regions of patients with epilepsy after administration of the benzodiazepine lorazepam (Monto et al., 2007), which is a GABAer- Supporting Information gic modulator enhancing inhibitory action, as does zolpidem. The life- Additional supporting information may be found in the online version times of oscillation bursts also increased with the application of of this article: zolpidem, suggesting that GABAergic potentiation may stabilize the Fig. S1. The frequency of fast network oscillations depends on amplitude of fast network oscillations. temperature. Peak frequency from eight slices from two C57BL6 ⁄ J The life-time analysis indicated an increased probability of long- mice measured for 10 min for each temperature at 25 lm carbachol. duration bursts of fast network oscillations at intermediate concentra- Wash-in at 25 C for 30 min. tions (10–15 lm) of carbachol. Interestingly, temporal correlations Fig. S2. Overview of life-time, DFA, power and frequency in different were also maximal at these intermediate levels of carbachol, which are areas and carbachol (carbamylcholine chloride) (CCH) concentrations. thought to mimic the physiological level of cholinergic input in vivo Hippocampal diagrams show which area each row represents. (Menschik & Finkel, 1998). Thus, the intermediate levels of Significant changes from 10 to 25 lm CCH are indicated with cholinergic drive provide a maximal dynamic range for meta-stable asterisks. Note that two recordings were made at 1 lm. dynamics. We propose that this maximum may be viewed as a Fig. S3. Power is correlated with DFA and life-time. (A) Power and mechanism of stochastic resonance (Gammaitoni, 1998), where the life-time ranked and correlated. Spearman rho = 0.9, P < 0.05. (B) cholinergic activation provides just sufficient random excitation Power and DFA ranked and correlated. Spearman rho = 0.7, P < 0.05. (‘noise’) to support the formation of synchronous assemblies, whereas Data in A and B are from a single channel in one slice; sub-region high levels of excitatory drive would disturb the delicate synchrony CA3 stratum radiatum ⁄ lacunosum–moleculare at 10 lm carbachol. and stability of the neuronal assemblies. Fig. S4. Parametric statistical tests do not give the same P-value as non- Interestingly, a maximal dynamic range is increasingly associated parametric tests, if the data do not follow the normal distribution. (A) with a critical state, which is characterized by balanced network DFA exponents from CA3 stratum radiatum ⁄ lacunosum–moleculare. activity and thought to play an important role in the efficient Lilliefors’ composite goodness-of-fit test shows that DFA exponents do processing of information (Linkenkaer-Hansen et al., 2001; Beggs & not have a normal distribution (PL = 0.01), which is also reflected in the Plenz, 2003, 2004; Chialvo, 2004, 2006; Kinouchi & Copelli, 2006; asymmetry of the box plot. The P-value from the non-parametric Levina et al., 2007; Beggs, 2008; Poil et al., 2008; Montez et al., Friedman test (PF = 0.06) is different (and non-significant) from the 2009; Priesemann et al., 2009). We speculate that temporal correla- parametric two-way ANOVA (PA = 0.03). Note that the non-parametric tions at high concentrations of carbachol decrease because of a test is more conservative than the parametric counterpart. (B) Peak changed balance between the excitatory and inhibitory populations frequency from CA3 stratum radiatum ⁄ lacunosum–moleculare. Lillief- (i.e. increased inhibitory dominance due to decreased excitatory to ors’ composite goodness-of-fit test shows that peak frequency values excitatory connectivity and increased inhibitory excitability) (Pitler & have a normal distribution (P = 0.2, i.e. the distributions of data points do Alger, 1992; Hasselmo et al., 1995; Tiesinga et al., 2001) combined not differ significantly from a normal distribution), which is also reflected with excessively random spiking activity. The increased inhibitory in the symmetry of the box plots. The P-value from the non-parametric dominance will give rise to oscillations with more uniform amplitudes. Friedman test (P = 0.001) is equal to the parametric two-way ANOVA The power spectrum analysis may not have captured these subtle (P = 0.001). (C) Histogram showing the non-Gaussian probability differences in oscillatory dynamics, because it only measures the distribution of DFA exponents obtained at 10 lm carbachol (see the ‘amount’ of activity, but not how this activity is distributed over time corresponding box plot enlarged in A). Box interval, 0.05. (D) Histogram (Fig. 2F). Overall, our analysis indicates that the amplitude modula- showing the Gaussian probability distribution of peak frequency from tion of oscillations can be described as coloured noise or a fractional 10 lm carbachol (see the corresponding box plot enlarged in B). Box Brownian process (Touboul & Destexhe, 2010). interval, 1. Note that two recordings were made at 1 lm. Fig. S5. Randomly shuffling the amplitude envelope time series removes the effect of carbachol concentration on DFA exponents. Boxplot of the DFA exponents of signals where the amplitude A possible link between oscillation life-time, memory and envelope has been randomly shuffled in mean oscillation cycle sized cholinergic deficits in Alzheimer’s disease? windows. No significant variation was found between 0 and 25 lm A proper level of cholinergic activation is thought to be crucial for carbachol (Friedman test, P > 0.05, uncorrected). Note that two learning and memory (Hasselmo, 2006). For example, it is believed recordings were made at 1 lm. ª 2011 The Authors. European Journal of Neuroscience ª 2011 Federation of European Neuroscience Societies and Blackwell Publishing Ltd European Journal of Neuroscience, 34, 394–403 402 S.-S. Poil et al. Please note: As a service to our authors and readers, this journal Fell, J., Klaver, P., Lehnertz, K., Grunwald, T., Schaller, C., Elger, C.E. & provides supporting information supplied by the authors. Such Ferna´ndez, G. (2001) Human memory formation is accompanied by rhinal- hippocampal coupling and decoupling. Nat. Neurosci., 4, 1259–1264. materials are peer-reviewed and may be re-organized for online Fisahn, A., Pike, F.G., Buhl, E.H. & Paulsen, O. (1998) Cholinergic induction delivery, but are not copy-edited or typeset by Wiley-Blackwell. of network oscillations at 40 Hz in the hippocampus in vitro. Nature, 394, Technical support issues arising from supporting information (other 186–188. than missing files) should be addressed to the authors. Fisahn, A., Neddens, J., Yan, L. & Buonanno, A. (2008) Neuregulin-1 modulates hippocampal gamma oscillations: implications for schizophrenia. Cereb. Cortex, 19, 612–618. Francis, P.T., Palmer, A.M., Snape, M. & Wilcock, G.K. (1999) The cholinergic Acknowledgements hypothesis of Alzheimer’s disease: a review of progress. Br. Med. J., 66, 137–147. This work was supported by The Netherlands Organization for Scientific Freund, T. & Buzsaki, G. (1996) Interneurons of the hippocampus. Hippo- Research (NWO) [Toptalent grant to S.-S.P.; R.J. was supported by a campus, 6, 347–470. Computational Life Sciences grant (635.100.005); Innovative Research Incen- Friedman, M. (1937) The use of ranks to avoid the assumption of normality tive Schemes grant to K.L.-H.], the Neuro-Bsik Mouse Phenomics consortium implicit in the analysis of variance. J. Am. Stat. Assoc., 32, 675–701. (http://www.neurobsik.nl) (grant to A.B.B), the Royal Netherlands Academy of Gammaitoni, L. (1998) Stochastic resonance. Rev. Mod. Phys., 70, 223–287. Arts and Sciences (KNAW) (fellowship to H.D.M.), and the Danish Research Gao, J., Hu, J., Tung, W., Cao, Y., Sarshar, N. & Roychowdhury, V.P. (2006) Agency (to K.L.-H.). Assessment of long-range correlation in time series: how to avoid pitfalls. Phys. Rev. E. Stat. Nonlin. Soft Matter Phys., 73, 016117. Goldstein, P.A., Elsen, F.P., Ying, S.W., Ferguson, C., Homanics, G.E. & Harrison, N.L. (2002) Prolongation of Hippocampal Miniature Inhibitory Abbreviations Postsynaptic Currents in Mice Lacking the GABAA Receptor alpha ACSF, artificial cerebrospinal fluid; DFA, detrended fluctuation analysis. 1 Subunit. J. Neurophysiol., 88, 3208. Greenhouse, S.W. & Geisser, S. (1959) On methods in the analysis of profile data. Psychometrika, 24, 95–112. Hasselmo, M.E. (2006) The role of acetylcholine in learning and memory. Curr. References Opin. Neurobiol., 16, 710–715. Axmacher, N., Mormann, F., Ferna´ndez, G., Elger, C.E. & Fell, J. (2006) Hasselmo, M.E., Schnell, E. & Barkai, E. (1995) Dynamics of learning and Memory formation by neuronal synchronization. Brain Res. Rev., 52, 170– recall at excitatory recurrent synapses and cholinergic modulation in rat 182. hippocampal region CA3. J. Neurosci., 15, 5249–5262. Bacci, A., Rudolph, U., Huguenard, J.R. & Prince, D.A. (2003) Major Heistek, T., Timmerman, J., Spijker, S., Brussaard, A. & Mansvelder, H. (2010) differences in inhibitory synaptic transmission onto two neocortical GABAergic synapse properties may explain genetic variation in hippocam- interneuron subclasses. J. Neurosci., 23, 9664. pal network oscillations in mice. Front. Cell Neurosci., 4, 18. Beggs, J.M. (2008) The criticality hypothesis: how local cortical networks Holm, S. (1979) A simple sequentially rejective multiple test procedure. Scand. might optimize information processing. Philos. Transact. R. Soc. A, 366, J. Stat., 6, 65–70. 329–343. Howard, M.W., Rizzuto, D.S., Caplan, J.B., Madsen, J.R., Lisman, J., Beggs, J.M. & Plenz, D. (2003) Neuronal Avalanches in neocortical circuits. Aschenbrenner-Scheibe, R., Schulze-Bonhage, A. & Kahana, M.J. (2003) J. Neurosci., 23, 11167. Gamma oscillations correlate with working memory load in humans. Cereb. Beggs, J.M. & Plenz, D. (2004) Neuronal Avalanches are diverse and precise Cortex, 13, 1369–1374. activity patterns that are stable for many hours in cortical slice cultures. Ikonomovic, M.D., Mufson, E.J., Wuu, J., Cochran, E.J., Bennett, D.A. & J. Neurosci., 24, 5216. DeKosky, S.T. (2003) Cholinergic plasticity in hippocampus of individuals Box, G.E.P. & Andersen, S.L. (1955) Permutation theory in the derivation of with mild cognitive impairment: correlation with Alzheimer’s neuropathol- robust criteria and the study of departures from assumption. J. R. Stat. Soc. B ogy. J. Alzheimers Dis., 5, 39–48. Met., 17, 1–34. Jansen, R., Linkenkaer-Hansen, K., Heistek, T., Timmerman, J., Mansvelder, Bragin, A., Jando´, G., Na´dasdy, Z., Hetke, J., Wise, K. & Buzsaki, G. (1995) H.D., Brussaard, A.B., de Gunst, M. & van Ooyen, A. (2009) Inbred mouse Gamma (40-100 Hz) Oscillations in the Hippocampus of the behaving rat. strains differ in multiple hippocampal activity traits. Eur. J. Neurosci., 30, J. Neurosci., 15, 47–50. 1092–1100. Buzsaki, G. (2006) Rhythms of the Brain. Oxford University Press, Oxford. Jensen, O., Gelfand, J., Kounios, J. & Lisman, J.E. (2002) Oscillations in the Chen, Z., Ivanov, P.C., Hu, K. & Stanley, H.E. (2002) Effect of nonstation- Alpha Band (9-12 Hz) Increase with Memory Load during Retention in a arities on detrended fluctuation analysis. Phys. Rev. E, 65, 41107–41122. Short-term Memory Task. Cereb. Cortex, 12, 877–882. Chialvo, D.R. (2004) Critical Brain networks. Physica A, 340, 756–765. Jokisch, D. & Jensen, O. (2007) Modulation of Gamma and Alpha Activity Chialvo, D.R. (2006) Are our senses Critical. Nat. Phys., 2, 301–302. during a Working Memory Task Engaging the Dorsal or Ventral Stream. Colgin, L.L., Denninger, T., Fyhn, M., Hafting, T., Bonnevie, T., Jensen, O., J. Neurosci., 27, 3244–3251. Moser, M.B. & Moser, E.I. (2009) Frequency of gamma oscillations routes Kinouchi, O. & Copelli, M. (2006) Optimal dynamical range of excitable flow of information in the hippocampus. Nature, 462, 353–357. networks at criticality. Nat. Phys., 2, 348–351. Conover, W.J. & Iman, R.L. (1981) Rank transformations as a bridge between Levina, A., Ernst, U. & Michael Herrmann, J. (2007) Criticality of avalanche parametric and nonparametric statistics. Am. Stat., 35, 124–129. dynamics in adaptive recurrent networks. Neurocomputing, 70, 1877–1881. Cope, D.W., Halbsguth, C., Karayannis, T., Wulff, P., Ferraguti, F., Hoeger, H., Linkenkaer-Hansen, K., Nikouline, V.V., Palva, J.M. & Ilmoniemi, R.J. (2001) Leppa, E., Linden, A.M., Oberto, A. & Ogris, W. (2005) Loss of zolpidem Long-Range Temporal Correlations and Scaling Behavior in Human Brain efficacy in the hippocampus of mice with the GABAA receptor [gamma] 2 Oscillations. J. Neurosci., 21, 1370–1377. F77I point mutation. Eur. J. Neurosci., 21, 3002. Linkenkaer-Hansen, K., Monto, S., Rytsa¨la¨, H., Suominen, K., Isometsa¨, E. & Csicsvari, J., Jamieson, B., Wise, K.D. & Buzsa´ki, G. (2003) Mechanisms of Ka¨hko¨nen, S. (2005) Breakdown of Long-Range Temporal Correlations in gamma oscillations in the hippocampus of the behaving rat. Neuron, 37, Theta Oscillations in Patients with Major Depressive Disorder. J. Neurosci., 311–322. 25, 10131. DiCiccio, T.J. & Efron, B. (1996) Bootstrap confidence intervals. Stat. Sci., 11, Linkenkaer-Hansen, K., Smit, D.J.A., Barkil, A., van Beijsterveldt, T.E.M., 189–212. Brussaard, A.B., Boomsma, D.I., van Ooyen, A. & de Geus, E.J.C. (2007) Dickinson, R., Awaiz, S., Whittington, M.A., Lieb, W.R. & Franks, N.P. (2003) Genetic Contributions to Long-Range Temporal Correlations in Ongoing The effects of general anaesthetics on carbachol-evoked gamma oscillations Oscillations. J. Neurosci., 27, 13882–13889. in the rat hippocampus in vitro. Neuropharmacology, 44, 864–872. Lisman, J.E., Talamini, L.M. & Raffone, A. (2005) Recall of memory Ernst, M.D. (2004) Permutation methods: a basis for exact inference. Stat. Sci., sequences by interaction of the dentate and CA3: a revised model of the 19, 676–685. phase precession. Neural Netw., 18, 1191–1201. Feldt, L.S. (1976) Estimation of the Box correction for degrees of freedom from Lu, C., Jefferys, J., Toescu, E. & Vreugdenhil, M. (2010) In vitro hippocampal sample data in randomized block and split-plot designs. J. Educ. Stat., 1, 69– gamma oscillation power as an index of in vivo CA3 gamma oscillation strength 82. and spatial reference memory. Neurobiol. Learn. Mem., 95, 221–230. ª 2011 The Authors. European Journal of Neuroscience ª 2011 Federation of European Neuroscience Societies and Blackwell Publishing Ltd European Journal of Neuroscience, 34, 394–403 Stability of fast network oscillations 403 Mann, E.O. & Paulsen, O. (2005) Mechanisms underlying gamma (‘40 Hz’) Palva, J.M., Monto, S., Kulashekhar, S. & Palva, S. (2010) Neuronal synchrony network oscillations in the hippocampus - a mini-review. Prog. Biophys. reveals working memory networks and predicts individual memory capacity. Mol. Biol., 87, 67–76. Proc. Natl Acad. Sci. USA, 107, 7580. Mann, E., Suckling, J., Hajos, N., Greenfield, S. & Paulsen, O. (2005) Peng, C.K., Buldyrev, S.V., Havlin, S., Simons, M., Stanley, H.E. & Perisomatic feedback inhibition underlies cholinergically induced fast Goldberger, A.L. (1994) Mosaic organization of DNA nucleotides. Phys. network oscillations in the rat hippocampus in vitro. Neuron, 45, 105–117. Rev. E, 49, 1685. McGill, R., Tukey, J.W. & Larsen, W.A. (1978) Variations of box plots. Am. Pitler, T.A. & Alger, B. (1992) Cholinergic excitation of GABAergic Stat., 32, 12–16. interneurons in the rat hippocampal slice. J. Physiol., 450, 127–142. Menschik, E.D. & Finkel, L.H. (1998) Neuromodulatory control of hippocam- Poil, S.S., van Ooyen, A. & Linkenkaer-Hansen, K. (2008) Avalanche pal function: towards a model of Alzheimer’s disease. Artif. Intell. Med., 13, dynamics of human brain oscillations: relation to critical branching processes 99–121. and temporal correlations. Hum. Brain Mapp., 29, 770–777. Michels, L., Moazami-Goudarzi, M., Jeanmonod, D. & Sarnthein, J. (2008) Priesemann, V., Munk, M.H.J. & Wibral, M. (2009) Subsampling effects EEG alpha distinguishes between cuneal and precuneal activation in working in neuronal avalanche distributions recorded in vivo. BMC Neurosci., 10, 40. memory. Neuroimage, 40, 1296–1310. Raghavachari, S., Kahana, M.J., Rizzuto, D.S., Caplan, J.B., Kirschen, M.P., Montez, T., Poil, S., Jones, B.F., Manshanden, I., Verbunt, J.P.A., van Dijk, Bourgeois, B., Madsen, J.R. & Lisman, J.E. (2001) Gating of Human B.W., Brussaard, A.B., van Ooyen, A., Stam, C.J., Scheltens, P. & Theta Oscillations by a Working Memory Task. J. Neurosci., 21, 3175– Linkenkaer-Hansen, K. (2009) Altered temporal correlations in parietal 3183. alpha and prefrontal theta oscillations in early-stage Alzheimer disease. Proc. Singer, W. (1999) Neuronal synchrony: a versatile code for the definition of Natl Acad. Sci. USA, 106, 1614–1619. relations. Neuron, 24, 49–65. Montgomery, S.M. & Buzsaki, G. (2007) Gamma oscillations dynamically Tiesinga, P.H.E., Fellous, J.M., Jose´, J.V. & Sejnowski, T.J. (2001) Compu- couple hippocampal CA3 and CA1 regions during memory task perfor- tational model of carbachol-induced delta, theta, and gamma oscillations in mance. Proc. Natl Acad. Sci. USA, 104, 14495–14500. the hippocampus. Hippocampus, 11, 251–274. Monto, S., Vanhatalo, S., Holmes, M.D. & Palva, J.M. (2007) Epileptogenic Touboul, J. & Destexhe, A. (2010) Can Power-Law Scaling and Neuronal Neocortical Networks Are Revealed by Abnormal Temporal Dynamics in Avalanches Arise from Stochastic Dynamics? PLoS ONE, 5, e8982. Seizure-Free Subdural EEG. Cereb. Cortex, 17, 1386–1393. Traub, R.D., Bibbig, A., Fisahn, A., LeBeau, F.E.N., Whittington, M.A. & Moretti, D.V., Babiloni, C., Binetti, G., Cassetta, E., Dal Forno, G., Ferreric, F., Buhl, E.H. (2000) A model of gamma-frequency network oscillations Ferri, R., Lanuzza, B., Miniussi, C. & Nobili, F. (2004) Individual analysis of induced in the rat CA3 region by carbachol in vitro. Eur. J. Neurosci., 12, EEG frequency and band power in mild Alzheimer’s disease. Clin. 4093–4106. Neurophysiol., 115, 299–308. Van Aerde, K., Mann, E., Canto, C., Heistek, T., Linkenkaer Hansen, K., Nikulin, V.V. & Brismar, T. (2005) Long-range temporal correlations in Mulder, A., Van Der Roest, M., Paulsen, O., Brussaard, A. & Mansvelder, H. electroencephalographic oscillations: relation to topography, frequency band, (2009) Flexible spike timing of layer 5 neurons during dynamic beta age and gender. Neuroscience, 130, 549–558. oscillation shifts in rat prefrontal cortex. J. Physiol., 587, 5177–5196. Palhalmi, J., Paulsen, O., Freund, T.F. & Hajos, N. (2004) Distinct properties of van Vugt, M.K., Schulze-Bonhage, A., Litt, B., Brandt, A. & Kahana, M.J. carbachol- and DHPG-induced network oscillations in hippocampal slices. (2010) Hippocampal Gamma Oscillations Increase with Memory Load. Neuropharmacology, 47, 381–389. J. Neurosci., 30, 2694. ª 2011 The Authors. European Journal of Neuroscience ª 2011 Federation of European Neuroscience Societies and Blackwell Publishing Ltd European Journal of Neuroscience, 34, 394–403 Lifetime DFA Power Peak 16 0.8 70 25 * Norm. Life-time * Peak Freq. [Hz] Power [mV2] 12 50 0.7 20 DFA 8 30 0.6 15 4 10 0 0.5 10 * * * * * * * * 0 1 1 5 10 15 20 25 0 1 1 5 10 15 20 25 0 1 1 5 10 15 20 25 0 1 1 5 10 15 20 25 CCH [µM] CCH [µM] CCH [µM] CCH [µM] Figure S3 (a) 16 14 12 Power -ranked 10 8 6 4 2 0 0 2 4 6 8 10 12 14 16 DFA - ranked (b) 16 14 12 Power -ranked 10 8 6 4 2 0 0 2 4 6 8 10 12 14 16 Lifetime - ranked Figure S4 pL = 0.01 pL = 0.2 A B 0.8 pF = 0.06 25 pF = 0.001 Peak Freq. [Hz] pA = 0.03 pA = 0.001 0.7 20 DFA 0.6 15 0.5 10 0 1 1 5 10 15 20 25 carbachol [µM] C D 8 6 # obs. 4 2 0.5 0.7 0.9 13 17 21 DFA Peak Freq. [Hz] Figure S6 0.8 0.7 DFA 0.6 0.5 0 1 1 5 10 15 20 25 carbachol [µM]