HOME | SEARCH | PACS & MSC | JOURNALS | ABOUT | CONTACT US

Management of synchronized network activity by highly active neurons

This article has been downloaded from IOPscience. Please scroll down to see the full text article. 2008 Phys. Biol. 5 036008 (http://iopscience.iop.org/1478-3975/5/3/036008) The Table of Contents and more related content is available

Download details: IP Address: 132.66.248.126 The article was downloaded on 08/06/2009 at 09:02

Please note that terms and conditions apply.

IOP PUBLISHING

PHYSICAL BIOLOGY

doi:10.1088/1478-3975/5/3/036008

Phys. Biol. 5 (2008) 036008 (13pp)

Management of synchronized network activity by highly active neurons Mark Shein1,2, Vladislav Volman3,4, Nadav Raichman1, Yael Hanein2 and Eshel Ben-Jacob1,3,5 1 School of Physics and Astronomy, Raymond and Beverly Sackler Faculty of Exact Sciences, Tel Aviv University, Tel Aviv 69978, Israel 2 Department of Physical Electronics, School of Electrical Engineering, Tel Aviv University, Tel Aviv 69978, Israel 3 Center for Theoretical Biological Physics, University of California at San Diego, La Jolla, CA 92093-0319, USA 4 Computational Neurobiology Laboratory, The Salk Institute for Biological Studies, La Jolla, CA 92037, USA

E-mail: [email protected]

Received 19 May 2008 Accepted for publication 11 August 2008 Published 9 September 2008 Online at stacks.iop.org/PhysBio/5/036008 Abstract Increasing evidence supports the idea that spontaneous brain activity may have an important functional role. Cultured neuronal networks provide a suitable model system to search for the mechanisms by which neuronal spontaneous activity is maintained and regulated. This activity is marked by synchronized bursting events (SBEs)—short time windows (hundreds of milliseconds) of rapid neuronal firing separated by long quiescent periods (seconds). However, there exists a special subset of rapidly firing neurons whose activity also persists between SBEs. It has been proposed that these highly active (HA) neurons play an important role in the management (i.e. establishment, maintenance and regulation) of the synchronized network activity. Here, we studied the dynamical properties and the functional role of HA neurons in homogeneous and engineered networks, during early network development, upon recovery from chemical inhibition and in response to electrical stimulations. We found that their sequences of inter-spike intervals (ISI) exhibit long time correlations and a unimodal distribution. During the network’s development and under intense inhibition, the observed activity follows a transition period during which mostly HA neurons are active. Studying networks with engineered geometry, we found that HA neurons are precursors (the first to fire) of the spontaneous SBEs and are more responsive to electrical stimulations.

PDMS SBE

Abbreviations MEA FI

FS

HA neurons 5

multi-electrode array fraction of inter-spike intervals between 102 and 104 ms, relative to the total intervals of a neuronal spike train fraction of spikes fired by a neuron during the network’s silent state (between synchronized bursting events), relative to all the spikes fired by this neuron highly active neurons

Introduction Electrical activity in the brain is comprised of evoked activity (induced by external stimuli) and spontaneous (internally generated) activity (Kandel et al 2000). Over the years, considerable effort has been focused on studying the evoked activity (Adrian 1928) in an attempt to reveal the relationship between cause (stimulus) and effect (responding activity). Recently, increasing research efforts have been

Corresponding author.

1478-3975/08/036008+13$30.00

poly-dimethylsiloxane synchronized bursting event

1

© 2008 IOP Publishing Ltd Printed in the UK

M Shein et al

Phys. Biol. 5 (2008) 036008

directed towards the investigation of spontaneous activity and to studying the physiological mechanisms underlying the generation and regulation of this activity. These studies are motivated by an accumulating body of experimental evidence, which suggests that spontaneous activity may play a decisive role in shaping the structure of neural cell assemblies (Katz and Shatz 1996, Zhang and Poo 2001, Hua and Smith 2004, Spitzer 2006). Examples include regulation of cell migration and neurite navigation (Hanson and Landmesser 2004, Komuro and Kumada 2005), modification of dendrite motility and morphology (Lohmann et al 2002) and the fine tuning of synaptic strength (Mooney et al 1993). The latter hints that hidden non-arbitrary spatio-temporal motifs may exist in spontaneously generated electrical activity. If so, it consequently implies that some innate, yet unknown, mechanisms of activity maintenance, initiation and regulation should exist. For example, according to the Hebbian learning postulate, the way in which synaptic strength is modified crucially depends on the relative temporal ordering (correlation) between pre- and post-synaptic spikes (Hebb 1949, Bi and Poo 2001). At the same time, theoretical models (van Rossum et al 2000) and experimental evidence (Turrigiano 1999, Marder and Goaillard 2006) indicate that neurons use intrinsic homeostatic regulatory mechanisms to prevent the unconstrained growth of synaptic strength. Altogether, the emerging picture is that a network level (rather than a single cell level) approach is necessary in studying the spatio-temporal patterns of spontaneous activity and in searching for the mechanisms that initiate and regulate this activity. Such an approach can be implemented by utilizing cultured neuronal networks. Cultured networks that can exhibit spontaneous activity in the absence of any external electrical stimulations or chemical cues have been serving as a valuable model system for studying spontaneous activity (Maeda et al 1995, Segev et al 2002, 2004, Wagenaar et al 2006). The spontaneous activity in cultured neuronal networks is often marked by the existence of synchronized bursting events (SBEs)—relatively short time windows of several hundreds of milliseconds during which a large fraction of the cells is engaged in intense firing (Maeda et al 1995, Segev et al 2001, Baruchi and Ben-Jacob 2004, Beggs and Plenz 2004, Wagenaar et al 2006). The specific characteristics of the SBEs, such as their duration and frequency, depend on experimental conditions and on the developmental stage of the network. Yet, this mode of collective behavior has been observed across many different brain structures and preparations (Spitzer 1995, Yuste 1997, Menendez de la Prida et al 1998, Feller 1999, BenAri 2001, Palva et al 2000, Garaschuk et al 2000, Khazipov et al 2001). The time series of the SBEs has a non-arbitrary temporal appearance with long-range temporal correlations (Segev et al 2002, Hulata et al 2004). In addition, a network can exhibit several types of SBEs. Each type is characterized by its own distinct spatio-temporal patterns of neuronal firings, reflected in a specific organization of inter-neuron correlations (Segev et al 2004, Beggs and Plenz 2004, Baruchi and BenJacob 2004, Volman et al 2005, Raichman et al 2006). Recent studies also reveal that the spatio-temporal pattern of the SBEs

can endure chemical treatments that inhibit activity and even hypothermia conditions (Rubinsky et al 2007). Even though SBEs spontaneously emerge in developing networks, they can also be artificially evoked by targeted electrical (Madhavan et al 2007) and chemical (Baruchi and Ben-Jacob 2007) stimulations. Combined with model simulations (Volman et al 2004a, 2004b, 2007), these observations suggest that SBEs may serve as a template for information processing and storage, and can thus represent a new computational principle utilized by brain networks (Sejnowski and Paulsen 2006). In the present study, we used multi-electrode arrays (MEAs) (Potter 2001, Segev et al 2002) in order to monitor the electrical activity of in vitro cortical cultured networks. The advantage of cultures grown on MEAs is that they afford parallel recording of the electrical activity of many neurons for very long periods of time. Additionally, cultured networks are susceptible to geometrical manipulations (Sorkin et al 2006, Segev et al 2002). Imposing such pre-defined morphological constraints on the network’s structure enables the study of small isolated networks. Consequently, it is possible to infer the spatio-temporal patterns of the network’s activity and the relations between the activities of many different recorded neurons. We confirmed previous results showing that while most of the neurons fire during the SBEs, there is a small subset of neurons (∼10% in our cortical cultures) whose activity persists between the SBEs (Mao et al 2001, Streit et al 2001, Darbon et al 2002, Volman et al 2004a, Sipila et al 2005, Eytan and Marom 2006). It has been proposed and tested in numerical simulations of realistic neural network models (Volman et al 2004a, 2007) that such highly active (HA) neurons may maintain and regulate the network’s spontaneous activity. Following this theoretical work, we performed a detailed statistical and time-series analysis of the special firing characteristics of the highly active neurons. We found that the activity patterns of these neurons are marked by distinctive firing characteristics—unimodal distribution (versus bimodal distribution of the rest of the neurons) in the inter-spike intervals and long time correlations—that may be associated with the regulation of the SBEs. To further investigate the putative role of HA neurons in the maintenance and regulation of the network’s spontaneous activity, we studied the activity of cultured networks during their development from a collection of isolated cells into mature wired networks of interconnected neurons that show intense spontaneous activity. Our results show that the early network development is marked by a transient period, during which a large fraction of HA neurons are detected. This fraction gradually decreases during the course of the network’s maturation. A similar effect was observed in the recovery of networks from intense chemical inhibition of glutamatergic synapses. The fraction of HA neurons gradually increased, following the complete abolishment of spontaneous activity in the network. In addition, we found that HA neurons act as precursors of the activation of SBEs, in the sense that they are typically the first to fire during a synchronized bursting event. To further understand this effect, we investigated the response 2

M Shein et al

Phys. Biol. 5 (2008) 036008

of regular neurons and HA neurons to electrical stimulations. We found that HA neurons responded with higher fidelity to electrical stimulations regardless of the stimulation location. However, no correspondence was found between electrodes whose stimulation triggered activity in individual neurons or in the whole network and electrodes on which HA neurons were identified. These observations may imply that the HA neurons respond to inputs from a large fraction of neurons and may contribute to the generation and synchronization of SBEs.

Electrical stimulation Electrical stimulations were applied to the cultured network by delivering voltage pulses to specific electrodes using a stimulus generator (STG1008, Multi Channel Systems). All pulses were biphasic (positive-then-negative) pulses, 500 mV in amplitude, with each phase lasting 1 ms. Electrodes that showed neuronal activity were selected, so as to enable the monitoring of changes in electrical activity at the locations of the stimulations. For each electrode, a stimulation session included three sub-sessions separated by 30 s intervals. Each sub-session included ten stimulations separated by 5 s intervals.

Materials and methods Preparation and growing of cultured networks

Patterning

Dissociated cortical cultures were prepared as follows: the entire cortices of (E18) Sprague Dawley rat embryos were finely removed. The cortical tissue was digested with 0.065% trypsin (Biological Industries, Beit Haemek, 03-046-1) in phosphate buffered saline (Beit Haemek, 02-023-1) for 15 min, followed by mechanical dissociation by trituration. Cells were resuspended in a modified essential medium with Eagle’s salts (Beit Haemek, 01-025-1), 5% horse serum (Beith Haemek 04-004-1), 5 mg ml−1 gentamycin (Beith Haemek 03-035-1), 50 μM glutamine (Beith Haemek 03-020-1) and 0.02 mM glucose (BDH 101174Y), and plated on a poly-D-lysine (PDL) (Sigma, catalog no. p-7889) covered electrode array with a cell density of 3000–4000 cells mm−2 (∼1.5 × 106 cells were plated per dish). The cultures were maintained at 37 ◦ C with 5% CO2 and 95% humidity using a specially designed life support chamber. The growth medium was partially replaced every 3–4 days. Since the medium replacements elicit a temporary activity transient that lasts up to 2 h, data recorded during this period were ignored.

Cultured networks with a defined geometry were constructed by manually placing thin strips of poly-dimethylsiloxane (PDMS), with cross-sections of 0.5 mm × 0.5 mm and several millimeters length, onto the PDL-coated surface, before placing the cells. PDMS strongly adheres to the MEA’s glass surface and prevents electrical connection between neurons on both sides of the PDMS strip. PDMS is a bio-compatible substance and does not interrupt neuronal growth (Kane et al 1999). These cultures did not show any significant differences in comparison to unrestricted networks in terms of the parameters we measured (also see Segev et al (2002), Raichman and Ben-Jacob (2008)). Prolonged chemical inhibition Chemical inhibition was performed by applying both a-amino3-hydroxy-5-methylisoxazole-4-propionic acid (AMPA) and N-methyl d-aspartate (NMDA) receptor antagonists (20 μM b-cyano-7-nitroquinoxaline-2,3-dione (CNQX) and 100 μM (2R)-amino-5-phosphonovaleric acid (APV)) to the medium. Due to thermal fluctuations and the small volume of the medium, homogeneity in concentration was assumed to be reached within several seconds.

Electrical activity recording We used a commercial micro-electrode array which consists of 60 substrate-integrated thin film micro-electrodes (MEA chip, Multi Channel Systems). The electrodes are 30 μm in diameter and are arranged in a rectangular array with a 500 μm spacing between the electrodes. Therefore, our recording area covers 15 mm2 out of the 3 cm2 total area of the dish. Extra-cellular recordings were collected utilizing a low noise pre-amplifier board (B-MEA-1060, amplifier, gain X1200 with a band-pass filter 200 Hz–5 kHz, Multi Channel Systems). The signals recorded from the microelectrodes were sampled at a 10 kHz sampling rate and stored on a personal computer equipped with a 128-channel, 12-bit data acquisition board (MC Card, Multi Channel Systems) and MC Rack data acquisition software (Multi Channel Systems, version 3.2.20). The impedance of the electrodes is 100–500 k at 1 kHz and their bandwidth ranges from 10 Hz to 3 kHz, which permits the recording of individual action potentials. Potential spikes were initially identified according to a threshold crossing detection (set to 4 standard deviation of the noise) and were later sorted and discriminated from noise using a wavelet packet decomposition-based algorithm (Hulata et al 2000).

Analysis methods Identification of synchronized bursting events. The time series of SBE occurrences (SBE loci) was calculated by a twostep analysis: identification of all possible events followed by elimination of false positive events. We first divided the recording into 10 ms bins and summed over all neuronal spikes within each bin to calculate the total population activity. Next, we convoluted the population activity with a Gaussian (40 ms in width) to achieve a smooth function (figure 1(C)). The exact occurrence of the SBE was then determined by taking the time at which the convoluted population activity function reached local maxima (figure 1(B)). Our second step was to exclude false positive identification of SBEs by merging maxima less than 1000 ms apart (corresponding to the SBE refractory period (Segev et al 2001)) and by eliminating maxima in which less than N neurons fired simultaneously (N being 10– 20% of the total number of recorded active neurons in the network). The duration of each SBE was defined as the time 3

M Shein et al

Phys. Biol. 5 (2008) 036008

Neuron number

1 (A ) *

*

Firing intensity

56

* (B )

(C )

5

10

Sec

15

20

25

Figure 1. Characterization of electrical activity in cultured cortical networks. The generic activity pattern consists of synchronized bursting events (SBEs) separated by relatively long periods of low-frequency sporadic firing. A raster plot of the network’s activity, shown in (A), demonstrates that each bursting event has its characteristic spatio-temporal structure (fingerprint). Yet, while most of the neurons are active only during the SBEs, a small fraction (marked by asterisks on the right) exhibit persistent firing even during the long (seconds) periods of network quiescence. In the population activity presentation, shown in (B), we plot the number of spikes per second per neuron as a function of time bin. This presentation reveals that each SBE possesses the characteristic profile of a fast rise followed by a slower decay. We define the peaks of the population activity as the temporal loci of SBEs. (C) Application of the burst detection algorithm (as explained in the Methods section) yields temporal loci of SBEs (black vertical lines), along with their corresponding width (gray regions).

interval between the nearest upward threshold crossing before the activity peak and the nearest downward threshold crossing after the activity peak of the convoluted population activity function. Consequently, on a coarse scale, we characterized two states of the network’s activity: (1) the excited state—a state exhibiting global network excitations and (2) the silent state—a state during which only sporadic firing was observed (figure 1(B)). The convolution of the population activity function increases the SBE duration. Consequently, choosing a particularly low value for the threshold to distinguish between the two states (0.1% of the peak activity in each SBE) reassures us that the active states included all spikes related to the SBEs, including the first. Nevertheless, the total temporal fraction of the SBE durations, out of the total recording length, was less than 1% in all recordings.

Results Characterization of the network’s activity We recorded the electrical activity of ten mature (>12 days in vitro) cortical cultures, all of which exhibited spontaneous collective activity. In figure 1, we show a typical raster plot of the recorded network activity. Visual inspection of the raster plot indicates that the network activity can be characterized by two states. Most of the time, the majority of the recorded neurons are silent and only sporadic action potentials in some of the neurons are seen. From time to time, however, many of the recorded neurons participate in SBEs—relatively short (several hundreds of milliseconds) time windows during which most of the recorded neurons are engaged in intense electrical activity. In order to identify the temporal occurrences along with the widths of SBEs, we performed the following analysis. We first constructed the population activity representation shown in figure 1(B) (as explained in the Methods section). Observations on the population activity during the SBEs revealed that it follows a stereotypical profile—a fast rise (tens of milliseconds) followed by a slower fall (hundreds of milliseconds)—suggesting that most of the recorded neurons are rapidly activated at the onset of synchronized bursting events. Using the population activity function, we identified the exact temporal loci of the SBEs (the peak of the activity function) along with their widths (figure 1(C)), according to the procedure explained in the Methods section. The above findings are in agreement with previous reports and suggest that synchronized bursting events may constitute a generic organizational motif shared by many neuronal networks.

Estimation of power spectral density. For long signals prone to large fluctuations and non-stationary trends, the power spectrum can be estimated by calculating a count-based periodogram (Lowen et al 2001). To build the periodogram of a spike time series, we proceed as follows. First, the binary vector of the spike time series of each neuron was partitioned into non-overlapping contiguous segments of equal length T. Within each segment, a discrete sequence was formed by further dividing T into M equal bins, and then counting the number of events within each bin. A periodogram was then formed for each of the segments according to SW (f ) = 1 |W (f )2 |, where W (f ) is a discrete Fourier transform of M length M for each of the segments. Finally, SW (f ) was averaged over all segments to obtain S(f )—the count-based periodogram of the sequence. In our analysis, we set T and M to be 1 h and 2 min respectively. 4

M Shein et al

Phys. Biol. 5 (2008) 036008 1

0.6

0.4

HA

Regular

Cumulative probability

0.8

0.2

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

FS

Figure 2. Definition of HA neurons in cultured cortical networks. For an individual neuronal spike time series (T = 10 h), we computed FS—the fraction of action potentials fired by this neuron during the network’s silent state (between SBEs), relative to the total number of spikes fired by that same neuron. This procedure was repeated for every recorded time series (N = 581 from ten cultures), and all results were pooled to obtain a probability distribution of FS values across neurons. In the figure we plot the cumulative distribution, i.e. the fraction of neurons with a value equal to or less than FS. The resulting plot shows that the majority of neurons (>90%) share a FS value that is less than 0.1, meaning that they fired more than 90% of their spikes during SBEs and less than 10% of their spikes during the network’s silent states. These neurons were termed regular neurons. The rest of the neurons (those that fired more than 10% of their spikes outside SBEs) were termed highly active neurons (HA neurons).

and HA neurons. Below, we show that, in agreement with the above classification, the activity of HA neurons has additional distinguished temporal features.

Activity patterns of single neurons Different neurons exhibit different characteristic firing patterns during the same SBE (as seen in figure 1(A)), yet most of them conform to the general rule of being strongly active during the SBE and weakly active inbetween the SBEs. An exception to this rule is a small subset of recorded neurons (∼10%), which fire a considerable fraction of their action potentials outside the SBEs, during the network’s silent state. In figure 1(A), three such neurons are clearly shown (marked by asterisks on the right), as their activity exhibits a significant number of spikes inbetween the network’s synchronized events. The appearance of such highly active neurons was consistently observed in all the cortical cultures studied here (N = 10). Hence, we proceeded by characterizing the special temporal features and assessing the dynamical significance of the activity patterns of these HA neurons. For each recorded neuron, we quantified the extent of its activity outside SBEs by calculating FS—the fraction of spikes generated by the neuron during the network’s silent state (between SBEs), relative to all the action potentials recorded from this cell (i.e. during both the active and the silent network states). We used this measure to define ‘regular’ neurons as neurons for which FS < 0.1 and highly active neurons as those for which FS > 0.1. In figure 2, we show results of this analysis for 10 h neuronal spike trains (N = 581 from ten cultures). Most of the recorded neurons (91%, N = 530) were classified as ‘regular’ while approximately 9% of the recorded neurons (N = 51) were classified as highly active neurons. To be precise, we selected the threshold in FS according to the sharp decrease in the fraction of neurons as a function of FS (which occurs at F S ∼ = 0.1). This sharp decrease is reflected by the change in the slope of the cumulative distribution shown in figure 2. Such a threshold best distinguishes between ‘regular’

Statistical analysis. Inter-spike interval (ISI) distribution is often used in the analysis of neuron firing profiles (Rieke et al 1997, Koch and Segev 1998, Strong et al 1998, Segev et al 2002, Beggs and Plenz 2003, Ayali et al 2004, Zochowski and Dzakpasu 2004, Linder 2004) as it provides information about the instantaneous firing rate of a neuron. As is shown in figure 3, the ISI distribution of the HA neurons is markedly distinguished from that of the regular neurons; the interspike interval distribution of the regular neurons is typically characterized by a bimodal distribution (figure 3(A)). This structured distribution is associated with the existence of two typical time scales—the intra-SBE time scale (hundreds of milliseconds) and the inter-SBE time scale (>10 s). On the other hand, the ISI distribution of a typical highly active neuron is centered inbetween the two maxima of the distribution of the regular neurons—between the inter-SBE times and the intra-SBE inter-spike intervals (figure 3(B)). Analysis of pooled data from several recorded neurons suggested that these are the generic profiles that characterize HA neurons (figures 3(C), (D)). The putative role of HA neurons during network development The activity of HA neurons between SBEs, in the absence of global network activation, may be particularly relevant for the development of dissociated cellular assemblies into active neuronal networks (Marom and Shahaf 2002, Spitzer 2006). Therefore we took advantage of the possibility of using MEAs for long-term, non-invasive recordings, in order to track 5

M Shein et al

Phys. Biol. 5 (2008) 036008

% of intervals (Average)

% of intervals

20

20

(A )

Regular neuron

15

15

10

10

5

5

0 0 10 20 (C )

2

4

10

0 0 10 20 (D )

6

10

10

Regular neuron

15

15

10

10

5

5

0 0 10

2

10

4

0 0 10

6

10 ISI [ms]

(B )

10

HA–neuron

2

10

4

10

6

10

HA–neuron

2

10

4

10 ISI [ms]

6

10

Figure 3. HA neurons are marked by a distinct spike time series. (A) An inter-spike interval (ISI) distribution of a typical distribution of a regular neuron has a characteristic bimodal shape. The bimodality reflects the clear separation of intra-SBE (∼100 ms) and inter-SBE (>10 s) time scales. In contrast, a typical HA neuron exhibits a considerable fraction of its ISIs in the [100 ms, 10 s] regime, as is shown in (B). The averaged inter-spike interval histograms for regular neurons ((C), N = 530, ten cultures) and for HA neurons ((D), N = 51, ten cultures) suggest that the different characteristic forms of ISI distributions may be used as a pivot to discriminate between the two types of cells. Error bars represent the standard deviation of the averaged values.

very small or no electrical activity during this early stage. Since only a small fraction of the neurons are active during early development, it is practically impossible to detect the SBEs at this stage. In light of this, the method used to discriminate between HA neurons and regular neurons during early development cannot depend on SBE detection. Consequently, we distinguished between the two types of neurons according to their ISI profiles; regular neurons, in contrast to the HA neurons, rarely exhibit ISIs in the interval of [102, 104] ms (as shown in figure 3). Consequently, we defined HA neurons to be those with higher values of FI—the fraction of ISIs in the interval [102, 104]. It is worth noting that due to the variability in the ISI histogram of different neurons, the transition in FI between regular neurons and HA neurons occurs in the interval of 0.1–0.2. However, changing the threshold between 0.1 and 0.2 did not yield qualitative changes in the results. As a benchmark test for this measure, we calculated the FI values for 10 h neuronal traces of regular neurons (N = 530, ten cultures) and HA neurons (N = 51, ten cultures) which were classified using the FS measure (FS = 0.1 was used as the classification threshold). The average FI for regular neurons and HA neurons was 0.05 ± 0.06 and 0.31 ± 0.25 (mean ± STD), respectively, providing support to the validity of this measure as a tool to differentiate between regular neurons and HA neurons at the early stages of network development. We used FI to analyze the network’s electrical activity during early development. Cultures were continuously recorded for several days, starting from the second day in vitro.

HA neurons

# neuron

Regular neurons

Last

First 0

5

10 Hours

15

20

Figure 4. Initiation of electrical activity in developing cultured networks. We show here the activity raster plot of 22 neurons taken at fifth day in vitro. For clarity of presentation, the individual neuronal traces are rearranged from bottom to top according to the time of the appearance of their first spike. During the first 5 h, mostly HA-like neurons (below the red line) were active. The collective activity emerged later as short time windows of rapid neuronal firings (the SBEs), which were separated by longer intervals of the quiescence state.

the correspondence between the activity of HA neurons and regular neurons during the network’s development. Visual inspection of a typical raster plot during the network’s development (figure 4) indicates that HA neurons are more abundant in the early stages of development. Two such neurons are clearly shown in figure 4. At the same time, most of the neurons in the developing network exhibit 6

M Shein et al

Phys. Biol. 5 (2008) 036008

seen, during the first stages of development, the activity is marked by a relatively high fraction of HA neurons and this trend gradually decreases along with the intensification of the network activity as it continues to develop. These observations suggest that HA neurons may play a role in the initiation of the network’s activity.

100 % HA-neurons

(A ) 80 60 40 20 0 0

20

40 60 Time [hours]

80

The putative role of HA neurons during recovery from chemical inhibition

100

Our findings in the previous section suggest that persistent firing in single HA neurons is evoked when activity levels in the network are low. This may hint that HA neurons are recruited by the network when the network fails to sustain a certain sufficient level of electrical activity. To further test this hypothesis, we analyzed the activity of networks during prolonged chemical inhibition of excitatory synapses, which strongly reduced the total level of activity in the network. More specifically, mature networks (>12 days in vitro) showing normal levels of network activity with characteristic patterns of synchronized bursting events (as in figure 1) were recorded for several hours, after which intense chemical blockage was applied by simultaneous application of AMPA and NMDA receptor antagonists (20 μM CNQX, 100 μM APV). This process is analogous to disconnecting the excitatory coupling between the neurons, thus resetting the network to a state of activity somewhat similar to that found during early development, when synaptic connections are still sparse. Immediately (up to 10 s) following the chemical treatment, we observed a strong decline in the network’s total activity. In most cultures, the activity was completely eliminated. Gradually (on a time scale of hours), an increasing number of neurons began to exhibit firing at an increasing rate. The firing profiles of these neurons had high values of FI, characteristic of HA neurons. Most of these neurons were not identified as HA neurons before the chemical inhibition. We proceeded to compute the fraction of HA neurons (using FI > 0.2 to detect HA neurons) in the network before and after the application of chemical inhibition. As is shown in figure 5(B), following chemical inhibition, cultured networks exhibited a high fraction of HA neurons. No synchronization was observed among individual neurons throughout the whole inhibition period and the ISI distribution of these neurons remained unimodal. This suggests that degradation of blockers was not a major factor in the observed activity patterns during the inhibition. We also verified that, following the washout of the synaptic blockers, the activity of other regular neurons was restored and the fraction of the HA neurons decreased. However, the neurons which began to exhibit a HA pattern during intense chemical inhibition continued to show these patterns after the wash of the blockers. These findings, together with our previous observations on the development of networks, suggest that HA neurons may be recruited by the network to sub-serve in the initiation and maintenance of proper system-level network activity.

100 % HA-neurons

(B ) 80

CNQX +APV

Wash

60 40

Control Chem. inhibition Wash

20 0 0

20

40 Time [hours]

60

80

Figure 5. The activity of HA neurons is correlated with the homeostatic regulation of the network’s activity. (A) In the developing cultured network, the fraction of HA neurons (blue circles) is initially high, and decreases gradually during the course of the network’s development. The black line is the smoothed average over data from several experiments (N = 7). (B) Expression of HA-neuron-like activity patterns is enhanced under acute blockage of glutamateric synaptic transmission. The network’s activity was completely abolished several seconds after the application of AMPA and NMDA antagonists. In the following hours, several neurons gradually began to exhibit firing patterns which were characterized by high values of FI. This was manifested as an increase in the observed number of HA neurons. Following the wash-out of antagonists, the network’s activity began to recover along with the simultaneous decrease in the fraction of active HA neurons. The delay between the application of inhibitors and the increase in HA-neuron activity varied between different cultures; however, all three examined cultures (12–22 days in vitro) showed the same trend of HA-neuron activation.

Individual neurons began to generate action potentials after 3 days in vitro, and the number of recorded active neurons increased with time. In most cases, the first neurons to start firing also exhibited activity patterns associated with those of the HA neurons (figure 4). After this initial increase, the number of HA neurons did not change significantly over the first several days of activity. In contrast, the number of neurons showing regular patterns of activity drastically increased throughout this period. The activity of single neurons did not exhibit transitions from HA patterns to a regular pattern or vice versa over this time of measurement. To quantitatively estimate the probability of a neuron in the network to express HA patterns of activity during early development, we first divided the entire recording into nonoverlapping time windows of 1 h each. For every window, we then calculated FI for all the active neurons in that same temporal window and classified every neuron as regular (FI < 0.13) or highly active (FI  0.13). In figure 5(A), we show the fraction of HA neurons as a function of time starting from the first detected spike in the network. As can be

Time-series analysis. We continued to investigate the special dynamical characteristics of the HA neurons by performing 7

M Shein et al

Phys. Biol. 5 (2008) 036008

(A ) –1

(B )

Regular neuron

(C )

HA–neuron

–1

10

HA–neuron (Chem. inhibition)

–1

10

PSD

10

–2

–2

10

–3

10

(D ) PSD (average)

–1

10

–2

10

10

–2

–3

10

(E )

Regular neuron

(F ) –1

–3

–2

–2

10

HA–neuron (Chem. inhibition)

–2

10 10 10 Frequency [Hz]

10 10

–2

10

–3

10

HA–neuron

–1

10

–2

–2

10

10 –3

–2

10 10 Frequency [Hz]

–3

–2

10 10 Frequency [Hz]

Figure 6. Long-term trends in the activity of the HA neurons. Power spectral density (PSD) was estimated using count-based periodograms (as explained in the Methods section) on long recordings of T > 50 h from early developing cultures (6–12 day in vitro) and networks under intense chemical inhibition (12–22 day in vitro). As is shown in (A), the PSD of a regular neuron in a developing network attains relatively low values in the low-frequency regime, as compared to the PSD of HA neurons (shown in (B)). Similar to developing networks, HA neurons under intense chemical inhibition exhibit high power spectral density at low frequencies (shown in (C)). (D)–(F) The averaged periodograms for regular neurons ((D), N = 43, five cultures) for HA neurons in developing networks ((E), N = 16, five cultures) and for HA neurons under intense chemical inhibition ((F), N = 9, two cultures) suggest that the characteristic long-term trends are generic. Error bars represent the standard deviation of the averaged values.

network area. As a result, all of the recorded SBEs were initiated in the area sampled by the electrodes. These isolated networks were created by surrounding the designated area for the network with thin strips of PDMS (see the Methods section), which physically prohibited processes from being sent through the strip to the outer network. This procedure was performed in early developing networks (N = 3) up to 10 h after the onset of intense synchronized activity. Next, we examined the temporal order of neuronal activation during SBEs for HA neurons and for regular neurons. We constructed a raster plot showing the neuronal activation pattern in every SBE (figure 7(A)). A survey of these activation patterns revealed that in each SBE, different neurons are the first to fire. In order to identify temporal order patterns in neuronal activation, we averaged the spike trains of all the recorded neurons over many SBEs (spike trains were aligned according to the SBE loci). We found that some of the neurons exhibited higher probabilities to fire at the beginning of the SBEs (figure 7(B)) or act as precursors of SBEs. Next, we calculated the FS for each neuron in order to identify HA neurons. We found that HA neurons were typically the first ones to fire during SBEs (figure 7(B)). To quantify this correspondence, we identified the first neuron to fire in every SBE (as shown in figure 7(A)) and calculated the average number of SBEs initiated per HA neuron and per regular neuron. Figure 7(C) shows that the relative number of initiations per HA neuron is considerably higher than that of regular neurons, indicating the preferential appearance of HA neurons at the beginnings of SBEs. Our next aim was to rule out an artifact in which HA neurons will appear to fire first, simply because they fire

power spectral analyses of a spike time series for both regular and HA neurons in developing networks and in networks under intense chemical inhibition. The analysis was performed using the count-based periodograms (explained in the Methods section) for long (>50 h) recordings in which HA neurons were defined according to FI > 0.13 for the developing networks and FI > 0.2 for networks under chemical inhibition. In figure 6, we show that HA neurons (figures 6(B), (C), (E), (F)) have a higher power spectral density at low frequencies in comparison with regular neurons (figures 6(A), (D)). Such preferential concentration of spectral density at low frequencies is indicative of long-time auto-correlations in the firing activity of these cells. Hence, it suggests the existence of some innate mechanisms which may regulate the firing of the HA neurons. The activity of HA neurons in networks with engineered geometry So far, we have shown that HA neurons are associated with the intensification of network activity on a long time scale (hours). To investigate the putative role of HA neurons in initiation of SBEs, which occurs on a time scale of tens of milliseconds, we focused on the correspondence between the time order of neuronal firing within SBEs and the firing of HA neurons. In order to locate the initiation sites of SBEs, we created small-scale networks with engineered geometry. The networks’ growth was restricted to the area covered by the electrodes, thus allowing uniform sampling of the entire 8

M Shein et al

Phys. Biol. 5 (2008) 036008

# Neuron

1

Investigating the response to electrical stimulations 1 (B )

(A )

1.8

Time [sec]

(C )

# Neuron

17 0

% initiation per neuron

To answer the above question, we investigated the response of neurons in the network to external electrical stimulations. More specifically, we examined the differences in triggering and response between electrodes recording from HA neurons to electrodes recording from regular neurons. For this purpose, we first identified the HA neurons in the network and afterwards stimulated all the electrodes in which neuronal activity was recorded (see the Methods section). In general, some of the stimulations did not elicit any response in the network, other stimulations triggered a response in individual neurons and a small fraction of the stimulations activated a synchronized burst in the network (figure 8(A)). The responses varied between the different stimulating electrodes and different stimulated neurons (figures 8(A), (D)). More specifically, some of the neurons consistently responded to the stimulations with a high yield while others were unresponsive (figure 8(B)). Additionally, some of the stimulating electrodes consistently activated more neurons and were more likely to trigger SBEs in the network while others had no influence on the network’s activity (figure 8(C)). In these tests, we did not identify correspondence between the electrodes whose stimulation triggered activity in individual neurons or the whole network and electrodes on which HA neurons (defined by FS > 0.1) were identified (figures 8(C), (D)). On the other hand, we found that HA neurons responded with higher fidelity to electrical stimulations in comparison to regular neurons (figures 8(B), (E)). Electrical stimulations using a multi-electrode array may trigger a response in more than one neuron. Additionally, passing by axons may also be triggered by the electrical stimulations (Wagenaar et al 2004). Consequently, the high responsiveness of HA neurons suggests that they are coupled to many physical locations in the network and that they are sensitive enough to be affected by these pathways. This suggests that the specific HA neurons we measured are not necessarily the initiators of SBEs, but rather act as precursors to the activation of collective events in the network.

HAneurons

Regular neurons

Regular neurons HA-neurons

100

50

0

1

2 Culture

3

24 0

HAneurons

Time [sec]

1

Figure 7. Correspondence between HA neurons and the order of neuronal activation during the SBEs. (A) A spatio-temporal firing pattern of a typical synchronized bursting event reveals that each SBE is characterized by a distinct pattern of neuronal activation. The neuron which fired the first action potential is defined as the precursor of this SBE. The HA neurons in this example (marked by arrows) clearly fire ahead of the other regular neurons. (B) Activity profiles of consecutive SBEs (N = 540) from a developing network were co-aligned and averaged according to the SBE loci, as defined in figure 2. The neurons were vertically ordered according to their FS values, so that HA neurons are at the bottom of the plot. This analysis disclosed that HA neurons (below the red line corresponding to FS = 0.1) expressed higher probabilities to be the first to fire in the SBEs. (C) The normalized (see text for details) percentage of SBEs in which HA neurons (black bars) were the first to fire is considerably higher than the normalized fraction of SBEs in which regular neurons (white bars) fired first. The results are consistent for all three examined cultures (1–3), in which HA neurons initiate on average 19 times more SBEs than regular neurons.

more intensely outside the SBEs. To rule out this possibility, we first created an artificial SBE sequence with the same average frequency as appears in the original data but with a different shuffling of the neurons for each SBE (HA neurons were deleted from this sequence). We then superimposed the original spike time series of the HA neurons on the time series of shuffled SBEs and tried to identify initiating HA neurons in the artificially constructed data set. The relative fraction of SBEs initiated per HA neurons was much lower in the shuffled scenarios as compared to the real data sets (0.94 on average in the original sets versus 0.07 after shuffling). These results indicate that the correspondence between HA neurons and the preferential firing at the beginning of SBEs is not an artifact, and that HA neurons act as precursors of SBEs. It should be noted that since the average fraction of HA neurons is small, it is quite likely that all SBEs are preceded by the firing of some HA neurons, but that due to their small fraction, these HA neurons are not recorded. These results raise the immediate question of whether HA neurons ignite the SBEs or whether their preferential activation in the beginning of SBEs is simply due to the fact that they are the first to respond.

Discussion In this work, we investigated the dynamics of HA neurons, identified in the spontaneous activity of cortical neuronal networks. We studied their dynamics in mature networks, in early developing networks, during recovery from intense chemical inhibition, in networks with engineered geometry and in response to electrical stimulations. Our main findings are as follows. (1) Activity patterns of single neurons in cortical cultures can be roughly classified as belonging to either one of two classes—regular patterns relating to neurons which fire mainly during the SBEs and highly active patterns relating to a small subset of neurons (HA neurons), which also show persistent firing during the network’s silent states (between SBEs). (2) The firing profiles of the HA neurons have special dynamical characteristics. 9

M Shein et al

Phys. Biol. 5 (2008) 036008

(A )

(B )

5

Regular Neurons

10 Electrode

(C )

HA Neurons

15 20 25

*

125

5 10 % Responding electrodes

(E )

*

100 * *

0.5

* 10

0.3

15 * 20

0.2 0.1

* * 5

10 15 20 Stimulating electrode

25

0

Probability

Responding electrode

0.4

% response per neuron

5

25

10 20 % Stimulated SBEs

80 60

HA

*

100

40 20 0

Reg

*

75 Time [sec]

HA

(D ) **

50

Reg

25

1

2 Culture

Figure 8. HA neurons and electrical stimulation. (A) An example of a raster plot from a network responding to electrical stimulations (blue lines). Some of the stimulations did not elicit any response while others activated individual neurons or the whole network. In most cases, SBEs were triggered in the first stimulation of a sub-session (see the Methods section for electrical session protocol). (B) The probability of response for different electrodes. For each stimulation, electrodes were marked as responding if they were triggered during the first 100 ms after the stimulus. The probability of response was then averaged over all stimulations in all electrodes. HA neurons (black) were triggered by the electrical stimulations with a higher probability compared to other neurons. The red line marks the average level. (C) The probability of different electrodes to successfully trigger a SBE in the network. A successful stimulation in an electrode was defined as one in which an SBE was detected up to 500 ms after stimulus. No significant correspondence was found between the stimulated electrodes which successfully triggered SBEs and electrodes on which HA neurons were identified (black). The red line marks the average level. (D) The response matrix, R(i, j ). Electrode i was marked as responding to stimulation in electrode j, if an action potential was recorded in electrode i during the first 100 ms after the stimulus was applied to electrode j. The response probability of every electrode to all stimulations was averaged and is presented in a gray-scale color code. Some electrodes consistently triggered responses in the network and other electrodes consistently responded to stimulations. There was a strong correspondence between electrodes on which HA neurons were detected (marked by asterisks and outlined by rectangles) and those which responded to stimulations but not those which triggered activity in the network. (E) The relative percentage of response, per HA neuron and per regular neuron, was calculated from the response matrix for two stimulated cultures (8–10 days in vitro). The results indicate that HA neurons are ∼4 times more responsive to electrical stimulations than regular neurons.

(3) The HA neurons seem to play an important role in the maintenance of spontaneous activity during development and under inhibition conditions. (4) The fraction of HA neurons is regulated according to the context of the network activity. (5) The HA neurons act as precursors of the SBEs but do not seem to directly ignite them. (6) HA neurons respond with higher fidelity to external electrical stimulations.

The temporal structure of a neuronal firing time series can be affected by many factors, ranging from the morphology of its dendritic tree (Mainen and Sejnowski 1996) to the selective regulation of its inward and outward current densities (Turrigiano et al 1995). In the cortical cultures studied here, we identified two qualitatively different patterns of individual neuronal activities. Most neurons were active almost exclusively during SBEs, firing only a small number of occasional spikes during the silent network state (i.e. the long time intervals between the SBEs). Most neurons have their own firing profile during the SBEs (Segev et al 2004, Raichman et al 2006), but, in general, the activity of these

Below we discuss in further detail our findings and their possible implication for understanding the regulation of the network’s electrical activity. 10

M Shein et al

Phys. Biol. 5 (2008) 036008

In order for the network to employ the HA neurons, there should be a mechanism that reduces the activity of the HA neurons when they receive high electrical input and vice versa. The mechanism proposed in Volman et al (2007) presents such a putative mechanism. It was shown that a self-synapse which is regulated by astrocytes can reduce the auto-transfer of action potentials when the network is active. From the single cell perspective, another putative mechanism may be based on activity-dependent redistribution of voltage-gated conductance (Desai et al 1999, Moody and Bosma 2005). Both mechanisms allow the regulated increase in neuronal excitability as observed for HA neurons. Such an increase in excitability complies with the observation that HA neurons are precursors of SBEs and are more responsive to external electrical stimulations. We now turn to reflect on the possible mechanisms which may underlie the spontaneous generation of the network’s collective activity. This activity is marked by the generation of synchronized bursting events, which start to appear in the spontaneous activity as the dissociated cell culture develops into a fully connected network. Several studies suggest that SBEs may be driven by a sub-population of neurons (Mao et al 2001, Voigt et al 2001, Darbon et al 2002, Sipila et al 2005, Hunt et al 2005, Ham et al 2008). In a recent study, Feinerman et al (2007) showed that in quasi-1D cortical cultures there exist ‘hot spots’ that reliably generate SBEs. In addition, focused electrical and chemical stimulation can reliably induce new SBEs which persist for long periods of time (Baruchi and Ben-Jacob 2007, Madhavan et al 2007). These findings suggest that the generation of SBEs may depend on a sub-network (sub-population) composed of a small number of neurons. An alternative hypothesis, which is supported by the recent studies of Eytan and Marom (2006), posits that spontaneously generated SBEs are an outcome of population-level interactions rather than being dependent on specific cells. In this view, neurons which initiate SBEs form a sub-population which is recruited in response to collective events, before the rest of the network. According to our observations, the same set of neurons which show increased activity when the global activity levels in the network are low appear to fire first during SBEs. This implies a link between the long-term homeostatic mechanisms and the short-term activity propagation in the network. This link is compatible with both the views of Eytan and Marom (2006) and Feinerman et al (2007). According to the former view, HA neurons can be looked upon as highly excitable neurons which are more sensitive to fluctuations in the network’s activity and whose sensitivity can be intrinsically regulated. For this reason, they are persistently triggered between SBEs, are the first neurons to fire during SBEs, are more responsive to electrical stimulations and are more active when the network’s activity is low. According to the latter view, HA neurons can be viewed as having intrinsic firing properties which ‘boost’ the activity whenever the network’s activity levels are low. For this reason, in a quazi-1D network, such as the one presented in Feinerman et al (2007), HA neurons can create localized initiation zones from which SBEs are triggered. Our observations may also support a new possibility that bridges between the above two views, based on the

neurons correlates well with the overall network activity. In addition to this class of regular neurons, whose activity is ‘network dependent’, a relatively small subset of recorded cells also exhibit persistent firing during the network’s silent states between SBEs. Such generic classification of neurons, based on the correspondence between the activity of a single neuron and the activity of the network, is consistent with previous findings (Mao et al 2001, Darbon et al 2002, Sipila et al 2005). At present, it is not clear which biophysical mechanisms are responsible for the appearance of the HA neurons. A putative mechanism based on the idea of self-synapses regulated by astrocytes was proposed by Volman et al (2007). Generally, neuronal activity can be affected by either cellextrinsic or cell-intrinsic factors. Two pieces of experimental evidence stand in favor of the second scenario. First, the HA neurons are active long before the network is fully connected and showing global patterns of collective activity. Second, the firing activity of these neurons persists during the absence of the network’s activity (the network’s silent states), as is also the case when the network undergoes intense chemical inhibition. Therefore, it appears that an input of action potentials from other neurons is not necessary in order for the HA neurons to be active, and thus the mechanism seems to be cell-intrinsic. It should be noted that even though our studies show that the HA neurons fire given limited or no electrical input (action potential signals from other cells), it does not mean that they do not receive chemical input from other cells such as astrocytes and microglia, which is likely to be the case. Mechanistically, the appearance of persistent firing with limited or no electrical input is reminiscent of pacemaker cell activity (Sipila et al 2005). However, in our opinion, the observed wide distribution of inter-spike intervals makes it unlikely that the HA neurons are simple pacemakers. It is also possible that more biophysically complicated processes take place, such as intrinsic bursting, driven by dynamical bi-stability (Loewenstein et al 2005), that may be triggered by external chemical inputs. Clearly, more detailed neurophysiological investigations of the HA neurons are needed in order to resolve this issue. Such investigations will also lead to a better understanding of the role that these special neurons play in the network’s activity. Our present findings suggest that the activity of the HA neurons is prominent when the overall network activity is quite weak (as was the case in early developmental stages and when the network underwent chemical inhibition). We also observed a decrease in the fraction of HA neurons as the network became fully active. These observations lead us to suggest that the HA neurons may be employed by the network as a means to maintain activity homeostasis. It is possible that in the case of early developing networks, HA firing patterns may appear due to some developmental constraint; however, our observation that HA firing patterns appear after intense chemical inhibition in mature networks supports the notion that these patterns appear due to the lack of activity. Additional support for this hypothesis is found in an observation of longrange (minutes to hours) auto-correlations in the activity of HA neurons. Memory on such long time scales may imply the involvement of homeostatic processes. 11

M Shein et al

Phys. Biol. 5 (2008) 036008

following reasoning: combined experimental and modeling studies (Segev et al 2004, Baruchi and Ben-Jacob 2004, 2007, Volman et al 2005, Baruchi et al 2006, 2008, Raichman and Ben-Jacob 2008) indicate that large (>1 M neurons) cultured cortical networks can sustain several SBEs, which differ from one another in their spatio-temporal patterns of neuronal activation. The ability of a network to exhibit different bursting events can be accounted for if the network is composed of overlapping sub-networks (this idea was discussed in Segev et al (2004), Baruchi and Ben-Jacob (2004)). To embrace these findings, while taking into account the observations reported here regarding the HA neurons, we follow the assumption that cortical cultures may be sub-divided into overlapping subnetworks and add the idea that the activity of each of the subnetworks is mainly regulated by its own set of HA neurons. In this view, spontaneous collective activity (as manifested by different SBEs) may still be generated by intrinsically active cells (in accord with experimental observations as in Sipila et al (2005), Feinerman et al (2007)). At the same time, activity feedback from a sub-network determines the propensity of its associated set of highly active cells to exhibit persistent firing. Therefore, synchronized bursting events may arise due to the synergistic action of cell assemblies and their associated sets of HA neurons. This perspective is additionally supported by the observation (data not shown) that in some cases, the activity of HA neurons during the network’s silent states is correlated. Future experiments with patterned cultured networks will delineate the interplay between highly active cells and the network’s architecture of synaptic connections. Meanwhile, the results of our analysis suggest that HA neurons may constitute an important pathway recruited by a developing cortical network in order to regulate its spontaneous activity.

Baruchi I, Volman V, Raichman R, Shein M and Ben-Jacob E 2008 The emergence and properties of mutual synchronization in in-vitro coupled cortical networks Eur. J Neurosci. at press Beggs J M and Plenz D 2003 Neuronal avalanches in neocortical circuits J. Neurosci. 23 11167–77 Beggs J M and Plenz D 2004 Neuronal avalanches are diverse and precise activity patterns that are stable for many hours in cortical slice cultures J. Neurosci. 24 5216–29 Ben-Ari Y 2001 Developing networks play a similar melody Trends Neurosci. 24 353–60 Bi G and Poo M 2001 Synaptic modification by correlated activity: Hebb’s postulate revisited Ann. Rev. Neurosci. 24 139–66 Darbon P, Scicluna L, Tscherter A and Streit J 2002 Mechanisms controlling bursting activity induced by disinhibition in spinal cord networks Eur. J. Neurosci. 15 671–83 Desai N S, Rutherford L C and Turrigiano G G 1999 Plasticity in the intrinsic excitability of cortical pyramidal neurons Nat. Neurosci. 2 515–20 Eytan D and Marom S 2006 Dynamics and effective topology underlying synchronization in networks of cortical neurons J. Neurosci. 26 8465–76 Feinerman O, Segal M and Moses E 2007 Identification and dynamics of spontaneous burst initiation zones in uni-dimensional neuronal cultures J. Neurophysiol. 97 2937–48 Feller M B 1999 Spontaneous correlated activity in developing neural circuits Neuron 22 653–6 Garaschuk O, Linn J, Eilers J and Konnerth A 2000 Large-scale oscillatory calcium waves in the immature cortex Nat. Neurosci. 3 452–9 Ham M I, Bettencourt L M, McDaniel F D and Gross G W 2008 Spontaneous coordinated activity in cultured networks: analysis of multiple ignition sites, primary circuits, and burst phase delay distributions J. Comput. Neurosci. 24 346–57 Hanson M G and Landmesser L T 2004 Normal patterns of spontaneous activity are required for correct motor axon guidance and the expression of specific guidance molecules Neuron 43 687–701 Hebb D O 1949 The Organization of Behavior vol 1 (New York: Wiley) Hua J Y and Smith S J 2004 Neural activity and the dynamics of central nervous system development Nat. Neurosci. 7 327–32 Hulata E, Baruchi I, Segev R, Shapira Y and Ben-Jacob E 2004 Self-regulated complexity in cultured neuronal networks Phys. Rev. Lett. 92 198105(1)–198105(4) Hulata E, Segev R, Shapira Y, Benveniste M and Ben-Jacob E 2000 Detection and sorting of neural spikes using wavelet packets Phys. Rev. Lett. 85 4637–40 Hunt P N, McCabe A K and Bosma M M 2005 Midline serotonergic neurones contribute to widespread synchronized activity in embryonic mouse hindbrain J. Physiol. 566 807–19 Kandel E R, Schwarz J H and Jessell T M 2000 Principles of Neural Science vol 1 4th edn (New York: McGraw-Hill) Kane R S, Takayama S, Ostuni E, Ingber D E and Whitesides G M 1999 Patterning proteins and cells using soft lithography Biomaterials 20 2363–76 Katz L C and Shatz C J 1996 Synaptic activity and the construction of cortical circuits Science 274 1133–8 Khazipov R, Esclapez M, Caillard O, Bernard C, Khalilov I, Tyzio R, Hirsch J, Dzhala V, Berger B and Ben-Ari Y 2001 Early development of neuronal activity in the primate hippocampus in utero J. Neurosci. 21 9770–81 Koch C and Segev I 1998 Methods in Neuronal Modeling: From Ions to Networks vol 1 2nd edn (Cambridge, MA: MIT Press) Komuro H and Kumada T 2005 Ca2+ transients control cns neuronal migration Cell Calcium 37 387–93 Linder B 2004 Interspike interval statistics of neurons driven by colored noise Phys. Rev. E 69 022901 Loewenstein Y, Mahon S, Chadderton P, Kitamura K, Sompolinsky H, Yarom Y and Hausser M 2005 Bistability of

Acknowledgments The authors thank I Baruchi for sharing his views with us and I Brainis for technical assistance. This research was supported in part by the NSF-sponsored Center for Theoretical Biological Physics (grant numbers PHY-0216576 and PHY-0225630), by the Israeli Science Foundation and by the Tauber fund at TelAviv University. VV acknowledges the support of the US National Science Foundation I2CAM International Materials Institute, grant DMR-0645461.

References Adrian E D 1928 The Basis of Sensation: The Action of the Sense Organs vol 1 (London: Christophers) Ayali A, Zilberstein Y, Robinson A, Shefi O, Hulata E, Baruchi I and Ben-Jacob E 2004 Contextual regularity and complexity of neuronal activity: from stand-alone cultures to task-performing animals Complexity 9 25–32 Baruchi I and Ben-Jacob E 2004 Functional holography of recorded neuronal networks activity Neuroinformatics 2 333–52 Baruchi I, Grossman D, Volman V, Shein M, Hunter J, Towle V L and Ben-Jacob E 2006 Functional holography analysis: simplifying the complexity of dynamical networks Chaos 16 1054–500 Baruchi I and Ben-Jacob E 2007 Towards neuro-memory-chip: imprinting multiple memories in cultured neural networks Phys. Rev. E 75 1539–3755 12

M Shein et al

Phys. Biol. 5 (2008) 036008

cerebellar Purkinje cells modulated by sensory stimulation Nat. Neurosci. 8 202–11 Lohmann C, Myhr K L and Wong R O 2002 Transmitter-evoked local calcium release stabilizes developing dendrites Nature 418 177–81 Lowen S B, Ozaki T, Kaplan E, Saleh B E and Teich M C 2001 Fractal features of dark, maintained, and driven neural discharges in the cat visual system Methods 24 377–94 Madhavan R, Chao Z C and Potter S M 2007 Plasticity of recurring spatio-temporal activity patterns in cortical networks Phys. Biol. 4 181–93 Maeda E, Robinson H P and Kawana A 1995 The mechanisms of generation and propagation of synchronized bursting in developing networks of cortical neurons J. Neurosci. 15 6834–45 Mainen Z F and Sejnowski T J 1996 Influence of dendritic structure on firing pattern in model neocortical neurons Nature 382 363–6 Mao B Q, Hamzei-Sichani F, Aronov D, Froemke R C and Yuste R 2001 Dynamics of spontaneous activity in neocortical slices Neuron 32 883–98 Marder E and Goaillard J M 2006 Variability, compensation and homeostasis in neuron and network function Nat. Rev. Neurosci. 7 563–74 Marom S and Shahaf G 2002 Development, learning and memory in large random networks of cortical neurons: lessons beyond anatomy Q. Rev. Biophys. 35 63–87 Menendez de la Prida L, Bolea S and Sanchez-Andres J V 1998 Origin of the synchronized network activity in the rabbit developing hippocampus Eur. J. Neurosci. 10 899–906 Moody W J and Bosma M M 2005 Ion channel development, spontaneous activity, and activity-dependent development in nerve and muscle cells Physiol. Rev. 85 883–941 Mooney R, Madison D V and Shatz C J 1993 Enhancement of transmission at the developing retinogeniculate synapse Neuron 10 815–25 Palva J M, Lamsa K, Lauri S E, Rauvala H, Kaila K and Taira T 2000 Fast network oscillations in the newborn rat hippocampus in vitro J. Neurosci. 20 1170–8 Potter S M 2001 Distributed processing in cultured neuronal networks Prog. Brain Res. 130 49–62 Raichman N and Ben-Jacob E 2008 Identifying repeating motifs in the activation of synchronized bursts in cultured neuronal networks J. Neurosci. Methods 170 96–110 Raichman N, Volman V and Ben-Jacob E 2006 Collective plasticity and individual stability in cultured neuronal networks Neurocomputing 69 1150–4 Rieke F, Warland D, van Steveninck R and Bialek W 1997 Spikes: Exploring the Neural Code vol 1 (Cambridge, MA: MIT Press) Rubinsky L, Raichman N, Baruchi I, Shein M, Lavee J, Frenk H and Ben-Jacob E 2007 Study of hypothermia on cultured neuronal networks using multi-electrode arrays J. Neurosci. Methods 160 288–93 Segev R, Baruchi I, Hulata E and Ben-Jacob E 2004 Hidden neuronal correlations in cultured networks Phys. Rev. Lett. 92 118102 Segev R, Benveniste M, Hulata E, Cohen N, Palevski A, Kapon E, Shapira Y and Ben-Jacob E 2002 Long term behavior of lithographically prepared in vitro neuronal networks Phys. Rev. Lett. 88 118102 Segev R, Shapira Y, Benveniste M and Ben-Jacob E 2001 Observations and modeling of synchronized bursting in two-dimensional neural networks Phys. Rev. E 64 011920

Sejnowski T J and Paulsen O 2006 Network oscillations: emerging computational principles J. Neurosci. 26 1673–6 Sipila S T, Huttu K, Soltesz I, Voipio J and Kaila K 2005 Depolarizing GABA acts on intrinsically bursting pyramidal neurons to drive giant depolarizing potentials in the immature hippocampus J. Neurosci. 25 5280–9 Sorkin R, Gabay T, Blinder P, Baranes D, Ben-Jacob E and Hanein Y 2006 Compact self-wiring in cultured neural networks J. Neural Eng. 3 95–101 Spitzer N C 1995 Spontaneous activity: functions of calcium transients in neuronal differentiation Perspect. Dev. Neurobiol. 2 379–86 Spitzer N C 2006 Electrical activity in early neuronal development Nature 444 707–12 Streit J, Tscherter A, Heuschkel M O and Renaud P 2001 The generation of rhythmic activity in dissociated cultures of rat spinal cord Eur. J. Neurosci. 14 191–202 Strong S P, Koberle R, van Steveninck R and Bialek W 1998 Entropy and information in neural spike trains Phys. Rev. Lett. 80 197–200 Tscherter A, Heuschkel M O, Renaud P and Streit J 2001 Spatiotemporal characterization of rhythmic activity in rat spinal cord slice cultures Eur. J. Neurosci. 14 179–90 Turrigiano G, LeMasson G and Marder E 1995 Selective regulation of current densities underlies spontaneous changes in the activity of cultured neurons J. Neurosci. 15 3640–52 Turrigiano G G 1999 Homeostatic plasticity in neuronal networks: the more things change, the more they stay the same Trends Neurosci. 22 221–7 van Rossum M C, Bi G Q and Turrigiano G G 2000 Stable Hebbian learning from spike timing-dependent plasticity J. Neurosci. 20 8812–21 Voigt T, Opitz T and de Lima A D 2001 Synchronous oscillatory activity in immature cortical network is driven by GABAergic preplate neurons J. Neurosci. 21 8895–905 Volman V, Baruchi I and Ben-Jacob E 2004a Self-regulated homoclinic chaos in neural networks activity Proc. 8th Experimental Chaos AIP Conf. vol 742 pp 197–209 Volman V, Baruchi I, Persi E and Ben-Jacob E 2004b Generative modelling of regulated dynamical behavior in cultured neuronal networks Physica A 335 249–78 Volman V, Baruchi I and Ben-Jacob E 2005 Manifestation of function-follow-form in cultured neuronal networks Phys. Biol. 2 98–110 Volman V, Ben-Jacob E and Levine H 2007 The astrocyte as a gatekeeper of synaptic information transfer Neural Comput. 19 303–26 Wagenaar D A, Pine J and Potter S M 2004 Effective parameters for stimulation of dissociated cultures using multi-electrode arrays J. Neurosci. Methods 138 27–37 Wagenaar D A, Pine J and Potter S M 2006 An extremely rich repertoire of bursting patterns during the development of cortical cultures BMC Neurosci. 7 1471–2202 Yuste R 1997 Introduction: spontaneous activity in the developing central nervous system Semin. Cell Dev. Biol. 8 1–4 Zhang L I and Poo M M 2001 Electrical activity and development of neural circuits Nat. Neurosci. 4 1207–14 (Suppl) Zochowski M and Dzakpasu R 2004 Conditional entropies, phase synchronization and changes in the directionality of information flow in neural systems J. Phys. A: Math. Gen. 37 3823–34

13