ARTICLE IN PRESSNeural Networks ( ) –

Contents lists available at ScienceDirect

Neural Networks

journal homepage: www.elsevier.com/locate/neunet

Spatio-temporal motifs ‘remembered’ in neuronal networks followingprofound hypothermiaLiel Rubinsky a,∗, Nadav Raichman b, Jacob Lavee c, Hanan Frenk a, Eshel Ben-Jacob b

a Department of Psychology, Tel-Aviv University, Tel-Aviv 69978, Israelb School of Physics and Astronomy, Raymond & Beverly Sackler Faculty of Exact Sciences, Tel-Aviv University, Tel-Aviv 69978, Israelc Department of Cardiac Surgery, Sheba Medical Center, Tel Hashomer 52621, Israel

a r t i c l e i n f o

Article history:Received 1 November 2007Received in revised form6 March 2008Accepted 14 June 2008

Keywords:Low temperatureNeuronal culturesMicroelectrode arraysSynchronized bursting events

a b s t r a c t

Surgical procedures using hypothermic temperatures have been linked to complications such as seizures,impairedmental development and impairedmemory. Although there is some evidence that the profoundhypothermia (<12 ◦C) used in these procedures may be contributing to these neurological impairments,skepticism remains because of lack of evidence from experimental studies isolating the effects ofhypothermia on neuronal networks. In order to attain a better understanding of profound hypothermiaeffects on neurons during surgical procedures, we applied cold to a cultured in-vitro neuronal network.The typical pattern of activity of such cultures is in the form of synchronized bursts, in which most ofthe recorded neurons fire action potentials in a short time period. In most cases, the bursting activityshows one or more repeating precise spatio-temporal patterns (motifs) that are sustained over longperiods of time. In this experimental study, neuronal networks grown on microelectrode arrays (MEA)are subjected to profound hypothermia for an hour and the collective dynamics of the network as awhole are assessed. We show, by using a similarity analysis that compares changes in the time delaysbetween neuronal activation at different burst motifs, that neuronal networks survive total inhibitionby profound hypothermia and retain their intrinsic synchronized burst motifs even with substantialgeneralized neuronal degeneration. By applying multiple sessions of cold, we also show a markedmonotonic reduction in the rate of burst firing and in the number of spikes of each neuron after eachsession.

© 2008 Elsevier Ltd. All rights reserved.

1. Introduction

Low temperatures are a common tool used to protect the brainduring traumatic injury, patient recovery and surgical procedures(Kumral et al., 2001; McIntyre, Fergusson, Hebert, Moher, &Hutchison, 2003; Sessler, 2001). Their specific therapeutic effectsin reducing cerebral injury associated with hypoxia, ischemia,inflammatory and traumatic injury are well documented (Barone,Feuerstein, & White, 1997; Muller et al., 2004).

In particular, profound hypothermia (temperatures below15 ◦C) have played a preferential role in protection of the brainduring many cardiovascular surgical procedures (Deleon et al.,1998; Gillinov et al., 1993). However, there is a controversy asto the possible detrimental effects with the use of profoundhypothermia on cortical functions following these surgeries

∗ Corresponding address: School of Physics and Astronomy, Raymond & BeverlySackler Faculty of Exact Sciences, Tel-Aviv University, Tel-Aviv 69978, Israel.Tel.: +972 3 640 7604; fax: +972 3 642578.

E-mail address: lielrubinsky@yahoo.com (L. Rubinsky).

(Deleon et al., 1998). Although many studies validate neuro-protection by profound hypothermia during surgical situations(Gillinov et al., 1993; Haka-Ikse, Blackwood, & Steward, 1978;Sailhamer et al., 2007), other studies show significant neurologicalcomplications such as impaired mental and memory developmentin patients undergoing surgeries using such low temperatures(Deleon et al., 1998; Khaladj et al., 2006). Several studies link theseneurological complications to swelling, cytoplasmic voculationand cytoskeletal disruption of the neurons caused by hypothermia(Emery & Lucas, 1995).

Nevertheless, identifying profound hypothermia’s effects onneurological functions in-vivo has been difficult (Deleon et al.,1998), suggesting a need for a better controlled experimentalsystem that allows the isolation of hypothermia as a variable.

In order to overcome these difficulties and assess the effectof profound hypothermia on neurons and neuronal networks,we have developed an in-vitro system that allows the study ofthe effects of low temperature on neuronal cultures (Rubinskyet al., 2007). In our experimental setup we cultivate neuronalnetworks on microelectrode arrays (MEA) and subject them tohypothermic parameters. MEAs enable high-resolution temporal

0893-6080/$ – see front matter© 2008 Elsevier Ltd. All rights reserved.doi:10.1016/j.neunet.2008.06.008

Please cite this article in press as: Rubinsky, L., et al. Spatio-temporal motifs ‘remembered’ in neuronal networks following profound hypothermia. Neural Networks(2008), doi:10.1016/j.neunet.2008.06.008

ARTICLE IN PRESS2 L. Rubinsky et al. / Neural Networks ( ) –

recordings of external membrane potentials of several dozen cellssimultaneously. Using methods based on similarity and clusteringanalysis (Raichman & Ben-Jacob, 2008), we are able to characterizethe spatio-temporal relations between neurons and betweenregions in the network, which allow for an understanding of thecollective dynamics of the network as a whole (Segev et al., 2002;Segev, Baruchi, Hulata, & Ben-Jacob, 2004).

Thus the current experiment was undertaken to assess theeffects of profound hypothermia (∼12 ◦C) on neurons andneuronal networks in an experimental study similar to those foundin surgical settings. By applying our methodology we are able toisolate the effects of profound hypothermia on an in-vitro neuronalculture by comparing the intensity, synchronized bursting patterns(motifs) and culture state before, during and after hypothermia.Herewe report on the neuronal and network activity following onehour of profound hypothermia and subsequent cooling sessions.We show thatmultiple sessions of cold aremarked by amonotonicreduction in the rate of burst firing and in the number of spikesof each neuron after each session. In addition we demonstratethe preservation of initial synchronized burst motifs even with ageneralized neuronal degeneration.

2. Experimental procedures

2.1. Network preparation

The neuronal networks used in this study were preparedfrom dissociated cortical cultures of neurons and glia of SpragueDawley rats in their 18 embryonic day (E18), prepared andmaintained according to the protocol described in Segev et al.(2002) and grown on a microelectrode array (Multi ChannelSystems, Germany). The MEA consists of 60 microelectrodes thatare 30 µm in diameter and are arranged in a square grid with adistance of 200 µm between electrodes. Before plating, the MEAis first coated with a poly-d-lysine (PDL, Sigma p-7886, Israel)substrate and left overnight. A well for the liquid medium isconstructed by gluing a glass cylinder with a diameter of 2 cm tothe MEA surface surrounding the recording cites. The neurons arethen plated into the well at a density of 2 × 106 cells/electrode onthe PDL substrate. The culture is maintained in a culture solutionof growth medium containing 25 ml Horse serum (Beith Haemek,Israel), 100 uL Gentamycin (Beith Haemek, Israel), and 10 mlGlucose 1 M to 464 ml of MEM (Beith Haemek, Israel) to whichwas added 125 uL of Glutamine (Beith Haemek, Israel). The cultureis kept in an incubator at 37 ◦C with 5% CO2 and 95% humidity.Cultures are incubated for a minimum of 10 days to mature intoself-organized connected and active neuronal networks. Twice aweek 1 ml of medium is replaced.

2.2. Data acquisition

The electrical signals recorded from the neurons are local actionpotentials from cells that have formed capacitive couplingwith themicroelectrodes. At 10 days in-vitro, the network is placed in anexternal recording chamber that maintains the same physiologicalconditions as those found in the incubator. The network electricalactivity is non-invasively recorded by a set of 60 amplifiers (Multi-Channel Systems, Germany) with a sampling rate of 12 kHzand transferred to a computer and saved to disk using AlphaMap data acquisition software (Alpha Omega Engineering, Israel).Spike detecting and sorting of the recorded action potentialswere performed by our wavelet packets decomposition method asdescribed in Hulata, Segev, and Ben-Jacob (2002).

Table 1Similarity values of SBE motifs from 5 cultures

Culture Session Similarity, mean (stdev) Cooling/Control (%)Control After cooling

1 1 0.76 (0.08) 0.97 (0.04) 1282 0.65 (0.12) 0.68 (0.09) 1053 0.57 (0.17) 0.51 (0.11) 88

2 1 0.72 (0.10) 0.31 (0.14) 433 1 0.20 (0.10) 0.88 (0.19) 437a

4 1 0.98 (0.02) 0.71 (0.14) 735 1 0.37 (0.18) 0.39 (0.13) 105

2 0.55 (0.08) 0.83 (0.15) 1513 0.90 (0.12) 0.99 (0.01) 111All: 0.64 (0.24) 0.70 (0.25) 100 (33)

For each cooling session, the similarity was computed between SBEs of the mostprominentmotif, taken from 1 h before and 5 h after application of cold. For control,we computed the similarity of the same motif between SBEs from 6 h before theapplication of cold and 1 h before. In two cultures (1 and 5) we applied multiplesessions of cold.

a This value was excluded from the calculation of the mean of all sessions.

2.3. Profound hypothermia application

Before application of profound hypothermia, the neuronal ac-tivity was recorded for 10 h under normal physiological growthconditions in order to establish a base line for normal neuronalactivity. The physiological atmosphere was maintained and con-tinuously supplied to the external recording chamber through-out the entire experiment. The recording was continued whileprofound hypothermia was applied by pumping chilled waterstored in a container submerged in a Neslab RTE-221 coolingbath (Neslab RTE-221, Thermo Electron Corporation, USA) throughchannels in the walls of the recording chamber. This effectivelyreduced the temperature to 12 ± 1 ◦C. Temperature was verifiedusing a thermocouple-measuring device (Digi-Sense Thermome-ter, Cole Parmer, USA). The thermocouplemeasurements indicatedthat cooling of the chamber began immediately and the designatedtemperature was achieved within 5 min. The cultures were main-tained in this hypothermic state for 1 h.

Following 1 h of profound hypothermia the cooling systemwasshut down and the temperature was returned to 37 ◦C by circulat-ing warm water through the walls of the recording chamber. Theculture activity was continuously recorded afterwards for a mini-mum of 10 h. Thermocouplemeasurements showed that the phys-iological temperature was achieved within 5 min.

We applied the above procedures on a total of 6 cultures. Ofthese, 2 of the networks were subjugated to repeated exposuresof profound hypothermia with a 1–2 day time period betweensessions. Thiswas done to assess the cumulative effects of repeatedhypothermia application on neuronal networks.

2.4. Synchronized bursting activity of in-vitro cultures

After 10 days in-vitro, the culture forms a dense interconnectednetwork that exhibits rich spontaneous dynamical behaviorcharacterized by the formation of synchronized bursting events(SBEs). SBEs are short time windows (200–1000 ms) during whichmost of the recorded neurons participate in relatively rapid firingof action potentials (Fig. 1A) (Segev et al., 2002). The SBEs areseparated by long intervals (above seconds) of quiescence withfew sporadic neuronal firings. A closer inspection reveals thateach SBE has a distinctive spatio-temporal activity pattern (amotif) between the neurons (Fig. 1B). It is noticed that in a singleculture a small number of different SBE spatio-temporal patternsmay appear, and these patterns repeat themselves overtime(Segev et al., 2002, 2004).

Please cite this article in press as: Rubinsky, L., et al. Spatio-temporal motifs ‘remembered’ in neuronal networks following profound hypothermia. Neural Networks(2008), doi:10.1016/j.neunet.2008.06.008

ARTICLE IN PRESSL. Rubinsky et al. / Neural Networks ( ) – 3

Fig. 1. Burst motifs in a single culture. A. A raster plot of 1 min of a typical network with 18 recorded neurons. Each black dash marks a neuron spike. SBEs are identified asblack lines in which most of the recorded neurons fire simultaneously. B. Closer views of the 4 SBEs observed in A (in the same order of appearance). Note that the left twoSBEs show a similar motif in which the order of neuronal initiation is sustained. Also the two SBEs on the right share a similar motif, which is different from the first twoon the left. C. Similarity matrix of 300 bursts from the same culture shown in A. Top: The similarity matrix when the SBEs are ordered according to their temporal order ofappearance. Bottom: The same matrix from the top, with the rows and columns re-ordered according to the dendrogrammethod (see method section for details). The largered square area around the diagonal on the top loft side of the image, points to a cluster of similar SBEs that share the same initiation motif. The two SBEs in B on the leftbelong to this cluster. Also recognizable is a smaller cluster in the bottom right corner of the image, in which the two SBEs on the right side of B belong to.

2.5. Identifying repeating motifs of SBEs

In order to identify repeating synchronized burst motifs, weused a new measure of similarity that compares the delaysbetween activation of neuron pairs, as described in Raichman andBen-Jacob (2008). In brief, for each recording session we selected300 consecutive SBEs and represent each SBE by an activationmatrix,A, whereA(i, j) is the delay inmilliseconds between the firstspike of neuron i and the first spike of neuron j in the SBE. A NULLvalue in the activation matrix was given to neurons that did notfire in the SBE. The similarity, S

(Ap, Aq

), between SBEs p and qwas

then defined as:

S(Ap, Aq

)=

1n (n − 1)

∑i6=j

H(th −

∣∣Ap (i, j) − Aq (i, j)∣∣) (1)

where H is the heaviside step function (H(x) = 0 if x < 0 andH(x) = 1 if x > 0) and th is a time threshold parameter. Accordingto Eq. (1), S

(Ap, Aq

)is the fraction of neuron pairs (i, j) that obey

the condition that the accuracy in delays between the two burstsis less than th :

∣∣Ap (i, j) − Aq (i, j)∣∣ < th. The summation is made

only on neuron pairs (i 6= j) that did not receive NULL values in anyof the two SBEs (i.e. both neurons fired a spike in both SBEs). In ouranalysis we used th = 30 ms to reflect the average spike precisionin bursts, as was shown in Bonifazi, Ruaro, and Torre (2005) andHarris, Csicsvari, Hirase, Dragoi, and Buzsáki (2003). A similaritymeasure of ‘‘1’’ (‘‘0’’) indicates two SBEs that share a similar (notsimilar) motif.

When plotting the similarity matrix of 300 consecutive SBEs(Fig. 1C) from a single recording, there seems to be no significantstructure to the similarity between bursts and the values are ratherrandom. By applying a standard clustering algorithm (dendrogram,matlab c©) on the similarity matrix, the rows and columns ofthe matrix are re-ordered so SBEs that share the least Euclideandistance between their vectors in the similarity matrix (i.e. SBEsthat share the most similar activation pattern) are placed nextto each other. Clusters of SBE motifs are identified in the newordered matrix by square areas of high values, centered on thematrix diagonal (Fig. 1C). Eachmotif can be represented by a single‘‘synchronized burst’’ by averaging over the k activation matricesrepresenting the collection of SBEs included in the motif’s cluster:

Amotif (i, j) =1N

∑k

Ak (i, j) . (2)

In order to evaluate the change in a motif pattern over differentsessions, such as over some later time or due to application of hy-pothermia, we computed the similarity between the matrices rep-resenting the motif at each session: S (Amotif 1, Amotif 2). Similarityanalysis was performed on 5 of the 6 cultures that demonstratedclear SBE motifs.

3. Results

Our results are based on the analysis of six separate culturesstarting at 10 days in-vitro, at the stage in which the networksexhibit rich spontaneous activity. When viewed under themicroscope these cultures take the form of a dense networkof inter-connected neurons and glia cells with a density ofapproximately 4000 cells/mm2. In a typical culture, most ofthe 60 available recording sites showed action potentials ofindividual neurons. The spontaneous activity after 10 days invitro was characterized by short bursts of synchronized activitylasting several hundreds of milliseconds at a rate of 3–10 burstsper minute. During the SBEs, most of the recorded neuronsparticipated in a relatively rapid firing of action potentials (Fig. 1A).Under normal conditions a typical culture can be recordedcontinuously for several weeks.

Before application of hypothermia each cultures activity wasinitially recorded for a minimum of 10 h to establish a baselineof the network activity. This was followed by the applicationof hypothermia in which the temperature of the medium waslowered from normal conditions of 37 ◦C to a hypothermictemperature of 12 ± 1 ◦C. Each culture remained in these lowtemperature conditions for 1 h, after which the initial temperaturewas returned for aminimumof 24 h. During and after hypothermiathe culture activity was continuously recorded. In 2 of the culturestested, 3 repeating sessions of 1 h of hypothermia and re-warmingwere applied in order tomeasure the cumulative effects ofmultiplesessions of cold on a network. The sessions were separated by24–48 h intervals.

The most pronounced and immediate observation of profoundhypothermia on a network was an overall decrease in the cultureslevels of activity. We observed an immediate decrease of the SBErate during the first few minutes of cold application, and in mostcases the network activity was eventually completely inhibited,showing no collective bursts or sporadic firing. In two out of thesix cultures, the network activity did show rare SBEs (2–3 bursts

Please cite this article in press as: Rubinsky, L., et al. Spatio-temporal motifs ‘remembered’ in neuronal networks following profound hypothermia. Neural Networks(2008), doi:10.1016/j.neunet.2008.06.008

ARTICLE IN PRESS4 L. Rubinsky et al. / Neural Networks ( ) –

during the 1 h of cold), indicating that neurons were still ableto trigger each other during profound hypothermia, though theSBEs were much shorter in duration, included very few spikes, andinvolved less neurons.

Upon termination of profound hypothermia the networkexhibited a temporary overshoot in the rate of SBEs compared toinitial baseline activity levels. This overshoot lasted for about anhour after which the network SBE rate gradually returned to pre-hypothermia levels.

A closer inspection of the SBEs duringnormal conditions prior tohypothermia revealed that they form distinctive spatio-temporalactivity patterns between the neuronal firing (Fig. 1B). In manycases, these patterns repeat in an exact manner with neuronsretaining the order and time delays in which they initiate duringthe SBE event.

Thus, the spatio-temporal order of neuron activation, or SBEspatio-temporal ‘‘image’’, can be termed as a ‘‘motif’’ that issustained and appears for long periods of time (hours and days).In some cultures, several motifs can co-exist simultaneously; witheach burst being associated with its particular motif (Fig. 1B). Thefact that motifs are sustained over long periods of time indicatesthat the overall connectivity between regions in the network andtheir neuronal pathways are largely maintained.

By applying ameasure of similarity between the identified SBEs(Fig. 1C), based on the temporal delays between neuron initiations,we were able to successfully cluster the recorded SBEs of eachculture into several observedmotifs. The different clusters (motifs)of SBEs are highlighted in the re-ordered similarity matrix as redsquare areas along the diagonal, indicating a set of SBEs that sharea similar pattern. The low levels of similarity between the clustersconfirm the distinction between different motifs.

By comparing the SBE motifs before and after the applicationof hypothermia, we found that the motifs remained relativelyunchanged. As shown in Fig. 2A, most of the neurons in the burstretain both their order of initiation and their relative time ofactivation following the application of cold. Moreover, it is alsoevident that the neurons’ individual characteristic is also sustained,since neurons that tended to fire relatively longer spike trains(the first firing neurons in Fig. 2A), are still firing more than otherneurons after the application of hypothermia. Thus neurons thatfired elongated spike trains relative to the rest of the neuronsretained that characteristic after cold application.

Although the SBE motif retention is relatively consistent andcan be maintained for long periods as a living network, changesin the motifs such as the addition or disappearance of neuronsare invariable seen over time. These changes may be associatedto modifications in synaptic connections between neurons and tothe movement of neurons from the recording sites on the plannerarea. Therefor, comparing similarity between bursts belonging toa motif at a certain time and bursts of the same motif at a latertime is expected to show a decrease in similarity. On average, thesimilarity level decreased to 64% ± 24% of its original level after5 h in ‘control’ conditions (Fig. 2B). 5 h following hypothermia,the decrease in similarity was 70% ± 25% compared to conditionsbefore hypothermia.Whenwe compared each culture individually,the relative change in the decrease in similarity 5 h afterhypothermia and 5 h during control conditions was 100% ± 33%on average (Fig. 2B). These results indicate that the 1 h applicationof hypothermia did not cause a significant change in the SBE motifthat is greater than that caused by time in control conditions. Thisanalysis was made over 5 cultures that exhibited clear SBE motifs.The individual similarity values of SBE motifs from the 5 culturesanalyzed can be found in Table 1.

In two of the cultures tested multiple sessions of hypothermiaand re-warming were applied in order to measure the cumulativeeffects of cold on a network. Each of the 2 networks received

Fig. 2. Retention of SBEmotif after application of cold. A. Two SBEs belonging to thesamemotif as appear before and after cooling. In the left raster plot, the neurons aresorted from top to bottom according to their order of initiation. In the raster plot onthe right the sorting of the neurons is the same as on the left. Note that the two SBEsretain a similar initiation order, with some small changes. Also note that neuronsthat fire elongated spike-trains relative to the rest of the neurons (top neurons)retained this characteristic after the application of cold. Also noticeable is thatseveral neurons have been inactivated after cold (see Fig. 3 for further discussion).B. Similarity between motifs before and after cold. For each culture we computedthe similarity between the most prominent burst motif from 1 h before and 5 hafter application of cold. For control, we also computed the similarity of the motiffrom 6 h before the application of cold and 1 h before. The bars indicate averageresults over 9 sessions of cold on 5 different cultures that showed a clear burstmotif (see Table 1 for individual similarity values). The average level for cold was70%±25% and for the control 64%±24%. The bar on the right indicates the averageratio between the changes in similarity after cold to the control over all sessions(100% ± 33%). The error bars represent Standard Deviations.

3 additional 1 h profound hypothermia sessions separated by24–48 h between each session.

Although there is an overall retention of burst motif patternsfollowing the application of low temperature, our results point toa quantifiable reduction in neuron activity after each hypothermiaapplication. This is evident in the raster plot depiction of SBEactivity in which each progressive hypothermic period resultedin greater neuron inactivity and reduction of neuron firing ratesper burst (Fig. 3A). This decline in neuron activity is quite evidentfollowing multiple applications of profound hypothermia anddemonstrates an approximate 10% reduction in neuron activityafter each cooling session.

As is shown if Fig. 3B, a monotonic decrease in the numberof active neurons is observed immediately following each sessionof hypothermia. This decrease in neuronal number, observed in 8out of 10 sessions, is attributed to the application of cold alone,since neuronal numbers show little fluctuations between sessions.Additionally multiple application of cold demonstrated a reducingeffect on the collective dynamics of the neuronal network. Eachsuccessive cooling period resulted in a greater decrease in thenumber of spikes per neuron (Fig. 3C) and in the rate of SBE firing(Fig. 3D), demonstrating a strong cumulative detrimental effect ofthe low temperature on the network.

4. Discussion

The results from the current experiments demonstrate thetypical behavior of a neuronal network undergoing profoundhypothermia. In all cases our cultured networks were able tosurvive 1 h of profound hypothermia. This network survival wasalso observed in our previous experiments in which networksendured 20 h of deep hypothermia (Rubinsky et al., 2007) otherin-vivo cooling experiments (Villa, Tetko, Dutoit, De Ribaupierre, &De Ribaupierre, 1999) and clinical situations inwhich brain activityreturns after one hour of applied profound hypothermia in surgicalprocedures (Tharion et al., 1982).

Each experimental network in the current study presentedthe typical spontaneous synchronized bursting events (SBE)associated with neurons grown in culture. Inspection of theSBEs recorded by our system during normal conditions prior

Please cite this article in press as: Rubinsky, L., et al. Spatio-temporal motifs ‘remembered’ in neuronal networks following profound hypothermia. Neural Networks(2008), doi:10.1016/j.neunet.2008.06.008

ARTICLE IN PRESSL. Rubinsky et al. / Neural Networks ( ) – 5

Fig. 3. Decrease in neuron number and neuronal activity after application of cold.A. Raster plots of SBEs of a single culture before and after 3 sessions of cold. Aftereach session the number of firing neurons decreases. B. Monotonic decrease in thenumber of neurons after each application of cold on a single culture. The error barsrepresent Standard Deviations. Each snowflake symbol marks a 1 h session of cold.Note that between sessions the number of active neurons remains relatively equal,and decreases only after each session of cold. C and D. Reduction in the number ofspikes of each neuron and in the rate of SBE firing due to multiple sessions of cold.The error bars represent Standard Deviations.

to hypothermia revealed that the cultures contained distinctivespatio-temporal activity patterns of neuronal firing. Our resultsdemonstrated that the spatio-temporal order of neuron activationor SBE spatio-temporalmotif prior to hypothermia applicationwasnot significantly altered following exposure to 1 h of profoundhypothermia, as compared to the changes observed by the sametime period in normal conditions.

It has been suggested that in-vitroneuronal network collectivedynamics may represent the building blocks of higher-levelcortical functions (Ikegaya et al., 2004; Tetko & Villa, 2001). TheSBE motifs associated with the networks may act as ‘memory’templates or precursors ofmemory related activitymodes of largerin-vivo networks. The neuronal firing could possibly representinnate system level information encoded in the correlationsbetween neuronal activities during the SBEs (Baruchi & Ben-Jacob,2007). The maintenance of the motifs observed in our studysuggests that ‘memory’ in neuronal networksmay be retained evenfollowing the strong suppression of network activity caused byprofound hypothermia. Most recently, this view was expressed byBaruchi and Ben-Jacob (2007)who successfully imprintedmultiplepre-designed and enduring memory templates in the activity ofin-vitro cultured networks.

In the current study the retention of these intrinsic motifs andindividual neuron characteristics suggest that the basic structure ofthe culture is not greatly affected following profound hypothermia,and that the network ‘remembers’ it prior modes of firing. Thisview is supported by clinical findings, which demonstrate theretention of cognitive functions in individuals recovering fromsurgical procedures in which body temperatures were greatly

reduced (Hövels-Gürich, Seghaye, Däbritz, Messmer, & Bernuth,1997). Our results demonstrating the retention of the basicstructure of the culture following profound hypothermiamay havean important contribution in supporting what is seen in clinicalstudies.

Remarkably, the burst motif maintenance is also seen afterapproximately 10% of the neuronal network suffered randomneuronal degeneration, as evidenced by the decreasing number ofobserved firing neurons after recovery. This inactivity is believedto be neuronal death in the network, as based on unpublished datafromour lab inwhich a decrease in neuron viability is observed dueto hypothermia application as assessed by XTT assays (L.Rubinsky,unpublished observations).

This detrimental effect of cold on the brain is supported bystudies that found that 4.5%–10% of patients have documentedpostoperative neurological morbidities (Bellinger et al., 1995;Svensson et al., 1993; Tharion et al., 1982), including dysfunctionin memory and learning following surgical procedures whichused low temperature (Reich et al., 1999). The neuronal inactivityfollowing the profound hypothermia witnessed in this studymay be implicated in playing a role in the mental and memorycomplications associated with surgical procedures in whichprofound hypothermia is used.

Our results with the use of our in-vitro model as a means ofassessing low temperature on neurons and neuronal networks areinline with results found in clinical studies. In our experimentalsetup we attempted to study only hypothermia on neurons. Thiswas achieved by limiting adverse effects from other variablessuch as hypoxia, in this case by supplying the network withthe appropriate physiological atmosphere. However, it must betaken into consideration that the effects of hypothermia in-vivoinvolve processes, such as blood flow, oxygen supply and extendedcooling/re-warming times, not found in our in-vitro model. Thisraises the possibility that the neuronal damage witnessed inthe current experiments may not only be associated with thehypothermia but the interaction of the re-warming speed of thenetwork.

Although hypothermia has been shown to cause neuronswelling, cytoplasmic voculation and cytoskeletal disruption(Emery & Lucas, 1995), the fast re-warming rate of the networkmay also play a role in some of the degeneration (Bissonnette, Pel-lerin, Ravussin, Daven, &Magistretti, 1999; Nakamura et al., 2003).We found that immediately following profound hypothermia thenetwork exhibited a dramatic increase in activity, with a tempo-rary overshoot reaching 3 times the SBE rate compared to the ac-tivity level prior to hypothermia. This resultmay be due to the sud-den termination of the cooling fluid on the network, and the rapidreturn to normal temperature within 5 min. We assume that if thenetwork had been gradually returned to 37 ◦C over the course ofseveral hours this overshoot would most likely not be evident.

The overshoot in SBE rate itself could be attributed to anincrease in glutamate and lactate release upon re-warmingfollowing the strong inhibition of the network activity duringhypothermia (Nakamura et al., 2003). Our statistical analysisshows no significant reduction in the number of neurons severalhours after return to physiological temperatures. The lack offurther reduction in neurons may demonstrate no consequentialsecondary damage due to glutamate excitoxicity.

The majority of our experimental results demonstrated a pro-nounced inactivity of neurons following profound hypothermia.However, in 2 of the experiments no significant degeneration wasobserved. These results may be explained by the fact that due tosample size inactivated neurons were simply not recorded, or pos-sibly that the recorded active neurons were more robust and re-silient to cold thus able to withstand the hypothermic toxic effect.

In summary we demonstrated that neurons and neuronalfunction are retained following one hour, and multiple sessions

Please cite this article in press as: Rubinsky, L., et al. Spatio-temporal motifs ‘remembered’ in neuronal networks following profound hypothermia. Neural Networks(2008), doi:10.1016/j.neunet.2008.06.008

ARTICLE IN PRESS6 L. Rubinsky et al. / Neural Networks ( ) –

thereafter, of profound hypothermia. A spatio-temporal analysisof our recorded data from the arrays was performed, and theproperties of the network such as activity and burst motifswere quantified. Our results demonstrate a measurable effectof low temperature on the neuronal network. These includeretention of burst motifs indicating the possible sustaining ofmemory templates after profound hypothermia and quantifiabledegenerative neuronal death.

Acknowledgments

Wewould like to thank Inna Brainis for the technical assistancein the preparations of the neuronal cultures. This research wassupported in part by the Tauber Funds and the Maguy-Glass Chairin Physics of Complex Systems.

References

Barone, F. C., Feuerstein, G. Z., & White, R. F. (1997). Brain cooling duringtransient focal ischemia provides complete neuroprotection. Neuroscience andBiobehavioral Reviews, 21, 31–44.

Baruchi, I., & Ben-Jacob, E. (2007). Towards neuro-memory-chip: Imprintingmultiple memories in cultured neuronal networks. Physical Review E. StatisticalNonlinear and Soft Matter Physics, 75(5 Pt 1).

Bellinger, D. C., Jonas, R. A., Rappaport, L. A., Wypij, D., Wernovsky, G., Kuban, K.C., et al. (1995). Developmental and neurologic status of children after heartsurgery with hypothermic circulatory arrest or low-flow cardiopulmonarybypass. New England Journal of Medicine, 332, 549–555.

Bissonnette, B., Pellerin, L., Ravussin, P., Daven, B. V., &Magistretti, J. P. (1999). Deephypothermia and rewarming alters glutamate levels and glycogen content incultured astrocytes. Anesthesiology, 91, 1763–1769.

Bonifazi, P., Ruaro, E., & Torre, V. (2005). Statistical properties of informationprocessing in neuronal networks. European Journal of Neuroscience, 22, 2953.

Deleon, S. Y., Thomas, C., Roughneen, P. T., King, N., Lehne, R., DeLeon, A. M., et al.(1998). Experimental evidence of cerebral injury from profound hypothermiaduring cardiopulmonary bypass. Pediatric Cardiology, 19, 398–403.

Emery, D. G., & Lucas, J. H. (1995). Ultrastructural damage and neuritic beading incold-stressed spinal neurons with comparisons to NMDA and A23187 toxicity.Brain Research, 18, 161–173.

Gillinov, A. M., Redmond, J. M., Zehr, K. J., Troncoso, J. C., Arroyo, S., Lesser, R. P.,et al. (1993). Superior cerebral protection with profound hypothermia duringcirculatory arrest. Annals of Thoracic Surgery, 55, 1432–1439.

Haka-Ikse, K., Blackwood, M. J., & Steward, D. J. (1978). Psychomotor developmentof infants and children after profound hypothermia during surgery forcongenital heart disease. Developmental Medicine and Child Neurology, 20,62–70.

Harris, K. D., Csicsvari, J., Hirase, H., Dragoi, G., & Buzsáki, G. (2003). Organization ofcell assemblies in the hippocampus. Nature, 424, 552–556.

Hövels-Gürich, H. H., Seghaye, M. C., Däbritz, S., Messmer, B. J., & Bernuth, G. (1997).Cognitive and motor development in preschool and school-aged childrenafter neonatal arterial switch operation. Journal of Thoracic and CardiovascularSurgery, 114, 578–585.

Hulata, E., Segev, R., & Ben-Jacob, E. (2002). A method for spike sorting and

detection based on wavelet packets and Shannon’s mutual information. Journalof Neuroscience Methods, 117, 1–12.

Ikegaya, Y., Aaron, G., Cossart, R., Aronov, D., Lampl, I., Ferster, D., et al. (2004).Synfire chains and cortical songs: Temporalmodules of cortical activity. Science,304, 559.

Khaladj, N., Peterss, S., Oetjen, P., von Wasielewski, R., Hauschild, G., Karck, M.,et al. (2006). Hypothermic circulatory arrest with moderate, deep or profoundhypothermic selective antegrade cerebral perfusion: Which temperatureprovides best brain protection? European Journal of Cardiothoracic Surgery, 30,492–498.

Kumral, E., Yuksel, M., Buket, S., Yagdi, T., Atay, Y., & Guzelant, A. (2001). Neurologiccomplications after deep hypothermic circulatory arrest: Types predictors, andtiming. Texas Heart Institute Journal, 28, 83–88.

McIntyre, A. L., Fergusson, A. D., Hebert, C. P., Moher, D., & Hutchison, S. J. (2003).Prolonged therapeutic hypothermia after traumatic brain injury in adults: Asystematic review. JAMA, 22, 2992–2999.

Muller, D., Fieguth, G. H., Wimmer-Greinecker, G., Wohleke, T., Kleine, P., & Moritz,A. (2004). Neurologic outcome after surgery of the aortic arch: Comparison ofdeep hypothermic arrest, antergrade and retrograde cerebral perfusion. IJTCVS,20, 72–76.

Nakamura, T., Miyamoto, O., Sumitani, K., Negi, T., Itano, T., & Nagao, S. (2003). Dorapid systemic changes of brain temperature have an influence on the brain?Acta Neurochirurgica, 145, 301–307.

Raichman, N., & Ben-Jacob, E. (2008). Identifying repeating motifs in the activationof synchronized bursts in cultured neuronal networks. Journal of NeuroscienceMethods, 170(1), 96–110.

Reich, D. L., Uysal, S., Sliwinski, M., Ergin, A., Kahn, R. A., Konstadt, S. N., et al. (1999).Neuropsychologic outcome after deep hypothermic circulatory arrest in adults.Journal of Thoracic and Cardiovascular Surgery, 117, 156–163.

Rubinsky, L., Raichman, N., Baruchi, I., Shein, M., Lavee, J., Frenk, H., et al. (2007).Study of hypothermia on cultured neuronal networks using multi-electrodearrays. Journal of Neuroscience Methods, 160, 288–293.

Sailhamer, E. A., Chen, Z., Ahuja, N., Velmahos, G. C., de Moya, M., Rhee, P.,et al. (2007). Profound hypothermic cardiopulmonary bypass facilitates survivalwithout a high complication rate in a swinemodel of complex vascular, splenic,and colon injuries. Journal of American College of Surgery, 204, 642–653.

Segev, R., Benveniste, M., Hulata, E., Cohen, N., Palevski, A., Kapon, E., et al. (2002).Long term behavior of lithographically prepared in vitro neuronal networks.Physival Review Letters, 88(11), 118102.

Segev, R., Baruchi, I., Hulata, E., & Ben-Jacob, E. (2004). Hidden neuronal correlationsin cultured networks. Physical Review Letters, 92(11), 118102.

Sessler, I. D. (2001). Complications and treatment of mild hypothermia. Anesthesi-ology, 95, 531–543.

Svensson, L. G., Crawford, E. S., Hess, K. R., Coselli, J. S., Raskin, S., Shenaq, S. A., et al.(1993). Deep hypothermia with circulatory arrest. Determinants of stroke andearly mortality in 656 patients. Journal of Thoracic and Cardiovascular Surgery,106, 19–28.

Tetko, I. V., & Villa, A. E. (2001). A pattern grouping algorithm for analysis ofspatiotemporal patterns in neuronal spike trains. 2. Application to simultaneoussingle unit recordings. Journal of Neuroscience Methods, 105, 15–24.

Tharion, J., Johnson, D. C., Celermajer, J. M., Hawker, R. M., Cartmill, T. B., &Overton, J. H. (1982). Profound hypothermia with circulatory arrest: Nineyears’ clinical experience. Journal of Thoracic and Cardiovascular Surgery, 84,66–72.

Villa, A. E., Tetko, I. V., Dutoit, P., De Ribaupierre, Y., & De Ribaupierre, F.(1999). Corticofugal modulation of functional connectivity within the auditorythalamus of rat, guinea pig and cat revealed by cooling deactivation. Journal ofNeuroscience Methods, 86(2), 161–178.

Please cite this article in press as: Rubinsky, L., et al. Spatio-temporal motifs ‘remembered’ in neuronal networks following profound hypothermia. Neural Networks(2008), doi:10.1016/j.neunet.2008.06.008