Keywords

1 Introduction

The brain is one of the most fascinating and complex organisms of the known Universe. Even if enormous advances have been done in the last decades, also thanks to the convergence of different disciplines into the study of the brain, we still know very little about how the brain works and we are not yet able to design artificial systems emulating its functionality. One of its most peculiar property is the capability to exploit plasticity to allow performing cognitive or motor task even when there is a damage. This is because the brain is redundant and intrinsically modular, being composed of local networks that are embedded in networks of networks [1], sparsely connected to each other [2]: the connections can reorganize bypassing the damage or reinforcing weak connections [3, 4].

Indeed, understanding the intricacy of brain signals, what is the effect of a damage on signal generation [5], and how this impacts on the electrophysiological behavior of brain networks and on their (re)-organization, has a twofold importance: from one side it is necessary in order to shape suitable intervention strategies [6] based on novel bioelectronic devices [3]; from the other side it will help in designing novel ‘neurobiohybrid’ technologies [7] which can exploit the brain self-repair capability in case of a damage. To reach the above goals it is fundamental to deeply characterize and understand how a damage affects the electrophysiological behavior of a network.

Within this framework, simplified in vitro models of cell assemblies can provide useful insights to investigate the interactions between networks of neurons, both in physiological and in pathological conditions. In vitro systems, by overcoming the limits of in vivo models imposed by the complexity of the surrounding networks and by the consequent low level of reproducibility, can thus serve as test bed for innovative solutions ranging from neuropharmacological to electroceutical applications [8,9,10]. Moreover, in silico models, either software or hardware, of the electrophysiological behavior of such reduced networks can be used to replace neuronal functionalities in the framework of novel neuroprosthetic devices [11, 12].

In the last decade, different groups have started to realize in vitro modular structures [13,14,15,16]. In particular, (multi) modular cell cultures plated over Micro Electrode Arrays (MEAs) represent an interesting bio-artificial experimental model for studying neuronal networks at the mesoscale level for different reasons. First, cell cultures on MEA can be manipulated in several ways (ranging from pharmacological to electrical, optical and other kinds of perturbations) and can survive longer with respect to other preparations [17]. Second, recording from multiple sites is crucial for investigating neural information processing, in case of neural networks and in particular when dealing with multimodular cell assemblies. Third, the high temporal resolution of MEAs allow characterizing the neuronal activity at a time scale that is critical to understand neuronal dynamics.

In the present work, we realized a multimodular structure able to reproduce the modular architecture of different interconnected subpopulations. We then characterized the electrophysiological activity of both spontaneous and electrically evoked activity in the obtained cell cultures confined in 3 or 4 modules. Neuronal modules, during development, projected to each other and therefore self-organized themselves in a network with intricate functional and anatomical connectivity to mimic at least a part of the modular properties of the neuronal tissue of origin. We then used a custom-made laser setup [18] able to produce a focal lesion between modules, thus affecting the anatomical connectivity among the neuronal modules. We thus observed the correlation of spike trains between modules and their changes due to the lesion.

This work lays the foundation for understanding the dynamical changes occurring after a brain lesion, a critical step towards the development of novel strategies to overcome the loss of communication between cell assemblies, with applications on both in vivo systems and in silico devices.

2 Methods

2.1 Ethics Statement

We obtained primary neuronal cultures from rat embryos at gestational day E18 (pregnant Sprague-Dawley female rats were delivered by Charles River Laboratories, Lecco, Italy). When performing the experiments, we minimized the number of sacrificed rats and the potential for nociceptor activation and pain-like sensations and we respected the three R (replacement, reduction and refinement) principle in accordance with the guidelines established by the European Community Council (Directive 2010/63/EU of September 22nd, 2010). Rat housing was in accordance with institutional guidelines and with the in force legislation of Italy (legislation N°116 of 1992). The procedures for preparing neuronal cultures are described in detail in previous studies [9, 19].

2.2 PDMS Structures for Multimodular Network Confinement

Through soft lithography, we realized a polymeric structure in polydimethylsiloxane (PDMS), composed of different modules, in order to provide the physical confinement of neuronal cultures [20, 21]. The PDMS mask was positioned on the MEA substrate before the coating procedure, performed by putting a 100-μl drop of laminin and poly-D-lysine solution on the mask and leaving it in the vacuum chamber for 20 min. The mask was then removed and cells were plated afterwards. MEAs (Multi Channel Systems, Reutlingen, Germany) are available in different geometrical layouts: “4Q”, where 60 electrodes are organized in 4 separate quadrants, “8 × 8”, where electrodes are placed according to a square grid; “6 × 10”, where electrodes are placed according to a rectangular grid. The nominal final cell concentration was around 500 cells/μl (~100 cells per network module).

2.3 Laser Ablation Setup

The entire optical system was described in a previous work [18]. Briefly, the laser dissection source consisted in a pulsed sub-nanosecond UV Nd:YAG laser at 355 nm (PNV-001525-040, PowerChip nano-Pulse UV laser – Teem Photonics), whose output was modulated with the aid of an acousto-optical modulator (MQ110-A3-UV, 355 nm fused silica, AA-Opto-electronic) driven by a custom low impedance linear driver. The laser dissector was integrated in a modified upright microscope (BX51 – Olympus) equipped with a 20x, 0.5 NA water dipping objective. A custom-made software interface based on LabVIEW (National Instruments) controlled the UV laser intensity, pulse repetition rate, and the number of pulses delivered to the sample. Synchronization signals between devices were sent through a D/A board (PCI-6529, 24 bit, 4 channels, 204.8 kSamples/second, National Instruments), in order to synchronize CCD image acquisition (Andor DU-897D-C00), sample positioning trough motorized stage (assembled 3-axis linear stages, M-126.CG1, Physics-Instruments), and the trigger of UV laser pulses.

2.4 Recording Set-Up and Experimental Protocol

The activity of all cultures was recorded by means of the MEA60 System (MCS). The signal from each channel was sampled at 25 kHz and amplified using a Multichannel System amplifier with a bandwidth of 1 Hz–3 kHz. Each recorded channel was acquired through the data acquisition card and on-line monitored through MC_Rack software (MCS). A commercial stimulator (MCS) was integrated in the system to deliver electrical pulses through one of the electrodes of the MEA. Figure 1A presents a scheme of the experimental setup.

Fig. 1.
figure 1

Setup and protocol description. (A) Schematic description of the setup. A personal computer equipped with MC-Card (Multichannel System, MCS) records the activity from the MEA 1060 Amplifier system. A commercial stimulator (STG 4002, MCS) delivered open-loop regular stimulation to the MEA amplifier. (B) Schematic of the experimental protocol, consisting of 60 min of basal (spontaneous) activity followed by 5 min of stimulation delivered to one electrode. Stimulation was delivered to different electrodes to test the propagation of the signal across modules. One of the clusters was then isolated by means of a laser cut. Following the laser ablation, the protocol was repeated as before cut. (C) Optical micrographs depicting a corner of a modular culture before (left) and after (right) laser cut.

Full size image

To reduce thermal stress of the cells during the experiment, MEAs were kept at 37 ℃ by means of a controlled thermostat (MCS) and covered by PDMS caps to avoid evaporation and prevent changes in osmolarity.

We performed experiments on neocortical modular networks, recorded between 20 and 25 Days In Vitro (DIVs). The experimental dataset consisted of 8 modular networks (Fig. 1B) consisted of 5 consecutive phases:

  1. (i)

    ‘Basal1’: one-hour recording of spontaneous activity;

  2. (ii)

    ‘Stim1’: stimulation session I, which consists of serially stimulating at least two electrodes per cluster using a train of 50 positive then- negative pulses (1.5 V peak-to-peak, duration 500 μs, duty cycle 50%) at 0.2 Hz;

  3. (iii)

    ‘Cut’: laser ablation of inter-cluster neural connections, whose aim is to isolate a cluster which is physically and functionally connected to at least another one (Fig. 1C), note that in control experiments no laser ablation was performed;

  4. (iv)

    ‘Basal2’: one-hour recording of spontaneous activity after performing the lesion;

  5. (v)

    ‘Stim2’: stimulation session II, from the same electrodes of phase ii.

Two control experiments underwent the same protocol, but in phase (iii) the laser setup was turned off (i.e. the ‘Cut’ phase was not present).

2.5 Data Analysis and Statistics

Raw data were bandpass-filtered in the Multi-Unit Activity - MUA (i.e. spikes) data band (~300 Hz to 3 kHz) and processed using custom MATLAB (The Mathworks, Natick, MA, USA) scripts. A custom offline spike detection algorithm, based on Precise Timing Spike Detection, was used to discriminate spikes [22]. For detecting bursting activity, we used a custom Burst Detection algorithm, according to which bursts were identified as sequences of at least 5 consecutive spikes spaced less than a 100 ms time threshold [23].

Once spikes and bursts were detected, we computed the following electrophysiological parameters:

  • Mean Firing Rate, MFR (mean number of spikes over an interval of time [spike/s]);

  • Mean Bursting Rate, MBR (mean number of bursts over an interval of time [burst/minute]);

  • Burst Duration (duration of burst in [ms])

  • Burstiness Index (burstiness level of the network, providing a normalized value between 0 - no bursts - and 1 - burst dominated [9, 24]).

To characterize the effect of the lesion induced by the laser ablation, we computed the ratios of MFR, MBR and Burst Duration between their values calculated post and pre lesion.

Furthermore, we evaluated the level of pairwise correlation between the spike trains, by exploiting the Cross-Correlation method implemented in SpyCode, a custom Matlab-based software developed in our lab [25]. We measured correlations within 100 ms time windows, using 1 ms bins and we normalized the Cross-Correlation peaks found in the post lesion phase over the average value of correlation found before the lesion.

In order to investigate the impact of the lesion on the evoked activity, we calculated the Post-Stimulus Time Histograms (PSTH) of stimulus-associated action potentials detected from each electrode (1 ms bins, normalized over the total number of stimulation pulses) [26]. The area under the normalized PSTH curve was used to quantify the total amount of stimulation-evoked neural activity during each stimulation phase.

The normal distribution of experimental data was assessed using the Kolmogorov-Smirnov normality test. Statistical comparisons were performed with Wilcoxon signed-rank test, with p-values <0.05 were considered as significant.

3 Results

Typical multimodular cultures were characterized by subpopulations of neurons interconnected by neurites allowing transfer of information from one module to another (Fig. 1C, left and Fig. 2A, left). The goal of the present study was to characterize both the spontaneous and evoked dynamics of multimodular networks and to subsequently evaluate the electrophysiological effect of the laser ablation aimed to isolate one of the neuronal module (Fig. 1C, right and Fig. 2B, left).

Fig. 2.
figure 2

Network activity in two representative experiments. (A) Control experiment. On the left MEA modules are graphically depicted, black dots representing the channels, where the active ones have background colored according to the module they belong to. In the center, a raster plot of the activity recorded during the Basal 1 phase is colored according to modules as on the left. In the right a raster plot of the activity recorded during the Basal 2 phase is colored according to modules as on the left. (B) Same as in A, for experiments with laser ablation (i.e. between Basal 1 and Basal 2, a laser ablation was performed - Cut phase). (Color figure online)

Full size image

3.1 Laser Cut Affects the Spontaneous Activity of a Neuronal Network

Spiking activity of multimodular networks appeared well synchronous for the entire duration of the experiment in control condition, as it can be qualitatively appreciated by looking at the raster plot of Fig. 2A. On the other hand, the laser dissection of a cluster induced a strong desynchronization of spiking activity between the isolated cluster and all the others (Fig. 2B).

As a consequence of the lesion, we also observed a significant global decrease in the network mean firing rate, both inside the isolated cluster (Isolated, p < 0.05), and in all the other clusters (Others, p < 0.001) that were previously connected to the isolated one (Wilcoxon signed-rank test, Fig. 3A). No changes of firing rate were found in the control condition (Fig. 3A). Moreover, the analysis of the pairwise correlation indicated a stable level of synchronicity for the activity of the electrodes belonging to the same cluster (Intra, Fig. 3B, left), while a dramatic drop was observed between the electrodes of the isolated cluster and all the other ones (Inter, Fig. 3B, right).

Fig. 3.
figure 3

Changes in spontaneous activity. (A) MFR variation in control experiments (CTRL) and in modules belonging to experiments with laser ablation (Isolated and Others). (B) Cross-Correlation pre and post lesion calculated in the same cluster (Intra) and among the isolated cluster electrodes and the other clusters (Inter) for the experiments with laser ablation. (C) Normalized cross-correlations pre and post lesion both for the lesioned (Lesion) and control (CTRL) experiments. **p < 0.01, *p < 0.05 Wilcoxon signed-rank test.

Full size image

Figure 3C underlines this effect by showing the Cross-Correlation peaks normalized over the average value of correlation found before the laser dissection, calculated both in the same cluster (Intra) and among the isolated cluster electrodes and the other clusters (Inter). While correlation remained rather stable in the control and in the lesioned intra condition, it significantly dropped in the lesioned intra condition (p < 0.001, Wilcoxon signed-rank test).

As shown in Fig. 4, laser dissection also influenced the bursting activity (Fig. 4A). We found that, after ablation, the tendency of both the rate (Fig. 4B) and the duration (Fig. 4C) of the detected bursting activity was to decrease for all the recorded clusters in the lesioned condition, while they remained stable for the control condition (values close to 1). Moreover the burst duration significantly decreased after ablation (p < 0.001, Wilcoxon signed-rank test). The level of burstiness of the network showed a tendency to increase within the isolated cluster after the lesion (Fig. 4D).

Fig. 4.
figure 4

Comparison of spontaneous and evoked bursting activity of healthy and damaged modular cultures. (A) 1.25-s raster plot of spontaneous activity of a channel before lesion. The red lines correspond to the detected bursting activity, the red bar above them represents the Burst Duration, and the grey lines corresponds to isolated spiking activity. (B) Variation of Burst Rate pre and post lesion either within the isolated cluster (Isolated) or in the other ones (Others) and in the no-lesioned ones (CTRL). (C) Comparison between the statistical distributions of the ratio between normalized Burst Durations calculated pre and post lesion for the isolated cluster (Isolated), the other clusters (Others) and for the no-lesioned (CTRL) condition. (D) Burstiness Index calculated pre and post lesion in all condition. **p < 0.01, Wilcoxon signed-rank test. (Color figure online)

Full size image

3.2 Laser Dissection Confines Evoked Activity

We also observed the effect of electrical stimulation in the different conditions. Before performing the lesion, electrical stimulation (Fig. 5A) was able to evoke activity both within the cluster hosting the stimulation channel and in the connected modules (Fig. 5B). After laser dissection, the evoked activity remained confined within the isolated cluster without spreading towards the other ones (Fig. 5C).

Fig. 5.
figure 5

Network response to stimulation. (A) Graphical representation of the stimulation site over the MEA. Stimulation was a sequence of 50 stimuli delivered at 0.2 Hz through an electrode in the top left module before lesion. (B) Stimulation effects in the pre lesion condition. Each graph represents the 60-electrodes MEA response at different time points with respect to the stimulus: ranging from −10 ms before the stimulation to 30 ms after the stimulus onset. Each pixel represents, in grey level, the probability to detect a spike in 1 ms bin. Around the stimulus onset, the spike detection algorithm identifies artifacts on all electrodes that were then blanked to avoid false positive detections. Stimulation site is highlighted in red. (C) Stimulation effects in the post lesion condition. Same as in A, but after lesion. (Color figure online)

Full size image

We quantified this effect in terms of PSTH area variation (Fig. 6). We observed a significant (Wilcoxon signed-rank test, p < 0.001) global decrease of PSTH Area after the laser cut in both local and distal responses (Fig. 6B): since lesion was aimed at reducing the amount of connections among clusters, responses of other clusters resulted to be more affected than that the isolated cluster. Single channel PSTH variation are highlighted in Fig. 6C.

Fig. 6.
figure 6

Impact of lesion on PSTH. (A) Graphical representation of MEA modules. (B) Ratio between normalized PSTH Areas calculated pre and post lesion for the isolated cluster (Isolated), the other clusters (Others) in the lesioned condition, and for the no-lesioned condition (CTRL). (C) Post-stimulus time histograms (PSTH) obtained when stimulating a channel belonging to either the isolated cluster (left) or one channel from the other clusters (right). The electrodes labelled in blue belong to the isolated cluster, while the red boxes indicate the stimulated channel. Black curves report the evoked activity before the lesion, while the red ones refer to the evoked activity after lesion. **p < 0.01, *p < 0.05 Wilcoxon signed-rank test. (Color figure online)

Full size image

4 Discussion

Despite their simplicity, multimodular cell cultures are a very useful tool to manipulate the neuronal networks dynamics. Modular networks were recently developed to impose a predefined directionality in functional information transfer between neighboring nodes [14]. A modular preparation was also recently used to prove the importance of the modular organization on dynamical richness in cortical networks [16].

In this work, we characterized the electrophysiological behavior of multimodular networks coupled to MEAs, devices able to both record and stimulate the neuronal activity of the neuronal cells sitting on the surface of the planar electrodes. Thanks to this system, we were able to show that selective laser dissection of interconnections among neural assemblies affected both spontaneous and evoked activity of multimodular networks, by inducing de-synchronization between the different modules during spontaneous activity, and preventing propagation of evoked responses among modules. The activity remained confined to the isolated cluster exhibiting the same level of synchronization as before the lesion. Moreover, the burstiness level showed a tendency to increase after the lesion, suggesting that the lack of incoming/outcoming connections further promoted the appearance of network-wide events. This is in line with recent findings related to the concept of ‘dynamical richness’ and ‘network complexity’ [16, 27], according to which interconnected subpopulation of neurons show a richer dynamics than single isolated clusters of cells.

Controlling the parameters of network information and studying the effect of a lesion can have a disruptive impact both in the neuroscience and in the computational neuroscience field. One of the open questions in neuroscience concerns our ability to decode the electrophysiological patterns of network activity [28]. The use of engineered neuronal networks can be the key to further investigate the neural code. Moreover, computational neuroscience can benefit from studies like ours in which simple and controllable biological networks are involved, to better tune the computational models that aims at mimicking the electrophysiological activity of the brain [29]. This kind of studies also opens up new avenues towards ‘wetware’ based technologies which can be employed in a synergistic way with pure silicon-based systems to truly emulate brain’s activity and reproduce intelligent behaviors [30].

In the coming years it will be increasingly necessary to take advantage of the complementary strengths of biological and computational studies to face the increasingly complex challenges related to study of the brain. All this will then promote interesting opportunities for innovative technologies to treat brain-related disabilities or deriving breakthrough ‘neuro-computational’ methodologies.