Biomedical Microdevices

, Volume 11, Issue 2, pp 495–501

Engineered neuronal circuits shaped and interfaced with carbon nanotube microelectrode arrays


  • M. Shein
    • School of Electrical EngineeringTel-Aviv University
  • A. Greenbaum
    • School of Electrical EngineeringTel-Aviv University
  • T. Gabay
    • School of Electrical EngineeringTel-Aviv University
  • R. Sorkin
    • School of Electrical EngineeringTel-Aviv University
  • M. David-Pur
    • School of Electrical EngineeringTel-Aviv University
  • E. Ben-Jacob
    • School of Physics and AstronomyTel-Aviv University
    • School of Electrical EngineeringTel-Aviv University

DOI: 10.1007/s10544-008-9255-7

Cite this article as:
Shein, M., Greenbaum, A., Gabay, T. et al. Biomed Microdevices (2009) 11: 495. doi:10.1007/s10544-008-9255-7


Standard micro-fabrication techniques which were originally developed to fabricate semi-conducting electronic devices were inadvertently found to be adequate for bio-chip fabrication suited for applications such as stimulation and recording from neurons in-vitro as well as in-vivo. However, cell adhesion to conventional micro-chips is poor and chemical treatments are needed to facilitate the interaction between the device surface and the cells. Here we present novel carbon nanotube-based electrode arrays composed of cell-alluring carbon nanotube (CNT) islands. These play a double role of anchoring neurons directly and only onto the electrode sites (with no need for chemical treatments) and facilitating high fidelity electrical interfacing–recording and stimulation. This method presents an important step towards building nano-based neurochips of precisely engineered networks. These neurochips can provide unique platform for studying the activity patterns of ordered networks as well as for testing the effects of network damage and methods of network repair.


Carbon nanotubesElectrodesStimulationCircuitNeurochip

1 Introduction

Recent studies have suggested the great potential of high-density, carbon nanotube (CNT) coated surfaces as an interfacing material with neural systems (Bekyarova et al. 2005; Gabay et al. 2007; Gabay et al. 2005; Hu et al. 2004; Lovat et al. 2005; Mattson et al. 2000; Mazzatenta et al. 2007; Sorkin et al. 2006; Zhang et al. 2005). Foremost, CNT surfaces act as an extremely efficient biocompatible substrate on which neurons adhere and proliferate. The pioneering work of Mattson et al. demonstrated that neurons can attach and grow on chemically modified CNT coated surfaces. Later studies confirmed that pristine CNT surfaces support cell adhesion and viability even without any surface modification (Gabay et al. 2005). Subsequent studies revealed that CNT coated surfaces can facilitate cell patterning, network engineering (Gabay et al. 2005), guided neurite growth (Zhang et al. 2005) and even boost neuronal electrical activity (Lovat et al. 2005). These findings complement a wealth of experimental results with other nano-patterned surfaces, suggesting a strong cellular sensitivity to nano scale topographical patterns (Craighead et al. 2001; Dowell-Mesfin et al. 2004).

The fact that, in addition to being biocompatible (Hu et al. 2004; Webster et al. 2004), CNTs are also electrically conducting and can be seamlessly integrated into micro-fabricated devices opens up new and exciting prospects in the realm of neurochips (Breckenridge et al. 1995; Grattarola and Martinoia 1993; Jimbo and Kawana 1992; Nguyen-Vu et al. 2007; Stenger et al. 2001; Wheeler et al. 1999; Yu et al. 2007). Indeed, previous studies, carried out in our group and by others, have demonstrated that CNT decorated micro electrodes are very promising as electro-chemical electrodes for neuronal recording and stimulation applications (Gabay et al. 2007; Wang et al. 2006).

Here we demonstrate, for the first time, a direct, unmediated electrical interfacing between pristine CNT micro-electrode array (CNT-MEA) and cultured neurons. The CNT coatings function as an adhesive, high specific capacitance interface material. Unlike conventional electrode materials, the islands of the CNT coatings are cell-alluring—the neurons and glia cells have high affinity for selective attachment to the islands. Hence the CNT electrodes automatically control the cells and the network arrangement in addition to facilitating electrical interfacing.

2 Experimental

2.1 Cell culturing

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, Kibbutz Beit Haemek, Israel, Cat. No. 03-046-1) in phosphate buffered saline (PBS) (Biological Industries, Kibbutz Beit Haemek, Israel, Cat. No. 02-023-1) for 15 minutes, followed by mechanical dissociation by trituration. Cells were re-suspended in a modified essential medium with Eagle’s salts (Biological Industries, Kibbutz Beit Haemek, Israel, Cat. No. 01-025-1), 5% horse serum (Biological Industries, Kibbutz Beit Haemek, Israel, Cat. No. 04-004-1), 5 mg/ml gentamycin (Biological Industries, Kibbutz Beit Haemek, Israel, Cat. No. 03-035-1), 50 μM glutamine (Biological Industries, Kibbutz Beit Haemek, Israel, Cat. No. 03-020-1) and 0.02 mM glucose (BDH, Cat. No. 101174Y), and plated onto the CNT patterned substrates at a density of 700 cells/mm2. To promote the long-term survivability of the cells on the CNT islands it was crucial to use a “feeder” colony of cells. To do so, a PDL-coated (Sigma, Cat. No. p7889) thin disk of poly-dimethylsiloxane (PDMS) was placed around the CNT patterned area. The surrounding feeder culture on the disk covered approximately 75% of the total neuro-chip area and did not directly contact the CNT patterned culture. The cultures were maintained at 37°C with 5% CO2 and 95% humidity. The growth medium was partially replaced every 3–4 days.

2.2 Chemical inhibition

Control inhabitation tests were carried out to validate the biological nature of the electrical signals measured. Inhibition of electrical activity in the neuronal network was performed by applying both a-amino-3-hydroxy-5-methylisoxazole-4-propionic acid (AMPA) and N-methyl d-aspartate (NMDA) receptor antagonists (20 μM 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) (Sigma, Cat. No. C239) and 100 μM (2R)-amino-5-phosphonovaleric acid (APV) (Sigma, Cat. No. A5282)) to the medium.

2.3 Electron microscopy

Scanning electron microscopy observations were preformed as follows: samples were fixated for 30 min (37°C) in PBS with 2.5% glutaraldehyde (Fluka, Cat. No. 49629) and 4% sucrose. The fixed samples were then dehydrated by rinsing for 5 min in increasing concentrations of ethanol (25%, 50% and 75%), keeping the sample covered with each of the ethanol solutions, followed by 10 min rinses with 96% and 100% ethanol solutions. Finally, the dehydrated samples were critical-point dried using a Balzers Union critical-point drier and chrome coated (6 nm layer, Emiteck K575X, SEM coating unit). The samples were examined using a JEOL 6700F high resolution scanning electron microscope (HRSEM).

2.4 Immunostaining

Immunostaining for glial cells and synapses was performed using the following procedure: samples were washed in PBS and fixed with 4% PFA (Paraformaldehyde), 4% sucrose solution for 20 min. Next, they were perforated and blocked with 0.25% triton X-100 (Sigma, Cat. No. T8787) with 10% NGS (normal goat serum)(Biological Industries, Kibbutz Beit Haemek, Israel, Cat. No. 04-009-1) in PBS for 20 min followed by an additional block using PBS solution with 10% NGS for another 20 min. Samples were then washed with 1% NGS in PBS solution and incubated with primary antibodies in a solution containing 1% NGS in PBS overnight at 4°C. To detect glial cells, mouse anti-glial fibrillary acidic protein (GFAP) monoclonal antibody, diluted 1:400 (Biotest, Cat. No. MAB3402), was used. Synapses were immunostained using rabbit anti-synapsin I polyclonal antibody, diluted 1:300 (Biotest, Cat. No. AB1543). After incubation, the samples were washed three times with PBS and incubated for 1 h (in the dark) with the secondary antibodies: Alexa fluor 488 goat anti-mouse IgG, diluted 1:800 (molecular probes, Cat. No. A-11029) for detection of GFAP, and Alexa fluor 546 goat anti-rabbit IgG, diluted 1:600 (molecular probes, Cat. No. A-11035) for detection of synapsyn-I. Finally, samples were mounted using a mounting medium (Sigma, Cat. No. G0918) and covered with a cover slip. The mounting medium was dried overnight at 4°C before fluorescence measurements. Confocal laser scanning microscope images were obtained using LSM 510 META NLO (Zeiss). LEXT OLS3100 (Olympus) confocal microscope was used to construct three-dimensional images of neuronal networks on the neurochips.

2.5 Electrophysiological recording and stimulation

Extra-cellular recording were conducted utilizing low noise pre-amplifiers board (B-MEA-1060, amplifier, gain ×1,200 with a band-pass filter of 200 Hz to 5 kHz, by Multi Channel Systems). The signals collected from the microelectrodes were sampled at a 10 kHz sampling rate and stored on a personal computer equipped with a 128-channel, 12-bits data acquisition board (MC_Card, Multi Channel Systems) and a MC_Rack data acquisition software (Multi Channel Systems—version 3.2.20). Focal electrical stimulations were applied to individual electrodes by delivering voltage pulses using a stimulus generator (STG1008, Multi Channel Systems). All pulses were biphasic (positive-then-negative) pulses, 500 mV in amplitude, with each phase lasting 400 μs. A stimulation session was composed of 60 pulses separated by 20 s intervals.

3 Results and discussion

To construct the CNT-MEA we utilized conventional micro-fabrication techniques combined with standard CNT chemical vapor deposition (CVD) synthesis method. The entire fabrication process was described in detail in a previous publication (Gabay et al. 2007). Briefly, underlying TiN lines are used as conducting tracks. These lines are passivated with sputtered Si3N4 which is later removed at the regions of the active electrode using a reactive ion etch step. A thin nickel layer is e-beam evaporated at the openings. The process is concluded with a CNT thermal chemical vapor (CVD) deposition growth procedure utilizing the nickel as a catalyst material. The complete device is presented in Fig. 1(a).
Fig. 1

The CNT based neuro-chip (a). The CNT based multi electrode array is fabricated using standard optical lithography combined with chemical vapor deposition process to grow the CNTs. The resulting chip includes passivated interconnecting TiN lines (bright thin lines) and CNT coated electrodes (dark disks). Electrode diameter is 80 μm. (b) A high resolution scanning electron microscope (HRSEM) image of an isolated 20 μm CNT island revealing the extremely rough morphology of the surface

Substrates with isolated CNT islands of 20 and 80 μm in diameter were also fabricated in a similar fashion and were used to test the interaction between the cells and the CNT surface (Fig. 1(b)). To perform electrical recordings from cultured networks using CNT-MEA chips, clean silicon chips were bonded to printed circuit board (PCB) supports and were adjusted with quartz tubes to contain the biological medium. In order to validate the electro-chemical properties of the CNT electrodes cyclic voltammetry (CV) measurements were conducted (Gabay et al. 2007). The CV data provided direct evidence for the success of the CNT growth process in promoting high interface capacitance, similarly to large scale CNT electrodes (Barisci et al. 2000; Chen et al. 2002; Li et al. 2002; Liu et al. 1999). The exceptionally high surface area of the CNT electrodes also facilitates a high charge injection limit of 2 × 10−3 C/cm2, measured by applying a current pulse of 2 mA for 50 ms on a CNT micro-electrode immersed in PBS (David-Pur et al. 2008).

The nature of the interface between the CNTs and the cells was first tested by culturing cortical neurons of rats on isolated CNT islands. Cells were cultured onto the chip surface and were allowed to adhere and develop for several days in an incubator. After several days neurons had aggregated and accumulated at the CNT coated regions, and the cell density on the CNT-free regions was very low (Fig. 2). Apparently, the entangled, three-dimensional CNT matrix provides neurons and glia cells with an appropriate bed for neurite development and cell adhesion. The propensity of cells to adhere to such surfaces is associated with the three-dimensional nature of the surface and is discussed in detail in a separate publication (Sorkin et al. 2008).
Fig. 2

Cells cultured on substrates with isolated CNT islands adhering preferentially onto CNT islands. HRSEM imaging (a, b) show two islands on the same sample. Both neurons (a) and glia cells (b) appear to adhere directly on to the CNT surface. Extensive neurite branching is also apparent (a). The arrangement of cell clusters on the CNT surface is typified by an underlying glia cell layer and mostly overlaying neurons with synapses clearly visible under the glia layer. This arrangement was validated using confocal fluorescence microscopy with glia cells (green) and neuronal synapses (red) specifically stained (c)

The HRSEM images in Fig. 2 clearly show isolated neuron-like cells positioned directly on the CNT matrix, forming close contact with the surface. Extensive neurite branching is also apparent. It should be noted that the morphology of glia cells on CNT surfaces is conspicuously different: unlike the three dimensional structure of the neurons, isolated glia cells appear to spread as thin carpets over the CNT surfaces. Additionally, due to the strong propensity of neuronal cells to aggregate in locally high cell densities, some CNT islands may be coated with clusters of glia and neurons containing several tens of cells.

To allow an efficient interface between the electrodes and the cells for recording applications, a most fundamental issue is the arrangement of the different cells (i.e. neurons and glia cells) on these rough surfaces. In fact, the specific manner by which neurons and glia cells are arranged on the underlying CNT material is crucial in determining the properties of the interface. Whether neurons in the tissue are intimately connected to the CNTs or adhere to an underlying glial tissue may have a critical impact on the properties of the interface. A double layer arrangement (underlying glia cells and overlying neurons) may hinder surface-specific effects on the neuronal elements. On the other hand, imposed and unnatural cell arrangements may negatively affect the activity of the cultured network.

The true mapping of the arrangement of mixtures of glia and neuronal cells is revealed using confocal fluorescence microscopy. Glia cells and synapses were specifically labeled and imaged (Fig. 2(c)). The affinity of the glia cells to the surface is readily seen, as is also the localization of the synapses at the CNT areas. This extensive formation of synapses provides further support to the notion that these CNT islands are a suitable substrate for network development. Cross sectional cuts provide direct evidence that while glia cells do often reside at the interface with the CNTs, neuronal processes are clearly found in direct contact with the CNT surfaces. These results demonstrate that neurons attach either directly to the rough surface or to an underlying layer of glia cells. Additionally, synapses can be clearly identified under the glia cell layer. These results provide a direct indication to the intimate contact between the CNTs layer and the cells.

While CNT electrodes are indeed very efficient electro-chemical electrodes as well as excellent substrates for neuronal culturing, the novelty of the CNT-MEA and its cardinal advantage, for in-vitro applications, rests in the unique manner by which neurons and glia cells, at appropriate densities, can self-assemble into ordered networks with pre-designed geometry and topology (Fig. 3). At specific cell culturing densities (Gabay et al. 2005), neurons and glia migrate towards the CNT electrodes and form small interconnected clusters, leaving behind taut bundles of axons and dendrites to connect neighboring clusters. The formed networks are highly organized, with geometries which faithfully follow the pattern of the CNT nodes. The quality of the network engineering depends on cell density, electrode geometry as well as the overall cell number in the entire dish. Optimized cell density yields cultured networks with compact engineered wiring following the electrode layout (Sorkin et al. 2006). Since the initial cell density is not entirely uniform on the chip during cell plating, high cell densities at some areas may result in large cell clusters extending more than a single CNT island (Fig. 3(a)).
Fig. 3

Self-assembled neuronal patterned network on a CNT neuro-chip. After several days of plating, the cells on the CNT neuro-chip interconnect to form networks with cells aggregates at the CNT islands and cell-free connections in-between (a, b). Cell plating density was around 1 × 10−3 cells/μm2. Images were obtained using a confocal microscope after fixation and drying

Once the neurons had self-organized into a connected circuit, in complete adherence to the electrode layout, the electrodes were used to directly record electrical activity with very high fidelity. Extra-cellular electrical activity measurements were taken from neurons on individual 80 μm electrodes after 12 days in vitro (DIV). This activity was maintained for time periods of up to 60 DIV. To verify that the activity recordings were of biological nature, we applied blockers of excitatory synapses. Simultaneous application of 20 μM CNQX (6-cyano-7-nitroquinoxaline-2,3-dione) and 100 μM APV (((2R)-amino-5-phosphonovaleric acid) completely blocked all the electrical activity in the network. This effect was reversible as the activity was regained following the wash of the blockers. Due to the relatively large size of the electrodes, each electrode could record the integrated activity of a clustered sub-population of several neurons. This activity was characterized by bursting events; short time windows (several hundreds of milliseconds long) of rapid collective neuronal firing, which were followed by long intervals (seconds) of sporadic firing (Fig. 4(a),(b),(c)). For networks with limited inter-cluster connectivity, bursting activity was mainly confined to each cluster, and only weak correlation in activity was observed between clusters (Fig. 4(c)). All the examined cultures showed similar patterns of bursting activity (115 electrodes from six cultures). Interestingly, the collective activity patterns observed in single CNT electrodes resemble the synchronized bursting events (SBEs) observed in uniform networks, recorded using PDL-coated commercial MEA electrodes (Ayali et al. 2004; Segev et al. 2002; Segev et al. 2003). This implies the effectiveness of the CNT electrodes in separating the network into well-defined sub-networks while retaining the hallmark of large network activity. In this respect, neural networks on CNT electrodes present an additional hierarchy in the bottom up approach for studying neuronal network, by enabling the monitoring of interaction between loosely connected neuronal sub-populations. Preliminary investigations with strong inter-cluster connectivity show clear inter-cluster signal correlation and will be reported separately.
Fig. 4

Spontaneous and stimulated electrical activity of neuronal clusters on CNT electrodes. (a, b) Voltage traces of spontaneous electrical activity recorded from a CNT electrode. (c) Raster plot of the spontaneous spiking activity in several CNT electrodes. Activity patterns are characterized by bursting events; short time windows (several hundreds of milliseconds) of rapid collective neuronal firing, which are followed by long intervals (seconds) of sporadic firing. (e) Activity response to 40 consecutive electrical stimulations of neurons on a CNT electrode. Each row represents the response of two neurons (red and blue) to one stimulation (marked by the black line) in an adjacent electrode. Out of the total 100 stimulations applied to the stimulated electrode, about half successfully triggered a response. The activity patterns in the stimulated electrode were marked by the spiking of single neurons (d) which greatly varied between consecutive stimulations. Nevertheless, repeating activity motifs such as spike-pairs from different neurons were identified (d, e). Spikes in raster plots were extracted using a spike sorting and detection algorithm (Hulata et al. 2002)

Individual CNT electrodes were additionally used to locally stimulate neurons on these electrodes. Biphasic voltage pulses were applied between two adjacent CNT electrodes, triggering a response in an intermediate electrode (Fig. 4(e)). In order to allow easy discrimination between the spontaneous and evoked activity in the stimulated electrode, an electrode exhibiting low activity levels (recording from only two neurons) was selected. Following a stimulation session, the response activity was marked by the spiking of the recorded neurons. The response activity lasted relatively long, about 300 ms. The exact response pattern (temporal order of the neuron iring) varied between consecutive stimulations (Fig. 4(e)). Nevertheless, some activity motifs, which were consistent over the whole stimulation session, could be identified. Utilizing a spike sorting algorithm (Hulata et al. 2002) it was possible to distinguish between the different spikes according to their distinct sources. An example of such a separation is shown in Fig. 4(d) and (e), in which a repeating motif of a large spike followed by a small spike was identified. The response activities to the stimulation sessions in addition to the ability to record the spontaneous activity exemplify the CNT electrode’s capability to bi-directionally electrically interface neuronal clusters.

4 Conclusions

To conclude, this work presents a new and complete approach to engineer and interface with electrically viable neuronal systems. Each micro-electrode in the new scheme is coated by a layer of several microns of dense and entangled CNTs, synthesized by a CVD process thus forming a CNT island. The islands strongly attract and anchor cells to pre-defined locations, and enable the formation of stable sub-networks on electrically active recording sites. Efficient cell patterning results with a stable neuronal network even though no adhesive agents were used. Low electrode impedance improves the electrochemical interface, and contributes to high quality recording and efficient stimulating signals.

The electrical viability of the cell cultures on the CNT substrates, and their long term survivability (up to 2 months), substantiate the biocompatibility of these surfaces, in agreement with previously reported results. Combined with their superior electrical performances, it was demonstrated here that CNT coated electrodes are, in fact, well-suited to assist the interfacing between electrically active biological cells and conventional electronic systems. The added advantage of network patterning provides a unique opportunity to form consistent, pre-defined networks. The study of signal propagation and the development of patterned networks with a single cell per electrode are currently underway. We expect that such CNT based neurochips can provide a valuable platform for studying network damage (e.g. by mechanical deletion of connections between islands), and for investigating network repair (e.g. by adding cells on specific islands).


The authors thank Inna Brainis for her technical assistance and Moti-Ben David, Itsik Kalifa, Itay Baruchi and Nadav Raichman for their assistance and useful discussions. This project was supported in part by a grant from the Israeli Science Foundation (1138/04) and by the Tauber Fund at Tel Aviv University.

Copyright information

© Springer Science+Business Media, LLC 2008