carbon nanotube films as flexible neural interfaces

Advances in neural interface technologies have sought to identify electroactive materials that are able to translate neural depolarisation events into digital signals or modulate neural firing through ionic or electrical stimulation with greater efficiency. An ideal material for neural recording and/or stimulation should possess low electrical impedance coupled with a high cathodic charge storage capacity (CSC C ), charge injection capacity (CIC) and electroactive surface area (ESA), as well as optimal mechanical biomimicry. In this study, we present the robustness of self-supporting CNT films as neural interfaces, combining advantageous electrical and mechanical properties with high cytocompatibility. Films were observed to possess a high CSC C (29.95 ± 0.91 mC cm -2 ), CIC (352 ± 5 µC V -1 cm -2 ) and ESA (0.908 ± 0.053 cm 2 ), low impedance (110 Ω at 1 kHz), low resistance (75 ± 13 Ω) and high capacitance (378 ± 9 µF cm -2 ), and outperformed Pt control electrodes. Self-supporting CNT films were also found to facilitate neuron growth and decrease the presence of reactive astrocytes in a mixed neural cell population. Self-standing CNT films were shown to be promising materials for the design of flexible and cytocompatible neural interfaces.


Introduction
By providing a direct communication route between neural tissue and a man-made digital system, neural interfaces can be used for recording and processing neural activity, or treating neurological disorders [1][2][3]. Advances in neural interface technologies have sought to identify electroactive materials that are able to translate neural depolarisation events into digital signals or modulate neural firing through ionic or electrical stimulation with greater efficiency [4]. In particular, neuromodulation systems employ stimulating neural electrodes which are implanted into a relevant region of the central or peripheral nervous system. In this way, they are used to deliver a specific prolonged and repetitive electrical signal to target tissues [5]. To provide efficacious and safe stimulation, the electric signal should be sufficient to trigger a physiological response, yet should not damage the peri-electrode tissue nor the electrode itself [6]. Consequently, most applied stimulation protocols employ biphasic current/potential pulses [6,7] with a width in the range from µs to ms [8,9] delivered with a frequencies ranging from 50 Hz to 300 Hz [10,11]. Recent studies, however, suggest that low frequencies can be clinically useful for treating symptoms of Parkinson disease, movement disorders and chronic pain [12][13][14][15], and current challenges in the design of neural interfaces are related to the fabrication of materials with optimal electrochemical properties at low frequencies.
Critically, the electrochemical impedance at the tissue-electrode interface should be as low as possible in order to increase the signal to noise ratio and decrease the potentials required to trigger a neuronal response [16]. Charge Storage Capacity (CSC), Charge Injection Capacity (CIC) and Charge Injection Limit (CIL) can be used to assess the amount of charge that can be injected during stimulation [17,18]. Coupled with the Electroactive Surface Area (ESA), a parameter used to describe the actual area of the interface that is able to transmit electric signal [19,20], the consideration of these descriptors are important in neuroelectrode design. Low electrical impedance coupled with high CSC, CIC and ESA represents an ideal material for neural recording and/or stimulation, facilitating low potential stimulation and high sensitivity neural recording.
Following implantation, however, the electrochemical properties of stimulating and recording electrode becomes compromised due to the onset of reactive gliosis, neural encapsulation [21] and chronic BBB disruption [22], a process initiated by mechanical trauma during electrode insertion [23], and exacerbated by chronic relative micromotion at the electrode-tissue interface [24]. Materials commonly used in neuroelectrode fabrication, i.e. crystalline silicon and noble metals, are rigid and exhibit Young's moduli from 70 to 180 GPa, about five orders of magnitude greater than that of i.e. brain tissue (1 kPa) [25]. Current approaches in neural engineering are focused on the design of implantable neural interfaces with optimal mechanical biomimicry, through the development of mechanically soft and flexible neuroelectrode arrays [26][27][28] which have been shown to reduce reactive gliosis in vivo and enhance chronic neuroelectrode functionality [29].
Significant efforts are underway on engineering materials with optimal electrical, biological and mechanical properties for the fabrication of robust electrically stimulating or recording neural interfaces. Notably, the application of nanotechnology and nanomaterials to design challenges in biomedical engineering have received much attention. Carbon nanotubes (CNT) [30,31], graphene [32,33] and metalic nanowires [34] have been explored as materials for neural tissue engineering, due to their tunable electronic properties, as well as unique mechanical and chemical characteristics [1,35]. In particular, CNTs have been widely studied as electrically active fillers for the fabrication of multifunctional composite materials [36][37][38][39][40][41][42][43]. Conducting CNT nanocomposites have been associated with ongoing limitations, principally related to challenges in the integration of CNTs within the supporting matrix, and in achieving a high CNT concentration. To address these challenges, self-supported, mechanically stable films made entirely of CNT aggregates have been explored as wearable and flexible conductors [44]. Due to their unique fabrication strategy [45] and beneficial electrical properties [44], CNT macroscopic films are promising candidates as neural interfaces. In this study, we present a self-supported, flexible CNT film as a neural interface material, through extensive characterization of the materials electrochemical, mechanical and cytocompatibility properties.

Fabrication
Self-supporting CNT films were fabricated from multi-wall CNT arrays formulated by chemical vapour deposition (CVD) of toluene and ferrocene according to procedures reported previously [45][46][47]. In short, to make a CNT film the material was sonicated (1h, 50% amplitude, Hielscher UP200St 200W sonicator) in iso-propanol in the presence of ethyl cellulose (1:1 weight ratio of polymer to CNTs). A uniform CNT dispersion was then spraycoated onto a Kapton foil, dried, detached from the substrate and annealed in a flame to remove residual ethyl cellulose. The procedure led to the formation of self-standing CNT sheets with the uniform thickness of 10 μm.

Chemical and morphological characterization
Raman spectra were recorded in the range of 100 cm −1 to 3200 cm −1 using inVia Renishaw Raman microscope at λ = 514 nm excitation wavelength. At least 50 accumulations were acquired for each sample for the data to obtain statistical significance. Transmission Electron Microscopy (TEM) images were acquired using FEI Tecnai F20 operating at 60 kV. Scanning Electron Microscopy (SEM) images were collected with a Hitachi S-4700 Scanning Electron Microscope operating at 15 kV.

Electrochemical characterization
Electrochemical studies were performed by means of a PARSTAT 2273 potentiostat in a three-electrode set-up, comprising a CNT film or Pt foil (0.1 mm thickness, 99.9 % purity, produced by Mennica-Warsaw, Poland) as a working electrode, Ag/AgCl as a reference electrode and glassy carbon rod as an auxiliary electrode. Cyclic voltammograms (CVs) were collected in 0.1 M KCl solution, within the potential range from −0.8 to 1.0 V (vs. Ag/AgCl) at 100 mV s −1 for 5 CV cycles. CVs were used to compare the electrochemical behaviour of materials and to determine their cathodic Charge Storage Capacity (CSC C ), calculated as the electric charge integrated under corresponding CV curve during the cathodic part of a CV cycle, according to the formula: where t 1 is the beginning of CV cycle, t 2 is the end of CV cycle, and I is the current where i p is the reduction/oxidation peak current (A), n is the number of electrons contributing to the redox reaction, A is the area of the electrode (cm 2 ), C is the concentration of Fe(CN) 6 4− in the bulk solution (mol cm −3 ), D is the diffusion coefficient of Fe(CN) 6 4− in KCl solution (6.3 × 10 −6 cm 2 s −1 [50]), and υ is the scan rate (V s −1 ). The measurements were performed in triplicate, the results were expressed as a mean ± standard deviation.
EIS spectra were collected in 0.1 M KCl solution within a frequency range from 100 mHz to 100 kHz, with AC amplitude of 40 mV (vs. Ag/AgCl) and DC potential equal to 0 V (vs. Ag/AgCl). The results were presented on Bode plots and compared to those of a planar Pt foil electrode. The data fitting analysis was performed using EIS Spectrum Analyzer 1.0 software with the application of the Powell algorithm. In order to assess the goodness of fit, χ 2 were calculated and restricted not to exceed 2 % relative deviation of the calculated spectrum from the measured data.
Capacitances were calculated basing on the parameters of constant phase element (CPE) according to the formula [51]: where C is the capacitance (F), R is the film resistance (Ω), Q 0 and n are CPE parameters. Charge Injection Capacity (CIC) was studied by integration of chronoamperometric curves used to simulate the parameters of neural stimulation [6][7][8][9][10][11], i.e. a single biphasic potential pulse consisting of a 5 ms application of a reduction potential followed by a 5 ms application of an oxidative potential. The potentials were kept in the range from -1.2 V to 1.2 V (vs. Ag/AgCl). The experiments were performed in 1x PBS solution.
All electrochemical experiments were performed in triplicates. The results were expressed as the mean of the values ± standard error of the mean.

Mechanical characterization
To gauge how flexible the CNT films are, a previously reported routine was employed [52]. The specimens (10 mm × 60 mm) were subjected to a three-point bending (Dynamic Mechanical Analysis, TA Instruments Q800DMA) during which electrical resistance was monitored by an on-line source meter (Keithley 2450) connected to a PC. The selected measurement parameters were: 2 bending cycles per minute (1/30 Hz), 1 mm amplitude, 960 cycles (8 h). The variation of electrical resistance as a function of the number of bending cycles gives information about the durability and mechanical integrity of the CNT films when subjected to bending. The measured resistance was normalized to the initial resistance to demonstrate the relative change of the resistance in response to mechanical loading.

Biological characterization
Primary cultures of a mixed neural population were obtained from the mesencephalon of embryonic Sprague-Dawley rats according to a procedure described previously [53,54]. To visualize neuron and astrocyte cell populations, indirect double-immunofluorescent labelling was applied as described previously [54,55].
An Olympus Fluoview 1000 Confocal Microscope was used to view immunostained samples at a fixed scan size of 1024 × 1024 at a ratio 1:1 and 60× magnification. Cell density was analyzed by counting the number of nuclei corresponding to neurons and astrocytes in an area of 211.97 μm × 211.97 μm in at least 20 random images taken from test and control groups [55]. The quantification of astrocyte area was through ImageJ software (NIH) [56]. The quantification of neurite length was through stereological methods as reported previously [57]. Nine random fields of view from three different technical replicas from five different biological samples were analysed. The average neurite length was calculated according to the formula [58]: where: L is neurite length (μm), n is the number of times neurites intersect with grid lines, T is distance between grid lines (μm).
The biological experiments were conducted to include five biological replicas for both test and control groups. The results were expressed as the mean of the values ± standard error of the mean. A t-test was performed to determine the statistical significance (p < 0.05).

Chemical and morphological characterization
A fast and convenient method of fabrication led to the formation of robust self-supporting, flexible CNT films with a uniform thickness of 10 µm. SEM micrographs of a CNT film (Fig.  1a) did not reveal the presence of any obvious amounts of contamination in the form of amorphous carbon or residual catalyst. As expected, the films prepared by our method were isotropic with entangled CNTs and their bundles pointing in every direction. Closer investigation by TEM (Fig. 1b) showed that the synthesized CNTs were multi-walled with a diameter of 20 nm. The outermost layer of individual CNTs contained some amorphous carbonaceous deposits, but the underlying CNTs were of reasonably high level of graphitization. This had a clear influence on the shape of the collected Raman spectra (Fig.  1c), wherein the I D /I G ratio reached 0.50 (the intensity of defect-induced D band to the G mode signalling the graphitic lattice is indicative of the quality of the sample). In our case, the I D /I G ratio was elevated by the presence of an extraneous surface coating commonly present on as-made nanocarbon materials [59].

.2 Capacitive behaviour
Neural interfaces can be divided into two group depending on the mechanisms of electron transfer [60]: capacitive materials involve charging and discharging of the electrodeelectrolyte double layer, while for other materials, faradaic reactions mediate the flow of electrons. Because faradaic reactions require some species to be oxidized or reduced, which can lead to the degradation of electrode, capacitive materials are more desirable as neural interfaces [60]. To assess whether the material acts through the capacitive or faradaic mechanism, its electrochemical behaviour should be examined e.g. via cyclic voltammetry (CV).
CV curves of bare Pt and CNT film presented in Fig. 2a show the lack of a distinctive redox peak system within the potential window from -0.8 V to 1.0 V (vs. Ag/AgCl), recognized as the water window for a CNT film [61]. A progressive increase in the anodic current starting from 0.4 V (vs. Ag/AgCl), observed in the CV of a CNT film, should be associated with the abundance of defects and amorphous carbonaceous deposits, which is in accordance with the results acquired through Raman spectroscopy and previous literature reports [62]. Both Pt and CNT film can be therefore considered as capacitive materials. For this type of materials, the parameter that is used to quantitatively assess their performance is the cathodic Charge Storage Capacity (CSC C ). Materials possessing high CSC C , calculated through the integration of CV curve with respect to time (equation 1), are able to store relatively large charges before reaching overpotentials where irreversible faradaic reactions occur [63]. CNTs are well known for their ability to accumulate charge [64]. This has been verified extensively with CNT/conducting polymer composites, i.e. PEDOT/PSS/MWCNT (CSC C = 8.6 mC cm −2 ) [65], in which CNTs were present as a conducting filler component. This charge accumulation was also noted here for self-supporting CNT films, for which the CSC C (29.95 ± 0.91 mC cm -2 ) was fourteen times higher relative to bare Pt foil electrodes (2.17 ± 0.10 mC cm -2 ), indicating their highly capacitive nature. It is worth noting that CNT films can be easily modified to further increase their CSC C , e.g. through electrochemical deposition of conducting polymers [66].
The increase in CSC C with respect to Pt is partially related to its elevated Electroactive Surface Area (ESA), describing the total of the surface which is accessible to the electrolyte and can participate in the electron transfer process [19]. Calculated based on the CV curve of a redox probe, K 4 [Fe(CN) 6 ] (Fig.2b), and with the use of Randles-Sevcik equation (equation 2), the ESA of CNT films (0.908 ± 0.053 cm 2 ) was increased by a factor of 4x relative to Pt control foils (0.238 ± 0.035 cm 2 ) of the same geometrical dimensions. This maximization of the electrochemically active surface area over a given geometric space resulted in a drastic increase in the capacitance of the material and reduction in electrochemical impedance, which is beneficial for both neural stimulation and recording [67]. By comparing the currents of the anodic and cathodic peaks of a redox probe at different scan rates (Fig.2c), the linear dependency with respect to the square root of scan rate can be observed, indicating a diffusion-limited process with no traces of undesired adsorption of the redox probe onto the surface of CNT film [68,69]. Even when the current density of CV curves is expressed in terms of ESA instead of the geometrical area of the electrodes (Fig.S1), self standing CNT films still outperform bare Pt control, indicating that their ability to accumulate charge is only partially derived from their expanded surface area.
Here Figure 2

Charge injection capability
In order to negate off-site stimulation or tissue damage, implantable electrodes utilized in neuromodulation and neural stimulation devices should deliver a stable, constant electrical signal for the duration of a stimulating pulse. A cathodal-led, biphasic potential pulse was used to examine the behaviour of the CNT film electrodes relative to control Pt foil electrodes while electrically stimulated (Fig.3a). It was observed that the current recorded with Pt electrodes was not stable, and its magnitude decreased when the potential signal was applied. Following 2.5 ms of applied potential, the current density decreased by 40% relative to that recorded at the onset of stimulation. What is more, a residual current could be noted on the Pt foil electrode after the stimulation pulse was terminated. This kind of electrode instability, derived from the polarization of the electrode, and considered as highly undesirable, has been noted previously for iridium and iridium-oxide alloys electrodes [70].
Conversely, CNT film electrodes demonstrated a current response higher than that of Pt, which was observed to remain constant for the pulse duration. Upon termination of the pulse, the current density abruptly dropped to zero. The integration of current with respect to time gave the Charge Injection Capacity (CIC), which was substantially higher with CNT film electrodes (352 ± 5 µC V -1 cm -2 ) relative to control Pt foil electrodes (138 ± 12 µC V -1 cm -2 ) [71]. It was also observed that the current density and CIC were linearly dependent on the stimulating potential (Fig.3b&c). Since the water window for both electrodes was determined as in the range from -0.8 V to 1 V (vs. Ag/AgCl), the maximum CIC that can be applied without triggering water electrolysis [72], namely Charge Injection Limit (CIL), is 352 ± 5 µC cm -2 for CNT film and 138 ± 12 µC cm -2 for a bare Pt electrode. Critically, neural implants employ charge densities ranging from 2.3 µC cm -2 to 2.3 mC cm -2 [60], and electrode materials with a high CIC allow for the use of low electrical potentials without compromising the therapeutic effects of stimulation. Indeed, the CIC of 352 ± 5 µC V -1 cm -2 , as observed with self-supported CNT film electrodes, indicates that therapeutic stimulation could be obtained through a 10 ms potential pulse with a potential of 70 mV [18].

.4 Electrochemical Impedance analysis
Since neuronal signals generally have a very small amplitude, low impedance is a crucial parameter in the development of neural interfaces designed to interface with the central or peripheral nervous system for electrical recording purposes [73]. Although possessing low impedance at high frequencies, metals such as platinum and platinum-iridium exhibit increased impedance at frequencies below 1 kHz. In contrast, the CNT film electrodes demonstrated a stable, low impedance module profile even at low frequencies (> 1 Hz) (Fig.4a), reaching the impedance module of only 430 Ω at 0.1 Hz, almost two orders of magnitude lower than for a Pt control electrode (12 kΩ) at the same frequency. Apart from the low impedance modules, CNT films were also characterized with the low values of the real component of the impedance (Z re ), significantly decreased when compared with Pt control electrodes (Fig.S2). Since Z re is directly related to the electrochemical noise of the electrode/electrolyte interface, the decrease in the value of this parameter should result in the more precise collection of neural signals [74].
A substantial shift in the peak frequency in the phase diagram of CNT film (Fig.4b) indicated that the difference in the impedance profile is derived from the capacitive behaviour of this material at frequencies below 10 Hz. Experimental data were subsequently fitted to a simple Randles circuit presented in Fig.4c, which comprised a parallel combination of solution resistance (R s ), charge transfer resistance (R), constant phase element (CPE) and Warburg diffusion element (W s ) [75,76]. The constant phase element replaces typically used doublelayer capacitance, showing that the surfaces of both electrodes behave as imperfect capacitors. The constant phase element is in parallel with the resistance R representing the combination of a charge transfer resistance and polarization resistance, which are the intrinsic properties of the electrode material and can be used to assess its conductivity. Furthermore, low polarization resistance is of a significant importance for the stability of the electrode, which was discussed in the previous section. The simulated data (Tab.S1) confirmed that CNT film electrodes are characterized by low resistance (75 ± 13 Ω) and high capacitance (378 ± 9 µF cm -2 ), relative to Pt control electrodes (R = 703 ± 151 Ω, C = 18 ± 1 µF cm -2 ). Interestingly, the electrical resistivity and capacity of experimental CNT film electrodes were superior to materials currently being explored in neural interfaces, such as polypyrrole (R = 6.6 ± 0.1 kΩ, C = 6.6 ± 0.2 µF cm -2 ) [77] and PEDOT (R = 4.2 ± 0.3 kΩ, C = 13.8 ± 0.1 µF cm -2 ) [77]. All analysed electrical and electrochemical properties of CNT film and Pt foil electrodes are presented in Table.1.
Here Figure 4 and Table 1 3.5 Mechanical characterization Self-supporting CNT films have been already explored as wearable and flexible conductors [44]. As shown in [45], various CNT films possess a tensile strength in the range of 85 to 122 MPa, and can be easily bent or strained (Fig.5a) without any loss of performance. To investigate the flexibility of CNT films, the electrical resistance was monitored while films were subjected to repetitive bending (Fig. 5b). After an initial increase of electrical resistance by about 2%, we did not observe any further decrease of conductivity, which remained relatively constant over time. Selected high bend angle may have caused some CNTs and their bundles to lose contact, and so the electrical resistance increased. It is important to note that the bend angles selected for the study represented extreme mechanical conditions.

.6 Biological characterization
To verify the cytocompatibility of experimental CNT film electrodes, primary ventral mesencephalic (VM) mixed cell population was seeded onto the surface of self-supporting CNT films and Pt foil control electrodes, and cultured for 3, 7 and 14 days. At each time point, cells were fixed and processed for immunocytochemical analysis. The cell nuclei (DAPI), neurons (anti-β III tubulin) and astrocytes (anti-GFAP) were stained and cell populations were imaged using scanning confocal microscopy (Fig.6a). Although mixed cell cultures contained oligodendrocytes and microglia populations [78], astrocytes staining was performed in this study to assess the effects of self-supporting CNT films on astrocyte proliferation, neuronal support cells which become activated in response to foreign body implantation [79,80]. Critically, in a mixed population of neural cells, the ratio between astrocytes and neurons can be used to assess in vitro the cytocompatibility of a material in neural systems. The neuron-to-astrocyte ratio (Fig.4b) show the prevalence of neurons after 3 days in culture for both CNT films and Pt control. After 7 days, the ratio of neurons to astrocytes is normalized for Pt but it is increased on self-supporting CNT films. This increased presence of neurons with respect to astrocytes may indicate that this surface promotes neurite viability or adhesion, as described in previous studies [81,82]. Following 14 days of culture, the neuron-to-astrocyte ratio equilibrates for both CNT films and Pt control and equals approximately 1:1.
The differences between CNT film and Pt control materials are observable when the average neurite length of neurons and the mean size of astrocytes are compared. Following 14 days of culture, the average length of neurons when cultured on CNT films was 866 ± 7 µm and only 528 ± 6 µm when cultured on Pt control materials, indicating that self-support CNT films facilitated neurite outgrowth. Furthermore, the mean area of astrocytes soma was significantly lower in cells cultured on self-supporting CNT films (637 ± 54 µm 2 ) relative to cells cultured on Pt control foils (839 ± 83 µm 2 ). This difference was significant as an increased cell area is characteristic of astrocyte activation, as seen with reactive gliosis and other neuropathologies [83].
Here Figure 6

Conclusions
In this study, the self-supporting CNT films are presented as advantageous materials for the design of neural interfaces. This flexible material made entirely from carbon moieties was found to possess a highly capacitive nature, elevated electrically active surface area and superior charge injection behavior, in theory, allowing for the use of low electrical potentials in neuromodulation approaches without compromising on the therapeutic effects of stimulation. EIS results confirmed that CNT film electrodes are characterized by low resistance and a high capacitance relative to Pt control electrodes and other materials currently being explored as neural interfaces. In contrast to Pt and PtIr electrodes, CNT films demonstrated a stable, low impedance profile at the frequencies below 1 kHz. Cytocompatibility studies indicated the increased presence of neurons with respect to astrocytes, suggesting that CNT film can promote neurite viability or adhesion. Moreover, CNT films were found to facilitate neurite outgrowth, as well as decrease the presence of reactive astrocytes in vitro, observed by the decrease in astrocyte cell area. CNT films demonstrated minimal loss of performance following repeated deformation, indicating the potential for this material for the development of flexible neural devices.