D-nanostructured boron-doped diamond for microelectrode array neural interfacing

The electrode material is a key element in the design of long-term neural implants and neuroprostheses. To date, the ideal electrode material offering high longevity, biocompatibility, low-noise recording and high stimulation capabilities remains to be found. We show that 3D-nanostructured boron doped diamond (BDD), an innovative material consisting in a chemically stable material with a high aspect ratio structure obtained by encapsulation of a carbon nanotube template within two BDD nanolayers, allows neural cell attachment, survival and neurite extension. Further, we developed arrays of 20-mm-diameter 3D-nanostructured BDD microelectrodes for neural interfacing. These microelectrodes exhibited low impedances and low intrinsic recording noise levels. In particular, they allowed the detection of low amplitude (10e20 mV) local-field potentials, single units and multiunit bursts neural activity in both acute whole embryonic hindbrain-spinal cord preparations and long-term hippocampal cell cultures. Also, cyclic voltammetry measurements showed a wide potential window of about 3 V and a charge storage capacity of 10 mC.cm 2, showing high potentiality of this material for neural stimulation. These results demonstrate the attractiveness of 3D-nanostructured BDD as a novel material for neural interfacing, with potential applications for the design of biocompatible neural implants for the exploration and rehabilitation of the nervous system. © 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
Neuroengineering more efficient neural interfaces is crucial to better explore neural networks [1e6] and to offer advanced clinical rehabilitation solutions based on neural prosthesis and braincomputer interfaces to target pathologies such as hearing loss [7], pathological tremors [8], visual impairments [9,10], or paralysis [11e15]. These devices make use of microelectrode arrays (MEAs) [16e19] to record the electrical activity from neural structures and can deliver electrical microstimulation in these structures to restore previously lost functions. One way to improve such systems is to increase their electrode number and density. For instance, increasing the number of recorded channels improves the efficiency of brain machine interfaces [14] and neural prosthesis [20]. In the case of retinal prosthesis, Nirenberg et al. have shown that driving a high density stimulator network with the retina's neural code could considerably rise the reconstructed image resolution [21]. Also, probing the electrical activity of neural circuits at multiple locations with miniaturized microelectrodes is expected to decipher their dynamics beyond characterizing their structural aspects [22e24]. However, increasing the density of microelectrodes implies the reduction of their dimensions [22,24e27], which has two major consequences [28]. First, small diameter electrodes lead to a high intrinsic noise level and thus low signal to noise ratio, precluding sensitive neural recordings. Second, charge injection capacity is reduced, hindering efficient electrical neural stimulation. Therefore, a current challenge is to find new biocompatible and more efficient electrode materials to optimize the electrodeetissue interface [3,29].
Boron doped diamond (BDD) is one of the materials of choice for electrodes of long-term neural implants since it combines properties of biocompatibility [30e39], chemical inertness [40], and structural stability at current charge densities typically applied for neural stimulation [41,42]. Previous studies have shown that BDD MEAs can successfully be used to resolve electrochemical signals with high-time resolution from whole living cells [43] or microareas of functional neuroendocrine cells [44]. Moreover, it exhibits a wide potential window in aqueous media [45], making it particularly suitable for neural stimulation [46e48]. Although BDD normally suffers from a low double layer capacitance and high impedance in comparison to other competitive materials for neural interfacing, Hebert et al. recently proposed an innovative solution where the specific surface area of BDD electrodes is increased using vertically aligned carbon nanotubes (VACNTs) as a template interlayer [49], a strategy in line with previous studies using nanostructuration strategies to enhance the electrode performance for neural recording and stimulation [28,50e55].
Moreover, the advantage of using nanostructures to modify the electrode geometry seems to be two-fold. First, in vitro studies have indeed evidenced that specific nanotopographies can support the sprouting of neurons from the central and peripheral nervous system without favoring the spread of glial cells [56e61]. Second, the use of nanostructures appears to be beneficial to interface neural tissues [61e65].
There has been only few attempts of neural recording using BDD microelectrodes [66e68], however yielding to rather high recording noise due to the high interface impedance and low double-layer capacitance of diamond. Moreover, Halpern et al. have shown that a single needle microelectrode coated with BDD could successfully be used to stimulated single neurons [46]. Such technology has however not yet been integrated into microelectrode arrays. Here, we investigated the possible use of 3D-nanostructured BDD as a material for efficient bidirectional neural interfacing in a configuration of 20-mm-diameter-microelectrode arrays. For this, the suitability of nanostructured BDD for culturing embryonic neural cells was first evaluated. We then developed a microfabrication process to design microelectrode arrays made of either conventional BDD or 3D-nanostructured BDD, consisting of vertically aligned carbon nanotubes (VACNTs) inter-layer template embedded in two BDD nanolayers. We finally evaluated the performance of both types of BDD microelectrodes for ex vivo neural tissue and in vitro cell culture recording and stimulation.

Methods
All experimental protocols conformed to recommendations of the European Community Council Directive of November 24, 1986 (86/609/EEC) and local French legislation for care and use of laboratory animals, and have been approved by the Bordeaux ethical committee for experimental research (Approval No A5012082).

Fabrication of BDD substrates
BDD films were produced on silicon substrates. Detonation diamond nanoparticules (6-nm) dispersed in water (0.1% wg) were spread by spin coating on these substrates. Then a 600-nm-thick nano-crystalline diamond film was grown at low temperature in a AX6500X Microwave Plasma Chemical Vapor Deposition (MPCVD) system (Seki Technotron Corp.). The growth was performed during 10 h under the following parameters: MW power of 2.6 kW, temperature of 650 C, pressure of 26 mbar, gas concentration: 1% methane in approx. 99% hydrogen. Trimethylboron was added to the gas phase as dopant so that the resulting Boron concentration in the diamond film was approximately 2.10 21 at.cm À3 as determined by Secondary Ion Mass Spectrometry measurements. This was found to be optimum in terms of electrode performance [69]. Surfaces were then oxygenated using an ozone treatment during 2 h in order to obtain an oxygen-terminated BDD and thus a hydrophilic surface.

Fabrication of 3D-nanostructured BDD substrates
VACNTs were grown on BDD films, and then coated with a second BDD film. The full fabrication process has been described previously [49]. In brief, a 7 nm nickel layer was deposited on the diamond surface by e-beam evaporation and turned into 50 nm nanoparticles by heating at 700 C for 3 min. The samples are then transferred into a plasma enhanced CVD reactor ("Black Magic" AIXTRON) where the nickel particles are used for the catalysis of 3-mm-long vertically aligned carbon nanotubes. These VACNTs are then coated with a layer of 25 nm diamond particles using an electrostatic grafting method described elsewhere [70]. Finally the diamond seeds were grown in a home-made MPCVD reactor until a BDD thin film of 50 nm was obtained on the bundles. Boron concentration in this case was also 2.10 21 at.cm À3 . Surfaces were then oxygenated using an ozone treatment during 2 h in order to obtain an oxygen-terminated BDD and thus a hydrophilic surface.

Spinal cord and hippocampal cell cultures
A digestion solution was prepared containing papain (30 Units/mL, Roche Life Science, code 10108014001) diluted at 0.1 mg/mL (3 Units/mL) and L-cysteine (Sigma Aldrich; code C7477) at 0.5 mg/mL in Neurobasal Medium (1Â, Gibco Life Technologie; code 21103-049). The digestion solution was incubated at 37 C during 10 min, filtered with a 0.22-mm-pore filter (SARSTEDT, Germany; code 83-1826-001) and used within 1 h. The hippocampi of both hemispheres of four E14.5 OF1 mice (Charles River, l'Arbresle, France) embryos were dissected and collected in a dissection medium [F-12 Nut Mix (Ham) medium with 2% of L-glutamine-penicillinstreptomycin (Sigma Aldrich, code G6784) and 1% of Gentamycin (Gibco Life Technologie; code 15710_049)]. Another E14.5 mouse embryo was decapitated, its spinal cord below the medulla was dissected and collected in the same dissection medium. Each tissue type was then treated separately but using the same procedure. The tissue was digested in a falcon tube for 30 min at 37 C, carefully rinsed with 3 Â 3 mL Neurobasal medium. Subsequently, 3 mL of culture medium [Neurobasal medium with 0.5% of L-glutamine-penicillin-streptomycin (Sigma Aldrich, code G6784), 0.25% of Gentamycin (Gibco Life Technologie; code 15710_049), 1% of Glutamax (Gibco Life Technologie; code 35050_038) and 1% of B27-supplement (Invitrogen Life Technologies; code 17504-044)] were added to the tissue, which was then gently triturated mechanically. The suspension was then spun down for 5 min at 1220 rpm at 20 C (Eppendorf AG, Centrifuge 5810R). The supernatant was removed and the cell pellet was re-suspended in 3 mL of culture medium, dissociated mechanically and respun down. The cell pellet was then resuspended again in 800 mL of culture medium and passed through a 40 mm cell strainer (Biologix Research, ref 15e1040). 100 mL were then seeded onto the 3D-nanostructured BDD MEA and onto each substrate type. Before cell seeding, substrates were placed overnight in an oven at 60 C, then incubated in 70% ethanol for 10 min, air-dried overnight, and finally placed separately in one well of a sterile 12-well plate (Nunc, Cat No 150628). No chemical coating was performed to augment cell adhesion so as to evaluate the bare material and the effect of the 3D-nanostructured BDD topography compared to the conventional BDD one. The plate was placed in the incubator during 15 min in order to allow for cell seeding. Further, 1.9 mL of culture medium were added to each well. Cultures were performed on different days and on 9 replicates of each substrate type.

Immunocytochemistry and fluorescence microscopy
Following 8 DIV, the cultured cells were fixed in 4% paraformaldehyde (PFA) in PBS (10 mM), rinsed 3 x 10 min in PBS (10 mM, pH 7.2), and then blocked and permeabilized by pre-incubation for 30 min at room temperature with PBS (10 mM, pH 7.2) containing 0.25% Triton X-100 (PBS-T) and 2% bovine serum albumin (BSA). Cells were subsequently incubated for 60 min at room temperature with a mouse monoclonal antibody against b-tubulin, isotype III (Covance, MRB435P0100; 1:500 in PBS-T 1% BSA) followed by three 10 min washes with PBS-T with 1% BSA. The cells were further incubated with the secondary antibody: Alexa Fluor 488 goat antirabbit (Invitrogen; ref A11008) at 1:1000 for 60 min at room temperature, followed by three 10 min washes with PBS-T with 1% BSA. Cell samples were mounted with Slowfade gold antifade reagent with DAPI (Invitrogen molecular probes, 536938) examined with a wide field fluorescence microscope (Olympus I81) using a plan-neofluar objective (20Â).

Conventional BDD MEA
The fabrication process of diamond MEAs is shown in Fig. 1A. Planar 60-channel MEAs were developed with electrodes arranged in a 4 Â 15 layout without corners covering an area of 900 Â 12600 mm 2 specifically adapted to the geometry of embryonic hindbrain-spinal cord preparations ( Fig. S1A and Fig. 4A), and also with electrodes arranged in a 1 Â 60 layout with a 150-mm spacing (Fig. S1B). To fabricate the arrays, an innovative approach was developed. At first, detonation diamond nanoparticles were spread onto a 4 inch fused silica (this substrate was used to avoid any electrical cross talk between the electrodes) using a process described earlier [71] (Step 1). Next, an aluminum layer, composed of 80-mm disks, was deposited to define the electrode patterns by photolithography (using the AZ4562 photoresist and AZ 351 B developer) (Step 2) and the diamond nanoparticles outside these protected areas were etched away using Reactive Ion Etching (RIE) under a pure oxygen plasma (Step 3). The aluminum hard mask was then chemically removed to reveal the diamond nanoparticles patterns, from which diamond electrodes were grown using Microwave Plasma Enhanced Chemical Vapor Deposition (MPECVD) in a diamond growth reactor (Seki AX6500) housing a gas mixture of methane, hydrogen, and trimethylboron. The fabricated diamond electrodes exhibited a thickness of 500 nm (Step 4). The electrodes were then individually contacted by depositing 10/150-nm Cr/Au metal tracks using lift-off with AZ nLof 2020 (Micro-Chemicals GmbH, Germany) as photoresist material. Contact to the electrodes was achieved by depositing a metal ring across the edges of the electrodes with a 5-mm overlap (Step 5). Finally a 2-mm-thick SU8 layer (MicroChem, USA) was deposited onto the substrate in order to isolate the metal tracks from the electrolyte solution and opened with SU8 developer to define the 20-mm-diameter of the microelectrode (Step 6).

3D-nanostructured BDD MEAs
We recently reported on enhanced electrochemical properties of the BDD electrode using a template of carbon nanotubes scaffolds [49]. This technology was integrated on the MEA fabrication using the process described in Fig. 1B. The first two steps consist in growing a boron doped diamond layer from diamond seeds on the fused silica (this substrate was used to avoid any cross talk between the electrodes). Then an 8-nm-thick nickel layer was selectively deposited at the locations of the future electrodes (step 3). This metallic layer was used as a catalyst for the growth of an inter-layer of 3 mm vertically aligned carbon nanotubes (VACNTs) using PECVD after a growth time of 20 min (step 4). The VACNTs were cleaned in hot aqua regia [HCl(Sigma Aldrich):HNO 3 (Merck), 3:1] to remove the catalyst that eventually remained at the tip of the VACNTs. The whole wafer was then coated with a highly dense nanodiamond layer to fully embed the VACNTs in the subsequent diamond growth step (step 5). A 50 nm boron doped diamond coating was performed with soft growth parameters as described in our previous study [49] (step 6, Fig. 1BeD). The result is the formation of a 3D-nanostructured BDD structure that has the shape of bundles as shown on Fig. 1D. The electrodes were locally masked with an aluminum layer using the AZ4562 photoresist and AZ 351 B developer. The boron doped diamond that grew outside of the electrodes was etched away using RIE under a pure oxygen plasma, and the aluminum mask was chemically removed (step 7). The electrodes were then individually contacted by depositing 10/150-nm Cr/Au metal tracks using lift-off with AZ nLof 2020 (MicroChemicals GmbH, Germany) as photoresist material. Contacts to the electrodes were achieved by depositing a metal ring across the edges of the electrodes with a 5 mm overlap. Finally a 2-mm-thick SU8 layer (MicroChem, USA) was deposited onto the substrate in order to isolate the metal tracks from the electrolyte solution and opened with SU8 developer to define the 20-mm-diameter of the microelectrode (Step 8, and see Fig. S1C).

Cyclic voltammetry
Ultrapure deionised (DI) water (Millipore Direct Q3) was used to prepare all solutions. For electrochemical activation and potential window measurement of the electrodes using cyclic voltammetry (CV), four solution were prepared: an aqueous solution of phosphate buffered Saline (PBS, Sigma Aldrich), a phosphate buffer prepared without NaCl, an artificial cerebrospinal fluid solution (aCSF) and lithium perchlorate (LiClO 4 , Sigma Aldrich) at 0.5 mM. The CV was performed using a Biologic SP200 potentiostat in a three-electrode setup where the studied BDD MEA 3D-nanostructured BDD MEA  microelectrode was the working electrode, a platinum wire the pseudo-reference electrode, and a platinum mesh the counter electrode.

Microelectrode impedance measurements
Electrochemical impedance spectroscopy (EIS) was performed in PBS and LiClO 4 over a frequency range from 0.1 Hz to 1 MHz with logarithmic point spacing and potential amplitude of 0.01 V rms while the diamond electrode was maintained at open circuit potential. The EIS was performed using a Biologic SP200 potentiostat in a three-electrode setup where the studied microelectrode was the working electrode, a platinum wire the pseudo-reference electrode, and a platinum mesh the counter electrode. Impedance measurements were also performed at 1 kHz using either a IMP-I Electrode Impedance Tester from Bak Electronics Inc (Mount Airy, USA) or a NanoZ (Multi Channel Systems GmbH, Reutlingen, Germany).

Microelectrode intrinsic noise measurements
To measure the intrinsic noise level of the electrodes, the electrical potential was recorded for 1 min in artificial cerebrospinal fluid (aCSF) between each microelectrode and an Ag/AgCl ground electrode pellet. The aCSF was composed of NaCl , where s a (¼1.42 mV) was measured for each channel with the amplifier inputs connected to the ground. Noise evaluation was performed for conventional BDD microelectrodes, 3D-nanostructured BDD microelectrodes, as well as platinum and black platinum microelectrodes (commercially available from QWANE). Noise evaluation was also performed after physical cleaning of the MEA to assess its mechanical robustness. For this purpose, the MEA chamber was filled with a standard housekeeping dishwashing detergent (Paic citron) and the array was brushed using a soft painting brush. This procedure was repeated 10 times after which the noise was recorded again.
2.6. Neural recording and stimulation 2.6.1. Acute embryonic spinal cord recording and stimulation To test the performance of BDD microelectrodes for neural recordings, we considered a whole acute embryonic mouse hindbrain-spinal cord preparation. Indeed this preparation displays rhythmic activity occurring every few minutes as episodes consisting of local field potentials (LFPs) and bursts of spikes, and is particularly suitable to assess the performances of the MEA in terms of sensitivity since the signals include spikes of relatively small amplitudes typically around 10e20 mV [28,72]. This is due to the high impedance of the immature cells in this preparation [73] thus requiring small trans-membrane currents to be depolarized. E14.5 embryos were surgically removed from pregnant OF1 mice (Charles River Laboratories, L'Arbresle, France), previously killed by cervical dislocation. Embryos were decapitated and their whole spinal cord and hindbrain were dissected in aCSF gassed with carbogen (95% O 2 , 5% CO 2 ), meninges were removed, and the neural tube was opened along the rostroecaudal axis (open-book preparation). The embryonic hindbrain-spinal cord preparation was then placed over the MEA (Fig. 4A) and superfused with physiological liquid, gassed with carbogen. A plastic net with small holes (70 Â 70 mm 2 ) was laid on the neural tissue, in order to achieve a tight and uniform contact with the microelectrodes. Signals were x1100 amplified and bandpass filtered between 1 Hz and 3 kHz using MCS MEA1060-Up-BC filter amplifiers from Multi Channel Systems (Reutlingen, Germany). Data were acquired at 10 kHz using two synchronized CED Power1401 AD converters and the Spike2 v6 software from Cambridge Electronic Design (Cambridge, England). Rhythmic activity of this immature preparation was recorded for several hours at room temperature. LFP activity was obtained by smoothing the raw signals with a 200-ms time window. Spiking activity was extracted as follows: for each data sample, two moving averages of the signal computed over a 10-ms (DC removing) and a 1-ms (smoothing) time windows centered on this sample were subtracted from the raw data. The MEA was cleaned and used several times without any increase of the noise level, nor degradation of the performances of the BDD microelectrodes to be observed.

Long-term hippocampal cell culture recording and stimulation
Following 14 DIV, the cells cultured on the 3D-nanostructured BDD MEA were superfused with physiological liquid, gassed with carbogen and recorded for hours at room temperature. As for the embryonic spinal cord recording, signals were x1100-amplified, bandpass filtered between 1 Hz and 3 kHz and acquired at 10 kHz. Spiking activity was extracted as described above for acute recordings.

Hippocampal and spinal cord cell cultures on 3Dnanostructured BDD substrates
Cultures of hippocampal and spinal cord cells were used to evaluate the biocompatibility of the new 3D-nanostructured BDD substrate, fabricated with a protocol recently published by H ebert et al. [49], and presenting a novel 3D surface topography different from that of conventional BDD substrates ( Fig. 2A and B). Cultures on 3D-nanostructured BDD substrates were compared to cultures on conventional BDD substrates, already known to be suitable for neural cultures [30e39]. The cell attachment and distribution was assessed using DAPI to stain cell nuclei. The cell ability to express the neuron-specific marker b-tubulin III and their neurite extension were also investigated. After 8 DIV, although cells were passed through a cell strainer just before cell seeding on the studied substrates, clusters and single cells were found to attach and extend neurites on both substrate types in both cell culture types (Fig. 2CeF). Similarly for the hippocampal cell culture and for the spinal cord cell culture, cells were found to express b-tubulin III and no difference in their neurite lengths could be observed on 3Dnanostructured BDD substrates ( Fig. 2D and F) when compared with conventional BDD substrates ( Fig. 2C and E).

Characterization of 3D-nanostructured BDD MEAs
The fabricated MEAs were characterized using cyclic voltammetry and impedance spectroscopy. Fig. 3A and B display three CV plots corresponding to 20 mm diameter conventional BDD microelectrodes measured in LiClO 4 (black dotted), and to 20 mm diameter 3D-nanostructured BDD microelectrodes measured in LiClO 4 (red) and in PBS (green). Within the 3e3.5 V wide potential window, a 3D-nanostructured BDD microelectrode leads to double layer capacitive currents approximately 44 times greater than values probed on conventional BDD microelectrodes. Indeed, the double layer capacitance (C dl ) as well as the charge storage capacity (CSC) of both electrode types can be deduced using the two following formula; C dl ¼ Di 2v and CSC ¼ 1 v R i a dV where Di is the difference between the cathodic and anodic currents at the open circuit potential, n is the scan speed, i a the anodic current density and V is the scanning potential. Hence, in LiClO 4 , a conventional BDD microelectrode exhibits a double layer capacitance of 70 mF cm À2 and a CSC of 220 mC cm À2 whereas a 3D-nanostructured BDD microelectrode exhibits a double layer capacitance of 3 mF cm À2 and a CSC of 10 mC cm À2 (for a potential window of 3.3 V). In PBS, the structured diamond presents a large peak at 1.3 V and a lower anodic current (260 mA, Fig. 3B), reducing the potential window and inducing a double layer capacitance of 2.6 mF cm À2 and a CSC of 6.8 mC cm À2 (for a potential window of 2.6 V). This large peak at 1.3 V can also be found when performing CV measurements in aCSF (Fig. S2A). However, it is not visible when aCSF is replaced by a phosphate buffer, free of NaCl (Fig. S2B). Fig. 3C and D show the Bode representation of the typical impedance spectra for one conventional BDD microelectrode (black dotted) and one 3Dnanostructured BDD microelectrode (red) in PBS. These figures show that the impedance modulus is significantly lower (approximately 40 times) for the 3D-nanostructured BDD microelectrode, which corroborates the larger double layer capacitance observed using cyclic voltammetry. The lower phase at high frequency for the 3D-nanostructured BDD microelectrode indicates the lower cut-off frequency of this interface, a direct result of its higher capacitance when compared with conventional BDD microelectrodes. Fig. 3E presents the impedance measurements at 1 KHz for all microelectrodes of one MEA as recorded in LiClO4, and thus providing a view of the microfabrication process steadiness. There is indeed a small dispersion of the impedance modulus around 50 kU with more than 75% of the microelectrodes displaying an impedance modulus below 100 kU. The intrinsic noise level was recorded for conventional and 3D-nanostructured BDD MEAs. As reported in Fig. 3F, the median noise level of 3D-nanostructured BDD microelectrodes was found to be 3.1 mV. This is lower than the 6.9 mV found for Pt microelectrodes and considerably lower than the 10.7 mV found for conventional BDD microelectrodes. The smallest noise corresponds to black Pt microelectrodes showing a noise of 2.0 mV. As black platinum microelectrodes are highly friable [28], we also assessed the mechanical robustness of the 3D-nanostructured BDD microelectrodes after repetitive standard cleaning (see methods) and found no noticeable change in the noise level of the MEA (Fig. S3).

Neural recording and microstimulation
Both conventional and 3D-3 BDD microelectrodes were used for recording rhythmic activity from the whole embryonic hindbrainspinal cord. Consistently with their high intrinsic noise (Fig. 3F), conventional BDD microelectrodes did not allow reliable recordings of the low-amplitude activity displayed by this preparation. By contrast, 3D-nanostructured BDD allowed the detection of episodes of activity propagating as waves from the hindbrain down the spinal cord (Fig. 4). These episodes could be recorded as LFPs on the 3D-nanostructured BDD microelectrodes covering the hindbrain (Fig. 4C and D). Bursts of spikes could be detected on 3Dnanostructured BDD microelectrodes along all the preparation, as shown in Fig. 4E. Noticeably, even small amplitude spiking signals (within the 10e20 mV range) could be clearly detected. Activity could also be triggered by applying a 1 ms biphasic (cathodic first) current of 10 mA through the 3D-nanostructured BDD microelectrodes located under the hindbrain. Fig. 4F shows the recording of a typical burst of spikes elicited by an electrical stimulation delivered by a BDD microelectrode and lasting about three hundreds milliseconds.
We also recorded spiking activity from a 14 DIV hippocampal cell culture on a few 20-mm-diameter 3D-nanostructured BDD microelectrodes of a 1 Â 60 linear array with a 100-mm interelectrode spacing (Fig. 5). Microelectrodes that recorded signals were generally the ones covered by a dense layer of cells (Fig. 5A). As shown in Fig. 5B, the activity recorded in these cultures mainly consists of bursts of spikes, including some with an amplitude  [74]. Also, the application of a 1-ms biphasic current of 10 mA typically triggered a multiunit response characterized by a burst of spikes (Fig. 5C).

Discussion
To date, although BDD displays excellent biocompatibility [30e32, 35,37], chemical inertness [40], and structural stability [41,42] properties, it has been poorly considered for neural application and there is no report for low-noise neural recording and stimulation with BDD microelectrode arrays [66e68,75]. Indeed, the intrinsic impedance and capacitance of conventional BDD do not match those of its competitor materials. Here however, we fabricated MEAs with 3D-nanostructured BDD microelectrodes (20-mm-diameter) and found that this material provides good performance for neural recording and stimulation.
We found interface impedance of 3D-nanostructured BDD microelectrodes 40 times lower than that of conventional BDD microelectrodes, a result consistent with observations previously reported on larger macroelectrodes [49]. This improvement can be explained by the increased surface area available on the microelectrode due to its nanostructuration [52,76]. Moreover, the microelectrode surfaces consist of a BDD layer deposited on a template of vertically aligned carbon nanotube (VACNT). The vertical CNT inter-layer has the advantage to be highly conductive, thus acting as an efficient current collector and ensuring a high conductivity through the thickness of the electrode [77].
Altogether, the porosity and resulting enhanced conductivity of 3Dnanostructured BDD microelectrodes explain their much lower impedances compared to microelectrodes made of conventional BDD. The median impedance at 1 kHz, the current standard frequency to evaluate the electrode performance corresponding to the main spike event frequency, was found to be of 50 kU for 20-mmdiameter microelectrodes. This value remains comparable to values obtained for current state of the art neural electrode materials such as titanium nitride (TiN) [50 kU for 30-mm-diameter electrodes, [78]], poly(3,4-ethyl enedioxythioph ene) (PEDOT) and PEDOT-CNT [20 kU for 30-mm-diameter electrodes, [79]], iridium oxide (IrOx) [14 kU for 100-mm-diameter electrodes, [80]], and black Pt [80 kU for 12-mm-diameter electrodes, [28]].
As a consequence of the low impedance level of these 3Dnanostructured BDD microelectrodes, we found that their noise level was notably reduced in comparison to conventional BDD MEAs. This low intrinsic noise level allows the detection of small amplitude signals in the range of 10e20 mV in both types of neural preparations tested in the study. In the hippocampal cell culture recording, microelectrodes recording signals were generally the ones covered by a dense layer of cells. This is in line with previous studies characterizing MEA recording of dissociated cell cultures [81]. Our results show that when compared with black platinum electrodes with identical diameter, 3Dnanostructured BDD exhibits higher intrinsic noise. However, black platinum microelectrodes are highly friable [28], while the 3D-nanostructured BDD MEA were very robust and reusable several times. Further electrochemical performances of the electrodes were characterized by cyclic voltammetry both in LiClO 4 and in PBS. The composition of the first electrolyte is typical for characterizing the potential window of an electrode, while the second is a physiological medium. The potential window of the diamond, typically of 3 V, was not altered during the MEA micro-fabrication steps, in the case of neither conventional nor 3D-nanostructured BDD. This large potential window is a strong asset to limit electrochemical reaction of water hydrolysis when the microelectrode is driven to high potential values. This is important to ensure safe charge injection avoiding harmful reactions during electrical neural stimulation, and particularly with microelectrodes of small sizes and thus subject to higher charge density thresholds [29].
The CV measurements allowed to estimate the gain in developed surface via the ratio of the capacitive currents measured on the 3D-nanostructured versus conventional BDD microelectrodes.
Most employed charge-injection electrodes present three main types of charge-injection mechanisms [29]. Among the modelled materials, titanium nitride is a chemically stable metallic conductor that presents a wide potential window À0.9 to 0.9 V with no irreversible reactions and thus uses a capacitive mechanism. Iridium oxide instead goes through a faradaic mechanism exploiting the hydrated oxide film created via the reversible reaction of iridium oxidation and reduction. Platinum displays a pseudo-capacitive mechanism where faradaic surface reactions of oxide formation and hydrogen atom plating as well as double layer charging play a role in most neural stimulation conditions. CV measurement in PBS evidenced a large peak beyond a potential of þ1.3 V suggesting that the 3D-nanostructured BDD microelectrode undergoes an electrochemical reaction at high positive potentials. This phenomenon has been observed previously for conventional BDD in NaClO 4 [41] and for platinum in PBS [84] that was respectively attributed to an oxidation of the BDD and to the presence of the chloride ion Cl À . In our case, this peak can mostly be associated to the chloride ion present in PBS and aCSF since the peak is not visible when CV measurements are performed in a phosphate buffer free of NaCl. Though, the charge injection mechanism for 3D-nanostructured BDD may remain capacitive within a very wide window À1.5 to þ1.1 V where no reactions took place. This is a much larger window and a much greater CSC of 6.8 mC cm À2 than that of TiN.
A critical aspect for an electrode material is to avoid, at the electrode-tissue interface, not only the electrochemical reactions during charge injection [29], but also other chemical reactions occurring during passive use of the electrode and leading to longterm material degradation, a further concern for long-term recordings [6]. In comparison to conducting polymers [51,85] and most metals [86e88], diamond and CNT display far better chemical stability. Diamond erosion only takes place in severe conditions that include very high temperatures, much above 37 C, or when submitted to high current densities [41]. Also, although CNT is a chemically robust nanomaterial, the potential release of single CNTs or CNT bundles from the electrode in the tissue and their toxicity is not well known and can be a concern for clinical applications [89]. In such case, it should be noted that other strategies for BDD nanostructuration could further be envisioned. Yet, in our 3Dnanostructured BDD electrode approach, CNTs were firmly encapsulated between two BDD layers. The embryonic neural cell culture indicated that neurons could attach, survive, and grow neurites on 3D-nanostructured BDD surfaces, with no visible difference in neurite extension. The biocompatibility of BDD surfaces was already assessed with different cell types, among which neural cells [31e36,38]. Our results thus confirm the suitability of BDD for neural cell cultures. It could have been expected however that the 3D-nanostructured BDD surfaces would lead to a greater neurite extension compared to the conventional BDD surface. Indeed, CNT have been shown to promote the neurite extension [57,90,91]. Also, the role of the geometry of nanostructures in neurite outgrowth was evidenced in numerous studies [56,92e94]. Since BDD surfaces are actually not smooth surfaces (see Fig. 2A), it is possible that their morphology already favors the neurite outgrowth [36,38]. 3Dnanostructured BDD thus appears as a reliable and powerful material for neural interfaces, which should be investigated and characterized further with in vivo studies.

Conclusion
In conclusion, our results demonstrate that 3D-nanostructured BDD offers good performances for neural recording and stimulation. The impedance of 20-mm-diameter microelectrodes were found to be 50 kU at 1 KHz, which is comparable to values obtained for current state-of-the-art neural microelectrode materials. These new MEAs also display a high safe charge injection capacity as they combine advantages of the large electrochemical potential window of diamond with the large surface area provided by nanostructuration. The MEA ability in stimulating and recording low amplitude neural signals, together with the established diamond biostability and biocompatibility, should significantly benefit to the future development of new types of MEA-based neural prostheses and implants for the exploration and rehabilitation of the nervous system.