Carbon fiber thread (7 µm diameter, H2550-12K; Hyosung Corporation, Seoul, Korea) is used for individual electrode shanks. Each carbon fibers were coated with a Parylene-C insulating layer (1 µm thickness) using Parylene Deposition System 2010 (SCS Inc., Indianapolis, IN, USA). In order to make electrical connections and fit into the guide structure, the insulated carbon fibers were cut into approximately 1-cm-long pieces.

An acrylic structure was designed using SolidWorks software (Dassault Systems, Wahltham, MA, USA) to guide carbon fibers and the support structure as shown in Fig. 1A and fabricated via mechanical machining (Yunwoo precision manufacture company, Hwaseong, Korea). The acrylic structure has one central groove (width, 200 µm; depth, 300 µm) to locate an insertion support and four grooves (width, 100 µm; depth, 200 µm) to guide carbon fibers. A custom-designed printed circuit board (PCB) was designed to fit with 4 carbon fibers in the grooves and to connect with a header socket for neural recording system (Fig. 1B). Four Parylene-coated carbon fibers were manually placed in each two grooves on either side of the acrylic structure and the four fibers are joined to the lower support guide under a microscope. The individual fiber was electrically connected with pads on the PCB using conductive silver epoxy (H20E; Epoxy Technology, Billerica, MA, USA) as shown in Fig. 1C.

The entire electrode assembly was placed on a hotplate to cure the conductive silver epoxy at 160ºC for 30 minutes. Afterwards, an insulating epoxy (AR-5; Huntsman, The Woodlands, TX, USA) was applied to protect the contact part and was cured at 120ºC for 30 minutes.

In order to sharpen the carbon fiber electrodes and lower the electrochemical impedance, fire-sharpening process was employed as previously reported.18 Briefly, the whole electrode including acrylic structure and the PCB was immersed in the water except a specific length of carbon fibers as shown in Fig. 1D. Then torch fire was applied briefly to ablate the exposed carbon fibers. After fire-sharpening, the morphology of the carbon fiber tip was obtained using scanning electron microscopy (SEM).

Following the fire-sharpening process, an insertion support was positioned in the central groove of the acrylic structure. For the insertion support, two different materials of tungsten and biodegradable silk were used to compare the characteristics after chronic implantation (Fig. 1E). Both tungsten and silk supports have the diameter of 100 µm. The silk support is prepared to be dissolved approximately for 2 days after implantation as described in the following section. Finally, the pin header socket was soldered to the PCB and one extra pin of the socket was soldered with a silver wire (AG-401265; Nilaco corporation, Tokyo, Japan) for connection with a reference electrode (Fig. 1F).

To make a biodegradable silk fibroin support, a polydimethylsiloxane (PDMS) mold was prepared to have a trench structure shaped in a semicircle (200 µm diameter). Silk solution (50 mg/mL; Advanced Biomatrix, San Diego, CA, USA) was applied to the trench of the mold using a syringe and cured on the hotplate at 90ºC for 20 minutes. The layer-by-layer processes of silk coating and drying were repeated until the thickness of silk support reaches about 200 µm. We repeated 7 times of the application and drying process. Then, the silk support had enough stiffness to penetrate into 0.6% (wt/vol) agarose gel mimicking brain tissues. To induce crystallization of silk material for delayed biodegradation, the silk fibroin support was treated with 70% ethanol for 5 seconds. After drying and ethanol treatment, the support shrank into a cylindrical shape with a diameter of 100 µm. This treatment condition resulted in the silk fibroin support which dissolves approximately for 2 days in the 0.5 mg/mL Protease solution (Protease from Streptomyces griseus; Sigma Aldrich, St. Louis, MO, USA). After ethanol treatment, the biodegradable silk fibroin support was located in the acrylic structure and attached with carbon fibers with slight application of additional silk fibroin solution (Fig. 1E). The silk fibroin coating for attachment was expected to dissolve in the brain shortly after electrode implantation.

The electrochemical impedances of carbon fiber electrodes before and after fire-sharpening process was measured at 1 kHz in phosphate-buffered saline (PBS) solution using a 3-electrode based potentiostat system (ZIVE SP2; ZIVE lab., Seoul, Korea) with Ag/AgCl electrode (MF-2052; Bioanalytical Systems, Inc., West Lafayette, IN, USA) as a reference electrode and Pt electrode (CHI115; CH Instruments, Inc., Austin, TX, USA) as a counter electrode. After electrode implantation into the rat brain, the impedances of the carbon fiber electrode embedded with silk and tungsten supports were measured every two days for 1 month. Since there was no counter electrode, the impedance was measured using 2-electrode configuration with the implanted silver wire as a reference electrode and the carbon fiber as a working electrode using RHD2000 interface software (Intan technologies, Los Angeles, CA, USA). The impedance magnitudes measured at 1 kHz were compared between tungsten- and silk-support electrode arrays.

The carbon fiber based microelectrode arrays with silk fibroin or tungsten supports were implanted in hippocampal regions of 10 male Sprague-Dawley rats, weighing 250–300 g (6 rats for silk fibroin support electrodes, 6 rats for tungsten support electrodes). Rats were anesthetized with a mixture of ketamine (75 mg/kg) and xylazine (4 mg/kg). After the incision area was shaved, the rats were placed on the stereotaxic frame and the shaved region was cleaned with 70% ethanol. The eyes were covered with ointment to prevent drying eyes during surgery. After mid-line incision, the skull surface was cleaned and 4 holes were drilled on the skull for four fixation screws. A hole for the microelectrode array was drilled at 2.8 mm posterior and 1.4 mm lateral from bregma. The dura was removed using syringe needle and the carbon fiber based microelectrode array was inserted in the hippocampus region (DV, −3.5 mm). The reference wire from the electrode array was connected to the fixation screws. After implantation, the microelectrode array was fixated on the skull using dental cement (Jet denture repair; Lang dental, Wheeling, IL, USA).

Using the implanted carbon fiber based microelectrode array, electrophysiological recording was performed every two days when the rat was freely behaving in the cage. The neural recording was performed to record spontaneous activity without stimulation.

The electrode was connected with a neural signal acquisition system (RHD2132 amplifier board and an RHD2000 evaluation system, Intan technologies). Acquired signal was sampled at 25 kHz sampling rate and bandpass-filtered from 0.3 to 20 kHz for neural spike recording. Recorded neural signal was analyzed using MATLAB (Mathworks, Natick, MA, USA) program. Each channel was thresholded and waveforms sorted using a spike sorting algorithm “Wave-Clus” (http://www.le.ac.uk/neuroengineering). The spikes were detected using an amplitude threshold (threshold = 5 × median $\left\{\frac{\left|x\right|}{0.6745}\right\}$, x = band-pass filtered signal). And then with a sampling frequency of 25 kHz, this plots 64 datapoints. Spike waveforms were overlapped and the mean values at each time point are extracted as sorted signal.

The peak-to-peak amplitudes of the sorted waveforms were averaged for both groups of electrode arrays with silk- and tungsten-support. The percentage of electrode channels where the sorted spikes were recorded was also calculated for both groups of electrode arrays.

To determine the noise floor of each channel, 1 second of recording data was chosen where no sorted spike and no motion artifact existed. SNR was calculated as a peak-to-peak signal amplitude of the sorted waveforms divided by the six times of root mean square (RMS) value of the noise floor because the peak-to-peak amplitude of the noise floor is calculated as six times the RMS value of the noise floor.24

At 1 week and 4 weeks after implantation, animals were transcardially perfused with normal saline and then with 4% paraformaldehyde. After perfusion, the microelectrodes were explanted and brains were carefully extracted and postfixed in 4% paraformaldehyde for 1 hour. Then, the brains were transferred to 30% (wt/vol) sucrose solution in PBS, and stored 3 days. For immunohistochemical staining of cells around electrode tracks, the brains were horizontally sectioned with the thickness of 20 µm using cryostat-microtome (CM1850; Leica, Wetzlar, Germany). Prior to staining, the brain slices were rinsed in PBS for 10 minutes 3 times and blocked with blocking solution (4% bovine serum albumin [BSA] and 0.2% Triton X-100 in PBS) for 30 minutes. Slices were gently agitated for 2 days in a solution of 0.2% BSA and 0.2% Triton X-100 in PBS with anti-glial fibrillary acidic protein (GFAP) primary antibody (rabbit polyclonal, 1:400; Abcam, Cambridge, UK) for visualization of astrocytes and anti-CD68 primary antibody (mouse monoclonal, 1:200; Abcam) for activated microglia and macrophages. Slices were rinsed 3 times in PBS and incubated in PBS solution containing secondary antibodies (Alexa 488 goat anti rabbit, 1:500; Texas red goat anti mouse, 1:250; ThermoFisher; Waltham, MA, USA) mixed with 0.2% BSA and 0.2% Triton X100 for 4 hours. Finally, the slices were rinsed with PBS for 10 minutes 3 times and mounted on slide glasses and allowed to dry overnight at room temperature. After one drop of mounting solution (Vectashield; Vector laboratories, Inc., Burlingame, CA, USA), a coverglass was affixed on the brain slices for imaging. The fluorescence images were collected under an inverted microscope (IX71; Olympus, Tokyo, Japan) using a CMOS camera (Zyla 5.5 sCMOS; ANDOR, Belfast, UK). The fluorescence images were analyzed in terms of the radial profile of fluorescence intensity around the location of implanted electrodes using a ‘Radial Profile Angle Ext’ macro plug-in for ImageJ (National Institute of Health, https://imagej.nih.gov/ij/). After the averaged intensity profiles were normalized with background fluorescence intensity, the fluorescence profiles for GFAP and CD68 were plotted using GraphPad Prism (GraphPad Software Inc., La Jolla, CA, USA).

The experimental procedures and the care of animals were approved by the Institutional Animal Care and Use Committee (IACUC) of Ewha Womans University (IACUC No. 2018002).

To verify the ability of the carbon fiber for neural recording, SEM images of carbon fiber tips and the electrochemical impedance were obtained. Before fire sharpening process, the carbon fiber had a blunt tip with Parylene C insulation as shown in Fig. 2. The impedances of a carbon fiber electrode were 15.76 ± 3.19 MΩ and 0.74 ± 0.15 MΩ (n = 48) in PBS at 1 kHz before and after fire sharpening, respectively (Fig. 3). This is because it increases the surface area of the exposed carbon fibers as a result of the fire sharpening process. In addition, it might be also possible that the fire sharpening induced the carbon surface modification to facilitate ionic exchange at the interface between the carbon surface and the aqueous solution.25 However, further electrochemical investigations were not performed because the impedance after fire sharpening was low enough to detect neural spikes.

Four-channel, carbon fiber based, depth-type microelectrode arrays embedded with insertion supports were fabricated as depicted in Fig. 4. The electrodes were prepared with two different support materials of a biodegradable silk-fibroin and a non-degradable tungsten. The impedances of the electrodes with silk support and tungsten support were monitored and compared for 1 month after implantation into the hippocampal regions of rat brain (Fig. 5). The first measurement at day 3 did not show statistically different impedances between both groups of electrodes for each group (4.48 ± 0.74 MΩ from silk support electrodes and 3.24 ± 0.69 MΩ from tungsten support electrodes, 24 carbon electrodes [6 implanted arrays]). However, the impedance of the electrodes with silk support gradually increased up to 10.48 ± 0.31 MΩ until day 17 while that of the electrodes with tungsten support decreased to 1.44 ± 1.07 MΩ during the same time period. Afterwards, however, the impedance of the electrodes with silk supports decreased down to 5.01 ± 1.67 MΩ at day 31. The impedance of the electrodes with tungsten supports was 1.43 ± 0.91 MΩ on the day 29.

In order to access the chronic recording capability of the fabricated microelectrode arrays with silk- and tungsten-supports, the neural activity was recorded and compared for 1 month. Spike sorting was performed to analyze the recorded multi-unit spiking signals. The representative sorted waveforms at different time points were presented in Fig. 6A. First, among 24 carbon fiber electrodes (6 electrode arrays) with each support, the percentage of active electrodes was calculated over time and plotted in Fig. 6B. The active electrode is defined as the electrode channel from which the spike signals were detected during recording session. On day 3 after electrode implantation, the percentages of channels with recorded units were 68.2% and 77.5% with silk and tungsten supports, respectively. Until day 15, the percentages of the channels with units were similar between two groups. However, the electrodes with silk supports showed a fluctuation of the percentage of active channels between 0% to 87.5% while the percentage gradually increased up to 100% from the electrodes with tungsten support (Fig. 6B). After day 27, the percentage of silk support electrodes was maintained above 70%, which is still higher than that of conventional microelectrode arrays made of micro-wires or microfabricated silicon as reported in previous literatures.15172026

The amplitude of sorted spike signals was also monitored over time and compared between two groups of electrode arrays with different supports (Fig. 6C). At day 3, the mean signal amplitudes recorded from the electrode with silk support and tungsten support was 76.0 ± 7.8 µV and 76.6 ± 9.2 µV, respectively. Until day 15, the amplitude from tungsten support group decreased to 48.4 ± 2.1 µV while that of silk support was maintained at the similar level 84.5 ± 5.1 µV. After then, even though a sudden amplitude increase was observed from electrodes with tungsten supports at day 17, in general, the mean signal amplitudes from both groups were remained as similar as the initial values. At day 29, the mean signal amplitudes recorded from the electrodes with silk and tungsten supports were 81.3 ± 13.5 µV and 89.4 ± 11.0 µV, respectively.

The amplitude of baseline activity were monitored for noise analysis and SNRs were obtained in order to more precisely evaluate the recording capability of the proposed carbon fiber based electrode arrays. The mean baseline activity was measured in both silk support electrode and tungsten support electrode. The average baseline activity calculated from RMS noise was 9.4 ± 0.7 µV of electrode with silk support and 9.4 ± 1.1 µV of electrode with tungsten support on day 3 (Fig. 6D). It rose at day 5 in silk support (12.3 ± 1.3 µV) and tungsten support (11.86 ± 1.7 µV) then the level of noise floor decreased until about day 15. The RMS value of noise level of electrode with silk support was 9.2 ± 1.2 µV and the noise amplitude of electrode with tungsten support was 8.5 ± 1.3 µV on the last day.

The SNR of the silk support electrodes was 1.3 ± 0.1 and that of tungsten support electrodes was 1.3 ± 0.1 on day 3 (Fig. 6E). The electrode with silk support had a peak SNR level on day 21 (1.9 ± 0.4, n = 5) and the electrode with tungsten support got a peak SNR value on day 17 (2.1 ± 0.7). After 1-month implantation, the SNR levels of both electrode types (silk support, 1.4 ± 0.1; tungsten support, 1.9 ± 0.3) slightly increased compared with the initial SNR levels, which indicated that the quality of the recorded signal was not significantly decreased for 1 month.

To investigate the chronic reactive tissue responses of the implanted microelectrode arrays with two different types of supports, after 1-month implantation, the immunohistological fluorescence images were obtained with GFAP staining to define activated astrocytes and CD68 staining to define activated microglia and macrophages. Overall, the images showed the chronic foreign body reaction characterized by increased reactive astrocyte sheath and activated microglia around the implanted support as reported by previous literatures.92728 Compared between two different materials of support, the images from tungsten support clearly showed the larger void region with the same diameters of the support (Fig. 7). However, the smaller void area was observed in the fluorescence images from silk support electrodes, indicating that the silk support was dissolved before the imaging. The GFAP response from astrocytes in tungsten support electrode showed a higher fluorescence intensity and a wider region around the electrode compared with silk support electrode (Fig. 7A and D). The activated microglia and macrophages were also observed with high intensity around the both probe (Fig. 7B and E). The CD68 staining from tungsten supports showed a diffuse fluorescence intensity around the support which is similar to that of GFAP. However, interestingly, it appears that the CD68 staining from silk supports was observed exclusively inside of GFAP sheath. Accordingly, it can indicate that the activated microglia and macrophages migrated and filled in the spaces where the silk support existed before completely dissolved. It is also important to note that the GFAP showed that the astrocytic sheath formation occurred before the biodegradation of silk support.

The normalized fluorescence profiles in Fig. 8 shows that peak GFAP fluorescence intensity of the tungsten support is approximately 50% higher than that of the silk support (Fig. 8A). Even though the peak CD68 fluorescence intensity was not different between different supports, the intensity profile from the tungsten support was much higher along the distance than that from the silk support (Fig. 8B). In addition, the intensity profile from the tungsten support electrode showed a clear void from 0 to 50 µm due to the radius of the support in both GFAP and CD68 staining.