Adult male Sprague Dawley rats (280–300 g at the beginning of experiments) were provided by the Department of Laboratory Animal Sciences, Peking University Health Science Center (Beijing, China). All animals were housed in standard cages with a 12 h alternating light/dark cycle and food and water available ad libitum. All experimental procedures were approved by the Animal Care and Use Committee of Peking University Health Science Center and were in accordance with the Guidelines of International Association for the Study of Pain.

A desktop FDM 3D printer (Z603S, JG Aurora) was used to perform all 3D printing. Polylactide (PLA) filament was used to build all parts. Electrode rack and drive rack were printed with a 0.05-mm slice thickness, and others used a 0.10-mm slice thickness. Printed parts were manually adjusted afterward with file and drill.

Once the parts were printed, the drive system was manually assembled as followed:

This electrode was initially designed for pain-network study (Moisset and Bouhassira, 2007; Paxinos and Watson, 2009; Ploner et al., 2016). We have selected the eight brain regions that have been reported related to pain (Table 1 and Figure 3A).

During surgery, the rat was anesthetized with 1% sodium pentobarbital (0.5 ml/kg) and supplementary doses of less than 1/3 of the initial dose were provided when necessary to maintain anesthesia. Mannitol (20%, 5 ml/kg) was injected from the tail vein in advance to reduce intracranial pressure. The animal was head-fixed by a Kopf stereotaxic apparatus (David Kopf Instruments, Tujunga, CA, United States) and the skull was exposed. Two stainless screws were tapped into the skull in front of the recording sites with the un-pierced dura as an anchor for headset stabilization. Another two stainless screws were tapped into the skull post lambda for stabilization and served as ground and reference sites, respectively. The exposed skull was strengthened with dental Superbond (Metabond Inc.). After 3D printed hat base was cemented to the skull, craniotomies were made to expose all sites for implantation. Dura was carefully removed in all of the craniotomy sites. 3D printed drive and electrode was fixed to the stereotaxic apparatus and slowly lowered to the brain (Figure 2C). Drive stand was cemented to the skull. Gel-foam was put in all of the craniotomy sites. A 1:1 mixture of paraffin oil and paraffin (58°C melting point) was put above the gel-foam and then melted with a cauter. This will cover the implantation site without limiting the movement of the electrodes. 3D printed hat was fixed to its base through four pairs of 3 mm M1.1 screws and nuts. Ground and reference wires were connected to the amplifier adapter board in the final step (Figure 2D). The animal was allowed to rest for a week in its home cage with free access to food and water (Figure 2E).

A 32-channel electrophysiology system (RHD2000, Intan Technologies) was used to record Ephys data. Signal was sampled at 20 KHz with a 7.5 KHz low-pass filter. The frame-synchronized video was also recorded (acA1920, Basler). Drive turning was carried out by a day-by-day basis based on the numbers of recorded single-unit activities in each brain area. A step of 50 μm was used when no units were recorded, and a step of 12.5 μm was used when low signal-noise ratio spikes shown.

Electrophysiology data was visualized in Neuroscope and Klusters (Hazan et al., 2006). Spike sorting was done by an automatic clustering software (Kadir et al., 2013). Manual curation was made to all automatic clustered units. We have inspected SUAs by autocorrelogram (ACG) and crosscorrelogram (CCG), noise clusters and MUA clusters are removed and clusters belong to the same unit are merged. Spectra of the LFP were computed in MATLAB with the Wavelab toolbox⁴. All the rest computation was done by custom MATLAB scripts.

Statistics of histology locations are computed in MATLAB. Electrode placement error are computed first and plotted with 95% confidential interval in Origin (OriginLab Inc.).

We have verified the placement of the implanted electrodes and biocompatibility of the system. Histology was performed to all of the chronically implanted animals and four extra animals with acute implantation for implantation precision verification (n = 7). 100 μA pulses were delivered for 10 s to each recording electrode to create a marker at the recording site. The animal was then perfused with 0.9% saline, followed by 4% paraformaldehyde in 0.12 M sodium phosphate buffer (pH = 7.4). Fixative solution of 300 ml was used per 100 g of body weight. After perfusion, the animal’s brain was removed from the skull and post-fixed in the same fixative at 4°C for more than 24 h before being moved to 20% sucrose in 1 × PBS solution to dehydrate. The brain samples were then switched to 30% sucrose solution. The fixed brain was frozen by liquid nitrogen for 10 s and then cut into 50 μm thick slices. After the Nissl staining, the slices were inspected by a light field microscope.

A lightweight 3D printed multi-channel electrode with an independent drive system was developed for multi-site micro-electrode implantation. This system consisted of six 3D-printed parts, two printed circuit boards, a flexible printed cable, and a laser-drilled stencil. Complete system weighted around 14 g (varies with drive numbers). Protective shell (Hat body, Hat base, and Hat cap) occupies 53% of the total weight. The weight of each component was as followed: drive rack 1.05 g, drive base 1.02 g, hat base 1.62 g, hat body 4.89 g, hat cap 2.34 g, screws (8 × copper screw) 0.67 g, and PCB parts 2.31 g. The weight could be further reduced by adopting a thinner PCB or replace the brass screw to titanium.

We have tested the microdrive’s movement accuracy. Drive movement hysteresis was 250 ± 45 μm (median value, standard error, n = 24, Figure 2A). Electrode movement were 1,250 ± 52 μm (n = 24) for 5 turns, 2,500 ± 50 μm (n = 24) for 10 turns, 3,750 ± 47 μm (n = 24) for 15 turns (Supplementary Figure S1).

We implanted eight tetrodes in eight different pain-related areas in rats. After the implantation surgery, the electrode was slowly lowered to the target depth by adjusting the microdrive. Histology showed that our electrodes were precisely implanted into the target areas (Figure 3). The overall implantation accuracy of all the recording site was 365.7 ± 28.7 μm (n = 56). In the next step, we checked each direction separately: 10 ± 31.7 μm for AP, 150 ± 24.1 μm for ML, −50 ± 42.7 μm for DV (Figure 3B). To better visualize the accuracy of our system, target locations and electrode locations are put together in a 3D scatter plot (Figures 3C,D).

We have implanted three rats chronically with our system. All animal showed no sign of limited movement ability, and all the electrodes are functioning throughout the recording period. We started the recording once all electrodes had been lowered to the target area with a descend rate lower than 250 μm/day.

We performed multiple long recordings (>8 h) to check the stability of the system. First, we checked the LFP recording quality. Figure 4A shows a recording sample during a sleep recording session. Prominent ripple activities in the CA1 region were observed (Figure 4B), indicating that the electrode was precisely placed in the pyramidal layer of the hippocampal CA1 region (Buzsáki, 2015). Figure 4C shows the spectra of one of the long recording sessions. The recording shows our system has very low movement-related noise and free of 50-Hz power-line noises.

Multiple single-unit activities (SUAs) were successfully recorded in all implanted regions. Low firing rate neurons (<5 spikes per minute) were excluded from the analysis. However, it is challenging to get SUA in all the areas simultaneously due to the turn-over time of recursive microdrive adjusting and verification. However, most of the recordings contained SUA from multiple brain regions. Figure 5 shows one of the best recording session. Among 9-hour recording period, 33 SUAs were recorded. In total, 355 single units were recorded over 7 days in three animals. The unit numbers of each brain region were as followed: AMY: 32, S1: 37, CA1: 218, VPL: 17, ACC: 14, PL: 13, NAc: 23, and AIV: 1. Multiple units with low firing rate were found in AIV but were not included in the unit count. Detailed spike amplitude characters are shown in Supplementary Figure S2.

Single-unit activities of noxious laser-evoked responses were recorded and analyzed (Figure 6A). The trial-averaged raw-waveform was not suitable for the analysis of the evoked oscillatory LFPs, so we analyzed the trial-averaged wavelet instead (Figure 6B). An increase of gamma-band oscillation was observed in CA1, ACC, NAc, and AIV. A gamma epoch in NAc was also found as shown in Figure 6A. Post-stimulation time histogram of SUAs was analyzed, and an example of noxious-stimulation responsive units in each recording site was shown in Figure 6C.

Brain’s building block is functionally connected as neural assemblies. Most of the brain functions involve the interplay between several brain areas. For example, memory consolidation involves the collaboration between the hippocampus and the medial prefrontal cortex. Some of the brain function involves more brain areas. For example, more than eight different brain regions are thought to take part in the neural process of pain (Melzack and Casey, 1968; Ploner et al., 2016). To answer the question about the neural activity on a brain-wide scale, imaging-based approaches like functional magnetic resonance imaging (fMRI) and immunofluorescence could take the snapshot of the neural activity over a large spatial extent. However, those methods do not have an action potential level temporal resolution. Therefore, it is hard for those techniques to capture the fast interplays between different brain regions. Multi-site electrophysiology is especially useful for research on neural network dynamics. Till now, many high-density micro-fabricated electrodes have been developed for this purpose (Khodagholy et al., 2015; Jun et al., 2017). However, although those probes have a state of art precision and electrode density, they have certain limitations. Firstly, due to the need for a clean-room facility, most micro-fabricated electrodes are expensive. Secondly, they have certain geometry limitations. Michigan probes are etched from the silicon wafer. Therefore, all electrodes have to be limited in the same plane. High-density surface electrocorticography (ECoG) could cover a large area of the cortex but is limited to the superficial layers of cortex (Khodagholy et al., 2015). The soft material based probe could be placed in several regions at the same time, but it needs a unique implantation tool and also suffers from placement error (Zhao et al., 2019). The micro-wire based electrode has been used for neuroscience research for over 40 years (Strumwasser, 1958; O’Keefe and Bouma, 1969), and its life cycle has not come to an end yet. Many attempts have been made to use microwires to record from multiple brain areas. However, due to the limited spatial resolution, it is crucial to be able to adjust the depth of the electrode continuously after the implantation. Many investigators implanted multiple individual microdrives (Connell et al., 2013; Patel et al., 2013). A continuous effort has been put to the improved design of the microdrive system to improve the microwire implantation. Battaglia et al. (2009) introduced a less-than-2 g microdrive with six drives. There are several works adopted similar cylindrical design to hold maximal number of microdrives (Siegle et al., 2011; Freedman et al., 2016; Liang et al., 2017) and also capable of recording from a few brain regions (Lansink et al., 2007).

Moreover, a few novel designs were introduced dedicated to recording from multiple brain regions. Headley et al. (2015) come up with a modular solution to move individual electrode groups independently in different locations. Billard et al. (2018) developed a multisite microdrive dedicated to recording from multiple brain regions.

Compared to the previous works, our drive’s major difference here is the adoption of a position board to keep the implantation process easy while maintaining a high implantation accuracy. Other modifications like the mechanism to separate the pre-amplification from the microdrive were also introduced. We have shown the quality of the signal unit recording in all recorded brain areas. A simple laser-evoked pain diagram was used to show the laser-evoked responses in all recorded regions. The results showed that our design is a powerful tool to study the complex neural dynamics involving multiple distant brain regions like pain. With this tool, we can analyze the correlation between SUA/between SUA and LFP/between LFP in multiple brain regions. This would significantly increase the feature space of the analysis and leading to more findings of how neural assembly interplays with each other.

There are several successful multi-drive systems developed. HyperDrive (Brunetti et al., 2014), Omni-drive (Neuralynx⁵) are both designed with a cylindrical shape. We adopted similar microdrive holding systems to ensure a large number of microdrives with a relatively small form factor. However, the main difference in our system is the electrode targeting mechanism. Instead of aiming at only one target, our system could have a set of custom locations. Also, compare to the systemDrive (Billard et al., 2018), no individual cannula needs to be pre-implanted for our system, and all electrodes could be implanted at once. This would require much less time for implantation surgery. Another benefit is the chance that the electrode gets kinked is almost zero.

Our design is also the first one to introduce the flexible interconnect between the pre-amplifier connector and the microdrive in the tetrode system. This design has removed one of the strongest external forces applied to the drive. During the recording, no any signal changes due to the daily plug/unplug of pre-amplification were observed.

The only chance that a large external force will apply to the microdrive itself left is the microdrive adjustment. We have trained the rat to eat fruit-loops while we were adjusting the screw on top of the head with a screwdriver. Sometimes the animal’s movement increased the force we applied to the one microdrive, and we observed signal change in other microdrives. To solve this problem, we should try to avoid excess force applying to the microdrive and, if necessary, anesthetize the animal. We also come up with an add-on device that could turn the microdrive by micro-steppers (Supplementary Figure S3). Eight micro-steppers were mounted to the same rack and installed on the top of the microdrive. We also designed a multidrive controller board that received the command from a micro-USB and turns all electrodes. However, this design would add an extra ∼10 g weight to the system. The performance of this electronic microdrive is still under validation and not included in the present study.

This probe can be further improved or modified. One direction is the integration of optogenetics parts. Optical fiber could be attached to the microdrive instead of the tetrodes. This would enable a flexible configuration of the recording/optogenetic stimulation in multiple target areas.

We designed the EIB board and microdrive for support up to record from 16 tetrodes, but in this study, we only showed data collected with eight tetrodes. Sixteen tetrodes will add more weight (<2 g) to the previously reported weight by eight brass screws and one other adapter board. More assembly time and micro-drive adjusting time would be expected for 16 tetrodes setup.