3D printed guide tube system for acute Neuropixels probe recordings in non-human primates

All experimental procedures were approved by the Institutional Animal Care and Use Committee and complied with United States Public Health Service policy on the humane care and use of laboratory animals. A single adult female rhesus macaque (Macaca mulatta, animal J, 21 years) was used for this study. Surgical placement of recording chambers, screws, and a head stabilization device occurred under aseptic conditions using isoflurane anesthesia. Cephalic recording chamber implant locations were determined using the surgical planning software Cicerone (Miocinovic et al 2007) and implanted bilaterally over the dorsal premotor cortices. In-vivo testing using the NP probe was performed after this animal completed a separate study. The dura mater at the time of testing, approximately two years and three months after the craniotomy surgery, was thick and no underlying cortical sulci were visible. The primate was trained to perform a food reward reaching task where a successful trial consisted of the NHP reaching from the starting position (a hand resting location in front of the primate) to the food reward, bringing the food to its mouth, and returning its hand to starting position.

Our general approach was to design a 'Neuropixels guide tube system' that could interface with an existing microdrive and enable alignment and insertion of a NP probe through a small cannula positioned within the cranial chamber. The Alpha Omega Flex MT microdrive (figure 2(a)) was identified to be well suited for implanting a NP because the Flex MT microdrive has an open chamber adapter, a positioning ring that provides flexibility for positioning the NP probe within the chamber, and it has a lowering stage for the NP probe and one for the guide tube. With respect to moving components, the gross adjustment stage translates the 'positioning' ring and attached fine adjustment stage by hand turning a ball screw. Given the measurement scale on the drive tower, the depth of the gross adjustment stage can be measured with 0.5 mm resolution. The fine adjustment stage, also known as the electrode drive tower, translates the NP probe along a guide channel with micron resolution via the attached motorized microdrive, also called the electrode positioning system (EPS). Used in conjunction, both stages provide independent control of the guide tube and NP probe (figure 2(b)). Finally, the electrode drive tower can be placed at multiple positions around the positioning ring to increase the targetable area on the cortical surface. A puncture assist guide from Narishige microdrive was used to assist dura puncturing while maintaining stereotaxic and chamber alignment (further details described below in section: 'Neuropixels Implantation Sequence—Aligning the Narishige and the Flex MT system').

Zoom In Zoom Out Reset image size

The NP guide tube system was designed in two parts, a probe holder and guide channel. The probe holder and guide channel were designed in SolidWorks and printed using FormLabs 3B stereolithography (SLA) material with a printing tolerance of 50 µm. Designs are provided in supplementary material. The probe holder was designed to fit around the probe base. The probe was manually slid into the probe holder until the probe base was at the distal end of the probe holder. Hot melt adhesive (Surebonder, 725R10-1P) was applied to the probe and probe holder to fix the probe in position. The probe holder attaches to the probe (figure 2(c), front and side view) and acts as an alignment carriage to keep the probe shank centered within the guide channel (figure 2(c), bottom view). The guide channel functions as a path for the probe holder and holds a 4.5 mm long, 18-gauge guide tube. The top 1.5 mm of the guide tube is seated in the guide channel platform and the remaining 3 mm of guide tube was exposed for bridging tissue (figure 2(c), isometric view). The guide tube was fabricated by removing the tip from an 18-gauge stainless steel needle (Monoject 18 × 1A) via a grinding stone on a Dremel rotary tool. Another stone was used to create a bevel on the edge of the remaining tube and smoothed with high grit (fine) sandpaper to remove any residual roughness left over from the beveling process. The guide tube gauge provided probe clearance and matched the puncture tube gauge. The guide tube was glued into a pre-aligned hole in the platform at the bottom of the guide channel. Additionally, the small square platform surrounding the guide tube (figure 2(c), isometric view bottom left) was designed to stabilize tissue surrounding the probe shank. The NP probe, probe holder, and guide channel have a total cross-sectional area of 67 mm².

To integrate the NP guide tube system into the Flex MT microdrive, the guide channel was bolted into the base of the electrode drive tower using an existing hole and bolt (figure 2(b), attachment bolt). The probe holder with the attached probe was then inserted into the guide channel and clamped into the electrode drive tower. Prior to experiments, the probe shank's alignment to the guide tube was confirmed by visually inspecting the probe shank as it slowly passed through the guide tube, ensuring that the NP shank was centered as it entered and exited the guide tube. Theoretically, given a guide tube inner diameter of 0.838 mm and NP shank length of 10 mm, the maximum allowable angular deflection is 2.1°. If adjustments were required, the hot melt adhesive was rewarmed to reposition the probe shank. After confirming proper alignment, visualization of the probe shank passing through the guide tube during in-vivo implantation was not required.

To target cortical areas and perform in-silico testing, 3D Slicer (Fedorov et al 2012) and 3D CAD software SolidWorks were used due to their complementary functionality. 3D Slicer provided excellent handling of the medical images and SolidWorks provided complete control over the microdrive model. Additionally, both software can import and export stereolithography files allowing for efficient transfer of models between software.

3D Slicer was used to align the primates 7 T magnetic resonance imaging (MRI) and computed tomography (CT) images and segment bone (tan), brain (pink), and instrumentation (grey) regions (figure 2(d), i-iii). Segmentations were spatially fixed together in 3D Slicer and exported into SolidWorks. The 3D model of the Flex MT microdrive was attached to the cephalic chamber segmentation for targeting with NP guide tube system. This approach provided the ability to manipulate all microdrive components (figure 2(e), i) to target various cortical locations (figure 2(e), ii-iii). Utilizing the radial targeting of the ring, the targetable area was 76% of the chamber (19 mm ID) or a 8.79 mm targetable radius from the chamber center (figure 2(e), ii). Since the Flex MT model has a 1:1 (real life) scale, x–y coordinates for the guide tube and gross stage translation measurements were collected to inform the puncture x–y coordinates and physical gross lowering stage movements during in-vivo implantation (figure 2(e), iv). After planning and targeting within SolidWorks, the microdrive model was exported in the fully implanted position into 3DSlicer and attached to the cephalic chamber to visualize the probe depth across MRI scans (figure 2(f)).

Radial targeting with the NP guide tube system increased the accessible area within the chamber but required the alignment of the NP guide tube system to the puncture site in the dura. A benchtop testing apparatus consisting of a cephalic chamber, paper to mimic the dura mater, and support structure was used to perform alignment of the NP guide tube system to the puncture hole. The paper within the testing apparatus was punctured using the Narishige guide cannula puncture apparatus at the x–y coordinates determined from the 3D model. The Alpha Omega Flex MT system was then placed on the testing apparatus and the guide tube on the NP guide system was aligned to the punctured hole in the paper. In-vivo implantation occurred after the two systems were physically aligned and tested in the benchtop apparatus. Designs for benchtop testing apparatus are included in supplementary material.

Implantation occurs in three general phases: puncturing the dura mater at the desired location using the Narishige microdrive, attaching the Flex MT microdrive and gross lowering of the guide tube system, and lowering the NP probe. Initially, the dura mater was manually punctured using a long, 18-gauge puncture cannula via a puncture assist device in the Narishige (figures 3(a) and (b), Step 1). The Narishige microdrive was then removed to place the Flex MT microdrive on the cephalic chamber.

Zoom In Zoom Out Reset image size

The Flex MT microdrive positioning ring was kept in a raised position to protect the NP guide tube system while fixing the microdrive to the cephalic chamber. Next, the ball screw was turned to descend the positioning ring and electrode drive tower with attached NP guide tube system (figures 3(a) and (b), Step 2). As the 3 mm guide tube approached the dura mater surface (estimated with measurements from the testing apparatus and under visualization with a surgical microscope), the ball screw was slowly turned to press the guide tube into the punctured dura mater hole until the bottom platform of the guide channel slightly depressed the dura surface (figures 3 (a) and (b), Step 3). We found that appropriate placement of the guide tube within the punctured dura was associated with a filling of the guide tube with cerebrospinal fluid.

After seating the NP guide tube system with the ball screw, the EPS drive wire was placed in the fine lowering stage clamp to drive the NP along the channel in micrometer increments (figures 3 (a) and (b), Step 4). EPS depth measurements and movement speeds were controlled via the Alpha Omega software. NP probe movement speed was approximately 50 µ s⁻¹ during implantation. The distance traveled in one movement was 0.5 mm while the probe was above the guide tube, 0.25 mm per movement within the guide tube, and 0.1 mm per movement in cortical tissue. The 0.1 mm displacement per movement was to help ensure the probe shank enter the cortical tissue safely. After each movement, a few seconds were taken to assess signal changes across channels and then another movement was initiated. Probe insertion time ranged from 20–30 min in total, including 10 min to lower the probe tip to the opening in the guide tube or cerebral spinal fluid (CSF) and approximately 10–20 min to lower the probe through the guide tube to the final depth, until there was approximately 1 mm distance between the top of the guide tube and probe base. This final depth was chosen because there is a widening in probe width at the point where the probe shank affixes to the probe base and to reduce risk of probe shank breakage. With a guide tube length of 4.5 mm, we estimate an insertion depth of approximately 5 mm past the bottom of the guide tube. A grounding wire was attached to the head post and an interface cable was connected from the recording system prior to probe lowering (figures 3(a) and (b), Step 4) to monitor signals in real time using Open Ephys (Siegle et al 2017). The signal noise floor decreased once probe contacts entered the cerebral spinal fluid that filled the guide tube. Neuronal firing was observed once the probe entered brain tissue, and the shift in spiking activity along the recording contacts was monitored as the probe advanced (figure 3(c)), further confirming implantation depth change and probe integrity. The NP probe was advanced to the final depth (figures 3(a) and (b), Step 5); in our recordings, we inserted until ∼1 mm distance remained between the top of the guide tube and base of the probe. Finally, the NP probe holder clamp was fixed in place with a set screw to immobilize the probe (figure 2(a)) for the remainder of the recording.

Neural data were collected using NP 1.0 probes or NHP NP 1.0 probes. NHP NP 1.0 probes are identical to the NP 1.0 probes except that NHP 1.0 probes have a shank cross-sectional area of 90 × 100 µm as compared to the 70 × 20 µm shank for the NP 1.0 probe. NP recording technology allows users to select up to 384 of the 960 recording channels to record from at any given time. The 384 channels closest to the tip, which span 3.84 mm, were selected for testing in this study.

Data sampled at 30 kHz and filtered with a 500 Hz high pass filter was spike sorted using Kilosort 2.5 (Steinmetz et al 2021) and manually curated using Phy following the guidelines written by S. Lenzi and N. Steinmetz (https://phy.readthedocs.io/en/latest/sorting_user_guide). Briefly, putative single unit (SU) were identified as clusters with waveforms distinct from other waveforms or noise on the same channels and had clear refractory periods. Clusters containing two or more groups of waveforms that were not distinct enough to be reliably separated into individual SU were grouped into a multi-unit (MU) cluster. Clusters with a firing rate lower than 0.1 Hz were excluded from further analysis. Neural data analysis for spiking data and plot generation was performed using MATLAB code modified from the spikes repository made by N. Steinmetz (https://github.com/cortex-lab/spikes).

Behavioral data were collected with two Logitech USB webcams recording at 20 frames per second. One camera recorded eye behavior and the other camera recorded reaching behavior. Cameras were synchronized and video data was saved using a Tucker Davis Technologies (TDT) RZ2 and Synapse software. The NI DAQ chassis (NI PXIe-1071) containing a NP PXIe acquisition card and a DAQ (PXIe-6341, X Series DAQ) were synchronized together with Open Ephys online synchronization and then synchronized with the TDT recording system using a 1 V square wave pulse train at the beginning and end of each recording. Behavioral timepoints during the reaching task were manually extracted in MATLAB (MathWorks, 2021) for use in analysis.

The NHP performed a reaching task where the NHP was trained to hold their hand on a start pad location, reach to one of three positions where food would be presented, bring the food to their mouth, and then return to the starting location. Successful trials are defined as a completion of all steps. Multiple behavioral time points can be extracted from this reaching task, however, for most of the analysis we have focused on examining data with respect to reach initiation. Additionally, the reaching task provided a good test of implant and recording stability due to forces imposed on the system by the primate during the reaching task.

Following the described implant methodology, we performed three implantations using commercially available NP 1.0 probes, two with NHP NP 1.0 probes, and successfully reimplanted both versions of the probe one time (table 1, figure 4). Two NP 1.0 probes and one NP NHP 1.0 probe were used. Experimental sessions were typically 3–4 h in duration, with multiple recording blocks obtained throughout the session. No NP probe broke after insertion into the brain; however, during a second implantation attempt, one probe broke on the dura mater due to a rotation in the X stage while fixing the Flex MT microdrive on the cephalic chamber (i.e. figure 3(a), Step 2). At the base of the probe where the probe shank integrates, there is a widening to ∼0.14 mm. Therefore, we drove the probes to a position where the base of the probe was 1 mm above the top of the guide tube due to concerns of breaking the probe shank on the wider portion. Utilizing Kilosort 2.5 and following guidelines noted in section 2 we identified 1198 (627 SU/571 MU) units across the five recording sessions (table 1).

Zoom In Zoom Out Reset image size

To test the ability of our setup to enable stable neuronal recordings, we analyzed data collected while the monkey was actively engaged in a food reach and retrieval task. Neural data were collected from all probes during reaching (table 1). Recording depths for each probe implantation were approximated using the model measurements and physical recording measurements (figure 2). Data were recorded on the bottom (most ventral) 384 channels, displayed as the yellow line in the estimated recording trajectory pane. A raster plot of SU and MU spike times across the length of the probe during a reaching trial are presented in figure 5(a). Behavioral epochs and time points are demarcated by vertical lines and labeled in the raster plot.

Zoom In Zoom Out Reset image size

Black vertical lines indicate trial start, food presentation, reach start, and the end of the trial (figure 5(a)). Of the 384 channels recorded, a subset of channels was selected to illustrate the types of raw neural traces collected by the probe and the heterogeneous responses captured across the span of the recording channels during the reaching task (figure 5(b)). The same units are held on a group of channels (channels 159–163) throughout the reaching task which enabled robust spike sorting. Small chewing artifacts were present in the signal (at time ∼56.5 s). Bursting activity (channel 163 and 164), tonic firing (channel 311), regions of low or high modulation (channel 82, 87, 98) were present during the reaching task. This showed that the NP probe shank can detect a variety of responses across the column of sampled neural tissue.

Waveforms from an isolated SU were displayed from recordings obtained 1 h apart (figure 5(c)) and show consistency in waveform amplitude and shape throughout the reaching task. Peristimulus time histograms were generated for three SU/MU clusters selected at different depths across the probe and demonstrate three different responses that were observed relative to reach onset: inhibited (SU), excited (MU), and mixed (MU) (figure 5(d)). The sample waveforms (figure 5(b)) and consistent response of cells during each reaching task trial (figure 5(d)) show that, with proper fixation of the recording equipment, this implantation methodology can accomplish stable recordings (e.g. holding cells) throughout NHP reaching behavior.

Supplementary figure 1 shows additional raster plots across multiple behavioral trials. Data from multiple trials further illustrate stability in the method to record using a NP probe. Additionally array raster plots show the variety of the spiking activity at different depths collected using a NP probe in PMd cortex.

Two useful outputs of Kilosort 2.5 based on sorted spiking data information include drift maps and drift traces. Drift maps provide a visual assessment of the spike location (figures 6(a)–(c)) and amplitude. Drift traces (figures 6(d) and (e)) are a batch by batch (2.2 ms per batch) calculation of the relative position of the neurons detected on the array. Recording 1 drift map (figure 6(a)) displays the spikes across the 384 channels closest to the tip of the probe as a function of time. Each consecutive drift map (moving left to right) follows a separate step in the spike sorting process and displays the overall drift correction and curation of the spiking data. Figure 6(a), recording 1 drift map, displays data from 0 to 100 s and recording 1 drift trace (figure 6(d)) displays the drift for the entire recording. This data was drift corrected and initially sorted (figure 6(b)) via Kilosort 2.5 and then manually curated following the methods described previously. The total spike count remains similar which is why figures 6(b) and (c) look similar.

Zoom In Zoom Out Reset image size

Drift data provide the amount of displacement up or down that occurs during a recording session. Kilosort 2.5 (Steinmetz et al 2021) best handles slow drift which can be seen as a gradual trend in the drift trace. Slow drift physically occurs when the tissue slowly passes against the NP shank. We allowed the probe to settle for at least 30 min to mitigate aspects of slow drift from the initial probe insertion. Fast drift is a second form of drift which occurs when the probe moves up and down with respect to the tissue. These events are more transient in nature and harder to account for with sorting algorithms (Rey et al 2015). Fast drift events, sharp transient peaks in a drift trace, indicate the probe moved with respect to the tissue seen in figure 6(d). These events were better addressed and mitigated after the initial recording session by using a set screw to stabilize the final recording position of the probe. Drift decreased in sessions after the initial recording except for recording 4 (figure 6(f)). Low drift recordings sessions are a marker of stability and expedite the spike sorting process. Drift data from this implantation approach show that this implantation system can accomplish low drift (i.e. stable) recording sessions.