Experience and safety of intraoperative Neuropixels: a case series of 56 patients

In Brief

In a case series of 56 patients undergoing craniotomy, researchers utilized the Neuropixels probe, a high-density, penetrating silicon microelectrode array, for intraoperative single-neuron recording. Recordings were conducted in tissue designated for resection. As operator experience increased, the incidence of probe fractures decreased and the success rate of recordings increased. These findings demonstrate that intraoperative use of the Neuropixels probe can be performed safely.

Abstract

OBJECTIVE

The Neuropixels probe, a high-density silicon microelectrode array, has been a transformative tool for extracellular recording of large numbers of single neurons across animal models. Traditional tungsten microelectrodes available for clinical neurophysiology typically only record 1–2 neurons at a given time. Human intraoperative Neuropixels recording increases access to single-neuron spiking by an order of magnitude and is poised for rapid adoption. The authors’ objective was to determine the safety and yield of intraoperative Neuropixels for large-scale simultaneous neuronal recordings.

METHODS

This study examined safety and the success rate from the authors’ case series of 56 consecutive patients who participated in Neuropixels intraoperative recording at the authors’ institution. All participants were undergoing craniotomy for resection for various pathologies (intractable epilepsy [n = 33], tumor [n = 19], and vascular lesion [n = 4]). Among these craniotomies for resection, the vast majority were done under monitored anesthesia care with awake mapping (50/56). In all cases, a Neuropixels probe was placed into tissue that was resected, allowing histology to be recovered from a subset of cases. All intraoperative probe fractures, cases that did not yield any putative single-neuron recordings, and postoperative complications are reported here.

RESULTS

No medical or neurological complications, surgical complications, 30-day readmissions, or deaths were observed. Overall, probe fractures were rare (n = 3) and all occurred within the first 6 cases. All fractures occurred at the base of the probe shank and were easily recovered. In recovered acute histology from 1 case, the histological evidence of insertion was minimal and proportional to the 70 mm × 132 mm cross-sectional area. In terms of yield, most cases had good neuronal yield, with only 9 (16.1%) cases that did not result in at least 1 putative single-neuron recording, with the etiology of failure being attributable to either electrical noise (6/9) or there being no detectable spikes (3/9). All cases that did not result in at least 1 putative single-neuron recording occurred within the first 20 cases, suggesting significant improvements in the learning curve and techniques for high-yield and safe intraoperative recordings.

CONCLUSIONS

Intraoperative Neuropixels recordings can be done safely for large-scale neuronal recordings. Over time, the authors experienced no probe fractures and improved success rate of microelectrode recording. Further measures to float the electrodes, limit forces on the probe, and prevent probe fracture during insertion may improve the yield. Neuropixels has tremendous potential for future applications in clinical functional mapping over traditional microelectrodes.

The Neuropixels high-density silicon electrode array¹ has seen widespread adoption across the field of neuroscience. The Neuropixels 1.0 probe is a single-shank device, 1-cm long × 70-mm wide × 20-mm thick (97- or 132-mm thick in the version used in humans), with 960 electrodes, each electrode measuring 12 × 12 mm. The probe is directly inserted into brain parenchyma and able to simultaneously record from 384 of these 960 sites. The 2.0 version² lets users select 384 sites from 1280, thereby providing the ability to simultaneously record from hundreds to thousands of single neurons in rodent models. The power of this approach to provide a mechanistic understanding of neural computation is evident in its ability to drive insights from the level of single-neuron coding to brain-wide functional connectivity.^3,4

The Neuropixels probe has recently been translated to human neurophysiology, through intraoperative recordings.^5–7 Its first application was applied to single-neuron coding of human speech in the temporal lobe, and then in the frontal lobe.^8,9 As this technology gains traction, it is critical to understand the safety of intraoperative Neuropixels recording and to share the experience and strategy in order to make safety and success reproducible. There are two key adverse outcomes to avoid for Neuropixels or any other intraoperative penetrating devices: fracture risk resulting in retained foreign bodies, and damage to brain parenchyma that would not otherwise occur. The Neuropixels probe has potential as a tool in neurosurgery for functional brain mapping, but at the cellular level with large-scale simultaneous recordings. This may someday have clinical applications to intraoperative localization of function, localization of interictal epileptiform activity, and identification of neoplastic or lesional tissue.

Here, we report safety, histology, and recording success data from our case series of all 56 patients who have participated in intraoperative recording with single-shank Neuropixels probes at our institution—the largest series at this time.

Methods

The experimental protocol was approved by the UCSF institutional review board. Fifty-six participants were included in the study, and all provided written informed consent. Patients provided consent to the research coordinator or members of the presurgical team. Participants were selected from patients undergoing craniotomy for a resection procedure at the University of California, San Francisco, with the senior author of this paper. All participants underwent surgery between October 2020 and June 2023.

Preoperative Preparation and Sterilization of Neuropixels

See Chung et al.⁵ for more details. Briefly, Neuropixels 1.0 (nonhuman primate short probe with metal dovetail caps) (IMEC) was used for all recordings. In earlier versions of the probe, two 27-gauge subdermal needle electrodes (Ambu) were soldered separately to the probe flex-interconnect to serve as ground and reference using lead-free solder and two strands of twisted 36 AWG (American wire gauge) copper wire. Later versions of the probe were delivered with stainless steel wire attached to ground and reference, terminating on Touchproof connectors, which in turn were attached to 27-gauge subdermal needle electrodes (Ambu). Probes were then installed into custom CleanCut mounting cards (Oliver Healthcare Packaging) before sterilization according to standard protocols of ethylene oxide sterilization. The sterile field was extended part way along the length of the cable connecting the headstage to the Neuropixels PXIe acquisition module. No probes were reused across participants.

Selection of Insertion Site

Clinical mapping, using direct cortical stimulation as well as depth and electrocorticography (ECoG) recordings for seizure monitoring, was done prior to insertion site selection to establish the cortical resection boundaries. Depending on the location of the pathology, either a single mapping modality or combination of modalities was used. Once the cortical resection boundaries were established, a site within these boundaries was selected. Sites were selected to be at the middle of a gyrus and carefully avoided any surface vasculature (Fig. 1B).

FIG. 1.

Intraoperative Neuropixels recording workflow. A: Configuration of Neuropixels intraoperative recording. The probe is attached to the patient via a Mayfield head holder, Greenberg C-clamp, articulating bar, 3-ball joint arm, micromanipulator, and Neuropixels mounting rod with dovetail adapter. B: Neuropixels array in the cortex, which was previously determined to be resected. C: Neuropixels array being inspected after recording with intact recording electrode. D: Fractured Neuropixels array. All fractures occurred at the base of the shank as shown.

Probe Insertion and Recording

The probe is mounted via a Mayfield head clamp, Greenberg C-clamp, an articulating bar, 3-ball joint arm, a micromanipulator, and Neuropixels rod with dovetail adapter (Fig. 1A). The Neuropixels probe was then removed from custom-designed holders for sterilization and mounted to the micromanipulator. The probe was then advanced against the pia arachnoid. Several millimeters of dimpling of the pial surface were common before the probe pierced the pia. In cases where the probe was unable to pierce the pia, the silicon shank would visibly bend. In these cases, the shank was withdrawn, the cortical surface was irrigated, and reinsertion was attempted. If the shank did not penetrate the pia before bending of the shank, a new insertion site was selected. Once piercing pia, the micromanipulator was then advanced to a final targeted depth of 7 mm. In cases where there was anticipated large amplitude motion, such as a patient under monitored anesthesia care (MAC) needing to cough or reposition, the probe was withdrawn at least a centimeter above the surface of the cortex. If there were no such events, recording was completed and the probe was retracted using the micromanipulator. The micromanipulator was then removed from the field where the probe was inspected under ×3.5 magnification for any fractures (Fig. 1C and D).

Electrical Noise Isolation

Once data acquisition began, all nonessential equipment was unplugged from wall power. Electrocautery and saline warmers were important in most of our operating room configurations. Battery power was also used for any equipment capable of being run on battery. It was important to run syringe pumps on battery power in all our recording configurations.

Resection of Recording Site

In cases where the recording site was at least 1 cm from the planned resection boundary and at least 1 cm away from any large surface vasculature, a 1–2 cm en bloc piece containing the recording site was resected. The en bloc tissue was then handed off the field, submerged into 10% buffered formalin, and sent to the department of pathology for sectioning and immunohistochemistry. The histology shown here was reviewed by a neuropathologist (C.R.C.). In most cases, the recording site was unable to be resected in this en bloc fashion due to proximity to vasculature and/or resection boundary.

Electrophysiological Data Collection and Processing

Briefly, data were collected using a custom-constructed rig, including a Windows machine, PXI chassis (NI PXIe-1071), PXI Multifunction I/O Module (NI PXIe-6341), NI SHC68-68-EPM shielded cable, Neuropixels PXIe Acquisition Module, and NI BNC-2110 Connector Block with SpikeGLX 3.0 acquisition software (http://billkarsh.github.io/SpikeGLX/). The Neuropixels probe was configured to acquire in "long column" layout, providing the greatest possible depth span of recording while acquiring from a contact at each depth. Data was then spike sorted using Kilosort 2.5.^2,10,11 Automatically labeled "good" clusters were followed by partial manual curation using Phy¹² by an experienced electrophysiologist. Recordings were then classified on the basis of the presence of at least 1 putative single unit. Waveforms were generated as the average of all events on the highest amplitude channel, as calculated from the raw data.

Ethical Considerations

Ethical considerations of invasive neurophysiological recordings must be at the forefront of responsible research. Clinical safety must not be compromised for the sake of research, and principles for ethical adoption of this technology should be followed as relevant for the particular experiment.^13,14 In the work here, a patient’s participation was only considered after a surgical plan had been determined. Informed consent was obtained in advance of the procedure, and the experiments were done by a team that included both treating physicians and nontreating researchers. Re-evaluation for inclusion in the study was done 3 more times on the day of surgery: first, in the preoperative area in discussion with the patient; second, after craniotomy and durotomy were performed but prior to bringing sterile research equipment into the operative field; and third, immediately before insertion of the probe into brain parenchyma. Constant re-evaluation of patient safety was done during recording to consider if the probe should be retracted and the experiment halted.

Results

Demographic and Clinical Characteristics

Overall, 56 patients were included in this study (Tables 1 and 2). The mean ± SD age at time of surgery was 41.4 ± 16.7 years. Intractable epilepsy was the surgical indication in 33 (59.0%) participants, brain tumor in 19 (33.9%), and vascular lesions (i.e., underlying cavernoma) in 4 (7.1%). The pathology and recording location were in the left hemisphere in 39 (69.6%) participants and in the right hemisphere in the other 17 (30.4%). The pathology and recording location were in the temporal lobe in 34 (60.7%) participants, frontal in 20 (35.7%), and parietal in 2 (3.6%). The type of anesthesia was MAC in 50 (89.3%) participants and general anesthesia in the remaining 6 (10.7%). In cases where the participant was under MAC, a combination of dexmedetomidine and remifentanil was administered, with propofol sometimes being used during initial induction. In cases under general anesthesia, a combination of dexmedetomidine, propofol, and fentanyl or remifentanil was administered.

TABLE 1.

Participant summary information

	Value
Age, yrs	41.4 ± 16.7
Sex
Male	31 (55.4)
Female	25 (44.6)
Anesthesia
MAC	50 (89.3)
General anesthesia	6 (10.7)
Surgical indication
Epilepsy	33 (59.0)
Tumor	19 (33.9)
Vascular lesion	4 (7.1)
Recording location
Hemisphere
Lt	39 (69.6)
Rt	17 (30.4)
Lobe
Frontal	20 (35.7)
Temporal	34 (60.7)
Parietal	2 (3.6)
Probe fracture
No	53 (94.6)
Yes	3 (5.4)
MAC	3 (100)
Fragment recovered	3 (100)
During insertion	2 (66.7)
After insertion	1 (33.3)
132-mm-thick probe	0 (0)
Sharpened probe	0 (0)
Successful recording (putative single units)
Yes	47 (83.9)
No	9 (16.1)
Etiology of failure
Noise	6 (66.7)
No detectable spikes	3 (33.3)
Pathology
Epilepsy	1 (33.3)
Tumor	2 (66.7)

Values are shown as number (%) or mean ± SD.

Adverse Events

There were no observed surgical or medical complications in the 56 patients. Specifically, there were no infections or hematomas observed nor 30-day readmissions. There were no wound complications, and no patients received antibiotics beyond standard 24-hour postoperative prophylaxis.

Probe Fractures

Three probe fractures occurred (3/56 [5.4%]). In all cases, the participants were under MAC. Two fractures occurred at probe insertion, 1 of which occurred in the first participant who had no prior surgery (UC01); the other fracture occurred in a patient who had undergone 3 prior surgical procedures involving the region of interest, including craniotomy for grid implantation and responsive neurostimulation (RNS) implantation, both 10 months prior to the craniotomy for resection, as well as craniotomy for RNS removal 6 months prior to the craniotomy for resection (UC06). In this participant, the pia arachnoid was thickened relative to the cases with no prior surgical procedures. One of the fractures occurred after insertion, when the participant coughed (UC05 with no prior surgery). The time of probe fracture was obvious by both direct visual inspection and loss of electrophysiological data at the time of fracture. All 3 fractures occurred at the base of the probe shank, generating a 1-cm fragment (Fig. 1C) that was easily identified. In the cases where the fracture occurred during insertion, the fragment was located on the pial surface within 2–3 cm from the attempted insertion location and removed from the field using forceps. In the case where the fracture occurred while the probe was in the brain, an approximately 3-mm portion of the fragment was protruding from the surface of the brain and removed using nontoothed forceps. All fractures occurred within the first 6 cases, and thus occurred with 97-mm-thick probes before the implementation of thicker 132-mm probes (the first participant with a thicker probe was UC24). Similarly, sharpened probes² were first implemented for participant UC47, and thus occurred after our observed probe fractures (Fig. 2A).

FIG. 2.

Improved success with experience. A: The x-axis shows the participant ID in chronological order. The y-axis shows the cumulative proportion of cases where there was a shank fracture. The vertical red dotted line marks when 3D sharpened probes started to be used for all of the following cases. B: Similar to panel A, but for the cumulative proportion of cases where there was at least 1 detectable putative single unit. C: Similar to panels A and B, but for the cumulative proportion of cases where there was electrical noise that prevented isolation of single units.

Recording Success

The primary experimental goal of the technique is to record from single neurons. Among the 56 recordings, 47 yielded at least 1 putative single neuron (83.9%). Recordings with at least 1 single neuron improved with experience. In cases where there was a shank fracture, a second probe was used to attempt recording at a nearby site and thus shank fracture was not included as a cause of recording failure. Four of the first 10 recordings of single neurons were able to be resolved (40%), and 11 of the first 20 recordings of single neurons were able to be resolved (55%) (Table 2). The 20th recording (participant UC20) was the last recording in this series that did not yield at least 1 single unit (Fig. 2B).

TABLE 2.

Participant data

ID	Age (yrs)/Sex	Anesthesia	Hemisphere	Lobe	Pathology	Successful Recording	Shank Thickness (mm)	OR ID	Probe Fracture	Noise	No Detectable Spikes	Sharpened Probe	Breakage Time
UC01	31/M	MAC	Lt	Temporal	Tumor	No	97	A	Yes	Yes	No	No	Insertion
UC02	25/F	GA	Rt	Frontal	Epilepsy	No	97	B	No	Yes	No	No
UC03	49/M	GA	Rt	Frontal	Tumor	Yes	97	C	No	No	No	No
UC04	33/F	MAC	Lt	Temporal	Epilepsy	Yes	97	C	No	No	No	No
UC05	22/F	MAC	Lt	Temporal	Epilepsy	No	97	B	Yes	Yes	No	No	Recording
UC06	32/F	MAC	Lt	Frontal	Epilepsy	Yes	97	C	Yes	No	No	No	Insertion
UC07	54/F	MAC	Lt	Temporal	Epilepsy	No	97	C	No	Yes	No	No
UC08	38/M	MAC	Lt	Parietal	Epilepsy	No	97	C	No	Yes	No	No
UC09	57/F	MAC	Lt	Temporal	Tumor	No	97	C	No	No	Yes	No
UC10	36/F	GA	Rt	Temporal	Epilepsy	Yes	97	C	No	No	No	No
UC11	49/M	MAC	Rt	Frontal	Vascular	Yes	97	C	No	No	No	No
UC12	46/F	GA	Rt	Temporal	Epilepsy	Yes	97	C	No	No	No	No
UC13	25/M	MAC	Lt	Temporal	Epilepsy	Yes	97	C	No	No	No	No
UC14	35/M	MAC	Lt	Temporal	Epilepsy	Yes	97	C	No	No	No	No
UC15	27/M	MAC	Lt	Parietal	Vascular	Yes	97	C	No	No	No	No
UC16	26/F	GA	Rt	Temporal	Epilepsy	Yes	97	C	No	No	No	No
UC17	31/M	MAC	Rt	Temporal	Tumor	No	97	C	No	No	Yes	No
UC18	30/M	MAC	Lt	Temporal	Epilepsy	No	97	C	No	No	Yes	No
UC19	51/F	GA	Rt	Temporal	Epilepsy	Yes	97	D	No	No	No	No
UC20	40/F	MAC	Lt	Temporal	Epilepsy	No	97	C	No	Yes	No	No
UC21	37/M	MAC	Rt	Temporal	Epilepsy	Yes	97	D	No	No	No	No
UC22	28/M	MAC	Lt	Temporal	Epilepsy	Yes	97	E	No	No	No	No
UC23	24/M	MAC	Lt	Temporal	Epilepsy	Yes	97	C	No	No	No	No
UC24	24/M	MAC	Lt	Temporal	Epilepsy	Yes	132	F	No	No	No	No
UC25	52/M	MAC	Lt	Temporal	Epilepsy	Yes	132	C	No	No	No	No
UC26	24/M	MAC	Lt	Temporal	Epilepsy	Yes	132	C	No	No	No	No
UC27	53/F	MAC	Lt	Temporal	Tumor	Yes	132	C	No	No	No	No
UC28	55/F	MAC	Rt	Frontal	Tumor	Yes	132	C	No	No	No	No
UC29	22/M	MAC	Lt	Frontal	Epilepsy	Yes	132	B	No	No	No	No
UC30	74/M	MAC	Lt	Frontal	Tumor	Yes	132	G	No	No	No	No
UC31	69/F	MAC	Lt	Temporal	Tumor	Yes	132	C	No	No	No	No
UC32	40/F	MAC	Rt	Frontal	Tumor	Yes	132	C	No	No	No	No
UC33	24/M	MAC	Lt	Temporal	Epilepsy	Yes	132	C	No	No	No	No
UC34	42/F	MAC	Rt	Temporal	Epilepsy	Yes	97	C	No	No	No	No
UC35	37/M	MAC	Rt	Temporal	Epilepsy	Yes	132	C	No	No	No	No
UC36	69/F	MAC	Lt	Temporal	Tumor	Yes	132	C	No	No	No	No
UC37	36/F	MAC	Lt	Temporal	Tumor	Yes	132	B	No	No	No	No
UC38	38/M	MAC	Lt	Frontal	Tumor	Yes	132	C	No	No	No	No
UC39	53/F	MAC	Lt	Temporal	Epilepsy	Yes	132	C	No	No	No	No
UC40	43/M	MAC	Lt	Temporal	Epilepsy	Yes	132	C	No	No	No	No
UC41	39/M	MAC	Lt	Frontal	Epilepsy	Yes	132	D	No	No	No	No
UC42	62/M	MAC	Rt	Frontal	Tumor	Yes	132	C	No	No	No	No
UC43	67/M	MAC	Lt	Temporal	Tumor	Yes	132	B	No	No	No	No
UC44	29/F	MAC	Lt	Frontal	Epilepsy	Yes	132	C	No	No	No	No
UC45	70/M	MAC	Rt	Frontal	Vascular	Yes	132	C	No	No	No	No
UC46	67/F	MAC	Lt	Frontal	Tumor	Yes	132	G	No	No	No	No
UC47	23/F	MAC	Rt	Frontal	Epilepsy	Yes	97	C	No	No	No	Yes
UC48	22/M	MAC	Lt	Temporal	Epilepsy	Yes	132	C	No	No	No	Yes
UC49	34/M	MAC	Lt	Temporal	Epilepsy	Yes	97	C	No	No	No	Yes
UC50	69/F	MAC	Lt	Temporal	Epilepsy	Yes	132	C	No	No	No	Yes
UC51	18/F	MAC	Lt	Frontal	Vascular	Yes	132	C	No	No	No	Yes
UC52	87/M	MAC	Lt	Frontal	Tumor	Yes	97	B	No	No	No	Yes
UC53	29/M	MAC	Lt	Temporal	Epilepsy	Yes	132	C	No	No	No	Yes
UC54	63/F	MAC	Lt	Frontal	Tumor	Yes	132	C	No	No	No	Yes
UC55	31/M	MAC	Lt	Frontal	Tumor	Yes	132	C	No	No	No	Yes
UC56	27/M	MAC	Rt	Frontal	Tumor	Yes	132	G	No	No	No	Yes

GA = general anesthesia; OR = operating room.

There were two primary observed causes for the inability to record single units across these 9 cases, with 6 (66.7%) being due to electrical noise that prevented single-unit isolation, and the remaining 3 (33.3%) having no detectable single units despite low electrical noise. Five of the first 10 cases had electrical noise that prevented single-unit isolation (50%), as did the 20th case (Table 2, Fig. 2C). Of the 3 cases with low noise and no detectable single units, 2 had underlying tumor pathology (2 of 19 cases for tumor resection), and the remaining case’s underlying pathology was epilepsy (1 of 33 cases for epileptic foci resection). There were a total of 7 operating rooms used for the 56 surgical procedures, with most occurring in operating room C (n = 42), and 5 of the operating rooms were used only 1 to 3 times (Table 2). Electrical noise preventing single-unit recording occurred across multiple operating rooms (3/42 recordings in operating room C, 2/6 recordings in operating room B, and 1/1 recording in operating room A). In the cases with high noise, we cannot rule out coincident inability to detect single units.

Motion and Recording Quality

Electrode movement relative to the brain can be visualized by seeing the high-amplitude action potentials of high-rate neurons move across multiple channels. This can occur on multiple timescales. Brain pulsations from cardiac and respiratory cycle cause reliable, relatively fast drift of signals across electrodes, and overall stability such as tissue settling can occur on minutes-long timescales. Finally, there can be rapid shifts from sudden movements. Altogether, this can cause putative neurons to be detectable for a relatively short period in the total recording. Recording stability varied among recordings. As expected,⁵ recordings with more motion and recording instability (Fig. 3A and B) had lower overall recording quality. Stable recordings, those without much motion of electrodes relative to brain, had higher cluster quality (Fig. 3C and D).

FIG. 3.

Representative recording quality. Representative recordings with high (upper) and low (lower) motion. A: Scatterplot of detected spikes, where each dot represents an individual spike. Spike amplitude is depicted in grayscale, with darker shades indicating larger amplitudes. The x-axis shows time in seconds. The y-axis shows the distance from tip of the shank, with 0 mm indicating the deepest recording site. Example clusters, precuration, with colors corresponding to the waveforms on the right. Data from participant UC35 are shown. B: Cluster waveforms, calculated as the average of all events on the highest amplitude channel, as calculated from the raw data. The x-axis shows time in milliseconds, and the y-axis shows amplitude in microvolts. The shaded regions indicate ± 1 standard deviation around the average waveform for the cluster. C: As in panel A, but for representative recording with low motion. Data from participant UC32. D: As in panel B but for representative recording with low motion.

Acute/Subacute Histological Damage

En bloc cortical tissue was recovered from both the Neuropixels insertion site and the stereoelectroencephalography (sEEG) insertion site in 1 patient who had undergone ECoG and sEEG monitoring in the epilepsy-monitoring unit. To further evaluate the extent and chronicity of tissue damage surrounding the Neuropixels and sEEG tracts, we performed routine hematoxylin and eosin (H&E) staining, as well as immunohistochemical staining for NeuN, GFAP, CD68, neurofilament (NF), and amyloid β (Aβ). Consistent with prior studies in animal models,^15–19 tissue damage from Neuropixels probe recording was minimal (Fig. 4), consisting of an approximately 100-mm-wide tract of edema and ischemic appearing neurons with shrunken and pyknotic nuclei (on H&E and NeuN staining), focal axonal loss (NF), and occasional axonal spheroids (NF and Aβ). There was no significant neuronal loss (NeuN), astrogliosis (GFAP), or macrophage infiltration (CD68) around the Neuropixels probe tract. In contrast, the sEEG tract (from sEEG implantation 17 days prior to resection) revealed a cavitary lesion approximately the size of the 2-mm-diameter sEEG electrode on H&E staining (Fig. 3). The rim of tissue surrounding the cavitary lesion also showed substantial neuronal loss (NeuN), macrophage infiltration (CD68), astrogliosis (GFAP), axonal loss (NF), and prominent axonal spheroids (NF and Aβ), consistent with a subacute injury extending 500 mm to 1 mm from the edge of the cavity. Compared to the nearly indiscernible tissue injury surrounding the Neuropixels probe tract, the tissue injury surrounding the sEEG tract was much more extensive.

FIG. 4.

Histological damage is nearly indiscernible along the Neuropixels probe tract. Histological slices from participant UC48 after intraoperative Neuropixels recording (left) and sEEG recording (right) (implanted 17 days prior) at ×4 and ×20 magnification. The black arrows indicate the Neuropixels insertion tract. The rows, from top to bottom, correspond to different histological stains. The size of each scale bar is indicated at the bottom of each column. The area inside the dotted lines was magnified to ×20.

Discussion

High-density single-neuron recording provided by the Neuropixels probe has transformed neuroscience and has been recently translated to humans. Here, we demonstrate safety and yield from a large case series, showing that intraoperative Neuropixels recording can be done safely with minimal if any tissue injury. Furthermore, with experience and technical improvements, elimination of probe fractures along with increased single-neuron recording yields can be accomplished.

Minimizing Fracture Risk

Fracture risk and a retained fragment can be avoided using multiple layers of precaution. Prior to opening the probe into the sterile field, preparations should be made to make identification and recovery of a potentially fractured probe shank as easy as possible. We recommend lining the edges of the craniotomy site with gauze sponges for this purpose. There are three stages where fractures can occur after the silicon probe enters the sterile field. The first stage is when the probe is on the field but not yet in the process of insertion or recording. The probe should only be brought onto the sterile field when insertion is imminent, and once there, it should be protected through direct visualization and physical safeguards.

The second stage occurs during insertion, requiring continuous observation to detect any bending of the probe, which may indicate an impending fracture. If bending is observed, an alternative insertion site should be selected or aborting the procedure may be necessary. Structural modifications to the probe have helped reduce the risk of fracture during insertion. Increasing the thickness of the probe increases its resistance to buckling forces, and sharpening of the probe tip reduces insertion force. Piotomy over the insertion site can be considered to further reduce the insertion force. Further study is necessary to determine if these techniques alter the number of recorded single neurons.

The final stage of fracture risk occurs after insertion. Once the probe is in place, any large movements can cause fracture. Constant direct visualization and attention are essential during this period. In our approach, at least 1 surgeon-experimenter directly watched the probe and was ready to rapidly withdraw the probe if any large movements occurred. In the future, risk could be further reduced with the floating implantation method,²⁰ which would allow the probe to move with the brain, thereby reducing stress on the probe shank.

Deep Targets

The current devices in use are largely limited to surface cortical targets given the 10-mm shank length. Longer devices (45-mm long × 125-mm wide × 97- or 132-mm thick and 25-mm long × 125-mm wide × 97- or 132-mm thick, with 384 sites selectable for simultaneous recording) are currently deployed in nonhuman primate work²¹ and would be able to reach deep structures in humans. However, concerns of fracture in these longer devices must be addressed. Specifically, if a fracture was to occur within the brain itself, and not at the base of the shank as in our experience with the 10-mm probe, it may not be possible to retrieve the fragment. Transventricular trajectories should explicitly be avoided early in human use for this reason. Strategies utilizing guide tubes such as those implemented in nonhuman primate work,²² as well as testing in tissue that will be resected, should be done before other-use cases.

Tissue Insertion Tracts: Neuropixels Versus Clinical Penetrating Electrodes

The acute insertion tract caused by the Neuropixels probe is limited to about the size of the device cross-sectional area of 70 × 132 mm. Note that this is an order of magnitude smaller than currently available sEEG leads or deep brain stimulator leads, which have diameters of around 1–2 mm. The acute damage caused by Neuropixels probe insertion can be contained within the damage caused by a clinical penetrating electrode, and this is a strategy used by others.^6,9 This is a valid approach based on our experience, as it allows for the retrieval of all fragments without requiring tissue resection or causing any additional damage. As longer devices²¹ and insertion trajectories are used, it is critical that the Neuropixels trajectory still falls within the tract that the clinical electrode will follow, as it is expected that any of the tissue that is not contained within the tract of the clinical electrode will still have a permanent microstab injury.¹⁵ While the damage caused by Neuropixels insertion is diminutive and clinically insignificant, any unnecessary permanent brain injury, even if microscopic, should be avoided.

In this series, all Neuropixels insertions were done in craniotomies for resection into tissue that was subsequently resected as part of the previously determined plan. This contrasts with the other human Neuropixels recordings that were obtained after the Neuropixels was inserted into a burr hole but before insertion of a deep brain stimulator lead.^6,9 Future study will be necessary to determine if there are differences in the risks of probe fracture between these different exposure types.

Electrical Noise Isolation

The order of magnitude difference in acceptable root-mean-squared electrical noise between typical ECoG or depth electrode local field potential recordings (mV scale) and those of the Neuropixels probe (mV) makes it difficult to simulate isolation of electrical noise with clinical devices. Furthermore, noise isolation must be done within clinical equipment and time constraints. Given the variability in operating room setups, using the same operating room between cases is helpful to quickly remove noise sources that were previously identified. In our experience, 5 of the first 10 cases had noise isolation issues, and 4/5 of these initial cases with noise issues were in different operating rooms.

Cases of No Detectable Spiking

With only 3 cases where there were no detectable spikes during recording, it is difficult to draw conclusions as to the underlying causes of the lack of observed spiking activity. It is possible that there is an association of pathology or edema with lower yields, as many recordings were situated within or directly adjacent to the pathology. As more data are collected across multiple pathologies and their distinct tissue sequelae, an association between recording yields may arise.

Conclusions

Intraoperative Neuropixels recording can be done safely. Fractures are uncommon, occurring in 5% (3/56) of cases, and probe fragments were easily retrieved. Acute histological damage from probe insertion is on the order of tenths of a millimeter and undetectable in many cases. Experience reduces the chance of probe fracture and improves the success rate of microelectrode recording. As the technique is adopted by more surgeons and institutions, ongoing efforts to limit fractures and prevent retained fragments are necessary.

Acknowledgments

This work was supported by the Howard Hughes Medical Institute. J.E.C. is supported by UE5 (5UE5NS070680-15 from NIH/NINDS). C.R.C. is supported by K08 (K08NS126573 from NIH/NINDS). There were no corporate sponsors of this research.

Disclosures

Dr. Chang reported nonfinancial support from Echo Neurotechnologies outside the submitted work.

Author Contributions

Conception and design: Chung, Leonard, Sellers, Dutta, Cadwell, Chang. Acquisition of data: Chung, Xu, Greicius, Leonard, Sellers, Cadwell, Chang. Analysis and interpretation of data: Chung, Xu, Andrews, Cadwell, Chang. Drafting the article: Chung, Chang. Critically revising the article: Chung, Xu, Greicius, Leonard, Andrews, Welkenhuysen, Cadwell. Reviewed submitted version of manuscript: Chung, Xu, Greicius, Leonard, Sellers, Andrews, Welkenhuysen, Dutta, Cadwell, Chang. Approved the final version of the manuscript on behalf of all authors: Chung. Statistical analysis: Chung, Poddar. Administrative/technical/material support: Chung, Greicius, Poddar, Welkenhuysen, Dutta, Cadwell, Chang. Study supervision: Chung, Chang. Design and fabrication of the device used in the experiment: Dutta.

Supplemental Information

Previous Presentations

Abstract and poster presentation, Society for Neuroscience, Washington, DC, November 11–15, 2023.