The use of Neuropixels probes for chronic neural recordings is not yet widespread, and initial studies leave questions about long-term stability and probe reusability unaddressed. Here, we demonstrate a new approach for chronic Neuropixels recordings over a period of months in freely moving rats. Our approach allows multiple probes per rat and multiple cycles of probe reuse. We found that hundreds of units could be recorded for multiple months, but that yields depended systematically on anatomical position. Explanted probes displayed a small increase in noise compared to unimplanted probes, but this was insufficient to impair future single-unit recordings. We conclude that cost-effective, multi-region, and multi-probe Neuropixels recordings can be carried out with high yields over multiple months in rats or other similarly sized animals. Our methods and observations may facilitate the standardization of chronic recording from Neuropixels probes in freely moving animals.

Subjects
A total of 18 male Long-Evans rats (Rattus norvegicus) were used in this study. Four of them were BAC transgenic rats expressing Cre recombinase. These rats were used for the purpose of targeting optogenetic constructs to cell types of interest for experiments not described in the present report. These rats came from the following three lines: LE-Tg(Pvalb-iCre)2Ottc (n=1), LE-Tg(Gad1-iCre)3Ottc (n=2), LE-Tg(Drd2-iCre)1Ottc (n=1). These lines were made by the Transgenic Rat Project at the National Institute of Drug Abuse (NIDA) and were obtained from the Rat Resource and Research Center (RRRC) at the University of Missouri.  All rats were water restricted to motivate them to work for water as reward, and obtained a minimum volume of water per day equal to 3% of their body mass. If rats consumed less than this minimum during performance of the task, additional water was offered after training. Rats were kept on a reversed 12-hr light-dark cycle and were trained in their dark cycle. Rats were pair housed whenever possible during early training, but were always single housed after implantation to prevent damage to the implant. Starting from the day before each surgery to at least three days after the surgery, rats were given ab lib access to water. Animal use procedures were approved by the Princeton University Institutional Animal Care and Use Committee and carried out in accordance with National Institute of Health standards.

Implant Construction
The probes used were commercially available Neuropixels 1.0 probes (IMEC, Leuven, Belgium), with the option of a flat silicon spacer attached parallel to the plane of the probe, except for one (among 20), which was phase 3A option 1 probe. A protocol for constructing the implant is available, with photographs illustrating each step, at https://github.com/Brody-Lab/chronic_neuropixels. We summarize these procedures here. First, to prepare a probe for implantation, the silicon spacer was glued to a 3D-printed plastic base with dovetail rails, inspired by the stainless steel dovetail adapter now optionally shipped on the probes. This dovetail connection mated with matching dovetail rails on an internal holder that could be manipulated using a stereotaxic holder (Kopf, Tujunga, CA, USA; Model 1766-AP Cannula Holder). The gold pads on the probe flex cable corresponding to ground and external reference were shorted by soldering a 0.01” silver wire to them. The entire apparatus was then enclosed with a 3D-printed external chassis and the holder and the chassis were connected using screws. The chassis was then wrapped in conductive tape to provide electromagnetic shielding and coated in a thin layer of C&B Metabond (Parkell, Edgewood, NY, USA) for better adherence to the dental acrylic during implantation.  The bottom surface of the chassis was sealed with petroleum jelly, dispensed in liquid droplets from a low-temperature cautery (Bovie), to prevent blood from entering the chassis after implantation. During the surgery, after the probe was inserted into the brain, the chassis, but not the holder, was fixed to the skull surface using adhesive cement and acrylic. The internal holder did not come into contact with the adhesives and therefore can be detached from the chassis by removal of the screws.

During recording, a 3D-printed “headstage-mount” mated directly with the chassis through a latching mechanism, and mounted headstages were connected to the flex cable of a probe. Up to four headstages can be mounted so that up to four probes can be simultaneously implanted and recorded using this system. (Although, note that the size and geometry of the implant places constraints on the set of coordinates that can be simultaneously targeted, constraints that become more significant when more probes are implanted at once). The recording cables were placed inside a corrugated plastic sleeving to avoid small bending radiuses (1/2” internal diameter, McMaster 7840K73). This sleeving was permanently attached to the recording cap using electrical tape.

All parts except the probe itself and the screws were 3D printed in-house using Formlabs SLA printers (Form 2 and Form 3). Prints used standard black Formlabs resin, except for the recording cap which was printed in “Tough” Formlabs resin, required for the flexible latching mechanism.
A subset of implants (n=7) were made using an earlier version of the 3D design that embodied a similar design principle but were slightly larger in size. Results were similar between the two versions of the design and were therefore combined.

CAD files, detailed protocols for implant assembly and surgeries, code, and data are provided at https://github.com/Brody-Lab/chronic_neuropixels.

Implantation
Surgeries were performed using techniques that were similar to those reported previously (Erlich et al., 2011). Rats were placed in an induction chamber under 4% isoflurane for several minutes to induce anesthesia and then injected with 10 mg ketamine and 6µg buprenorphine intraperitoneally to further assist induction and provide analgesia. Rats were then placed on a stereotaxic frame (Kopf Instruments) and anesthesia was maintained with 1-2% isoflurane flowing continuously through a nose cone. After verifying surgical levels of anesthesia, rats were secured in ear bars, shaved and cleaned with ethanol and betadine. A midline incision was then made with a scalpel and a spatula was used to clean the skull of all overlying tissue.  

A small, crater-like craniotomy roughly 1mm in diameter was made at the site of probe implantation. A needle was used to cut an approximately 0.5 mm slit in the dura into which the probe was later lowered. This procedure was repeated at the site of each probe to be implanted. Next, a craniotomy and durotomy were performed at a site distal to the brain region(s) to be recorded, typically in the olfactory bulb or cerebellum, into which a ground (wire, pin, or cannula) would later be inserted. Saline soaked Gelfoam (Pfizer Injectables) was placed in the craniotomies to protect the brain while a thin coat of C&B Metabond (Parkell, Inc) was applied to the skull surface. Then, the Gelfoam was removed from the ground craniotomy and the ground was lowered into the brain. This craniotomy was then sealed with Kwik-Sil (WPI) and the ground was fixed in place using a small amount of dental composite (Absolute Dentin, Parkell, Inc.). 
Gelfoam was removed one by one from the remaining craniotomies and the probes inserted. Probes were manually lowered at a rate of 0.1 mm every 10-30 s. After a probe was fully inserted, a small quantity (<5uL) of soft silicone gel (DOWSIL 3-4680, Dow Chemical) was injected to seal the craniotomy. We found that harder silicones, such as Kwik-Sil (WPI), could damage the probe shank upon application. Dental composite (Absolute Dentin, Parkell) was then used to create a wall-like barrier covering the gaps below the four ventral edges of the external chassis and the skull, creating a seal around the probe shank. After these steps were completed for each probe, the silver wires soldered to the ground and external reference pads on each probe flex cable were then soldered to the common animal ground. Acrylic resin (DuraLay, Reliance Dental) was applied to secure the entire implant assembly to the Metabond-coated skull. Rats were given ketoprofen after the surgery and again 24 and 48 hours post-operative and recovered with ad lib access to water for 5 days before returning to water restriction.

Explantation
To explant a probe, the animal was first anesthetized and placed in the stereotaxic frame. Then, a stereotaxic arm was attached to the probe’s internal holder and the screws fixing the internal holder and external chassis were removed. The stereotaxic arm was raised until the internal holder, carrying the probe, was fully removed. The internal holder sometimes adhered to the external chassis after screw removal. In these cases, the external chassis was carefully drilled away with a dental drill until the internal holder could be easily removed. After explantation, the probe shank was fully immersed in 1% tergazyme (Alconox) for 24-48h, followed by a brief rinse in distilled water and isopropyl alcohol, in that order.

Electrophysiological recordings 
Electrophysiological recordings were performed using either commercially available Neuropixels 1.0 acquisition hardware (Putzeys et al., 2019) or earlier test-phase IMEC acquisition hardware. The former was used in conjunction with PCI eXtensions for Instrumentation (PXI) hardware from National Instruments (including a PXIe-1071 chassis and PXI-6133 I/O module for recording analog and digital inputs.) We used SpikeGLX software (http://billkarsh.github.io/SpikeGLX/) to acquire the data. For measurement of signal stability over time, the selected reference was a silver wire shorted to the ground wire and penetrating the brain at a different location from the probe insertion site. The amplifier gain used during recording was 500. The recording channels addressed either the deepest 384 recording sites (“bank 0”) or the second deepest 384 recording sites (“bank 1”), and recordings lasted approximately ten minutes. Recordings were made on days with logarithmically-spaced intervals.

Spikes were sorted offline using Kilosort2 (Pachitariu, 2020), using default parameters and without manual curation. A unit was considered a single-unit if Kilosort2 categorized that unit as “good.” A putative single unit meets both of the following conditions: 1) the proportion of refractory violations was less than 0.05. This proportion was computed by calculating a unit’s autocorrelogram in one millisecond bins, summing the values in the central bins, and normalizing the sum by the expected sum from the mean firing rates. This normalized value was compared to a normalized sum similarly calculated from the shoulder of the autocorrelogram. The span of the central bins range varied from +/-1.5 ms to +/-10.5 ms, and a separate ratio was computed for each span. The span of the shoulder bins were either +/-10.5 ms to +/-50 ms or +/-25.5 ms to +/-499.5 ms. A separate ratio was calculated for each combination central bin span and shoulder bin span. The minimum ratio across combinations was the proportion of refractory violations. 2) For each span of central bins, a probability was calculated for the number observed coincidences in the central bins of the autocorrelogram, or fewer, being produced by a Poisson distribution with observed mean firing rate of the unit. The minimum probability across spans of central bins has to be less than 0.05. 
 
Behavioral task
Training took place in an operant box with three nose ports (left, center and right) and two speakers, placed above the right and left nose ports. Each trial began with a visible light-emitting diode (LED) turning on in the center port. This cued the rat to insert its nose into the center port and keep it there until the LED was turned off (1.5 s “fixation” period). In each trial, rats were concurrently presented with two trains of auditory pulses, one train from a loudspeaker on its left, and the other from a loudspeaker on its right (Brunton et al., 2013). At the end of each trial, the subjects received a reward if they oriented to the port on the side of the auditory train with the greater total number of pulses. The timing of pulses varied both within and across individual trials and between the left and right streams. Behavioral control and stimulus generation used Bcontrol software (brodylabwiki.princeton.edu/bcontrol). Before probe implantation, subjects were trained in a semi-automated, high-throughput training facility, each in a behavioral enclosure 13” wide, 11” deep and 18.5” tall within a sound-attenuated chamber (Coulbourn, Holliston, MA, USA). After surgery, tethered behavioral performance took place in a larger behavioral enclosure that was 13” wide, 11” deep and 40” tall within a sound-attenuated chamber and built-in Faraday cage (IAC, Naperville, IL, USA). Untethered performance took place in the training facility, as before surgery. 

Behavioral performance metrics
The performance of each rat (pooled across multiple sessions) was fit with a four-parameter logistic function:

P(right) =?0 + ?1/(1+exp(-ꞵ(x-ɑ))                           

where x is the click difference on each trial (number of right clicks minus number of left clicks), and P(right) is the fraction of trials when the animal chose right. The parameter ɑ is x value (click difference) at the inflection point of the sigmoid, quantifying the animal’s bias; ꞵ is the slope of the sigmoid at x = ɑ, quantifying the sensitivity of the animal’s choice to the stimulus; ?0 is the minimum P(right); and ?0+?1 is the maximum P(right). The lapse rate is (1-?1)/2. The number of trials completed excludes trials when the animal prematurely broke fixation and trials after the animal in which the animal failed to choose a side port after five seconds. 

Measurement of input-referred noise
The RMS noise of the AP band (300Hz - 10kHz) of each recording site was measured in a 0.9% phosphate buffered saline (PBS) solution, following the procedure described in the Neuropixels User Manual (IMEC). To accurately compare across recording sites and probes, this procedure requires first determining a gain correction factor for each recording site, so that the nominal and actual gains match. To do this, we measured the response to a 4mVpp, 1.8kHz sine wave generated using an arbitrary waveform generator (PXIe-5413, National Instruments) in 0.9% PBS. We compared the measured amplitudes on each recording site to the amplitude measured on an independent, calibrated I/O module (PXI-6133, National Instruments) to determine the gain correction factor.

Choice selectivity
This metric is based on the receiver operating characteristic (ROC) and indexes how well an ideal observer can classify left- versus right-choice trials using the spike counts of an isolated unit. Spikes were counted in 0.1 s bins stepped in 0.02 s, from 1 s before movement onset to 0.5 s afterwards. Trials were included for analysis if the rat did not prematurely disengage from the center port and also reported its choice within five seconds after it was cued to do so. An ROC curve classifying left- and right-choice trials was constructed based on the spike counts of each unit in each time bin. The area under the ROC curve ranged from 0 to 1, with values greater than 0.5 indicating a larger mean spike count associated with right-choice trials. Because the present analysis concerns only the magnitude and not the directionality of the choice selectivity, a value x less than 0.5 was flipped to the corresponding value above 0.5, i.e. |x-0.5| + 0.5. The choice selectivity results were from the first recording session for each animal after the implantation when the animal had completed more than a hundred trials (4-11 days after surgery).