Chronically implanted Neuropixels probes enable high-yield recordings in freely moving mice

Ashley L Juavinett; George Bekheet; Anne K Churchland

doi:10.7554/eLife.47188

. 2019 Aug 14;8:e47188. doi: 10.7554/eLife.47188

Chronically implanted Neuropixels probes enable high-yield recordings in freely moving mice

Ashley L Juavinett ¹, George Bekheet ², Anne K Churchland ^3,^✉

Editors: Laura L Colgin⁴, Laura L Colgin⁵

PMCID: PMC6707768 PMID: 31411559

Abstract

The advent of high-yield electrophysiology using Neuropixels probes is now enabling researchers to simultaneously record hundreds of neurons with remarkably high signal to noise. However, these probes have not been well-suited to use in freely moving mice. It is critical to study neural activity in unrestricted animals for many reasons, such as leveraging ethological approaches to study neural circuits. We designed and implemented a novel device that allows Neuropixels probes to be customized for chronically implanted experiments in freely moving mice. We demonstrate the ease and utility of this approach in recording hundreds of neurons during an ethological behavior across weeks of experiments. We provide the technical drawings and procedures for other researchers to do the same. Importantly, our approach enables researchers to explant and reuse these valuable probes, a transformative step which has not been established for recordings with any type of chronically-implanted probe.

Research organism: Mouse

Introduction

Observing behavior and recording neural activity in freely moving animals is crucial for our understanding of how the brain operates. Electrophysiology in freely moving rodents has been used to observe place and grid cell dynamics (Hafting et al., 2005; O'Keefe and Dostrovsky, 1971), cortical dynamics during attentional control (Bolkan et al., 2017), the role of oscillations during fear learning (Stujenske et al., 2014), whisking behavior during exploration (Kerekes et al., 2017), the effect of environmental context on neural activity (Whitlock et al., 2012), and the control of sensory selection in divided attention (Wimmer et al., 2015), to name a few. Although freely moving recordings can be challenging, recording from unrestrained mice enables researchers to investigate behaviors that involve natural movements and offers ethologically valid insight into neural activity (Juavinett et al., 2018; Markowitz et al., 2018). Electrophysiology in freely moving animals is commonly performed with static electrode arrays or microdrives (Okun et al., 2016; Vandecasteele et al., 2012; Voigts et al., 2013). These techniques have contributed much to the field, but are not at pace with the spatiotemporal coverage of cutting edge recording techniques, such as Neuropixels probes (Jun et al., 2017; Steinmetz et al., 2018). Given the experimental tractability of the mouse and the increasing interest in ethological approaches in neuroscience research, we sought to develop a system that would enable researchers to perform repeatable high-yield recordings.

Recent advancements in semiconductor technology have enabled the development of high-density silicon probes known as Neuropixels (Jun et al., 2017). The linear recording shank can record from 384 contacts across 3.84 mm (selectable from 960 available sites on a 10 mm length shank). In the mouse brain, which is at most 6 mm deep, this span of contacts means researchers can simultaneously record from more than half of the depth of the brain. Further, Neuropixels probes have low baseline noise levels (<6 µV RMS), comparable to other silicon probes (Steinmetz et al., 2018). However, Neuropixels probes also have on-site amplification and digitization, thereby enabling simultaneous recording of hundreds of cells across brain regions in an unprecedented low-noise, high-throughput manner. Importantly, methods have also been developed to automatically sort spikes from these recordings, and even correct for probe drift (Jaeyoon et al., 2017; Pachitariu et al., 2016).

Neuropixels probes have already proved invaluable for neuroscientists conducting acute experiments in mice, or chronic experiments in freely moving rats (Jaeyoon et al., 2017; Krupic et al., 2018; Vélez-Fort et al., 2018). However, there is limited work with these probes in unrestrained mice (Evans et al., 2018), likely because of the difficulty designing small, lightweight recording devices. Still, there is plentiful interest in behaviors and computations that involve movements of the animal’s head in space (Vélez-Fort et al., 2018), foraging (Lottem et al., 2018), pup retrieval (Marlin et al., 2015), or naturalistic fear responses (Evans et al., 2018). Further, although these probes have been very successful in freely moving rats (Jaeyoon et al., 2017; Krupic et al., 2018), there is not an established method to recover them after the experiment.

The opportunity to explant and reuse Neuropixels probes is transformative. Given the cost ($1000 each, https://www.neuropixels.org/) and limited availability of the probes, many researchers will only be able to use them if it is possible to recycle them after experiments. The ability to recover these probes would enable researchers to repeat their experiments in different animals, boost the statistical power of their experimental findings, and thus enhance reproducibility of experimental data. We therefore sought to design a device for the Neuropixels probe that would allow experimenters to chronically implant it, run an experiment, and explant it for future experiments.

Several major innovations are required to design a removable holder for chronic implants of Neuropixels probes in unrestrained mice. First, the current design of the probe has several components that need to be securely mounted onto the small mouse skull. Further, these sensitive onboard electronics need to be protected while the mouse is in its home cage. Most importantly, the shank of the probe must be secured to ensure consistent recordings across weeks of recording. In previous work, this required permanently mounting the biosensor using adhesive, which was effective but made it nearly impossible to remove the probe afterwards (Okun et al., 2016). We also opted to use a 3D printed device in order to limit the use of acrylic in our design and ensure that it would be lighter than alternative designs.

To address these needs, we designed the Apparatus to Mount Individual Electrodes (AMIE), a device that fully encases and protects the sensitive onboard electronics of the Neuropixels probe, allowing long term, freely moving experiments. Moreover, the Neuropixels AMIE allows explantation and recycling. Our design and protocol is applicable to laboratories that wish to adapt the Neuropixels probe, or similar silicon probes, for recording in freely moving mice. With the drawings, materials and instructions, our device can be implemented not only by labs with years of expertise in electrophysiological recordings, but also by labs with different expertise that encounter a new need to study neural activity during behavior. Researchers that are using this technology in primates, rats, or in acute mice experiments may also find aspects of this approach useful.

With this design we have successfully recorded ~100 neurons simultaneously from unrestrained mice while observing freely moving behavior, and explanted the Neuropixels probe with a functioning recording shank. Further, because the AMIE is designed to allow implantation of a headbar (if desired), we recorded from the same mice in head-fixed experiments using systematic presentation of traditional visual stimuli. This feature of the AMIE allows experimenters to study neural activity in both psychophysical and ethological paradigms, affording the chance to build a bridge between the two.

Results

Design overview

The entire AMIE device weighs ~1.5 g (with cement:~2.0 g) and is assembled from three parts: the Neuropixels probe, the internal mount (IM), and external casing (EC) (Figure 1A,B; Video 1). The IM attaches directly to the Neuropixels PCB board with adhesive and is the core of the assembly (Figure 1a). On the backside of the IM is a slot for a stereotax adapter (SA) which allows for easy handling of the probe (Figure 1A). The IM attaches to the EC via a rail system (Figure 1B). During the implantation procedure, all adhesive binding the assembly to the rodent’s skull exclusively contacts the EC, which acts as a protective shell (Figure 1D).

Figure 1. — (A) Probe base mounted onto 3D printed internal casing and attached to machined metal stereotax adapter. Inset: Rear view, with screws that attach the internal mount (IM) to the stereotax adapter (SA). (B) Entire assembly in a. within 3D printed external casing. Inset: Rear view. (C) The headstage is positioned on the back of the encasing, with the flex wrapped in an ‘S’ shape. (D) Entire assembly in relation to size of mouse brain and skull. The EC is attached to the skull with cement. Silicon gel is used to as an artificial dura to protect the open craniotomy.

Video 1. 3D rendering of the AMIE device demonstrating the configuration of internal mount (IM), external casing (EC), and stereotax adapter.

Download video file^{(591.2KB, mp4)}

DOI: 10.7554/eLife.47188.003

Open in a new tab

One difficulty in adapting the current Neuropixels design for freely moving experiments in mice is the ~3 cm long flex cable attached to a 1 g headstage (see Jaeyoon et al., 2017) for details). In early testing, we suspended the flex and headstage above the mouse’s head during recording. However, we found that the flex very quickly twisted, potentially damaging it. In addition, the headstage added swinging weight above the mouse’s head. With these observations in mind, we designed the encasing with a space for the headstage to be semi-permanently affixed. The probe flex wraps in an ‘S’ shape behind the implant, and attaches to the bottom (Figure 1C). In this way, the recording cable can be attached to the top of the implant, suspended above the mouse’s head.

Protocol overview

At least one day prior to implant, we attach the probe to the internal mount (Figure 2A). Silicone is added to further secure the base of the recording shank (Figure 2B). Once this is dry, the internal mount is slid into the rails of the external casing and secured with cement (Figure 2C,D). This cement will be drilled away in order to explant the probe. When the entire AMIE assembly is dry, it is ready to be implanted (Figure 2E). The surgery to implant the probe and encasing typically takes ~3 hr (see Materials and methods for details). During this surgery, a headbar can also be implanted, which does not interfere with the encasing. The external casing is the only part of the assembly that is attached to the skull (Figure 2F). In a typical experiment, we implant the probe and encasing without the headstage attached. We wait ~3–4 days for the mouse to recover, and then add the headstage. The headstage can be removed after each experiment, if desired. After ~1 day of habituation to the additional weight of the headstage (~1 g), we begin recording during behavior.

Mice are mobile with the implant

Neuropixels probes were not designed for chronic implants in freely moving mice, and the entire probe assembly is quite bulky in comparison to a mouse’s head (Figure 1D; Jaeyoon et al., 2017). However, we have designed a very slim encasing for the probe, and mice can adjust to the weight and size of the implant (Video 2).

Video 2. Behavior of mouse implanted with Neuropixels AMIE.

Download video file^{(1.8MB, mp4)}

DOI: 10.7554/eLife.47188.005

Open in a new tab

Mouse was free to move around a 16”x16’ arena while implanted and tethered. Video is shown at 2x speed.

By approximately 48 hr post-surgery, mice were mobile with the Neuropixels AMIE (Figure 2G). To evaluate the suitability of the AMIE for use during behavior, we assessed the impact of the device on both spontaneous and stimulus-driven movements. For spontaneous behavior, we analyzed video data taken while mice explored an open arena (Figure 3A,B). Even while tethered, implanted mice were typically agile and active (Video 2). To quantify behavior and compare for implanted vs. naive mice, we calculated three metrics from video data: the percentage of time spent moving, the maximum velocity and the maximum acceleration. For all three metrics, considerable overlap was apparent in the distribution of values for implanted and naive mice (Figure 3D). Although implanted mice moved slightly less and were slightly slower, the differences failed to reach significance for any metric. In fact, the mouse with the highest max acceleration was implanted (Figure 3D, middle panel).

To examine stimulus driven behavior, we measured responses to overhead visual looming stimuli (Figure 3E), which are known to elicit strong escape responses in mice (De Franceschi et al., 2016; Evans et al., 2018; Yilmaz and Meister, 2013). The distribution of values for the metrics tested (mean/max velocity and max acceleration) again overlapped considerably for naïve vs. implanted mice (Figure 3E). Although we observed no significant changes, a few naive mice achieved max acceleration during their escapes at values unobserved in implanted mice (Figure 3E, right). A possible explanation is that naive mice were free from the weight of the device and thus were able to accelerate very quickly when motivated to do so by a threatening stimulus. Taken together, these behavioral observations argue that although the presence of the AMIE may have idiosyncratically slowed mice slightly, they remained active in an open arena and showed species-typical responses to threatening stimuli.

Neuropixels AMIE allows for 60–100 simultaneously recorded neurons across weeks of freely moving behavior

We recorded spiking activity across multiple brain areas during freely moving behavior over the course of 1–2 weeks. Figure 4 illustrates an experiment with the probe implanted in medial visual cortex, subiculum, and midbrain. We isolated ~60–100 units for each session in this experiment (Figure 4D,E), during which the mouse moved freely around the arena and was exposed to looming stimuli. The number of single units we were able to isolate ranged across mice and experiments from ~20 to 145, but these numbers were fairly consistent within each mouse across recording sessions (Figure 4C). This variability is likely dependent on the probe that was used (Phase 3A Option four probes used in mouse #3 and #4 had 270 rather than 374 recordable channels; see Materials and methods and Jun et al., 2017), recording noise, and brain region. The absolute number of isolated units depends on the quality of the sorting and the experimenter’s manual curation of Kilosort output, which does present challenging edge cases and can be difficult to assess with drift in the experiment. Overall, these numbers are less than has been previously reported with acute experiments in mice (Jun et al., 2017), possibly because of the chronic recording environment or inability to completely reduce noise. The longest we left a probe in was 41 days, without any noticeable decay in the signal.

To test how automatic unit sorting and classification would compare with our approach, we also sorted one of these freely moving sessions with Kilosort2, which automatically classifies units as ‘good.’ Indeed, for mouse #7, Kilosort2 identified 77 well-isolated units, compared to 69 with Kilosort1 and manual post-Kilosort designations in phy, confirming that our manual criteria were effective.

We elected to be conservative about any claims that the same neurons were recorded across days of the experiment, because demonstrating a stable recording of the same neurons from day-to-day is difficult and often regarded with skepticism. However, we did indeed observe waveforms that were consistent in both shape and depth across recordings, and it is entirely possible that these originate from the same neurons (Figure 4F).

Researchers can also conduct headfixed recordings to further characterize neurons

A major limitation of many chronic implant designs is that they do not enable researchers to also implant a headbar to restrain the animal. The ability to head-fix animals critical for two reasons. First, it allows the experimenter to easily restrain the mouse during experiments, for example to attach/replace the headstage or fix twisting in the tether. Moreover, it affords the opportunity to measure neural activity in response to traditional psychophysical stimuli after the freely moving recording (Figure 5). This makes it possible to connect the neural responses obtained during an unrestrained, ethological task with those obtained during more traditional sensory electrophysiology context (simple stimuli defined by parameters that are systematically varied). This opportunity could prove invaluable in bridging observations from these two very different contexts which are normally studied in separate laboratories.

Figure 5. — (A) Schematic of headfixed setup. The mouse was implanted with a headbar (see Materials and methods) enabling it to be restrained above a wheel. Visual stimuli was presented on a monitor above the mouse’s head (similar to the unrestrained condition). The mouse’s pupil can be tracked with a high resolution IR camera, and movement can be tracked using a rotary encoder on a 3D printed wheel. (B) Eight sample waveforms (n = 200, mean in yellow) from a restrained recording, same mouse as Figure 4D–F (Mouse #2). (C) Distribution of sorted units across the probe, same mouse as in Figure 4D–F (Mouse #2). (D) Peristimulus time histograms for three example neurons from different locations on the probe. The stimuli were a pseudorandomized set of 2 s full contrast sinusoidal drifting gratings in eight different directions. Shaded region is standard error of the mean. Stimuli began at the dotted line. (E) Raster plots for the neurons in (D). Shaded area indicates the duration of the stimulus.

For example, after six days of recording freely moving behavior, we presented a battery of visual stimuli while the mouse was head-fixed to determine whether cells were visually responsive (Figure 5A). We were able to isolate 60 units (63 with Kilosort2) in the restrained condition, just as in the freely moving condition (Figure 5B). The distribution of units was similar to previous experiments where the mouse was not restrained.

Implant allows researchers to recover the probe after the experiment

Beyond providing a stable implant over many days, we also sought to design an implant that would allow for recycling of the Neuropixels probes. As demonstrated in Figure 1, the internal mount is separate from the external casing that is cemented to the mouse. After the completion of the experiment, researchers can drill away the cement and slowly remove the probe (see Materials and methods and Figure 6A). This same probe, still attached to the internal mount, can then be re-secured within an external casing and implanted in another mouse.

We were able to record from a mouse for over 2 weeks, explant the probe, and re-implant for a second experiment (Figure 6B). We were easily able to isolate clear units in both (Figure 4C and 6C). Although we initially were able to isolate comparable numbers of units to the probe’s first implant, the number of isolated units fell over time (compare mouse #5 and #7 in Figure 4C). Still, the re-implanted probe yielded 43.6 ± 17.8 neurons which is ample for many studies, especially those conducted in labs for which Neuropixels probes are a scarce resource.

In another experiment, were able to explant a probe from a mouse that did not recover from surgery and reimplant it in a second mouse (see Table 1). Although this second mouse ultimately also had complications resulting from a poorly positioned ground wire, one successful session of recording yielded 145 units (Figure 4C; Table 1). Further experiments will determine the unit yields that can be typically expected following reimplantation.

Table 1. Overview of experiments, with the Neuropixels probe option used and the outcome of the experiment.

For each of these experiments, even the unsuccessful explants, neural data was obtained from the initial implant and recording sessions. For an explanation of the probe options, see Materials and methods. Starred mice are included in the paper; + sign indicates the experiment was sorted with Kilosort2; M = mean; SD = standard deviation.

Mouse	Probe option	Recordable channels	M ± SD Isolated units	Silicone on shank	Outcome
NP6* (Figure 4; Mouse #3)	4	276	19.8 ± 6.23	No	Shank broke during explant
NP7* (Figure 4; Mouse #4)	4	276	20.0 ± 4.36	no	Shank broke during freely moving recording
NP8* (Figure 4; Mouse #2)	1	384	77.2 ± 13.7	no	Shank broke during explant
NP9* (Figure 4; Mouse #1)	1	384	117.4 ± 16.3	no	Shank broke during explant
NP11	1	384	-	no	Shank broke during freely moving recording
NP12	3	384	-	no	Mouse didn’t recover from surgery, probe successfully explanted and re-implanted in NP13
NP13 (Figure 4; Mouse #8)	3	384	145⁺	yes	Ground wire issues after surgery; one session successfully recorded. Successful explant
NP14* (Figure 4 and 6; Mouse #5)	3	384	80.6 ± 13.6	yes	Successful explant, re-implanted in NP16
NP15* (Figure 4; Mouse #6)	3	384	64.3 ± 19.1	yes	Successful explant
NP16* (Figure 4 and 6; Mouse #7)	3	384	43.6 ± 17.8	yes	Successful explant

Open in a new tab

To assess the stability of the probe and our ability to detect spikes, we computed the event rate (sum of temporally coincident spikes on a group of sites for which the maximum amplitude exceeds the threshold) and signal-to-noise (SNR; see Materials and methods) ratio for the first implant of the probe as well as its re-implantation in another animal. There was a drop in the event rate and a small drop in the SNR in the re-implanted probe (Figure 6D,E). However, even with the first implant there was a significant drop in both event rate and SNR on the 12th day of recording, suggesting that this may not be due to the re-implantation itself.

Successful explant of probes depended on several factors. First, applying silicone to the base of the shank to add extra support appears to be necessary (Figure 1B). With silicone added to the base of the shank, 4/4 explant attempts were successful, whereas 1/6 explants were successful without the silicone (Table 1). Second, careful alignment of the probe, internal mount, and external casing will help ensure that the shank is being removed at the appropriate angle. Third, we only had success with Phase 3A Option three probes, suggesting that it may be easier with these, possibly due to the fact that the recording shanks on these probes are longer (10 mm) than Option 1 (5 mm). Fortunately, the shank of Phase 3B probes (now on the market) is also 10 mm long.

Discussion

Here, we present a significant advance in our ability to use and recycle high-density silicon probes such as Neuropixels. Our device, the AMIE, and accompanying methods, allow researchers to perform recordings in both restrained and unrestrained conditions, and critically, to explant and reuse probes after experiments. This approach will enable researchers to capitalize on important technological advances to understand the complexity of brain activity during ethological behaviors, and to bridge the gap between ethological and psychophysical behaviors (Gomez-Marin et al., 2014).

Although Neuropixels probes were not designed for unrestrained recording in mice, our AMIE customizes them so that they are ideally suited to this purpose. The AMIE has a slim enclosure for the probe as well as the headstage (Figures 1 and 2), that mice can easily handle (Figure 3). It is worth noting that the Neuropixels design featured here is 3A, but the 3B (Neuropixels 1.0) version is the one currently commercially available. AMIE designs for both probe generations are available in the resources for this paper (see Materials and methods). Our design can also be readily adapted to other types of silicon probes (e.g. Neuronexus).

Unlike other electrophysiology systems, the current Neuropixels recording tether is not easily commutated due to heavy data demands. While this has not been a problem for recording from chronically-implanted rats in large arenas (Jun et al., 2017), it can be challenging for recordings from mice in smaller arenas, requiring constant monitoring of the mouse’s position and occasional intervention from the experimenter to untangle the cord. In our experience, this is manageable, requiring the experimenter to stop an hour-long session once or twice to unplug the cord and untangle. Importantly, here we report similar behavior both during open field exploration and looming-evoked escape responses in implanted and naive mice (Figure 3).

The Neuropixels AMIE can be used to record in both restrained and unrestrained conditions, with similar yields in numbers of isolated units (Figures 4 and 5), although a direct comparison of yields across publications is challenging due to potential differences in spike sorting criteria across labs. The ability to restrain the mouse for passive stimulation enables researchers to obtain additional information about their recordings that may ultimately aid in uncovering the function of cells and brain regions. Remarkably, during our headfixed experiments we found that even cells deep in the midbrain showed clear visual responses to drifting gratings (Figure 5D,E). This demonstrates the power of Neuropixels to uncover signals relevant to decision-making and other behaviors in uncharted brain territories.

Materials and methods

Key resources table.

Reagent type/resource	Designation	Source or reference	Identifiers	Additional information
Chemical compound/drug	Medical-grade clear silicon adhesive	Mastersil	912MED
Chemical compound/drug	Loctite Instant Adhesive 495	ULINE	S-7595
Chemical compound/drug	Medigel CPF	Clear H20	74-05-5022
Chemical compound/drug	Isoflurane	Allivet	50562
Chemical compound/drug	C and B Metabond 'B' Quick Base	Parkell	S398
Chemical compound/drug	C and B Metabond 'C' Quick Base	Parkell	S371
Chemical compound/drug	C and B Metabond Radiopaque L-Power	Parkell	S396
Chemical compound/drug	Optibond Solo Plus	Kerr	31514
Chemical compound/drug	Vetbond	Santa Cruz Biotechnology	sc-361931
Chemical compound/drug	Charisma A1 Syringe	Net32	66000085
Chemical compound/drug	Eye Ointment	Rugby	370435
Chemical compound/drug	Dental Cement	Stoelting	5217307
Chemical compound/drug	DiI	ThermoFisher Scientific	D282
Chemical compound/drug	Silicone Gel Kit	Dow Coning	3–4860.
Chemical compound/drug	Bleach	Amazon	B01K8HT54G	Any brand bleach ok
Other	Neuropixel Probe	Neuropixel Stock Center (Neuropixels.org)	Neuropixel 1.0 Probe
Other	3D Printed Internal Mount	‘this paper’ - Github repository	IM_Neuropixel1.stl	Internal mount design file (.stl) can be downloaded from the following github repository: https://github.com/churchlandlab/ChronicNeuropixels
Other	3D Printed External Casing	‘this paper’ - Github repository	EC_Neuropixel1.stl	same as above
Other	Sterotax Adapter	‘this paper’ - Github repository	stereotax adapter v4.ipt	same as above
Other	2-56A Screws	Amazon	B00F34U238
Other	Silver Wire	WPI	AGW1010
Other	4’ post holder with thumbscrew	Thorlabs	PH4
Other	Slim right angle bracket	Thorlabs	AB90B
Other	Aluminum Breadboard	Thorlabs	MB624
Other	M6 Cap Screw	Thorlabs	SH6MS20
Other	M6 Nut	Thorlabs	HW-KIT2/M
Other	Kapton Tape	ULINE	S-7595
Other	Kimwipes	Kimtech	34120
Other	Oxygen Cylinders	Airgas	OX USP300
Other	Mouse Anesthesia System with Isoflurance Box	Parkland Scientific	V3000PK
Other	Small rodent sterotax fitted with anesthesia mask	Narishige	SG-4N
Other	Dental Drill	Osada	EXL-M40
Other	0.9 mm burrs for micro drill	Fine Science Tools	19007–09
Other	T/Pump Warm Water Recirculator	Kent Scientific	TP-700
Other	Warming Pad for warm water recirculator	Kent Scientific	TPZ
Other	Cotton Applicators	Fisher Scientific	19-062-616
Other	Surgical Spears	Braintree Scientific Inc.	SP 40815

Open in a new tab

Printing and machining parts

To conduct this experiment, researchers will need Neuropixels probes. We recommend performing the entire process of preparing and implanting the probe using a dummy probe for practice. We printed and tested in VeroWhite material using a Stratasys Eden 260VS PolyJet 3D Printer with 16 µm resolution. The stereotax adaptor should be machined from aluminum or stainless steel. The parts featured here were designed for Neuropixels 3A probes, but we have since adapted these for Neuropixels 3B probes (Neuropixels 1.0). All designs can be found on the CSHL repository (http://repository.cshl.edu/36808/) as well as on Github (Juavinett et al., 2019; (copy archived at https://github.com/elifesciences-publications/ChronicNeuropixels).

The probe options for Neuropixels 3A differ based on their probe length (and corresponding site count), as well as whether they are active or passive electrodes. Probe options 1 and 3 are both passive, and contain 384 (5 mm long shank) and 960 sites (10 mm long shank), respectively. Probe options 2 and 4 are both active, and contain 384 (5 mm long shank) and 966 sites (10 mm long shank) respectively. All the options have the option to record from 374 channels, with the exception of Option 4, which only has 270 recording channels. Readers should refer to the Supplementary Information in Jun et al. (2017) for additional details. Neuropixels 3B probes have the Phase 3A Option three shank.

Mounting the probe

First, the internal mount is secured to the stereotax adapter (SA) using two 2-56A screws (Amazon, B00F34U238). As depicted in Figure 2A, we then attached the Neuropixels probe to the internal mount (IM) usingLoctite Instant Adhesive 495 (ULINE S-17190). Using a needle, we applied a medical-grade clear silicone adhesive, Mastersil 912MED, to the base of the shank (Figure 2B). The IM and probe was slid into the rails of the external casing (EC), and secured with cement (Figure 2D–F).

Surgical methods

All surgical and behavioral procedures conformed to the 316 guidelines established by the National Institutes of Health and were approved by the Institutional 317 Animal Care and Use Committee of Cold Spring Harbor Laboratory. We used male 3–4 month old C57/BL6 mice (Jackson Laboratories, 000664). Male mice were used because they are typically larger, and we expected that they would better handle the weight of the implant. Mice were given medicated (carprofen) food cups (MediGel CPF, Clear H20 74-05-5022) 1–2 days prior to surgery.

During surgery, the mouse was anesthetized with isoflurane. We cut away the skin and cleared any connective tissue. Tissue at the edges of the skull was glued down with Vetbond (Santa Cruz Biotechnology, cat. no. sc-361931). The skull was cleared and dried, using a skull scraper or blade to add additional texture. A boomerang shaped custom Titanium headbar was cemented to the skull, just posterior to the eyes, near Bregma. A burr hole was drilled for the ground screw, which was carefully screwed into the skull. We applied Optibond Solo Plus (Kerr, cat No. 31514) to the skull, and used UV light to cure it. We used Charisma (Net32, cat. No. 66000085) to create a base for the implant, and add additional support around the ground screw. Using a dental drill, a small craniotomy (1–2 mm) was made over visual cortex (2–2.5 ML,−3.4–3.5 AP relative to Bregma). The entire Neuropixels assembly (SA, IM, and EC) was placed in the stereotax. After carefully applying DiI (ThermoFisher Scientific, cat. no. D282, 0.5% in DMSO) to the probe, the shank was slowly lowered into the brain at a ~ 16 degree angle (Figure 2E). The ground wire is wrapped around the ground screw, and Metabond cement was carefully applied to attach the EC to the skull. The entire assembly was wrapped in Kapton Tape (ULINE S-7595) and the mouse was allowed to recover for 3–4 days. Mice were housed individually after surgery.

Once the mouse recovered, we removed the tape and added the headstage to the back of the implant. The entire assembly was re-wrapped with tape. On the next day, we began behavioral testing.

Behavioral data

To compare the behavior of implanted mice with naïve/unimplanted mice, we tracked mice using a Basler Pylon camera and Ethovision XT13 in a 16’ x 16’ open arena. For open field tests, naïve mice were allowed to explore a bare arena for 15 min. Implanted mice were tested in an arena with an inset nest; the data presented here are random excerpts of the mouse’s activity while outside of the nest. We excerpted the same length time segments from the naive mice for comparison. Our behavioral data were not normally distributed, so a Wilson Rank Sum Test was used to test for differences between naïve and implanted mice. We computed effect sizes using Cohen’s d.

Visual stimulation

For visually-evoked responses during freely moving behavior (Figure 4), a linearly expanding dot (40 cm/s) was presented on a monitor directly over the mouse’s head. This stimulus is known to elicit an escape response in mice (De Franceschi et al., 2016; Yilmaz and Meister, 2013). Unimplanted mice could escape into a small nest: a triangular prism with a 13 cm opening. Implanted mice could escape into a nest inset into the wall – this modification was necessary to enable mice to enter the enclosure with the implant. We found that being able to easily enter the nest increased the probability of flight (vs. freezing) responses. For visually evoked responses during head restraint (Figure 5), a set of full contrast, full field drifting gratings in eight different directions (10 repeats) were presented above the mouse’s head while the mouse was free to move on a wheel.

Electrophysiology data

Electrophysiology data was collected with SpikeGLX (Bill Karsh, https://github.com/billkarsh/SpikeGLX). The data were first median subtracted across channels and time (see Jun et al., 2017). Unless otherwise noted, experiments were first sorted with Kilosort spike sorting software (Pachitariu et al., 2016) and manually curated using phy (https://github.com/kwikteam/phy). Numbers of recorded neurons here may be more conservative than previously published reports because we were careful to exclude any units that exhibited drift or had evidence of being more than one neuron. Specific experiments (as noted in the text) were sorted with Kilosort2 for comparison (https://github.com/MouseLand/Kilosort2). Additional analyses and plotting with data were done with MATLAB code modified from N. Steinmetz (https://github.com/cortex-lab/spikes). To assess the quality of our recordings, we computed two metrics. First, we calculated the rate of spikes above the noise floor (‘event rate’). Events are temporally (<1 ms) and spatially (~50 µm radius) consistent events with amplitudes (on any site) that exceed six times the median absolute deviation (MAD, Jun et al., 2017). In addition, we computed the signal-to-noise ratio SNR for each event. As previously described, the event SNR is the ratio of peak amplitude of the site with largest amplitude (negative peak) in the event to 0.6745 × MAD (Jun et al., 2017).

Probe explantation

To explant the probe, we first anesthetized the mouse with isoflurane and loosely positioned the mouse into the earbars. The SA was placed in the stereotax and aligned with its slot in the IM. We carefully lowered the SA into the IM, and put the two screws back into place. It was important that the SA was properly aligned with the IM so that no unnecessary tension was placed on the implant. We carefully drilled away the cement at the boundary of the IM and EC, unraveled or cut the ground wire, and slowly raised the SA+IM+probe assembly. The mouse was perfused and the brain was fixed in 4% PFA for sectioning. We were able to find DiI signals in the brain even 1 month after implantation (we did not test later time points).

A detailed surgical protocol for mounting, implanting, and explanting the probe is located on Github (Juavinett et al., 2019; copy archived at https://github.com/elifesciences-publications/ChronicNeuropixels).

Acknowledgements

This work represents the collective input and knowledge of a burgeoning Neuropixels community (http://www.neuropixels.org).

We would like to acknowledge Tim Harris for his leadership on the development of the Neuropixels probes and his constant encouragement of this project.

The UCL Neuropixels course, taught by Nick Steinmetz, Matteo Carandini, Andrew Peters, Adam Kampff, was imperative in getting this project off the ground (http://www.ucl.ac.uk/neuropixels/courses). In particular, we would like to thank Nick Steinmetz for his critically important feedback, code, and upkeep of the Neuropixels Wiki page (https://github.com/cortex-lab/neuropixels/wiki).

We would also like to acknowledge Claudia Boehm and Albert Lee (Janelia Research Campus) for allowing us to observe their rat Neuropixels implant. Their protocol served as an important starting point for the protocol we developed in mice.

We have also benefitted from troubleshooting help from many individuals, including Wade Sun, James Jun, Marius Bauza, and Bill Karsh (SpikeGLX).

We are also grateful to the CSHL Undergraduate Research Program, which yearly provides a diverse group of students with funding and resources to complete invaluable research experiences at CSHL. This program funded GB for his initial summer in our lab.

We welcome feedback from the community regarding the diversity of methods used to implant and record with these probes.

Funding Statement

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Contributor Information

Anne K Churchland, Email: churchland@cshl.edu.

Laura L Colgin, University of Texas at Austin, United States.

Funding Information

This paper was supported by the following grants:

Simons Foundation Simons Collaboration on the Global Brain to Anne K Churchland.
Pew Charitable Trusts Pew Scholars to Anne K Churchland.
Eleanor Schwartz Fund Scholar Award to Anne K Churchland.
Cold Spring Harbor Laboratory Marie Robertson to Anne K Churchland.
National Science Foundation 1559816 to George Bekheet.

Additional information

Competing interests

No competing interests declared.

Author contributions

Conceptualization, Data curation, Software, Formal analysis, Supervision, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing.

Data curation, Software, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing.

Conceptualization, Resources, Supervision, Funding acquisition, Methodology, Writing—original draft, Project administration, Writing—review and editing.

Ethics

Animal experimentation: All surgical and behavioral procedures conformed to the guidelines established by the National Institutes of Health and were approved by the Institutional Animal Care and Use Committee of Cold Spring Harbor Laboratory (protocol # 16-13-10-7). All surgery was performed under isoflurane anesthesia and every effort was made to minimize suffering.

Additional files

Transparent reporting form

elife-47188-transrepform.pdf^{(345.1KB, pdf)}

DOI: 10.7554/eLife.47188.014

Data availability

We have made all the materials related to this device available to the community via GitHub (https://github.com/churchlandlab/ChronicNeuropixels; copy archived at https://github.com/elifesciences-publications/ChronicNeuropixels). The technical drawings, the methodological instructions, the photographs and supporting code will, together, allow any researcher to rapidly adopt this new technology and begin to benefit from Neuropixels probes. Data from the electrophysiological recordings available is available here: http://churchlandlab.labsites.cshl.edu/code/.

The following dataset was generated:

Ashley L Juavinett, George Bekheet, Anne K Churchland. 2019. Chronically-implanted Neuropixels probes enable high yield recordings in freely moving mice: dataset. CSHL Institutional Repository. 38304

References

Bolkan SS, Stujenske JM, Parnaudeau S, Spellman TJ, Rauffenbart C, Abbas AI, Harris AZ, Gordon JA, Kellendonk C. Thalamic projections sustain prefrontal activity during working memory maintenance. Nature Neuroscience. 2017;20:987–996. doi: 10.1038/nn.4568. [DOI] [PMC free article] [PubMed] [Google Scholar]
De Franceschi G, Vivattanasarn T, Saleem AB, Solomon SG. Vision guides selection of freeze or flight defense strategies in mice. Current Biology. 2016;26:2150–2154. doi: 10.1016/j.cub.2016.06.006. [DOI] [PubMed] [Google Scholar]
Evans DA, Stempel AV, Vale R, Ruehle S, Lefler Y, Branco T. A synaptic threshold mechanism for computing escape decisions. Nature. 2018;558:590–594. doi: 10.1038/s41586-018-0244-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
Gomez-Marin A, Paton JJ, Kampff AR, Costa RM, Mainen ZF. Big behavioral data: psychology, ethology and the foundations of neuroscience. Nature Neuroscience. 2014;17:1455–1462. doi: 10.1038/nn.3812. [DOI] [PubMed] [Google Scholar]
Hafting T, Fyhn M, Molden S, Moser MB, Moser EI. Microstructure of a spatial map in the entorhinal cortex. Nature. 2005;436:801–806. doi: 10.1038/nature03721. [DOI] [PubMed] [Google Scholar]
Jaeyoon JJ, Mitelut C, Lai C, Gratiy S, Anastassiou C, Harris TD. Real-time spike sorting platform for high-density extracellular probes with ground-truth validation and drift correction. bioRxiv. 2017 doi: 10.1101/101030. [DOI]
Juavinett AL, Erlich JC, Churchland AK. Decision-making behaviors: weighing ethology, complexity, and sensorimotor compatibility. Current Opinion in Neurobiology. 2018;49:42–50. doi: 10.1016/j.conb.2017.11.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
Juavinett AL, Bekheet GB, Churchland AK. ChronicNeuropixels. Github. 2019 https://github.com/churchlandlab/ChronicNeuropixels
Jun JJ, Steinmetz NA, Siegle JH, Denman DJ, Bauza M, Barbarits B, Lee AK, Anastassiou CA, Andrei A, Aydın Çağatay, Barbic M, Blanche TJ, Bonin V, Couto J, Dutta B, Gratiy SL, Gutnisky DA, Häusser M, Karsh B, Ledochowitsch P, Lopez CM, Mitelut C, Musa S, Okun M, Pachitariu M, Putzeys J, Rich PD, Rossant C, Sun W-lung, Svoboda K, Carandini M, Harris KD, Koch C, O’Keefe J, Harris TD. Fully integrated silicon probes for high-density recording of neural activity. Nature. 2017;551:232–236. doi: 10.1038/nature24636. [DOI] [PMC free article] [PubMed] [Google Scholar]
Kerekes P, Daret A, Shulz DE, Ego-Stengel V. Bilateral discrimination of tactile patterns without whisking in freely running rats. The Journal of Neuroscience. 2017;37:7567–7579. doi: 10.1523/JNEUROSCI.0528-17.2017. [DOI] [PMC free article] [PubMed] [Google Scholar]
Krupic J, Bauza M, Burton S, O'Keefe J. Local transformations of the hippocampal cognitive map. Science. 2018;359:1143–1146. doi: 10.1126/science.aao4960. [DOI] [PMC free article] [PubMed] [Google Scholar]
Lottem E, Banerjee D, Vertechi P, Sarra D, Lohuis MO, Mainen ZF. Activation of serotonin neurons promotes active persistence in a probabilistic foraging task. Nature Communications. 2018;9:1000. doi: 10.1038/s41467-018-03438-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
Markowitz JE, Gillis WF, Beron CC, Neufeld SQ, Robertson K, Bhagat ND, Peterson RE, Peterson E, Hyun M, Linderman SW, Sabatini BL, Datta SR. The striatum organizes 3D behavior via Moment-to-Moment action selection. Cell. 2018;174:44–58. doi: 10.1016/j.cell.2018.04.019. [DOI] [PMC free article] [PubMed] [Google Scholar]
Marlin BJ, Mitre M, D'amour JA, Chao MV, Froemke RC. Oxytocin enables maternal behaviour by balancing cortical inhibition. Nature. 2015;520:499–504. doi: 10.1038/nature14402. [DOI] [PMC free article] [PubMed] [Google Scholar]
O'Keefe J, Dostrovsky J. The Hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Research. 1971;34:171–175. doi: 10.1016/0006-8993(71)90358-1. [DOI] [PubMed] [Google Scholar]
Okun M, Lak A, Carandini M, Harris KD. Long term recordings with immobile silicon probes in the mouse cortex. PLOS ONE. 2016;11:e0151180. doi: 10.1371/journal.pone.0151180. [DOI] [PMC free article] [PubMed] [Google Scholar]
Pachitariu M, Steinmetz NA, Kadir SN, Carandini M, Harris KD. Fast and accurate spike sorting of high-channel count probes with KiloSort. Advances in Neural Information Processing Systems. 2016;29:4448–4456. [Google Scholar]
Steinmetz NA, Koch C, Harris KD, Carandini M. Challenges and opportunities for large-scale electrophysiology with neuropixels probes. Current Opinion in Neurobiology. 2018;50:92–100. doi: 10.1016/j.conb.2018.01.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
Stujenske JM, Likhtik E, Topiwala MA, Gordon JA. Fear and safety engage competing patterns of theta-gamma coupling in the basolateral amygdala. Neuron. 2014;83:919–933. doi: 10.1016/j.neuron.2014.07.026. [DOI] [PMC free article] [PubMed] [Google Scholar]
Vandecasteele M, M S, Royer S, Belluscio M, Berényi A, Diba K, Fujisawa S, Grosmark A, Mao D, Mizuseki K, Patel J, Stark E, Sullivan D, Watson B, Buzsáki G. Large-scale recording of neurons by movable silicon probes in behaving rodents. Journal of Visualized Experiments. 2012:e3568. doi: 10.3791/3568. [DOI] [PMC free article] [PubMed] [Google Scholar]
Vélez-Fort M, Bracey EF, Keshavarzi S, Rousseau CV, Cossell L, Lenzi SC, Strom M, Margrie TW. A circuit for integration of head- and Visual-Motion signals in layer 6 of mouse primary visual cortex. Neuron. 2018;98:179–191. doi: 10.1016/j.neuron.2018.02.023. [DOI] [PMC free article] [PubMed] [Google Scholar]
Voigts J, Siegle JH, Pritchett DL, Moore CI. The flexDrive: an ultra-light implant for optical control and highly parallel chronic recording of neuronal ensembles in freely moving mice. Frontiers in Systems Neuroscience. 2013;7:8. doi: 10.3389/fnsys.2013.00008. [DOI] [PMC free article] [PubMed] [Google Scholar]
Whitlock JR, Pfuhl G, Dagslott N, Moser MB, Moser EI. Functional split between parietal and entorhinal cortices in the rat. Neuron. 2012;73:789–802. doi: 10.1016/j.neuron.2011.12.028. [DOI] [PubMed] [Google Scholar]
Wimmer RD, Schmitt LI, Davidson TJ, Nakajima M, Deisseroth K, Halassa MM. Thalamic control of sensory selection in divided attention. Nature. 2015;526:705–709. doi: 10.1038/nature15398. [DOI] [PMC free article] [PubMed] [Google Scholar]
Yilmaz M, Meister M. Rapid innate defensive responses of mice to looming visual stimuli. Current Biology. 2013;23:2011–2015. doi: 10.1016/j.cub.2013.08.015. [DOI] [PMC free article] [PubMed] [Google Scholar]

eLife. doi: 10.7554/eLife.47188.019

Decision letter

Editor: Laura L Colgin¹

Reviewed by: Nick Steinmetz²

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Chronically-implanted Neuropixels probes enable high yield recordings in freely moving mice" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by Laura Colgin as the Senior and Reviewing Editor. The following individual involved in review of your submission has agreed to reveal their identity: Nick Steinmetz (Reviewer #1).

The reviewers have discussed the reviews with one another, and the Senior Editor has consolidated their comments and discussion into this decision letter to help you prepare a revised submission.

Summary:

The authors have presented a method and specifications for achieving chronic Neuropixels recordings in freely-moving mice and, importantly, for recovering the probes afterwards. Due to the financial cost and limited availability of Neuropixels probes, chronic implantation of these probes is currently challenging (i.e., researchers may not want to "waste" a probe by chronically implanting it). This paper presents a potential solution to this problem. Many labs have ordered Neuropixels probes since they were made available, so reviewers agreed that these methods provide a useful resource for many in the community. The proposed system consists of two main components: 1) an internal mount permanently fused to the probe base and 2) a removable casing (Apparatus to Mount Individual Electronics: AMIE) that adapts the probe and internal mount to a chronic implant. In addition to carrying the probe, the AMIE protects and holds the flex-cable and head stage. Reviewers found the paper to be clear and appreciated the authors' honesty about potential pitfalls (e.g., broken probes, early attempts with low yields). Reviewers appreciated that the authors made their surgical protocol and 3D printer files publicly available. However, a number of major concerns were raised that need to be addressed before the paper can be deemed suitable for publication.

Essential revisions:

1) Reviewers felt that the manuscript misses key data showing that the Neuropixel can be re-used without a significant drop in signal quality. Specifically, in Table 1 average numbers of isolated units are missing for NP13, 15 and 16 (i.e. the cases in which Neuropixels were implanted and then successfully explanted and re-implanted in a second mouse). These numbers are key (particularly NP16), as the authors claim that the "quality of the recording did not noticeably change in the second mouse", but do not provide any quantitative data to support their claim. Instead, they show 5 example units from NP14 and 16, which only demonstrates they could find at least 5 units in both recordings (Figure 6C). Additional quantification would also be bolster the manuscript's reusability claims. It would be useful to compare noise levels on the first implantation to that on the second for a given probe. This can be computed straightforwardly e.g. with this function (https://github.com/cortex-lab/spikes/blob/master/analysis/computeRawRMS.m). The authors should also run kilosort2 on all recordings (https://github.com/MouseLand/Kilosort2), which automatically assigns 'good' and 'mua' labels. This is not necessarily perfectly accurate, but at least it would provide a rapid and unbiased analysis that would be comparable to yields reported in other recordings (e.g. comparable to this dataset https://janelia.figshare.com/articles/Eight-probe_Neuropixels_recordings_during_spontaneous_behaviors/7739750). Alternatively, the authors could repeat the manual process that they did for some recordings already for at least one recording from each of the other mice in Table 1. The same argument applies to the comparison between restrained and unrestrained, which needs similar automatic (i.e., KS2) or manual quantification or both. The reusability of the probes is critical for the paper's significance, because (as the authors note) techniques without explanation already exist to chronically implant Neuropixels in freely moving mice and re-use (Introduction, fifth paragraph, Okun, 2016).

2) A further concern is how the authors demonstrate that behavior in implanted vs. naïve mice is comparable. Quantitatively, they do this by reporting non-significant p-values on rank-sum tests of differences between behavioral measures in implanted vs. naïve mice. However, using non-significant p-values to argue for the null-hypothesis is not statistically valid; failing to reject the null hypothesis is not evidence for the null hypothesis. It's just as likely that the authors didn't use enough mice to have sufficient statistical power to show the effect. Indeed, visual inspection of Figure 3 makes this appear highly likely – clear changes in several activities can be seen in Figure 3 (most notably for % time moving, max acceleration and mean escape velocity). For instance, time moving declines from about 40% to 30%, and it's hard to believe that with additional samples this would remain non-significant, especially given that p = .0502. Relatedly, the authors claim that the correct critical value should be p = .0167 due to multiple comparisons. The meaning of a multiple-comparisons correction is difficult to interpret when the researcher's goal is to show that a difference isn't significant. Then, one could perform rank-sum tests on an arbitrarily large number of behavioral measures to reduce the critical value, and find no differences whatsoever between populations. A simple and more interpretable approach would be to simply report the effect sizes instead of p-values. Though there is a small reduction in exploratory behavior/ escape velocity, this is relatively small and essentially unavoidable due to the weight of any probe/casing being placed on the mouse's head.

3) Currently shipping probes do not have the same geometry as the probes described in this manuscript. That is, the probes that are now on sale at neuropixels.org have a slightly different shape than the "Phase 3A" probes employed in this paper. The authors need to describe how their system could be adapted to other probe geometries (and future designs will almost certainly evolve in this regard). It would be extremely valuable if the authors could provide an updated design based on the changes to the probes. Even though it may only be subtly different, it will save others from having to replicate the work.

4) Reviewers had trouble visualizing the exact shape of the pieces and how they fit together. One suggestion was that the authors could make a short video that shows someone putting the pieces together by hand with a dummy probe. Some sort of additional visual aid such as this would greatly help readers to understand what screws onto what, how things slide, etc. Another suggestion was that it could perhaps somehow be done with the 3D renderings from Figure 1.

5) A concern was raised about whether the authors' system for recovering the probe might compromise the stability of the probe, since most of the body of the probe itself is not really directly attached to the skull, but only via a thin 3D printed piece. A potential way to test this that was suggested could be to plot a comparison of waveforms like in Figure 4F, but for the same neuron recorded during periods of high acceleration versus stationary periods. An even better test would be to compute a metric of waveform similarity between moving and stationary periods, which could be compared to a recording in which the probe was attached in a fixed way. Perhaps such a recording would be available from Evans et al., who may have performed chronic recordings in freely moving animals with a probe fixed in place in the manner of Okun et al., 2016. Other reasonable ideas about how to address this concern would also be acceptable.

6) Some discussion of why this technology would be useful even if probes were less expensive and more accessible (or how/whether the approach would be modified under such conditions) would help extend the impact of this paper, assuming that limitations of the neuropixels purchasing scheme will be alleviated in the future.

eLife. 2019 Aug 14;8:e47188. doi: 10.7554/eLife.47188.020

Author response

Essential revisions:

1) Reviewers felt that the manuscript misses key data showing that the Neuropixel can be re-used without a significant drop in signal quality. Specifically, in Table 1 average numbers of isolated units are missing for NP13, 15 and 16 (i.e. the cases in which Neuropixels were implanted and then successfully explanted and re-implanted in a second mouse). These numbers are key (particularly NP16), as the authors claim that the "quality of the recording did not noticeably change in the second mouse", but do not provide any quantitative data to support their claim. Instead, they show 5 example units from NP14 and 16, which only demonstrates they could find at least 5 units in both recordings (Figure 6C). Additional quantification would also be bolster the manuscript's reusability claims.

We appreciate this concern from the reviewers. We have added the number of isolated units for our re-implanted probe (NP16/Mouse #7) as well as an additional mouse (NP15/Mouse# 6) to Figure 4C, and have added the average number of isolated units to Table 1. Although the probes in NP13 and NP15 were successfully explanted, they were not implanted in a second mouse.

It would be useful to compare noise levels on the first implantation to that on the second for a given probe. This can be computed straightforwardly e.g. with this function (https://github.com/cortex-lab/spikes/blob/master/analysis/computeRawRMS.m).

We wholeheartedly agree. We have added new panels to Figure 6 that address the concern about noise levels on the probe. These panels (6D and 6E) show the detected event rate and SNR for the re-implanted probe (using the recommended code and following the example in Jun et al., 2017). Although there is a drop in detected event rate and a slight drop in SNR in the re-implanted probe, we note that it is still possible to isolate many units (43.6 ± 17.8). Two points should be considered when evaluating this yield. These are described below.

First, variability across yields in individual animals even for a new probe (Figure 4C) leaves open the possibility that yields on re-implanted probes might sometimes be higher than what we observed. To test this, we sorted an additional, previously unsorted dataset from a new mouse (NP13) which had a probe that was re-implanted after being in NP12. We had not originally included NP13 in this dataset because of issues with the grounding wire and screw after surgery, enabling us to only record one session. We are pleased to report that sorting of this session yielded 145 well-isolated, ‘good’ units. This number is now included on Figure 4C, highlighting that high yields are possible even for re-implanted probes.

Second, yields of ~40 neurons, while not ideal, are likely acceptable to many researchers because Neuropixels probes are still available in very limited numbers. The recent release of 3B probes from IMEC was a first step towards increased availability, but even many well-established systems neuroscience labs were only granted 10/year (indeed, the last author of this paper was provided with that number). This is a small number even for a single project, especially one that involves a trainee who will have to learn to use the probes and may even break a few. For such labs, which are probably the vast majority in Neuroscience, a yield of 43 simultaneously recorded neurons for a re-implanted probe has tremendous value.

The corresponding results text can be found in the subsection “Printing and machining parts”. We thank the reviewer for pushing us to include additional data on yields and feel that this has greatly strengthened the paper.

The authors should also run kilosort2 on all recordings (https://github.com/MouseLand/Kilosort2), which automatically assigns 'good' and 'mua' labels. This is not necessarily perfectly accurate, but at least it would provide a rapid and unbiased analysis that would be comparable to yields reported in other recordings (e.g. comparable to this dataset https://janelia.figshare.com/articles/Eight-probe_Neuropixels_recordings_during_spontaneous_behaviors/7739750). Alternatively, the authors could repeat the manual process that they did for some recordings already for at least one recording from each of the other mice in Table 1. The same argument applies to the comparison between restrained and unrestrained, which needs similar automatic (i.e., KS2) or manual quantification or both. The reusability of the probes is critical for the paper's significance, because (as the authors note) techniques without explanation already exist to chronically implant Neuropixels in freely moving mice and re-use (Introduction, fifth paragraph, Okun, 2016).

We agree with the reviewer that the automatic assignment of ‘good’ and ‘mua’ labels could be useful in comparing of neuron yields. To evaluate whether Kilosort2 differed from our approach, we implemented it for two separate analysis. First, we implemented Kilosort2 on the first experiment with the re-implanted probe in NP16. Neuron yields with Kilosort2 were similar to our original sorting method, which was reassuring (77 for

Kilosort2 compared to 69 with our original sorting). The slightly higher yield for KS2 may reflect the conservativeness of our original approach. Second, we implemented Kilosort2 during a head-restrained experiment and again, our original yields were similar, if slightly more conservative (63 for Kilosort2 compared to 60 for our original approach). The similar neuron yields for our original approach and KS2 suggest that our criteria for spike sorting were effective: robust units with clear waveforms, reasonable interspike intervals, and minimal drift in depth. The similarity of our original approach to Kilosort2 made us disinclined to re-analyze all the datasets used for the comparisons in the paper. Nevertheless, we include the numbers above in the revised manuscript (subsection “Researchers can also conduct headfixed recordings to further characterize neurons”) as readers may find these observations useful in evaluating spike sorting methods.

2) A further concern is how the authors demonstrate that behavior in implanted vs. naïve mice is comparable. Quantitatively, they do this by reporting non-significant p-values on rank-sum tests of differences between behavioral measures in implanted vs. naïve mice. However, using non-significant p-values to argue for the null-hypothesis is not statistically valid; failing to reject the null hypothesis is not evidence for the null hypothesis. It's just as likely that the authors didn't use enough mice to have sufficient statistical power to show the effect. Indeed, visual inspection of Figure 3 makes this appear highly likely – clear changes in several activities can be seen in Figure 3 (most notably for % time moving, max acceleration and mean escape velocity). For instance, time moving declines from about 40% to 30%, and it's hard to believe that with additional samples this would remain non-significant, especially given that p = .0502. Relatedly, the authors claim that the correct critical value should be p = .0167 due to multiple comparisons. The meaning of a multiple-comparisons correction is difficult to interpret when the researcher's goal is to show that a difference isn't significant. Then, one could perform rank-sum tests on an arbitrarily large number of behavioral measures to reduce the critical value, and find no differences whatsoever between populations. A simple and more interpretable approach would be to simply report the effect sizes instead of p-values. Though there is a small reduction in exploratory behavior/ escape velocity, this is relatively small and essentially unavoidable due to the weight of any probe/casing being placed on the mouse's head.

We have re-written this section and amended our methods to no longer apply corrected p-value criteria. Ultimately, this does not change our claim because none of our comparisons achieved p<0.05, better yet the corrected p<0.0167 criteria. Text in the Results, Materials and methods, and Figure 3 legend has been changed to reflect this. Further, the overall language is toned down: the key conclusion is that implanted mice remain active and show species-typical behavior. Small and idiosyncratic changes in speed are in keeping with this conclusion. The new text in the Results section reads:

“To evaluate the suitability of the AMIE for use during behavior, we assessed the impact of the device on both spontaneous and stimulus-driven movements. […] Taken together, these behavioral observations argue that although the presence of the AMIE may have idiosyncratically slowed mice slightly, they remained active in an open arena and showed species-typical responses to threatening stimuli.”

We appreciate the reviewer’s recommendation to report effect sizes for these comparisons. A Cohen’s d for each comparison is also now reported on the figure.

3) Currently shipping probes do not have the same geometry as the probes described in this manuscript. That is, the probes that are now on sale at neuropixels.org have a slightly different shape than the "Phase 3A" probes employed in this paper. The authors need to describe how their system could be adapted to other probe geometries (and future designs will almost certainly evolve in this regard). It would be extremely valuable if the authors could provide an updated design based on the changes to the probes. Even though it may only be subtly different, it will save others from having to replicate the work.

We have modified our design to fit the Neuropixels 1.0 probes, and these designs are now available on our GitHub resource (https://github.com/churchlandlab/ChronicNeuropixels). We are working with collaborators at CSHL and UC San Diego to test these new designs with the Neuropixels 1.0 probes and will continue to update the available designs as needed, including making them available in Autodesk, which is freely available to students. We have also made corresponding notes in the text.

4) Reviewers had trouble visualizing the exact shape of the pieces and how they fit together. One suggestion was that the authors could make a short video that shows someone putting the pieces together by hand with a dummy probe. Some sort of additional visual aid such as this would greatly help readers to understand what screws onto what, how things slide, etc. Another suggestion was that it could perhaps somehow be done with the 3D renderings from Figure 1.

We agree and have generated a video (Video 1) which shows a 3D rendering.

5) A concern was raised about whether the authors' system for recovering the probe might compromise the stability of the probe, since most of the body of the probe itself is not really directly attached to the skull, but only via a thin 3D printed piece. A potential way to test this that was suggested could be to plot a comparison of waveforms like in Figure 4F, but for the same neuron recorded during periods of high acceleration versus stationary periods. An even better test would be to compute a metric of waveform similarity between moving and stationary periods, which could be compared to a recording in which the probe was attached in a fixed way. Perhaps such a recording would be available from Evans et al., who may have performed chronic recordings in freely moving animals with a probe fixed in place in the manner of Okun et al., 2016. Other reasonable ideas about how to address this concern would also be acceptable.

We agree that this is an important concern and have taken the reviewer’s suggestion to compare waveforms during moving versus stationary periods. For a session from Mouse

#7 (implanted with a recycled probe), we defined periods where the mouse was moving

(>15 cm/sec) or stationary (<5 cm/sec). A velocity of 15 cm/s is quite fast, while also common enough to enable the extraction of enough waveforms to compute a mean. We computed mean waveforms from 50 different spikes from 9 different units. As Author response image 1 shows, these waveforms are almost identical regardless of whether the mouse is moving or stationary. We computed correlation coefficients for moving vs. stationary mean waveforms, finding that all waveforms were very highly correlated (r > 0.96, p <.00001).

Several of these units show notable changes in absolute voltage, which could be for a number of reasons. For one, there are slow oscillations in the absolute voltage of the signal over time, so extracting a limited number of waveforms from different periods of the recording may reflect these slow changes in voltage. Second, it could be that the probe does slightly move relative to the recorded neuron during movement periods, leading to a change in the extracellularly recorded voltage. Additional analyses across multiple mice and probe depths would be needed to identify the cause of this shift. Regardless, it is clear that the waveforms recorded during movement and stationary periods are almost identical, providing further evidence that these recordings using our device are stable and that it is possible to isolate the same unit even in periods of fast movement. During manual spike sorting, we were also able to observe whether the waveforms changed over the course of the experiment, and found them to be quite stable.

Although we feel this was an important check, a full treatment of this topic warrants many additional analyses (e.g., multiple mice, different parameters to define movement periods) that are beyond the scope of this paper. We are however open to suggestions to include a reference in the manuscript to the preliminary analyses we have conducted if the reviewers or editors think this would be useful.

6) Some discussion of why this technology would be useful even if probes were less expensive and more accessible (or how/whether the approach would be modified under such conditions) would help extend the impact of this paper, assuming that limitations of the neuropixels purchasing scheme will be alleviated in the future.

We agree. In the revised Introduction, we highlight 3 advantages of our design. First, the AMIE can be readily adapted to different types of silicon probes, extending its significance beyond Neuropixels probes. Second, the AMIE is lighter than alternative designs because of the limited use of acrylic. Finally, by providing the drawings, materials and instructions, our device can be implemented not only by labs with years of expertise in electrophysiological recordings, but also by labs with different expertise that discover a new need to study neural activity during behavior. Existing methods for implanting electrodes are non-standardized, and rely on designs and surgical protocols that are part of lab lore and are not publicly available. This leaves new researchers in the dark about how to implant probes securely. Our thorough protocols make the AMIE within reach of these labs. Even if such labs are amongst those rare researchers without the need to re-use probes, they will still benefit from the use of standardized, open source hardware that this manuscript provides.

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Citations

Ashley L Juavinett, George Bekheet, Anne K Churchland. 2019. Chronically-implanted Neuropixels probes enable high yield recordings in freely moving mice: dataset. CSHL Institutional Repository. 38304 [DOI] [PMC free article] [PubMed]

Supplementary Materials

Figure 3—source data 1. Behavioral data for implanted and naive mice in an open field and in response to a looming stimulus.

elife-47188-fig3-data1.xlsx^{(12KB, xlsx)}

DOI: 10.7554/eLife.47188.007

Figure 4—source data 1. Number of isolated units for each probe across post-surgery days.

elife-47188-fig4-data1.xlsx^{(10.3KB, xlsx)}

DOI: 10.7554/eLife.47188.009

Figure 6—source data 1. Computed event rate and SNR ratio for the initial implant and re-implant of the same probe across post-surgery days.

elife-47188-fig6-data1.xlsx^{(8.5KB, xlsx)}

DOI: 10.7554/eLife.47188.012

Transparent reporting form

elife-47188-transrepform.pdf^{(345.1KB, pdf)}

DOI: 10.7554/eLife.47188.014

Data Availability Statement

The following dataset was generated:

Ashley L Juavinett, George Bekheet, Anne K Churchland. 2019. Chronically-implanted Neuropixels probes enable high yield recordings in freely moving mice: dataset. CSHL Institutional Repository. 38304

[bib1] Bolkan SS, Stujenske JM, Parnaudeau S, Spellman TJ, Rauffenbart C, Abbas AI, Harris AZ, Gordon JA, Kellendonk C. Thalamic projections sustain prefrontal activity during working memory maintenance. Nature Neuroscience. 2017;20:987–996. doi: 10.1038/nn.4568. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib2] De Franceschi G, Vivattanasarn T, Saleem AB, Solomon SG. Vision guides selection of freeze or flight defense strategies in mice. Current Biology. 2016;26:2150–2154. doi: 10.1016/j.cub.2016.06.006. [DOI] [PubMed] [Google Scholar]

[bib3] Evans DA, Stempel AV, Vale R, Ruehle S, Lefler Y, Branco T. A synaptic threshold mechanism for computing escape decisions. Nature. 2018;558:590–594. doi: 10.1038/s41586-018-0244-6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib4] Gomez-Marin A, Paton JJ, Kampff AR, Costa RM, Mainen ZF. Big behavioral data: psychology, ethology and the foundations of neuroscience. Nature Neuroscience. 2014;17:1455–1462. doi: 10.1038/nn.3812. [DOI] [PubMed] [Google Scholar]

[bib5] Hafting T, Fyhn M, Molden S, Moser MB, Moser EI. Microstructure of a spatial map in the entorhinal cortex. Nature. 2005;436:801–806. doi: 10.1038/nature03721. [DOI] [PubMed] [Google Scholar]

[bib6] Jaeyoon JJ, Mitelut C, Lai C, Gratiy S, Anastassiou C, Harris TD. Real-time spike sorting platform for high-density extracellular probes with ground-truth validation and drift correction. bioRxiv. 2017 doi: 10.1101/101030. [DOI]

[bib7] Juavinett AL, Erlich JC, Churchland AK. Decision-making behaviors: weighing ethology, complexity, and sensorimotor compatibility. Current Opinion in Neurobiology. 2018;49:42–50. doi: 10.1016/j.conb.2017.11.001. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib8] Juavinett AL, Bekheet GB, Churchland AK. ChronicNeuropixels. Github. 2019 https://github.com/churchlandlab/ChronicNeuropixels

[bib9] Jun JJ, Steinmetz NA, Siegle JH, Denman DJ, Bauza M, Barbarits B, Lee AK, Anastassiou CA, Andrei A, Aydın Çağatay, Barbic M, Blanche TJ, Bonin V, Couto J, Dutta B, Gratiy SL, Gutnisky DA, Häusser M, Karsh B, Ledochowitsch P, Lopez CM, Mitelut C, Musa S, Okun M, Pachitariu M, Putzeys J, Rich PD, Rossant C, Sun W-lung, Svoboda K, Carandini M, Harris KD, Koch C, O’Keefe J, Harris TD. Fully integrated silicon probes for high-density recording of neural activity. Nature. 2017;551:232–236. doi: 10.1038/nature24636. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib10] Kerekes P, Daret A, Shulz DE, Ego-Stengel V. Bilateral discrimination of tactile patterns without whisking in freely running rats. The Journal of Neuroscience. 2017;37:7567–7579. doi: 10.1523/JNEUROSCI.0528-17.2017. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib11] Krupic J, Bauza M, Burton S, O'Keefe J. Local transformations of the hippocampal cognitive map. Science. 2018;359:1143–1146. doi: 10.1126/science.aao4960. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib12] Lottem E, Banerjee D, Vertechi P, Sarra D, Lohuis MO, Mainen ZF. Activation of serotonin neurons promotes active persistence in a probabilistic foraging task. Nature Communications. 2018;9:1000. doi: 10.1038/s41467-018-03438-y. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib13] Markowitz JE, Gillis WF, Beron CC, Neufeld SQ, Robertson K, Bhagat ND, Peterson RE, Peterson E, Hyun M, Linderman SW, Sabatini BL, Datta SR. The striatum organizes 3D behavior via Moment-to-Moment action selection. Cell. 2018;174:44–58. doi: 10.1016/j.cell.2018.04.019. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib14] Marlin BJ, Mitre M, D'amour JA, Chao MV, Froemke RC. Oxytocin enables maternal behaviour by balancing cortical inhibition. Nature. 2015;520:499–504. doi: 10.1038/nature14402. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib15] O'Keefe J, Dostrovsky J. The Hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Research. 1971;34:171–175. doi: 10.1016/0006-8993(71)90358-1. [DOI] [PubMed] [Google Scholar]

[bib16] Okun M, Lak A, Carandini M, Harris KD. Long term recordings with immobile silicon probes in the mouse cortex. PLOS ONE. 2016;11:e0151180. doi: 10.1371/journal.pone.0151180. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib17] Pachitariu M, Steinmetz NA, Kadir SN, Carandini M, Harris KD. Fast and accurate spike sorting of high-channel count probes with KiloSort. Advances in Neural Information Processing Systems. 2016;29:4448–4456. [Google Scholar]

[bib18] Steinmetz NA, Koch C, Harris KD, Carandini M. Challenges and opportunities for large-scale electrophysiology with neuropixels probes. Current Opinion in Neurobiology. 2018;50:92–100. doi: 10.1016/j.conb.2018.01.009. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib19] Stujenske JM, Likhtik E, Topiwala MA, Gordon JA. Fear and safety engage competing patterns of theta-gamma coupling in the basolateral amygdala. Neuron. 2014;83:919–933. doi: 10.1016/j.neuron.2014.07.026. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib20] Vandecasteele M, M S, Royer S, Belluscio M, Berényi A, Diba K, Fujisawa S, Grosmark A, Mao D, Mizuseki K, Patel J, Stark E, Sullivan D, Watson B, Buzsáki G. Large-scale recording of neurons by movable silicon probes in behaving rodents. Journal of Visualized Experiments. 2012:e3568. doi: 10.3791/3568. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib21] Vélez-Fort M, Bracey EF, Keshavarzi S, Rousseau CV, Cossell L, Lenzi SC, Strom M, Margrie TW. A circuit for integration of head- and Visual-Motion signals in layer 6 of mouse primary visual cortex. Neuron. 2018;98:179–191. doi: 10.1016/j.neuron.2018.02.023. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib22] Voigts J, Siegle JH, Pritchett DL, Moore CI. The flexDrive: an ultra-light implant for optical control and highly parallel chronic recording of neuronal ensembles in freely moving mice. Frontiers in Systems Neuroscience. 2013;7:8. doi: 10.3389/fnsys.2013.00008. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib23] Whitlock JR, Pfuhl G, Dagslott N, Moser MB, Moser EI. Functional split between parietal and entorhinal cortices in the rat. Neuron. 2012;73:789–802. doi: 10.1016/j.neuron.2011.12.028. [DOI] [PubMed] [Google Scholar]

[bib24] Wimmer RD, Schmitt LI, Davidson TJ, Nakajima M, Deisseroth K, Halassa MM. Thalamic control of sensory selection in divided attention. Nature. 2015;526:705–709. doi: 10.1038/nature15398. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib25] Yilmaz M, Meister M. Rapid innate defensive responses of mice to looming visual stimuli. Current Biology. 2013;23:2011–2015. doi: 10.1016/j.cub.2013.08.015. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Chronically implanted Neuropixels probes enable high-yield recordings in freely moving mice

Ashley L Juavinett

George Bekheet

Anne K Churchland

Roles

Abstract

Introduction

Results

Design overview

Figure 1. Schematic of Neuropixels AMIE.

Video 1. 3D rendering of the AMIE device demonstrating the configuration of internal mount (IM), external casing (EC), and stereotax adapter.

Protocol overview

Figure 2. Mounting and implanting the Neuropixels probe.

Mice are mobile with the implant

Video 2. Behavior of mouse implanted with Neuropixels AMIE.

Figure 3. Behavior in implanted mice is comparable to naive mice.

Neuropixels AMIE allows for 60–100 simultaneously recorded neurons across weeks of freely moving behavior

Figure 4. Chronic Neuropixel probe implants in cortex and subcortical regions can record ~20–145 units across multiple days.

Researchers can also conduct headfixed recordings to further characterize neurons

Figure 5. Implant design enables researchers to further characterize brain regions recorded during freely moving behavior.

Implant allows researchers to recover the probe after the experiment

Figure 6. AMIE design allows for probe explantation and subsequent re-implantation with the same Neuropixels probe.

Table 1. Overview of experiments, with the Neuropixels probe option used and the outcome of the experiment.

Discussion

Materials and methods

Key resources table.

Printing and machining parts

Mounting the probe

Surgical methods

Behavioral data

Visual stimulation

Electrophysiology data

Probe explantation

Acknowledgements

Funding Statement

Contributor Information

Funding Information

Additional information

Competing interests

Author contributions

Ethics

Additional files

Data availability

References

Decision letter

Roles

Author response

Author response image 1. Comparison of mean waveforms (n=50) across periods where the mouse was moving (>15 cm/sec) or stationary (<5 cm/sec).

Associated Data

Data Citations

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases