Chronic interfacing with the autonomic nervous system using carbon nanotube (CNT) yarn electrodes
SCiENtifiC REPORTS |
Chronic interfacing with the autonomic nervous system using carbon nanotube (CNT) yarn electrodes
Grant A. McCallum
Thomas E. Eggers
Dominique M. Durand
OPEN Published: xx xx xxxx The ability to reliably and safely communicate chronically with small diameter (100-300 ?m) autonomic nerves could have a significant impact in fundamental biomedical research and clinical applications. However, this ability has remained elusive with existing neural interface technologies. Here we show a new chronic nerve interface using highly flexible materials with axon-like dimensions. The interface was implemented with carbon nanotube (CNT) yarn electrodes to chronically record neural activity from two separate autonomic nerves: the glossopharyngeal and vagus nerves. The recorded neural signals maintain a high signal-to-noise ratio (>10 dB) in chronic implant models. We further demonstrate the ability to process the neural activity to detect hypoxic and gastric extension events from the glossopharyngeal and vagus nerves, respectively. These results establish a novel, chronic platform neural interfacing technique with the autonomic nervous system and demonstrate the possibility of regulating internal organ function, leading to new bioelectronic therapies and patient health monitoring.
The autonomic nervous system provides non-directed, unconscious control over bodily functions, such as
digestion, breathing and heartbeat. Interfacing with this system to first detect a physiologic anomaly from the neural
activity and then modulate its signaling behavior using electrical stimulation could create revolutionary
diagnostic and therapeutic modalities targeting chronic health conditions and diseases1?10. However, these autonomic
nerve bundles (fascicles) have small diameters (100?300 ?m) and are contained within a mechanically strong
and a highly, electrically insulated membrane called the perineurium11. In addition to the implant device
biocompatibility, these limitations make it challenging to create a chronic, neural interface that is both reliable and
safe for implantation. The leading theory why chronic neural interfaces fail is the mechanical mismatch between
the implant and the surrounding biological tissue, causing micromotion between the two media12. This
triggers the body?s natural inflammation process enhancing the fibrotic encapsulation barrier between the implant
and the neural tissue and can result in recording signal losses with higher stimulation current thresholds that
could damage the surrounding tissue13. Another important consideration is the dimension of the implant. Recent
data in the central nervous system (CNS) indicate that decreasing the size of the implant generates little or no
reaction14. Therefore, two important specifications are that small nerve implants should have axon-like
dimensions and flexibility.
Cuff electrodes are the most widely used neural interface type based on their perceived safety compared to
other neural interface implementations, ease of implantation and clinical results from nerve stimulation trials15,16.
A true, feedback-based neuromodulation system must be able to successfully record and process neural activity
with a high signal-to-noise ratio (SNR), which requires a stable, low electrical impedance between the recording
electrode and the neural tissue. However, extrafascicular cuff electrodes fail to produce neural recordings that are
highly selective with high SNR because they detect weak, local field potentials outside the electrically insulating
perineurium layer17. Others have attempted to record chronically using wire hook electrodes around sympathetic
nerves but the signals they acquire can be highly contaminated with triboelectric noise18?20 (Supplementary
Fig.?5). To address these limitations, microfabrication techniques have been employed to create intrafascicular
electrodes designed to penetrate the perineurium and place the electrode contacts in closer proximity to the nerve
fibers. The three most prevalent designs are: (1) the longitudinal intrafascicular electrode (LIFE)21, which is sewn
into the nerve along its length; (2) the transverse intrafascicular multi-channel electrode (TIME)22, which
penetrates through the nerve?s cross-section; and (3) the Utah Slanted Electrode Array (USEA)23, which is an array of
multi-height, cone-shaped, needle structures that penetrate the nerve similar to the TIME. Although these
intrafascicular electrodes have been able to initially record high SNR neural signals24,25, they have yet to demonstrate
chronic implant reliability because, we hypothesize, the electrode flexural rigidity is at least five orders of
magnitude greater than the surrounding neural tissue14,26. To achieve chronic neural recordings with high selectivity
and SNR, a neural interface should satisfy three important design criteria: high flexibility, small dimensions and
intrafascicular penetration. To achieve such a design, we used carbon nanotube fabrication techniques to create
an intrafascicular CNT yarn (CNTY) interface. The CNT yarn possesses an electrical conductivity and flexural
rigidity that is superior to existing intrafascicular implementations. To insert such a flexible material such as a
CNT yarn into a nerve, we have developed a novel implantation method based on microneurography techniques.
We have chronically implanted this new electrode into the rat vagus and glossopharyngeal nerves to determine if
gastric extension and hypoxic events can be identified based on the chronically recorded autonomic nerve
activity. Preliminary chronic stimulation experiments were performed on the rat sciatic nerve and charge thresholds
were recorded over a six week period.
Chronic recordings of autonomic nervous system activity. Achieving chronic recordings from small
diameter autonomic nerves requires both novel electrode development and implantation process (Fig.?1a?c).
Owing to their excellent electrical, thermal, and mechanical properties, carbon nanotubes have attracted a great
deal of interest for various applications, particularly as new flexible electrical wires in advanced biomedical
systems. However, the connection of nanotubes in biomedical systems to the outside world is challenging. Recent
efforts have led to the creation of CNT yarns by continuous, high-rate spinning vertically-aligned multi-walled
CNT (VA-MWCNT) arrays27 (Fig.?1a). The availability of flexible, mechanically strong, and highly conducting
CNT yarns with micro-sized diameters has offered ideal alternatives to replace rigid and expensive noble metal
wires. Compared to platinum iridium wires of identical diameter, the CNT yarns are both 10 times more
conductive and more flexible (Fig.?2a and Supplementary Fig.?1a). These two parameters are of critical importance for
interfacing with small diameter nerves since the high conductance will reduce the electrical baseline noise in the
neural recordings and the flexibility will be more mechanically similar to the surrounding tissue, reducing nerve
damage and the inflammation response.
The CNT yarns were coated with a thin insulating layer except for a ~500 ?m long segment at one end.
This uninsulated segment is wound around the tip of a Tungsten microneedle similar to those used in clinical
microneurography procedures28 (Fig.?1b). Under sterile surgical conditions, a portion of the target nerve was
dissected away from all connective tissue. The microneedle loaded with the CNT yarn electrode was then advanced
into the nerve until the exposed yarn segment was completely inside. The microneedle was then pulled back out
of the nerve leaving the CNT yarn electrode embedded within (Fig.?1c, Supplementary Fig.?3b). For each target
nerve, two CNT yarn electrodes were implanted within a 2 mm separation distance. Fibrin glue was then applied
to completely surround the nerve at the implant sites to help secure the CNT yarns in place. The opposite ends
of the electrode directly interfaced to a percutaneous connector. A removable neural amplifier recording board
was used to amplify, filter, digitize and store the neural activity to a laptop computer (Fig.?1d, Supplementary
Fig.?4c). Figure?1e depicts the entire neural recording assembly from the implanted CNT yarns to the external
data acquisition computer. Spontaneous, multi-unit neural spike activity recorded using CNT yarn electrodes
after being implanted in the vagus nerve for four months (Fig.?1f) shows signal amplitudes ranging from 30?60
Furthermore, signal processing techniques can be applied to these recorded neural waveforms to identify
individual spike activity based on a desired threshold level, clustering algorithms to separate the activity into
groups based on the waveform shape and temporal firing patterns, and ultimately determine the firing rate for
each identified neural group (Fig.?1f).
Figure?2 summarizes three important points regarding this interface technology: (1) the flexural rigidity value
of the CNT yarn electrode compared to a Platinum-Iridium (PtIr) wire of a similar 10 ?m diameter is an order
of magnitude lower; (2) the measured, in vivo, chronic impedance values are stable over time; (3) the resulting
SNR from the recorded waveforms is also stable over time. As shown in Fig.?2a, the flexural rigidity was measured
using two methods, gravity and atomic force microscopy (AFM). In both cases, the flexural rigidity of the CNT
yarn was found to be an order of magnitude less when compared to the PtIr wire demonstrating its superior
flexibility compared to traditional electrode materials29. Figure?2b shows the CNTY 1 kHz impedance magnitude
measurements over a 10 week, post implant period. Data was collected from five different subjects each with two
implanted CNT yarns (n = 10). The average impedance value over the 10 weeks was 17.8? 7.7 k-Ohms
(average ? std. dev). Linear regression was performed on the data and a t-test was run on the regression line slope. The
results determined the slope was not significantly different from zero (p= 0.428) indicating a stable impedance
magnitude over the 10 week period. Similarly, the SNR over a 10 week period is shown in Fig.?2c from three
different subjects. The recorded spikes were analyzed separately in three different amplitude groups: High, Medium
and Low. For each group, the SNR was calculated by taking the spike peak-to-peak voltage levels and dividing
it by the baseline standard deviation value in a period with no neural activity. For the three SNR groups high,
medium and low, the average SNR was found to be 62.2 ? 2.5, 25.11 ? 1.1 and 11.20 ? 0.2 (average ? std. dev.),
respectively. Linear regression was also performed to determine if each SNR group?s regression line slope was
significantly different from zero. The results indicate none were significantly different from zero with P-values of
0.420, 0.298 and 0.448 for the high, medium and low SNR groups, respectively.
Chronic recordings of glossopharyngeal nerve activity. The glossopharyngeal nerve (GPN) is known
to contain afferent and efferent nerve fibers that innervate the upper airways, taste and temperature receptors on
the tongue, and chemoreceptor afferents that originate in the carotid body and baroreceptor afferents that
emanate from the carotid sinus and join the GPN through the carotid sinus nerve (CSN) (Fig.?3a)30. Chemoafferent
activity from the carotid body and its role in cardiovascular and respiratory diseases has received
considerable interest31?33. Increased activity of carotid body chemoreceptors have been linked to sleep apnea and heart
failure34?36. In addition, chemoreflex-evoked sympathetic responses are enhanced in humans and animal models
of systemic hypertension37?40. We have investigated the GPN?s response to a hypoxic challenge given to an
anesthetized rat that has been implanted with CNT yarn electrodes for 16 days (Fig.?3b). The recorded neural activity
is presented with an overlay of the hypoxic events in which the gas used to anesthetize the rat was switched to a
highly concentrated nitrogen gas (90%/10%, nitrogen/oxygen) until the animal?s blood oxygen saturation
percentage was reduced to 75%. The gas was then switched back to the anesthetic gas, allowing the rat to recover. In
some cases, a second hypoxic event was administered following recovery (Fig.?3c). A total of 11 hypoxic events
were administered at 7, 10, and 16 days post implant.
The post-processed data revealed two distinct groups of neural activity (Fig.?3c) with the individual spike
signals shown in the magnified trace and again in the sorted spike clusters plot for a single experiment, shaded
regions indicated the standard deviation for each cluster. Each group?s firing rate over the course of six
experiments is displayed in Fig.?3d, aligned with the onset of hypoxia at time = 0. During several experiments, it took
an extra 10 seconds for the rat to reach 75% blood oxygen saturation; thus the data from these experiments is
excluded from Fig.?3d, though it is still used in the statistical analyses of firing rates. The table in Fig.?3d shows
the firing rates for each of the two spike groups (?standard deviation) immediately before the onset of hypoxia,
at the end of the hypoxic event, and approximately 90 seconds after the hypoxic event. Experiments 1 and 2 did
not have sufficient recovery data recorded, but were still included in the analysis for the firing rates before and at
the end of hypoxia.
Cluster separation was analyzed using the UltraMegaSort2000 MATLAB software package41. Quality
metrics (false positive and false negative events) for both clusters are shown at 7, 10, and 16 days in Tables?1 and 2.
Figure?4a shows separation of the two groups in the first two principal components at 10 days post-implant;
Supplementary Figs?6a and 7a show the feature-space of the two groups at 7 and 16 days, respectively. Figure?4b
shows cross-correlation of the spike times of the two clusters at 10 days, with Supplementary Figs?6b and 7b
showing cross-correlation of spike times at 7 and 16 Days. Finally, Fig.?4c,d, and Supplementary Figs?6c,d, 7c and d
show the autocorrelation of each cluster?s spike times at 7, 10, and 16 days.
The Group 1 firing rate decreased significantly from before the onset of hypoxia to the end of the hypoxic
event (p = 0.00036), and that firing rate increased significantly from the end of the event to end of the 90second
recovery time (p = 0.0023). However, the firing rate was not significantly different between the pre-hypoxia and
post-recovery measurements (p = 0.34). Similarly, the Group 2 firing rate increased significantly from before the
onset of hypoxia to the end of the hypoxic event (p = 0.00019), and that firing rate decreased significantly from
the end of the event to end of the 90 second recovery time (p = 0.0018). Additionally, the firing rate at the end
of the recovery period was slightly higher than the firing rate before the hypoxia was administered (p = 0.012).
We hypothesize that Group 1 could be baroreceptor activity as the blood pressure is known to decrease for
anesthetized rats under hypoxia42, which reduces baroreceptor activity. Furthermore, Group2 could be
chemoreceptor activity based on their known behavior of increased activity during falling blood oxygen level
concentrations43. This experiment was performed on three different days over a 16day period with identical results,
demonstrating the possibility of identifying hypoxic events from recorded neural activity in the glossopharyngeal
Chronic recordings of vagus nerve activity. The vagus nerve (VN) contains fibers that innervate many
internal organs, including the stomach and digestive system44. It is likely that vagal activity plays an important role
in a variety of pathologies, such as irritable bowel syndrome (IBS). We investigated vagal response to a stomach
Refractory violations, f1p
Overlap of clusters, f2p
Composite, max ( f1p , f2p)
distention45?47 given to an anaesthetized rat that has been implanted with CNT yarn electrodes in the left cervical
vagus for up to 50 days. The rat was fasted overnight before the procedure, and then a tube was inserted into the
rat?s stomach. 3 mL of sterile saline was infused into the stomach and the tube was immediately removed to
prevent inhalation of saline into the lungs. Vagal activity was recorded for several minutes before, during, and after
In response to the gastric distention, we observed intermittent bursting behavior (Fig.?5) during and after
saline infusion into the stomach. These bursts appeared to occur randomly in response to the gastric
stimuli, but were observed each of the five times the experiment was conducted over a 21-day span (40 to 61 days
post-implant). The repeatability of this result illustrates the possibility of identifying patient risk for pathological
gastrointestinal responses via vagal activity, or patient responses to treatments and therapies. In addition, due to
the nearly ubiquitous innervations of the VN, this technology may be useful in identifying responses to other
stimuli, including hypoxia, changes in heart rate or blood pressure, or bronchoconstriction. It may also be
possible to specifically target specific branches of the VN to isolate physiological responses.
Chronic CNT nerve histology using CLARITY. Figure?6 shows a VN implanted with a CNT yarn
electrodes for seven days. Figure?7 shows a GPN implanted with CNT yarn electrodes for 138 days. Figure?6b shows
no evidence of response from antigen-presenting immune cells (activated microglia) after one week. There is no
significant localized expression of oligodendrocyte specific protein (OSP, shown in Fig.?6c), suggesting a small
population of Schwann cells, and thus no significant nerve injury. Similarly, Fig.?7a shows a lack of localized
neutrophil activity, which suggests a lack of prolonged inflammatory response after chronic implantation. DAPI and
collagen I show a thin layer of cellular encapsulation around the CNT yarns (Figs?6d, 7b, and c); thickness was
estimated to be 35.0 ? 2.67 ?m in the VN, and 31.7 ? 4.5 ?m in the GPN.
To estimate the thickness of the layer of encapsulation surrounding the implant, the perpendicular distance
from the Parylene-C structure to the rim of the capsule (visually identified by change in DAPI/Collagen I density)
was measured at several points along the electrode implant. Several measured locations are shown in Fig.?6d.
Chronic CNT yarn electrode stimulation charge thresholds. To test the ability of the nanowire
interface to stimulate nerves, CNT wires were implanted in the tibial and peroneal branches of the sciatic nerve in five
rats. Isolated current stimulation pulses of charge balanced, 100 ?sec pulse width per phase were applied with
variable amplitude to activate the nerve. The threshold for activation was estimated by detecting muscle twitches
in the medial gastrocnemius and tibialis anterior muscles. Figure?8 shows the charge threshold measured from
the sciatic nerve branches for five rats through 4 weeks and 2 rats at six weeks. The results across animals were
fairly consistent, with similar thresholds from week 1?2 followed by an increase in threshold by week 4. Much
of this change could be attributed to a single wire pair (out of 10) with a threshold increasing from 4 to 25 nC.
It is unclear if this represents an increase due to natural encapsulation of the electrode, or if this electrode was
somehow displaced within the fascicle. Regression analysis revealed that the trend line had a small but statistically
significant positive slope over the six week period (p < 0.05).
We have developed the first successful use of CNT yarns to be implanted intrafascicularly to record/stimulate
autonomic nerve activity in long-term chronic rats. The CNT yarns are fabricated using a dry-spinning method27,
which is superior to chemical bath, wet-spinning methods48 for this specific application as the latter could leave
behind non-biocompatible residues detrimental for long term implantation. The CNT yarns were connected to
a more mechanically robust DFT wire to enable tunneling to a percutaneous port. The entire electrode implant
assembly was insulated with Parylene-C using vapor deposition except for a small portion of the distal end for
nerve implantation and the proximal end to connect to an external electrical connector. The CNT yarns were
chronically implanted into two different autonomic nerves and electroneurogram data was collected during
physiological challenges on rats, under anesthetized conditions, for up to 16 weeks. Statistical analysis of the measured
impedance and SNR of the chronically implanted subjects, indicate that both metrics are stable over a 10 week
implant period as shown in Fig.?2.
Chronic neural spike activity was successfully recorded from two separate autonomic nerves:
glossopharyngeal and vagus. In both nerves the signal-to-noise ratio was over 10 dB for the duration of the implant. Different
unit activity could be identified from the recordings and separated into distinct functional groups using
classification and clustering algorithms off-line. This is clearly apparent in the glossopharyngeal recordings where two
functional neural groups were repeatedly identified during hypoxia challenges throughout the duration of the
study. The two spike groups appear to be distinct across 3 days of recording (7, 10, and 16 days post implant),
with the first two principal components appearing relatively stable across those days (Fig.?4a, Supplementary
Figs?6a, 7a). The cross-correlation of spike times for the two groups shows an upward trend (Fig.?4b); usually
independent spike clusters would show a flat cross-correlation. However, the regular nature of the spiking
activity (lined up with the respiratory rhythm), and its response to hypoxia, could account for the upward trend in
the cross-correlation plot. The autocorrelation of spike times for each group (Fig.?4c,d) shows a small number
of refractory violations for Group 1 and a slightly larger number of refractory violations for Group 2. These
values vary day-to-day, but the fraction of refractory violations is less than 0.17 for both groups on all three days
(Table?1). Due to the size of our electrodes, each of the spike clusters could contain spikes from more than one
neuron, which could account for the refractory violations. We can see in Table?1 that the number of false positive
events due to cluster overlap for at 7, 10, and 16 days remains small. In Table?2, we can see that Group 1 has a
larger fraction of undetected spikes, which is expected since that group has smaller amplitude, which is close to
our threshold. However, both clusters have a small fraction of censored and overlapping false negative events.
The data shown in Fig.?4, Supplementary Figs?6, 7, and Tables?1 and 2 show that the two groups shown in Fig.?3
are distinct and repeatable over 3 recording days, up to 16 days post-implant. These two groups are believed to be
chemoreceptor and baroreceptor activity based on the location of the CNT yarn implants where the carotid sinus
nerve joins the glossopharyngeal nerve (Fig.?3). More investigation is needed to prove this claim is true, however,
a hypoxic event can clearly and repeatedly be identified from the recorded neural spike activity. The vagus nerve
activity proved to be more challenging to correlate to physiological function as the cervical vagus contains afferent
fibers from a multitude of organ systems within the body. Each vagus implant yielded different neural activity
patterns such as the spontaneous firings shown in Fig.?1, which are believed to be intestinal signals, or the elicited
gastric extension activity (Fig.?5). Based on the implant depth, CNT yarn electrodes could detect activity within
the entire cross-section of the nerve. Therefore, rather than implanting in the nerve trunk, targeting specific
nerve branches that contain a limited number of functional nerve fiber groups would be ideal to identify specific
This new chronic electrode interface has significant benefits over the existing intrafascicular implants in
regards to long-term stability. While some intrafascicular solutions have focused on using mechanically soft
materials21,22,49 and others on implants with small cross-sectional areas but mechanically hard materials23, we
hypothesize that to minimize the encapsulation process required for stable long-term intrafascicular interfaces one must
design for a flexural rigidity as close as possible to that of the surrounding neural tissue. Any relative motion to
the nerve caused by a non-flexible implant is believed to exacerbate the encapsulation process. Implanted CNT
yarns have a lower flexural rigidity value (Fig.?2a) compared to existing intrafascicular electrodes as they have
both a smaller second moment of inertia and lower modulus. Preliminary evidence of this is presented in Figs?6
and 7 as the cellular encapsulation observed in these CNT yarn experiments (~32 ?m, 19 weeks post-implant) is
smaller than the 135 ? 56 ?m shown in the TIME electrodes22, the 113 ? 2.8 ?m estimated for the thin-film LIFE,
and the 64.3 ? 2.5 ?m estimated for the PtIr LIFE49.
The results in Fig.?8 demonstrate the viability of using CNT yarn electrodes for chronic stimulation. While
difficult to directly compare to other studies as authors tend to use varying pulse widths and stimulation frequencies,
the injected charges here are slightly higher than previously published work investigating similar in vivo
intrafascicular recording/stimulation but utilizing different electrodes50. The increase in threshold at week four is not
unexpected as it typically marks the completion of the encapsulation layer formation and foreign body response
to the implant. These thresholds stabilized at one month time point, although the data shown is limited to only six
weeks post-op. However, three additional rats were implanted in the tibial nerve and left in vivo for 16 months. A
single stimulation experiment was run on these animals, as their percutaneous connectors were destroyed early
in the study. Of the four CNT yarn electrodes tested, three had normal impedance (~20 k) while two elicited
twitches at relatively low thresholds, roughly 20 nC, versus ~70 nC for the other two electrodes. Together these
results indicate that CNT yarn electrodes are relatively stable within the nerve and can be used for chronic
stimulation. The charge capacity for stimulation is significant higher (~300 times) than that of platinum iridium wires51.
Fundamentally, stimulating autonomic and somatic fibers is the same. The only difference would be differences in
charge thresholds, with the unmyelinated fibers to likely require higher thresholds than reported for the somatic
nerves. The data presented serves to demonstrate empirically that the CNT yarns can provide the necessary
charge to activate nerve fibers, as well as to provide an additional metric for the stability of the interface over time.
The preliminary histology data and high quality neural recording results we have presented indicate that it is
indeed possible to obtain information from the autonomic nervous system for long periods of time. This
capability opens the possibility to reliably obtain and decode autonomic nerve signals to understand the physiological
signaling mechanisms of the nervous system. The recording capability and the low stimulation charge thresholds
indicate the feasibility of chronic implanted CNT yarn electrodes for autonomic nerve control and modulation to
provide therapeutic and patient monitoring capabilities for various diseases and health conditions.
Carbon nanotube (CNT) yarn fabrication. VA-MWCNT arrays (with MWCNTs of ca.250 ?m in length
and ca.30 nm in diameter) (Fig.?1a), were first synthesized on Fe/Al-coated Si/SiO2 wafers using a low pressure
chemical vapor deposition (CVD) method according to our previously reported procedure52. In a typical
experiment, a 10-nm thick Al layer was coated on the silicon wafers before the deposition of 3-nm thick Fe film. The
catalyst coated substrate was then inserted into a quartz tube furnace for the VA-MWCNT array growth using
high-purity (99.99%) acetylene as the carbon source and Ar/H2 (2.5:1 v/v) as a carrier gas under 0.2 atm pressure
at 760 ?C52. For spinning VA-MWCNT arrays into the CNT yarn (Fig.?1a), one piece of the VA-MWCNT array
of approximately 3-mm width on Si/SiO2 wafer was initially peeled off from the edge of the sidewall, drawn
away from the CNT array, and attached onto a tip of the rod, which was connected to a motor27. A CNT yarn
was spun by the rotating rod that was moved away from the array, and at the same time the bundle was rotated
anti-clock-wise to form the CNT yarn (Supplementary Fig.?1b). The rotating speed was in the range of 2000?
20000 rpm and the diameter of the as-prepared yarn was in the range of 10?20 ?m, which was controlled by the
number of CNTs initially pulled out from the array. The yarn diameter was determined by SEM images.
CNT yarn electrode assembly process. The implant electrode assembly is comprised of two different
wire types: CNT yarn for implantation in the nerve and a more robust lead wire used to route the electrode under
the skin to a percutaneous electrical connector. The electrode lead wire is a 25cm long (1 ? 7 ? 25.4 ?m: 127 ?m
diameter) metal-to-metal composite wire with a silver core (28%) and a stainless steel outer tubing coated with
PFA insulation (35NLT?-DFT? wire, Fort Wayne Metals). The DFT? wire insulation was stripped by
approximately 3 mm at one end and joined to the CNT yarn by conductive epoxy resin (H20E, EPO-TEK). These wire
assemblies were then placed and fixed on a custom-made acrylic rack (Supplementary Fig.?2a) for insulating
the CNT yarn wire and junction. Parylene-C is chosen to be the insulation material for the CNT yarn. The
Parylene-C thickness was 3.5 ?m and was deposited using chemical vapor deposition (CVD). During the CVD
coating process, a 500 ?m portion of the CNT yarn end was left uninsulated by using tape to mask the desired
area. After the Parylene-C insulation was deposited, the individual electrode wire assemblies were removed from
the rack. The junction between the CNT yarn and the DFT? lead wire was placed within a 6 mm long (0.51 mm
ID, 0.94 mm OD) silicone tube (2415500 - Dow Corning Corporation). Any empty space inside the silicone tube
was filled with silicone elastomer (MED-4211/MED-4011, NuSil Silicone Technology) to further insulate the
Next, the uninsulated CNT yarn end and a short portion of the insulated yarn was wound around the tip of a
Tungsten microneedle (UEWSGKSNXNND, FHC Inc.) (Fig.?1b). Then the silicone tube between the CNT yarn
and DFT? wire was fixed to the microneedle using a slip knot made of thread (Supplementary Fig.?2b). After
winding, 3 mm of DFT? wire insulation at the free end was stripped to be eventually soldered to a percutaneous
electrical connector. A ground electrode (Supplementary Fig.?2c) was made with a 3 ? 4 cm, 25 ?m diameter
Platinum-Iridium (Pt-Ir) wire and DFT? wire. The Pt-Ir wire and DFT? wire were joined together by conductive
epoxy. This ground electrode was placed chronically under the dorsal skin of the animal.
Chronic In vivo implantation procedure. All surgical and experimental procedures were done with the
approval and oversight of the Case Western Reserve University, Institutional Animal Care and Use Committee to
ensure compliance with all federal, state and local animal welfare laws and regulations.
Adult male Sprague Dawley rats (350?450 grams, Charles River Laboratories) were used for all peripheral
nerve implants discussed. Prior to surgery, all surgical instruments and implant assemblies were autoclave
sterilized at 250 ?C for 1 hour. During surgery, the animals were gas anesthetized under 2% isoflurane in 1 L/min
oxygen. For each of the nerves, a section approximately 3?4 mm was completely separated from and devoid of all
connective tissue. A custom made hook (Fig.?1c, Supplementary Fig.?3a) was attached to a three-axis
micromanipulator and positioned to provide a small amount of tension and suspend the nerve in the air. The slip knot was
then removed, thereby releasing the silicone tube junction from the microneedle. The CNT yarn electrode
assembly was then advanced into the nerve until all of the uninsulated CNT yarn segment was within the perineurium.
Micro tweezers were then used to apply a small amount of pressure at the implant site while the Tungsten
mirconeedle was carefully removed leaving the CNT yarn inside the nerve (Supplementary Fig.?3b). The hook was
then advanced along the nerve to the next implant site, approximately 2 mm away, and the implant process was
repeated. The CNT yarns and the silicone tube junctions were then carefully positioned within the incision site
and completely encapsulated with 1 mL of fibrin glue (Tisseel, Baxter International Inc.) (Supplementary Fig.?3c)
to secure the implant in place while the normal physiological process of collagen encapsulation took place while
the fibrin glue was absorbed by the body.
The free DFT? wire ends are then tunneled and externalized at a location on the back between the shoulder
blades. The free DFT? wire ends and the ground wire are then passed through a custom made, percutaneous
silicone plug and soldered to an electrical connector (MCP-05-SS, Omnetics Connector Corporation). The
connector was then fitted inside the silicone tube of the percutaneous plug and the tube was filled with UV curable,
medical grade epoxy (OG603, Epoxy Technology Inc.).
Blunt dissection is then performed to create a pocket in the back between the skin and muscle. The implanted
wires and ground wire are placed within this pocket. The percutaneous plug?s mesh base was then sutured to the
underlying muscle fascia using biodegradable 3-0 sutures (Vicryl Plus, Ethicon). Finally, all skin incision sites
were closed using 4-0 sutures (Webpro?, Patterson Veterinary).
Chronic autonomic nerve recordings. The recording system was designed to accommodate two,
differential recording channels. Each recording channel has eight amplifier hardware averaging for increased
signal-to-noise recording results53. This was achieved by creating, two custom, printed circuit boards (PCB)
(Supplementary Fig.?4a and b). The first PCB contains the mating connector (MCS-05-DD, Omnetics Connector
Corporation) to the connector which resides in the implanted percutaneous plug. DFT? wires are soldered to
make electrical connections between the two PCB boards for all the CNT yarn electrode implants (up to four
total) and the implanted ground wire.
The second PCB routes each differential electrode pair to an electrical connector (NPD-36-AA-GS, Omnetics
Connector Corporation) that connects directly to a commercially available 16-amplifier, neural instrumentation
amplifier board (C3313/RHD2216, Intan Technologies LLC). The electrode connections are routed such that
one differential channel is simultaneously recorded with amplifiers 0?7 and the other is captured with amplifiers
8?15. The RHD2216 performs the signal filtering, analog signal multiplexing, and analog to digital conversion
and is controlled via a digital SPI connection to a commercially available acquisition system (C3100/RHD20000
Evaluation System, Intan Technologies, LLC) (Fig.?1e and Supplementary Fig.?4c). Neural signal acquisition was
performed and stored to a laptop computer using the following system configurations: 20k samples/sec/amplifier,
DSP offset removal was enabled with the high-pass filter cutoff frequency set at 100Hz, the amplifier bandpass
filters were configured to be between 100Hz and 9 kHz.
A manual trigger switch signal is recorded via a separate analog-to-digital input line on the RHD2000
evaluation system. This trigger signal is synchronized with the recorded neural amplifier data and is also sampled at
a rate of 20 k samples/sec. Holding down a simple push button switch, produces a high logic-level trigger signal
which is used to mark different events during the various experiments. Releasing the push button returns the
trigger signal to a low logic-level (Supplementary Fig.?4c).
Impedance magnitude and signal-to-noise ratio measurements and analysis. Impedance
magnitude measurements were taken, over a 10 week period, between each implanted CNT yarn and an implanted
ground wire at 1 kHz using a handheld meter (Protek, Z580 LCR Meter) while the animal was anaesthetized. SNR
values were calculated for each recording data set over a ten week period. For each recording session, the baseline
noise standard deviation was calculated using the MATLAB function std() between two time points where no
neural activity was present. Because of the multi-unit recordings obtained with the CNT yarns, neural spikes were
placed into three groups (High, Medium and Low) based on their peak-to-peak amplitudes. For each group, in
each recording session, three spikes were used to calculate the voltage peak-to-peak values and divide them by
the baseline noise value to determine the SNR. Minitab v17.3.1 was used to perform a linear regression analysis
to determine if there was a significant change in impedance and SNR over the 10 week period using a two-tailed,
paired t-test on the slope of the regression line.
Post signal processing of acquired neural recordings. Using Matlab (version R2016b, Mathworks), the
eight parallel amplifier channels of differential recording data from each CNT yarn implant pair was first
averaged. The data was then processed using a 7th order, zero-phase, digital bandpass filter with the lower and upper
cutoff frequencies equal to 300 Hz and 1,000 Hz, respectively. Filtered data is then imported into a MATLAB
algorithm that uses fuzzy c-means clustering (fcm) to threshold, cluster, and count spikes54,55. This algorithm first
identifies spikes as local maxima that exceed a user-specified threshold. These spikes are then aligned with the
maxima centered and 1.5 ms (30 samples) of data before and after the maximum saved for analysis. These spike
waveforms are run through MATLAB?s principal component analysis function (pca) to reduce the dimensionality
of the spikes from 61 to 5 without any significant loss in information (greater than 99% of variance is maintained
using the top 5 principal components). The principal component scores associated with each spike are then
processed via the MATLAB fcm function and sorted into clusters. The number of clusters is user-specified and is
generally determined based on qualitative assessment of clusters and quantitative measures of cluster separation
(i.e. L-Ratio)56. Firing rates for each cluster are calculated using a 2 second sliding window and are linearly
interpolated using a version of MATLAB?s envelope function. Average firing rates were calculated during 5-second
intervals before hypoxia, at the end of the hypoxic event, and after a 90-second recovery period. A two-tailed,
paired t-test was then conducted to compare firing rates of each spike cluster at each of these three time-points.
Spike sorting analysis was confirmed using the spike sorting package UltraMegaSort2000 (UMS2000),
described previously by Hill et al.41. Spikes were separated using the splitmerge_tool GUI, and plots showing the
spikes in feature space, spike time correlations, and cluster firing rates were generated using the UMS2000 plot
functions (plot_features, plot_xcorr, plot_isi, plot_stability). Cluster quality metrics (fractions of false negative
and false positive events) were calculated for each day using the UMS2000 functions, and are displayed in Tables?1
and 2. Refractory period violations were calculated with ss_rpv_contamination; undetected spike fractions were
calculated using ss_undetected; censored spike fractions were calculated using ss_censored; and overlap of
clusters was calculated using ss_gaussian_overlap.
Histology. Nerves were extracted from rats after implantation, and were immediately placed into hydrogel
monomer solution (clarityresourcecenter.org). After incubation in the hydrogel monomer solution for 2?3 weeks
at 4 ?C, samples were then polymerized at 37 ?C and low oxygen (a layer of peanut oil was added to the tube to
decrease the concentration of oxygen in the solution). After polymerization, the excess gel was removed, and
the nerve is placed into clearing solution for passive clearing, as described on the Clarity Resource Forum57.
Primary antibodies (W and X) were diluted 1:500 in 1 mL of PBST. After a 1-day wash in PBST, the sample was
incubated in secondary antibody solution, which contained secondary antibodies (Y and Z), diluted 1:500, and
DAPI, diluted 1:1000. After another 1-day PBST wash PBST, the sample was placed in VectaShield with DAPI
(Vector Laboratories) on a glass-bottom petri dish (Ted Pella, Inc.). Samples were imaged on a Leica SP8 gSTED
Super-Resolution Confocal microscope (Leica Microsystems).
Chronic sciatic nerve charge threshold stimulation. Five Sprague Dawley rats each had two CNT yarn
electrodes implanted into their tibial and peroneal fascicles, and stimulation thresholds were subsequently
measured. While anesthetized under isoflourane, biphasic square pulses were applied differentially using an isolated
stimulator (A-M Systems Analog Isolator 2200) to these implanted electrodes (i.e. cathode and anode both in
the fascicle) while the leg was monitored via visual inspection and measured electromyogram (EMG) response.
One hertz stimulation was performed to avoid muscle fatigue. Each phase of the stimulation lasted 100 ?s with
no inter-phase interval.
Data Availability. The datasets generated during and/or analysed during the current study are available from
the corresponding author on reasonable request.
We thank W. Marcus for the animal veterinary assistive care throughout the study; S. Lewis and Y.H. Hsieh?s
input on surgical approaches and discussion on the acquired neural signals; J. Vanderford and A. Hlavacik of
SMART Microsystems, Elyria, OH for the Parylene-C deposition; P. Conrad for confocal microscopy assistance
at the Light Microscopy Imaging Facility at Case Western Reserve University, made available through the Office
of Research Infrastructure (NIH-ORIP) Shared Instrumentation Grant (S10OD016164); This work is supported
by GlaxoSmithKline ? Bioelectronics Medicine and National Institutes of Health SPARC Award Number
D. Durand conceived the project. G. McCallum program managed, performed glossopharyngeal implants, created
the recording electronics system, performed animal experiments, analyzed data and wrote the manuscript. X. Sui
and J. Marmerstein performed vagus implants, animal experiments and data analysis. J. Marmerstein performed
CLARITY histology and spike sorting analysis. X. Sui, C. Qui and Y. Zheng fabricated the CNT yarn implant
assemblies. T. Eggers performed sciatic implants, performed animal experiments and analyzed data. C. Hu and L.
Dai fabricated the CNTs and CNT yarns.
Supplementary information accompanies this paper at doi:10.1038/s41598-017-10639-w
Competing Interests: The authors declare that they have no competing interests.
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and
Open Access This article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons license, and indicate if changes were made. The images or other third party material in this
article are included in the article?s Creative Commons license, unless indicated otherwise in a credit line to the
material. If material is not included in the article?s Creative Commons license and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the
copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
39. Trzebski , A. , Tafil , M. , Zoltowski , M. & Przybylski , J. Increased sensitivity of the arterial chemoreceptor drive in young men with mild hypertension . Cardiovascular research 16 , 163 - 172 ( 1982 ).
40. Somers , V. K. , Mark , A. L. & Abboud , F. M. Potentiation of sympathetic nerve responses to hypoxia in borderline hypertensive subjects . Hypertension 11 , 608 - 612 ( 1988 ).
41. Hill , D. N. , Mehta , S. B. & Kleinfeld , D. Quality metrics to accompany spike sorting of extracellular signals . The Journal of neuroscience: the official journal of the Society for Neuroscience 31 , 8699 - 8705 , doi:10.1523/jneurosci.0971- 11 . 2011 ( 2011 ).
42. Mendoza , J. P. et al. Role of nitric oxide-containing factors in the ventilatory and cardiovascular responses elicited by hypoxic challenge in isoflurane-anesthetized rats . Journal of applied physiology (Bethesda , Md.: 1985 ) 116 , 1371 - 1381 , doi:10.1152/ japplphysiol.00842. 2013 ( 2014 ).
43. Fukuda , Y. , Sato , A. & Trzebski , A. Carotid chemoreceptor discharge responses to hypoxia and hypercapnia in normotensive and spontaneously hypertensive rats . Journal of the autonomic nervous system 19 , 1 - 11 ( 1987 ).
44. Berthoud , H. R. & Neuhuber , W. L. Functional and chemical anatomy of the afferent vagal system . Autonomic neuroscience: basic & clinical 85 , 1 - 17 , doi:10.1016/s1566- 0702 ( 00 ) 00215 - 0 ( 2000 ).
45. Ozaki , N. , Sengupta , J. N. & Gebhart , G. F. Mechanosensitive properties of gastric vagal afferent fibers in the rat . Journal of neurophysiology 82 , 2210 - 2220 ( 1999 ).
46. Schwartz , G. J. , McHugh , P. R. & Moran , T. H. Gastric loads and cholecystokinin synergistically stimulate rat gastric vagal afferents . The American journal of physiology 265 , R872 - 876 ( 1993 ).
47. Takahashi , T. & Owyang , C. Characterization of vagal pathways mediating gastric accommodation reflex in rats . The Journal of physiology 504(Pt 2) , 479 - 488 ( 1997 ).
48. Behabtu , N. et al. Strong, light, multifunctional fibers of carbon nanotubes with ultrahigh conductivity . Science 339 , 182 - 186 , doi:10.1126/science.1228061 ( 2013 ).
49. Lago , N. , Yoshida , K. , Koch , K. P. & Navarro , X. Assessment of biocompatibility of chronically implanted polyimide and platinum intrafascicular electrodes . IEEE transactions on bio-medical engineering 54 , 281 - 290 , doi:10.1109/tbme. 2006 . 886617 ( 2007 ).
50. Badia , J. et al. Comparative analysis of transverse intrafascicular multichannel, longitudinal intrafascicular and multipolar cuff electrodes for the selective stimulation of nerve fascicles . J Neural Eng 8 , 036023 , doi:10.1088/ 1741 -2560/8/3/036023 ( 2011 ).
51. Vitale , F. , Summerson , S. R. , Aazhang , B. , Kemere , C. & Pasquali , M. Neural stimulation and recording with bidirectional, soft carbon nanotube fiber microelectrodes . ACS nano 9 , 4465 - 4474 , doi:10.1021/acsnano.5b01060 ( 2015 ).
52. Qu , L. , Dai , L. , Stone , M. , Xia , Z. & Wang , Z. L. Carbon Nanotube Arrays with Strong Shear Binding-On and Easy Normal LiftingOff . Science 322 , 238 - 242 , doi:10.1126/science.1159503 ( 2008 ).
53. Dweiri , Y. M. , Eggers , T. , McCallum , G. & Durand , D. M. Ultra-low noise miniaturized neural amplifier with hardware averaging . J Neural Eng 12 , 046024 , doi:10.1088/ 1741 -2560/12/4/046024 ( 2015 ).
54. Lewicki , M. S. A review of methods for spike sorting: the detection and classification of neural action potentials . Network: Computation in Neural Systems 9 , R53 - R78 , doi:10.1088/ 0954 -898X_ 9 _4_ 001 ( 1998 ).
55. Oliynyk , A. , Bonifazzi , C. , Montani , F. & Fadiga , L. Automatic onlinespike sorting with singular value decomposition and fuzzy C-mean clustering . BMC Neuroscience 13 , 1 - 19 , doi:10.1186/ 1471 -2202-13- 96 ( 2012 ).
56. Schmitzer-Torbert , N. , Jackson , J. , Henze , D. , Harris , K. & Redish , A. D. Quantitative measures of cluster quality for use in extracellular recordings . Neuroscience 131 , doi:10.1016/j.neuroscience. 2004 . 09 .066 ( 2005 ).
57. CLARITY Resources, clarityresourcecenter.org ( 2016 ).