Organ- and function-specific anatomical organization of vagal fibers supports fascicular vagus nerve stimulation

Naveen Jayaprakash; Weiguo Song; Viktor Toth; Avantika Vardhan; Todd Levy; Jacquelyn Tomaio; Khaled Qanud; Ibrahim Mughrabi; Yao-Chuan Chang; Moontahinaz Rob; Anna Daytz; Adam Abbas; Zeinab Nassrallah; Bruce T Volpe; Kevin J Tracey; Yousef Al-Abed; Timir Datta-Chaudhuri; Larry Miller; Mary F Barbe; Sunhee C Lee; Theodoros P Zanos; Stavros Zanos

doi:10.1016/j.brs.2023.02.003

. Author manuscript; available in PMC: 2023 May 30.

Published in final edited form as: Brain Stimul. 2023 Feb 10;16(2):484–506. doi: 10.1016/j.brs.2023.02.003

Organ- and function-specific anatomical organization of vagal fibers supports fascicular vagus nerve stimulation

Naveen Jayaprakash ^a,¹, Weiguo Song ^a,¹, Viktor Toth ^a,¹, Avantika Vardhan ^a, Todd Levy ^a, Jacquelyn Tomaio ^a, Khaled Qanud ^a, Ibrahim Mughrabi ^a, Yao-Chuan Chang ^a, Moontahinaz Rob ^a, Anna Daytz ^b, Adam Abbas ^a, Zeinab Nassrallah ^b, Bruce T Volpe ^a, Kevin J Tracey ^a, Yousef Al-Abed ^a, Timir Datta-Chaudhuri ^a, Larry Miller ^a, Mary F Barbe ^c, Sunhee C Lee ^a, Theodoros P Zanos ^a, Stavros Zanos ^a,^b,^d,^*

PMCID: PMC10228508 NIHMSID: NIHMS1898332 PMID: 36773779

Abstract

Vagal fibers travel inside fascicles and form branches to innervate organs and regulate organ functions. Existing vagus nerve stimulation (VNS) therapies activate vagal fibers non-selectively, often resulting in reduced efficacy and side effects from non-targeted organs. The transverse and longitudinal arrangement of fibers inside the vagal trunk with respect to the functions they mediate and organs they innervate is unknown, however it is crucial for selective VNS. Using micro-computed tomography imaging, we tracked fascicular trajectories and found that, in swine, sensory and motor fascicles are spatially separated cephalad, close to the nodose ganglion, and merge caudad, towards the lower cervical and upper thoracic region; larynx-, heart- and lung-specific fascicles are separated caudad and progressively merge cephalad. Using quantified immunohistochemistry at single fiber level, we identified and characterized all vagal fibers and found that fibers of different morphological types are differentially distributed in fascicles: myelinated afferents and efferents occupy separate fascicles, myelinated and unmyelinated efferents also occupy separate fascicles, and small unmyelinated afferents are widely distributed within most fascicles. We developed a multi-contact cuff electrode to accommodate the fascicular structure of the vagal trunk and used it to deliver fascicle-selective cervical VNS in anesthetized and awake swine. Compound action potentials from distinct fiber types, and physiological responses from different organs, including laryngeal muscle, cough, breathing, and heart rate responses are elicited in a radially asymmetric manner, with consistent angular separations that agree with the documented fascicular organization. These results indicate that fibers in the trunk of the vagus nerve are anatomically organized according to functions they mediate and organs they innervate and can be asymmetrically activated by fascicular cervical VNS.

1. Introduction

The autonomic nervous system maintains physiological homeostasis in organs and organ systems through neural reflexes. Autonomic reflexes are comprised of sensory neurons that detect changes in physiological functions, and motor neurons that regulate organ functions in response to physiologic changes. With its thousands of afferent (sensory, directed from the periphery to the brain) and efferent fibers (motor, directed from the brain to the periphery), the vagus nerve is the main conduit for bidirectional communication between the brain and body organs and it participates in numerous autonomic reflexes regulating cardiovascular, respiratory, gastrointestinal, endocrine and immune functions.

In the human vagus nerve, nerve fibers are arranged in fascicles, longitudinal bundles within the nerve separated by connective tissue [1,2]. Along the cervical and thoracic vagus nerve, afferent and efferent fibers leave the fascicles and emerge from the nerve trunk to form branches, which in turn provide sensory and motor innervation to the heart, the airways and the lungs [3,4]. Several aspects of the macroscopic anatomy of the vagus nerve have been described in detail, including perineural and epineural tissue [2], numbers and sizes of fascicles [5] and levels and patterns of emergence of organ-specific vagal branches [4,6-9]. Despite this extensive body of work, the transverse (horizontal) and longitudinal (vertical) arrangement of fascicles in the vagal trunk, with respect to the organs to which fascicles project and the functions they mediate, is limited. It was recently shown that fascicles forming the superior and recurrent laryngeal branches in the swine are located on one side of the vagal trunk [5]. It is unclear whether this spatial separation of fascicles concerns other organs, in addition to the larynx, and whether it persists throughout the vagal trunk, away from the nerve levels at which organ branches emerge. Answering these questions requires tracking the trajectories of individual fascicles along many centimeters of nerve length. Using histological methods would necessitate staining, imaging and image analysis of thousands of nerve sections, a technically prohibitive undertaking. Microscopic 3-D imaging techniques could be used instead, e.g., micro-computed tomography (micro-CT) [10], but those have not been applied to long sections of nerves from humans or large animals.

Likewise, elements of the microscopic anatomy of the vagus nerve, most importantly the types and morphological characteristics of fibers of the vagal trunk, have been well characterized in humans and in other species [1,6,9,11-23]. Fibers in the vagus nerve span a large range of sizes, from myelinated fibers with diameters >10 μm, to unmyelinated fibers with diameters <0.5 μm, with a variety of neurochemical phenotypes, including cholinergic, monoaminergic, glutamatergic and peptidergic [24,25]. Past studies have arrived at estimates of fiber counts by analyzing spatially limited nerve areas and extrapolating to the entire nerve, assuming a uniform spatial distribution within and across nerve fascicles. However, it is unknown whether fibers are indeed uniformly distributed, or they show preference for certain nerve areas, fascicles or even sub-fascicular sectors. Answering such questions requires identifying the majority, ideally the entirety, of the thousands of fibers in the vagal trunk, classifying them into several morphological types and assigning them to a location within fascicles. This task has not been feasible, as it requires processing of high-power microscopy images, able to resolve fibers with diameters ranging from <0.5 to >10 μm, across large nerve areas, using several stains to identify different morphological fiber types.

How fascicles and fibers are arranged inside the vagal trunk has implications for bioelectronic therapies based on vagus nerve stimulation (VNS) and for the rational design of selective VNS devices. VNS relies on an electrode device implanted on the vagus nerve, that delivers electrical stimuli to alter activity of nerve fibers and modulate organ function in diseases in which the vagus nerve is implicated. VNS has been tested in heart failure [26], inflammatory bowel disease [27] and rheumatoid arthritis [28], among other diseases. However, current VNS devices deliver stimuli all around the vagus nerve, without regard to the underlying spatial arrangement of fascicles and fibers. This non-selective mode of stimulation often limits therapeutic efficacy and results in clinically significant side effects from non-targeted organs [29,30]. For instance, a VNS device tested in patients with heart failure, targeting vagal fibers projecting to the heart, produced side effects from the airways by activating laryngeal fibers; that prevented investigators from delivering the relatively high stimulus intensities required to activate cardioprotective fibers, thereby limiting therapeutic efficacy [31,30]. Fiber-selective VNS using optogenetic methods has been demonstrated in experimental animals, e.g. Refs. [32-34], however the clinical translation of optogenetic stimulation in humans is still unclear. Placement of electrodes on vagal branches to engage specific organ functions is an active area of research (e.g., Refs. [35-38]), however it poses significant surgical challenges. On the other hand, stimulation devices placed on the cervical vagus nerve that accommodate the anatomical of the vagal trunk could in principle activate specific sub-populations of fibers and lead to increased efficacy with reduced side-effects [38,39]. Such devices could expand the indications of nerve stimulation to organs not targeted by existing therapies and allow the mechanistic study of specific vagal reflexes in health and disease, in humans and large animals. VNS targeting specific fascicles has been demonstrated in rats [35] and in sheep [39]. However, both those species have vagus nerves with one or two fascicles, unlike the multi-fascicular human and swine vagus nerves. Therefore, the ability of fascicular VNS to deliver organ- and function-specific neuromodulation in the human vagus is unknown.

In this study, we used surgical dissections, micro-anatomical tracking and microscopic imaging methodologies to characterize the transverse and longitudinal anatomical organization of the vagus, or any peripheral nerve, at the level of organ branches, nerve fascicles and nerve fibers. We applied those methodologies to resolve the fascicular and fiber organization of the vagus nerve of the swine, which most closely resembles that of humans [2,5]. We discovered specific spatial organization in the vagal trunk, with fascicles and fibers grouped in distinct clusters, with respect to the physiological functions they mediate, the organs they innervate and the morphological types of fibers. To test whether this fiber and fascicle arrangement can support fascicular VNS, we developed a multi-contact cuff electrode to accommodate the fascicular organization of the cervical vagus nerve and delivered fascicular VNS in anesthetized and awake swine. By targeting different sections of the nerve, we were able to evoke compound action potentials from nerve fibers of distinct functional types and elicit differential responses from organs innervated by different branches of the vagus nerve. Those electrical and physiological responses are asymmetrically distributed around the nerve in a manner consistent with the anatomical organization of the vagal trunk. Our findings indicate that fascicles and fibers in the vagus nerve are spatially organized according to the organs they innervate and the functions they mediate, and that fascicular VNS is, in principle, feasible in the swine vagus nerve, a multi-fascicular nerve similar to the human vagus. Our study provides quantitative data that could aim the construction of anatomically realistic computational models of the vagus nerve and nerve-electrode interfaces [40]. Finally, our findings have implications for anatomical constraints on the coding of interoceptive information in the central nervous system and the autonomic control of physiological homeostasis [41-43].

2. Methods

2.1. Animals and animal surgeries

In our study, we chose the swine as animal model because the vagus nerve of swine more closely resembles that of humans in its macroscopic fascicular structure, effective nerve diameter and relative thickness of the endoneurium, perineurium and epineurium, compared to rodents and non-human primates [2,5]. These macroscopic features are relevant to the translational testing of neurostimulation therapies based on implanted devices [30]. The effects of VNS on physiological and neural response were examined in 14 male swine (Yucatan; n = 13, ~50 kg; Yorkshire, n = 1, ~30 kg). All of the animal protocols and surgical procedures were reviewed and approved by the animal care and use committee of Feinstein Institutes for Medical Research and New York Medical College.

2.1.1. Animals and nerve samples used in different studies

In the swine organ-fascicle micro-CT tracking study: 2 right vagus nerves were used in total, both from Yucatan swine (RVN1, RVN2).
In the swine sensory/motor-fascicle micro-CT tracking study: 4 right vagus nerves were used in total, 3 from Yucatan swine (RVN1, RVN2, RVN3) and 1 from Yorkshire swine (RVN4). Two of these nerves (RVN1, RVN2) were used in both organ- and sensory/motor-fascicle micro-CT tracking studies.
In the swine histology/IHC study: 8 swine nerves were used in total (5 right nerves, 3 left nerves). All these nerves were different than the ones used in the micro-CT studies.
In the human micro-CT tracking and histology/IHC study: 1 left vagus nerve was used, of which only one segment that included the RLN was further analyzed (Suppl. Figs. S23-1; Suppl. video 2, also available at https://youtu.be/gacFJ-nV_G0), and eventually sectioned and IHC stained and imaged (human-LVN).
In swine physiology studies: a separate cohort of 12 animals was used, all Yucatan; 8 of those animals received right VNS, 4 received left VNS. The nerves from these animals were not used for imaging or histology/IHC studies.

2.1.2. Animal care during terminal surgeries

For a 12-h period before surgery, the animals were maintained on a no-food and no-fluid regimen (NPO). On the day of surgery, the animals were weighed and sedated with mixture of Ketamine (10–20 mg/kg) and Xylazine (2.2 mg/kg) or Telazol (2–4 mg/kg). Propofol (4–6 mg/kg, i.v.) was used to induce anesthesia, and following intubation, the anesthesia was maintained with isoflurane (1.5–3%, ventilation). Mechanical ventilation was provided and was turned off only at time breathing rate changes were measured during VNS. Animals were placed on a table in a supine position with normal body temperature maintained between 38 °C and 39 °C using a heated blanket and/or a hot air warmer. Blood pressure and blood oxygen level were monitored with a BP cuff and a pulse oximeter sensor. The depth of anesthesia was monitored by assessing heart rate, blood pressure, respiration, and mandibular jaw tone. All surgeries were performed using sterile techniques.

2.1.3. Animal care during survival surgeries

Pre-surgical care was identical as for terminal surgeries. After each survival surgery, a fentanyl patch (25–50 mcg/hr) and buprenorphine (0.01 mg/kg s.q.) were provided to alleviate postoperative pain. After gradually weaning off the ventilator and extubating, and when physiological signs returned to normal, animals were returned to the home cage and were monitored 3 times/day for 72 h. Cefazolin (25 mg/kg p.o.) was given for 5 days. Carprofen (2–5 mg/kg, s.q.) was administered when the animal showed signs of pain. Animals were allowed 10 days for full recovery before any physiology tests were performed.

2.2. Cervical vagus nerve and laryngeal muscle implants

Under anesthesia, a 4–5 cm long incision was cut down to the subcutaneous tissue on the right or left cervical region of the neck, 2 cm lateral to the midline. Subcutaneous tissue was dissected and retracted to expose the underlying muscle layers. The sternocleidomastoid muscle was bluntly separated and retracted laterally away from the surgical field. The carotid sheath was exposed, and the vagal trunk was identified and isolated away from the carotid artery. In all non-survival experiments (n = 12 animals), two cuffs, fabricated by Cortec GmBH (Germany) were implanted at the cervical level. One rostral cuff was implanted 2 cm from the nodose ganglion, used for fascicular VNS, and a second cuff was used for eCAP recording. There was 4–5 cm gap between the two cuffs. In chronic experiments (n = 2 animals), only one cuff was implanted at the cervical level for fascicular VNS. The vagal trunk between the two exposed sites with the cuffs was left intact within the carotid bundle, to minimize the extent of surgical manipulation and trauma to the nerve. For laryngeal muscle recording, after the hyoid bone was blunt dissected, Teflon-insulated single or multi-stranded stainless-steel wires were de-insulated at the tip for about 1 mm, and inserted in the thyroarytenoid laryngeal muscle with a needle. In survival surgeries, the incision was closed, the leads and connectors of the implanted cuff electrodes and EMG electrodes were tunneled subcutaneously from the surgical field in the neck area to the back of the neck and then to the top of the skull, where they were externalized. Several stainless-steel skull screws were implanted in the exposed area of the skull by drilling small burr holes. The connectors and the anchoring screws were secured to the skull with dental acrylic; finally, a custom-made titanium cylindrical case surrounding the connector was attached to the skull and the skull screws with acrylic cement, to protect the externalized connectors (Fig. 7 E, a).

Fig. 7. — (A) Example in a single animal of threshold intensities for different physiological responses, as determined for each of the 8 contacts in the multi-contact cuff electrode. (a) Threshold for HR response, (b) BR threshold, (c) laryngeal EMG threshold. Threshold was defined as the minimum intensity of a stimulus train (200 μs pulse width, 30 Hz) that produces a change of 5–10% in the corresponding physiological variable. The contact with minimum HR threshold is placed at 12 o'clock direction (origin of polar plot); the remaining plots are aligned to that orientation. In each polar plot, the origin represents the minimum threshold (e.g. 1500 μA for HR threshold, 1250 μA for BR threshold), and the outer circle the maximum threshold (e.g. 3000 μA for HR threshold, 2500 μA for BR threshold). Each point in the polar plots is the average determined from 3 to 5 stimulus trains delivered to each contact. The red line represents the resultant (vector sum) of thresholds across all eight contacts: its direction points towards the side of the nerve with the highest thresholds overall, i.e. away from the side that is most selective for that physiological response. (B) (a) Average ( ± SEM) normalized HR thresholds and resultant vectors in a total of 12 animals, 8 with right VNS and 4 with left VNS (all 12 animals had HR thresholds, 10 of those also had BR thresholds and 9 laryngeal EMG thresholds; 7 animals had all 3 thresholds determined on all contacts). Each blue vector represents the resultant of HR thresholds around the nerve in a single animal. The red vector represents the sum of all individual resultant vectors, and it points towards the side of the nerve with the largest heart rate threshold; the dashed line opposite to that, and the associated angle, represents the most cardiac selective direction. p-value represents statistical significance of repeated measures ANOVA (dependent variable: threshold, independent variables: animal id and contact angle (nested); Suppl. Table S4). (b) Average BR thresholds and resultant vectors from 10 animals. (c) Average laryngeal EMG thresholds and resultant vectors from 9 animals (C) Example in a single animal of eCAP responses and fiber amplitudes, elicited from different contacts. (a) Traces of eCAPs elicited by stimulation delivered to 2 different contacts located at the 12 and 6 o'clock directions, with the fast (red) fiber components highlighted. Polar plot of the amplitudes of fast fiber responses, ranging between minimum (origin) and maximum (outer circle). The contact eliciting the maximum fast fiber response is placed at 12 o'clock direction. The red vector represents the resultant vector of fast fiber amplitudes from all 8 contacts. (b) Same as (a), but for amplitudes of slow fiber responses. The 2 polar plots are aligned by the 12 o'clock contact. (D) (a) Average ( ± SEM) normalized eCAP response amplitudes for fast fibers (shaded trace) and resultant vectors (blue vectors) from 11 individual animals. Before averaging, individual polar plots were aligned to the contact associated with the maximum fast fiber response. The red vector represents the sum of all individual resultant vectors. p value represents statistical significance of repeated measures ANOVA (dependent variable: fiber response amplitude, independent variables: animal id, contact angle). (b) Same as (a), but for slow fibers. Plot is aligned by the contact associated with the maximum fast fiber response (12 o'clock in panel (a)). (E) Physiological responses elicited by fascicular VNS delivered for up to 3 weeks in chronically implanted swine (n = 2). (a) Schematic of the chronic vagus nerve implant in swine, involving the helix cuff. (b) Cough-reflex thresholds in two animals. (b1) Average ( ± SEM) normalized cough-reflex thresholds and the resultant vectors at 8, 12, and 16 days post-implantation (shown in respective colors) in animal K. Red vector represents the sum of all days, pointing towards the side of the nerve with the largest cough-reflex threshold; the dashed line opposite to that, and the associated angle, represents the most coughing selective direction. (b2) Same as (b1), but for Animal E.

2.3. Multi-contact cuff electrode device

2.3.1. Methodology to specify contact configuration in cuff electrode

To determine the number of contacts that could in principle reveal differential engagement of fibers around the vagal trunk of the swine, we calculated (estimated) the stimulation efficacy and selectivity of different contact configurations that could be expected when stimulating the cervical vagus nerve of the swine (Suppl. Fig. S33). This theoretical methodology was a rough approximation of the expected efficacy and selectivity and was by no means a thorough effort to anatomically and physiologically simulate the nerve, nerve fibers and the nerve-electrode interface, as has been done in other modeling studies [39,44,45]. As a result, the contact configuration at which we arrived with this methodology cannot be considered “optimal”; it is a configuration that could theoretically allow for some degree of stimulation efficacy and selectivity, enough to demonstrate differential engagement of fibers across fascicles in this multi-fascicular nerve. The methodology we followed comprised the following 5 steps.

Outlines of the epineurium of entire nerves and of the perineurium of each of its fascicles were extracted from H&E images of sections taken through the mid cervical region of 8 extracted vagus nerves. Each outline of a nerve or a fascicle was then converted to a matrix of x-y coordinates and fitted with an equivalent ellipse centered at the center of mass of the outline. The equivalent ellipse was calculated by iteratively determining the length of short and long axes and the angular rotation that minimized the difference between the area of the ellipse and the area of the nerve or fascicle outline. The ellipse surrounding the nerve outline roughly represents the epineural cuff.
Points were placed on the equivalent ellipse surrounding the nerve at regular intervals, representing individual electrode contacts on the epineural cuff (e.g. for 4 contacts, the angular separation between points was 90°) (Suppl. Fig. S33A). For each nerve outline, placement of points was repeated 100 times, with a random angular placement of the configuration on the perimeter of the ellipse.
Circles centered on each of the points representing contacts were drawn; each circle represents the electrical field generated by that contact, with a field strength that activates all fibers within the area of the circle. The radius of the circle, expressed as the fraction of the length of short axis of the nerve-equivalent ellipse, represents the strength of that electrical field. For example, a circle with a radius 0.2 means that all nerve fibers that lie within the area of a circle surrounding the point contact with a radius equal to 0.2 the short axis of the (equivalent eclipse of the) nerve, will be activated; Suppl. Fig. 33-A shows examples of a “low strength” field (0.2 radius) and a “high strength” field (0.5 radius), for 4- and 8-contact configurations.
The level of activation of fibers within a fascicle was taken to be dependent on the degree of overlap between the area of that fascicle and the circular fields generated by contacts in each electrode configuration. That spatial overlap was expressed as % of the fascicle area; if the overlap was >75% of the fascicle area, that fascicle was considered “activated”; if overall was <15%, the field was considered “spared”.
The calculated spatial overlap between fascicles and fields generated by a given contact configuration was used to estimate efficacy and selectivity of that configuration. Efficacy of a configuration for a given field strength was defined as the percentage of fascicles that were activated by all contacts in a configuration, averaged across all 100 random configuration placements on a nerve. Selectivity of a configuration for a given field strength was defined as the percentage of nerve fascicles that were activated by each 1 contact, averaged across all contacts in a configuration and across all 100 random configuration placements on a nerve (Suppl. Fig. S33B).

2.3.2. Electrode fabrication

The helical design of the cuff is based on a self-spiraling cuff, fabricated with a laser-machining method [46,47]. At first, a polymeric release layer was made on a ceramic substrate (Al2O3, 100 × 100 mm², thickness: 0.635 mm) through a mechanical carrier. Then, silicone rubber 1:1 n-heptane-diluted (Carl Roth GmbH + Co. KG, Karlsruhe, Germany) and MED-1000 (NuSil Technology LLC, Carpinteria, CA, US) is spin-coated on top, and followed by laser-structured using a picosecond Nd:YAG laser (HyperRapid NX, γ = 355 nm, Coherent, Inc., Santa Clara, CA, US). Platinum/Iridium (90/10) foil (thickness:25 μm, Goodfellow GmbH, Bad Nauheim, Germany) is laminated, and laser-cut and glued to the pre-stretched MED-1000 silicone rubber foil, which was laser cut.

The helical cuff includes 8 square contacts, based on theoretical considerations (Suppl. Fig. S33). Each contact has exposed dimensions 0.5 mm × 0.5 mm, evenly distributed around the circumference, with approximately 1 mm separation between 2 neighboring contacts; and 2 ring-shaped “return” contacts, each with exposed dimensions 0.5 mm × 7.4 mm, placed 1 mm above and 2 mm below the row of 8 square contacts (Fig. 5A).

Fig. 5. — (A) (a) The design of the helical cuff electrode. The cuff is comprised of 3 silicone loops that wrap around the nerve. The middle loop houses 8 small square contacts made of platinum iridium (500 μm side, 500 μm apart from each other) and each of the 2 outer loops houses 1 ring-shaped return contact (500 × 7400 μm). (b) Close up of the actual helix cuff. (c) Cuff placed on a swine cervical vagus nerve. (B) Electrochemical impedance spectroscopy shows the impedance magnitude and phase of a representative electrode across a measurement frequency range from 0.2 Hz to 100 KHz; impedance measured at 1 kHz is ~4 kΩ. The top inset shows a representative cyclic voltammetry trace used to characterize charge storage capacity (~4 mC/cm2), scan range from −0.6 to 0.8 V at rate of 100 mV/s. (C) Individual impedance values at 1 kHz (mean ± SEM), for each of the 8 square contacts in 3 cuff electrodes, measured after the first and sixth use in acute in vivo VNS experiments.

2.3.3. Electrode characterization

Cleaning of electrodes:

Prior to characterization, the electrodes were soaked in a 1% tergazyme solution and placed in an ultrasonic cleaner for 5 min. They were then allowed to soak in the tergazyme solution (room temperature) for at least 1 h, following which the solution was decanted. The electrodes were rinsed with DI water, then placed back in the ultrasonic cleaner for 5 additional minutes in DI water. The water was then decanted and the electrodes were rinsed once more before performing measurements.

Electrochemical impedance spectroscopy (EIS):

The EIS measurements were done using a 3-electrode set up. A graphite rod was used as a counter electrode along with an Ag/AgCl reference electrode. The electrode under test along with the reference electrode were placed in a 7.2 pH 1x PBS solution, then sonicated for 5 min to remove any air bubbles from around the cuff structure. They were then placed in a Faraday Cage, where the counter electrode was introduced into the solution. The EIS measurements of each electrode were taken separately, from a range of 100,000 Hz–0.2 Hz (Fig. 5B). The impedance for each contact of the array at 1 Khz was determined to be less than 10 KΩ, and the impedance of the return contacts was less than 1 KΩ. The charge storage capacity for each electrode ranged from 3000 to 5000 μC/cm². Each cuff could be used multiple times without significant changes in impedance (Fig. 5C).

Charge storage capacity (CSC):

The cyclic voltammetry (CV) plots were taken in a similar manner as the EIS measurements. The cuff electrode and reference electrode were again placed in the PBS solution and sonicated for 5 min to remove air bubbles, after which the counter electrode was introduced. The CV measurements were performed over 10 cycles and swept from −0.6 - 0.8 V at a scan rate of 100 mV/s. In calculating the charge storage capacity, the first and last sweep were disregarded, and the remaining sweeps were averaged and integrated below zero to yield the cathodic CSC.

2.4. Physiology experiments, vagus nerve stimulation and physiological data analysis

2.4.1. Physiological and neural signals

Breathing rate:

Breathing was monitored by using a respiratory belt transducer (TN1132/ST), which was placed around the animal's chest and connected to a bridge amplifier (FE221, ADI). The change of the belt tension with breathing movements represents the breathing cycle and was used to derive breathing rate. The mechanical ventilator was turned off when measuring the effects of VNS on breathing.

Heart rate:

The skin around the wrists was shaved and conductive gel was applied to patch electrodes which were subsequently adhered to the skin; ECG was recorded using a 3-lead patch electrode configuration and amplified using a commercial bio-amplifier (FE238, ADI), and was used to derive heart rate.

All physiological signals were digitized at 1 Ksamples/s (PowerLab 16/35, ADI); they were visualized in real-time using LabChart (ADInstruments Inc) and stored on a hard drive.

Neural and EMG signals:

Neural and EMG signals were collected through a second data acquisition system that included a headstage (RHS-32, Intan Tech) and a recording controller (Recording/stimulation Controller, Intan Tech), digitized at a sampling frequency of 30 Ksamples/s.

2.4.2. Physiological thresholds of VNS

Biphasic stimulus pulses were delivered using an isolated constant current stimulator (STG4008, Multichannel Systems) through each of the 8 contacts of the cuff electrode device, using both ring-shaped electrodes as stimulus returns. By splitting the stimulus return between the 2 ring-shaped contacts on both sides of the active contacts and by increasing the surface area of the return electrodes (2 instead of 1), the risk of fiber activation by electric fields generated at the stimulus return is reduced.

Physiological thresholds were determined using stimulus trains of 10 s durations (30 Hz, pulse width 200 μs). There was at least 30 s-long interval between successive trains to ensure that physiological parameters had reached a steady state before a new train was delivered. We defined as threshold the lowest stimulation intensity that induced a 5–10% drop in heart rate (HR threshold), ~50% drop in breathing rate (BR threshold), or a visible response in laryngeal EMG (EMG threshold). The 50% drop in BR was chosen because of the sharp, sigmoid relationship between intensity and drop in BR, documented in Ref. [45]. In experiments in awake animals we measured the threshold for evoking cough reflex, as the lowest stimulation intensity to induce coughing. In those experiments, we also measured changes in heart rate at cough reflex threshold intensities. The thresholds for each of the 8 contacts were determined in random order.

2.4.3. Evoked compound action potentials (eCAPs)

After removing DC and very low frequency components from neural recordings, with a high-pass filter at 1 Hz, stimulus-evoked compound action potentials (eCAP) were compiled by averaging voltage sweeps triggered from all stimulus pulses within a train. The peak-to-peak amplitude of eCAP components within predefined temporal windows was used to quantify fiber activation; those windows were chosen on the basis of conduction velocities of different fiber types and the distance between the stimulating and recording electrodes. For a (typical) distance between recording and stimulation electrode of 45–60 mm, the earliest component, typically occurring within 1.5 ms from the onset of the stimulus, was classified as fast fiber response, consistent with activation of large, myelinated A-type fibers (conduction speed >33 m/s); the component occurring at 2 ms or later was classified as slow fiber response, consistent with activation of smaller, myelinated fibers (Aδ- and B-type) (conduction speed, <20 m/s). Contamination of eCAPs from EMG from stimulus-elicited muscle contraction was minimal, as confirmed in experiments in which eCAPs were recorded before and after administration of vecuronium, a muscle relaxant (Fig. 6A, e, f). Long latency C-fiber responses were not observed in our experiments. We registered eCAP responses from all eight contacts of the recording cuff, when stimulating each of the eight contacts of the stimulating cuff; however, because of the high degree of similarity between the 8 individual eCAP traces elicited from the same stimulated contact (Suppl. Fig. S22), and in order to minimize the contribution of anatomical variability at the level of the recording cuff (discussed in detail in Discussion-Section 3), eCAP traces were analyzed after averaging across all 8 recording contacts.

Fig. 6. — (A) (a) Examples of cardiovascular, respiratory and laryngeal physiological responses to a stimulus train (shaded area) through one of the 8 contacts (contact 5) include a drop in heart rate (HR), a moderate drop in mean blood pressure (BP) and slowing of breathing rate (BR); at the same time, minimal evoked laryngeal muscle activity (EMG) and strong evoked responses in the nerve potential (ENG) were observed. (b) Example responses to the same stimulus train delivered through a different contact (contact 1) included minimal drop in HR, BP and BR but strong laryngeal EMG. The red line indicates the mean cycle rate from the cycling raw data. The two sketches in (a) and (b) indicate the relative position of the tested contacts around the nerve. (c) Evoked laryngeal EMG (top) and compound nerve action potential (eCAP, bottom) triggered by stimuli in the train delivered to contact 5. Small EMG is generated and the eCAP shows a response at a latency consistent with slow A-fiber activation. (d) EMG and eCAP triggered by stimuli delivered to contact 1; EMG response and a short latency eCAP, consistent with fast A-fiber activation, are observed. (e) Nerve eCAP is not affected by neuromuscular blocker vecuronium. (f) EMG of laryngeal muscle is blocked by neuromuscular blocker. (B) Example physiological responses and eCAPs mediating them have different recruitment curves from different contacts. (a) Amplitude of laryngeal EMG is shown for different stimulation contacts (abscissa) and at different intensities (ordinate), represented by the color scale. (b) Same for amplitude of fast fiber response in the eCAP; fast fibers (e.g. Aα) mediate efferent signaling to laryngeal muscles. (c) Same for the magnitude of breathing response, quantified as % change in BR. (d) Same for amplitude of slow fiber response in the eCAP; slow fibers (e.g., Aδ [64]) mediate afferent signaling to the brain that, through a reflex mechanism (Herring-Breuer r4.3 Responses to fascicular VNS occur asymmetrically around the vagal trunk.

2.4.4. Analyses of physiological responses across contacts and animals

Recruitment maps of physiological and eCAP responses.

To demonstrate that physiological responses to VNS delivered through specific electrode contacts correlate with eCAP components from respective fiber types, we calculated the magnitude of physiological responses, including peak-to-trough evoked EMG and breathing rate change, as well as the amplitude of fast and slow components of eCAPs, in response to increasing stimulus intensities through each of the 8 contacts. The magnitude of physiological responses and amplitude of eCAP components were then plotted against the contact number and the stimulus intensity, in respective recruitment maps (Fig. 6B; Suppl. Fig. S21).

Statistical assessment of asymmetry of responses to fascicular VNS.

To test whether responses to fascicular VNS are asymmetric (i.e., they depend on contact location), intensity thresholds for different organs (heart rate, breathing rate, laryngeal EMG) were determined separately for each contact; heart rate thresholds were determined in 12 animals, breathing rate thresholds in 10 animals, laryngeal EMG thresholds in 9 animals, fast eCAPs in 11 animals, slow eCAPs in 10 animals; in 7 animals, all three types of physiological thresholds were determined. A separate repeated measures ANOVA model was used to test whether contact location affects each of the physiological thresholds: respective threshold was the dependent variable, animal id and (nested) contact location were the categorical independent variables; p < 0.05 in each ANOVA test was considered a statistically significant result. Similar procedures were followed for the 2 eCAP components (fast and slow). To test whether the spatial distribution of responses to fascicular VNS is different for different organs, normalized thresholds determined on the same contact were pairwise compared using several paired t-test comparisons; p < 0.05, Bonferroni-corrected for multiple comparisons, in each paired t-test was considered a statistically significant result. Normalization of thresholds was performed by dividing the absolute value of the threshold by the respective minimum threshold value among all contacts in that animal. Similar procedures were followed for the fast and slow eCAP components.

Visualization of asymmetry of responses to fascicular VNS.

To visualize the asymmetry of organ-specific physiological responses to stimulation delivered through the 8-contact cuff, physiological thresholds for the 8 contacts were plotted on a polar plot. Organ-specific thresholds were normalized between the lowest value (set to 0) and highest value (set to 1) for each animal. Thresholds were aligned on the polar plots with respect to the contact with the minimal heart rate threshold (i.e., aligned to the most “heart-specific” contact), which was placed at 12 o'clock (90°). The resultant vector was then calculated in each animal, for each type of threshold (Fig. 7). The vector sum of the resultant vectors from all animals was used to estimate the overall “preferred” angular direction for each type of threshold. Similarly, eCAP magnitude plots were aligned by placing the contact with the highest response (peak-to-trough amplitude) of the fast eCAP component at the 12 o'clock position; resultant vectors and vector sums were then calculated as described above.

2.5. Surgical dissection and micro-computed tomography of nerve samples

Branches emerging from the cervical and thoracic vagus nerve provide sensory and motor innervation to visceral and internal organs [4], including the larynx and pharynx, through the superior and recurrent laryngeal nerves [48], the heart, through the cardiac branch [4,48,49], and the lower airways and lungs, through the bronchopulmonary branches [50]. To understand the spatial arrangement of fascicles in the vagal trunk with respect to the formation of organ-specific branches, we performed high resolution micro-computed tomography (micro-CT) imaging of the trunk, from above the nodose ganglion, rostrally, to the lower thoracic portion of the nerve, caudally. Fascicles with any visible contribution to the formation of vagal branches were manually annotated at individual micro-CT sections taken through the emergence of each of the branches (Suppl. Figs. S2, S3, S4) and were considered to project to the corresponding organ (Fig. 1A). The longitudinal trajectories of those fascicles were then tracked along the length of the vagus nerve rostrally with the use of a user-assisted, semi-automated, deep learning-based, image segmentation algorithm.

Fig. 1. — (A) Schematic to provide a visual of a segment of the vagal trunk, demonstrating 3 branches emerging from it (bottom to top): bronchopulmonary (BP, yellow), recurrent laryngeal nerve (RLN, red), cardiac branch (blue). The trajectories of fascicles contributing to each branch are traced rostrally inside the trunk, even after they merge with other fascicles to form “mixed” fascicles: BP and RLN (orange), BP, RLN and cardiac (green). (B) (left) Macroscopic image of an intact right vagal trunk, from the nodose ganglion to the lower thoracic level (left), with the 3 major branches labeled; several levels along the trunk are identified with lower-case letters (a: above nodose, w: BP emergence). (right) Multiple micro-computed tomography images of the vagal trunk taken through the identified levels (panels a–w). Fascicles contributing to each of the identified branches or mixed fascicles are shown shaded in the corresponding colors. (C) Macroscopic and organ-specific fascicular structure of an intact left vagal trunk from a second animal.

Surgical dissection of swine nerve samples.

Animals were euthanized by injection of Euthasol (1 ml/10 pounds BW, i.v.); death was confirmed using ECG and absence of arterial pulse. After euthanasia, both the left and right cervical vagus nerve were dissected from above the nodose ganglion to the end of the thoracic vagus; during that time, nerve branches, still attached to the nerve trunk, were isolated using blunt dissection up to the respective end organ (heart, lung or larynx). A fine suture loop (6-0) was placed on the epineurium of each branch, close to its emergence from the trunk, to label that branch and maintain a record of the innervated organ during subsequent imaging studies, as the sutures used are radiopaque and visible in micro-CT images. The nerve trunk, along with organ-specific labels was photographed before and after extraction (Fig. 1; Suppl. Figs. S2 and S3). The samples were fixed in 10% formalin for 24 h, then transferred to Lugol's solution (Sigma, L6146) for five days to achieve optimal fascicular contrast for the micro-CT scan, similar to a previous study [10].

Micro-CT imaging of swine nerve samples.

Each nerve trunk was sectioned into several 6 cm-long segments and the rostral end of each segment was marked with a suture knot, to maintain the rostral-caudal direction. Each nerve segment was scanned individually in the micro-CT scanner. Nerve segments were mounted in position on a vertical sample holder tube. The samples were scanned using Bruker micro-CT Skyscan 1172 with a voxel size of 6 μm. Volume rendering was done using Bruker CTvox v3.3.1 to obtain some 3D views of the nerve (Suppl. Figs. S2-S4).

Micro-CT video rendering.

Micro-CT images are 2D images (Suppl. Fig. S29). Those 2D images show fascicles and connective tissue around them in shades that remain relatively uniform across the length of the nerve. Videos were rendered using z-stack methodology to improve 3D perspective. Because 3D images are rendered by post processing of 2D images, the uniformity of shade present in 2D images is reduced. Further contrast, brightness, tissue density and lightning parameters are changed in the 3D rendered images to obtain a better signal to noise ratio, so that fascicles can be better visualized. For these reasons, there is variation in white to black balance across the nerve. Each fascicle was manually identified as fascicle by tracing it to a minimum of 1 cm in both rostral and caudal direction. Connective tissue sections that appeared circular were not confused with fascicles, as they usually appear or disappear abruptly in consecutive frames, without any continuity. These circular shapes or pseudo-fascicles were not included in our analysis.

Surgical dissection and micro-CT imaging of a human nerve sample:

An embalmed human cadaver was obtained through Donald and Barbara Zucker School of Medicine at Hofstra/Northwell. Dissection at the left side of the neck was performed to expose the carotid sheath. Further dissection was performed keeping the carotid sheath intact to expose the left vagus nerve starting from the cervical to the thoracic inlet region (Supplementary Figs. S23-1 & 2). Vagal branches including RLN, esophageal plexus and cardiac plexus were identified and marked with sutures knots. The left vagus nerve was then extracted, starting from the cervical region to the thoracic inlet region along with identified branches. The nerve was post fixed in 10% formalin for a week. A single, 5 cm-long segment of the nerve containing the RLN branch (Suppl. Fig. S23-1) was cut and was transferred to Lugol's solution (Sigma, L6146) for 5 days to achieve optimal fascicular contrast for the micro-CT scan. The nerve sample was wrapped in a less dense material (e.g., saran wrap) to prevent dehydration; micro-CT imaging and fascicle tracking was performed similarly to the swine vagus nerve, as described above.

2.6. Segmentation and tracking of fascicles in micro-CT data

For segmentation and tracking of fascicles in the micro-CT data we followed a manually-assisted, semi-automated approach.

2.6.1. Segmentation of fascicles

Instance segmentation on micro-CT images was performed using the Matlab Deep Learning toolbox, Mask R–CNN model implementation (Mathworks Inc.). The mask R–CNN model [51] comes pretrained with the COCO image dataset [52]. Transfer learning was applied by freezing the Resnet 50-coco backbone during training, forcing the model to utilize the existing COCO features from the Resnet50 subnetwork, while updating the weights of the region proposal network and mask head [51].

Micro-CT images are significantly different from COCO images (Suppl. Fig. S29): for example, in grayscale intensity images, as opposed to color photographs, content extends over a much smaller range of shades. To achieve adequate performance, initial training with manually annotated images and subsequent supervised training was required. At a first stage, all images from the first scanned nerve were manually annotated. About 80% of those manually annotated images were included in the training set; the remaining 20% were held out for testing the mask R–CNN model on the first scanned nerve. At a second stage, the mask R–CNN model was updated for each new nerve sample with a smaller number of manually annotated images: on subsequent nerves, only 1 out of every 50 micro-CT images was manually annotated, whereas 49 out of 50 were automatically annotated. The automatically annotated images included 3 categories: fascicle, false alarm on nerve, and false alarm off nerve. The false alarm classes were created by storing the segmentation results that didn't overlap with the ground truth annotations and then subcategorizing them based on the average pixel intensity within the detection masks.

In both stages, we chose an initial learning rate during training, γ₀ = 1.2 * 10⁻⁵. The overall learning rate decreased with each epoch based on the formula γ = γ₀/(1 + τ(epoch-1)), where τ = 0.01 is a decay rate. The model was trained for 18 epochs resulting in a total number of training iterations which is on par with the number of training iterations used to train the original Mask R–CNN model [51]; stochastic gradient descent with momentum was used to update the network weights with a momentum of 0.9 and a minibatch size of 1. We chose a mask R–CNN model as our detector because its threshold can be tuned to perform well on the detection task, and can readily be updated in individual animals if necessary. Tuning the mask R–CNN threshold with a high sensitivity and relatively low specificity allows the subsequent tracking stage to increase the specificity while still maintaining a relatively high sensitivity. Shape-based, pixel-based, and texture-based features were extracted for each detected fascicular cross-section. Shape-based features consist of the centroid, circularity, eccentricity, irregularity, area, orientation, major and minor axes lengths, solidity, and the weighted centroid [53]. Pixel-based features consisted of the mean pixel intensity and central moments like variance, skewness, and kurtosis [53]. Texture-based features were derived from the gray-scale co-occurrence matrix and include contrast, correlation, energy, and homogeneity [54]. A minimum area constraint of 120 pixels in total was imposed based on the smallest area in the annotated data.

2.6.2. Tracking of fascicles

After fascicles were detected, segmented and annotated in each slice of micro-CT data, fascicles were tracked longitudinally along the length of the nerve. The tracker filters and structures detections to construct a graph that captures properties of the fascicular cross-sections (shape and texture metrics) at each node. We used a windowed, density-based-clustering approach. The distances between the detected fascicles within the micro-CT slice window were computed as 1 – J, where J is the Jaccard similarity: Jaccard similarity was originally used to quantify similarity between flora species in different districts, and it calculates the intersection of 2 sets over their union [55]. A detected fascicle that is constant from one slice to the next will have a Jaccard similarity of 1 and a distance of 0. At the other extreme, completely nonoverlapping fascicles in adjacent slices will have a Jaccard similarity of 0 and a distance of 1. Partially overlapping fascicles on adjacent slices will have a distance between 0 and 1. DBSCAN was used to cluster the detected fascicular cross-sections at each window position [56], with a minimum of 3 samples per cluster and an epsilon neighborhood of 0.6. A track was extended if more than 2/3 of the detections were the same between the adjacent window positions since some jitter from the clustering algorithm is expected as the sliding window is shifted. Any detections that were not included in a track were discarded, except for detections that correspond to manual annotations.

The final algorithm-tracked trajectories were manually validated to confirm that no fascicle tracks were missed or discarded erroneously. During this manual track validation step, any instances of tracks that end or emerge abruptly without an obvious split/merge event were reviewed and corrected at the corresponding cross-sections and the tracking step was rerun. Because of the irregular fascicle shapes at the level of branch emergence, individual fascicles contributing to a vagal branch of interest were manually identified in 5–10 sections taken through and rostrally to the level of emergence (Suppl. Figs. S2-S4); afterwards, fascicle trajectories were tracked as described previously.

The fascicle segmentation algorithm can be downloaded at https://github.com/tlevy-nds/fascicleMaskRCNN.git.

2.7. Quantitative immunohistochemistry of nerve samples

2.7.1. Sectioning of samples

A total of 8 nerves were used in the IHC studies (Fig. 3 and Suppl Fig. S17): N = 5 right, and N = 3 left vagus nerve. From all 8 nerves, a 1 cm-long nerve segment was paraffin sectioned at the mid cervical region, approximately 5 cm caudal to the nodose ganglion. A total of the 200 sections were taken from each of the segments, each section was 5 μm thick. Out of the 200 sections, about 25 sections were used for the IHC study. Two replicates of 5 sections were used for the each of the combinations of the triple stain (for example NF, MBP and ChAT). Out of those 2 replicates of 5 sections, in one section, all fascicles were imaged at 100× magnification. Two of the 5 right vagus nerves were sectioned at 3 different levels: first at nodose ganglion, second at mid-cervical level, 5 cm from nodose ganglion, and third at the thoracic level, 20 cm from nodose ganglion.

The same human left vagus nerve segment used for the micro-CT was also used for the IHC study. The nerve was rinsed with phosphate buffer solution (PBS) 5 times to clear off the Lugol's stain. The nerve was then processed and sectioned using the same technique as the swine nerve.

2.7.2. Staining and imaging of sections

Sections were stained for myelin basic protein (MBP), neurofilament (NF), choline acetyltransferase (ChAT), tyrosine hydroxylase (TH) or voltage-gated sodium channels 1.8 (Nav1.8) using standard IHC protocols [57]. Briefly, sections were subjected to deparaffinization procedure using xylene and ethanol rinse and then washed with distilled water for 10 min. Antigen retrieval was performed by briefly subjecting the samples to 1 x citrate buffer to unmask the antigens. Sections were rinsed with 1x Tris-buffered saline and subjected to 1 h of blocking step using one normal goat serum. Sections were incubated with anti-NF (1:500, Abcam AB8135), anti-MBP (1:500, Abcam AB7349) and either anti-ChAT (Millipore Sigma, AB144P) or anti-TH (Abcam, ab112 (“efferent IHC panel”) or SCN10A monoclonal antibody for swine (Nav1.8, MA5-27663 Fisher) and SCN10A polyclonal antibody for humans (Alomone labs, ASC-016) (“afferent IHC panel”), overnight at 4 °C in a shaking humidifier. The following day, sections were rinsed and incubated with the secondary antibody for 2 h at room temperature. Slides were then rinsed thoroughly with TBS buffer three times, and cover glass was mounted on the sides with the fluoromount (Vector labs, H-1700). The slides were then imaged using ZEISS LSM 900, confocal laser scanning microscope, and BZ-X800 all-in-one fluorescence microscope at 100× magnification. The human nerve sample was stained and imaged using the same technique, except for the Nav1.8 staining, for which a different polyclonal antibody was used (Alomone labs, ASC-016).

One of the advantages of the immunofluorescence method is its ability to use fluorescence markers of varying emission spectra to identify multiple markers in the same cell. For example, here we used 3 markers: blue for NF, red for MBP and green for ChAT. Addition of a fourth or fifth marker would result in emission overlap, making it hard to discern the difference between markers. Thus, we were restricted to a maximum of 3 different stains per section. In this study NF and MBP was kept as a standard for the all sections and the third stain was either ChAT, Nav1.8 or TH. Minimum of 12 adjacent replicate sections were used for each of the combinations of triple staining (e.g., NF, MBB and ChAT or NF, MBP or Nav1.8). Out of 12 sections, one of the sections was imaged at 100x for detailed fiber analysis. Also, a separate set of 12 adjacent sections were used for H&E staining.

2.7.3. Quality controls and validation of IHC

To ensure the quality of IHC and validate the specificity of the antibodies, we undertook the following quality control studies.

Transgenic experiments using pigs are not feasible, hence we used transgenic mice to validate antibody specificity. In our study, we used transgenic mice that express enhanced fluorescent protein (EYFP) with ChAT expression as a positive control. Vagus nerves from these transgenic animals were used to validate binding specificity of the ChAT antibody with the same specific antibody used in the pigs (Suppl. Fig. S19) and also to define the ChAT staining pattern of single fibers, information that was used to train the segmentation and classification algorithm.
By omitting the primary (ChAT) antibody run and using only the secondary antibody in tandem sections as negative control, we observed no nonspecific binding in tissue.
We studied the localization of antigens to well-characterized specific locations in the cell to confirm the validity of binding. Antibodies against NF, MBP, ChAT and Nav 1.8 are well characterized with regard to their binding location in axons; our results are consistent with those studies ([1,5,[61]]). We found that NAV1.8+ fibers have a punctate expression pattern. To confirm that that staining pattern represents C-fibers and not merely precipitation of the Nav1.8 antibodies, we double-labeled the sections with Schwann cells marker (S100) and Nav1.8. We found that Schwann cells engulf Nav1.8 positive fiber clusters (Fig. 3B). Unmyelinated small diameter C-fibers clusters are known to be ensheathed by nonmyelinating Schwann cells called Remak Schwann cells [59,60]; Nav1.8 is known to have a clustered distribution along the length of the axons [58] both consistent with our current findings. Together, our results indicate that the anti-Nav1.8 antibody used in this study selectively stain C-fibers.
Higher or lower dilution factors of the antibody may result in false positive or false negative results. To address this, we diluted the antibody to the optimal point just enough to detect the antigen of interest leading to minimal background staining. Further during imaging, exposure levels were kept at the optimal levels to obtain higher signal to noise ratio.
BLAST homology analysis: To identify unmyelinated afferent fibers in swine tissue, we used a monoclonal anti-Nav1.8 antibody, which, compared with polyclonal antibodies, binds with higher affinity and specificity to its antigen. However, this antibody failed to identify afferent fibers in human vagus nerve sections. Therefore, we tested a polyclonal anti-Nav1.8 antibody and our preliminary results show comparable staining of afferent fibers as seen in pig nerve sections. To further validate the new polyclonal antibody, we ran a BLAST analysis of the amino acid sequence of the rat antigen against which it was raised (amino acids 1724–1956: ENFNVA-TEESTEPLSEDDFDMFYETWEKFDPEA) and found that it had 100% identity match with human Nav1.8 (several isoforms).

2.8. Fiber segmentation and feature extraction from IHC images

We developed a set of algorithms and a data organization and analysis methodology to segment and extract nerve, fascicle and single-axon level features from 2-D IHC images of the vagus nerve. First, we developed a fiber segmentation and feature extraction algorithm by combining standard computer vision methods with statistical modeling and geometrical manipulations, including Gaussian mixture models and Voronoi tessellation, and achieved a single-axon detection accuracy of 82% and myelination classification accuracy of 89% [61].

To improve the detection accuracy, we further trained a Mask R–CNN deep convolutional neural network on axon instance segmentation. The model was pretrained on the COCO Instance Segmentation task [62]. We generated training images by first annotating fibers using the more standard algorithm, followed by manual correction of 80 fascicles, done in an ordinary image editing software. Fiber segmentation and annotation was done in a total of 701 fascicles. We estimated that pre-annotating using the original algorithm reduced the time of correction by four times compared to manual annotation from scratch.

2.8.1. Single fiber identification and characterization

Neurofilament:

To detect neurofilament positive pixels, we empirically selected a threshold brightness in the neurofilament color channel (154 out of 256) above which pixels were classified as positive. We further discovered connected components of such pixels to arrive at “blobs” of neurofilament positive staining. Blobs containing less than 10 pixels were excluded as noise. The remaining blobs were used to estimate counts of Nav1.8+ fibers, as described in detail below.

Myelin:

Myelin positive pixels were detected using the same technique as described previously [61]. Briefly, myelin channel pixels surrounding each detected neurofilament positive axon were analyzed. Pixels were classified as myelin positive when their brightness reached 128 out of 256. First, the outside pixel shell of neurofilament positive fibers were taken and neighboring pixels were recursively checked (for a maximum of 5 recursive steps) for myelin positive pixels. If more than 30% of fiber neighboring pixels were found to be myelin positive, the said fiber was classified as myelinated.

ChAT:

Similar to neurofilament and myelin pixel classification, we empirically selected a brightness threshold, applied to the chat color channel, of 131 out of 256. ChAT + pixels overlapping myelinated axons were counted separately for each fiber and if more than 5% of pixels were found to be chat positive, we regarded the fiber as chat positive.

The final set of segmentations was performed by the combination of the standard computer vision and deep learning algorithms: we let Mask R–CNN detect axons first, then opted for traditional methods to detect and divide overlapping neurofilament blobs that were ignored by the deep learning model. This combination achieved a fiber detection accuracy of 96% [63].

After the instance segmentation of neurofilament and the assignment of myelin to detected fibers, we extracted a set of features for each axon. These features include the longer and shorter diameters, the perimeter, and area of the fiber, in addition to the thickness of the myelin, and the polar coordinates of the fiber relative to the center of mass of the enclosing fascicle.

Fascicles:

We manually annotated nerve cross-sections on the fascicle level to derive polar coordinates of fascicles and their distance from the epineurium. Moreover, we extracted fascicle-level features including the perimeter of the perineurium, the area of the fascicle enclosed by the perineurium, and the total count and area of each fiber type within a fascicle.

Nav1.8:

To generate ground truth data for the deep learning instance segmentation network, we performed manual annotation of Nav1.8-stained images to determine the density of C-fibers within Remak bundles. For this study, due to the low resolution of light microscopy (1 μm), we used Airyscan-equipped super high resolution confocal microscopy images of portions of fascicles, in which single Nav1.8+ C-fibers with diameters <1 μm can be identified (Suppl. Fig. S20, a). High-resolution images taken from 4 randomly selected fascicles in 2 nerves, each from a different animal, were manually annotated and single Nav1.8+ fibers contained within more than 300 bundles were manually counted; the areas of the same Remak bundles were also manually measured. Using single Remak bundle areas and fiber counts, we compiled a linear equation that allowed us to estimate an approximate Nav1.8+ fiber count from a known Remak bundle area (Suppl. Figs. S20, c, d). We used that equation to estimate Nav1.8+ fiber counts from thousands of Remak bundles in lower resolution, light microscopy images, of which areas could be automatically determined (Suppl. Fig, S20, e). As the linear equation outputs are real numbers, we round those numbers to the closest integer for each detected bundle to get actual counts.

All images are pre-processed before the segmentation to enhance contrast and to remove noise. MBP–NF–ChAT stained images were first adjusted by adaptive histogram equalization, by dividing each fascicle image into a 32-by-32 grid, then color-clipped, by defining upper and lower threshold of the clip for separately each cross section manually. The thresholds for each cross-section and channel were selected to remove background noise and to render the stains close to binary (distinct foreground and background), which removed the brightness variance between cross-sections allowing us to use the same parameters for all cross-sections in the subsequent segmentation process. MBP–NF–Nav1.8 images were not processed by adaptive histogram equalization, but were color clipped as described above. In the subsequent segmentation process, we ignore all individual neurofilament and Nav1.8 blobs that are smaller than 0.225 μm², which we found to be background noise rather than actual fibers.

Each detected and characterized fiber was classified, using a set of criteria, into one of 4 morphologically- and functionally-distinct types: myelinated efferents (MEff), myelinated afferents (MAff), unmyelinated efferents (UEff) and unmyelinated afferents (UAff). The criteria are explained in Table 1, Suppl. Table 3, Suppl. Fig. S28, and discussed in detail in Discussion section 2.

Table 1.

Fiber types in the cervical vagus nerve of the swine, fiber counts and areas they occupy. Different fiber types (rows 1–8) according to criteria used in our pipeline (A) and corresponding fiber types according to Erlanger-Gasser classification (B). (C) Nerve side (left or right). (D) Percentage of fiber-covered area in the entire nerve corresponding to a given fiber type; shown is mean and range across 8 nerves. (E) Absolute fiber count of given fiber type in the entire nerve. (F) Percentage of given fiber type among all fibers in the whole nerve. (G) Percentage of area covered by given fiber type in single fascicles; mean [range] across all fascicles, of all nerves. (H) Absolute count of given fiber type in single fascicles. (I) Fiber type-specific features used in our pipeline to classify single fibers: positive (+) or negative (−) in NF (neurofilament) stain, MBP (myelin basic protein) stain, ChAT (choline-acetyl-transferase) stain, and Nav1.8 stain. Measured diameters for each fiber type (in μm, mean and interquartile range).

	A	B	C	D	E	F	G	H	I
	Fiber type	Erlanger Gasser class.	Side	% Fiber area in nerve, mean [range]	Fiber count in nerve, mean [range]	% Fiber count in nerve, mean [range]	% Fiber area in fascicle, mean [range]	Fiber count in fascicle, mean [range]	Fiber type features
	Fiber type	Erlanger Gasser class.	Side	% Fiber area in nerve, mean [range]	Fiber count in nerve, mean [range]	% Fiber count in nerve, mean [range]	% Fiber area in fascicle, mean [range]	Fiber count in fascicle, mean [range]	NF	MBP	ChAT	Nav 1.8	Diameter (μm), mean (IQR)
1 Myel. eff. (MEff)		Aα, Aγ, B	L	33 [28–35]	14.8K [12.7–16.6]	4.2 [2.7–6.4]	29 [0–83]	335 [0–2419]	+	+	+	NA	3.47 (1.46)
1 Myel. eff. (MEff)			R	31 [15–43]	14.6K [4.9–21.4]	5.1 [0.7–8.7]	29 [0–83]	403 [0–1879]
2 >10 μm		Aα	L	1.9 [0.4–4.6]	52 [12–130]	0.01 [0–0.03]	1.25 [0–27]	1.2 [0–33]	+	+	+	NA	15.7 (5.93)
			R	1.8 [0.5–4.5]	54 [12–127]	0.01 [0–0.02]	1.20 [0–19]	1.5 [0–21]
3 <3 μm		B	L	15 [12–20]	11.6K [9.5–13.5]	3.3 [2.0–5.2]	13.8 [0–48]	14 [0–48]	+	+	+	NA	3.06 (1.05)
			R	16 [4–27]	11.8K [3.4–18.7]	4.2 [0.5–7.6]	14.5 [0–57]	32 [0–57]
4 Myel. aff. (MAff)		Aβ, Aδ	L	52 [50–56]	34.0K [29.7–37.6]	9.1 [7.8–11.2]	54 [0–86]	766 [0–2753]	+	+	–	NA	3.10 (1.42)
4 Myel. aff. (MAff)			R	58 [48–74]	28.7K [25.8–31.4]	8.7 [5.2–10.5]	59 [12–92]	790 [20–3401]
5 >5 μm		Aβ	L	3.6 [1.2–5.7]	409 [163–623]	0.10 [0.06–0.14]	3.1 [0–20]	9 [0–56]	+	+	–	NA	8.49 (2.98)
			R	14 [1.7–36]	1348 [178–3109]	0.30 [0.02–0.59]	10 [0–54]	37 [0–510]
6 2–5 μm		Aδ	L	40 [37–42]	20.4K [17.7–22.4]	5.5 [4.7–6.8]	41 [0–72]	41 [0–72]	+	+	–	NA	3.72 (1.21)
			R	37 [33–42]	17.6K [13.9–21.2]	5.3 [3.2–7.1]	40 [0–83]	48 [0–84]
7 Unmyel. eff. (UEff)		C-Eff.	L	6.9 [5.1–8.8]	21.8K [15.1–25.7]	6.3 [3.1–9.8]	8.9 [0.6–54]	492 [9–3084]	+	–	–	NA	1.83 (0.78)
7 Unmyel. eff. (UEff)			R	4.7 [2.9–7.4]	15.8K [10.8–23,3]	4.9 [2.4–7.8]	6.6 [0.7–35]	434 [26–1626]
8 Unmyel. aff (UAff)		C-Aff.	L	7.3 [5.1 – 10.2]	315K [190–414]	80 [72–86]	7.2 [1–28]	7.0K [0.3–63.0]	–	–	NA	+	0.55 (0.32)
8 Unmyel. aff (UAff)			R	6.3 [5.1–7.9]	294K [186–470]	81 [75–91]	5.1 [1 – 16]	8.0K [0.6–78.0]

Open in a new tab

2.8.2. Population analyses of single fiber data

Once we detected, localized and characterized fibers and fascicles, we performed subsequent analyses to map the distributions of different fiber types within single fascicles, and across fascicles within the whole nerve.

Fiber counts and areas occupied.

We calculated within-fascicle counts and areas of the 4 fiber types, MEff, MAff, UEff and UAff. The per-fascicle fiber counts or areas occupied by those fibers are either presented in their absolute values (Fig. 3A, I-L) or as percentages normalized by the total fiber count (or area occupied by all fibers) per fascicle (Fig. 3C).

Spatial distributions of fibers within fascicles.

For each fiber type we estimated the ratio of its neighboring fiber types in a 15 μm radius (Fig. 3E, c). For each fiber in every fascicle at the mid-cervical level, we captured its neighboring fibers using k-nearest neighbors, then filtered out the ones outside of the 15 μm radius of the fiber. We accumulated the counts of neighboring fiber types for each center fiber type; normalized by the total count of neighbors we calculated the overall ratio of a fibers a given type being in the vicinity of a given fiber type. To further describe the tendency of different fiber types to share space, we compared their total numbers in each fascicle (Fig. 3D, a & b). Within-fascicle mixing was analyzed by splitting each mid-cervical fascicle into a grid of 30-by-30 μm non-overlapping spatial sectors, establish the fiber counts by type in the sectors, then finally count the number of sectors of a certain composition and represent the propensity of two fiber types to share a space of 30-μm-width fascicle area by showing a 2D histogram of the distribution of compositions of the two fiber types (Fig. 3E, a & b).

Distribution of fibers at different distances from the epineurium.

To characterize the spatial distribution of fiber types at different distances from the epineurium at the mid-cervical level, we calculated the percentage of fibers of every type located within concentric rings of 100-μm width, at different distances from the epineurium. We then displayed the distribution of fiber types as the function of epineurium distance combining data extracted from nerves of different animals (Suppl. Fig. S14, A). We also present the same information for individual nerves as cumulative distributions (Suppl. Fig. S15). We grouped fascicles according to their effective diameter into 50 -μm bins and measured the percent area occupied by each fiber type within each fascicle normalized by the area of the fascicle. We then computed the average and standard deviation of such fiber type percentages per effective diameter bin (Suppl. Fig. S14, B).

Manual validations.

As we manually annotated single fibers of MEff, MAff, UEff and UAff to generate ground truth data for our instance segmentation and NAV1.8 positive C-fiber counting model, we needed access to accurate single-fiber features. To gauge the size of the different fiber types, we measured the effective diameter of the manually annotated single fibers of each type and presented their distribution (Suppl. Fig. S13).

3. Results

3.1. Fascicles form organ-specific clusters along the vagus nerve

The vagus nerve in swine is multi-fascicular, with 40-50 fascicles of varying sizes in the cervical region (Suppl. Fig. S1). Fascicles in the vagal trunk contributing to the bronchopulmonary, the cardiac or the recurrent laryngeal branch were identified manually at micro-CT sections taken through the levels of emergence and their longitudinal trajectories were semi-automatically tracked rostrally (Suppl. Figs. S2, S3, S4).

Bronchopulmonary fascicles form a cluster close to its level of emergence (yellow fascicles, Fig. 1B, panels x and w; and Fig. 1C, panels r and q); bronchopulmonary fascicles remain clustered for several centimeters rostrally (Fig. 3B, panels v–s; Fig. 3C, panels p–o). The recurrent laryngeal nerve (RNL) emerges from the trunk through another cluster of fascicles (red fascicles; Fig. 1B, panels r–q; Fig. 1C, panels n–m). Some of those fascicles ascend in parallel to the BP cluster and some merge with it, in a “mixed” BP-RLN cluster (orange fascicles; Fig. 1B, panels p–o; Fig. 1C, panels m–l). BP and RLN fascicles merge within a few centimeters into a cluster of mixed BP-RLN fascicles, which runs rostrally along the upper thoracic (Fig. 1B, panel n) and the cervical vagus nerve (Fig. 1C, panel l). The cardiac branch emerges rostrally to the RLN level from a separate cluster of fascicles (blue fascicles; Fig. 1B, panel o; Fig. 1C, panel l). Some of the cardiac fascicles maintain their separate trajectories all the way through the upper thoracic and cervical vagus nerve (Fig. 1B, panels l–d; Fig. 1C; panels k–d), whereas some merge with the mixed BP-RLN fascicles to form a larger cluster of mixed BP-RLN-cardiac (green fascicles; Fig. 1B, panels m and above; Fig. 1C, panels e and above). Many of the mixed BP-RLN-cardiac fascicles terminate in the nodose ganglion, and some of them bypass it (Fig. 1B, panels b, a; Fig. 1C, panels b, a). Fascicular splitting, where one fascicle splits to form two fascicles, or fascicular merging, where two fascicles merge to form one fascicle is observed throughout the length of the nerve (Suppl. Fig. S5, and Supplementary Video 1: https://www.youtube.com/watch?v=gaNHijoVB_U).

These findings indicate that fascicles in the thoracic and cervical vagal trunk form spatially distinct organ-specific clusters close to the levels of entry of branches to those organs; those clusters progressively merge into larger, mixed clusters in the rostral direction.

Supplementary video related to this article can be found at https://doi.org/10.1016/j.brs.2023.02.003

3.2. Fascicles form separate sensory and motor clusters along the vagus nerve

To understand the spatial arrangement of fascicles with respect to sensory or motor functions, we analyzed the micro-CT images in the opposite direction, from rostral to caudal: fascicles that originate in the nodose ganglion were considered primarily sensory and those that bypass the nodose ganglion were considered primarily motor (Fig. 2A; Fig. 2B, a, b). By tracking fascicular trajectories in the caudal direction (Suppl. Video 1; also available at https://www.youtube.com/watch?v=gaNHijoVB_U), we found that sensory and motor fascicles form 2 distinct clusters throughout most of the cervical vagus nerve (Fig. 2B, c-i); the 2 clusters begin to merge in the lower cervical region (Fig. 2B, j, k), and the sensory-motor separation rather abruptly disappears in the thoracic region (Fig. 2A, l-o). Similar results from two other vagus nerves are shown in Suppl. Figs. S6 and S30. Furthermore, when we combined the tracing of organ- and function-specific fascicles in the same nerve, we found that cardiac fascicles that maintain their separate trajectories all the way through the upper thoracic and cervical vagus nerve primarily belong to the efferent group of fascicles, and that the mixed RLN/bronchopulmonary/cardiac fascicles form a separate sensory and motor group (Suppl. Fig. S31).

Fig. 2. — (A) Schematic showing a segment of the vagal trunk with several fascicles along its path. At rostral levels, sensory fascicles (green) converge into the nodose ganglion, while motor fascicles (red) by-pass the nodose. At more caudal levels, sensory and motor fascicles merge to form mixed fascicles (yellow). (B) Micro-computed tomography imaging of an intact right vagus nerve trunk, from just above the nodose ganglion to the upper thoracic level (n = 2). Shown here is results from one animal, similar results from a second nerve are shown in Suppl. Fig. S6. Panels (a–i) show the trajectories of sensory and motor fascicles in consecutive sections through the cervical vagus, at different levels (shown in cm from the rostral end of the sample). (C) Imaging of afferent and efferent fibers inside sensory and motor fascicles, n = 8. (a) H&E section at the level of the nodose ganglion. An area inside the sensory nodose ganglion (1) and a motor fascicle (2) are selected. (b1) Nodose ganglion stained with anti-myelin basic protein antibody (MBP, red) that labels myelin, and choline acetyltransferase (ChAT, green) that labels efferent, cholinergic, fibers. (b2) The same area in the nodose ganglion stained with MBP and *anti*-Nav1.8 antibody (cyan) that labels afferent fibers and sensory neurons. (c1) Motor fascicle (2) stained with ChAT/MBP. (c2) The same motor fascicle (2) stained with Nav1.8/MBP. (d) H&E section at the mid-cervical level. A sensory (27) and a motor fascicle (7) are selected and are stained with the same antibodies as before (e1-2 and f1-2 panels, respectively). (g) H&E section at the thoracic level. Two fascicles are selected, stained like before (h1-2 and i1-2 panels).

To examine whether the separation of micro-CT-resolved sensory and motor fascicles persists when single fibers inside those fascicles are accounted for, we performed immunohistochemistry (IHC) in sections through the nodose ganglion, mid-cervical and thoracic region (Fig. 2C, Suppl. Figs. S7, S8, and S24). At the nodose ganglion level (Fig. 2C, a), choline acetyltransferase (ChAT)-positive (ChAT+) fibers, indicative of motor function, are almost absent (Fig. 2C, b1), however voltage-gated sodium 1.8 (Nav1.8)-positive (Nav1.8+) sensory “pseudo-unipolar” cells are numerous (Fig. 2C, b2; Suppl. Fig. S9). In a motor fascicle, ChAT + fibers are abundant (Fig. 2C, c1); Nav1.8+ fibers are seen but are not as common (Fig. 2C, c2). At the mid-cervical region (Fig. 2C and d), fascicles poor in ChAT + fibers (Fig. 2C, e1) but rich in Nav1.8+ fibers (Fig. 2C, e2) are separated from fascicles rich in ChAT+ (Fig. 2C, f1, Suppl. Fig. S10) and in Nav1.8+ fibers (Fig. 2C, f2). Finally, at the thoracic region, all fascicles show intense staining for both ChAT and Nav1.8 (Fig. 2C, g-i; Suppl. Fig. S11).

These findings indicate that sensory and motor fascicles are spatially separated in the upper and mid-cervical trunk of the vagus nerve; sensory and motor fascicles rather abruptly merge in the lower cervical region, and in the thoracic region all fascicles are mixed (sensory-motor). Together with the organ-specific arrangement of fascicles, these findings suggest that effects of vagus nerve stimulation are expected to be dependent on the level at which stimulating electrodes are placed on the vagal trunk.

3.3. Morphologically distinct fiber types are organized across and within fascicles in a specific, nonuniform pattern

The human and swine vagus nerve contain fibers of several functional types, including myelinated and unmyelinated afferents and efferents, with distinct physiological functions (Suppl. Fig. S28) [1,14,30,49]. To characterize the spatial organization of vagus nerve fibers of different types across and within fascicles in the cervical vagus nerve, we developed a quantitative IHC methodology, consisting of IHC staining, imaging and automated fiber segmentation, feature extraction and analysis, at the single fiber level. Based on the expression pattern of several IHC markers, we classified each fiber in the cervical vagus nerve as myelinated efferent (MEff, Fig. 3A, a), myelinated afferent (MAff, Fig. 3A and b), unmyelinated efferent (UEff, Fig. 3A, c) or unmyelinated afferent (UAff, Fig. 3A, d and 3B; Suppl. Fig. S12). Applying additional fiber diameter criteria, we classified fiber types compatible with the Erlanger-Gasser scheme (Table 1); for a detailed discussion of criteria used in the classification of fiber types, see Suppl. Table 3, Suppl. Fig. S28, and Discussion section 2.

Overall, we counted 200,000–450,000 individual fibers in each nerve. The 4 main fiber types have distinct fiber diameter distributions (Suppl. Fig. S13). Approximately 70–85% of them are of the UAff type; however, UAff fibers occupy less than 10% of the total fiber-covered area in the nerve (Table 1). On the other hand, myelinated fibers, MAff and MEff, despite representing less than 15% of total fiber count, occupy 60–80% of total area (Table 1; and Suppl. Fig. S13). There is substantial variability in fiber counts, and in respective occupied areas across fascicles, for all fiber types (Table 1). More than 50% of all fibers in the nerve, independently of fiber type, lie within 500 μm from the epineurium (Suppl. Figs. S14A and S15). The area occupied by fibers of a specific type is not significantly different between smaller, intermediate-size and larger fascicles, and this is true for all fiber types (Suppl. Fig. S14B).

In both left and right cervical vagus, fascicles with large MEff fiber counts are found on one side of the nerve, forming an “motor cluster” (Fig. 3C, a1 and a2) and fascicles with large MAff fibers are found on the opposite side, forming an “sensory cluster” (Fig. 3C, b1 and b2), in agreement with our micro-CT and qualitative IHC findings (Fig. 2). The very few, larger Aα-type efferents (diameter >10 μm), are concentrated in specific fascicles (Fig. 3C, bottom left in a1 and a2), whereas the smaller (2–5 μm in diameter), cholinergic (ChAT+) B-type fibers are more uniformly distributed within the motor cluster (Fig. 3C, bottom right in a1 and a2). Importantly, the distribution of UEff fibers, a subset of which are tyrosine-hydroxylase positive (TH+) sympathetic fibers (Suppl. Fig. S16), has little overlap with that of cholinergic B-fibers as fascicles with numerous UEff fibers are commonly localized in the sensory cluster (Fig. 3C, c1 and c2). Most large MAff fibers are found in specific fascicles of the sensory cluster, with some of them overlapping with fascicles with large MEff fibers (Fig. 3C, bottom right in b1 and b2). Many more MAff fibers are intermediate size, presumably Aδ-type, fibers, widely distributed within the sensory cluster (Fig. 3C, bottom left in b1, b2). UAff fibers are present in fascicles of both sensory and motor clusters, even though fiber counts vary between 55% and over 80% (Fig. 3C, d1 and d2). Maps with fiber distributions across fascicles from additional nerves are given in Suppl. Fig. S1. Across all nerves, most fascicles with many UEff fibers have few MEff fibers and vice-versa (Fig. 3D, a), in agreement with the non-overlapping fascicular distribution of cholinergic and non-cholinergic efferents. In contrast, most fascicles rich in UEff fibers are also rich in MAff fibers (Fig. 3D, b). Interestingly, these relationships persist even at the sub-fascicular level: sectors within a fascicle with high UEff counts have low MEff counts (Fig. 3E, a) and high MAff counts (Fig. 3E, b). In general, within sub-fascicular sectors, fibers of the same type tend to cluster together (main diagonal of Fig. 3E, c; Suppl. Fig. S18).

To illustrate whether the anatomical methodologies we developed in swine can also be applied to human vagus nerves, we subjected to micro-CT and then to IHC, the same segment of a vagus nerve from a formalin-fixed cadaver (Fig. 4A). We found that the staining patterns of different morphological fiber types within the vagal trunk and the RLN appear similar to those in swine nerves (Fig. 4B-D), suggesting that this methodology can potentially be used to characterize the organization of human vagus nerve samples at the single fiber level. We also found that micro-CT is able to resolve fascicles in the human vagus nerve, suggesting that it can potentially be used to longitudinally track fascicular trajectories (Suppl. Figs. S23–1; Suppl. Video 2; also available at https://youtu.be/gacFJ-nV_G0).

Fig. 4. — (A) Left vagus nerve sample extracted from a human cadaver, including the recurrent laryngeal nerve (RLN) branch and part of the esophageal plexus. (B) IHC image of the sample stained with anti-tyrosine hydroxylase antibody (TH, green, catecholaminergic fibers), anti-myelin basic protein antibody (MBP, red, myelinated fibers), and anti-choline-acetyl-transferase (ChAT, green, cholinergic fibers). Insert image shows H&E image of the nerve at the same level. Panel (a), shows magnified a region of the nerve rich in ChAT+ (white arrows) but poor in TH + fibers. Panel (b) shows a different region, rich in TH+ (white arrows), but poor in ChAT + fibers. (C) IHC image of the same sample stained with *anti*-Nav1.8 antibody (green, unmyelinated afferents, UAff), MBP (red) and NF (blue). A selected area is shown at 2 levels of magnification. Highest power image shows super-resolution aryscan confocal image of individual UAff fibers (white arrows). Notice the lack of Nav1.8-NF co-localization, similar to the swine vagus nerve. (D) (a)H&E image from the level of RLN emergence. Two fascicles are selected for IHC imaging, one in the RLN branch and one in the main trunk. Panels (c–f), and (g–j) show the 2 fascicles stained with MBP, NF and ChAT.

Supplementary video related to this article can be found at https://doi.org/10.1016/j.brs.2023.02.003

Taken together, these findings indicate consistent, specific and highly nonuniform patterns in the way morphologically distinct fiber types are organized in the cervical vagus nerve. Afferent and efferent myelinated fibers mediating somatic and visceral sensory and motor functions are found in, largely non-overlapping, sensory and motor fascicles. Cholinergic and noncholinergic efferent fibers are found in separate fascicles and separate sub-fascicular sectors. Unmyelinated afferent fibers are widely distributed across fascicles. These findings suggest that spatially selective, or fascicular, VNS at the cervical level could in principle be used to differentially activate distinct morphological types of vagal fibers.

3.4. Fascicular vagus nerve stimulation asymmetrically elicits function- and organ-specific nerve potentials and physiological responses

3.4.1. Development and characterization of a multi-contact cuff electrode for fascicular VNS

Activation by bioelectronic devices of different populations of efferent and afferent vagal fibers innervating peripheral organs determines the effects of vagus neuromodulation therapies [30,45,65]. We sought to test whether spatially selective, fascicular VNS can differentially activate fibers innervating specific organs, as well as fibers for controlling sensory or motor functions. For that, we first developed a multi-contact cuff electrode device to deliver stimulation from the nerve's surface. Based on theoretical calculations on the geometric overlap between fascicles and circular representations of electrical fields generated by point contacts on the periphery of the nerve (Suppl. Figure 33A), an 8-contact configuration was chosen (Fig. 5A). Even though our calculations do not constitute an optimization procedure and the chosen configuration is cannot be considered “optimal”, the 8-contact configuration could in principle, according to our calculations, activate most nerve fascicles with the same selectivity as configurations with more contacts (Suppl. Figure 33B). The cuff has a helical design to accommodate nerves of different external diameters and includes 8 small, square-shaped contacts, made of platinum-iridium (Pt–Ir), evenly distributed around the circumference, and 2 ring-shaped return contacts (Fig. 5A). The electrodes were characterized after repeated use to determine durability, using electrochemical impedance spectroscopy and cyclic voltammetry (Fig. 5B); no change in electrode impedance was found between the 1st and 6th use (Fig. 5C). Charge storage capacity (CSC) measurements performed at a scan rate of 100 mV/s over the water window (after the 6th use) yielded 3.9 ± 0.63 mC/cm2, which is comparable to previously reported values for Pt–Ir electrodes [66,67].

3.4.2. Physiological and nerve potential responses to fascicular VNS

The multi-contact cuff device was used in acute experiments in anesthetized swine. Stimulation through different contacts of the multi-contact cuff produces different physiological responses specific to organs receiving vagal innervation: changes in heart rate (heart), changes in breathing rate (lungs), and contraction of laryngeal muscles (larynx) (Fig. 6A, a and b). Stimulation also elicits evoked compound action potentials (eCAPs), recorded through a second cuff electrode placed at a distance from the stimulating device, reflecting direct activation of different fiber types. Depending on which contact is used for stimulation, differential activation of fast fibers, with a conduction velocity of Aα/β-type fibers, and of slow fibers (Aδ/B-type fibers), is observed (Fig. 6A, c and d). Administration of a neuromuscular blocking agent (vecuronium), abolishes EMG activity recorded from laryngeal muscle (Fig. 6A,f) but does not alter eCAPs from the vagus nerve (Fig. 6A, e), confirming that minimal EMG activity contaminates eCAP recordings and that EMG is not elicited by direct activation of muscles by leakage current from the cuff. Graded physiological organ responses and, accordingly, eCAP amplitudes of fibers mediating those responses are observed at a range of stimulus intensities, resulting in recruitment curves that are asymmetric across contacts (Fig. 6B). Similar recruitment maps are seen in another series of tests in a second animal (Suppl. Fig. S21).

3.4.3. Responses to fascicular VNS occur asymmetrically around the vagal trunk

To statistically assess the asymmetry in responses to fascicular VNS, threshold intensities for eliciting organ-specific responses and associated eCAP components [68] were determined for each contact. Organ-specific physiological responses have different intensity thresholds: heart rate (HR) threshold is 2500 ± 170 μA (grand mean ± SEM; tested in 86 contacts from 12 animals), breathing rate (BR) threshold is 1130 ± 68μA (71 contacts, 10 animals) and laryngeal EMG threshold is 440 ± 110 μA (66 contacts, 9 animals). Within a single animal, physiological thresholds depend on which contact is stimulated; for example, a “cardiac-selective” contact, placed over a “heart-specific” part of the nerve, will be associated with a lower heart rate threshold. On average, HR threshold is 2.7-fold greater in the least cardiac-selective compared to the most cardiac selective contact (range: 1.2- to 5-fold), BR threshold is 2.2-fold greater in the least vs. the most lung-selective contact (1.6–3.3-fold), and EMG threshold is 4.4-fold greater in the least vs. the most larynx-selective contact (1.5–11-fold). These asymmetries result in contact location statistically significantly affecting HR, BR and EMG thresholds, as well as the amplitude of fast and slow eCAP components (p < 0.05, in each ANOVA test) (Suppl. Table S4). This indicates that each of the organ-specific thresholds and fiber-specific eCAP components is asymmetrically distributed around the vagal trunk.

We then sought to determine whether the “shapes” of the asymmetric distributions of pairs of responses are similar or different across the same contacts. After normalizing threshold values (or eCAP components) by their respective minimum in each animal, to adjust for the fact that, for example, HR threshold is significantly different than BR threshold and fast eCAPs than slow eCAPs, we compared pairs of different organ-specific thresholds or eCAP components measured on the same contacts. We found that, when measured on the same contacts, normalized HR threshold is significantly different than BR threshold, HR threshold is significantly different that EMG threshold, and BR threshold is significantly different than EMG threshold (Bonferroni-corrected p < 0.05 in each of the 3 comparisons, paired t-test); the same for fast and slow eCAPs (Suppl. Table S5). We arrived at similar, statistically significant results from pairwise comparisons of the angles of the resultant vectors between pairs of responses obtained in the same animals (data not shown). This indicates that the shapes of distributions around the vagal trunk of responses to fascicular VNS are different from one another.

3.4.4. Responses to fascicular VNS have preferred directions consistent with spatial separation of organ-specific fascicles in the vagal trunk

To visualize the directionality of the asymmetrical spatial distributions of the three organ-specific physiological thresholds across contacts in different animals, we arbitrarily assigned the radial location of the contact associated with the lowest heart rate threshold in each animal (the most “heart-specific” nerve location) at 12 o'clock (90° angle) in the polar plot. After aligning polar plots in individual animals by that contact, and normalizing thresholds between a minimum and maximum value, we compiled average maps of the directionality of organ thresholds across animals (Fig. 7B). The vector sum of average heart rate thresholds points towards the location with the greatest threshold and away from the most heart-specific nerve location, which lies at an angle of 134° (Fig. 7B, a). The lung-specific location lies at an angle of 155° (Fig. 7B, b) and the larynx-specific location lies at an angle of 44° (Fig. 7B and c). All these angles are expressed with respect to the most heart-specific nerve location in every animal, as heart rate thresholds were measured in all contacts, in all animals, and were used to consistently align breathing rate or laryngeal EMG thresholds, in animals in which those thresholds were measured (See Methods-Section 1.1). In additional analyses in a subset of 7 animals in which all three of the physiological thresholds were measured, relative angles between heart- and lung-/larynx-specific locations were similarly separated (Suppl. Fig. S34). Stimulus-elicited CAPs also depend on the stimulated contact, which reflects in differential recruitment of fast and slow fibers (Fig. 7C). After aligning data in individual animals by the contact with the maximal fast fiber amplitude (placed at 90° angle) and normalizing fiber amplitudes, we compiled maps of the directionality of fast and slow fiber responses in several animals. The fast fiber-specific location lies at 90° (Fig. 7D, a), and the slow fiber-specific location at 154° (Fig. 7D, b).

Finally, to test whether the preferred directions for fascicular VNS remain consistent across days, and do not arise from “random” fluctuations in terminal experiments, we performed chronic vagus nerve implants with the multi-contact cuff electrode device, in 2 animals (Fig. 7E). By stimulating through different contacts during awake experimental sessions, we found that threshold intensities for eliciting the cough reflex, whose sensory arc is mediated by myelinated afferents innervating the laryngeal and pharyngeal mucosa [19,43,69,70], are asymmetrically distributed around the nerve and, despite the gradual increase in thresholds, the overall shape of the spatial distribution is maintained for up to 3 weeks post-implantation (Fig. 7E, b). The chronic implants were associated with minimal damage to the nerve fibers (Suppl. Fig. S25).

These findings indicate that the nonuniform spatial organization of fascicles and fibers in the vagus nerve can in principle be leveraged with a multi-contact cuff electrode to elicit differential nerve responses from vagus-innervated organs and functionally-distinct fiber types.

4. Discussion

In this study, we demonstrate rich, nonuniform structure in the anatomical organization of fibers in the cervical and thoracic vagus nerve of the swine, with specific organ- and function-specific arrangement in the transverse and longitudinal directions. We also show that a multi-contact cuff electrode device that delivers fascicular vagus nerve stimulation (VNS), can in principle differentially activate vagal fibers of different functional types, mediating different organ-specific functions. These findings have implications for the development and testing of selective vagus neuromodulation therapies targeting specific organs and functions mediated by the cervical vagus nerve in humans.

4.1. Anatomical organization of fascicles in the vagal trunk

A first level of organization of the vagus nerve concerns organs that are innervated by branches emerging along its course: cervical branches, including the superior and recurrent laryngeal nerves for the larynx and pharynx, thoracic branches, including the cardiac and bronchopulmonary nerves for the heart and the respiratory system, and numerous lower thoracic and abdominal branches, innervating organs of the gastrointestinal system [1]. By imaging fascicles contributing to individual organ-specific branches with micro-CT [3] and tracking the 3D trajectories of individual fascicles along the nerve trunk, we discovered that organ-specific fascicles form clusters close to the level at which those branches emerge and progressively merge with clusters from other organs, at more rostral levels (Fig. 1). Cardiac fascicles are an exception to this rule, as some of them, mostly motor fascicles, do not merge with other fascicles (Fig. 1; Suppl. Fig. S31; Suppl. Fig. S32), an attribute that may facilitate selective VNS for cardiac diseases [71]. Recently, it was reported that fascicles contributing to the formation of laryngeal nerves in swine are grouped separately from other, non-laryngeal, fascicles in the vagal trunk [5], consistent with our findings in swine (Fig. 1) and in one human vagus nerve segment (Suppl. Figs. S23–1). We suggest that this separation is a manifestation of an organotopic anatomical organization of the vagal trunk, in which clusters of fascicles form organ-specific branches, at least along segments of the thoracic and cervical regions. Even though we only observed organ-specific clusters in 2 swine nerves, a recent preliminary study that used a similar, micro-CT methodology to characterize the microscopic structure of the swine mid-cervical vagus nerve arrived at a similar conclusion in 4 nerves [72], offering additional support to our findings.

The presence of an organotopic organization of fibers in the trunk of the vagus nerve is reminiscent of the topographic organization of motor vagal neurons in the dorsal motor nucleus of the brainstem [73,74] and of sensory vagal neurons in the nodose ganglion [74] and the nucleus tractus solitarius [75]. The progressive merging of organ-specific fascicles in the cephalad direction suggests that fascicles of the cervical vagus nerve are organized in a “fanning-out” manner with respect to their organ connectivity (Fig. 1A). Merging and splitting of fascicles has recently been reported in short segments of human vagus nerves [76], suggesting that, over longer distances, progressive cephalad merging of fascicles from different organs may also apply to the human vagus nerve.

A second level of organization of the vagus nerve concerns the sensory and motor innervation of organs, provided by afferent and efferent fibers, respectively. By tracking the 3D trajectories of fascicles emerging from the sensory nodose ganglion (primarily sensory fascicles) and of fascicles by-passing the nodose ganglion (primarily motor fascicles), we document for the first time spatial separation between the 2 groups, that persists throughout most of the cervical region; motor and sensory fascicles start merging in the lower cervical and are completely merged in the upper thoracic region (Fig. 2B; Suppl. Figs. S6 and S30). The sensory-motor separation involves large myelinated afferent and efferent fibers (Fig. 3C, a and b), in agreement with [5]. Such fibers provide somatic sensory and motor innervation to the larynx and pharynx, even though our IHC studies could not resolve which of those fibers and fascicles specifically contribute to the RLN. This suggests that fascicular VNS may be able to suppress laryngeal muscle contraction by avoiding motor fibers, and/or the coughing reflex, by avoiding sensory fibers, depending on whether RLN fibers do or do not co-localize in the same fascicles. We additionally find for the first time that this separation includes smaller afferent and efferent fibers that provide visceral sensory and motor innervation to visceral organs (Fig. 3C, panels a, fibers <3 μm, panel b, fibers 2–5 μm, and panels c, d). Therefore, the sensory-motor separation of fascicles may represent an organizing principle for both somatic and visceral fibers in the vagus nerve throughout most of the cervical region, similar to the distinct topographic arrangement of some somatic peripheral nerves [6]. This finding suggests that sensory-motor functions in the cervical vagus nerve are organized, at a large scale, in a manner resembling a multi-conductor cable: the 2 functions are mediated by separate fascicles that descend in parallel to each other, similar to 2 separate cords running down a cable, only to merge at the lower cervical and upper thoracic vagal trunk (Fig. 2A) [6]. Finally, we found that the surgical dissection and micro-CT imaging methodology we developed in swine vagus nerves is effective in a single vagus nerve sample, obtained from a formalin-fixed human cadaver (Suppl. Fig. S23). This suggests that, after additional validation in more nerve samples, this methodology can potentially be used to characterize the trajectories of organ- and function-specific fascicles in human vagus nerves.

Taken together, the increasing separation of sensory-motor fascicles in the cephalad direction and of organ-specific fascicles in the caudad direction, suggests that the level at which an electrode is implanted on the cervical vagus nerve is likely to affect the function- and organ-specific responses to VNS. Placing an electrode close to the emergence of organ-specific branches, e.g., on the lower cervical and thoracic vagal trunk, is likely to attain better selectivity related to organ functions; in contrast, placing an electrode closer to the nodose ganglion, e.g., on the mid and upper cervical vagal trunk, is likely to attain better sensory vs. motor selectivity. Mid-cervical placement of a stimulation device may be the preferred option for spatially-selective VNS, given the ease of surgical access of that location. A major difference between the human and the swine vagus nerve is the number of fascicles in the cervical vagus nerve: the swine vagus nerve has 5–10 times more fascicles than the human vagus nerve and accordingly smaller fascicle diameters (Suppl. Figure 1), offering a more “granular” anatomical substrate for testing methods and technologies for selective vagus neuromodulation [1,30,38,77,78].

4.2. Anatomical organization of morphological fiber types

Another layer of the anatomical organization in the vagus nerve concerns the several morphologically-distinct fiber types that comprise it [79]. Historically, to obtain high resolution mapping of fibers in the vagus nerve, electron microscopy (EM) imaging has been used to visualize, classify and quantify single fibers based on their ultrastructural characteristics. To discriminate between afferent and efferent fibers, selective vagotomies are performed at different levels, causing selective loss of injured afferent or efferent axons, whose distributions are then compared to those in intact nerves. In those studies, fiber quantification is performed by manual annotation of EM images from randomly selected regions of the nerve; the number of fibers in the sampled area is then extrapolated to estimate the total number of fibers in the entire nerve [11,13,20,80,81]. This procedure assumes uniform distribution of fibers across and within fascicles, something that our data does not support. Because it involves EM imaging and selective vagotomies, this process is time-consuming and expensive and cannot realistically be used to fully resolve fiber arrangement in many fascicles and/or in many nerves [20]. This procedure assumes uniform distribution of fibers across and within fascicles, something that our data does not support.

To address these issues, we developed a methodology for staining, imaging and automated annotation and quantification of single fibers. We were particularly interested in myelinated and unmyelinated, efferent and afferent fibers, that mediate specific functions of the vagus nerve. Myelinated efferents provide motor innervation to laryngeal muscles (Aα-type) or parasympathetic innervation to visceral organs (B-type); unmyelinated efferents are adrenergic hitchhiking postganglionic fibers from the sympathetic ganglion (C-type) [8,22,49,82,83], even though there is evidence for the presence of adrenergic and cholinergic afferents in the vagus nerve [84,85]. Myelinated afferents convey sensory signals from the pharyngeal and laryngeal mucosa and laryngeal muscles (Aβ -type), and signals from mechanical and chemical receptors in visceral organs, including the heart, lungs and vessels [86] (Aδ-type); unmyelinated afferents convey chemical, inflammatory and possibly nociceptive signals from visceral organs (C-type) [16].

In some of our IHC studies we used a panel of antibodies against choline acetyltransferase (ChAT), neurofilament (NF) and myelin basic protein (MBP). The colocalized expression of all three stains together was used to identify large myelinated efferent (MEff) fibers (Fig. 3A). Fibers that lack ChAT staining but are positive for NF and MBP were classified as myelinated afferents (MAff) and fibers that lack ChAT and MBP are classified as unmyelinated efferents (UEff). In our study we imaged the entire nerve and all fascicles at a resolution of 0.150 μm/pixel. At that resolution, individual axons as small as 0.3 μm diameter can be visualized and annotated. To the best of our knowledge, this is the first methodology using light microscopy that allows complete characterization of all the main morphological fiber types in a peripheral nerve, including small, unmyelinated fibers. ChAT + fibers in previous reports were quantified using a DAB chromogen technique, at a relatively low power e.g. 20x, which does not allow accurate estimation of fiber diameter [5]. In this work, we employed immunofluorescence (IF) to detect ChAT expression. The advantage of the IF technique is the ability to colocalize the ChAT expression with NF and MBP. Also, IF enabled us to visualize individual fibers at 100 × magnification, accessing the subcellular localization of the ChAT, NF, and MBP in the same axon with high selectivity and sensitivity.

Small, unmyelinated afferent (UAff) fibers do not express NF in our stains (Suppl. Fig. S12) and therefore could not be identified using that marker. We instead relied on an antibody against Nav1.8, a tetrodotoxin-resistant voltage-gated channel encoded by the SCN10A gene specifically expressed by small diameter sensory neurons in spinal and cranial nerves, to identify clusters of unmyelinated afferents (Fig. 3B; Suppl. Fig. S11; Suppl. Fig. S20). Since non-myelinating Schwann cells in Remak bundles are known to encompass small unmyelinated fiber clusters, we confirmed detection of C-afferent clusters by testing for co-localization with the S100 antibody expressed by Schwann cells (Fig. 3B). In our study, Nav1.8+ fibers do not stain for MBP or NF (Suppl. Fig. S12) which is consistent with their expression in unmyelinated, nociceptor fibers but not myelinated, mechanoreceptor afferent fibers in the vagus nerve [87]. Sizes and counts of Nav1.8+ fibers in our study are consistent with those of small, unmyelinated afferents in EM studies of the cervical vagus nerve in humans [12,13]. However, additional investigations directly comparing IHC to EM imaging will be needed for unequivocal confirmation that Nav1.8 is an accurate and precise marker of UAff fibers in the vagus nerve.

Until now, analysis of IHC data in nerves to segment single fibers has relied on inherently slow manual or semi-automated processes, e.g. Ref. [88]. Likewise, analysis of micro-CT data in nerves has used software-aided manual segmentation methods [10]. In this study, we used a mask-RCNN deep-learning architecture to segment both micro-CT and IHC images [10,89]. Deep-learning based algorithms have recently received attention in anatomically guided, medical image segmentation [90]. With regard to segmentation of anatomical features of nerves, deep-learning algorithms have been used on EM images, to resolve single fibers, and on ultrasound [91] and histological images [92], to resolve fascicles. To the best of our knowledge, this is the first use of convolutional neural networks on micro-CT and IHC data, to segment and extract anatomical features from micro-CT or IHC data in a user-assisted, semi-automated manner. After additional validation, similar methodologies to the ones described here may be useful in quantitative anatomical studies of peripheral nerves in health and disease, by improving efficiency and by minimizing sampling errors [93], to resolve fascicles. Our semi-automated segmentation and tracking methodology has limitations. While semi-automated annotation of 98% of the micro-CT images is significantly speeding up segmentation and tracking, it still requires a human user to validate and correct missed or misclassified segmented fascicles, and to manually validate the extracted tracks. One possible extension of this work, in a way that could increase specificity and require less manual validation, would be to utilize segmentation algorithms that take advantage of the three-dimensional structure of the fascicular trajectories (such as 3D U-Nets), rather than relying on the 2D projections of those fascicles. Moreover, our fascicle tracking involves heuristic criteria to exclude certain traces that are considered spurious. While these criteria, paired with the manual validation of the resulting tracks, worked in our datasets, their generalization was not formally tested and further evaluation on other datasets is warranted.

Using the IHC methodology, we were able to count, rather than estimate, the entirety of fibers in several vagus nerves (3 left, 5 right), at the cervical level, and provide their detailed spatial distributions across all fascicles; complete characterization of all fibers in a large peripheral nerve is reported here for the first time. The fiber counts we documented in the swine vagus nerve are comparable to those reported in the, much less fascicular, ferret [11] and cat nerves [20]: MEff ~4% of all fibers in swine (4% in ferret), MAff ~8% (9% in cats and ferrets), UEff ~5% (8% in ferrets, 26% in cats), and UAff >80% (79% in ferrets, 54% in cats). Ours is the first quantification of the distribution of UEff fibers in the swine vagus nerve. Many of the UEff fibers are small, TH+, adrenergic fibers (Suppl. Fig. S16). The remaining UEff fibers (TH−) stain for NF but not for Nav1.8, consistent non-adrenergic non-cholinergic (NANC) efferents; for example, the vagus nerve contains nitrinergic and peptidergic efferent fibers [94-97]. Our study is also the first to describe the complete arrangement of vagal UAff fibers across the entire vagus–or any large size nerve–using the Nav1.8 marker, a product of the SCN10A gene, which is highly expressed in sensory neurons in the nodose ganglion [98]. UAff fibers occupy 5–10% of the nerve area on average and 1–28% of area in fascicles (Table 1); accordingly, a recent EM study in the human cervical vagus nerve estimated that UAff fibers occupy up to 18% of the nerve area [98]. UAff fibers occupy 5–10% of the nerve area on average and 1–28% of area in fascicles (Table 1); accordingly, a recent EM study in the human cervical vagus nerve estimated that UAff fibers occupy up to 18% of the nerve area [13]. [98]. UAff fibers occupy 5–10% of the nerve area on average and 1–28% of area in fascicles (Table 1); accordingly, a recent EM study in the human cervical vagus nerve estimated that UAff fibers occupy up to 18% of the nerve area [13]. The fascicular distribution of UAff fibers in the vagus nerve is described here for the first time. We find significant numbers of UAff fibers in every fascicle; this means that even motor fascicles have significant numbers of UAff fibers; given that motor fascicles bypass the nodose ganglion, a percentage of identified UAff fibers may represent peripheral axons of sensory cells outside of the nodose ganglion (e.g. jugular ganglion) or “hitchhiking” fibers from sensory fascicles. For all morphological fiber types, the variation of fiber counts and occupied areas between fascicles is far greater than the variation between nerves or subjects (Table 1), indicating that sampling from a subset of fascicles in a nerve to estimate total fiber counts for the entire nerve likely introduces significant estimation errors.

We show for the first time that most morphological fiber types in the vagus nerve have highly nonuniform distributions across and within fascicles. Fascicles with high UEff fiber counts tend to have low MEff fiber counts, and vice versa (Fig. 3D, a); because MAff and MEff fibers are found in separate fascicles, that also means UEff and MAff fibers tend to co-localize in the same fascicles (Fig. 3D, b). Many of the MEff fibers are cholinergic, B-type efferents, the “canonical” vagal fibers targeted by VNS therapies for heart failure and inflammatory disorders. On the other hand, a significant portion of UEff fibers are TH+ (Fig. 4B & Suppl. Fig. S16), while others could be dopaminergic or non-adrenergic-non-cholinergic fibers. The spatial separation of UEff fibers from MEff fibers suggests that a portion of catecholaminergic (UEff) and cholinergic (MEff) efferent fibers in the cervical vagus nerve occupy different nerve sections and could in principle be differentially modulated by fascicular VNS. For the first time, we document nonuniform distribution of vagal fibers within single fascicles, as well as sub-fascicular colocalization of different fiber types. For instance, myelinated fibers are frequently located in close proximity with other myelinated fibers (Fig. 3D, c-e); UAff fibers also form clusters (Fig. 3E, c), in agreement with what has been found in other peripheral nerves with C-type fibers [99].

Our choice of IHC criteria for classifying morphological fiber types has limitations. First, we classified fibers that are positive for NF, MBP and ChAT as myelinated efferents (MEff) (Table 1; Suppl. Table 3; Suppl. Fig. S12). However, studies have shown ChAT activity in the nodose (sensory) ganglion in various species, suggesting the presence of ChAT⁺ sensory fibers in the vagus. In those studies, chemoluminescence methods were used in the entire ganglia, whose output reflects ChAT activity in both the cell bodies of sensory neurons and the efferent bundles that pass around the nodose ganglion. Using IHC, only few ChAT⁺ neurons were observed in the nodose ganglion in rats and humans [85,100,101]. Recently, quantitative IHC studies in transgenic mice (ChAT-EGF) found that sensory neurons in the nodose ganglion are completely devoid of ChAT; instead, ChAT expression is limited to the efferent fibers passing around the nodose ganglion [43]. Despite their potentially small numbers, ChAT⁺ afferent fibers in the vagus nerve will need to be quantified in humans and large animals in future studies. Second, we classified fibers that are positive for NF but negative for BMP and Nav1.8 as unmyelinated efferents (UEff); because only myelinated fibers are ChAT+, those fibers are also negative for ChAT (Table 1; Suppl. Table 3; Suppl. Figure 12). We found that the vast majority, if not all, of those fibers are positive for TH (Suppl. Figure 16), something that suggests that a significant portion of UEff fibers may represent sympathetic efferents. However, we did not perform a formal quantitative assessment of TH + fibers in all our samples. Importantly, vagal fibers that are positive for both ChAT and TH have been described [102]. Therefore, the exact count of TH + fibers in our samples is unknown, as is what fraction of ChAT-unmyelinated efferents they represent. Third, a portion of UEff-classified fibers may in reality be afferent. Retrograde neuronal tracing studies have shown that some TH + fibers distributed to the esophagus originate from cells in the nodose ganglion with central axons projecting to the (sensory) nucleus tractus solitarius, suggesting that some TH + fibers are afferent. In fact, it has been estimated that up to 17% of TH + fibers in the vagus originate in the nodose ganglion, with most of the remaining from cells in the DMV [23] suggesting that some TH + fibers are afferent [84].A study found that as many as 8% of (sensory) cells in the nodose ganglion are TH+ [103], even though more recent studies showed sparse TH staining in the nodose ganglion, corroborated by low overall TH gene expression in nodose neurons [98]. Single-cell RNA sequencing in nodose and jugular ganglion cells showed that all but one cell type express Vglut2 (Slc17a6), a marker for peripheral sensory neurons. The one cell type that does not express Vglut2-displays a “sympathetic” expression profile (expressing Hand 2, Tbx20, and Ecel1 genes) and its axons could therefore be considered “efferent” [104].

The role of TH + fibers in the vagus has been a matter of debate. TH + fibers in the cervical and sub-diaphragmatic vagus nerve have been well-documented in rats [23], cats [105] and dogs [82]. In humans, TH + fibers are present throughout the vagal trunk at the cranial level, jugular and nodose ganglion, cervical and thoracic trunk, and many vagal branches [15]. In some studies, many TH + fibers are postganglionic sympathetic fibers with the cell bodies located in the sympathetic chain, entering the vagus nerve through the jugular ganglion [15]. In cats and dogs, the sympathetic trunk and the vagal trunk are fused together, to form the vagosympathetic trunk [105]. In pigs, the sympathetic trunk and the vagus nerve are separate and they run parallel to each other [5]. The presence of the postganglionic sympathetic fibers in the vagus nerve could be due to projection of hitchhiking branch from the sympathetic trunk entering the cervical vagus nerve [8,22,82] and the presence of these fibers have been suggested as possible sources of variability in the clinical responsiveness to VNS [22].

In a proof-of-principle study of a segment from a single human vagus nerve sample from a preserved cadaver, we documented potential similarities with the swine vagus nerve. For example, MEff, ChAT(+), fibers and many of the UEff, TH(+), fibers occupy separate fascicles and areas within the sample we examined (Fig. 4A); this is reminiscent of the separate distribution of MEff and UEff fibers in the swine (Fig. 3C, a vs. c). We found that UAff, Nav1.8+, fibers are widely distributed throughout most fascicles in the human sample (Fig. 4C); reminiscent of their distribution in the swine (Fig. 3C and d). Since these observations were made in the same nerve segment as the micro-CT imaging (Suppl. Figs. S23–1), it may be feasible to image the microscopic anatomic of the same nerve, at both the fascicular and single fiber levels, a method that could generate a wealth of cross-registered anatomical data. We did not follow this approach with the swine vagus nerves we studied here: because of tissue dehydration and insufficient rinse steps to remove the Lugol’s stain, we were unable to use the same swine nerve samples for both micro-CT and IHC studies. It is likely that the approach followed here with the human nerve sample is applicable in swine nerve samples, but that needs to be tested.

Our findings on the distribution of morphological fiber types across and within fascicles have implications for how vagal projections between the brain and peripheral organs may contribute to strategies for coding interoceptive stimuli [41] and for the autonomic control of organ homeostasis [106]. The co-localization of certain afferent and efferent fiber types in the same fascicles, and in the same sub-fascicular sectors, suggests that the microscopic structure of the vagus nerve may follow a “multiplexing” architecture, in which vagal signals, from/to the same or different organs, are conveyed through projections that lie in close proximity; this organization may pose physical constraints in the spatial arrangement of vagal sensory and motor neurons in the ganglia and the brainstem [41]. Our quantitative data on how fascicles and fibers are arranged in the vagal trunk will be applicable to building anatomically realistic computational models of the vagus nerve, commonly used to optimize design of bioelectronic devices and explain functional outcomes of neuromodulation therapies on the basis of variability in the underlying anatomy and in the electrode-tissue interface [44,107-109].

4.3. Fascicular vagus nerve stimulation using a multi-contact cuff electrode

Following the example of classic studies that investigated how structure of autonomic nerves explains the physiological effects arising from their electrical stimulation (e.g. Ref. [110]) we sought to determine whether the anatomical information gained in our studies could be leveraged for more selective stimulation of the vagus. Selective VNS may provide a means to personalize therapy to account for variability in individual nerve anatomy and engage the innervation of desired organs with fewer off-target effects [3]. The use of multi-contact electrodes for steering of electrical fields during nerve stimulation is an established method for delivering spatially selective stimulation, as it permits activation of sub-sets of fascicles or fibers. Such an approach has been previously followed in a rat model of VNS, in which reduction of blood pressure was attained without affecting heart rate and breathing rate [111]; it has also been used in a sheep model of VNS, in which selective stimulation attained changes in breathing rate without affecting the heart rate [39]. In this study, we tested fascicular stimulation in the multi-fascicular vagus nerve of the swine, with direct relevance to the human vagus nerve, which is also multi-fascicular.

Because a large percentage of vagal fibers lie within less than 500 μm from the epineurium (Fig. 5A), a cuff electrode may be a viable selective interface with the vagus nerve. Cuff electrodes for VNS, like the one used in this study, are less invasive than penetrating probes, and have an established safety record, e.g., Ref. [112]. In previous studies, contact configurations were tested and optimized, with regard to selectivity, in detailed, in silico studies, e.g. Refs. [39,44]. The contact configuration used in our cuff was partially based on a theoretical/geometric approximation (Suppl. Fig. S33; Methods, Section 3.1); additional studies will be required to optimize contact configuration for large animal and human studies. Even though the cuff has not undergone formal preclinical device testing, in principle, it could be used in future clinical tests: the 8-contact configuration is compatible with the fascicular anatomy of the human vagus nerve (Suppl. Fig. S26), the helical structure can accommodate a range of nerve diameters (Fig. 5A), and the silicone substrate is biocompatible [113] [https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfPMN/pmn.cfm?ID=K183437].

Delivering stimulation through single contacts results in differential engagement of fibers innervating the larynx, the lungs and the heart, as demonstrated by the different patterns of physiological and evoked potential responses (Fig. 6A). Correlations between physiological and corresponding nerve fiber responses are seen in the recruitment curves, which are also different among different contacts (Fig. 6B; Suppl. Fig. S21). This indicates that the differences among contacts in the stimulation intensity thresholds arise because of differential activation of distinct fiber populations. The radial distributions of organ-specific thresholds are asymmetric (Fig. 7; Suppl. Table S4), indicating that different sections of the nerve can be activated by specific contacts to differentially engage each of the organs. The different radial distributions of cardiac vs. respiratory vs. laryngeal EMG responses and of fast vs. slow eCAPs (Fig. 7; Suppl. Table S5), indicates that between-contact differences arise because of differences in the underlying fiber distribution, rather than factors that could affect one side of the nerve or the interface (e.g., thicker epineurium or looser placement of the cuff), in which case all three registered physiological responses would have similar radial distributions.

The angular separation between the cardiac-, lung- and larynx-selective contacts (Fig. 7B) can be explained by the underlying anatomical arrangement of fascicles contributing to cardiac, bronchopulmonary and laryngeal branches, respectively (Fig. 1). However, we did not co-register the exact contact locations with the specific functional anatomy of the stimulated nerves, that being a major limitation of our study. Recently, preliminary results from 2 studies in which spatially-selective VNS was delivered in swine suggest that, when stimulated fascicles are tracked to the organ branches they contribute to, contacts associated with stronger laryngeal responses tend to be over laryngeal branch-projecting fascicles, and contacts associated with stronger heart rate responses over cardiac branch-projecting fascicles [72,114]. The angular separation between the fast-fiber-selective contacts and slow-fiber-selective contacts (90° counter-clockwise, Fig. 7D) agrees with the separation between contacts eliciting corresponding physiological functions, i.e., fast-fiber-mediated laryngeal EMG and slow-fiber-mediated changes in breathing rate (also 90° counter-clockwise, Fig. 7B). Even though both physiological and eCAP responses were registered in most animals, we did not align eCAP responses by the contact with the minimum heart rate threshold, but to the contact with the maximum fast fiber amplitude. eCAPs are recorded through a second multi-contact cuff (Suppl. Fig. S22), placed at a distance of at least 5 cm from the stimulating cuff and, over that distance, fascicles merge or split and their radial location in the trunk changes significantly (Suppl. Video 1; also available at https://www.youtube.com/watch?v=gaNHijoVB_U). For that reason, the spatial arrangement of fascicles at the stimulating cuff, assessed through physiological thresholds, is different than the arrangement at the recording cuff, and aligning the two response patterns would be unlikely to produce interpretable results. eCAP traces from each of the 8 recording contacts reflect the fascicular structure of the nerve at the recording site, when stimulation is delivered to the same site (e.g., Suppl. Fig. S22). When comparing eCAPs elicited from different stimulation contacts, we wanted to minimize that source of eCAP variability, hence we averaged the 8 individual eCAP traces elicited by stimuli delivered to each of the stimulation contacts. That way, differences between eCAPs primarily reflect the fascicular structure of the nerve at the stimulating site.

In principle, there is possibility that some of the responses to VNS are mediated by leakage currents, rather than direct activation of fibers in the vagal trunk. In all experiments, the vagus nerve was dissected and isolated from the accompanying vessels in the vagal sheath, before the cuff was implanted and VNS was delivered; despite that, current leakage into the vessel wall cannot be ruled out. Direct activation of baroreceptors that lie close to the bifurcation of the common carotid artery cannot be ruled out; however, in 3 animals, we did not observe any changes in arterial blood pressure that could not be explained by the concomitant bradycardic effect of VNS (data not shown). The mid-cervical cuff placement was >12 cm away from the emergence of the cardiac and recurrent laryngeal branches, and even further away from bronchopulmonary branches (Fig. 1); given those distances, current leakage and direct activation of organ branches is unlikely, even though it cannot be excluded. Direct activation of the end organs themselves by VNS is unlikely. We found no evidence of direct activation of laryngeal muscles by VNS, as laryngeal muscle responses to VNS disappeared when neuromuscular junctionblocking agents were used (Fig. 6A, f). Direct activation of the sinoatrial node from bipolar stimuli of <5 mA is unlikely, especially from electrodes placed >30 cm away. Finally, slower breathing cannot be explained by direct activation of lung or tracheal tissue.

In our experiments in anesthetized animals, heart rate responses were almost uniformly bradycardic, and bradycardia was by far the most common cardiac response to stimulation at and above threshold intensities. Previous studies in anesthetized swine showed that relatively low stimulus intensities promote tachycardia, whereas higher intensities promote bradycardia e.g., Ref. [115]. It is likely that, in anesthetized animals, at intensities below the threshold for direct activation of cardioinhibitory efferent fibers, lower threshold afferent fibers are instead activated, causing reflex changes in sympathetic drive and tachycardia [115]. Synaptic transmission along those brainstem reflexes is likely sensitive to general anesthetics: for example, isoflurane anesthesia in rats is associated with a significantly higher VNS intensity threshold for afferent responses [68]. In our study, isoflurane was used at relatively higher concentrations (1.5–3% vs. 1–2% in Ref. [115]), possibly shifting the tachycardic response to higher intensities; at higher intensities, cardioinhibitory efferents may be activated as well, leading to a net no-change or even drop in heart rate. Even though we documented many unmyelinated efferents in the cervical vagus, some of which may be catecholaminergic, hitch-hiking sympathetic fibers (Table 1; Fig. 4), it is unlikely that those fibers were directly engaged in our physiology experiments, due to their high activation thresholds. However, the spatial segregation of unmyelinated efferents from myelinated cholinergic efferents (e.g., Fig. 3C) may represent an opportunity for differential engagement of sympathetic vs. parasympathetic efferent vagal projections using fascicular VNS.

To test whether fascicular VNS is feasible in awake conditions, and to establish its stability across experimental sessions, we administered fascicular VNS from chronically-implanted cuffs for up to 3 weeks, in 2 swine (Fig. 7E, a). In awake conditions, activation of the cough reflex was the physiological response observed at lowest stimulus intensities; indeed, cough and throat pain are common side effects in patients receiving cervical VNS [116]. Intensity thresholds for eliciting the cough reflex are asymmetrically distributed across stimulating contacts in both animals (Fig. 7E, b1 and c1), suggesting that the severity of this off-target effect can in principle be reduced by fascicular VNS. To our knowledge, this is first report of a chronic VNS implant in swine, and the first report of cough reflex thresholds from VNS in awake swine. Cough reflex thresholds progressively increase post-implantation (Suppl. Fig. S27), similar to what has been observed in chronic VNS implants in mice [117] and rats [68]. The progressive increase of cough reflex threshold likely reflects the progressive development of fibrotic tissue around and inside the implanted cuff rather than major loss of nerve fibers, for which we found no histological evidence (Suppl. Fig. S25). Despite the increase in thresholds, the shape of threshold distributions around the nerve remains consistent across time, in both animals (Fig. 7E, b1 and c1; Suppl. Fig. S27), suggesting that fascicular VNS may be a feasible strategy to suppress this off-target effect in chronically implanted subjects.

During fascicular VNS, we observed eCAPs comprised of only fast and slow A-fiber components. This may be related to the large areas occupied by large myelinated fibers (Table 1). At the same time, we did not activate small, unmyelinated fibers (e.g., C-type), as evidenced by the lack of slow fiber eCAP responses. Small, unmyelinated fibers constitute >80% of the total vagal fiber count, however, they occupy <10% of the entire area the nerve (Table 1). Small, unmyelinated fibers have high activation thresholds [118,119]. In our experiments, we used smaller current intensities; by using intensities high enough to activate C-fibers, larger fibers would have been activated as well, and hence the spatial selectivity of other physiological effects would have disappeared. At the same time, most unmyelinated fibers have a more uniform distribution across fascicles (Fig. 3). For that reason, it is less likely that fascicular VNS would result in asymmetric C-fiber responses in eCAPs–however our study has not tested that possibility, e.g., by using C-fiber selective electrical stimuli.

Our physiology studies demonstrate that spatially selective, fascicular stimulation using a multi-contact cuff electrode is a meaningful strategy for selective VNS in the multi-fascicular vagus nerve of the swine, and it may be feasible in the human cervical vagus nerve, which is also multi-fascicular. However, the degree of organ- and function-selectivity of this approach in our study was limited: no single contact is 100% selective for a single organ or for a single fiber type. In addition, our data show that spatially-selective stimulation by itself is not a viable strategy to alter the default order of recruitment of vagal fibers: larger fibers are recruited at low stimulus intensities, both larger and smaller fibers are recruited at high stimulus intensities. In many cases, it would be desirable to elicit a response mediated by smaller fibers, e.g. cardioinhibition, at a stimulus intensity that does not produce an undesired side-effect mediated by larger fibers, e.g. activation of sensory and motor fibers to the larynx. Indeed, in most experiments in our study, with one exception, the cough reflex threshold was lower than the heart rate threshold, independently of which contact was stimulated. This is in agreement with a recent report in which fascicular VNS in swine resulted in EMG thresholds that were consistently smaller than heart rate thresholds, independently of contact location on the vagal trunk [114]. The limited selectivity in these studies may be expected given the many, small fascicles in the swine nerve, the epineural placement of the stimulating electrode, and the nonfiber-selective nature of stimulus waveforms we used. Additional studies are needed to establish whether fascicular VNS is indeed an essential component of a methodology for attaining truly function- and organ-selective VNS.

To translate these findings into clinically-relevant selective VNS therapies, anatomy-guided electrode contact configurations used with fiber-selective stimulus parameters will need to be tested [120]. To deliver precision VNS therapies, the functional anatomy of the vagus nerve will need to be mapped in individual subjects. Towards that end, the large-scale fascicular structure of the swine vagus nerve has been visualized with noninvasive, high resolution ultrasound imaging [78] or with electrical impedance tomography of nerve activity [121,122]. Our results show that functional nerve mapping can be also performed by quantifying physiological and/or nerve fiber responses delivered through different contacts and visualizing their spatial distributions around the nerve (Figs. 6B and 7A). Physiological responses and eCAPs elicited by stimulation can be registered and processed in almost real-time in humans or experimental animals to generate functional nerve maps and calibrate and optimize VNS parameters [45].

Supplementary Material

NIHMS1898332-supplement-1.docx^{(44.6MB, docx)}

Download video file^{(9MB, mp4)}

Download video file^{(715.1KB, mp4)}

Acknowledgments

This work was supported by a grant from United Therapeutics Corp. To SZ and by an NIH-SPARC grant to LM (1OT2OD026539). The authors would like to acknowledge Drs. Betty Diamond, Thomas Coleman, Valentin Pavlov, Sangeeta Chavan and Eric Chang for helpful discussions.

Abbreviations:

VNS: vagus nerve stimulation
eCAP: stimulus-evoked compound action potential
micro-CT: microscopic computed tomography
IHC: immunohistochemistry
NF: neurofilament
MBP: myelin basic protein
ChAT: choline acetyl-transferase
TH: tyrosine hydroxylase
Nav1.8: sodium voltage-gated channel 1.8
RLN: recurrent laryngeal nerve
BP: bronchopulmonary
EMG: electromyography

Footnotes

CRediT authorship contribution statement

Naveen Jayaprakash: designed experiments, performed experiments and collected data, Formal analysis, Writing – original draft. Weiguo Song: designed experiments, performed experiments and collected data, Formal analysis, Writing – original draft. Viktor Toth: designed experiments, performed experiments and collected data, Formal analysis, Writing – original draft. Avantika Vardhan: Formal analysis, Writing – original draft. Todd Levy: Formal analysis, Writing – original draft. Jacquelyn Tomaio: performed experiments and collected data. Khaled Qanud: performed experiments and collected data. Ibrahim Mughrabi: collected and analyzed data. Yao-Chuan Chang: collected and analyzed data. Moontahinaz Rob: collected and analyzed data. Anna Daytz: collected and analyzed data. Adam Abbas: collected and analyzed data. Zeinab Nassrallah: Writing – original draft. Bruce T. Volpe: Writing – original draft. Kevin J. Tracey: Writing – original draft. Yousef Al-Abed: Writing – original draft. Timir Datta-Chaudhuri: Writing – original draft. Larry Miller: performed experiments and collected data, Writing – original draft, Funding acquisition. Mary F. Barbe: performed experiments and collected data, Writing – original draft. Sunhee C. Lee: performed experiments and collected data, Writing – original draft. Theodoros P. Zanos: designed experiments, Formal analysis, Writing – original draft. Stavros Zanos: designed experiments, performed experiments and collected data, Formal analysis, interpreted data, Writing – original draft, Funding acquisition.

Declaration of competing of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.brs.2023.02.003.

References

[1].Stakenborg N, et al. Comparison between the cervical and abdominal vagus nerves in mice, pigs, and humans. Neuro Gastroenterol Motil 2020;32(9):e13889. [DOI] [PMC free article] [PubMed] [Google Scholar]
[2].Pelot NA, et al. Quantified morphology of the cervical and subdiaphragmatic vagus nerves of human, pig, and rat. Front Neurosci 2020;14:601479. [DOI] [PMC free article] [PubMed] [Google Scholar]
[3].Thompson N, Mastitskaya S, Holder D. Avoiding off-target effects in electrical stimulation of the cervical vagus nerve: neuroanatomical tracing techniques to study fascicular anatomy of the vagus nerve. J Neurosci Methods 2019;325:108325. [DOI] [PMC free article] [PubMed] [Google Scholar]
[4].Kenny BJ, Bordoni B. Neuroanatomy, cranial nerve 10 (vagus nerve). In: StatPearls; 2021. [Treasure Island (FL: ]. [PubMed] [Google Scholar]
[5].Settell ML, et al. Functional vagotopy in the cervical vagus nerve of the domestic pig: implications for the study of vagus nerve stimulation. J Neural Eng 2020;17(2):026022. [DOI] [PMC free article] [PubMed] [Google Scholar]
[6].Agostoni E, et al. Functional and histological studies of the vagus nerve and its branches to the heart, lungs and abdominal viscera in the cat. J Physiol 1957;135(1):182–205. [DOI] [PMC free article] [PubMed] [Google Scholar]
[7].Hammer N, et al. Human vagus nerve branching in the cervical region. PLoS One 2015;10(2):e0118006. [DOI] [PMC free article] [PubMed] [Google Scholar]
[8].Verlinden TJ, et al. Morphology of the human cervical vagus nerve: implications for vagus nerve stimulation treatment. Acta Neurol Scand 2016;133(3):173–82. [DOI] [PubMed] [Google Scholar]
[9].Yoshida Y, et al. Sensory innervation of the pharynx and larynx. Am J Med 2000;108(Suppl 4a):51S–61S. [DOI] [PubMed] [Google Scholar]
[10].Thompson N, et al. MicroCT optimisation for imaging fascicular anatomy in peripheral nerves. J Neurosci Methods 2020;338:108652. [DOI] [PMC free article] [PubMed] [Google Scholar]
[11].Asala SA, Bower AJ. An electron microscope study of vagus nerve composition in the ferret. Anat Embryol 1986;175(2):247–53. [DOI] [PubMed] [Google Scholar]
[12].Gabella G, Pease HL. Number of axons in the abdominal vagus of the rat. Brain Res 1973;58(2):465–9. [DOI] [PubMed] [Google Scholar]
[13].Havton LA, et al. Human organ donor-derived vagus nerve biopsies allow for well-preserved ultrastructure and high-resolution mapping of myelinated and unmyelinated fibers. Sci Rep 2021;11(1):23831. [DOI] [PMC free article] [PubMed] [Google Scholar]
[14].Hoffman HH, Schnitzlein HN. The numbers of nerve fibers in the vagus nerve of man. Anat Rec 1961;139:429–35. [DOI] [PubMed] [Google Scholar]
[15].Kawagishi K, et al. Tyrosine hydroxylase-immunoreactive fibers in the human vagus nerve. J Clin Neurosci 2008;15(9):1023–6. [DOI] [PubMed] [Google Scholar]
[16].Kollarik M, Ru F, Brozmanova M. Vagal afferent nerves with the properties of nociceptors. Auton Neurosci 2010;153(1–2):12–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
[17].Kressel M, Radespiel-Troger M. Anterograde tracing and immunohistochemical characterization of potentially mechanosensitive vagal afferents in the esophagus. J Comp Neurol 1999;412(1):161–72. [DOI] [PubMed] [Google Scholar]
[18].Lundberg J, et al. Catecholamine-containing nerve fibres in the human abdominal vagus. Gastroenterology 1976;70(3):472–4. [PubMed] [Google Scholar]
[19].Mazzone SB, Undem BJ. Vagal afferent innervation of the airways in health and disease. Physiol Rev 2016;96(3):975–1024. [DOI] [PMC free article] [PubMed] [Google Scholar]
[20].Mei N, Condamin M, Boyer A. The composition of the vagus nerve of the cat. Cell Tissue Res 1980;209(3):423–31. [DOI] [PubMed] [Google Scholar]
[21].Schemann M, et al. Identification of cholinergic neurons in enteric nervous system by antibodies against choline acetyltransferase. Am J Physiol 1993;265(5 Pt 1):G1005–9. [DOI] [PubMed] [Google Scholar]
[22].Seki A, et al. Sympathetic nerve fibers in human cervical and thoracic vagus nerves. Heart Rhythm 2014;11(8):1411–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
[23].Yang M, Zhao X, Miselis RR. The origin of catecholaminergic nerve fibers in the subdiaphragmatic vagus nerve of rat. J Auton Nerv Syst 1999;76(2–3):108–17. [DOI] [PubMed] [Google Scholar]
[24].de Lartigue G. Putative roles of neuropeptides in vagal afferent signaling. Physiol Behav 2014;136:155–69. [DOI] [PMC free article] [PubMed] [Google Scholar]
[25].Berthoud HR, Neuhuber WL. Functional and chemical anatomy of the afferent vagal system. Auton Neurosci 2000;85(1–3):1–17. [DOI] [PubMed] [Google Scholar]
[26].De Ferrari GM, et al. Long-term vagal stimulation for heart failure: eighteen month results from the NEural Cardiac TherApy foR Heart Failure (NECTAR-HF) trial. Int J Cardiol 2017;244:229–34. [DOI] [PubMed] [Google Scholar]
[27].Mikami Y, et al. Vagus nerve-mediated intestinal immune regulation: therapeutic implications of inflammatory bowel diseases. Int Immunol 2022;34(2):97–106. [DOI] [PubMed] [Google Scholar]
[28].Koopman FA, et al. Vagus nerve stimulation inhibits cytokine production and attenuates disease severity in rheumatoid arthritis. Proc Natl Acad Sci U S A 2016;113(29):8284–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
[29].Donega M, et al. Human-relevant near-organ neuromodulation of the immune system via the splenic nerve. Proc Natl Acad Sci U S A 2021;118(20). [DOI] [PMC free article] [PubMed] [Google Scholar]
[30].Nicolai EN, et al. Sources of off-target effects of vagus nerve stimulation using the helical clinical lead in domestic pigs. J Neural Eng 2020;17(4):046017. [DOI] [PMC free article] [PubMed] [Google Scholar]
[31].Musselman ED, Pelot NA, Grill WM. Empirically based guidelines for selecting vagus nerve stimulation parameters in epilepsy and heart failure. Cold Spring Harb Perspect Med 2019;9(7). [DOI] [PMC free article] [PubMed] [Google Scholar]
[32].Booth LC, et al. Selective optogenetic stimulation of efferent fibers in the vagus nerve of a large mammal. Brain Stimul 2021;14(1):88–96. [DOI] [PMC free article] [PubMed] [Google Scholar]
[33].Fontaine AK, et al. Optogenetic stimulation of cholinergic fibers for the modulation of insulin and glycemia. Sci Rep 2021;11(1):3670. [DOI] [PMC free article] [PubMed] [Google Scholar]
[34].Tanaka S, et al. Vagus nerve stimulation activates two distinct neuroimmune circuits converging in the spleen to protect mice from kidney injury. Proc Natl Acad Sci U S A 2021;118(12). [DOI] [PMC free article] [PubMed] [Google Scholar]
[35].Plachta DT, et al. Blood pressure control with selective vagal nerve stimulation and minimal side effects. J Neural Eng 2014;11(3):036011. [DOI] [PubMed] [Google Scholar]
[36].Kong SH, et al. Intraoperative neurophysiologic testing of the perigastric vagus nerve branches to evaluate viability and signals along nerve pathways during gastrectomy. J Gastric Cancer 2019;19(1):49–61. [DOI] [PMC free article] [PubMed] [Google Scholar]
[37].Furukawa N, Okada H. Effects of selective vagal stimulation on the gallbladder and sphincter of Oddi and peripheral vagal routes mediating bile evacuative responses induced by hypothalamic stimulation. Jpn J Physiol 1992;42(2):321–34. [DOI] [PubMed] [Google Scholar]
[38].Fitchett A, Mastitskaya S, Aristovich K. Selective neuromodulation of the vagus nerve. Front Neurosci 2021;15:685872. [DOI] [PMC free article] [PubMed] [Google Scholar]
[39].Aristovich K, et al. Model-based geometrical optimisation and in vivo validation of a spatially selective multielectrode cuff array for vagus nerve neuromodulation. J Neurosci Methods 2021;352:109079. [DOI] [PubMed] [Google Scholar]
[40].Musselman ED, et al. ASCENT (Automated Simulations to Characterize Electrical Nerve Thresholds): a pipeline for sample-specific computational modeling of electrical stimulation of peripheral nerves. PLoS Comput Biol 2021;17(9):e1009285. [DOI] [PMC free article] [PubMed] [Google Scholar]
[41].Zhao Q, et al. A multidimensional coding architecture of the vagal interoceptive system. Nature 2022;603(7903):878–84. [DOI] [PMC free article] [PubMed] [Google Scholar]
[42].Ran C, et al. A brainstem map for visceral sensations. Nature 2022;609(7926):320–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
[43].Chang RB, et al. Vagal sensory neuron subtypes that differentially control breathing. Cell 2015;161(3):622–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
[44].Pelot NA, Behrend CE, Grill WM. Modeling the response of small myelinated axons in a compound nerve to kilohertz frequency signals. J Neural Eng 2017;14(4):046022. [DOI] [PMC free article] [PubMed] [Google Scholar]
[45].Chang YC, et al. Quantitative estimation of nerve fiber engagement by vagus nerve stimulation using physiological markers. Brain Stimul 2020;13(6):1617–30. [DOI] [PubMed] [Google Scholar]
[46].Naples GG, et al. A spiral nerve cuff electrode for peripheral nerve stimulation. IEEE Trans Biomed Eng 1988;35(11):905–16. [DOI] [PubMed] [Google Scholar]
[47].Schuettler M, et al. Fabrication of implantable microelectrode arrays by laser cutting of silicone rubber and platinum foil. J Neural Eng 2005;2(1):S121–8. [DOI] [PubMed] [Google Scholar]
[48].Erman AB, et al. Disorders of cranial nerves IX and X. Semin Neurol 2009;29(1):85–92. [DOI] [PMC free article] [PubMed] [Google Scholar]
[49].Yuan H, Silberstein SD. Vagus nerve and vagus nerve stimulation, a comprehensive review: Part I. Headache 2016;56(1):71–8. [DOI] [PubMed] [Google Scholar]
[50].Weijs TJ, et al. Topography and extent of pulmonary vagus nerve supply with respect to transthoracic oesophagectomy. J Anat 2015;227(4):431–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
[51].He K, et al. Mask R-CNN. arXiv; 2017. p. 1703.06870. [Google Scholar]
[52].Lin T-Y, et al. Microsoft COCO: common objects in context. 2014. arXiv: 1405.0312. [Google Scholar]
[53].Gonzalez RC, Woods RE. Digital image processing. third ed. NJ: Pearson Prentice Hall; 2008. [Google Scholar]
[54].Haralick RM, Shanmugan K, Dinstein I. Textural features for image classification. IEEE Trans Syst Man Cybern 1973;3(6):610–21. SMC-. [Google Scholar]
[55].Jaccard P. The distribution of flora in the alpine zone. New Phytol 1912;11(2):37–50. [Google Scholar]
[56].Ester M, et al. A density-based algorithm for discovering clusters in large spatial databases with noise. In: KDD'96: proceedings of the second international conference on knowledge discovery and data mining. Portland, OR: AAAI; 1996. [Google Scholar]
[57].Zaqout S, Becker LL, Kaindl AM. Immunofluorescence staining of paraffin sections step by step. Front Neuroanat 2020;14:582218. [DOI] [PMC free article] [PubMed] [Google Scholar]
[58].Pristera A, Baker MD, Okuse K. Association between tetrodotoxin resistant channels and lipid rafts regulates sensory neuron excitability. PLoS One 2012;7(8):e40079. [DOI] [PMC free article] [PubMed] [Google Scholar]
[59].Mata M, Alessi D, Fink DJ. S100 is preferentially distributed in myelin-forming Schwann cells. J Neurocytol 1990;19(3):432–42. [DOI] [PubMed] [Google Scholar]
[60].Harty BL, Monk KR. Unwrapping the unappreciated: recent progress in Remak Schwann cell biology. Curr Opin Neurobiol 2017;47:131–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
[61].Toth V, et al. Single-axon level automatic segmentation and feature extraction from immuhistochemical images of peripheral nerves. Annu Int Conf IEEE Eng Med Biol Soc 2020;2020:1859–62. [DOI] [PubMed] [Google Scholar]
[62].Yuxin Wu AK, Massa Francisco, Lo Wan-Yen, Girshick Ross. Detectron2: A PyTorch-based modular object detection library. https://ai.facebook.com/blog/-detectron2-a-pytorch-based-modular-object-detection-library-/. [Google Scholar]
[63].Viktor Tóth NJ, Adam Abbas, Rob Moontahinaz, Khan Ariba, Zanos Stavros, Theodoros P. Zanos Single-axon level segmentation and feature extraction pipeline combining traditional and deep learning computer vision algorithms. Abstracts from the Fourth Bioelectronic Medicine Summit; 2020. p. P16. [Google Scholar]
[64].Schelegle ES. Functional morphology and physiology of slowly adapting pulmonary stretch receptors. Anat Rec A Discov Mol Cell Evol Biol 2003;270(1):11–6. [DOI] [PubMed] [Google Scholar]
[65].Yoo PB, et al. Modulation of heart rate by temporally patterned vagus nerve stimulation in the anesthetized dog. Phys Rep 2016;4(2). [DOI] [PMC free article] [PubMed] [Google Scholar]
[66].Dalrymple AN, et al. Electrochemical and biological performance of chronically stimulated conductive hydrogel electrodes. J Neural Eng 2020;17(2):026018. [DOI] [PubMed] [Google Scholar]
[67].Staples NA, et al. Conductive hydrogel electrodes for delivery of long-term high frequency pulses. Front Neurosci 2017;11:748. [DOI] [PMC free article] [PubMed] [Google Scholar]
[68].Ahmed U, et al. Implant- and anesthesia-related factors affecting cardiopulmonary threshold intensities for vagus nerve stimulation. J Neural Eng 2021;18(4). [DOI] [PubMed] [Google Scholar]
[69].Kalia M, Mesulam MM. Brain stem projections of sensory and motor components of the vagus complex in the cat: II. Laryngeal, tracheobronchial, pulmonary, cardiac, and gastrointestinal branches. J Comp Neurol 1980;193(2):467–508. [DOI] [PubMed] [Google Scholar]
[70].Prescott SL, Liberles SD. Internal senses of the vagus nerve. Neuron 2022;110(4):579–99. [DOI] [PMC free article] [PubMed] [Google Scholar]
[71].Zafeiropoulos S, et al. Autonomic neuromodulation for atrial fibrillation following cardiac surgery: JACC review topic of the week. J Am Coll Cardiol 2022;79(7):682–94. [DOI] [PubMed] [Google Scholar]
[72].Thompson N, et al. Organotopic organization of the cervical vagus nerve. bioRxiv; 2022. p. 2022. 02.24.481810. [Google Scholar]
[73].Geis GS, Wurster RD. Cardiac responses during stimulation of the dorsal motor nucleus and nucleus ambiguus in the cat. Circ Res 1980;46(5):606–11. [DOI] [PubMed] [Google Scholar]
[74].Broussard DL, Altschuler SM. Brainstem viscerotopic organization of afferents and efferents involved in the control of swallowing. Am J Med 2000;108(Suppl 4a):79S–86S. [DOI] [PubMed] [Google Scholar]
[75].Huang XF, Tork I, Paxinos G. Dorsal motor nucleus of the vagus nerve: a cyto- and chemoarchitectonic study in the human. J Comp Neurol 1993;330(2):158–82. [DOI] [PubMed] [Google Scholar]
[76].Upadhye AR, et al. Fascicles split or merge every ~560 microns within the human cervical vagus nerve. J Neural Eng 2022;19(5). 0.1088/1741-2552/ac964. [DOI] [PMC free article] [PubMed] [Google Scholar]
[77].Peclin P, Rozman J. Alternative paradigm of selective vagus nerve stimulation tested on an isolated porcine vagus nerve. Sci World J 2014;2014:310283. [DOI] [PMC free article] [PubMed] [Google Scholar]
[78].Settell ML, et al. In vivo visualization of pig vagus nerve "vagotopy" using ultrasound. Front Neurosci 2021;15:676680. [DOI] [PMC free article] [PubMed] [Google Scholar]
[79].Kenny BJ, Bordoni B. Neuroanatomy, cranial nerve 10 (vagus nerve). In: StatPearls; 2022. [Treasure Island (FL: ]. [PubMed] [Google Scholar]
[80].Jammes Y, et al. Afferent and efferent components of the bronchial vagal branches in cats. J Auton Nerv Syst 1982;5(2):165–76. [DOI] [PubMed] [Google Scholar]
[81].Schwaber JS, Cohen DH. Electrophysiological and electron microscopic analysis of the vagus nerve of the pigeon, with particular reference to the cardiac innervation. Brain Res 1978;147(1):65–78. [DOI] [PubMed] [Google Scholar]
[82].Onkka P, et al. Sympathetic nerve fibers and ganglia in canine cervical vagus nerves: localization and quantitation. Heart Rhythm 2013;10(4):585–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
[83].Bonaz B, Sinniger V, Pellissier S. The vagus nerve in the neuro-immune Axis: implications in the pathology of the gastrointestinal tract. Front Immunol 2017;8:1452. [DOI] [PMC free article] [PubMed] [Google Scholar]
[84].Kummer W, et al. Tyrosine-hydroxylase-containing vagal afferent neurons in the rat nodose ganglion are independent from neuropeptide-Y-containing populations and project to esophagus and stomach. Cell Tissue Res 1993;271(1):135 – 44. [DOI] [PubMed] [Google Scholar]
[85].Palouzier B, et al. Cholinergic neurons in the rat nodose ganglia. Neurosci Lett 1987;80(2):147–52. [DOI] [PubMed] [Google Scholar]
[86].Porzionato A, et al. The carotid sinus nerve-structure, function, and clinical implications. Anat Rec 2019;302(4):575–87. [DOI] [PubMed] [Google Scholar]
[87].Gautron L, et al. Genetic tracing of Nav1.8-expressing vagal afferents in the mouse. J Comp Neurol 2011;519(15):3085–101. [DOI] [PMC free article] [PubMed] [Google Scholar]
[88].Sun K, et al. Three-dimensional reconstruction and visualization of the median nerve from serial tissue sections. Microsurgery 2009;29(7):573–7. [DOI] [PubMed] [Google Scholar]
[89].Prats-Galino A, et al. 3D reconstruction of peripheral nerves from optical projection tomography images: a method for studying fascicular interconnections and intraneural plexuses. Clin Anat 2018;31(3):424–31. [DOI] [PubMed] [Google Scholar]
[90].Liu L, et al. Anatomy-aided deep learning for medical image segmentation: a review. Phys Med Biol 2021;66(11). [DOI] [PubMed] [Google Scholar]
[91].Horng MH, et al. DeepNerve: a new convolutional neural network for the localization and segmentation of the median nerve in ultrasound image sequences. Ultrasound Med Biol 2020;46(9):2439–52. [DOI] [PubMed] [Google Scholar]
[92].Tovbis D, et al. Automatic three-dimensional reconstruction of fascicles in peripheral nerves from histological images. PLoS One 2020;15(5):e0233028. [DOI] [PMC free article] [PubMed] [Google Scholar]
[93].Mayhew TM. Efficient and unbiased sampling of nerve fibers for estimating fiber number and size. In: Conn MP, editor. Quantitative and qualitative microscopy. NY: Academic Press; 1990. p. 172–87. [Google Scholar]
[94].Moffatt jD, Dumsday B, McLean JR. Characterization of non-adrenergic, non-cholinergic inhibitory responses of the isolated Guinea-pig trachea: differences between pre- and post-ganglionic nerve stimulation. Br J Pharmacol 1999;128(2):458–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
[95].Wu M, et al. Anatomical and neurochemical features of the extrinsic and intrinsic innervation of the striated muscle in the porcine esophagus: evidence for regional and species differences. Cell Tissue Res 2003;311(3):289–97. [DOI] [PubMed] [Google Scholar]
[96].Funakoshi K, et al. Monoaminergic and peptidergic axonal projections to the vagal motor cell column of a teleost, the filefish Stephanolepis cirrhifer. J Comp Neurol 2002;447(4):351–65. [DOI] [PubMed] [Google Scholar]
[97].Covenas R, et al. Mapping of neurokinin-like immunoreactivity in the human brainstem. BMC Neurosci 2003;4:3. [DOI] [PMC free article] [PubMed] [Google Scholar]
[98].Sapio MR, et al. Comparative analysis of dorsal root, nodose and sympathetic ganglia for the development of new analgesics. Front Neurosci 2020;14:615362. [DOI] [PMC free article] [PubMed] [Google Scholar]
[99].Murinson BB, Griffin JW. C-fiber structure varies with location in peripheral nerve. J Neuropathol Exp Neurol 2004;63(3):246–54. [DOI] [PubMed] [Google Scholar]
[100].Ternaux JP, et al. Presence of cholinergic neurons in the vagal afferent system: biochemical and immunohistochemical approaches. J Auton Nerv Syst 1989;28(3):233–42. [DOI] [PubMed] [Google Scholar]
[101].Andrei F, Didilescu AC, Rusu MC. Evidence of nNOS and ChAT positive phenotypes in nervous ganglia of the retrostyloid space. J Med Life 2012;5(4):482–5. [PMC free article] [PubMed] [Google Scholar]
[102].Cai QQ et al. Distribution of dopamine receptors D1- and D2-immunoreactive neurons in the dorsal motor nucleus of vagus in rats. Auton Neurosci 2013;176(1–2):48–53. [DOI] [PubMed] [Google Scholar]
[103].Nishiyama K, et al. Tyrosine hydroxylase and NADPH-diaphorase in the rat nodose ganglion: colocalization and central projection. Acta Histochem Cytoc 2001;34(2):135–41. [Google Scholar]
[104].Kupari J, et al. An atlas of vagal sensory neurons and their molecular specialization. Cell Rep 2019;27(8):2508–2523 e4. [DOI] [PMC free article] [PubMed] [Google Scholar]
[105].Muryobayashi T, et al. Fluorescence histochemical demonstration of adrenergic nerve fibers in the vagus nerve of cats and dogs. Jpn J Pharmacol 1968;18(3):285–93. [DOI] [PubMed] [Google Scholar]
[106].Porges SW. The polyvagal perspective. Biol Psychol 2007;74(2):116–43. [DOI] [PMC free article] [PubMed] [Google Scholar]
[107].Rattay F, Tanzer T. A simple model considering spiking probability during extracellular axon stimulation. PLoS One 2022;17(4):e0264735. [DOI] [PMC free article] [PubMed] [Google Scholar]
[108].Bhadra N, et al. Simulation of high-frequency sinusoidal electrical block of mammalian myelinated axons. J Comput Neurosci 2007;22(3):313–26. [DOI] [PubMed] [Google Scholar]
[109].Lertmanorat Z, Durand DM. Extracellular voltage profile for reversing the recruitment order of peripheral nerve stimulation: a simulation study. J Neural Eng 2004;1(4):202–11. [DOI] [PubMed] [Google Scholar]
[110].Armour JA, Randall WC. Functional anatomy of canine cardiac nerves. Acta Anat 1975;91(4):510–28. [DOI] [PubMed] [Google Scholar]
[111].Gierthmuehlen M, et al. Haemodynamic responses to selective vagal nerve stimulation under enalapril medication in rats. PLoS One 2016;11(1):e0147045. [DOI] [PMC free article] [PubMed] [Google Scholar]
[112].FDA, PMA P970003/S207. FDA Summary of Safety and effectiveness data FDA. White Oak, MD: FDA; 2017. p. 1–30. [Google Scholar]
[113].FDA. K183437, 510(K) premarket notification of subtantial equivalence. White Oak, MD: FDA; 2019. [Google Scholar]
[114].Blanz SL, et al. Spatially selective stimulation of the pig vagus nerve to modulate target effect versus side effect. bioRxiv; 2022. p. 2022. 05.19.492726. [DOI] [PMC free article] [PubMed] [Google Scholar]
[115].Ardell JL, et al. Defining the neural fulcrum for chronic vagus nerve stimulation: implications for integrated cardiac control. J Physiol 2017;595(22):6887–903. [DOI] [PMC free article] [PubMed] [Google Scholar]
[116].Ben-Menachem E. Vagus nerve stimulation, side effects, and long-term safety. J Clin Neurophysiol 2001;18(5):415–8. [DOI] [PubMed] [Google Scholar]
[117].Mughrabi IT, et al. Development and characterization of a chronic implant mouse model for vagus nerve stimulation. Elife 2021;10. [DOI] [PMC free article] [PubMed] [Google Scholar]
[118].Ordelman SC, et al. An indirect component in the evoked compound action potential of the vagal nerve. J Neural Eng 2010;7(6):066001. [DOI] [PubMed] [Google Scholar]
[119].Evans MS, et al. Intraoperative human vagus nerve compound action potentials. Acta Neurol Scand 2004;110(4):232–8. [DOI] [PubMed] [Google Scholar]
[120].Ahmed U, et al. Strategies for precision vagus neuromodulation. Bioelectron Med 2022;8(1):9. [DOI] [PMC free article] [PubMed] [Google Scholar]
[121].Aristovich K, et al. Imaging fast neural traffic at fascicular level with electrical impedance tomography: proof of principle in rat sciatic nerve. J Neural Eng 2018;15(5):056025. [DOI] [PubMed] [Google Scholar]
[122].Ravagli E, et al. Optimization of the electrode drive pattern for imaging fascicular compound action potentials in peripheral nerve with fast neural electrical impedance tomography. Physiol Meas 2019;40(11):115007. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

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

Supplementary Materials

NIHMS1898332-supplement-1.docx^{(44.6MB, docx)}

Download video file^{(9MB, mp4)}

Download video file^{(715.1KB, mp4)}

[R1] [1].Stakenborg N, et al. Comparison between the cervical and abdominal vagus nerves in mice, pigs, and humans. Neuro Gastroenterol Motil 2020;32(9):e13889. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R2] [2].Pelot NA, et al. Quantified morphology of the cervical and subdiaphragmatic vagus nerves of human, pig, and rat. Front Neurosci 2020;14:601479. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R3] [3].Thompson N, Mastitskaya S, Holder D. Avoiding off-target effects in electrical stimulation of the cervical vagus nerve: neuroanatomical tracing techniques to study fascicular anatomy of the vagus nerve. J Neurosci Methods 2019;325:108325. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R4] [4].Kenny BJ, Bordoni B. Neuroanatomy, cranial nerve 10 (vagus nerve). In: StatPearls; 2021. [Treasure Island (FL: ]. [PubMed] [Google Scholar]

[R5] [5].Settell ML, et al. Functional vagotopy in the cervical vagus nerve of the domestic pig: implications for the study of vagus nerve stimulation. J Neural Eng 2020;17(2):026022. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R6] [6].Agostoni E, et al. Functional and histological studies of the vagus nerve and its branches to the heart, lungs and abdominal viscera in the cat. J Physiol 1957;135(1):182–205. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R7] [7].Hammer N, et al. Human vagus nerve branching in the cervical region. PLoS One 2015;10(2):e0118006. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R8] [8].Verlinden TJ, et al. Morphology of the human cervical vagus nerve: implications for vagus nerve stimulation treatment. Acta Neurol Scand 2016;133(3):173–82. [DOI] [PubMed] [Google Scholar]

[R9] [9].Yoshida Y, et al. Sensory innervation of the pharynx and larynx. Am J Med 2000;108(Suppl 4a):51S–61S. [DOI] [PubMed] [Google Scholar]

[R10] [10].Thompson N, et al. MicroCT optimisation for imaging fascicular anatomy in peripheral nerves. J Neurosci Methods 2020;338:108652. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R11] [11].Asala SA, Bower AJ. An electron microscope study of vagus nerve composition in the ferret. Anat Embryol 1986;175(2):247–53. [DOI] [PubMed] [Google Scholar]

[R12] [12].Gabella G, Pease HL. Number of axons in the abdominal vagus of the rat. Brain Res 1973;58(2):465–9. [DOI] [PubMed] [Google Scholar]

[R13] [13].Havton LA, et al. Human organ donor-derived vagus nerve biopsies allow for well-preserved ultrastructure and high-resolution mapping of myelinated and unmyelinated fibers. Sci Rep 2021;11(1):23831. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R14] [14].Hoffman HH, Schnitzlein HN. The numbers of nerve fibers in the vagus nerve of man. Anat Rec 1961;139:429–35. [DOI] [PubMed] [Google Scholar]

[R15] [15].Kawagishi K, et al. Tyrosine hydroxylase-immunoreactive fibers in the human vagus nerve. J Clin Neurosci 2008;15(9):1023–6. [DOI] [PubMed] [Google Scholar]

[R16] [16].Kollarik M, Ru F, Brozmanova M. Vagal afferent nerves with the properties of nociceptors. Auton Neurosci 2010;153(1–2):12–20. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R17] [17].Kressel M, Radespiel-Troger M. Anterograde tracing and immunohistochemical characterization of potentially mechanosensitive vagal afferents in the esophagus. J Comp Neurol 1999;412(1):161–72. [DOI] [PubMed] [Google Scholar]

[R18] [18].Lundberg J, et al. Catecholamine-containing nerve fibres in the human abdominal vagus. Gastroenterology 1976;70(3):472–4. [PubMed] [Google Scholar]

[R19] [19].Mazzone SB, Undem BJ. Vagal afferent innervation of the airways in health and disease. Physiol Rev 2016;96(3):975–1024. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R20] [20].Mei N, Condamin M, Boyer A. The composition of the vagus nerve of the cat. Cell Tissue Res 1980;209(3):423–31. [DOI] [PubMed] [Google Scholar]

[R21] [21].Schemann M, et al. Identification of cholinergic neurons in enteric nervous system by antibodies against choline acetyltransferase. Am J Physiol 1993;265(5 Pt 1):G1005–9. [DOI] [PubMed] [Google Scholar]

[R22] [22].Seki A, et al. Sympathetic nerve fibers in human cervical and thoracic vagus nerves. Heart Rhythm 2014;11(8):1411–7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R23] [23].Yang M, Zhao X, Miselis RR. The origin of catecholaminergic nerve fibers in the subdiaphragmatic vagus nerve of rat. J Auton Nerv Syst 1999;76(2–3):108–17. [DOI] [PubMed] [Google Scholar]

[R24] [24].de Lartigue G. Putative roles of neuropeptides in vagal afferent signaling. Physiol Behav 2014;136:155–69. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R25] [25].Berthoud HR, Neuhuber WL. Functional and chemical anatomy of the afferent vagal system. Auton Neurosci 2000;85(1–3):1–17. [DOI] [PubMed] [Google Scholar]

[R26] [26].De Ferrari GM, et al. Long-term vagal stimulation for heart failure: eighteen month results from the NEural Cardiac TherApy foR Heart Failure (NECTAR-HF) trial. Int J Cardiol 2017;244:229–34. [DOI] [PubMed] [Google Scholar]

[R27] [27].Mikami Y, et al. Vagus nerve-mediated intestinal immune regulation: therapeutic implications of inflammatory bowel diseases. Int Immunol 2022;34(2):97–106. [DOI] [PubMed] [Google Scholar]

[R28] [28].Koopman FA, et al. Vagus nerve stimulation inhibits cytokine production and attenuates disease severity in rheumatoid arthritis. Proc Natl Acad Sci U S A 2016;113(29):8284–9. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R29] [29].Donega M, et al. Human-relevant near-organ neuromodulation of the immune system via the splenic nerve. Proc Natl Acad Sci U S A 2021;118(20). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R30] [30].Nicolai EN, et al. Sources of off-target effects of vagus nerve stimulation using the helical clinical lead in domestic pigs. J Neural Eng 2020;17(4):046017. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R31] [31].Musselman ED, Pelot NA, Grill WM. Empirically based guidelines for selecting vagus nerve stimulation parameters in epilepsy and heart failure. Cold Spring Harb Perspect Med 2019;9(7). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R32] [32].Booth LC, et al. Selective optogenetic stimulation of efferent fibers in the vagus nerve of a large mammal. Brain Stimul 2021;14(1):88–96. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R33] [33].Fontaine AK, et al. Optogenetic stimulation of cholinergic fibers for the modulation of insulin and glycemia. Sci Rep 2021;11(1):3670. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R34] [34].Tanaka S, et al. Vagus nerve stimulation activates two distinct neuroimmune circuits converging in the spleen to protect mice from kidney injury. Proc Natl Acad Sci U S A 2021;118(12). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R35] [35].Plachta DT, et al. Blood pressure control with selective vagal nerve stimulation and minimal side effects. J Neural Eng 2014;11(3):036011. [DOI] [PubMed] [Google Scholar]

[R36] [36].Kong SH, et al. Intraoperative neurophysiologic testing of the perigastric vagus nerve branches to evaluate viability and signals along nerve pathways during gastrectomy. J Gastric Cancer 2019;19(1):49–61. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R37] [37].Furukawa N, Okada H. Effects of selective vagal stimulation on the gallbladder and sphincter of Oddi and peripheral vagal routes mediating bile evacuative responses induced by hypothalamic stimulation. Jpn J Physiol 1992;42(2):321–34. [DOI] [PubMed] [Google Scholar]

[R38] [38].Fitchett A, Mastitskaya S, Aristovich K. Selective neuromodulation of the vagus nerve. Front Neurosci 2021;15:685872. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R39] [39].Aristovich K, et al. Model-based geometrical optimisation and in vivo validation of a spatially selective multielectrode cuff array for vagus nerve neuromodulation. J Neurosci Methods 2021;352:109079. [DOI] [PubMed] [Google Scholar]

[R40] [40].Musselman ED, et al. ASCENT (Automated Simulations to Characterize Electrical Nerve Thresholds): a pipeline for sample-specific computational modeling of electrical stimulation of peripheral nerves. PLoS Comput Biol 2021;17(9):e1009285. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R41] [41].Zhao Q, et al. A multidimensional coding architecture of the vagal interoceptive system. Nature 2022;603(7903):878–84. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R42] [42].Ran C, et al. A brainstem map for visceral sensations. Nature 2022;609(7926):320–6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R43] [43].Chang RB, et al. Vagal sensory neuron subtypes that differentially control breathing. Cell 2015;161(3):622–33. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R44] [44].Pelot NA, Behrend CE, Grill WM. Modeling the response of small myelinated axons in a compound nerve to kilohertz frequency signals. J Neural Eng 2017;14(4):046022. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R45] [45].Chang YC, et al. Quantitative estimation of nerve fiber engagement by vagus nerve stimulation using physiological markers. Brain Stimul 2020;13(6):1617–30. [DOI] [PubMed] [Google Scholar]

[R46] [46].Naples GG, et al. A spiral nerve cuff electrode for peripheral nerve stimulation. IEEE Trans Biomed Eng 1988;35(11):905–16. [DOI] [PubMed] [Google Scholar]

[R47] [47].Schuettler M, et al. Fabrication of implantable microelectrode arrays by laser cutting of silicone rubber and platinum foil. J Neural Eng 2005;2(1):S121–8. [DOI] [PubMed] [Google Scholar]

[R48] [48].Erman AB, et al. Disorders of cranial nerves IX and X. Semin Neurol 2009;29(1):85–92. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R49] [49].Yuan H, Silberstein SD. Vagus nerve and vagus nerve stimulation, a comprehensive review: Part I. Headache 2016;56(1):71–8. [DOI] [PubMed] [Google Scholar]

[R50] [50].Weijs TJ, et al. Topography and extent of pulmonary vagus nerve supply with respect to transthoracic oesophagectomy. J Anat 2015;227(4):431–9. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R51] [51].He K, et al. Mask R-CNN. arXiv; 2017. p. 1703.06870. [Google Scholar]

[R52] [52].Lin T-Y, et al. Microsoft COCO: common objects in context. 2014. arXiv: 1405.0312. [Google Scholar]

[R53] [53].Gonzalez RC, Woods RE. Digital image processing. third ed. NJ: Pearson Prentice Hall; 2008. [Google Scholar]

[R54] [54].Haralick RM, Shanmugan K, Dinstein I. Textural features for image classification. IEEE Trans Syst Man Cybern 1973;3(6):610–21. SMC-. [Google Scholar]

[R55] [55].Jaccard P. The distribution of flora in the alpine zone. New Phytol 1912;11(2):37–50. [Google Scholar]

[R56] [56].Ester M, et al. A density-based algorithm for discovering clusters in large spatial databases with noise. In: KDD'96: proceedings of the second international conference on knowledge discovery and data mining. Portland, OR: AAAI; 1996. [Google Scholar]

[R57] [57].Zaqout S, Becker LL, Kaindl AM. Immunofluorescence staining of paraffin sections step by step. Front Neuroanat 2020;14:582218. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R58] [58].Pristera A, Baker MD, Okuse K. Association between tetrodotoxin resistant channels and lipid rafts regulates sensory neuron excitability. PLoS One 2012;7(8):e40079. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R59] [59].Mata M, Alessi D, Fink DJ. S100 is preferentially distributed in myelin-forming Schwann cells. J Neurocytol 1990;19(3):432–42. [DOI] [PubMed] [Google Scholar]

[R60] [60].Harty BL, Monk KR. Unwrapping the unappreciated: recent progress in Remak Schwann cell biology. Curr Opin Neurobiol 2017;47:131–7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R61] [61].Toth V, et al. Single-axon level automatic segmentation and feature extraction from immuhistochemical images of peripheral nerves. Annu Int Conf IEEE Eng Med Biol Soc 2020;2020:1859–62. [DOI] [PubMed] [Google Scholar]

[R62] [62].Yuxin Wu AK, Massa Francisco, Lo Wan-Yen, Girshick Ross. Detectron2: A PyTorch-based modular object detection library. https://ai.facebook.com/blog/-detectron2-a-pytorch-based-modular-object-detection-library-/. [Google Scholar]

[R63] [63].Viktor Tóth NJ, Adam Abbas, Rob Moontahinaz, Khan Ariba, Zanos Stavros, Theodoros P. Zanos Single-axon level segmentation and feature extraction pipeline combining traditional and deep learning computer vision algorithms. Abstracts from the Fourth Bioelectronic Medicine Summit; 2020. p. P16. [Google Scholar]

[R64] [64].Schelegle ES. Functional morphology and physiology of slowly adapting pulmonary stretch receptors. Anat Rec A Discov Mol Cell Evol Biol 2003;270(1):11–6. [DOI] [PubMed] [Google Scholar]

[R65] [65].Yoo PB, et al. Modulation of heart rate by temporally patterned vagus nerve stimulation in the anesthetized dog. Phys Rep 2016;4(2). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R66] [66].Dalrymple AN, et al. Electrochemical and biological performance of chronically stimulated conductive hydrogel electrodes. J Neural Eng 2020;17(2):026018. [DOI] [PubMed] [Google Scholar]

[R67] [67].Staples NA, et al. Conductive hydrogel electrodes for delivery of long-term high frequency pulses. Front Neurosci 2017;11:748. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R68] [68].Ahmed U, et al. Implant- and anesthesia-related factors affecting cardiopulmonary threshold intensities for vagus nerve stimulation. J Neural Eng 2021;18(4). [DOI] [PubMed] [Google Scholar]

[R69] [69].Kalia M, Mesulam MM. Brain stem projections of sensory and motor components of the vagus complex in the cat: II. Laryngeal, tracheobronchial, pulmonary, cardiac, and gastrointestinal branches. J Comp Neurol 1980;193(2):467–508. [DOI] [PubMed] [Google Scholar]

[R70] [70].Prescott SL, Liberles SD. Internal senses of the vagus nerve. Neuron 2022;110(4):579–99. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R71] [71].Zafeiropoulos S, et al. Autonomic neuromodulation for atrial fibrillation following cardiac surgery: JACC review topic of the week. J Am Coll Cardiol 2022;79(7):682–94. [DOI] [PubMed] [Google Scholar]

[R72] [72].Thompson N, et al. Organotopic organization of the cervical vagus nerve. bioRxiv; 2022. p. 2022. 02.24.481810. [Google Scholar]

[R73] [73].Geis GS, Wurster RD. Cardiac responses during stimulation of the dorsal motor nucleus and nucleus ambiguus in the cat. Circ Res 1980;46(5):606–11. [DOI] [PubMed] [Google Scholar]

[R74] [74].Broussard DL, Altschuler SM. Brainstem viscerotopic organization of afferents and efferents involved in the control of swallowing. Am J Med 2000;108(Suppl 4a):79S–86S. [DOI] [PubMed] [Google Scholar]

[R75] [75].Huang XF, Tork I, Paxinos G. Dorsal motor nucleus of the vagus nerve: a cyto- and chemoarchitectonic study in the human. J Comp Neurol 1993;330(2):158–82. [DOI] [PubMed] [Google Scholar]

[R76] [76].Upadhye AR, et al. Fascicles split or merge every ~560 microns within the human cervical vagus nerve. J Neural Eng 2022;19(5). 0.1088/1741-2552/ac964. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R77] [77].Peclin P, Rozman J. Alternative paradigm of selective vagus nerve stimulation tested on an isolated porcine vagus nerve. Sci World J 2014;2014:310283. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R78] [78].Settell ML, et al. In vivo visualization of pig vagus nerve "vagotopy" using ultrasound. Front Neurosci 2021;15:676680. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R79] [79].Kenny BJ, Bordoni B. Neuroanatomy, cranial nerve 10 (vagus nerve). In: StatPearls; 2022. [Treasure Island (FL: ]. [PubMed] [Google Scholar]

[R80] [80].Jammes Y, et al. Afferent and efferent components of the bronchial vagal branches in cats. J Auton Nerv Syst 1982;5(2):165–76. [DOI] [PubMed] [Google Scholar]

[R81] [81].Schwaber JS, Cohen DH. Electrophysiological and electron microscopic analysis of the vagus nerve of the pigeon, with particular reference to the cardiac innervation. Brain Res 1978;147(1):65–78. [DOI] [PubMed] [Google Scholar]

[R82] [82].Onkka P, et al. Sympathetic nerve fibers and ganglia in canine cervical vagus nerves: localization and quantitation. Heart Rhythm 2013;10(4):585–91. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R83] [83].Bonaz B, Sinniger V, Pellissier S. The vagus nerve in the neuro-immune Axis: implications in the pathology of the gastrointestinal tract. Front Immunol 2017;8:1452. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R84] [84].Kummer W, et al. Tyrosine-hydroxylase-containing vagal afferent neurons in the rat nodose ganglion are independent from neuropeptide-Y-containing populations and project to esophagus and stomach. Cell Tissue Res 1993;271(1):135 – 44. [DOI] [PubMed] [Google Scholar]

[R85] [85].Palouzier B, et al. Cholinergic neurons in the rat nodose ganglia. Neurosci Lett 1987;80(2):147–52. [DOI] [PubMed] [Google Scholar]

[R86] [86].Porzionato A, et al. The carotid sinus nerve-structure, function, and clinical implications. Anat Rec 2019;302(4):575–87. [DOI] [PubMed] [Google Scholar]

[R87] [87].Gautron L, et al. Genetic tracing of Nav1.8-expressing vagal afferents in the mouse. J Comp Neurol 2011;519(15):3085–101. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R88] [88].Sun K, et al. Three-dimensional reconstruction and visualization of the median nerve from serial tissue sections. Microsurgery 2009;29(7):573–7. [DOI] [PubMed] [Google Scholar]

[R89] [89].Prats-Galino A, et al. 3D reconstruction of peripheral nerves from optical projection tomography images: a method for studying fascicular interconnections and intraneural plexuses. Clin Anat 2018;31(3):424–31. [DOI] [PubMed] [Google Scholar]

[R90] [90].Liu L, et al. Anatomy-aided deep learning for medical image segmentation: a review. Phys Med Biol 2021;66(11). [DOI] [PubMed] [Google Scholar]

[R91] [91].Horng MH, et al. DeepNerve: a new convolutional neural network for the localization and segmentation of the median nerve in ultrasound image sequences. Ultrasound Med Biol 2020;46(9):2439–52. [DOI] [PubMed] [Google Scholar]

[R92] [92].Tovbis D, et al. Automatic three-dimensional reconstruction of fascicles in peripheral nerves from histological images. PLoS One 2020;15(5):e0233028. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R93] [93].Mayhew TM. Efficient and unbiased sampling of nerve fibers for estimating fiber number and size. In: Conn MP, editor. Quantitative and qualitative microscopy. NY: Academic Press; 1990. p. 172–87. [Google Scholar]

[R94] [94].Moffatt jD, Dumsday B, McLean JR. Characterization of non-adrenergic, non-cholinergic inhibitory responses of the isolated Guinea-pig trachea: differences between pre- and post-ganglionic nerve stimulation. Br J Pharmacol 1999;128(2):458–64. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R95] [95].Wu M, et al. Anatomical and neurochemical features of the extrinsic and intrinsic innervation of the striated muscle in the porcine esophagus: evidence for regional and species differences. Cell Tissue Res 2003;311(3):289–97. [DOI] [PubMed] [Google Scholar]

[R96] [96].Funakoshi K, et al. Monoaminergic and peptidergic axonal projections to the vagal motor cell column of a teleost, the filefish Stephanolepis cirrhifer. J Comp Neurol 2002;447(4):351–65. [DOI] [PubMed] [Google Scholar]

[R97] [97].Covenas R, et al. Mapping of neurokinin-like immunoreactivity in the human brainstem. BMC Neurosci 2003;4:3. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R98] [98].Sapio MR, et al. Comparative analysis of dorsal root, nodose and sympathetic ganglia for the development of new analgesics. Front Neurosci 2020;14:615362. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R99] [99].Murinson BB, Griffin JW. C-fiber structure varies with location in peripheral nerve. J Neuropathol Exp Neurol 2004;63(3):246–54. [DOI] [PubMed] [Google Scholar]

[R100] [100].Ternaux JP, et al. Presence of cholinergic neurons in the vagal afferent system: biochemical and immunohistochemical approaches. J Auton Nerv Syst 1989;28(3):233–42. [DOI] [PubMed] [Google Scholar]

[R101] [101].Andrei F, Didilescu AC, Rusu MC. Evidence of nNOS and ChAT positive phenotypes in nervous ganglia of the retrostyloid space. J Med Life 2012;5(4):482–5. [PMC free article] [PubMed] [Google Scholar]

[R102] [102].Cai QQ et al. Distribution of dopamine receptors D1- and D2-immunoreactive neurons in the dorsal motor nucleus of vagus in rats. Auton Neurosci 2013;176(1–2):48–53. [DOI] [PubMed] [Google Scholar]

[R103] [103].Nishiyama K, et al. Tyrosine hydroxylase and NADPH-diaphorase in the rat nodose ganglion: colocalization and central projection. Acta Histochem Cytoc 2001;34(2):135–41. [Google Scholar]

[R104] [104].Kupari J, et al. An atlas of vagal sensory neurons and their molecular specialization. Cell Rep 2019;27(8):2508–2523 e4. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R105] [105].Muryobayashi T, et al. Fluorescence histochemical demonstration of adrenergic nerve fibers in the vagus nerve of cats and dogs. Jpn J Pharmacol 1968;18(3):285–93. [DOI] [PubMed] [Google Scholar]

[R106] [106].Porges SW. The polyvagal perspective. Biol Psychol 2007;74(2):116–43. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R107] [107].Rattay F, Tanzer T. A simple model considering spiking probability during extracellular axon stimulation. PLoS One 2022;17(4):e0264735. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R108] [108].Bhadra N, et al. Simulation of high-frequency sinusoidal electrical block of mammalian myelinated axons. J Comput Neurosci 2007;22(3):313–26. [DOI] [PubMed] [Google Scholar]

[R109] [109].Lertmanorat Z, Durand DM. Extracellular voltage profile for reversing the recruitment order of peripheral nerve stimulation: a simulation study. J Neural Eng 2004;1(4):202–11. [DOI] [PubMed] [Google Scholar]

[R110] [110].Armour JA, Randall WC. Functional anatomy of canine cardiac nerves. Acta Anat 1975;91(4):510–28. [DOI] [PubMed] [Google Scholar]

[R111] [111].Gierthmuehlen M, et al. Haemodynamic responses to selective vagal nerve stimulation under enalapril medication in rats. PLoS One 2016;11(1):e0147045. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R112] [112].FDA, PMA P970003/S207. FDA Summary of Safety and effectiveness data FDA. White Oak, MD: FDA; 2017. p. 1–30. [Google Scholar]

[R113] [113].FDA. K183437, 510(K) premarket notification of subtantial equivalence. White Oak, MD: FDA; 2019. [Google Scholar]

[R114] [114].Blanz SL, et al. Spatially selective stimulation of the pig vagus nerve to modulate target effect versus side effect. bioRxiv; 2022. p. 2022. 05.19.492726. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R115] [115].Ardell JL, et al. Defining the neural fulcrum for chronic vagus nerve stimulation: implications for integrated cardiac control. J Physiol 2017;595(22):6887–903. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R116] [116].Ben-Menachem E. Vagus nerve stimulation, side effects, and long-term safety. J Clin Neurophysiol 2001;18(5):415–8. [DOI] [PubMed] [Google Scholar]

[R117] [117].Mughrabi IT, et al. Development and characterization of a chronic implant mouse model for vagus nerve stimulation. Elife 2021;10. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R118] [118].Ordelman SC, et al. An indirect component in the evoked compound action potential of the vagal nerve. J Neural Eng 2010;7(6):066001. [DOI] [PubMed] [Google Scholar]

[R119] [119].Evans MS, et al. Intraoperative human vagus nerve compound action potentials. Acta Neurol Scand 2004;110(4):232–8. [DOI] [PubMed] [Google Scholar]

[R120] [120].Ahmed U, et al. Strategies for precision vagus neuromodulation. Bioelectron Med 2022;8(1):9. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R121] [121].Aristovich K, et al. Imaging fast neural traffic at fascicular level with electrical impedance tomography: proof of principle in rat sciatic nerve. J Neural Eng 2018;15(5):056025. [DOI] [PubMed] [Google Scholar]

[R122] [122].Ravagli E, et al. Optimization of the electrode drive pattern for imaging fascicular compound action potentials in peripheral nerve with fast neural electrical impedance tomography. Physiol Meas 2019;40(11):115007. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Organ- and function-specific anatomical organization of vagal fibers supports fascicular vagus nerve stimulation

Naveen Jayaprakash

Weiguo Song

Viktor Toth

Avantika Vardhan

Todd Levy

Jacquelyn Tomaio

Khaled Qanud

Ibrahim Mughrabi

Yao-Chuan Chang

Moontahinaz Rob

Anna Daytz

Adam Abbas

Zeinab Nassrallah

Bruce T Volpe

Kevin J Tracey

Yousef Al-Abed

Timir Datta-Chaudhuri

Larry Miller

Mary F Barbe

Sunhee C Lee

Theodoros P Zanos

Stavros Zanos

Abstract

1. Introduction

2. Methods

2.1. Animals and animal surgeries

2.1.1. Animals and nerve samples used in different studies

2.1.2. Animal care during terminal surgeries

2.1.3. Animal care during survival surgeries

2.2. Cervical vagus nerve and laryngeal muscle implants

Fig. 7. Fascicular vagus nerve stimulation produces asymmetric organ-specific physiological responses and fiber-specific evoked nerve potentials.

2.3. Multi-contact cuff electrode device

2.3.1. Methodology to specify contact configuration in cuff electrode

2.3.2. Electrode fabrication

Fig. 5. A multi-contact cuff electrode for fascicular stimulation of the cervical vagus nerve.

2.3.3. Electrode characterization

Cleaning of electrodes:

Electrochemical impedance spectroscopy (EIS):

Charge storage capacity (CSC):

2.4. Physiology experiments, vagus nerve stimulation and physiological data analysis

2.4.1. Physiological and neural signals

Breathing rate:

Heart rate:

Neural and EMG signals:

2.4.2. Physiological thresholds of VNS

2.4.3. Evoked compound action potentials (eCAPs)

Fig. 6. Physiological and evoked nerve potential responses to fascicular vagus nerve stimulation.

2.4.4. Analyses of physiological responses across contacts and animals

Recruitment maps of physiological and eCAP responses.

Statistical assessment of asymmetry of responses to fascicular VNS.

Visualization of asymmetry of responses to fascicular VNS.

2.5. Surgical dissection and micro-computed tomography of nerve samples

Fig. 1. Fascicles form organ-specific clusters along the trunk of the swine vagus nerve.

Surgical dissection of swine nerve samples.

Micro-CT imaging of swine nerve samples.

Micro-CT video rendering.

Surgical dissection and micro-CT imaging of a human nerve sample:

2.6. Segmentation and tracking of fascicles in micro-CT data

2.6.1. Segmentation of fascicles

2.6.2. Tracking of fascicles

2.7. Quantitative immunohistochemistry of nerve samples

2.7.1. Sectioning of samples

Fig. 3. Morphologically distinct fiber types are organized across and within fascicles in the cervical vagus nerve of the swine in a specific, nonuniform pattern.

2.7.2. Staining and imaging of sections

2.7.3. Quality controls and validation of IHC

2.8. Fiber segmentation and feature extraction from IHC images

2.8.1. Single fiber identification and characterization

Neurofilament:

Myelin:

ChAT:

Fascicles:

Nav1.8:

Table 1.

2.8.2. Population analyses of single fiber data

Fiber counts and areas occupied.

Spatial distributions of fibers within fascicles.

Distribution of fibers at different distances from the epineurium.

Manual validations.