Designing a bioelectronic treatment for Type 1 diabetes: targeted parasympathetic modulation of insulin secretion

Abstract

The pancreas is a visceral organ with exocrine functions for digestion and endocrine functions for maintenance of blood glucose homeostasis. In pancreatic diseases such as Type 1 diabetes, islets of the endocrine pancreas become dysfunctional and normal regulation of blood glucose concentration ceases. In healthy individuals, parasympathetic signaling to islets via the vagus nerve, triggers release of insulin from pancreatic β-cells and glucagon from α-cells. Using electrical stimulation to augment parasympathetic signaling may provide a way to control pancreatic endocrine functions and ultimately control blood glucose. Historical data suggest that cervical vagus nerve stimulation recruits many visceral organ systems. Simultaneous modulation of liver and digestive function along with pancreatic function provides differential signals that work to both raise and lower blood glucose. Targeted pancreatic vagus nerve stimulation may provide a solution to minimizing off-target effects through careful electrode placement just prior to pancreatic insertion.

The pancreas is an organ which works to maintain metabolic homeostasis, most notably maintenance of a euglycemic state. This control of energy is maintained through multiple signaling pathways that control the release of hormones from the islets of Langerhans (Figure 1) [1]. These functional units are well-vascularized, innervated structures, which are made up of multiple endocrine types and supporting cells [2]. The β-cells are the most numerous within islets and respond to increased blood glucose concentrations by secreting insulin, a hormone which allows for glucose to be transported into liver and muscle cells for use or storage [3]. This delicate control mechanism over blood glucose requires signaling that has both a high degree of accuracy and temporal resolution which is achieved through both local chemical and neural signaling.

Figure 1. . Parasympathetic control over insulin release.

Type 1 diabetes results in the autoimmune destruction of many insulin secreting β-cells. One way these cells respond to systemic glucose needs is through neural signaling. Acetylcholine released from the vagus nerve signals β-cells through the M3 receptor activating both calcium dependent and independent release of insulin. In T1D, remaining β-cells can receive exogenous signaling through vagus nerve stimulation to increase the amount of insulin secreted.

In Type 1 diabetes (T1D), an autoimmune attack occurs over the course of several years which ultimately results in loss of sufficient β-cells to disrupt the ability to maintain glucose homeostasis and hyperglycemia ensues [4]. T1D leads to permanent hyperglycemia due to the loss of a majority of β-cells [5,6]. Over 1.6 million Americans have T1D and require life-long replacement insulin therapy [7]. Much progress has been made in T1D management; however, there remains an opportunity to better treat the disease. For instance, β-cells have been observed in islets of T1D patients decades after disease onset [8,9]. As such, novel therapies may be proposed to protect and utilize these remaining β-cells thereby improving chances for β-cell regeneration.

The emerging field of bioelectronic medicine offers new opportunities to treat diseases through modulation of natural organ function. Bioelectronic medicine relies on the use of neural interfaces to apply electrical stimulation to neural tissue resulting in modulation of organ function for the treatment of disease [10]. A common neural target in bioelectronic medicine is the vagus nerve, or cranial nerve X. The vagus nerve is multifunctional containing the parasympathetic motor and visceral sensory axons to many visceral organs. While all of the functions of the nerve are not fully understood, it is commonly accepted that the nerve contains between 65 and 90% afferent fibers [11–14]. Devices for vagus nerve stimulation (VNS) have been US FDA approved for implantation at the cervical level to treat epilepsy for decades- the first device approved for the treatment of epilepsy in 1997 [15]. Since then, VNS has also been used to treat patients for headache, depression and obesity with over 110,000 devices already implanted in patients [16]. As the pancreas is a target of the vagus nerve, there is a theoretical mechanism by which the pancreas, particularly the endocrine pancreas, can respond to VNS neuromodulation. Like currently approved VNS therapies, a fully implantable system can be utilized in which neuromodulation of the pancreas results in residual, healthy β-cells to release insulin on demand.

Successful development of neuromodulation of the pancreas relies on understanding the neurophysiology and neuroanatomy of the organ. Once the neural circuitry is understood, a neural interface for organ specific neuromodulation must be identified that will enable in vivo studies. Finally, it is vital that these studies are carried out parallel with development of technology that will facilitate rapid translation to human pancreatic neuromodulation. Here, we will establish the framework for a closed-loop system in which neuromodulation of the pancreas can be used to drive surviving β-cells in a T1D patient. First, we will review the literature demonstrating control over pancreatic function through VNS and identify shortcomings in this body of work. Next, we will review neural interfaces relevant for pancreatic neuromodulation. Finally, we will discuss other technologies that must be advanced to implement a closed loop neuromodulatory device to treat T1D.

Bioelectronic neuromodulation of the pancreas

The need for a bioelectronic approach to T1D

Today, the prognosis for patients with T1D is much better than it has historically been; however, management of T1D is predicated on strict patient compliance to their treatment regimen. The current standard of care for T1D includes monitoring blood glucose concentration and injecting exogenous insulin [17]. Clinically, this is often accomplished by basal dosing of insulin paired with bolus dosing insulin before consuming meals [18]. While insulin therapies are commonly successful, small and frequent insulin administration must be used to maintain tight control over blood glucose and reduce the risk of developing secondary neuropathies [19]. This dosing regimen results in a negative impact on patients’ quality of life (QoL). Furthermore, frequent dosing of insulin increases the risk for patients to become hypoglycemic. This risk translates to high mortality with between 4 and 10% of T1D patients dying due to insulin-induced hypoglycemia [20].

To restore metabolic control with less of a patient-active burden, semi-automated approaches have been developed including both external devices and biological implants. Fairly autonomous, closed-loop systems (termed ‘the artificial pancreas’) allow for closed-loop glycemic control with minimal user input [21,22]. These continuous glucose monitors working in concert with insulin pumps continuously deliver a basal rate of insulin, suspend basal delivery when a hypoglycemic event is predicted, and bolus dose in response to a hyperglycemic event [23,24]. They also allow user input to suspend basal delivery or bolus deliver. Closed-loop systems are effective at lowering glycated hemoglobin (A1C), a standard marker of long-term disease control, compared with traditional treatments [25]. Another option when looking at restoring lost homeostatic control is to replace the destroyed cells. There are two methods for this: whole pancreatic transplantation [26,27] which is done in limited cases, or islet transplantation [28]. These strategies have allowed for normal insulin responses [29] and resulted in an increased QoL for patients [30].

While these treatments have been proven successful, each have significant drawbacks or current challenges. Closed-loop devices require constant access to exogenous insulin which limits a patient’s autonomy. The delivery of insulin from these devices is into the subcutaneous space. This delivery presents numerous potential problems including infusion set occlusion, insulin precipitation as a result of changes in local pH and insulin complex formation. These all can hinder the delivery of insulin to the patient through subcutaneous administration [31]. Additionally, subcutaneous delivery introduces a lag time between dosing and effect when compared with normal pancreatic secretions [32]. Pancreatic transplants can be used but are subject to donor availability and typically are only used in cases where kidney transplantation is also needed. These procedures have limited long-term success with only 76% pancreases functional after 5 years [33]. Additionally, as with all organ transplants, recipients will need lifetime immunosuppressants and potentially have complications as a result. Islet transplantations have also been plagued with losses in transplanted cells with a median insulin independence period of only 15 months [34].

Data have recently shown that there may be some neural contribution to the progression of T1D. In recent onset T1D patients, fewer vasoactive intestinal peptide nerves, considered parasympathetic fibers, were observed to the exocrine pancreas but not islets. These patterns were reversed in patients long after disease onset [35]. The changes in sympathetic innervation of islets have not been clearly identified. Some studies have observed decreased innervation however other studies have not replicated this neuronal loss [36,37]. In mouse models, Schwann cell death near islets has also been observed before the death of β-cells is detected [38]. Recently, sympathetic pancreatic nerve stimulation has demonstrated an ability to protect islets within the pancreas [39]. Mice that received stimulation of the pancreatic branch of the vagus nerve had fewer autoreactive T cells in pancreatic lymph nodes. These mice also showed a slowed rise in glycemia suggestive of delayed disease onset. These data together implicate some neural degeneration and dysregulated signaling that is an important and overlooked pathology of T1D. Electrical stimulation of peripheral nerves has previously been shown to stimulate neural regeneration [40,41]. Peripheral neuromodulation may benefit T1D by helping to restore this damaged circuitry.

These challenges motivate the design of better control systems that improve patients' QoL by alleviating constant active management. Toward this, a fully automated system which augments the function of remaining β-cells through a bioelectronic approach may help restore homeostasis.

Pancreatic neuromodulation

Neuromodulation of the pancreas may offer a method to control glycemic state with high temporal resolution, tunability and consistency. Control over the pancreas via electrical stimulation relies on responsiveness of endocrine cells to neurotransmitters. Innervation of pancreatic islets have been investigated to help understand this mechanism in cat [42], rat [43], mouse [39,44], baboon [45] and pig [46]. Parasympathetic innervation from the vagus nerve modulates islet function both through direct innervation and through modulation of intra pancreatic ganglia (Figure 1). This induces the release of both insulin and glucagon as a result of acetylcholine signaling through the M3-muscarinic receptor [47–50]. Activation of the M3 receptor causes activation of both phospholipase C and phospholipase A2 resulting in downstream insulin secretion. Hydrolysis of phosphoinositides by phospholipase C results in the formation of trisphosphate and diglyceride [51]. The trisphosphate in turn causes release of calcium stored in the endoplasmic reticulum resulting in an increase in cytosolic Ca²⁺ concentration. To oppose this, the cell removes Ca²⁺ via exocytosis and secretes cytosolic insulin [52]. Diglyceride activates protein kinase C, which works to maintain insulin secretion once it has begun [53]. Activation of phospholipase A2 results in the production of arachidonic acid, which allows for a calcium independent exocytosis of insulin [54]. While neuronal mapping of the human pancreas has been limited [55], methods to study the neural structure in the pancreas using 3D analysis of optically clear human tissue have been developed and showing greater innervation density than previously reported by 2D methods [2,56–58]. Widespread innervation of islets across many species suggests that innervation plays an important role in the function of islets and may provide a route for glycemic control in humans.

In the 1960's and 1970's, many experiments were carried to better understand the effects of parasympathetic stimulation on pancreatic function. Early data from Kaneto et al. show in a canine model that cervical vagus nerve stimulation (cVNS; Figure 2B) of either the right or left nerves resulted in the release of insulin [59]. Interestingly, their stimulation paradigm did not result a decrease in blood glucose concentration. Stimulation of the right vagus nerve has been replicated in a nonhuman primate model and similar rises in insulin were also observed [45]. While many of these early studies utilized cVNS, it was understood that this nontargeted approach had confounding effects from off-target stimulation. Exogenous over-activation of the vagus nerve at the cervical level results in confounding systemic changes in multiple known ways. First, false afferent signals of lung over-expansion pause respiration. Second, efferent signals to the heart slow heart rate. When combined, these two effects dramatically drop systemic blood pressure and confound experiments by altering pancreatic perfusion and hormone transport. Third, cVNS recruits efferent signaling to the liver and stomach. These signals may directly affect glucagon and ghrelin secretion thereby indirectly confounding measurements of pancreatic neuromodulation.

Figure 2. . Methods for interfacing with the vagus nerve.

The vagus nerve provides locations to interface with which can be used for neuromodulation. (A) Auricular vagus can be used in a non-invasive manner to modulate vagal tone. (B) Cervical vagus nerve is the standard current site of interface for approved VNS treatments. (C) Subdiaphragmatic vagus nerve provides stimulation site distal to many abdominal organ innervation points potentially avoiding unwanted side effects. (D) Pancreatic VNS provides the most targeted method of applying stimulation to an organ target.

QoL: Quality of life; VNS: Vagus nerve stimulation.

Early attempts to remove some of these confounding factors were carried out using a targeted stimulation approach by interfacing with the subdiaphragmatic abdominal trunks of the vagus (Figure 2C). In these experiments, electrical stimulation was applied to a distal trunk of the vagus nerve circumventing thoracic cavity off-target effects. In canines, it was observed that stimulation of the dorsal trunk of the vagus nerve resulted in an increase in insulin [59], as well as glucagon [60]. Both of these responses were diminished when stimulation was combined with a pretreatment of the anticholinergic drug atropine [60]. Ultimately during these subdiaphragmatic stimulations an increase in blood glucose concentrations was observed suggesting that other mechanisms may be active other than just effects on the pancreas.

To further understand these circuits, the roles of both afferent and efferent signaling on glycemic control have been investigated. Bilateral cervical vagotomy in a canine model resulted in a decrease in insulin 15 min after transection [50]. Upon stimulation of the distal end of the transected nerve, there was a sharp rise in insulin as measured in the portal vein. In conscious calves, similar observations of efferent stimulation of dorsal and ventral trunks of the subdiaphragmatic vagus led to a rise in plasma insulin and glucagon levels that were abolished with parasympathetic blocker atropine [61]. Recent experiments in anesthetized rats also observed that efferent cervical stimulation resulted in the slight increase in plasma insulin 120 min poststimulation, as well as decrease in blood glucose concentration [62]. Meyers et al. also found that both stimulation of the intact nerve and afferent-only stimulation led to an increase in blood glucose without significant change to insulin. These data suggest that there is a direct efferent mechanism for controlling the release of insulin using VNS; however, there also seems to be some glycemic control by afferent signaling. Normal afferent signaling in the vagus nerve conveys information related to a hypoglycemic state [63]. Stimulation the vagus nerve may have multifaceted effects by directly driving release of insulin from β-cells while enacting a centrally mediated homeostatic mechanism to raise blood glucose due to a sensed hypoglycemic state. Due to this complex interaction, selective afferent or efferent stimulation should be investigated.

Pancreatic neuromodulation relies on exogenous activation of neurons innervating the β-cells. These neurons can be activated through the extracellular delivery of electrical charge which depolarizes the cell resulting the production of an action potential. Application of various waveform parameters changes the ability of the stimulus to recruit various sizes and quantities of fibers [64–67]. Over the decades, a wide variety of stimulation parameters have been used for VNS for pancreatic neuromodulation (Table 1); however, no consensus on the ideal parameter set has yet been made. Both voltage-controlled and current-controlled stimuli have been used with ranges from 3 to 60 V and 3 to 13.5 mA respectively. Pulse widths have ranged from 200 μs to 8 ms, frequencies have ranged from 0.25 to 100 Hz and durations have ranged from 1 s to up to 2 h. The variability of stimulation parameters has made it difficult to compare data between studies. Location of stimulation has also varied greatly between studies. Both subdiaphragmatic and cervical VNS result in the release of insulin and glucagon. This suggests that even subdiaphragmatic stimulation may not be specific enough and is likely signaling the liver to release glucagon.

Table 1. . Historical stimulation parameters used for pancreatic neuromodulation.

Location	Amplitude	Pulse width (mS)	Freq (Hz)	Train duration	Wave form	Year	Model	Insulin	Glucagon	Blood glucose	Ref.
Cervical/Sub diaphragmatic	14 V	5	100	10 s burst every 15 s for 3 min	Biphasic	1966	Dog	Increase (left, right, posterior). No change (anterior)	–	Increase (posterior). No change (left, right, anterior)	[59]
Subdiaphragmatic	15–30 V	1	5	10–15 min	–	1966	Cat	–	–	–	[68]
Sub diaphragmatic	5 V	1.5	10	10 min	–	1967	Baboon	Increase	–	–	[45]
Cervical	3 mA	8	60	20 min	Mono-phasic	1967	Dog	Increase	–	No change	[50]
Subdiaphragmatic	10 V	–	–	90 s, 20 min	–	1967	Dog	No change	–	No change	[69]
Thoracic vagus	5 mA	1.5	10	10 min	Mono-phasic	1973	Dog	Increase	–	–	[70]
Pancreatic neuro-vascular bundle	10 mA	1	40	10 min	–	1973	Dog	Increase	–	–	[71]
Subdiaphragmatic	8 mA	2	50	10 min	–	1974	Dog	Increase (posterior) No change (anterior)	Increase (posterior) No change (anterior)	Increase (posterior) No change (anterior)	[60]
Cervical	7–10 V	2–3	1.0–20	5–15 min	–	1978	Cat	Increase (left and right) correlated with pulse count	–	–	[72]
Thoracic	8 mA	4	0.25–20	5–30 min	–	1979	Pig	Increase correlated with BG concentration	Increase correlated with inverse of BG concentration	–	[46]
Pancreatic	6 V	1	20	–	–	1980	Rat	Increase	–	Decrease	[73]
Efferent Subdiaphragmatic	10–15 V	0.5	10	10 min	–	1981	Calf	Increase	Increase	Increase	[61]
Thoracic	10–20 V	0.5	4 or 40	1 s every 10 s for 10 min	–	1983	Calf	Increase	Increase	Increase	[74]
Subdiaphragmatic	20 V	1	50	12 min	Biphasic	1983	Rat	Increase	Increase	–	[75]
Subdiaphragmatic	10 mA	4	4–8	4 or 15 min	–	1986	Ex vivo pig	Increase	Increase	–	[76]
Thoracic	13.5 mA	5	10	10 min	–	1986	Dog	Increase	Increase	Increase	[77]
Subdiaphragmatic	10 V	1	2–10	5 min	–	1987	Ex vivo rat	Increase	Increase	–	[78]
Cervical	10 mA	–	40	1 s per 10 s for 10 min	–	1999	Sheep	Increase (efferent)	–	–	[79]
Cervical and pancreatic branch	10 mA	0.2	20	–	–	2002	Dog	Increase (cervical). No change (pancreatic)	No change (cervical and pancreatic)	–	[80]
Cervical	3 V	1	5	120 min	Mono-phasic	2016	Rat	Increase (efferent). No change (afferent, combined)	Increase (efferent, combined). No change (afferent)	Increase (afferent, efferent, combined)	[62,81]

When designing future studies, the field could benefit from using a standard parameter set that can be used to compare between studies. From Table 1, most commonly stimulation waveforms consist of current controlled pulses with an amplitude of approximately 10 mA held for 2 ms repeated at 10 Hz. This stimulation waveform may be suggested as a control waveform to be used between studies so that data can be compared between studies While it is important to understand the effects of changing stimulation waveform parameters, it is unknown if these parameters would translate into clinically relevant parameters for human neuromodulation due to differences in nerve diameter & innervation patterns in preclinical animal models [82]. As such, these animal data should be used for starting places for stimulation parameters, but proof-of-principle experiments in humans need to be carried out before parameters for ideal waveforms can be established. Ultimately when used clinically, stimulation parameters may be changed in real time to achieve a desired therapeutic outcome; however, the location of stimulation cannot be changed without repeated surgery. As such, the best location of stimulation must be defined before surgical intervention.

Cervical VNS as a therapeutic intervention

Recently, a focus of research in the field of bioelectronic medicine has been on understanding the mechanism by which VNS has an effect. Application of cVNS results in activation of sensory and motor axons innervating the heart, lungs and abdominal viscera, albeit in a nonselective manner without regard to innervation target. Treatments that are aimed at diseases of CNS origin such as epilepsy and depression may benefit from this approach. Nonselective afferent stimulation results in widespread neurotransmitter release in the nucleus of the solitary tract which is thought to increase modulation of secondary projections to the locus coeruleus and raphe nucleus [83] and other regions. The effect of these increased projections and the underlying mechanism of action is not fully understood. This leads to the possibility of the stimulus inducing some unknown, off-target mechanism that is not well understood.

Similar to many other therapies, cVNS does have some known side effects that are associated with its use. These side effects can include dyspnea, coughing, hoarseness, as well as changes in voice [84]. While there are side effects, for many patients the end goal of efficacy is met and QoL is improved [85], which demonstrates the therapeutic potential of neuromodulation. In practice, clinical dosing guidelines are not based on targeting specific fiber activation, but instead increasing the stimulus amplitude to maximize efficacy. This typically is implemented by gradually increasing the amplitude to the highest current the patient can tolerate before side effects become too cumbersome. By design, this aims to minimize known side effects, but it gives no consideration to unknown side effects. One such concern arises from the cVNS anti-inflammatory reflex [86]. While this effect may be therapeutic in the case of acute systemic inflammation, the effects of decades of altered inflammatory state due to chronic cVNS are unknown. To ensure that VNS is safe and effective over a long period of time, it is important that the mechanism of action of these treatments are further studied. Due to the combination of extensive innervation patterns through the viscera and incomplete knowledge of central processing of afferent vagus nerve activity, this task becomes difficult for cVNS.

Targeted pancreatic neuromodulation

Informed and targeted neuromodulatory therapy design needs to consider the entire anatomy of the targeted nerve to understand all systems that the electrical stimulation will engage. The vagus nerve provides both parasympathetic efferent (motor) and visceral afferent (sensory) signaling to and from most of the organs and structures within the thoracic and abdominal cavities (Figure 2). In humans, the vagus nerves at the cervical level consist of a 4–5 mm bundle of fibers containing between 1 and 21 fascicles [87]. The cervical vagus nerve is easily identified by its proximity to the carotid artery and jugular vein. As it further descends, the nerve branches to innervate the heart and lungs before joining the esophagus and forming a plexus eventually giving rise to the anterior and posterior vagal trunks as it passes through the diaphragm at the esophageal hiatus. At the subdiaphragmatic level, these trunks continue down the esophagus toward the stomach while giving rise to two major branches [88]. The common hepatic branch arises from the anterior trunk and innervates the liver, pancreas and gall bladder (if present) [89] and the celiac branch provides signaling to the celiac plexus and modulates downstream signaling to the stomach, through the rest of the gastrointestinal tract from duodenum to as far as the transverse colon, along with the pancreas, kidneys and spleen [90,91]. Distal to these bifurcations, the trunks splay out into multiple gastric branches at the gastroesophageal junction. Through numerous afferent and efferent pathways, this allows for signaling to multiple organs which together modulate metabolic state.

Location of stimulation should be determined based on the target organ of control when designing a bioelectronic therapy. Common sites for neuromodulation include the cervical vagus as previously discussed (Figure 2B), auricular branch of the vagus (Figure 2A), and subdiaphragmatic vagus (Figure 2C). Stimulation of the auricular vagus (auVNS) offers an attractive way to neuromodulate a branch of the vagus nerve as it can be accessed superficially in the external ear canal; however, it does not offer any way to directly modulate specific visceral organ function and contains skin sensory axons with cell bodies in the jugular ganglion of the vagus, not the nodose ganglion. As such, if specific visceral organ stimulation is desired, then the auVNS should not be used. Subdiaphragmatic vagal stimulation (sdVNS) prevents stimulation of sensory and motor nerves to and from the heart and lungs, which eliminates some possible thoracic organ side effects and laryngeal muscles, via the recurrent laryngeal branch of the vagus, which eliminates, hoarseness, cough and change in the voice. However, numerous off-target visceral organs still receive stimulation with subdiaphragmatic stimulation. Targeted VNS reaches the highest degree of spatial selectivity by placing the electrode at the most distal branch of the nerve which exclusively innervates the organ of interest. In the case of pancreatic vagus nerve stimulation (pVNS), electrical stimulation can be applied to a distal region of the hepatic branch, termed the pancreatic branch, just before it enters the pancreas (Figure 2D). This method can provide the desired effects on the organ while minimizing off-target signaling to the other innervated systems [92]. Similar approaches of targeted stimulation are commonly used to achieve high spatial resolution for prosthetic control in both central [93] and peripheral [94] neuromodulation. Other techniques such as selectively activating specific fibers may provide an alternative way to achieve targeted neuromodulation; however, neural interfaces which can identify and selectively activate these fibers do not yet exist. Careful selection of stimulation parameters may activate populations of fibers based on fiber size; however, this does not ensure that all fibers activated will innervate the target organ.

Identification of neural interfaces for organ specific neuromodulation

Interfacing with the peripheral nervous system requires a device that can chronically deliver electrical current to a target nerve. As with any device, noninvasive approaches are preferred as they avoid the implantation-related foreign body response and possibility of device failure. Inversely, as devices are distanced from target nerves, the specificity of stimulation decreases. To balance this conundrum, an array of electrodes has been made with a range of invasiveness from transcutaneous to neural penetrating.

Transcutaneous electrical nerve stimulation (TENS) devices offer an FDA approved method to deliver current to the cervical or auVNS nerve in a noninvasive approach. TENS devices have successfully been used for the treatment of migraine or chronic pain [95] by applying stimulus through electrodes to the skin above neural targets. While minimizing invasiveness is good when interfacing with biological systems, TENS is only effective at stimulating superficial nerve targets such as the cervical or auricular branch [96]. As such, TENS is not an appropriate method for use when targeting nerves deep within the visceral cavities. To limit the amount of intermediate tissue between electrode and neural target, implanted electrodes can be used.

Implantable devices with extraneural electrodes can increase specificity by directly applying current to neural targets. Cuff electrodes that encircle the nerve of interest allow the selective stimulation of specific nerve trunks by careful surgical placement [97]. These devices consist of wire or conductive foil strips that wrap around the epineurium of the nerve with an insulator, commonly silicone, preventing electrical contact with surround tissue. The flat interface nerve electrode is a refined version of a cuff electrode with multiple electrodes that can independently be controlled [98]. Silicones used for the dielectric in cuffs are soft (in the 1 MPa range [99]), but they still are 1–2 orders of magnitude stiffer than peripheral nerve. This increases the possibility for neural damage due to any movement of the electrode if not appropriately anchored, particularly with nerves of small diameter. To address this, new polymers such as Thiol-ene/Acrylate-based materials have been investigated as novel substrate for cuff electrodes [100]. These advanced polymers may offer a direction forward in research to interface with small diameter peripheral nerves.

Higher specificity can be achieved using more invasive intraneural devices that position electrode sides inside the perineurium and reside within microns of the axons of interest. These devices can either run axially, such as the longitudinal intrafascicular electrode [101], or perpendicularly such as the transverse intrafascicular multichannel electrode [102]. Devices such as the tissue engineered electronic neural interface require nerve transection and regrowth through the electrode toward distal targets [103,104]. High density interfaces such as these can achieve high spatial specificity compared with extraneural devices by containing dozens of electrodes each interfacing with few axons.

When implanting high density electrodes into nerves with multiple functions, it is necessary to map the electrode sites to separate functions. In sensorimotor tasks, this can be accomplished by asking patients to describe the sensations that they are feeling or to move prosthetic devices. This task becomes more challenging when the goal is organ-specific modulation. Without conscious knowledge over the state of organs, a patient cannot communicate the specificity of the stimulation. Effective mapping of these devices would rely on real time hormone analysis that is not currently available. Furthermore, these high channel devices are still research grade have not been FDA approved for use in humans. Bypassing these limitations by using nonspecific electrodes such as a cuff electrodes around a branch of the nerve which exclusively innervates a target organ may prove more useful in these cases.

Toward human pancreatic neuromodulation

The success of translating bioelectronic medicine from the lab is predicated on multiple factors. First, approaches need to be demonstrated to be safe and effective. This relies on using clinically relevant disease models; these models should accurately represent changes in interactions between the nervous system and diseased organ of interest. In the case of T1D, this includes observed changes to parasympathetic innervation patterns. Second, bioelectronic systems have the potential to incorporate feedback mechanisms to monitor their efficacy in real time, maximizing treatment and minimizing side effects. However, implementing these feedback signals is heavily dependent on the need to develop new real-time peptide hormone qualification technologies.

Animal models of pancreatic neuromodulation

There are two primary types of T1D animal models, chemically and genetically induced models. While these disease models offer ways to study the disease and treatments, neither model is a complete representation of the onset or progression seen in humans. Chemical induction relies on toxic glucose analogs such as streptozotocin or alloxan transport by GLUT-2 which results in β-cell death. This death, however, is only semi-specific to β-cells since GLUT-2 is expressed in other tissues including the CNS [15,105]. Genetic models of T1D including the nonobese diabetic mouse and biobreeding rat offer other spontaneous onset mechanisms which induce a diabetic phenotype in animals; however, similar to chemical induction, these genetic models are not identical to what is seen clinically in humans. Hundreds of treatments have prevented or reversed T1D in nonobese diabetic mice and biobreeding rats yet have not been successfully translated to humans [106]. Given that there are differences in innervation patterns between species that may change the efficacy of neuromodulation-based treatments, animal models are necessary for proof-of-principle studies but should only be used to provide preliminary tests of treatment paradigms. After safety is proven and a potential for efficacy is demonstrated, it is vital to rapidly translate these technologies to humans.

Animal models also provide unique challenges to development of pancreatic neuromodulation techniques. When interfacing with the small diameter nerves in rodent models, researchers may encounter anatomical problems which may not be clinically relevant. The cervical vagus nerve in the rat is approximately 750 μm in diameter, compared with a human cervical vagus nerve which is 5.1 ± 1.5 mm [87]. Nerves in the 750 μm diameter range are reliably interfaced using cuff electrodes in rodents [107], and can translate easily to human nerves in the similar size range. When attempting to achieve spatial targeting by interfacing with the pancreatic branch of the vagus nerve in a rodent, the nerve which must be interfaced is between 50–300 μm [39]. It is possible that chronic interfacing with a small nerve may damage it given the normal mobility of the abdominal organs. Additionally, animal models have different proportions of β-cells within islets, that may change efficacy in animals studies compared with human trials [108]. These challenges ultimately exist whenever animal models are used and divert from the goal of clinical translation. Minimizing variability from humans by using large animal models should be emphasized to enable clinical translation.

Considerations for clinical translation

Bioelectronic therapies are unique to pharmacological interventions in both their potential for both high temporal and spatial selectivity. To translate bioelectronic therapies to the clinic and take advantage of this high temporal specificity, progress must be on the development of technology to enable real-time closed-loop devices. Closed-loop systems require high fidelity feedback signals that are not clinically available due to technological limitations. Additionally, techniques to improve spatial selectivity using unidirectional stimulation must be developed to improve stimulation of a specific organ.

Measuring metabolic biomarkers gives a patient valuable information that can be used to direct their course of treatment. Day-to-day management of T1D primarily involves monitoring blood glucose, managing diet and delivering exogenous insulin. As such, measuring glucose concentration accurately has been important for treating patients in pseudo-real-time. Glucose-oxidase based sensors allow for the accurate measurement of glucose by patients [109]. Clinical measurements most often collected as interstitial fluid due to its ease of access; however, interstitial fluid is separated from blood via anatomical barriers. These barriers cause up to a 10 min lag time behind intravascular glucose which may complicate the use of this in predictive closed-loop devices [110]. While the development of glucose oxidase-based sensors has improved the treatment of diabetes, there remains an opportunity to measure hormones which control blood glucose for better predictive modeling than based on glucose alone.

Real-time measurement of circulating hormones such as insulin and glucagon could help these systems predict glucose fluctuations and better deliver insulin as needed. Current closed-loop systems take advantage glucose oxidase-based meters and insulin infusion pumps to automate some level of treatment. As we move toward better treatments for diabetes, such as a neuromodulation approach, it will be advantageous to measure circulating hormones with the same (or better) temporal resolution as we currently measure glucose. This will generate a more complete picture of metabolic and disease state and ultimately allow us to develop better algorithms for a completely closed-loop, biological system for controlling blood glucose.

Intraoperative positive identification of the pancreatic branch of the vagus nerve is crucial to the success of the therapy. This can be accomplished through inducing neural signaling via electrical stimulation at a known trunk of the nerve such as the cervical vagus nerve, and recording induced neural activity at a suspected target nerve [92]. While this identifies a nerve as of vagal origin, it does not verify that the branch innervates the pancreas. Real-time intraoperative measurement of hormones such as insulin and glucagon can ensure that the electrode is correctly placed and that the patient will respond to the therapy.

Devices that can selectively stimulate afferent or efferent pathways are important for achieving selective modulation of target organ systems. Controlling the directionality of stimulation will also likely be important in pancreatic neuromodulation [62]. Directionality has been observed to be important in the anti-inflammatory reflex and has been hypothesized to be important independent parameter for modulation of future organ systems [111]. Directionality can be assured by nerve transection of the undesired direction of stimulus spread; however, while the use of nerve transection is appropriate in experimental paradigms, it will lead to neuronal death and rapid loss of function in chronic cases. As such, this is not safe approach in chronic cases. A potentially viable option is to use kilohertz frequency electrical stimulation combined with VNS to block neural conduction in one direction [112], in other words to block ascending vagal sensory axons. This method of blocking is reversible offering the potential to be used clinically but its long-term viability is not yet understood.

Bioelectronic therapies present an interesting concern regarding developing tolerance to the treatment. VNS-induced plasticity has been observed in the CNS [113]. Any afferent mechanism of pancreatic neuromodulation provided by the enhanced signaling through VNS may filter out over time. Chronic studies of pancreatic neuromodulation have not yet been carried out in a way that can characterize the development of tolerance to this approach.

Future perspective

Pancreatic neuromodulation offers a unique testbed to develop an approach that can be extended to other organ modulation paradigms. This is due to the pancreas’ well characterized inputs and outputs, large body of literature demonstrating its neuroresponsiveness and a clear need for modulation to treat disease. As work continues into developing bioelectronic therapies into clinically viable treatments, a clear path for the field must be set to ensure that studies are rigorous, comparable and represent a meaningful stepping stone toward human translation. Future work to address the knowledge gap in chronic efficacy should be carried out in a location-based targeting model in large animals. This work should be accomplished using a standardized parameter set that results can be better compared between laboratories. Due to the differences in innervation patters, these studies should not focus on optimizing parameter sets, but instead show titratability before transitioning to human clinical trials. The creation of a closed-loop therapy requires large technological strides. Separate to and in parallel with pancreatic neuromodulation work, efforts should focus on the development of real time hormone sensing technologies. Aptamer-based sensors offer an approach for high specificity detection of proteins which would be useful in these closed-loop devices. Neural interface technologies need to be developed to allow for selective stimulation at the cervical level avoiding a more complicated abdominal surgery. Devices with a high density of electrodes are currently being created but are not ready for widespread adoption. Minimizing electrode site size and maximizing channel count potentially may allow for the direct communication with organ specific fibers in mixed population nerve preventing off-target stimulation. Within a decade, it is possible that these puzzle pieces can come together to form a therapy that can be easily implanted and help patients with T1D.

Conclusion

Restoring homeostasis through the modulation of organ neuromodulation offers an new route to replace traditional pharmaceuticals. Historical experiments have laid the foundation for understanding of how neural inputs control organs. The United States’ NIH as well as the Department of Defense (DoD) have recently invested over 100 million US dollars in research programs to unlock the potential of electrical stimulation as a therapeutic through their Stimulating Peripheral Activity to Relive Conditions (SPARC) [10] and Electrical Prescriptions (ElectRx) programs [114]. As technology has developed, neuromodulatory therapies are now emerging as technologically feasible [10].

Successfully transitioning bioelectronic medicine from the lab to the demands the use of targeted neuromodulatory approaches. Targeted stimulation paradigms allow for the understanding mechanisms that can benefit patients while minimizing obfuscating off target effects. In the preclinical environment, experiments need to be carried out in models that mimic human neuroanatomy and disease state as best as possible. Neural interfaces should be selected such that they can stimulate nerves in a minimally invasive way while achieving specificity. As treatments are identified in animal models, they should quickly be followed up with clinical data to assess efficacy in humans using already FDA approved devices in an off-label manner.

Executive summary.

• The pancreas is a well innervated organ in which endocrine cells receives input through parasympathetic and sympathetic signaling.

• Parasympathetic signals provided by the vagus nerve result in release of insulin from pancreatic β-cells. A large body of literature suggests this release can be enhanced through electrical stimulation of the vagus nerve. A proposed therapy for Type 1 diabetes using vagus nerve stimulation relies on this mechanism.

• Historically, stimulation of the vagus nerve has been applied at the cervical level. Due to numerous organ systems which are innervated by the vagus nerve, this stimulation paradigm results in heavily obfuscated results. Furthermore, due to a wide variety of stimulation parameters and animal models, it is difficult to compare between experiments within the field.

• Targeted pancreatic neuromodulation offers an approach to modulate only the function of the pancreas while eliminating off target effects on other organs. This can be achieved through selective modulation of specific fibers via anatomical placement of electrode or by interfacing with a subpopulation of fibers using neural interfaces containing a high density of electrodes. While these devices are not ready for the clinic, research should be carried out using an anatomically targeted paradigm.

• This research should be carried out in parallel with designing better biomarker detection methods such as aptamer-based hormone sensors to allow for eventual closed-loop neuromodulation.