Implantation of a passive electrical neurostimulation device achieves inflammatory modulation in rodents

Abstract

Vagus nerve stimulation (VNS) has been identified as a treatment for diseases ranging from depression to rheumatoid arthritis. Despite widespread interest, the modulation of both pro- and anti-inflammatory cytokines using a chronically implanted VNS device in rodents has not yet been demonstrated. Existing neuromodulation devices typically utilize application-specific integrated circuits for miniaturization. Here, we present a neuromodulation system comprising a passive, wirelessly driven, implantable neurostimulator capable of bipolar, multisite, and temporal interference stimulation. The device incorporates a bidirectional current-limiting circuit to protect tissue from excessive cathodal and anodal stimulation. System performance is validated through sciatic nerve stimulation to demonstrate functionality and versatility, and through VNS to demonstrate chronic reliability and inflammatory control in rats. After three weeks of implantation, VNS significantly modulates blood pressure and cytokine concentrations, increasing anti-inflammatory and decreasing pro-inflammatory cytokines following lipopolysaccharide injection. Passive, miniaturized neuromodulation systems may enable chronic VNS studies and facilitate clinical translation.

Fully passive, wireless neurostimulation device with bidirectional current limiting enables modulation of both pro- and anti-inflammatory cytokines following lipopolysaccharide challenge in rodents after chronic implantation on the vagus nerve.

Introduction

Electrical neuromodulation, the act of electrically stimulating nerves, has emerged as a treatment option for many diseases and disorders that are poorly treated or untreatable through traditional pharmaceutical and surgical approaches (e.g., cardiac pacemakers and cochlear implants)^1–4. Devices such as cochlear implants, deep brain stimulators, and vagus nerve stimulators (VNS) are among the most commonly used neuromodulation devices that have made it to market, enabling patients to hear or alleviate symptoms such as tremors or seizures. As clinicians and engineers alike search for new neuromodulation applications, the vagus nerve remains a focal point due to its widespread connectivity.

Named after the Latin root for “wanderer,” the tenth cranial nerve, or vagus nerve, meanders throughout the body, forming the core of the parasympathetic nervous system. It innervates numerous structures, originating in the medulla oblongata and projecting to the larynx, the heart, the spleen, and much of the gastrointestinal tract, before ultimately terminating in the colon. It is composed of a mixture of afferent sensory and efferent motor and autonomic fibers. The versatility of VNS is illustrated by its U.S. Food and Drug Administration (FDA)-approved treatments for drug-resistant epilepsy⁵, clinical depression⁶, and migraines⁷, with ongoing clinical trials exploring its potential in stroke rehabilitation^8,9, inflammatory bowel disease (IBD)^10–12, and heart failure^13,14. Most recently, VNS was approved by the FDA for the treatment of rheumatoid arthritis^15,16, the first purely inflammatory indication.

One of the most exciting applications for VNS is its role in modulating inflammation. The vagus nerve is involved in the cholinergic anti-inflammatory pathway, a neural circuit functioning to dampen the body’s inflammatory response^17,18. The ability to reduce inflammation would greatly benefit patients with autoimmune¹⁹, gastrointestinal²⁰, metabolic²¹, or neurodegenerative^22,23 diseases and acute injuries such as traumatic brain injury (TBI)^22,24 or ischemic stroke^8,25. Moreover, the theorized benefits of noninvasive VNS were substantial enough to prompt the Canadian government to grant an emergency authorization for treating long-COVID symptoms during the recent global pandemic²⁶. Inflammation can be reduced with pharmaceutical intervention, but it comes at the cost of side effects. Additionally, neuromodulation therapies open the door to real-time, closed-loop treatments.

Interestingly, no studies have successfully modulated both pro- and anti-inflammatory cytokine concentrations in a rodent model with a chronically implanted VNS device, though some researchers have modulated cytokines with acutely implanted VNS devices^27–29. In 2018, our lab demonstrated that chronic cuffing of the vagus nerve in rodents may impair retrograde transport in efferent fibers³⁰. We hypothesized that the stress applied to the nerve by the placement of a chronically tethered cuff electrode was responsible for this compromised behavior, suggesting that a lightweight, untethered stimulator could resolve this issue. While not currently used for VNS, some groups engineer peripheral nerve cuffs housing application-specific integrated circuits (ASICs) to achieve significant miniaturization^31–35, but these devices are typically prohibitively difficult and expensive to produce and not easily adaptable to alternative applications. Other groups create systems using commercial off-the-shelf (COTS) components or specially designed cuff electrodes with techniques to reduce stress on the nerve^36–40. Notably, the Feinstein Institutes were able to modulate cytokines in mice with a head cap-based VNS system³⁹ by administering VNS prior to LPS injection and demonstrated chronic VNS in mice with a fully implantable device⁴⁰. However, there has yet to be a chronically implanted device that enables long-term rodent studies on inflammatory modulation.

Consequently, we developed a wirelessly driven neuromodulation system that comprises only passive, COTS components, featuring a bidirectional current-limiting circuit to protect tissue from excessive stimulation. The neurostimulator is fully implantable and consists of a flexible printed circuit board (PCB) containing an integrated coil and a parylene C nerve cuff with platinum-iridium electrodes. It is small and lightweight enough to be sutured directly around the vagus nerve. We were able to modulate both blood pressure and pro- and anti-inflammatory cytokines in rats when applying VNS following the injection of lipopolysaccharide (LPS) from Escherichia coli three weeks after implantation. This differs from administering VNS prior to LPS injection, which is a preemptive therapy rather than a rescue therapy and less clinically relevant for treating systemic inflammation⁴¹. Furthermore, we demonstrated the versatility of this system in the rat sciatic nerve, customizing its wirelessly driven stimulation waveforms to elicit different behaviors, including preferential EMG activation of plantar or hamstring muscles depending on relative proportions of stimulation amplitudes. This system can be used for a variety of applications, as its wirelessly driven nature allows the stimulus waveform to be easily customized. We designed the system with the ability to independently drive different implanted neurostimulators with their own external transmitters at distinct waveform frequencies and amplitudes, allowing for multisite stimulation techniques and temporal interference stimulation (TIS). TIS, also known as interferential current stimulation, takes advantage of multiple stimulators whose outputs interfere in 3D space, creating an amplitude-modulated pattern⁴². Some researchers have shown that noninvasive TIS of deep structures without stimulating superficial structures may be a viable noninvasive option in peripheral nerves since the superficial tissues have different compositions and therefore different thresholds for electrical stimulation^43,44. Regardless of its noninvasive capability, TIS remains valuable in peripheral nerve stimulation by allowing localized regions of conduction block and phasic stimulation, fascicular nerve mapping, or causing different fibers to fire in or out of phase with each other⁴⁵. Simple, miniaturized, versatile neuromodulation tools can open the door to future chronic VNS studies in animal models and their translation to humans.

Results

Passive Neuromodulation System

We developed a stimulation system consisting of an external driving system and an implantable neurostimulator (Fig. 1). The driving system contains a signal generator, a matching network, and a transmit coil to generate and transmit amplitude-modulated (AM) radio frequency (RF) signals to the implanted stimulator, which demodulates these signals to produce the stimulation waveform. The implantable stimulator is composed entirely of passive circuitry to minimize size, cost, and complexity. The packaged device weighs 19.5 mg and measures 7 × 5 × 1 mm with a total volume of 7.56 mm³, making it small enough to be sutured directly around a nerve (Fig. 1b). It comprises a receive coil, a matching network, an AM demodulator, and a bidirectional current-limiting circuit. Each matching network matched the impedance of its corresponding coil to the impedance of the signal generator or rectifier, forming a resonant structure at the RF carrier frequency. The AM demodulator consists of a Schottky diode half-wave rectifier followed by a lowpass filter.

Fig. 1. Overview of the wireless neuromodulation system and signals used for temporal interference stimulation (TIS).

a Passive stimulator circuit with impedance matching, amplitude-modulated (AM) demodulator, and bidirectional current limiting. b A single packaged neurostimulator shown next to a U.S. penny for scale. c System overview illustrating externally generated AM radio frequency (RF) signals transmitted from the transmit (Tx) coil to the receive (Rx) coil on the device, producing stimulation waveforms at the nerve. RF₁ and RF₂ carriers (13.56 and 27.12 MHz) are modulated by intermediate frequencies IF₁ and IF₂ (2 kHz and 2.02 kHz), producing a 20 Hz electromyography (EMG) response through temporal interference (TI). These signals are delivered through separate electrodes into the tissue where they superimpose in 3D space within the nerve. In this figure, a 2.1 kHz signal is plotted instead of a 2.02 kHz signal, and the harmonic distortion caused by the rectification diode is neglected to better demonstrate the phase difference between the two interfering IF signals. d Fast Fourier transform of the signals shown in c.

The bidirectional current-limiting circuit is based on the principle of constant current diodes (CCDs), or current regulator diodes⁴⁶, typically used for current regulation with light-emitting diodes or other applications such as reference generation⁴⁷. A CCD is constructed from a depletion-mode field-effect transistor (FET) – either a junction FET (JFET) or depletion-mode metal-oxide-semiconductor FET (MOSFET) – and a resistor (Fig. 2a). Neglecting channel length modulation, the drain current, I_D, through a FET in saturation is defined by

$$I_D=\kappa {\left(V_{GS}-V_{th}\right)}^2$$ 1

where κ is a device-specific constant (Supplementary Note 1, Supplementary Eqs. 1 and 2), $V_{GS}$ is the gate-source voltage, and $V_{th}$ is the FET threshold voltage. From Ohm’s Law, we know that $V_{GS}$ of the JFET in Fig. 2a is defined by

$$V_{GS}=-I_{D}R$$ 2

Fig. 2. Bidirectional constant current circuit design.

a Traditional constant current diode (CCD) circuit. b Bidirectional CCD configuration. c Calculated current-voltage (I-V) curves (lines) compared with experimentally measured data (markers) for combinations of two FETs and two resistors. A 1 kΩ resistor was used as the load.

This can be substituted into Eq. (1) and solved for the drain current:

$$I_D=\frac{\left(1-2\kappa R{V}_{th}-\sqrt{-4\kappa R{V}_{th}+1}\right)}{2\kappa R^2}$$ 3

Of the two solutions to the quadratic equation, only Eq. (3) is possible, as the drain current cannot exceed the saturation current $I_{DSS}$ without forward biasing the gate-drain diode. Once saturated, the drain current is entirely dependent on the FET parameters and resistor value, not the applied voltage or load resistance. With negative applied voltages, the gate-drain diode of the FET will become forward-biased, allowing large currents to flow.

The bidirectional design, shown in Fig. 2b, consists of a JFET, two resistors, and two Schottky diodes and operates in the same manner as the CCD, with only one of the antiparallel Schottky diodes forward-biased at a time. This combined topology enables bidirectional current limiting under alternating voltage using a single FET, rather than two separate CCDs. Thus,

$$V_{GS}=V_f-I_{D}R$$ 4

where $V_f$ is the forward voltage drop across the diode. This can once again be substituted into Eq. (1) to yield the bidirectional current limit:

$$I_D=\frac{\left(2\kappa R{V}_f-2\kappa R{V}_{th}+1\pm \sqrt{4\kappa R{V}_f-4\kappa R{V}_{th}+1}\right)}{2\kappa R^2}$$ 5

Neither Schottky diode conducts DC current because both are antiparallel with the JFET gate diode. Each can, however, be individually reverse-biased or forward-biased to a 0 V drop. While this device is limiting the current in the positive direction, the gate-drain diode is reverse-biased, and the gate-source diode is forward-biased to 0 V ($V_f$ = 0 V). When a voltage of the opposite polarity is applied, the diode voltages are flipped. This allows the circuit to limit the current to the same value in both directions without forward biasing the gate-drain diode:

$$I_D=\frac{\left(1-2\kappa R{V}_{th}-\sqrt{-4\kappa R{V}_{th}+1}\right)}{2\kappa R^2}$$ 6

The primary tradeoff of adding bidirectional current limiting is the increased voltage required to reach the current limit. Since the transistor needs to be saturated for the device to work, a minimum voltage must be applied. In the unidirectional constant current diode, to saturate the FET, the applied voltage must be

$$V_{app}\ge V_{DS,sat}+I_{D}R$$ 7

where $V_{DS,sat}$ is the required drain-source voltage to place the transistor in saturation. In the bidirectional case, since there are two resistors, a larger applied voltage is needed to reach the current limit:

$$V_{app}\ge V_{DS,sat}+2I_{D}R$$ 8

The current during triode operation can be determined. The drain current of a JFET operating in its triode region is defined by

$$I_D=2\kappa \left(V_{GS}-V_{th}-\frac{V_{DS}}{2}\right)\left(V_{DS}\right)$$ 9

In both the unidirectional and the bidirectional cases, $V_{GS}$ is defined by Eq. (2). When substituting this into the triode-region drain current equation, we get a relationship between I_D and $V_{DS}$:

$$I_D=\frac{-2\kappa V_{DS}{V}_{th}-\kappa V_{DS}^2}{1+2\kappa V_{DS}R}$$ 10

In the unidirectional case, since $V_{DS}=V_{app}-I_{D}R$, we can solve for the drain current in terms of the applied voltage:

$$I_D=\frac{-2\kappa R{V}_{th}+1-\sqrt{{\left(2\kappa R{V}_{th}-1\right)}^2-4\kappa R^2\left(-2\kappa V_{app}{V}_{th}-\kappa V_{app}^2\right)}}{2\kappa R^2}$$ 11

For the bidirectional case where $V_{DS}=V_{app}-2I_{D}R$, we get:

$$I_D=\frac{-2\kappa V_{app}{V}_{th}-\kappa V_{app}^2}{1-4\kappa R{V}_{th}-2\kappa V_{app}R}$$ 12

This relationship was validated empirically across multiple JFET and resistor combinations and compared to the theoretical results from Eq. (12) to demonstrate the current-limiting ability (Fig. 2c).

Finally, the packaging process, depicted in Fig. 3, was used to create the cuff electrode and package the device before implantation. A 20 μm ultrathin coating of parylene C was conformally deposited on the device and its stimulation electrodes, enabling our simple, flexible, and lightweight design. The parylene thickness used in this work was chosen to be flexible enough not to damage the nerve but mechanically robust enough to last throughout the chronic study. The parylene coating not only acts as the electrical and biological insulation of the PCB, further reducing the overall size of the implant, but is also versatile enough to be used for virtually any cuff diameter, length, thickness, or topology, even enabling multiple electrodes. Post-explant examination of chronically implanted devices found that the parylene encapsulant withstood implantation well. Only a few devices showed fluid ingress, likely caused by damage during explant, as tissue encapsulating the devices made removal challenging. Mechanical stress during explant flexes the platinum-iridium wires between the PCB and cuff, which can crack the parylene or cause cuff delamination. With practice and slow, careful dissection, we were able to remove most devices without damaging them.

Fig. 3. Fabrication process for the leadless vagus nerve stimulation (VNS) device.

a Rendering of the unpopulated printed circuit board (PCB). b PCB after impedance matching and component population. c Cuff scaffold – an uncoated copper wire with platinum-iridium wires adhered using ultraviolet-cure cyanoacrylate – soldered to the PCB. d Device conformally coated with parylene C. e Cuff formation by trimming the parylene from the ends of the copper wire and cutting longitudinally along the cuff between the electrodes. f Dissolution of the copper wire using copper etchant. g Dissolution of the cyanoacrylate using acetone. h Image of the fully packaged device on a finger.

Fiber-Selective Temporal Interference Stimulation

We also utilized our system for multisite stimulation or TIS. To independently control multiple implanted stimulators, their receive coils need to be tuned to different resonant frequencies. They should then be driven by the corresponding RF carrier frequencies modulated by the desired stimulation signals, whether that is a square wave for traditional bipolar stimulation or a kilohertz frequency sinusoid for TIS. An overview of the signals used in the system and their spectra for the TIS case is shown in Fig. 1c, d. In summary, each stimulation signal requires a single stimulator circuit with two electrode outputs (e.g., traditional bipolar stimulation uses one stimulator; two-channel TIS requires two stimulators and four electrodes).

We validated the system’s capabilities by implanting a modified device adapted for the sciatic nerve of a urethane-anesthetized Long-Evans rat. The two devices were tuned to resonate at 13.56 MHz and 27.12 MHz so both would operate within ISM communication bands. Electromyography (EMG) of the plantar muscles of the foot was recorded as done by Budde et al.⁴⁵. The stimulators were operated in three configurations: a single stimulator operating with a 20 Hz square wave (bipolar case), both stimulators operating with sinusoids at 2 kHz and 2.02 kHz (TIS case), and a single stimulator operating with a 2 kHz sinusoid at an increased amplitude (conduction block case). The resulting EMG signals and associated waveforms are shown in Fig. 4. Both former methods produced 20 Hz muscular contractions, matching the pulse frequency of the square wave in the bipolar case and the beat frequency of the resulting AM signal in the TIS case. The conduction block setup produced a tonic muscle contraction throughout the stimulation cycle, indicative of neural conduction block^45,48.

Fig. 4. Wirelessly powered leadless stimulator elicits muscle activity under various stimulation conditions.

a 20 Hz contractions initiated by 20 Hz bipolar stimulation. b 20 Hz contractions initiated by 2000 Hz/2020 Hz temporal interference stimulation (TIS). c Tonic muscle activity induced by 2000 Hz stimulation.

We further validated the TIS case to prove we could perform fiber-selective stimulation. We implanted a 12-electrode cuff with longitudinal, as opposed to circumferential, contacts on the sciatic nerve of a urethane-anesthetized Long-Evans rat, as done by Budde et al.⁴⁵. Using two neurostimulator boards tuned to resonate at 13.56 MHz and 27.12 MHz, we connected them to different electrodes to map the fascicular nerve to find the firing threshold for both the plantar and hamstring muscles with EMG electrodes (Fig. 5a). After investigating the map, we were able to preferentially stimulate the hamstring muscles or the plantar muscles by applying TIS to electrodes 2 and 6 at different amplitudes to steer the current (Fig. 5b, c). Increasing current on one electrode pair preferentially recruited hamstring muscles without activating plantar muscles, and vice versa. At higher currents, both hamstring and plantar muscles could be recruited simultaneously.

Fig. 5. Nerve mapping and preferential fiber activation using temporal interference stimulation (TIS).

a Muscle activation thresholds across six distinct electrode configurations around the sciatic nerve; lower stimulation thresholds indicate fibers closer to electrodes. b Approximate nerve fiber map generated from the muscle activation thresholds in (a). Stimulation between electrodes 2 and 6 provided the most efficient configuration for preferential activation of both hamstring and plantar muscles. c Electromyography (EMG) results from the fiber selectivity experiment. Hamstring activation is elicited primarily by electrode 6 stimulation; plantar activation is elicited primarily by electrode 2 stimulation; simultaneous activation occurs when interferential current from both electrodes is steered toward the green overlap region. Asterisks (*) indicate deliberate changes in delivered power resulting in changes in EMG amplitude.

At the end of both experiments, the nerve was cut distal to the cuff. Stimulation was repeated, but no response was observed in the muscle or EMG traces, indicating true neural stimulation was achieved. These scenarios validate the capability of our device to perform several types of neural stimulation in vivo.

Autonomic Modulation through VNS

To test our device in a more clinically applicable setting, we performed VNS in a series of acute rodent experiments. Twelve urethane-anesthetized Long-Evans rats were implanted with a stimulator (Fig. 6a), and catheters were inserted into the femoral artery and vein to the abdominal aorta and inferior vena cava to allow for blood pressure recording and venous blood sampling, as done by Biggs et al.⁴⁹. In 6 of the animals, VNS was conducted with a 5 Hz signal driven by a 1 W wireless signal for a period of 5 minutes. Control animals received no stimulation. Blood pressure was recorded from 5 minutes before the start of stimulation until 5 minutes after stimulation ceased. Blood pressure was not recorded for one acute control animal due to an experimental error, and blood pressure data for one acute stimulation animal were excluded because of poor quality, likely from a clogged or mispositioned arterial catheter. Example mean arterial pressure (MAP) traces from both stimulation and control animals are shown in Fig. 6c.

Fig. 6. Vagus nerve stimulation (VNS) significantly affects mean arterial pressure (MAP) in both acute and chronic implant settings.

a Rendering of the neurostimulator device used for VNS implanted in a rodent (created in BioRender. Lawlor, G. (2025) https://BioRender.com/k1l5wmv). b Timeline of the VNS immunomodulation experiments showing device implantation, lipopolysaccharide (LPS) injection, blood pressure monitoring, 5 minutes of VNS treatment, and blood collection time points. c Example MAP traces for acute and chronic stimulation and control animals. Periods of active stimulation are highlighted in blue, and corresponding time periods for control animals are highlighted in red. The three gray vertical lines represent the blood collection procedure: waste collection, sample collection, and fluid replacement. d Mean difference in MAP for stimulation vs. pre-stimulation and stimulation vs. post-stimulation across experimental groups: acute stimulation (n = 5), acute control (n = 5), chronic stimulation (n = 6), and chronic control (n = 5). e Rolling mean absolute deviation (MAD) difference in MAP for stimulation vs. pre-stimulation and stimulation vs. post-stimulation for the same groups shown in (d). Box plots show the median and interquartile range (IQR), with whiskers extending to 1.5 × IQR. Individual data points are overlaid. *p < 0.05, **p < 0.01. Statistical comparisons were performed using two-tailed, unpaired t-tests with FDR correction.

To demonstrate the efficacy of this device in the context of chronic implantation, we implanted it on the vagus nerve of 12 isoflurane-anesthetized Long-Evans rats. The animals recovered for a period of at least 3 weeks (21.83 ± 0.94 days), after which they were anesthetized with urethane to avoid the effect of isoflurane on the inflammatory response^50,51. During this second operation, rats were catheterized as in the acute case for blood sampling and pressure recording, and LPS was intraperitoneally injected (5 mg/kg) to provoke an inflammatory response^28,30. Blood samples were collected every half hour over a 2.5-hour period, starting 30 minutes prior to the LPS injection. As before, VNS was administered to 7 of the animals (no electrical stimulation was delivered to the remaining control animals) at 5 Hz for 5 minutes, 30 minutes after LPS injection, with the goal of suppressing the inflammatory response. Blood pressure was not recorded in one of the stimulation animals because a clot formed in the catheter. Figure 6b shows a timeline of this experimental procedure.

MAP changed predictably during blood collection, with decreases during waste and sample collection and increases during fluid replacement in all animals. However, stimulation-specific effects were only present in stimulated animals, indicating effective efferent fiber activation (Fig. 6c). In acute animals, mean MAP during the stimulation period was significantly lower than post-stimulation in stimulated vs control animals (-22.78 ± 8.60 mmHg vs 0.87 ± 5.13 mmHg; t(8) = -5.28, p = 0.006, 95% CI [-34, -13], Hedges’ g = -3.02). The rolling mean absolute deviation (MAD) during the stimulation period was significantly higher than pre-stimulation (1.56 ± 1.08 mmHg vs -0.18 ± 0.63 mmHg; t(8) = 3.12, p = 0.049, 95% CI [0.46, 3.03], Hedges’ g = 1.78) and post-stimulation (1.54 ± 1.16 mmHg vs -0.30 ± 0.88 mmHg; t(8) = 2.83, p = 0.049, 95% CI [0.34, 3.34], Hedges’ g = 1.61). In chronic animals, the rolling MAD during the stimulation period was significantly higher than pre-stimulation in stimulated vs control animals (0.95 ± 0.56 mmHg vs 0.14 ± 0.39 mmHg; t(9) = 2.70, p = 0.049, 95% CI [0.13, 1.49], Hedges’ g = 1.49) (Fig. 6d, e).

To assess the effect of VNS in chronically implanted rats, we analyzed serum cytokines using multiplexed immunoassays to measure both pro- and anti-inflammatory cytokines (Fig. 7). The pro-inflammatory cytokine TNF-α, a “necessary and sufficient” modulator of inflammation that promotes production of other cytokines and physiological responses⁵², showed a decrease in peak concentration in stimulation animals compared to controls (1103 ± 320 pg/mL vs 2096 ± 1083 pg/mL; t(10) = −2.33, p = 0.042, 95% CI [−1943, −43], Hedges’ g = −1−0.26). Conversely, the anti-inflammatory cytokine IL-10 showed an increase in peak concentration in stimulated animals compared to controls (1090 ± 321 pg/mL vs 534 ± 376 pg/mL; t(10) = 2.76, p = 0.040, 95% CI [107, 1005], Hedges’ g = 1.49). Peak concentrations were derived from spline-interpolated data.

Fig. 7. Serum cytokine concentrations (TNF-α and IL-10) as a function of time relative to lipopolysaccharide (LPS) injection (t = 0 minutes) in chronic animals.

Seven experimental animals (n = 7; blue squares, dashed line) received vagus nerve stimulation (VNS; 5 Hz, 5 minutes) 30 minutes after LPS injection. Five control animals (n = 5; red circles, dotted line) received no stimulation. Square and circle markers represent measured data points; dashed and dotted lines indicate interpolated concentrations; solid lines represent group mean concentrations. Insets show interpolated peak concentrations as mean ± standard error (SE). *p < 0.05. Statistical comparisons were performed using two-tailed, unpaired t-tests with FDR correction.

Discussion

In this work, we present a system for wireless electrical neuromodulation containing only passive circuitry and use that system to evoke neural activity in both motor and autonomic nerves in a rodent model. Moreover, we modulated blood pressure and concentrations of pro- and anti-inflammatory cytokines through VNS with a chronically implanted device.

The bidirectional current-limiting circuit depicted in Fig. 2b is useful beyond medical implants, as it can be utilized practically anywhere bidirectional current-limiting is needed. Although bidirectional current-limiting can be achieved with two antiparallel CCDs (Fig. 2a), the symmetry of this system is dependent on the matching of two transistors and two resistors. For our system, symmetry is dependent on the matching of two resistors and two Schottky diodes, which is advantageous because even though high-quality transistor matching is easy to do on a single wafer, it is difficult with discrete components. From our preliminary testing, we found that the discrete resistors and Schottky diodes were easier to match than discrete JFETs, leading us to select our design. The discrepancies between the measured and theoretical values in Fig. 2c can most likely be attributed to the inaccuracies in the approximations of the JFET parameters and the imperfect symmetry in the resistors and Schottky diodes.

To the best of our knowledge, this is the first published instance demonstrating both pro- and anti-inflammatory serum cytokine modulation using a chronically implanted, untethered VNS device in a rodent model. Previous chronic studies have only shown modulation of pro-inflammatory serum cytokines^39,53,54. While our lab previously demonstrated that chronic cuffing of the vagus nerve in a rat compromised the efferent fibers³⁰ regardless of whether stimulation was administered, we believe that the leadless design used in this study helped alleviate the stress on the nerve, enabling efferent effects observed in our cytokine and blood pressure data. A recent study by Tu et al. showed an inverse relationship between electrode thickness and the evoked local inflammatory response⁵⁵. While their polyimide electrode is not a perfect analog to our parylene electrode, their results support this theory. When considering the thickness, diameter, and pre-molded cylindrical shape of our cuff (as opposed to their planar cuff that must be curled around the nerve), as well as the fact that the elastic modulus of parylene is less than that of polyimide, extrapolating their results to our system suggests that our device applied a relatively low mechanical stress to the nerve, which we believe contributed to our positive results. It is possible that other researchers attempted cytokine modulation through VNS with a chronically implanted device or cuff, but caused damage to the nerve and failed to see the desired results, so they did not publish the negative data. Mughrabi et al. successfully modulated the pro-inflammatory cytokine TNF-α using a chronically implanted cuff electrode connected to a head-capped system, and they highlighted the importance of the surgical technique and cuff placement to avoid unnecessary stress on the nerve³⁹. In their study, mice were stimulated 9–17 days after cuff implantation, receiving three 5-minute stimulation sessions (1 mA, 250 µs pulse width, 30 Hz) over two days prior to injecting LPS, with serum TNF-α levels assessed 90 minutes later. This protocol produced a significant reduction in TNF-α of about 1250 pg/mL, comparable in magnitude to our 1000 pg/mL decrease. Notably, most animals did not display changes in heart rate during stimulation, indicating that cytokine modulation can occur with subthreshold stimulation intensities insufficient to alter heart rate. In contrast, mice implanted for 2–6 weeks and treated with only a single stimulation session (individualized heart rate threshold current, 250 µs pulse width, 10 Hz, 5 minutes) prior to LPS injection did not show significant changes in TNF-α. Importantly, their stimulation occurred prior to treatment, making it markedly different from our stimulation after LPS was already injected. In many diseases where VNS is a potential treatment, such as TBI or acute inflammatory conditions, administering stimulation before injury or disease induction is not possible. Therefore, we focused on the rescuing effect of VNS as opposed to its preventative effects, making our findings more relevant to real-world scenarios. Furthermore, we show that this device can be used for fiber-selective stimulation through TIS, as well as multisite stimulation, traditional biphasic stimulation, and conduction block, enabling easier adoption in additional neuromodulation applications.

It is important to note that the strength of the inductive link determines the stimulation strength, leading to two effects. The first was that it became difficult to align the transmit and receive coils in the chronic animals. It was much easier in the acute animals since the incision site was clearly visible, giving the user a sense of where to place the coil. This may have contributed to why we saw more defined changes in blood pressure in the acute animals. The second is that in some of the animals, the neck occasionally twitched during stimulation, as we were likely activating motor fibers in the vagus as well as our targeted fibers. This twitching would, in some instances, slightly move the implanted coil relative to the externally fixed coil, changing the quality of the inductive link for better or worse, causing varying levels of stimulation. Prevention of off-target responses to neural stimulation is a significant challenge that many researchers are addressing with various techniques^45,56–61. In future work, we plan to combine fiber-selective stimulation of efferent fibers with refined current-limiting behavior to minimize overstimulation and activation of these motor fibers. Despite these limitations, there are also certain advantages to using a battery-free device, most importantly the reduction in size, as the battery frequently accounts for more than half of the device volume⁶². Additionally, avoiding battery requirements eliminates the concern that the device will run out of charge or pose a safety risk if the package fails.

Rodent models are excellent in research because they are relatively inexpensive, have shorter gestation times, are typically easier to work with, allow for genetic modification, and are usually met with a lower regulatory threshold than larger animals⁶³. However, up to this point, very little clinical research on VNS therapies has been done using rodent models because, while rodents are often used for in vivo models of biomedical research intended to translate to humans⁶³, they differ in some major ways: primarily in size. The rat cervical vagus nerve is roughly ten times smaller in diameter than the human cervical vagus nerve⁶⁴. On this scale, accounting for the stress applied to the vagus nerve is essential toward functional stimulator design. Our device, measuring 7.56 mm³ and weighing 19.5 mg, is small enough to be sutured directly around the rat vagus nerve, eliminating concerns for damage caused by tethered leads, while its passive circuitry and ultrathin packaging technique allow for it to be implanted chronically and still achieve significant autonomic modulation.

To our knowledge, this is the first demonstration of pro- and anti-inflammatory cytokine modulation using a chronically implanted VNS device in a rodent model. This work serves as a turning point not only in inflammatory control but also in other VNS applications. It would be interesting to apply this technique to a disease model in an animal other than the LPS/endotoxic shock model. Applying this VNS technique to an animal model of depression, irritable bowel syndrome, rheumatoid arthritis, or TBI and measuring physiological and behavioral outcomes could produce insightful, disease-specific results. This technique may also be utilized to create a rodent model for VNS treatment of epilepsy, as there has yet to be a chronic VNS rodent model for the treatment of epilepsy. Finally, it might also be worth exploring the difference between preventative and rescuing VNS, as many endotoxemia VNS papers administer VNS prior to sepsis onset, which is not as clinically relevant. This work can be viewed as a stepping stone to other chronic animal treatment models for inflammatory diseases and future human clinical translation and impact (Fig. 8).

Fig. 8. Conceptual depiction of a human temporal interference stimulator (TIS).

Inset: rendering of two neurostimulators encased in a parylene C cuff, selectively stimulating a specific fiber within the fascicular vagus nerve (created in BioRender. Lawlor, G. (2025) https://BioRender.com/4swmdps).

Methods

Electrical Hardware

Before taking measurements from the unidirectional or bidirectional CCDs, accurate JFET parameters were measured. A 100 Ω resistor was placed in series with the drain of the JFET (2SK879-Y(TE85LF), Toshiba Semiconductor), and 5 V was applied to the free terminal of the resistor while the source of the JFET was grounded using a DC power supply (E36312A, Keysight). The gate voltage was swept while monitoring the current through the resistor using an oscilloscope (MSOX3014T, Keysight). The voltage at which the drain current exceeded the noise floor of the system was specified as the threshold voltage. The gate-source voltage was then set to 0 V, and the current through the resistor was measured to give the saturation current I_DSS while measuring the drain-source voltage to ensure the JFET remained saturated.

To test the current-limiting capabilities of the unidirectional and bidirectional CCDs, simplified circuits depicted in Fig. 2a, b were utilized. A 1 kΩ resistor was used as the load. The input signal was driven with a function generator (33500B, Keysight). For the unidirectional case, the input signal was a 100 Hz, 20 V_PP sinusoid with no DC offset; in the bidirectional case, the input signal was a 100 Hz, 10 V_PP sinusoid with no DC offset. Using an oscilloscope, the voltage across the current-limiting circuit and the current through the load resistor were measured.

To calculate the series and parallel capacitor values (C_P and C_S) for the matching/tuning networks, we did not include the loading effect of the inductive link in our calculations. Preliminary testing showed that the loading effect negligibly changed the impedance of the two coils as the coupling coefficient was so low at anatomically reasonable powering distances (benchtop measurements of the coupling coefficient show the value ranging from ~0.003–0.025 over distances of 5−25 mm). The impedance of the transmit coil was measured using a vector network analyzer (VNA) (E5071B, Agilent Technologies) at the desired RF carrier frequency. The required capacitances to match the transmit coil to the 50 Ω output impedance of the power amplifier (ZHL-1-2W-S + , Mini-Circuits) were calculated. Then the impedance of the capacitor-coil network was measured with the VNA, and the capacitance values were manually adjusted until the network was matched to the desired impedance. Typical values for 13.56 MHz were C_P $\cong$ 720 pF and C_S $\cong$ 300 pF. Typical values for 27.12 MHz were C_P $\cong$ 170 pF and C_S $\cong$ 34 pF. The input impedance of the stimulator circuit with a 1 kΩ load to emulate the cuff electrode impedance was measured using the VNA at the desired carrier frequency and a signal power of +10 dBm. The receive coil was then tuned in the same way as the transmit coil, but to match the impedance of the stimulator circuit. The transmit coil was created from 22 AWG (0.64 mm diameter) enameled magnet wire wound two turns around an 18 mm cylinder and soldered into a simple PCB with series and parallel tuning capacitors and an SMA connector. The device current across a 1 kΩ load as a function of distance is described Supplementary Note 2 and shown in Supplementary Figs. 1 and 2.

Device Packaging

The device packaging process, depicted in Fig. 3, was used to electrically insulate and hermetically seal the devices for implantation. First, custom-designed, flexible polyimide PCBs (PCBWay) were constructed out of a rectangular Archimedean spiral composed of seven turns, with each trace measuring 125 μm wide and 18 μm thick. Coil windings were printed on the backside of the passive stimulator circuit to minimize the footprint of the implanted device. Boards were populated and tested. Then, 7-strand twisted 75 μm platinum-iridium (Pt-Ir) wires (Fort Wayne Metals) were soldered to the PCB and adhered ~270° circumferentially around a 10 mm piece of 19 AWG (0.912 mm diameter) solid-core copper wire using UV-cure cyanoacrylate, 2 mm apart from each other. This entire assembly was then conformally coated with a ~20 µm layer of parylene C. The cylindrical parylene coating on the copper wire was sliced longitudinally along the copper wire, between the opening in the Pt-Ir wires, and circumferentially 0.5 mm outside the wires to create the cuff opening and cut it to length. The entire assembly was submerged in 80 °C copper etchant (Sigma-Aldrich) for ~24 hours to dissolve the copper wire. After rinsing with isopropyl alcohol, devices were placed under a microscope to ensure that all the copper was etched from the cuff and that the area around the device was free from any liquid, which would indicate damage to the encapsulating parylene. Next, the entire assembly was submerged in acetone for another 24 hours to dissolve the cyanoacrylate and expose the Pt-Ir electrodes on the inside of the cuff. Once the packaging process was complete, we tested the device by carefully connecting the Pt-Ir electrodes to a 1 kΩ load and measuring the voltage across the load on an oscilloscope while applying the stimulation waveform with the transmit coil to ensure it still had the same output signal as before packaging. Resistive changes or shorts due to liquid inside the package were not observed. The final packaged device is shown in both Figs. 1 and 3. Prior to chronic implantation, the devices were sterilized using a liquid sterilant (CIDEX OPA Solution, Advanced Sterilization Products) and then dipped in sterilized water to remove any residues.

Animal Experiments

All procedures involving animals were approved by and performed in accordance with the Johns Hopkins University Institutional Animal Care and Use Committee. We have complied with all relevant ethical regulations for animal use. All animals were housed with a 12-hour light/dark cycle with ad libitum access to food and water. A total of 26 Long-Evans rats were used.

Fiber Selectivity Experiments

To surgically isolate the sciatic nerve, a male Long-Evans rat (557 g) was induced with 5% inhaled isoflurane mixed with oxygen at a rate of 2 L/min. Then, isoflurane was discontinued, and the rat was injected intraperitoneally with 1.5 g/kg urethane (Sigma-Aldrich) dissolved in water. Additional urethane was given intraperitoneally in increments of 10% of the original dose every 30 minutes until a surgical plane of anesthesia was achieved, verified by the toe-pinch reflex. A constant body temperature of 36.5 °C was maintained using a closed-loop heater with a rectal thermometer (PY2 50-7212, Harvard Apparatus). The dorsal side of the hindlimb was shaved superficial to the gluteal and biceps femoris muscles, extending medially to the midline. A 1.5 cm incision was made in the skin following the curvature of the gluteal muscles. Blunt dissection between the gluteus maximus and biceps femoris was used until reaching the sciatic nerve. The nerve was separated from the surrounding tissue and placed in the cuff. A silk suture was passed around the cuff, circumferentially, between the wires extending from the cuff to the device, and tied to close and secure it around the nerve.

For the first sciatic experiment, a single cuff with four circumferential electrodes and two neurostimulator boards was used. The incision site was filled with lactated Ringer’s solution to prevent drying and sutured closed with a simple uninterrupted pattern. The leg was bent to a 90-degree angle and secured in place with tape. Hook electrodes made by crimping stainless steel wire to 30 gauge stainless steel needles bent into a U-shape were placed near the fifth digit and at the apex of the sole of the foot as in Budde et al. to record electromyography from the plantar muscles of the foot⁴⁵. A second-order Butterworth bandpass filter (AM 1700, AM-Systems), ranging from 100–500 Hz with a gain of 100 and a 60 Hz notch was used to filter and amplify the signal. The signal was digitally sampled at 10 kHz using a National Instruments data acquisition unit (NI-DAQ) (USB-6343, National Instruments) controlled by a custom MATLAB script running in MATLAB 2022a (MathWorks). The RF AM waveforms were generated using a signal generator (N5172B, Agilent Technologies) and amplified through a power amplifier (ZHL-1-2W-S + , Mini-Circuits). The power output of the signal generator was set so that the total output power from the power amplifier was 1 W. The transmit coils were aligned with the receive coils and held in place using a “third hand” clamp. The different stimulation setups (20 Hz bipolar, 20 Hz TIS, 2 kHz conduction block) were each administered for 5 seconds with a 10 second break in between. The signals were generated by amplitude modulation of the carrier RF signal. For the 20 Hz bipolar stimulus, the carrier waveform was modulated to be at 100% amplitude for 25 ms, and then at 0% amplitude for 25 ms, making a 20 Hz square pulse. For the 20 Hz TIS case, one device was modulated with a 2 kHz sine wave and the other with a 2.02 kHz sine wave, as shown in Fig. 1c. Then, the incision site was reopened, and the nerve was cut distal to the cuff electrode. The stimulation setups were repeated to ensure the results were caused by neural stimulation and not direct muscle stimulation.

For the fiber-selective sciatic experiment, a male Long-Evans rat (291 g) was anesthetized as described above. A 12-electrode cuff with longitudinal contacts was placed on the sciatic nerve, as in Budde et al.⁴⁵. The incision site was filled with lactated Ringer’s solution and covered with a small piece of plastic wrap to prevent drying. The leg was secured at a 90-degree angle with tape, and hook electrodes were placed in the biceps femoris and foot (described above) to record electromyography from the hamstring and plantar muscles. The signals were filtered, amplified, and sampled as described previously. The electrode leads were also secured with tape to prevent the cuff from moving during the experiment. The two neurostimulator boards were aligned with their respective transmit coils, held in place using “third hand” clamps. Hook test clips were used to connect the stimulator output from the printed circuit boards to the respective electrode in the cuff. Two signal generators and power amplifiers were used to generate the RF signals for the devices. A 100 Ω resistor was connected in series between each device and the positive electrode to record the stimulation current amplitude. We mapped the fascicular nerve by applying TIS to circumferentially adjacent electrode pairs at equal amplitude, determining the activation threshold for each muscle, and then proceeding to the next electrode pair as in Budde et al.⁴⁵. The setup can be seen in Supplementary Fig. 3. Then, we investigated the activation map for the two muscles and chose electrodes to apply fiber-selective TIS, trying different stimulation parameters until we were first able to stimulate one muscle discretely, then modifying the stimulation parameters (while leaving the electrode positions unchanged) until distinct activation of the other muscle occurred. At the end of the experiment, the nerve was cut distal to the cuff electrode. The stimulation setups were repeated to ensure the results were caused by neural stimulation and not direct muscle stimulation (Supplementary Figs. 4 and 5).

Chronic Vagus Device Implantation

We utilized 12 Long-Evans rats – 5 male (319.6 ± 24.8 g) and 7 female (257.7 ± 34.7 g) – for the chronic VNS experiments. Sample size for experiments was determined using an a priori power analysis with an estimated effect size of 0.9, power of 0.8, and significance level of 0.05, resulting in a group size of 12 animals. This process was repeated independently for acute and chronic studies. Anesthesia was induced with 5% inhaled isoflurane mixed with oxygen at a rate of 2 L/min and maintained with 1.5–2.5% isoflurane mixed with oxygen at a rate of 2 L/min. A preoperative dose of butorphanol was given at a concentration of 1 mg/kg, subcutaneously. Anesthetic depth was monitored using a toe-pinch reflex, respiration rate, heart rate, and blood oxygenation. Implant devices were sterilized by soaking in CIDEX for at least 30 minutes and then washed off with sterile saline. The ventral side of the neck was shaved using an electric razor from the top of the ribcage to the jaw. A constant body temperature of 36.5 °C was maintained using a closed-loop heater with a rectal thermometer. Following proper aseptic technique, the incision site was cleaned with alternating scrubs of betadine and sterile alcohol pads and draped with a sterile drape. Using ethylene oxide-sterilized tools, a 15 mm incision was made through the ventral side of the neck, parallel to the trachea, roughly 2 mm lateral to the midline, on the animal’s left side. The tissue was bluntly dissected, exposing the sternohyoid and sternomastoid muscles. Both muscles were retracted to expose the carotid sheath. The vagus nerve was carefully separated from the carotid artery and placed within the cuff of the device. A silk suture was tied around the parylene cuff to close it and ensure the nerve remained inside. The PCB was situated below the sternomastoid muscle, and the retractors were removed. The incision site was filled with sterile lactated Ringer’s solution to prevent drying of the nerve. The incision site was closed with polyglycolic acid sutures and coated with triple-antibiotic ointment. Implanted animals received buprenorphine at a concentration of 0.1 mg/kg and cefazolin at a concentration of 30 mg/kg every 12 hours for 72 hours following surgery. Animals were monitored for positive signs of recovery and for signs of infection.

Vascular Catheterization

Rats were induced with 5% inhaled isoflurane mixed with oxygen at a rate of 2 L/min. Then, isoflurane was discontinued, and the rat was injected intraperitoneally with 1.5 g/kg urethane (Sigma-Aldrich) dissolved in water. Additional urethane was given intraperitoneally in increments of 10% of the original dose every 30 minutes until a surgical plane of anesthesia was achieved, verified by the toe-pinch reflex. The catheterization was performed in the same manner as in Biggs et al.⁴⁹. The ventral side of the portion of the leg that meets the abdomen was shaved with an electric razor. An 18 mm skin incision was made along the medial portion of the thigh where it meets the abdomen. The abdominal wall was bluntly dissected from the muscles of the thigh, exposing the femoral artery and vein. The two vessels were separated from each other. The distal ends of the vessels were occluded by ligating with sutures, and the proximal ends were lifted with loosely tied sutures to stop blood flow. Incisions were made in the vessels, and cannulae, preloaded with heparinized saline, were inserted into the vessels and advanced until the abdominal aorta and inferior vena cava, respectively. A suture was placed around each vessel and cannula to help keep it in place. The venous cannulae were made of silicone rubber (CX-2020S, BASi Research Products); the arterial cannulae were custom-made from polyethylene tubing (AM-Systems) to prevent distension from arterial pulse pressure. The incision site was then filled with lactated Ringer’s solution and covered with plastic wrap to prevent drying of the tissue.

Acute Vagus Device Implantation and Vascular Catheterization

We utilized 12 Long-Evans rats – 7 male (299.3 ± 13.0 g) and 5 female (249.6 ± 18.9 g) – for the acute VNS experiments. In these animals, the stimulator implantation and vascular catheterization procedures took place on the same day under urethane anesthesia; otherwise, they were performed as described above for the chronic animals.

Blood Collection Procedure and LPS Injection

Lipopolysaccharide (LPS) from Escherichia coli, serotype O111:B4 (L2630, Sigma-Aldrich), was supplied as a lyophilized powder and reconstituted in sterile saline to a final concentration of 5 mg/mL, then vortexed for 15 minutes and stored at 4 °C. Every 30 minutes, starting 30 minutes prior to the intraperitoneal injection of 5 mg/kg LPS, until 120 minutes after the injection, blood was collected from the venous cannula. First, the lock solution was drawn until fresh venous blood had traveled through the catheter and into the collection syringe. This waste syringe was discarded, and a fresh syringe was used to pull 200 µL of blood. The total fluid volume of the two syringes was then replaced by venous injection of lactated Ringer’s solution to prevent hypovolemia. The blood sample was transferred to a microcentrifuge tube and allowed to clot for 15 minutes. Then, the sample was centrifuged at 2000 × g for 10 minutes. The supernatant blood serum was transferred to another microcentrifuge tube and frozen at −20 °C until analysis. For the 3^rd blood sample, the time of the waste draw, sample draw, and fluid replacement were recorded in conjunction with the blood pressure.

VNS Procedure

Thirty minutes after the LPS injection, the 3^rd blood sample was collected. Immediately after replacing the withdrawn fluid, VNS was started for 5 minutes. The stimulation waveform was a 5 Hz square pulse (100 ms on, 100 ms off) with a 5 mA maximum current limit (actual delivered current was likely lower due to the inductive link, see Discussion and Supplementary Fig. 2) generated by modulating the carrier waveform at 5 Hz. The transmit coil was aligned with the implanted coil as closely as possible. In the acute animal experiments, a mark was made on the skin with a permanent marker directly superficial to the implanted device to facilitate coil alignment. This was not possible in the chronic animal experiments, so the coils were aligned to the best estimate of the experimenter. The coil was positioned as close to the animal’s neck as possible without touching and held in place by a flexible-arm clamp. The signal generator was configured to operate at a power level that produced a 1 W power output at the output of the power amplifier (input to the transmit coil matching network). The signal generator was active for a total of 5 minutes. The time when stimulation was turned on and off was recorded in conjunction with the blood pressure recording. Control animals did not receive electrical stimulation.

Blood Pressure Recording and Processing

Blood pressure data were recorded using the same custom amplifier circuit as in Biggs et al. and digitized using a NI-DAQ⁴⁹. Data was recorded starting from 5 minutes prior to the 3rd blood draw for a total of 15 minutes. The blood pressure signal was sampled at 10 kHz using a NI-DAQ controlled by a custom MATLAB script. To get mean arterial pressure (MAP), another custom MATLAB script running MATLAB 2021b was executed to apply a digital lowpass filter with a 6th-order Butterworth filter at 100 Hz to the data before applying a moving window average with a two-second window length. Data for each animal can be found in Supplementary Figs. 6 and 7.

To assess the impact of VNS on blood pressure, the MAP recordings were divided into three different regions: pre-stimulation, from the start of recording to just before the 3^rd blood collection started (waste collection); stimulation, from 30 seconds after the start of stimulation to 5 seconds after stimulation stopped; and post-stimulation, from 10 seconds after stimulation ended until the end of the recording. These bounds were chosen due to the interference of the blood collection during the stimulation window and because when stimulation stopped, the MAP often increased for 5−10 seconds, which we wanted to capture as a stimulation effect. For each region, we calculated the mean and the rolling mean absolute deviation (MAD), a measure of local variability. Rolling MAD was used because standard deviation and MAD are both based on each point’s difference from the mean or median, but for nonstationary data like MAP, this can artificially increase variance. The moving mean for each region was calculated in Python (v3.12.12) using the “uniform_filter1d()” function from the SciPy.ndimage library (v1.16.2) with a 60 second window. To avoid the shortening of the output that typically occurs with convolution, edge values were extended by holding the first and last values of the MAP constant, so the moving mean was the same length as the MAP. Since each animal could have a different mean or rolling MAD based on catheter placement and biological variance, the difference between the mean and rolling MAD for stimulation vs. pre-stimulation and stimulation vs. post-stimulation was used to make comparisons between groups.

Cytokine Immunoassay Analysis and Processing

Blood serum samples were analyzed at the Johns Hopkins University Core Facilities laboratory using custom MSD V-PLEX inflammatory assays (Meso Scale Diagnostics) measuring the concentration of TNF-α, IL-1β, IL-6, and IL-10 for rats. The samples were analyzed in duplicate, averaging the two values to get a single concentration value for each time point in each animal. Data for all cytokines can be found in Supplementary Fig. 8.

Polynomial smoothing spline interpolation was performed using Python and the “csaps” library (v1.3.3; Bochkarev, 2024), with a smoothing parameter of 1 to preserve all sample values. This approach generated a new set of interpolated values defined by one-minute increments, allowing for a better estimation of peak concentration, as validated in our previous work²⁸. While concentrations of TNF-α and IL-10 exhibited a peak for each animal, IL-1β and IL-6 did not, as we did not collect samples long enough to capture the entire cytokine profile, so we did not interpolate values²⁸.

Statistics and Reproducibility

All statistical analyses were performed using Python (v3.12.12) with the SciPy library (v1.16.2). Two-tailed, unpaired t-tests were used to compare: (1) differences in MAP features (mean and rolling MAD) between stimulation vs pre-stimulation and stimulation vs post-stimulation periods, and (2) group means of interpolated peak concentrations for cytokine data. Assumptions of normality were assessed using the Shapiro-Wilk test, while homogeneity of variance was tested with Levene’s test. Most comparisons met the assumptions, so Student’s t-test was applied, but Welch’s t-test was used for comparisons violating homogeneity of variance. Effect sizes were calculated using Hedges’ g to account for the small sample size. All reported p-values were corrected with the Benjamini-Hochberg procedure for false discovery rate (FDR). A significance level of α = 0.05 was considered statistically significant. No covariates were included, as sex did not influence grouping. Sample sizes are reported in the figure legends. Central tendency and variation in text are expressed as mean ± standard deviation. Biological replicates are individual rats, with one measurement per animal. For cytokine assays, technical replicates were averaged for analysis. MAP measurements consisted of single recordings per animal. All experiments are reproducible using the methods provided.

Reporting Summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.T.W. conceived the study, designed the neurostimulator system, fabricated the devices, performed the animal experiments, processed the data, and wrote the manuscript. G.L.L. assisted with the benchtop system testing, performed the animal experiments, processed the data, conducted statistical analyses, and wrote the manuscript. B.J.C. developed the packaging procedure, assisted with device fabrication and packaging, and edited the manuscript. M.T.W., G.L.L., and B.J.C. prepared the figures. P.P.I. contributed to the study conception and experimental design and secured funding. All authors reviewed and approved the final manuscript.

Peer review

Peer review information

Communications Biology thanks Kaiyuan Yang, Timir Datta-Chaudhuri, and the other anonymous reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Christina Karlsson Rosenthal.

Data availability

All numerical data for generating figures and supporting the findings of this study are included in Supplementary Data 1. The raw data, analysis scripts, and printed circuit board files are openly available at the following URL: 10.7281/T1DMJNO8⁶⁵.

Competing interests

The authors declare the following competing interests: filed patent for the system 63/280,652, Interferential Current Stimulator for Peripheral Nerves.

Supplementary Information

The online version contains supplementary material available at 10.1038/s42003-026-09989-x.