Vagus nerve stimulation

The vagus nerve is usually described by what it tells the organs, slowing the heart among them, but 65–80% of its fibres run the other way, carrying news of the viscera up to the brainstem. In the neck it lies beside the carotid artery, within reach of an incision, so a cuff around it there is a wire into the brainstem that needs no hole in the skull. Vagus nerve stimulation is that wire: a pacemaker-like generator under the left collarbone, a lead tunnelled up to the left cervical vagus, and a milliamp or two for 30 s, then five minutes’ rest, day and night, for epilepsy that drugs have not controlled.

From a dog’s neck to the chest

Jacob Zabara, a physiologist at Temple University, patented the device in 1987 as a “neurocybernetic prosthesis”: a constant-current pulse generator tunnelled to an electrode on the vagus nerve, switched on by a patient who senses a seizure coming or — the patent allowed — by sensors detecting the EEG changes that precede one. In his dog experiments, published in 1992, stimulating the cervical vagus stopped strychnine-induced seizures within 0.5–5 s, the protection outlasted the stimulation about fourfold, and cutting the nerve below the electrode changed nothing: the effect travelled up the nerve, towards the brain.

By then it was in people. The neurologist J. Kiffin Penry and the neurosurgeon William Bell implanted the first patient in 1988 at the Bowman Gray School of Medicine in North Carolina, and in 1990 Penry and J. Christine Dean reported four: seizures stopped in two, fell by 40% in one and did not change in the fourth, a result they called inconclusive. After two randomised trials the FDA approved the system on 16 July 1997 for drug-resistant partial-onset seizures. A fortnight later it approved thalamic deep brain stimulation for tremor.

Why the left, and why the voice changes

The two vagi are not mirror images at the heart. The usual rationale for the left is that it does not supply the sinoatrial node, the heart’s pacemaker; in dogs both nerves reach it, but stimulating the right slows the heart significantly more (Ardell and Randall, 1986). The left is not silent: the manufacturer counted roughly one implant in 875 in which the routine lead test during surgery slowed or briefly stopped the heart.

The larynx cannot be spared. The recurrent laryngeal nerve, which supplies all but one of the laryngeal muscles, leaves the vagus low down — on the left it hooks under the aorta — and climbs back, so its fibres are still in the trunk where the cuff sits. They are recruited well below therapeutic currents: during implantation, vocal-fold EMG appeared at 0.25–0.5 mA and saturated by 0.75–1.0 mA, while titration usually aims for 1.5–2 mA. In the second pivotal trial, 73% of the high-stimulation group reported voice alteration or hoarseness, against 6% at baseline.

An open-loop dose

Electrically, the generator is a pacemaker pointed at a nerve: constant-current, capacitively coupled, charge-balanced pulses, with DC-blocking capacitors in both lead connections. The electrode came from Huntington Medical Research Institutes and the NINDS Neural Prosthesis Program — platinum ribbons in pliable silicone helices that expand and contract with the nerve after surgery — and animal work there found electrical injury falling hardest on the largest myelinated axons, and reduced by less total stimulation, lower frequencies and lower duty cycles (Agnew and McCreery, 1990). The clinical dose is intermittent, as that work advised.

In the pivotal trials it was 30 Hz, 500 µs pulses, 30 s on and 5 min off, the current raised in 0.25 mA steps to whatever the patient tolerated. Nothing in that loop measures the brain. Today’s standard settings descend from those trials, and optimal parameters have never been systematically studied; what retunes them is the clinic — every couple of weeks for the first three months or so, then as response and side effects dictate — which is as far as adaptive control goes here. A sensor arrived only in 2015, and it reads the heart rather than the brain: newer generators record an ECG between the lead in the neck and their own case, and add a train when the rate jumps by a programmed 20–70%, a sign that may accompany a seizure. Reading the cortex itself is the job of responsive neurostimulation.

A constant-current output has to find whatever voltage the electrode demands. The programmer accepts lead impedances of 600–5,300 Ω around a typical 3 kΩ, at which 1.5 mA takes about 4.5 V. Impedance is also how the leads failed. More than fifty broke or went high-impedance before approval, prompting a redesign, and in the FDA’s tally of adverse-event reports from 1997 to 2004, high lead impedance was the commonest malfunction, 351 of 708. Tissue the generator cannot see changes what the current does, too: in one model, 110 µm of fibrous tissue around the nerve halved the fibres a given setting activated (Helmers and colleagues, 2012).

Neuromodulators on cue

Whatever it does to seizures, VNS gave neuroscience a switch on the brain’s neuromodulators. In rats, lesioning the locus coeruleus, the brain’s main source of noradrenaline, blunted its anticonvulsant effect (Krahl and colleagues, 1998), and noradrenaline in cortex and hippocampus rose during stimulation and not between periods of it (Roosevelt and colleagues, 2006). A release confined to stimulation can be timed to what an animal is doing.

Michael Kilgard’s group at the University of Texas at Dallas did that. Pairing tones with brief VNS trains removed the neural and behavioural signs of tinnitus in noise-exposed rats (Engineer and colleagues, 2011). Pairing it with one forelimb movement enlarged that movement’s territory in motor cortex, whereas the same training without VNS did not (Porter and colleagues, 2012), and the effect needed the cholinergic nucleus basalis (Hulsey and colleagues, 2016). That is what crossed the border: a stimulator in the neck made neuromodulator release something you can schedule, and with it the direction of cortical plasticity.

Vivistim put the schedule in a therapist’s hand. Each button press, timed to a movement, delivers 0.5 s of 0.8 mA, 30 Hz pulses. In its pivotal trial all 108 participants, at least nine months after an ischaemic stroke, were implanted; after six weeks of rehabilitation the arm-impairment score (Fugl-Meyer, upper extremity) rose 5.0 points with VNS and 2.4 in controls, who got real stimulation only for the first five movements of each session, to keep the blind (Dawson and colleagues, 2021). The FDA approved it in August 2021. For a controller this inverts the usual brief: the change in the plant is not a disturbance to track but the product.

What it cost

The effects are modest and hard to measure. A sham is hard to run when the stimulation can be heard in the patient’s voice, so both pivotal epilepsy trials compared high stimulation with low — 1–2 Hz, 30 s every 90–180 minutes — and in the second, even low stimulation altered the voice of a third of patients. High cut seizures by 28% on average, low by 15% (Handforth and colleagues, 1998); in the FDA’s summary, 23% of the high group at least halved them, against 16% of the low, a difference that was not significant. Among patients still followed, open-label reductions grew to a median of about 40% at two and three years, but the same summary noted that those data were uncontrolled and might reflect changes in drugs.

Depression is starker. The randomised trial missed its primary endpoint, 15% responders against 10% with sham after ten weeks (Rush and colleagues, 2005); approval followed in July 2005, after an alternate plan compared a year of open-label stimulation with a separate observational group on usual treatment.

And the mechanism is a route rather than an explanation. Zabara credited the thin unmyelinated C fibres, but destroying them with capsaicin left the effect intact in rats (Krahl and colleagues, 2001), and recordings in patients mostly show myelinated fibres responding at clinical currents. In 1997 the FDA’s advisory panel asked for a way to tell, before implantation, who would respond. The settings are still found by titration.