Fine-tuning brain stimulation to fight Parkinson’s disease

Sixteen years ago, Simon Little was a junior neurologist thinking of devoting his time to international health work. But then he attended a lecture by Peter Brown.

Brown, a neuroscientist at the University of Oxford, UK, described his studies of people with Parkinson’s disease who had had electrodes implanted in their brains. The electrodes were intended to treat symptoms of this neurodegenerative disease by targeting electrical pulses to structures deep inside the brain. But to start with, Brown used them to record the oscillating waves of neuronal electricity that these brain structures emitted.

The recordings showed that as a person’s symptoms ebbed and flowed — their movements slowing then normalizing, their tremors flaring and subsiding — the oscillations changed¹. “I was completely fascinated,” Little says. “The music of the brain was encoding something really powerful about people’s symptoms.”

Nature Outlook: Parkinson’s disease

After the lecture, Little asked Brown if he could do a PhD with him. Little laughs as he recalls how Brown tried to warn him off, saying that the field was a complete backwater and that Little would be pursuing this work at his own professional risk.

Undeterred, Little joined Brown’s laboratory, and the risk paid off. Little, now at the University of California, San Francisco (UCSF), and Brown took the signal most strongly related to the intensity of Parkinson’s symptoms — an oscillation in what’s termed the beta band, occurring 12–30 times per second — and used it as the trigger for deep brain stimulation (DBS).

Typically, DBS zaps the brain constantly. But by activating the device only when beta activity crested, Little and Brown could provide stimulation when a person most needed it. In an initial trial, numerical ratings of volunteers’ mobility were about 30% better than they were with constant stimulation². “We found that less was more,” Little says.

Last year, the first ‘closed loop’ or adaptive DBS (aDBS) system for Parkinson’s, which operates similarly to the set-up that Little and Brown used in their early experiments, was approved by the US Food and Drug Administration and earned a CE mark to be sold in the European Union. The technology — made by Medtronic in Minneapolis, Minnesota — uses the same electrodes to record beta-band activity and stimulate the brain.

Although this system is clinically effective, researchers think that it provides only a fraction of what smart implanted neurotechnologies could one day offer people with Parkinson’s. “This is a breakthrough,” says Wolf-Julian Neumann, a neuroscientist at Charité University Medicine in Berlin. “But it’s a super-tiny step towards the vision that we are pursuing.”

Neumann, Little and others say that emerging techniques and hardware now being tested in small clinical trials will lead to devices that decode, on the basis of neural activity, what someone with Parkinson’s is doing — say, sleeping, walking or speaking — and the exact symptoms they are experiencing. Then, with this real-time information, devices will deliver different forms of electrical stimulation to alleviate users’ symptoms and aid their movements³. “It’s a totally different way of conceptualizing the treatment of brain disorders,” says Neumann.

Constant to adaptive

Parkinson’s disease is characterized by degeneration of the brain’s substantia nigra, a cluster of neurons that produces the neurotransmitter dopamine. This dopamine regulates the brain networks that coordinate voluntary movements. Without it, that system malfunctions.

“In Parkinson’s disease, you’ve got an unwanted brain arrhythmia,” says Helen Bronte-Stewart, a neuroscientist at Stanford University in California. The extent to which neurons synchronize their firing at beta-band frequencies is abnormally high, producing a disruptive signal, she says. “It’s jamming the brain’s ability to coordinate movement.”

This leads to the hallmark symptoms of Parkinson’s: slowed movements, rigidity and tremors. The disease can also cause postural changes, speech difficulties and non-motor symptoms, such as depression or anxiety. In its later stages, cognitive decline is common. The standard treatment is to boost a person’s dopamine levels by dosing them with the drug levodopa. But levodopa can cause side effects, such as uncontrolled movements, and people can experience periods in which the drug’s effectiveness blinks on and off.

Since its clinical approval in 1997, DBS has emerged as a popular alternative for people whose Parkinson’s isn’t well controlled by drugs. Far and away the most common type of brain implant in use today, DBS has done much to normalize the process of undergoing neurosurgery to have a therapeutic device fitted. In its original form, the DBS electrodes — connected by wires that run down the recipient’s neck to a small power pack implanted in the chest — deliver a constant, high-frequency train of electrical pulses. Their target is either the subthalamic nucleus or the globus pallidus, two brain regions that are components of the movement network modulated by dopamine.

Andreas Horn, a neuroscientist at the University of Cologne in Germany, says that ‘stimulation’ is a misnomer in DBS. The high-frequency zapping, he says, “is more about disrupting or overriding some circuitry that’s become dysfunctional”.

The change from continuous DBS to aDBS demanded that Medtronic incorporate the necessary technology into a compact medical device. Little says that in his first experiments, volunteers were wired up to racks of computers, filters and stimulators. Medtronic miniaturized that hardware so that it all fitted into the small implanted power pack.

Bronte-Stewart led the trial that secured the device’s approval⁴. This showed that aDBS was safe and that it controlled recipients’ symptoms effectively. The trial — somewhat controversially, Bronte-Stewart acknowledges — didn’t directly compare aDBS with conventional, continuous DBS. But just sending people on aDBS home and trusting they’d be safe was a groundbreaking step, she says.

Since 2020, Medtronic has been supplying a system capable of running either continuous or aDBS, and roughly 25,000 recipients could potentially switch to aDBS now. However, scientists are already taking note of the limitations of this system — and of using beta-band oscillations as a trigger.

One issue is that current aDBS systems take weeks of inpatient programming, with neurologists observing the effects of different sets of manually selected stimulation parameters⁵. “For this to scale, you can’t have every Parkinson’s patient needing three weeks of a neurologist monitoring them and optimizing this,” says Doris Wang, a neurosurgeon at UCSF.

Another issue is that beta oscillations provide only a crude indication of overall symptoms. “Beta power is essentially a biomarker of global worsening of symptoms of Parkinson’s disease,” says Neumann. Although this information is no doubt useful — not least, says Neumann, for effectively up- and downregulating DBS as levodopa’s effects change — beta-band oscillations don’t discern exactly what someone is experiencing. “Tremor actually reduces beta,” he says. As a result, “the beta-based adaptive DBS algorithm can confuse tremor for a good symptom state.”

What’s more, in current systems that use the same electrodes to stimulate the brain and record activity, artefacts from the stimulation compromise the quality of recorded signals⁶. Plus, these systems are recording from sites in the brain that are selected because they are best places to stimulate, and not necessarily the most informative to record from.

To address these issues and offer much finer control of the disease’s various manifestations, researchers are seeking to develop next-generation aDBS systems. On the input side, they are harnessing advances in brain–computer interfaces, including fresh technologies for recording and decoding brain activity. These, the researchers say, will be able to decipher exactly what users are doing and what symptoms are occurring.

On the output side, clinical studies are showing that it’s possible to halt symptoms of Parkinson’s by targeting separate neural pathways. For instance, Horn’s group has found that, by targeting adjacent but distinct nerve-fibre tracts (using conventional continuous stimulation), they can selectively help with tremor, rigidity, slow movement, or postural and gait-related problems⁷. “This is not so much about when to stimulate,” Horn says, “but rather about where and what to simulate.”

Where, when, how

To map which nerve-fibre tracts passed close to the DBS electrodes, Horn’s team took brain scans of people who had had Medtronic devices implanted in the subthalamic nucleus. The scans showed tracts originating from different parts of the brain, such as the motor cortex or cerebellum, running near to the four contact sites on each electrode.

The researchers then stimulated various combinations of contact sites to steer currents to different chosen tracts. To their surprise, targeting specific fibre tracts ameliorated symptoms individually.

Horn and colleagues then developed an algorithm that, given a set of symptoms, will choose stimulation sites that should provide the most benefit. “If a patient has a lot of tremor, we tell the algorithm, ‘find me the stimulation volume that will maximally engage the tremor tract and not so much the other tracts’,” says Nanditha Rajamani, a neuroscientist at Harvard Medical School in Boston, Massachusetts, who collaborated with Horn on this project.

In a small, five-person trial, the algorithm’s suggested settings produced a bigger improvement in the symptoms of four participants than did the stimulation their neurologists had previously set up for them⁷. The fifth person’s symptoms neither improved nor worsened.

Looking ahead, Horn says that because the manifestations of Parkinson’s shift constantly, the ability to target each symptom type could enable adaptive devices that suppress only the symptoms a person is currently dealing with. But that will require the device to ‘know’ what those symptoms are, meaning that researchers must identify the neural correlates of each symptom type.

To do so, they are recording brain activity while volunteers use wearable devices, such as accelerometers, to precisely track arm and leg movements. Increasingly, recordings are acquired not just from DBS electrodes deep in the brain, but also from a separate set of electrodes placed on the brain surface. These cortical electrodes pick up the signals that arise in the motor cortex to initiate and control movements. “There is a lot more information in the cerebral cortex,” Wang says.

Neumann’s group, for example, has used machine learning to link hundreds of features captured in such neural recordings to both symptoms and drug side effects. The features the researchers used correlated with symptoms better than beta activity did. And importantly, the model worked in people whose data had not been used to develop it.

Neumann thinks that this approach might even allow researchers to identify neural signals that precede symptoms. If so, future systems might not just react to symptoms as they occur, but could also pre-emptively stimulate the brain to prevent tremors and other typical manifestations of Parkinson’s from occurring at all. “When you and I intend to move, there’s a huge spike of dopamine release,” he notes. A well-timed burst of electrical stimulation could functionally substitute for this in people with Parkinson’s and thereby ease upcoming movements.

Neumann also thinks that if a smart DBS system can decode what a person is doing, or trying to do, this could help to select the stimulation activity that best serves that person. For instance, he says, the high-intensity stimulation that suppresses tremor can make it difficult to speak, so if brain signals indicate that a person is attempting to talk, then the tremor-suppressing stimulation might be turned down, or even off.

Likewise, Bronte-Stewart, Little and Wang are searching for neural correlates of movements and symptoms, with a view to controlling stimulation⁸. Little and Bronte-Stewart are also keen to treat symptoms of Parkinson’s that are not related to movement by finding appropriate signals and target structures to combat mood symptoms and cognitive difficulties, respectively⁹.

In June, Wang published a small trial that began to put the pieces together. It used movement-related brain signals to address a hugely problematic aspect of Parkinson’s: gait freezing, in which people who are walking suddenly become unable to move and so fall¹⁰.

After identifying cortical and subcortical signals that indicated the motions of walking, Wang’s group used them to trigger brief stimulation on the relevant side of the brain — the right side to affect the movement of the left leg, and vice versa. People who received this intervention fell less frequently than those who did not. “It’s the first demonstration that such rapid, movement-specific aDBS or closed-loop DBS is possible,” Wang says. Trial participants said that movement felt smoother and more fluid when using the system, and that it was easier for them to get up and walk.

Next steps

It’s safe to say that Brown was wrong when he told Little that attempting to relate brain signals to disease symptoms was a dead end. But over the years, what it takes to bridge the gulf between exploratory research and a clinical product has become more apparent. “Our role as researchers is to show what’s possible,” Little says. Only then can the different questions regarding what can be scaled up into a viable clinical intervention begin to be addressed.

Already, there are questions about next-generation DBS that did not apply to the first adaptive systems. For instance, Little notes that if extra cortical recording electrodes are needed, surgeons will need to learn a new implantation procedure.

He is also wary of emerging systems being heavily dependent on artificial intelligence — either to decode brain activity or to get systems up and running without weeks of manual programming. Concerns swirl about embedding uninterpretable AI technology in medical devices, especially when people’s welfare and behaviour depend on it.

Little questions what would happen if someone using an AI-powered DBS system experienced a major mishap, such as freezing completely or crashing their car, because the algorithm changed its stimulation patterns. “Who bears responsibility for that?” Little says. This is an evolving space for both developers and regulators, who must find a sweet spot with which all parties are comfortable, he says.

The transformation of an idea into a clinical product is a balancing act between what’s possible and what’s plausible, reliable and safe. As Parkinson’s researchers aim to again lead the way in new forms of neurotechnology, they will need to navigate the known and unknown challenges that arise. As Bronte-Stewart says, today’s advances “are coming out of really good labs in academic centres doing a lot of computational work. How is that going to filter down into something that can be embedded in a system in your chest?”