The translucent chip on the mouse’s leg didn’t look like a neuron at all. Dotted with an array of sensors and channels, and smaller than a human finger, it looked – and flexed – like a band-aid. But when doused with dopamine, the chip worked its magic. The mouse’s leg began to twitch and stretch. Depending on the dose of dopamine, the chip controls the penis like a marionette.
The chip is an artificial neuron, but nothing like previous chips built to mimic the brain’s electrical signals. Rather, it takes over and adapts the brain’s other communication channel: chemicals.
These chemicals, called neurotransmitters, are the brain’s “natural language,” said Dr. Benhui Hu from Nanjing Medical University in China. An artificial neuron using a chemical language could, in theory, easily tap into neural circuits – to control a mouse’s leg, for example, or to build an entirely new family of brain-controlled prostheses or neural implants.
A new study led by Hu and Dr. Xiaodong Chen at Nanyang Technological University, Singapore, took a major step towards seamlessly connecting artificial and biological neurons into a semi-living circuit. Fueled by dopamine, the setup wasn’t a simple one-way call with one component activating another. Rather, the artificial neuron looped with multiple biological counterparts and pulsed dopamine while receiving feedback to alter its own behavior.
In a way, the system behaves like an interneuron, serving as a decision maker in the brain for fine-tuning neural circuits. “Much of intelligent information — including memory and emotion — is encoded in or transmitted by chemical molecules such as neurotransmitters, and we set out to build an artificial neuron that mimics how a real neuron communicates,” the authors said.
The other side of the story
You’ve heard this classic story about neural networks. A neuron receives an electric shock that travels down its convoluted branches. If the signal is strong enough, it will activate – or suppress – the next neuron, connecting the two into a network. This neuroscientific dogma, popularized as “neurons firing together, wiring together,” is the basis of many neuromorphic chips built to reverse engineer this electrical quirk for low-power, high-efficiency computation.
The data or “memory” of these activities is stored in synapses. I like to think of these intricate structures as two riverbanks with a stream flowing between them. One bank is part of the neuron that sends signals, the other is part of the receiving neuron.
But what helps signals to cross the stream?
Enter neurotransmitters. Once a neuron integrates its received electrical signals, the impulses travel down the branches until they reach a synapse. Here, the signals tell dozens of parked “boats”—imagine little soap bubbles—each filled with neurotransmitters, to launch across the bank. Once docked, the chemicals in the boats discharge to trigger another electrical signal in the downstream neuron. And the circuit continues to connect the intricate networks of the brain.
Chemical calculations are often ignored in the manufacture of nerve implants, but focusing solely on electrical signals is like ignoring transoceanic cargo routes when planning shipping routes.
“This discrepancy can potentially lead to misinterpretation of the transmitted neuron information,” said the team, which may mislead brain interfaces.
The new study reintroduced chemical thinking into artificial neurons. The team screened a whole range of potential neurotransmitter candidates and focused on dopamine – a multitasker that drives motivation, encodes rewards and controls movement – as the star of the artificial neuron.
The chip contains three main components that mimic a real neuron: it recognizes dopamine, encodes the resulting signal in a “synapse,” and releases dopamine to its neighbors.
The first part is an electrochemical sensor that can detect dopamine at a biological level. Composed of carbon nanotubes with an interspersed dose of graphene oxide, the nanostructure is particularly efficient at picking up tiny bits of dopamine from its surroundings, even when other biological chemicals are clouding the water.
Once recognized, the data is transmitted as an electrical impulse to the next component – a memristor. Like a synapse, a memristor has the built-in ability to change its resistance depending on previous activity – meaning it has “memory”. The higher the resistance, the less it can pass on electrical signals.
The device may sound exotic, but it is actually a (very expensive) piece of cheese bread. The two pieces of bread are made of silver and gold nanoparticles, and the “cheese” is a silk protein that adjusts the resistance of the memristor. It’s a neat setup: the component can support both short- and long-term changes in the “synapse,” mimicking memories that quickly come to mind or are burned into the brain.
It’s a sign of learning. “This means that the system has established a stronger connection with repeated stimuli and is more sensitive to familiar stimuli than to new ones,” the authors said.
Then comes the really cool part. Depending on its resistance, the memristor can heat a hydrogel so that it releases dopamine into pre-etched nanochannels.
All in all, the chip acts like a biological neuron. When stimulated with dopamine, it produces an electrical signal that is encoded at the “synapse”. When the signal is strong enough, it pumps dopamine onto its neighbors.
What remains to be done? Test it with live neurons.
A bio-hybrid bridge
As an initial plausibility check, the team placed the chip in a Petri dish containing cells with the ability to release dopamine, called PC12.
They mimicked the activation of neurons and pumped in a salty mixture that triggered the cells to release dopamine. Startled by the sudden surge of “awake,” the artificial neuron spiked with activity and in turn pumped out its own dose of dopamine to its PC12 neighbors. Once bathed in dopamine, the biological cells changed their electrical current in response (little did they know the chemical came from an artificial neuron).
This type of neural chatter is similar to interneurons. As their name suggests, these neurons act like rungs on a ladder, connecting neural networks and helping to refine circuit activity. Here, the artificial neuron behaved like an interneuron – a sort of ‘traffic controller’ that shapes neural networks and keeps their activity consistent.
Next, the team went one step further and tied the chip to a nerve in a mouse’s leg. Depending on dopamine levels, the leg flexed like a morning stretch and continued to spread as the chemical on the chip increased. In another proof of concept, the team attached the chip to a robotic hand. By varying the amount of dopamine on the chip, the team was able to control the robot in a chemically induced “handshake” — a downward twitch of the mechanical wrist powered solely by dopamine.
It’s not the first time scientists have constructed a chemically based neuron. Back in 2020, a Stanford team connected an artificial neuron to isolated biological ones and showed that an artificial neuron can hybridize to a biological neuron using dopamine as a trigger.
The difference here lies in the feedback capability: the new setup forms a loop with neurons capable of taking in and releasing dopamine while simultaneously altering the network’s “memory”. For the time being, the artificial neuron acts more as a “messenger bridge” that can transmit information. The setup is still too bulky for brain implants, although the authors are working to downsize each component and reduce power consumption.
For the authors, chemical and electrical neuromorphic chips are neither one nor the other. After all, neither is the brain.
“Such chemical BMIs [brain-machine interfaces] could complement electrical BMIs and potentially allow correct and comprehensive interpretation of neural information for use in neuroprosthetics, human-machine interactions, and cyborg construction,” the authors said.
Photo credit: ktsdesign / Shutterstock.com