‘Unprecedented’ Artificial Neurons Are Part Biological, Part Electrical—Work More Like the Real Thing

Most people wouldn’t give Geobacter sulfurreducens a second look. The bacteria was first discovered in a ditch in rural Oklahoma. But the lowly microbe has a superpower. It grows protein nanotubes that transmit electrical signals and uses them to communicate.

These bacterial wires are now the basis of a new artificial neuron that activates, learns, and responds to chemical signals like a real neuron.

Scientists have long wanted to mimic the brain’s computational efficiency. But despite years of engineering, artificial neurons still operate at much higher voltages than natural ones. Their frustratingly noisy signals require an extra step to boost fidelity, undercutting energy savings.

Because they don’t match biological neurons—imagine plugging a 110-volt device into a 220-volt wall socket—it’s difficult to integrate the devices with natural tissues.

But now a team at the University of Massachusetts Amherst has used bacterial protein nanowires to form conductive cables that capture the behaviors of biological neurons. When combined with an electrical module called a memristor—a resistor that “remembers” its past—the resulting artificial neuron operated at a voltage similar to its natural counterpart.

“Previous versions of artificial neurons used 10 times more voltage—and 100 times more power—than the one we have created,” said study author Jun Yao in a press release. “Ours register only 0.1 volts, which [is] about the same as the neurons in our bodies.”

The artificial neurons easily controlled the rhythm of living heart muscle cells in a dish. And adding an adrenaline-like molecule triggered the devices to up the muscle cells’ “heart rate.”

This level of integration between artificial neurons and biological tissue is “unprecedented,” Bozhi Tian at the University of Chicago, who was not involved in the work, told IEEE Spectrum.

Better Way to Compute

The human brain is a computational wonder. It processes an enormous amount of data at very low power. Scientists have long wondered how it’s capable of such feats.

Massively parallel computing—with multiple neural networks humming along in sync—may be one factor. More efficient hardware design may be another. Computers have separate processing and memory modules that require time and energy to shuttle data back and forth. A neuron is both memory chip and processor in a single package. Recent studies have also uncovered previously unknown ways brain cells compute.

It’s no wonder researchers have long tried to mimic neural quirks. Some have used biocompatible organic materials that act like synapses. Others have incorporated light or quantum computing principles to drive toward brain-like computation.

Compared to traditional chips, these artificial neurons slashed energy use when faced with relatively simple tasks. Some even connected with biological neurons. In a cross-continental test, one artificial neuron controlled a living, biological neuron that then passed the commands on to a second artificial neuron.

But building mechanical neurons isn’t for the “whoa” factor. These devices could make implants more compatible with the brain and other tissues. They may also give rise to a more powerful, lower energy computing system compared to the status quo—an urgent need as energy-hogging AI models attract hundreds of millions of users.

The Life of a Neuron

Previous artificial neurons loosely mimicked the way biological neurons behave. The new study sought to recapitulate their electrical signaling.

Neurons aren’t like light switches. A small input, for example, isn’t enough to activate them. But as signals consistently build up, they trigger a voltage change, and the neuron fires. The electrical signal travels along its output branch and guides neighboring neurons to activate too. In the blink of an eye, the cells connect as a network, encoding memories, emotions, movement, and decisions.

Once activated, neurons go into a resting mode, during which they can’t be activated again—a brief reprieve before they tackle the next wave of electrical signals.

These dynamics are hard to mimic. But the tiny protein cables G. sulfurreducens bacteria use to communicate may help. The cables can withstand extremely unpredictable conditions, such as Oklahoma winters. They’re also particularly adept at conducting ions—the charged particles involved in neural activity—with high efficiency, nixing the need to amplify signals.

Harvesting the nanocables was a bit like drying wild mushrooms. The team snipped them off collections of bacteria and developed a way to rid them of contaminants. They suspended the wispy proteins in liquid and poured the concoction onto an even surface for drying. After the water evaporated, they were left with an extremely thin film containing protein nanocables that retained their electrical capabilities.

The team integrated this film into a memristor. Like in neurons, changing voltages altered the artificial neuron’s behavior. Built-up voltage caused the protein nanowires to bridge a gap inside the memristor. With sufficient input voltage, the nanocables completed the circuit and electrical signals flowed—essentially activating the neuron. Once the voltage dropped, the nanocables dissolved, and the artificial neurons reset to a resting state like their biological counterparts.

Because the protein wires are extremely sensitive to voltage changes, they can instruct the artificial neurons to switch their behavior at a much lower energy. This slashes total energy costs to one percent of previous artificial neurons. The devices operate at a voltage similar to biological neurons, suggesting they could better integrate with the brain.

Beating Heart

As proof of concept, the team connected their invention to heart muscle cells. These cells require specific electrical signals to keep their rhythm. Like biological neurons, the artificial neurons monitored the strength of heart cell contractions. Adding norepinephrine, a drug that rapidly increases heart rate, activated the artificial neurons in a way that mimics natural ones, suggesting they could capture chemical signals from the environment.

Although it’s still early, the artificial neurons pave the way for uses that seamlessly bridge biology and electronics. Wearable devices and brain implants inspired by the devices could yield prosthetics that better “talk” to the brain.

Outside of biotech, artificial neurons could be a greener alternative to silicon-based chips if the technology scales up. Unlike older designs that require difficult manufacturing processes, such as extreme temperatures, this new iteration can be printed with the same technology used to manufacture run-of-the-mill silicon chips.

It won’t be an easy journey. Harvesting and processing protein nanotubes remains time consuming. It’s yet unclear how long the artificial neurons can remain fully functional. And as with any device including biological components, more quality control will be needed to ensure even manufacturing.

Regardless, the team is hopeful the design can inspire more effective bioelectronic interfaces. “The work suggests a promising direction toward developing bioemulated electronics, which in turn can lead to closer interface with biosystems,” they wrote. Not too bad for bacteria discovered in a ditch.