Originally published 2021-07-14.
Researchers in Germany and Greece have recorded a type of electrical signal in human brain cells that had never been seen before. It appears in the dendrites, the branching input fibres of neurons in the outer layers of the cortex, and it is carried by calcium rather than the sodium that drives most nerve impulses. Computer models suggest it lets a single neuron perform a logical operation that was long thought to need a whole network of cells.
The study, led by Albert Gidon and Matthew Larkum, was published in the journal Science in January 2020 and drew fresh attention when it was widely shared again in 2021. It is a real and important finding about how individual neurons compute. What it does not do, despite some headlines, is explain consciousness or hand engineers a blueprint for thinking machines. Below is what was found, how, and how much it really changes.
What the researchers actually found
The team studied pyramidal neurons in layers 2 and 3 of the human cerebral cortex. These cells have unusually large, heavily branched dendritic trees, and the layers they sit in are especially thick in humans compared with rodents.
When the researchers stimulated the dendrites, they saw electrical spikes that persisted even after they applied tetrodotoxin, a toxin that blocks sodium channels. Blocking calcium channels, by contrast, silenced the signal. They called the new events dendritic calcium action potentials, or dCaAPs.
The signals had an odd property. They were strongest for a particular intensity of input and became weaker when the input grew stronger. Most neural signals behave the other way round: more input, more output. That unusual response curve is the key to why the finding caught the attention of computational neuroscientists.
Larkum, of Humboldt University in Berlin, described the team's reaction as a "'eureka' moment" when they first saw the dendritic action potentials.
How the experiment was done
Living human brain tissue is rarely available for experiments. The team used small pieces of cortex removed during surgery on people with epilepsy, tissue that has to be taken out to reach deeper targets. To check that the results were not a quirk of epileptic brains, they compared them with samples removed from patients undergoing surgery for brain tumours.
They recorded from individual neurons with the patch-clamp technique, which uses a fine glass pipette sealed onto the cell membrane, and imaged the cells' structure with fluorescence microscopy. Earlier, similar experiments in rat neurons had not shown the same kind of signal, which is one reason the result surprised the researchers.
Why a single neuron solving XOR matters
In logic, an XOR (exclusive OR) function gives a "yes" when exactly one of two inputs is active, and a "no" when neither or both are. For decades, textbook models treated neurons as simple summing devices, and such a unit cannot solve XOR on its own. It famously takes at least two layers of artificial neurons to do it.
Because dCaAPs peak at a middle level of input and fade at higher levels, the researchers' modelling showed that a single human cortical neuron could classify inputs in an XOR-like way. In other words, one cell may be doing work that was assumed to require a small network.
If that holds up in living brains, it suggests that the computing power of the human cortex has been underestimated when each neuron is treated as a single switch. Each dendritic branch may act more like a small processing unit in its own right.
A second, separate study: signals that jump a gap
Popular reports at the time often blended this finding with a different piece of research. In 2019, biomedical engineers led by Dominique Durand at Case Western Reserve University reported in The Journal of Physiology that slow waves of activity in slices of mouse hippocampus could travel without normal synaptic transmission.
Most strikingly, the wave crossed a cut through the tissue when the two edges were placed close together. The team argued that only ephaptic coupling, communication through the weak electric fields neurons generate, could explain this. The waves travelled at about 0.1 metres per second, similar to the speed seen with synapses intact.
Clayton Dickson of the University of Alberta, who wrote a commentary on the study and was not involved in it, noted that the activity could be "not only modulated, but also eliminated or even regenerated by imposed electrical fields." It is a fascinating result, but it comes from mouse tissue, not human, and it is a different mechanism from the calcium spikes in Gidon and Larkum's work.
Does this change what we know about consciousness?
Not directly. Headlines in 2021 suggested the discovery could radically change our understanding of consciousness, but the study did not measure awareness, perception or experience. It measured voltage changes in fragments of tissue kept alive in a dish.
What it does is add to a growing view that dendrites are active, complex computers rather than passive wires. Some theories of consciousness, including ideas Larkum himself has explored, give special weight to how the outer layers of the cortex integrate information. Findings like dCaAPs feed into those debates, but they are one piece of a very large puzzle.
Several big questions remain open. It is not yet known how these signals behave across whole neurons in a living brain, how common they are, or whether they are unique to humans or also occur in other primates.
Will it lead to smarter computers?
Some coverage claimed the discovery could produce computers capable of human-like thought. That is a long stretch. Researchers in artificial intelligence have taken an interest in "dendritic" models, in which each artificial neuron has several sub-units, because they can be more efficient for some tasks. But today's AI systems already solve XOR trivially with simple networks, and building machines that think like people involves far more than copying one property of one cell type.
The more realistic payoff is scientific: better models of how real neurons compute, which may in turn help explain how brain circuits fail in conditions such as epilepsy.
Key takeaways
- Human layer 2/3 cortical neurons produce a calcium-based dendritic spike, the dCaAP, first described in Science in 2020.
- Its unusual response, strongest at moderate input, lets a single neuron perform an XOR-like operation in computer models.
- The tissue came from epilepsy and tumour surgery, and the result has not yet been shown in a living brain.
- The "signal crossing a cut" experiment is a separate 2019 mouse study on ephaptic coupling.
- Neither study explains consciousness, but both show that single neurons are more capable than older models assumed.
Sources: Neuroscience News / AAAS, 2 January 2020; ScienceAlert, Mike McRae; Case Western Reserve University, February 2019; Mind Matters, February 2019; Wikipedia: Dendrite.
COMMENTS