A decades-old challenge in neuroscience has been solved by harnessing artificial intelligence (AI) to identify the electrical signatures of different types of brain cells for the first time, as part of a study in mice led by researchers from UCL.
Brains are made up of many different types of neurons (nerve cells in the brain), each of which are thought to play different roles in processing information. Scientists have long been able to use electrodes to record the activity of neurons by detecting the electrical 'spikes' that they generate while performing brain functions.
Although recording spikes has proved invaluable for monitoring the activity of individual neurons deep in the brain, until now the method has been 'blind' to the type of neuron being recorded - making it impossible to identify how different neurons contribute to the brain's overall operation.
In a new study, published in Cell, the research team have overcome this problem by identifying the distinct 'electrical signatures' of different neuron types in the mouse brain, using brief pulses of blue light to trigger spikes in specific cell types (a method called optogenetics).
They created a library of the different electrical signatures for each type of neuron, which then allowed them to train an AI algorithm that can automatically recognise five different types of neurons with 95% accuracy without further need for genetic tools. The algorithm was also validated on brain recording data from monkeys.
The researchers say they have overcome a major hurdle in being able to use the technology to study neurological conditions such as epilepsy, but that there is still "a long way" to go before it can be used in practical applications.
Dr Maxime Beau, co-first author of the study from the UCL Wolfson Institute for Biomedical Research, said: "For decades, neuroscientists have struggled with the fundamental problem of reliably identifying the many different types of neurons that are simultaneously active during behaviour. Our approach now enables us to identify neuron types with over 95% accuracy in mice and in monkeys.
"This advance will enable researchers to record brain circuits as they perform complex behaviours such as movement. Like logic gates on a computer chip, neurons in the brain are elementary computing units that come in several types. Our method provides a tool to identify many of the brain's logic gates in action at the same time. Before, it could only be done one at a time, and at much greater cost."
The authors say the fact that the algorithm can be applied across different species gives it huge potential for being expanded to other animals and, eventually, to humans.
In the short term, the new technique means that, instead of requiring complex genetic engineering to study the brain, researchers could use any normal animal to study what different neurons do and how they interact with one another to generate behaviour.
One of the ultimate aims is to be able to study neurological and neuropsychiatric disorders such as epilepsy, autism and dementia, many of which are thought to involve changes to the way different cell types in the brain interact.
Professor Beverley Clark a senior author of the study from UCL Wolfson Institute for Biomedical Research, said: "Just as many different instruments in an orchestra contribute to the sound of a symphony, the brain relies on many distinct neuron types to create the complex behaviour that humans and other animals exhibit. Our work is analogous to learning the sound that each instrument makes and then teaching an algorithm to recognise the contribution of each of them to a symphony.
"Being able to observe this 'neural symphony' of the brain in action has been a fundamental challenge in neuroscience for over 100 years, and we now have a method for reliably doing this.
"Although the technology is a long way from being able to be used to study neurological conditions such as epilepsy, we've now overcome a major hurdle to reaching that goal. In fact, some recordings of living human brain activity have already been recorded in patients during surgery, and our technique could be used to study those recordings to better understand how our brains work, first in health and then in disease."
Improved understanding of how our brains work could pave the way for some ground-breaking advances in medical science, some of which are already on the horizon.
Human brain-to-computer interfaces, or neural implants, are one such possibility. Ongoing research at the UCSF Weill Institute for Neurosciences, for example, has enabled a paralysed man to control a robotic arm using a neural implant for a record seven months. Like the current study, this work was also informed by studying the electrical patterns in the brains of animals and using AI to automatically recognise these patterns.
The authors say the new technique to differentiate neuron types could help to improve neural implants by more accurately recording which types of cells are involved in particular actions, so that the implant can more easily recognise specific signals and generate the appropriate response.
Key to this technology is understanding how our brains work when they're healthy, so that any damage can be compensated for. If a person had a stroke and part of their brain was damaged, for example, you would need to understand how that bit worked before you could consider designing an implant to replicate that functionality.
Professor Michael Häusser, a senior author of the study from UCL Division of Medicine and The University of Hong Kong, said: "This project came to life thanks to the convergence of three critical innovations: using molecular biology to successfully 'tag' different neuron types using light, developments in silicon probe recording technology, and of course the fast-paced improvements in deep learning.
"Crucially, the synergy in our team was absolutely instrumental. The partner labs at UCL, Baylor, Duke and Bar Ilan University have all contributed critical pieces to the puzzle. Just like the brain, the whole is larger than the sum of its parts."
The database gathered by the team is freely available and the algorithm is open source, meaning scientists from across the world can use these resources for neurological research.
This research was funded by funding from Wellcome, National Institutes of Health (NIH), European Research Council (ERC), and the European Union's Horizon 2020 research and innovation programme.
Links
- Professor Beverley Clark's academic profile
- Professor Michael Häusser's academic profile
- UCL Wolfson Institute for Biomedical Research
- UCL Division of Medicine
- UCL Faculty of Medical Sciences
Image
peterschreiber.media on iStock.
