Thoughts flow in one direction instead of in loops

Charité study in Science decodes efficient wiring of the human cerebral cortex

Asia 728x90

 Photo Credit: Rekonstruktion eines Nervenzellnetzwerks in der menschlichen Hirnrinde © Charité | Sabine Grosser

Berlin, 18.04, 2024

Contrary to previous assumptions, nerve cells in the human cerebral cortex are wired differently to those in mice. This is the result of a study by Charité – Universitätsmedizin Berlin, which has now been published in the journal Science*. According to the study, the neurones in humans communicate in one direction, whereas the signals in mice often flow in loops. This makes information processing in humans more powerful and efficient. The findings could contribute to the further development of artificial neural networks.

The cerebral cortex, one of the most important structures for human intelligence, is less than five millimetres thick. Twenty billion nerve cells process countless sensory perceptions here, in the outermost layer of the brain. We plan our actions in the cerebral cortex; this is where our consciousness is located. How do the nerve cells manage to process this complex information? This depends largely on how they are interconnected.

More complex cerebral cortex – different information processing

“Our current understanding of the neuronal architecture in the cerebral cortex is largely based on knowledge gained from animal models such as the mouse,” explains Prof Jörg Geiger, Director of the Institute of Neurophysiology at Charité. “In them, the neighbouring nerve cells often communicate with each other in a two-way dialogue, with one neuron sending a signal to another and the latter sending a signal back. As a result, the information often flows in loops.”

Photo: Device for multipatch analysis of the activity of up to ten nerve cells. A human tissue sample in the centre of the glass pipettes © Charité | Yangfan Peng

The human cerebral cortex is significantly larger and more complex than that of the mouse. Nevertheless, it was previously assumed – partly due to a lack of data – that it functions according to the same wiring principles. A Charité research team led by Jörg Geiger has now been able to prove that this is not the case using particularly rare sample material and state-of-the-art technology.

A sophisticated method for eavesdropping on neuronal communication

For the study, the researchers analysed brain tissue from 23 people who had undergone neurosurgery at Charité due to epilepsy. During the procedure, brain tissue had been removed in order to gain access to the underlying diseased structures. The patients had consented to the use of this tissue for research.

In order to observe the signal flows between neighbouring neurons in the outermost layer of the human cerebral cortex, the team developed an improved version of the so-called multipatch technique. It allowed the researchers to listen to the communication of up to ten nerve cells simultaneously (details in “About the methodology”). This enabled them to carry out the number of measurements required to create a network map in the short time until the cells stop their activity outside the body. In total, they analysed the communication paths of almost 1,170 nerve cells with around 7,200 possible connections.

Straight ahead instead of in loops

The result: only a small fraction of the neurons conducted two-way dialogues. “In humans, the information instead flows primarily in one direction, rarely returning to the starting point directly or via loops,” explains Dr Yangfan Peng, first author of the publication. The scientist worked on the study at the Institute of Neurophysiology and now works at the Department of Neurology and the Neuroscience Research Centre at Charité. The scientists were able to prove that this forward-directed signal flow has advantages for data processing using a computer simulation that they created based on the principles of human network architecture.

The researchers gave the artificial neural network a typical machine learning task: to recognise audio recordings of numerical words and assign them the correct number. It turned out that the model modelled on a human was more often correct in this speech recognition task than a model modelled on a mouse. It was also more efficient: Converted, the same performance would require 380 nerve cells in the mouse, but only 150 nerve cells in humans.

Economic role model for AI?

“The directed network architecture in humans is more efficient and saves resources because more independent nerve cells can handle different tasks at the same time,” explains Yangfan Peng. “The network can therefore store more information. It is still unclear whether our results collected in the outermost layer of the cerebral cortex apply to the entire cortex and how well they can explain the unique cognitive abilities of humans.”

In the past, AI developers have been inspired by biological models when constructing artificial neural networks, but they have also optimised the algorithms independently of biology. “Many artificial neural networks already utilise some form of forward structure because it delivers better results for some tasks,” says Jörg Geiger. “It is fascinating to see that the human brain also exhibits related wiring principles. The insights we have now gained into the particularly resource-efficient information processing in the human cerebral cortex could now contribute to the refinement of AI networks.”

*Peng Y et al. Directed and acyclic synaptic connectivity in the human layer 2-3 cortical microcircuit. Science 2024 Apr 18. doi: 10.1126/science.adg8828

About the study

The study was the result of close co-operation between the basic science and clinical departments of the Charité. Under the direction of the Institute of Neurophysiology, the following departments were involved: the Department of Neurosurgery, the Department of Neurology with Experimental Neurology, the Institute of Integrative Neuroanatomy, the Institute of Neuropathology, the Neuroscience Research Centre and the NeuroCure Cluster of Excellence, supported by the Department of Neurosurgery at the Protestant Hospital Bethel and the Institute of Neuroinformatics at ETH Zurich.

About the methodology

In the surgical treatment of treatment-refractory epilepsy, it is often medically necessary to remove brain tissue. The prerequisite for analysing this valuable tissue in the recently published study was the explicit consent of the patients, for which the research group would like to express its gratitude. The authors used the so-called patch-clamp method to analyse synaptic communication between the neurons. This involves attaching a very fine glass pipette to a single nerve cell under a microscope in order to measure or stimulate its electrical activity. In the study, a further development of the technique was used in which several of these glass pipettes simultaneously record the activity and connectivity of ten nerve cells (multipatch method). In order to position the glass pipettes precisely, the device is equipped with robotic arms that enable movements in the nanometre range. The measurement is very demanding and time-consuming; by using two devices in parallel, the team was able to analyse several hundred connections between the nerve cells per tissue. The brain tissue can survive in an artificial nutrient solution outside the body for up to two days before it ceases to be active.