Groundbreaking study uncovers how our brain learns
Sophisticated synapse imaging used in NIH-funded project tracks changes within neurons as learning unfolds, offering new insights for brain-like AI systems
University of California - San Diego
image:
Neurons and their branch extensions known as dendrites are featured within a mouse’s cerebral cortex.
view moreCredit: Komiyama Lab, UC San Diego
How do we learn something new? How do tasks at a new job, lyrics to the latest hit song or directions to a friend’s house become encoded in our brains?
The broad answer is that our brains undergo adaptations to accommodate new information. In order to follow a new behavior or retain newly introduced information, the brain’s circuity undergoes change.
Such modifications are orchestrated across trillions of synapses — the connections between individual nerve cells, called neurons — where brain communication takes place. In an intricately coordinated process, new information causes certain synapses to get stronger with new data while others grow weaker. Neuroscientists who have closely studied these alterations, known as “synaptic plasticity,” have identified numerous molecular processes causing such plasticity. Yet an understanding of the “rules” selecting which synapses undergo this process remained unknown, a mystery that ultimately dictates how learned information is captured in the brain.
University of California San Diego neurobiologists William “Jake” Wright, Nathan Hedrick and Takaki Komiyama have now uncovered key details about this process. The main financial support for this multi-year study was provided by several National Institutes of Health research grants and a training grant.
As published April 17 in the journal Science, the researchers used a cutting-edge brain visualization methodology, including two-photon imaging, to zoom into the brain activity of mice and track the activities of synapses and neuron cells during learning activities. With the ability to see individual synapses like never before, the new images revealed that neurons don’t follow one set of rules during episodes of learning, as had been assumed under conventional thinking. Rather, the data revealed that individual neurons follow multiple rules, with synapses in different regions following different rules. These new findings stand to aid advancements in many areas, from brain and behavior disorders to artificial intelligence.
“When people talk about synaptic plasticity, it’s typically regarded as uniform within the brain,” said Wright, a postdoctoral scholar in the School of Biological Sciences and first author of the study. “Our research provides a clearer understanding of how synapses are being modified during learning, with potentially important health implications since many diseases in the brain involve some form of synaptic dysfunction.”
Neuroscientists have carefully studied how synapses only have access to their own “local” information, yet collectively they help shape broad new learned behaviors, a conundrum labeled as the “credit assignment problem.” The issue is analogous to individual ants that work on specific tasks without knowledge of the goals of the entire colony.
Finding that neurons follow multiple rules at once took the researchers by surprise. The cutting-edge methods used in the studied allowed them to visualize the inputs and outputs of changes in neurons as they were happening.
“This discovery fundamentally changes the way we understand how the brain solves the credit assignment problem, with the concept that individual neurons perform distinct computations in parallel in different subcellular compartments,” said study senior author Takaki Komiyama, a professor in the Departments of Neurobiology (School of Biological Sciences) and Neurosciences (School of Medicine), with appointments in the Halıcıoğlu Data Science Institute and Kavli Institute for Brain and Mind.
The new information offers promising insights for the future of artificial intelligence and the brain-like neural networks upon which they operate. Typically an entire neural network functions on a common set of plasticity rules, but this research infers possible new ways to design advanced AI systems using multiple rules across singular units.
For health and behavior, the findings could offer a new way to treat conditions including addiction, post-traumatic stress disorder and Alzheimer’s disease, as well as neurodevelopmental disorders such autism.
“This work is laying a potential foundation of trying to understand how the brain normally works to allow us to better understand what’s going wrong in these different diseases,” said Wright.
The new findings are now leading the researchers on a course to dig deeper to understand how neurons are able to utilize different rules at once and what benefits using multiple rules gives them.
As mice learned a new behavior, researchers closely tracked synaptic connections (depicted here as small protrusions) on the dendrites of neurons.
Credit
Komiyama Lab, UC San Diego
Journal
Science
Method of Research
Experimental study
Subject of Research
Animals
Article Title
Distinct synaptic plasticity rules operate across dendritic compartments in vivo during learning
Article Publication Date
18-Apr-2025
Surprise: Synapses on single neurons follow distinct rules during learning
Summary author: Walter Beckwith
Shedding light on how the brain fine-tunes its wiring during learning, a new study finds that different dendritic segments of a single neuron follow distinct rules. The findings challenge the idea that neurons follow a single learning strategy and offer a new perspective on how the brain learns and adapts behavior. The brain's remarkable ability to learn and adapt is rooted in its capacity to modify the connections within its neural circuits – a phenomenon known as synaptic plasticity, in which specific synapses are altered to reshape neural activity and support behavioral change. Neurons, unlike most other cell types, are characterized by their intricate, tree-like dendritic arbors, which extend from the cell body and serve as the primary site for receiving signals from other neurons via synaptic inputs. These dendrites are not uniform; instead, they are organized into distinct compartments with specialized anatomical and biophysical properties which likely influence how various patterns of neural activity trigger the biochemical processes that underlie synaptic plasticity. However, how the brain determines which synapses should be modified during learning and whether individual neurons apply the same plasticity rules uniformly across their structurally and functionally distinct dendritic compartments remains unknown.
To explore how synapses function and adapt during learning, William Wright and colleagues used advanced imaging to observe single synapses in the motor cortex of mice as the animals learned new motor skills. Wright et al. trained mice on a motor task known to drive synaptic plasticity in layer 2/3 motor cortex neurons, observing clear behavioral signs of learning over two weeks. Then, to investigate how individual synapses adapt during this process, Wright et al. used in vivo two-photon imaging with molecular sensors that simultaneously tracked synaptic input (via glutamate release) and neuronal output (via calcium activity). The authors discovered that learning-related patterns of neural activity drive synaptic plasticity differently across dendritic compartments. In apical dendrites, synapses were strengthened when they were coactive with nearby neighbors, suggesting that plasticity here is governed by local interactions between adjacent inputs. In contrast, plasticity in basal dendrites was linked to the neuron's overall output—strengthening or weakening depending on how synapse activity aligned with global action potential firing. Suppressing a neuron's activity selectively impaired plasticity in basal, but not apical, dendrites. In a related Perspective, Ayelén Groisman and Johannes Letzkus discuss the study and its findings in greater detail.
Journal
Science
Article Title
Distinct synaptic plasticity rules operate across dendritic compartments in vivo during learning
Article Publication Date
18-Apr-2025
Does your brain know you want to move before you know it yourself?
Neural activity in brain motor area largely coincides in time with the onset of the experience of intention
image:
Temporal binding between intention and action. Neural recording and experimental setup.
view moreCredit: Noel J-P, et al., 2025, PLOS Biology, CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
Researchers led by Jean-Paul Noel at the University of Minnesota, United States, have decoupled intentions, actions and their effects by manipulating the brain-machine interface that allows a person with otherwise paralyzed arms and legs to squeeze a ball when they want to. Published in the open-access journal PLOS Biology on April 17th, the study reveals temporal binding between intentions and actions, which makes actions seem to happen faster when they are intentional.
Separating intentions from actions was made possible because of a brain-machine interface. The participant was paralyzed with damage to their C4/C5 vertebrae and had 96 electrodes implanted in the hand region of their motor cortex. As they tried to squeeze a ball, a machine-learning algorithm looked at all the signals coming to the electrodes, and learned which activity pattern meant “squeeze” (close) and which meant “relax” (open). Once learned, the machine could send an electrical signal to the appropriate hand muscles, allowing the participant to squeeze the ball, which then delivered a sound. On average, the participant perceived the time from intention to action to be 71 ms, which was slightly faster than the real time duration.
The heart of the study involved systematically removing each part of the chain of events. By randomly stimulating the participant’s hand to squeeze the ball, the researchers were able to eliminate intentions. In this case, actions were judged to occur much later. On the other hand, when the participant tried to squeeze the ball, but actions were prevented by not stimulating the hand, intentions were perceived to happen much earlier if the sound was still played after the machine decoded the intention. Perceived timing did not change if the sound was not played. These results indicated a compressed temporal binding between intention and action. Recordings from the electrodes, which normally cannot be done in humans, showed that the motor cortex encodes these intensions.
The authors add, “The work builds on a long tradition trying to establish the temporal relationship between moving, the onset of the subjective experience of intending the movement, and the neural correlates of this intention. While there have been quite a number of non-invasive studies on this topic, only a single study prior to our was able to measure single neurons - the gold-standard in neuroscience - in humans while asking them to report when they first ‘felt the urge to move’.”
The authors continue, “This prior work (Fried et al., 2011) showed that certain brain areas (all in the frontal cortex) know about the intention to move up to a second prior to when we experience that intention. You can imagine this created quite the debate regarding whether humans have free will or not. Our study contributes to this debate by recording single neurons in the primary motor cortex. We show that firing of neurons in this area (the last cortical node before the spinal cord, which ultimately elicits movements) co-occur with the subjective experience of intending a movement.”
The authors conclude, “The work would have not been possible without expertise from a number of contributors, including neurosurgeons, neuroengineers and neuroscientists, among others.”
In your coverage, please use this URL to provide access to the freely available paper in PLOS Biology: https://plos.io/421hAZm
Citation: Noel J-P, Bockbrader M, Bertoni T, Colachis S, Solca M, Orepic P, et al. (2025) Neuronal responses in the human primary motor cortex coincide with the subjective onset of movement intention in brain–machine interface-mediated actions. PLoS Biol 23(4): e3003118. https://doi.org/10.1371/journal.pbio.3003118
Author countries: United States, Switzerland, United Kingdom
Funding: AS is supported by the Swiss National Science Foundation (grant PP00P3 163951/1; www.snf.ch), OB is supported by the Swiss National Science Foundation (www.snf.ch) and the Bertarelli Foundation (https://www.fondation-bertarelli.org/). MB was supported by the Craig H. Neilsen Foundation (Grant number: 651289; chnfoundation.org) and State of Ohio Research Incentive Third Frontier Fund (https://development.ohio.gov/business/third-frontier-and-technology). JPN is supported by NIH NINDS R00NS128075 (https://www.ninds.nih.gov/) and an Alfred Sloan Research Fellowship (https://sloan.org/). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.
Journal
PLOS Biology
Method of Research
Experimental study
Subject of Research
People
