Summary: Researchers used artificial intelligence to visualize and track changes in synapse strength in living animals. Synapses—the tiny junctions where neurons communicate—are central to learning, memory and age-related brain changes.
By applying machine learning to noisy in vivo microscopy, the team enhanced image quality enough to detect and follow individual synapses across days and weeks. This approach provides a new way to study how synaptic connections change with learning, injury, aging and disease.
Key Facts:
- The method applies AI to monitor changes in synaptic strength—the number and distribution of glutamate receptors—within live animal brains.
- Researchers trained a machine learning model using high-quality ex vivo images paired with low-quality in vivo images to restore and super-resolve in vivo data.
- The technique enables tracking of thousands of individual synapses over multiday experiments, revealing how experiences alter synaptic receptor levels.
Source: Johns Hopkins Medicine
Johns Hopkins researchers have developed an AI-based imaging pipeline that restores low-quality in vivo images and enables nanoscale tracking of synaptic plasticity in living animals.
Published in Nature Methods, the work offers a practical path to observe how synapses change with experience and how those changes might go awry in aging or disease.
“To understand an orchestra you must observe individual musicians over time. This method lets us do that for synapses in the living brain,” says Dwight Bergles, Ph.D., Diana Sylvestre and Charles Homcy Professor in the Solomon H. Snyder Department of Neuroscience at Johns Hopkins School of Medicine.
The study team includes Bergles and collaborators Adam Charles, Ph.D., M.E., Jeremias Sulam, Ph.D., and Richard Huganir, Ph.D. The interdisciplinary group spans neuroscience and biomedical engineering and is affiliated with the Kavli Neuroscience Discovery Institute at Johns Hopkins.
Synapses are submicron structures where neurons exchange chemical messages. Changes in the number or composition of glutamate receptors at these junctions are widely believed to underlie learning and memory. However, synapses are densely packed and extremely small, making it difficult to resolve individual receptor clusters in living brains using standard microscopy.
To detect which synapses change during a particular experience, the researchers combined biological labeling with computational image restoration. They used transgenic mice whose glutamate receptors fluoresce, so receptor quantity correlates with fluorescence intensity and thus synaptic strength.
In vivo imaging of intact brains produced noisy, low-resolution images in which individual receptor clusters were hard to distinguish. To overcome this, the team collected paired datasets: high-resolution images of brain slices (ex vivo) captured with super-resolution microscopy, and corresponding low-resolution images resembling those obtained in live animals.
Using these paired images, they trained a deep-learning image-restoration algorithm to translate low-quality in vivo images into enhanced, super-resolved reconstructions. This cross-modality supervised approach leverages the strengths of ex vivo super-resolution while preserving the dynamics captured in in vivo experiments.
With the trained model, the researchers substantially improved in vivo image quality and were able to detect and track thousands of individual synapses across multiday time courses. To test the system, they repeatedly imaged the same cortical region in mice over several weeks and introduced a brief, novel sensory experience—five minutes in an unfamiliar chamber with new sights, smells and textures.
After the experience, follow-up imaging every other day revealed a mix of increases and decreases in glutamate receptor fluorescence across synapses, indicating that some connections strengthened while others weakened. Overall, exposure to the novel environment produced a bias toward strengthening in many synapses, demonstrating the method’s ability to link behavior to nanoscale synaptic changes.
The work depended on collaboration among experts in molecular neuroscience, microscopy and machine learning—teams that do not always work closely together but are encouraged to collaborate at the Kavli Neuroscience Discovery Institute.
The researchers are extending this approach to study synaptic changes in animal models of Alzheimer’s disease and other conditions. They anticipate the technique will provide new insights into synaptic dysfunction in disease, injury and aging.
“We are excited to see how other groups will adapt and apply these tools,” says Jeremias Sulam, Ph.D.
Funding: Experiments were conducted by Yu Kang Xu (Ph.D. student and Kavli fellow), Austin Graves, Ph.D. (assistant research professor), and Gabrielle Coste (neuroscience Ph.D. student). Funding was provided by the National Institutes of Health (RO1 RF1MH121539).
About this AI and neuroscience research news
Author: Vanessa Wasta
Source: Johns Hopkins Medicine
Contact: Vanessa Wasta – Johns Hopkins Medicine
Image credit: Neuroscience News
Original Research (open access): “Cross-modality supervised image restoration enables nanoscale tracking of synaptic plasticity in living mice” by Dwight Bergles et al., Nature Methods.
Abstract
Cross-modality supervised image restoration enables nanoscale tracking of synaptic plasticity in living mice
Learning is thought to involve changes in glutamate receptors at synapses, submicron structures that mediate communication between neurons. Because synapses are small and densely packed, resolving receptor dynamics in vivo is challenging, limiting direct links between receptor changes and behavior.
The authors developed a combined computational and biological pipeline that uses paired ex vivo super-resolution and in vivo imaging data to train a deep-learning image-restoration algorithm. Applied to fluorescently labeled glutamate receptors in transgenic mice, this algorithm super-resolves in vivo images and enables nanoscale tracking of behavior-associated synaptic plasticity.
This method demonstrates how image enhancement trained on ex vivo data can improve the spatial resolution of in vivo imaging and permit large-scale tracking of synaptic changes over time.