The stepping stones of memory are more fluid than once thought
New research from Carnegie Mellon University reveals that dendritic spines—the tiny structures that help brain cells communicate and form memories—don’t fit neatly into a few categories. Instead, they display a continuous range of shapes, offering new insight into learning, development, and neurological disease.
If memories were like a path through the forest, dendritic spines would be the stepping stones. Every thought, memory, and skill we develop travels across these tiny structures that extend from our neurons to receive information. Their size and shape influence how well our neurons communicate, which makes them essential for our development. For decades, neuroscientists classified dendritic spines into a handful of categories such as “mushroom,” “thin,” and “stubby.” New research from the labs of Jessica Zhang and Phil LeDuc, professors in the Department of Mechanical Engineering suggests that the stepping stones of our memory are actually better understood as a continuum.
“We saw that dendritic spines are constantly changing in response to neural activity, and there are structures that exist beyond the traditional categories,” said Uma Sharma, Ph.D. candidate at Carnegie Mellon University. Because abnormalities in spine shape and density have been linked to conditions including Alzheimer’s, schizophrenia, down syndrome, and even chronic stress, better classifying these structures could enable clinicians to flag early indications for neurological disease.
The team developed a computational framework that combines dimensionality reduction, probabilistic clustering, and biological interpretation to model dendritic spine morphology as a continuum. This enables researchers to study how neural connections form, change, and potentially deteriorate from disease in a more nuanced way.
Our goal is to provide researchers with a principled, data-driven way to study biological structures that don’t fit neatly into a mold.
Jessica Zhang, Professor, Mechanical Engineering
The decision-based framework evaluates different machine learning and data visualization approaches before identifying the best representation of the spine morphology. The research team analyzed dozens of quantitative features extracted from microscopy images, including measurements of the shape, contour, and image intensity to create more comprehensive descriptions of the spine structures than what previously existed.
Rather than assuming the spines belong in one of the predefined groups, the framework identifies patterns in the data that allows the spines to show characteristics of multiple groups simultaneously. This revealed not only that the spines exist on a continuum but that there are additional subgroups within the existing classifications. This suggests that researchers who rely solely on the preexisting labels may be missing important biological information.
Beyond improving the study of dendritic spines, this framework could help researchers better understand how neural connections change throughout development and disease progression.
“Our goal is to provide researchers with a principled, data-driven way to study biological structures that don’t fit neatly into a mold,” said Zhang. “Because many biological systems also exhibit continuous variation, our approach could extend beyond neuroscience.”