New computational tool decodes the molecular rules of brain connectivity

A recent study has revealed that specific patterns of gene activity serve as a hidden map that guides the complex wiring of the entire brain. By using machine learning to analyze mouse brain data, researchers provided evidence that chemical gradients direct neurons to their correct target regions across the whole brain. The findings, published in the Proceedings of the National Academy of Sciences, offer new ways to understand how the brain develops and how developmental disorders might arise.

To function properly, the brain relies on an incredibly complex network of nerve connections. Neurons, the primary cells of the brain, communicate by sending out long, thread-like branches called axons. These axons must navigate across the brain to find and connect with specific target cells. The complete map of all these neural connections is known as the connectome.

Understanding exactly how these growing axons know where to go has been a major question in biology. In 1963, a scientist named Roger Sperry proposed the chemoaffinity theory. He suggested that neurons find their matching partners by following molecular concentration gradients. These gradients are essentially chemical signals that vary in strength across different areas of the brain, acting like a chemical GPS for growing nerve fibers.

The chemoaffinity theory was previously proven to work in simple sensory systems. For instance, in the visual system, specific chemical gradients guide nerve fibers from the eye to the visual processing centers in the brain. Yet, the sheer complexity of the entire brain made it difficult to test whether this same principle governed the larger, brain-wide network.

A team of scientists led by Jigen Koike from Hiroshima University and Naoki Honda from Nagoya University developed a new computational framework to solve this problem. They aimed to decode the hidden wiring rules of the brain by combining maps of gene activity with maps of neural connections.

Free daily newsletter

The researchers analyzed existing data from the Allen Mouse Brain Atlas, a comprehensive public database. This database provides a detailed map of both brain connections and gene activity in adult mice. The team focused on the long-range connections between 213 distinct brain regions. By filtering out the very short, local connections, they isolated a total of 2,213 major neural pathways to study.

For the genetic component, the researchers looked at the activity levels of 763 different genes across these 213 brain regions. Gene activity, or gene expression, refers to how much a specific gene is turned on or off in a cell. Because different areas of the brain express genes differently, they create unique chemical landscapes. These overlapping patterns of gene activity give each brain region a distinct molecular identity.

To find hidden relationships between the genetic data and the connection data, the team developed a machine learning tool. They named their method SPERRFY, which stands for Spatial Positional Encoding for Reconstructing Rules of axonal Fiber connectivity. The algorithm searched for matching patterns between the gene activity at a nerve fiber’s starting point and its final destination.

The machine learning algorithm successfully identified specific patterns of gene activity, or gradients, that predict which brain regions are likely to connect. The researchers then used these extracted patterns to build a simulated wiring map of the mouse brain. They operated under the assumption that brain regions with similar gradient values would be highly likely to form a connection.

When the researchers compared their simulated map to the actual biological connectome, the predictions were highly accurate. They measured this using a standard statistical performance score where zero means completely incorrect and one point zero means a perfect prediction. The gene-based model achieved a high score of 0.88, which suggests that the genetic patterns closely match the real wiring structure.

To confirm that the model was not simply predicting connections based on physical closeness, the scientists ran a second test. They tried to predict the brain connections using only the physical distance between the brain regions. This distance-based prediction scored much lower, at around 0.70. This drop in accuracy suggests that the genetic patterns provide unique biological instructions beyond simple spatial geography.

The scientists also wanted to ensure their algorithm was not just finding random patterns by chance. They randomized the original brain data to create fake, jumbled connection maps. When they ran their machine learning tool on this randomized data, the algorithm failed to find strong matching patterns. This failure on randomized data provides evidence that the biological results reflect genuine, meaningful wiring rules in the mouse brain.

By examining the extracted genetic gradients more closely, the researchers found that the brain’s wiring map seems to operate on a two-tiered system. Broad, sweeping patterns of gene activity tend to control the large-scale organization between different major brain areas. At the same time, more detailed and localized genetic patterns manage the specific, smaller connections within those distinct regions.

The algorithm also allowed the researchers to identify specific candidate genes whose activity patterns closely matched the predicted gradients. For example, the model highlighted genes like Ephb6 and Efnb2, which are already known to guide nerve growth in sensory systems. Finding these familiar genes in a brain-wide analysis suggests that the computational tool successfully captured real biological mechanisms.

Another gene highlighted by the model was Robo2, which is known to act as a guidance receptor that helps determine the paths of growing axons. The model also identified genes involved in synaptic transmission, which is the process neurons use to send chemical messages to one another. The presence of these specific genes provides evidence that the extracted gradients are biologically relevant.

While the findings offer new insights into brain wiring, the study does have some limitations. A primary limitation is its reliance on gene activity data from adult mice. The actual wiring of the brain primarily happens during early embryonic development. This means the adult gene activity patterns might only offer a partial or altered reflection of the original developmental signals that guided the axons.

Additionally, the neural connection data was simplified into a binary format for the analysis. This format simply records whether a connection exists or not, treating all connections equally. This approach leaves out detailed information about the strength, density, or volume of those neural connections, which could obscure more subtle wiring rules.

The study also relies on statistical correlations and does not prove a direct cause and effect relationship. Just because a gene’s activity pattern matches a connection pattern does not mean that specific gene caused the connection to form. Experimental testing in a laboratory setting will be necessary to confirm the exact roles of these candidate genes in brain wiring.

Future research tends to point toward applying this computational method to other species. Scientists could use this approach on available brain data from fruit flies, marmosets, and humans. Exploring different species could help determine if these genetic wiring rules are universally shared across the animal kingdom or if unique patterns exist.

Applying the machine learning tool to data collected from younger, developing brains could provide a more direct view of the biological wiring process. The researchers note that a better understanding of these basic connectivity rules could eventually help explain the origins of various developmental brain disorders. Such disorders often arise when the complex wiring map of the brain fails to form correctly.

The study, “A data-driven framework linking the connectome to spatial gene expression gradients inspired by chemoaffinity theory,” was authored by Jigen Koike, Ken Nakae, Riichiro Hira, Yuichiro Yada, and Naoki Honda.