Brain imaging method detects genetic markers of autism with over 90% accuracy

A new study published in Science Advances introduces a powerful brain imaging technique that can detect autism-linked genetic variations with up to 95% accuracy. This approach, developed by researchers from multiple universities, including Johns Hopkins and Carnegie Mellon, analyzes structural brain images to identify specific genetic patterns associated with autism, potentially offering a way to detect the condition earlier and more objectively than current behavior-based methods.

Autism spectrum disorder is a neurodevelopmental condition characterized by differences in social communication, interaction, and the presence of restricted interests or repetitive behaviors. It is understood to result from a complex interplay of genetic predispositions and environmental influences. Currently, autism is diagnosed based on observing an individual’s behavior, a process that can take time and may not occur until certain developmental milestones are missed.

However, research increasingly points to a strong genetic component in autism. Understanding this genetic basis offers a potential pathway to better comprehend the condition’s origins, potentially leading to more personalized approaches and earlier support. This study explored a “genetics-first” avenue, focusing on specific genetic alterations known as copy number variations. These variations involve segments of a person’s genetic code being deleted or duplicated. Certain copy number variations are known to substantially increase the likelihood of developing autism.

The researchers aimed to see if unique patterns in brain structure, visible through imaging, could be directly linked to these specific genetic variations, providing a potential biological marker, sometimes called an endophenotype, that connects genes to observable traits.

To investigate this possibility, the research team, involving experts from Carnegie Mellon University, the University of California San Francisco, and the Johns Hopkins University School of Medicine, utilized a specialized computer modeling technique they developed called transport-based morphometry. This method stands apart from many other image analysis techniques because its mathematical underpinnings are based on modeling the movement and distribution of mass, akin to how substances move within biological tissues. It essentially quantifies the shape and form (morphometry) of the brain based on these modelled transport processes.

Free daily newsletter

The researchers applied this technique to analyze brain scans from a group of 206 individuals sourced from the Simons Variation in Individuals Project. This cohort included 48 individuals with a deletion in a specific genetic region known as 16p11.2, 40 individuals with a duplication in the same region (both variations are strongly associated with increased autism risk), and 118 control participants without these specific genetic changes.

The control group was carefully selected to match the other participants in terms of age, sex, handedness, and non-verbal intelligence scores, and they were screened to exclude individuals with related neurological conditions or family histories of autism. High-resolution structural brain images (T1-weighted magnetic resonance imaging scans) were obtained for all participants using standardized procedures across different imaging sites.

The images underwent preprocessing steps to isolate brain tissue (gray matter and white matter), adjust for overall brain size differences, and normalize the data before the transport-based morphometry analysis was performed separately on gray matter and white matter distributions. The system was trained using machine learning principles to distinguish the brain structure patterns characteristic of the deletion group, the duplication group, and the control group.

The analysis revealed distinct patterns in brain structure associated with the 16p11.2 copy number variations. The transport-based morphometry system was highly effective at identifying which genetic group an individual belonged to based solely on their brain scan. When analyzing white matter structure, the system achieved an average accuracy of 94.6% in correctly classifying individuals into the deletion, duplication, or control group on previously unseen test data. Analyzing gray matter structure yielded an average accuracy of 88.5%. These results significantly outperformed classification attempts using only basic information like age, gender, or overall brain volume.

A key capability of the transport-based morphometry technique is that it is generative, meaning it allowed the researchers not just to classify the scans but also to visualize the specific brain structure differences driving the classifications. The analysis indicated that the 16p11.2 variations were associated with widespread, or diffuse, changes across the brain, rather than being confined to just one or two small areas.

There was a dose-dependent relationship observed: individuals with the 16p11.2 deletion tended to have larger overall brain volumes and relatively more gray matter tissue compared to controls, while those with the duplication tended to have smaller brain volumes and relatively less gray matter tissue. The visualization also revealed specific regional patterns.

For instance, areas involved in language processing, emotional regulation, visuospatial skills, and integrating information from multiple senses showed distinct patterns of relative tissue expansion or contraction depending on whether an individual had the deletion or duplication. Often, the effect was reciprocal, meaning a region might show relative expansion in the deletion group and relative contraction in the duplication group compared to controls. Some differences were also noted between the left and right sides of the brain.

Importantly, the researchers explored associations between these identified brain structure patterns and participants’ behavioral or cognitive characteristics. They found a strong association between one specific brain pattern (identified along what the researchers termed discriminant direction 1) and the presence of articulation disorders – difficulties producing speech sounds correctly.

This pattern was particularly prominent in individuals with the 16p11.2 deletion. Another distinct brain pattern (associated with discriminant direction 2) showed a significant association with participants’ intelligence quotient scores, explaining roughly 17-20% of the variation in full-scale, verbal, and nonverbal intelligence quotient measures across the groups. These findings suggest that the structural brain differences linked to the 16p11.2 copy number variations are related to observable functional outcomes.

The researchers acknowledge some limitations to their study. The participants were recruited through clinical genetics centers and patient networks, which might mean the sample doesn’t represent the full spectrum of individuals with these genetic variations, potentially missing those with milder or different presentations (ascertainment bias). The study focused on one specific genetic region, 16p11.2, and didn’t explore interactions with other genes.

While the study included individuals from childhood through adulthood, assuming relative stability of these brain patterns, further research focusing on early development is warranted. Also, while associations between brain structure patterns and behavioral measures like articulation or intelligence quotient were found, this type of study cannot establish a cause-and-effect relationship.

Future research could apply this transport-based morphometry approach to investigate other genetic variations linked to autism and related neurodevelopmental conditions. Larger studies involving more diverse populations and prospective studies tracking individuals over time are needed to validate these findings and explore their potential clinical utility for early detection, prognosis, or monitoring responses to interventions. Such work could significantly advance a genetics-first approach to understanding and supporting individuals with autism.

The study, “Discovering the gene-brain- behavior link in autism via generative machine learning,” was authored by Shinjini Kundu, Haris Sair, Elliott H. Sherr, Pratik Mukherjee, and Gustavo K. Rohde.