Four different subtypes of autism were identified in brain studies

People with autism spectrum disorders can be classified into four different subtypes based on their brain activity and behavior, according to a study by Weill Cornell Medicine researchers.

The study, published March 9 in Natural Neuroscience, utilized machine learning to analyze newly available neuroimaging data from 299 people with autism and 907 neurotypical people. They found patterns of brain connections associated with behavioral traits in people with autism, such as verbal ability, social influence, and repetitive or stereotyped behavior. They confirmed that the four autism subgroups could also be replicated in separate datasets and demonstrated that differences in regional gene expression and protein-protein interactions explain differences in brain and behavior.

“Like many neuropsychiatric diagnoses, individuals with autism spectrum disorder experience various types of difficulties in social interaction, communication, and repetitive behavior. Scientists believe there may be many types of autism spectrum disorder that may require different treatments, but there is no consensus on how to define them,” said co-senior author Dr. Conor Liston, a professor of psychiatry and neuroscience at the Feil Family Brain and Mind Research Institute at Weill Cornell Medicine. “Our work highlights new approaches to discovering subtypes of autism that may one day lead to new approaches to diagnosis and treatment.”

A previous study published by Dr. Liston and colleagues in Nature Medicine in 2017 used similar machine learning methods to identify four biologically distinct subtypes of depression, and subsequent work has shown that the subgroups respond differently to various depression therapies.

“If you put people with depression in the right group, you can give them the best therapy,” said lead author Dr. Amanda Buch, neuroscience postdoctoral fellow in psychiatry at Weill Cornell Medicine.

Building on that success, the team set out to determine whether similar subgroups exist among individuals with autism, and whether different gene pathways underlie it. He explained that autism is a highly heritable condition linked to hundreds of genes that have diverse presentations and limited therapeutic options. To investigate this, Dr. Buch pioneered new analyzes to integrate neuroimaging data with gene expression and proteomics data, introducing them to the lab and enabling testing and developing hypotheses about how risk variants interact in autism subgroups.

“One of the barriers to developing a therapy for autism is that the diagnostic criteria are broad, and thus apply to large and phenotypically diverse groups of people with different underlying biological mechanisms,” said Dr. Buch. “To personalize therapy for individuals with autism, it is important to understand and target this biodiversity. It’s difficult to identify optimal therapy when everyone is treated the same, when they are each unique.”

To date, there are not sufficiently large collections of functional magnetic resonance imaging data from people with autism to conduct large-scale machine learning studies, said Dr. Buch. But the massive datasets that Dr. Adriana Di Martino, research director of the Center for Autism at the Child Mind Institute, as well as other colleagues across the country, provided the big data sets needed for this study.

“New machine learning methods that can address thousands of genes, differences in brain activity, and a wide variety of behaviors make this research possible,” said senior co-author Dr. Logan Grosenick, assistant professor of neuroscience in psychiatry at Weill Cornell Medicine, who pioneered machine learning techniques used for biological subtyping in the study of autism and depression.

The advance allowed the team to identify four clinically distinct groups of people with autism. Two groups have above average verbal intelligence. One group also had severe deficiencies in social communication but less repetitive behavior, while the other had more repetitive behaviors and fewer social distractions. The link between the parts of the brain that process visual information and help the brain identify incoming information is most salient to hyperactivity in the subgroup with more social impairments. This same connection was weak in the group with more repetitive behavior.

“It is interesting at the brain circuit level that there are similar brain networks involved in these two subtypes, but connections within these same networks are not prevalent in opposite directions,” said Dr. Buch, who completed his doctorate from the Weill Cornell Graduate School of Medical Sciences in Dr. Liston and now works in Dr. Grosnick.

The other two groups had severe social impairment and repetitive behavior but had verbal abilities at opposite ends of the spectrum. Despite some behavioral similarities, the researchers found very different brain connection patterns in these two subgroups.

The team analyzed the expression of genes that explain the atypical brain connections that exist in each subgroup to better understand what causes the differences and found many genes previously associated with autism. They also analyzed network interactions between proteins associated with atypical brain connections, and looked for proteins that might serve as hubs. Oxytocin, a protein previously associated with positive social interaction, is a link protein in a subgroup of individuals with more social impairment but relatively limited repetitive behavior. Studies have looked at using intranasal oxytocin as a therapy for people with autism with mixed results, says Dr. Buch. He said it would be interesting to test whether oxytocin therapy was more effective in this subgroup.

“You can get a treatment that works in a subgroup of people with autism, but that benefit gets lost in larger trials because you don’t pay attention to the subgroup,” says Dr. Grosnick.

The team confirmed their results in a second human data set, finding the same four subgroups. As the final verification of the team’s results, Dr. Buch performed an unbiased text-mining analysis he developed from the biomedical literature that showed other studies had independently linked autism-associated genes with the same behavioral traits associated with subgroups.

The team will next study this subgroup and potential subgroup-targeted treatments in mice. Collaborations with several other research teams that have large human data sets are also under way. The team is also working to further refine their machine learning techniques.

“We are trying to make our machine learning more cluster aware,” said Dr. Grosnick.

Meanwhile, dr. Buch says they have received encouraging feedback from individuals with autism about their work. A neurologist with autism talks to Dr. Buch after the presentation and said his diagnosis was confusing because his autism was so different from others but the data helped explain his experience.

“Being diagnosed with an autism subtype can really help him,” says Dr. Buch.