Machine-learning model spots autism-linked variants that scramble splicing

Rare genetic variants associated with autism may disrupt the regulation of tiny DNA segments called microexons, according to a new study.

Researchers developed a machine-learning model to predict how genetic variants affect microexon splicing—the process by which these short sequences are included in or excluded from RNA transcripts. By applying the model to genetic data from thousands of families, the team found that rare variants predicted to disrupt microexon splicing are more common in autistic people than in their non-autistic siblings.

The findings, published last month in Nature, point to “an underappreciated source of [autism]-linked genetic variation,” says study investigator Benjamin Blencowe, professor of molecular genetics at the University of Toronto. 

Microexons, which are typically 3 to 27 nucleotides long, can influence how proteins interact and function, particularly in the brain. Microexon splicing is altered in the brains of some autistic people, according to previous work by Blencowe’s team. But it was unclear whether these alterations stemmed from genetic variants, including those in surrounding regions that do not encode proteins but contain important regulatory instructions.

“The major advance here is moving from observations of altered microexon splicing in autism toward identifying genetic variants that may contribute to their splicing or retention,” says Lilia Iakoucheva, professor of psychiatry at the University of California, San Diego, who was not involved in the work. The model could make it easier to investigate genetic changes whose effects have long been difficult to understand, she adds.

B

lencowe and his colleagues analyzed existing RNA-sequencing data from thousands of human and mouse tissue samples, including individual brain cells. They identified 1,514 previously unknown microexons in humans and another 1,115 in mice. Some showed distinct splicing patterns across neuronal types or stages of brain development. Only about 12 percent of the newly identified human microexons whose splicing differed between the brain and other tissues had clear counterparts in mice.

A subset of the microexons whose inclusion in RNA differed between the brain and other tissues appeared to trigger nonsense-mediated decay, a process that destroys certain RNA transcripts before they can produce proteins, suggesting that these gene snippets regulate protein abundance. 

Using the expanded collection of microexons identified through RNA sequencing, the team trained a machine-learning model to examine the corresponding DNA sequences, including regions surrounding those microexons, and to identify features that influence their splicing. The model identified regulatory sequences surrounding microexons, including sites recognized by proteins that promote or suppress their incorporation into RNA transcripts.

When the researchers used the model to analyze genetic data from 7,391 families across three large autism studies, they identified 72 rare variants predicted to disrupt microexon splicing. These variants occurred in 2.5 percent of autistic participants but not in their non-autistic siblings. About 80 percent of the participants carrying these variants had no previously identified genetic explanation for their autism.

Many of these rare variants occurred in genes involved in communication between neurons, including several already linked to autism. Testing three of the variants in human neurons, one at a time, confirmed that they each altered microexon splicing, reducing their inclusion in RNA transcripts.

“We predict that these variants will often affect protein-protein interactions that are important for the assembly and function of synapses,” Blencowe says. The team’s machine-learning model could also help researchers pinpoint noncoding variants involved in microexon splicing in other brain-related conditions, he adds. 

B

eyond its potential applications, the study sheds light on another way in which genetic variants outside protein-coding sequences could contribute to autism, says Hyejung Won, associate professor of genetics at the University of North Carolina at Chapel Hill, who was not involved in the research. But identifying these variants is only the first step, Won says. “Further studies are needed to determine how individual variants alter microexon inclusion or exclusion and how these splicing changes affect neuronal function.”

It’s also unclear whether microexons are particularly susceptible to autism-associated genetic changes, or whether they are affected in the same way as longer exons that undergo alternative splicing, Iakoucheva says. Experiments testing the effects of thousands of genetic variants could help determine whether microexons play a distinctive role in autism, she adds. 

The findings underscore how much remains to be understood about the consequences of altered splicing, from changes in individual proteins to effects on neural circuits and behavior, says Christopher Park, a research scientist at the Flatiron Institute, who was not involved in the work. (The Flatiron Institute is funded by the Simons Foundation, The Transmitter’s parent organization.) “This work just kind of  hammers in the importance of splicing in autism and, unfortunately, how complex it can be.”