Most people diagnosed with autism never learn of a genetic cause. Known variants explain roughly 15 percent of cases, leaving the large majority without an answer. A study in Nature now points to an overlooked part of the genome that could fill some of that gap: the stretches of DNA that control microexons.
Microexons are minute pieces of genes, spanning between three and 27 bases. Through alternative splicing, cells decide whether to include them in the messenger RNA used to build proteins, producing slightly different protein versions. Earlier work from Benjamin Blencowe’s lab in Toronto had shown that these snippets are often misregulated in autistic brains, and that mice unable to splice them display autism-like behaviour.
The new work, led by researchers in London, Toronto and New York, began with a huge search. Guillermo Parada used his software tool MicroExonator to comb through more than 100 terabytes of public RNA data and found over 1,500 microexons that had not been described before. Close to 300 of them seem to be used mainly by nerve tissue.
Next the team trained a machine learning model to learn which sequences next to a microexon decide whether it is spliced in. Strikingly, the features the model rated as important were the same ones that have barely changed across vertebrate evolution, a typical sign of biological importance.
The researchers then ran the model over complete genome sequences from several thousand families, comparing autistic people with brothers and sisters who are not autistic. Uncommon mutations stood out as the ones most likely to disrupt splicing in nerve cells. Most turned up in people who had no previously known genetic explanation. The authors stress these are predictions that need lab confirmation, and they hope to use the tool for schizophrenia and epilepsy as well.
