October 13, 2017

Tracy Hampton, PhD
Article Information
JAMA. 2017;318(13):1211-1212. doi:10.1001/jama.2017.13706


Infants at high familial risk of autism spectrum disorder (ASD) do not typically exhibit symptoms in their first year of life, but new research indicates that magnetic resonance imaging (MRI) may reveal signs of the disorder during this presymptomatic period. The findings point to a noninvasive method to detect autism at its earliest stages, when interventions may provide the most benefits.

Prospective neuroimaging findings suggest increased cerebral cortical growth between 6 and 12 months of age may predict autism diagnosis at 24 months of age.

An estimated 1 in 5 infants who have an older sibling with ASD develop the condition, compared with approximately 1 in 100 in the general population. Because the defining features of ASD tend to emerge over the latter part of the first year and into the second year, a diagnosis is not typically made until 24 months of age and beyond.

Two prospective neuroimaging studies published earlier this year in Nature and Science Translational Medicinetrained machine learning algorithms using longitudinal MRI data from infants at high familial risk of ASD to predict autism diagnosis. The first study published in Nature compared behavioral changes and brain structural changes in 106 infants at high familial risk of ASD and 42 low-risk infants at 6, 12, and 24 months of age.

“Behaviors in the first year or even first year and a half are not good predictors of who will develop autism, but in our Nature study we found that by using structural MRI scans from 6 and 12 months we could predict with high accuracy [81% positive predictive value and 88% sensitivity] 8 out of 10 infants who would go on to meet criteria for autism at 24 months of age,” said senior author Joseph Piven, MD, a professor of psychiatry at the University of North Carolina School of Medicine and director of the Carolina Institute for Developmental Disabilities.

“The implication is that we may be able to intervene in these children before the symptoms appear and prior to the brain changes concurrently associated with autism at 24 months of age becoming consolidated.” There is general consensus in the field that early interventions are more effective than those initiated later, and “given that our success with interventions are modest at best, this new possibility holds great promise,” said Piven.

The structural MRI findings suggest a cascading series of events in the development of ASD in high-risk children: presymptomatic increased growth rate of cerebral cortical surface area particularly in sensory processing areas from 6 months to12 months of age, followed by increased growth rate of cortical volume between 12 months and 24 months of age, and emergence of autistic behaviors at 24 months of age.

“The social deficits of autism noted at 24 months of age are significantly linked to the increased growth of the cortical volume in the second year,” Piven noted. He also explained that the expansion of surface area in the first year of life is likely the result of a hyperproliferation of neural progenitor cells.

“This publication makes a significant contribution to our understanding of one potential neurodevelopmental pathway to the ASD phenotype that might, provided replication of course, be relatively common among multiplex families—a pathway characterized by brain overgrowth,” said Allison Jack, PhD, who was not involved with the research and is an assistant research professor at George Washington University’s Autism and Neurodevelopmental Disorders Institute.

A more recent study by Piven and his colleagues published in Science Translational Medicine used functional connectivity MRI (fcMRI), which analyzes brain function rather than structure, to identify individual differences in brain region synchronization. The team scanned the brains of 59 infants with high familial risk of ASD and collected data from 26 335 pairs of functional connections between 230 different brain regions. Of the 59 infants, 11 were later diagnosed with ASD at 24 months of age.

The researchers trained a machine learning algorithm to look for differences in functional neuroimaging results at 6 months of age that could differentiate the 2 groups of participants—those later diagnosed with autism vs those that were not—and predict future diagnoses. This method correctly identified 9 of the 11 infants diagnosed with ASD without any false-positives and all of the infants who did not develop autism, indicating that a single fcMRI scan at 6 months of age could predict a diagnosis at 24 months of age with high accuracy (100% positive predictive value and 81.8% sensitivity).

“If replicated, the findings in high-familial-risk children should have a clear impact on clinical practice. I believe that most parents with a newborn and an older child with autism would be very willing to do a single scan—during natural sleep—in their baby to determine their risk for autism,” said Piven. “Once this work moves into the realm of treatment studies, we will be able to begin to sort out what treatments work best presymptomatically in this population.”

In the future, high-risk children may be identified by factors other than family history. For example, investigators are looking into genetic risk scores obtained from DNA samples that might indicate an individual’s risk of later disease. “Such a polygenic risk marker would identify a subset of children who would then go on to a phase 2 MRI scan in their first year of life and if positive might be recommended for presymptomatic intervention for later autism,” said Piven.

For now, using neuroimaging to predict ASD would not be practical in infants in the general population, according to Piven and others. Jack stressed the need for studies designed to better understand and predict the developmental trajectories that lead to ASD regardless of familial autism history. “It's something of a catch-22: it’s much more cost-effective and practicable to study infant sibs in a prospective design, but we are in much greater need of diagnostic predictors for children who don't already have a known risk factor.”