The sooner autism treatment starts, the better the chance for success. But early intervention demands early identification of children with the disorder, and this presents a challenge.
“We know that autism reflects errors in brain development,” says Charles Nelson, professor of pediatrics at Harvard University and director of the Laboratories of Cognitive Neuroscience at Children’s Hospital, Boston.
The underlying process may begin soon after birth, or even prenatally, “but we can’t see signs of it until the child has a behavioral repertoire that allows it to express differences in things like social communication and repetitive motor acts.” This is why autism is rarely diagnosed before age 3.
Nelson is one of a group of researchers who are looking for biomarkers— manifestations of brain function gone awry—that are detectable long before symptoms appear. Working toward this goal, his Infant Sibling Project is comparing babies at high risk of autism [they have an older sibling already diagnosed with autistic spectrum disorder (ASD)], and infants from families without such a history.
His team may have found such a biomarker in the electrical impulses that arise from brain activity—electroencephalography, or EEG. The traditional uses of EEG involve pronounced, visible changes in brain waves associated with seizures and brain disease. But more advanced analytical techniques reveal that “there is a lot more information in EEG signals that we can’t see with our eyes,” says William Bosl, a neuroinformatics researchers and instructor in pediatrics at Harvard and Children’s Hospital.
In a study reported in the Feb. 22 issue of BMC Medicine (PDF), Bosl, Nelson, and colleagues recorded EEG signals from 46 high-risk infants and 33 controls, using a hairnet-like cap that covered the entire scalp with 64 electrodes. They conducted tests, when possible, in the children at ages 6, 9, 12, 18, and 24 months.
The researchers computed “modified multiscale entropy” of signals from each electrode—a measure of their variability, randomness, and complexity. They then applied a computerized method of statistical analysis, derived from artificial intelligence, to seek out meaningful patterns hidden in the huge masses of data.
With this approach, they could differentiate high-risk from control infants by their EEGs. The separation was most marked at 9 months, when it was possible to identify members of the two groups with 80 percent accuracy.
There was less difference after 12 months, and it is unclear why. Bosl suggests that the EEG patterns they observed represent an endophenotype—the expression of a genetic predisposition to autism that high-risk infants share. “But mental abilities are not entirely genetically determined: the brain is very plastic and influenced by environment and response to experience,” he says.
Early on, “the developmental process splinters the high risk group into several groups, which will go on to develop autism [one-fifth, according to earlier studies]; show some autistic characteristics; or develop perfectly normally.”
Bosl points out that the 9-12 month period is when milestones in brain development occur, “some of which are critically involved in behaviors that characterize autism: social awareness, the ability to read faces and to communicate.”
What might the EEG entropy patterns reveal about the nature of autism? “Developments in neuroscience suggest that autism is a connectivity disorder,” he says; there is evidence that EEG complexity corresponds to brain connectivity, and it may be that the observed differences reflect alterations in how brain areas communicate with one another.
Just how communication within the brain goes wrong is in itself the subject of considerable debate among autism researchers. Much of the research has used functional MRI scanning and suggests a generalized underconnectivity between regions of the autistic brain.
But Ralph-Axel Müller, professor of psychology and director of the Brain Development Imaging Laboratory at San Diego State University, reviewed 32 connectivity studies in the March online issue of Cerebral Cortex, and found a more nuanced picture. “Most of those studies have looked at specific regions within networks,” parts of the brain that work together to perform cognitive functions, such as Broca’s and Wernicke’s areas in language processing, “and here the connections are reduced,” he said.
“But when you look at the rest of the brain, you see the opposite: a diffuse overconnectivity,” he said. “The picture that emerges from the empirical literature, I would say, suggests less-distinct network organization in ASD.”
In the normally developing infant brain, Müller explains, connectivity is initially broad and undifferentiated, and functional networks gradually emerge through synaptic plasticity: “The process has a lot to do with [brain] activity; if neurons fire together [to process speech or recognize faces, for example], synapses between them will be strengthened. If they don’t, the synapses are lost.”
The process “is probably disrupted early on” in autism, Müller says, perhaps because axons grow too rapidly (anatomical studies have found increased white matter), or basic mechanisms that should shape synapses fail. But experience may be involved as well. “ASD children don’t interact with their environment normally, and because synaptic pruning and stabilization depend on such interactions, the disorder itself probably has an effect on emerging networks.”
EEG is a relative newcomer to connectivity research. Research published in the December 2010 Clinical Neurophysiology suggests a mix of under- and overconnectivity somewhat different from what Müller proposes.
The researchers measured EEG signals in corresponding visual cortex areas in the left and right hemisphere as children were exposed to flashing light. These areas are normally activated together, but they were poorly synchronized in autistic, compared with developmentally standard children. Within each separate cortical area, however, the signals were stronger in the autistic kids.
“I lean toward the theory—and an interpretation of this paper supports it— that in autism, the brain is overconnected within local areas, but lack the long-range connectivity needed for specialized networks,” says Joseph Isler, associate research scientist at Columbia University, lead author of the paper.
His work, Isler says, has the same “overarching goal” as the research at the Infant Sibling Project, “to find a biomarker that would have good predictive value, even at birth.” He sees EEG as a potentially ideal tool in this regard: it can be used easily even with very young infants, which imaging techniques like fMRI cannot, and is far less expensive.
“It was interesting and encouraging,” he says of the Bosl-Nelson paper. “I think it will be really, really fascinating to see their longitudinal results with this measure.”
The big question here is whether EEG patterns at 9 months or even earlier will predict which of the high-risk children will show signs of autism. For that, “we have to wait until the kids are 3 years old, when a definitive diagnosis can be made,” says Nelson.
When the data are reanalyzed, it may be that “the whole trajectory, from 6 to 24 months, will provide more information than [readings at] any single age,” Bosl says.
Detailed analysis at that point, he adds, may provide a closer look at what is happening in the autistic brain. “If the entropy pattern in one part of the brain at 9 months looks more like a normal child at 6 months, it may tell us there’s a delay in development in that region.”
On a pragmatic level, defining EEG patterns that characterize normal and autistic development might help researchers develop interventions and therapists apply them. “Watching the trajectory could be a way to measure whether treatment is making a difference,” Bosl says. “If it turns toward normal, something is working.”