What differs in the brains of young children as the behaviors that characterize autism emerge? Researchers have sought to define that difference using all available tools, including functional imaging, blood-based markers, eye-tracking and scalp electrical recordings (EEG). Past discoveries have yielded new clues, but no single test has demonstrated the power to accurately indicate which infants might be at higher risk of developing autism. However, a new report published this week in BMC Medicine could change how autism risk is assessed in the future.
A team of Boston researchers have found a reliable marker of autism risk in a simple and non-invasive EEG test. Although many autism researchers measure EEG—some even from infants—the new study, partly funded by Autism Speaks, has a unique twist. Instead of thinking about EEG signals in the traditional way, the investigators have embraced subtle relationships in the wiggles of EEG waveforms. By capturing those relationships, the research team identified a signature that reflects the structure of neural connections in the brain.
“We’ve long assumed that there is much more signal in the EEG than we take for granted. Conventional approaches lack the power of these more advanced machine learning tools in detecting useful signals from noise,” says Charles Nelson, Ph.D. of Harvard and Children’s Hospital in Boston. Dr. Nelson is the senior author of the study and a member of Autism Speaks’ Scientific Advisory Board. In conjunction with long-time collaborator and Boston University Professor, Helen Tager-Flusberg, Ph.D., the researchers sought the skills of someone who could analyze the complex EEG signal. William Bosl, Ph.D., of the Informatics Program in the Children’s Hospital of Boston, brings the perspective of a physicist to this task, using a suite of tools designed to find subtle relationships among the EEG waveforms and creating a framework for identifying hidden patterns in those signals.
In the paper, the team draws links between the amount of complexity observed in the EEG signal and the self-similar patterns of connectivity observed in the brain. Self-similarity is a favorite pattern for mother nature. Just as the branching of a tree has a pattern that repeats from the trunk and large branches to the finer branches and extensions to leaves, one can think of the branching of neurons in the same way.
This pattern is, in fact, essential to the way the brain communicates. Each neuron is connected with relatively few others, given the billions of neurons in the brain. To make communication efficient in such a sparsely connected network, the principle of self-similarity must apply. Neurons communicate with neighboring neurons more frequently then they do with neurons in a distant brain region—this is typical. However, in the brains of individuals with autism, the bias toward local communication is even greater, and long-range communication is less than is expected. These patterns of communication between neurons create electrical signatures which can be measured using EEG.
Throughout development, many neural connections change. New synapses are formed and others are pruned as a child develops and experiences new things. Atypical connectivity may result when these normal processes fail to occur as they should. Experiments that use EEG could help us identify when development is starting to go off the normal course. Previous studies have shown that the brains of children with autism tend to have less synchronized activity between different EEG sensors than typically-developing children. This pattern would be expected from having too many neighboring neurons chatting and less communication with distant neurons.
The researchers compared EEG from infants aged 6 to 24 months from two groups: one group of children had an older sibling with autism and were therefore at greater risk themselves and a second group of children had no affected siblings. To ensure compliance of each little subject, a research assistant amused the infant by a blowing bubbles while an EEG cap—which resembles a space-age hairnet—was quickly situated on the baby’s head. The data for this analysis was measured during a baseline period, where the children were quietly observing their surroundings and not otherwise engaged in a specific task.
The period of the most dramatic EEG differences between risk groups—about 9 months—corresponds to a time of major milestones that form the foundation for later social and communication skills. Around this time babies develop the ability to perceive another person’s intention to act and they lose the ability to detect differences in speech sounds from languages that are not their own. In another study of high-risk infants, Sally Ozonoff, Ph.D. and colleagues found no differences in social and orienting behaviors before six months, however significant differences emerged in the following six months.
Nearly 80% of the 9 month old children were correctly categorized as high or low risk on the basis of a measure of disorder or randomness in the EEG signal compared across different scalp . For reasons that are as yet unclear, the signal varied more for girls. At 9 months, if only boys were considered, the percentage of correctly categorized infants rises to greater than 90%. The categorization degrades at later ages for all children, and in fact is best for baby girls at 6 months.
This study demonstrates the use of a simple, non-invasive measurement tool and an important time window for identifying children who may be at risk of developing ASD. “The next and most critical next step is to see if our brain measures can actually predict which of our children will develop autism and which will not,” says Dr. Nelson. That information will come soon as the children who participated in this study come to an age where traditional diagnosis is highly reliable. However, as Dr. Tager-Flusberg notes, “We are still a long way from understanding the clinical significance of these brain signatures. More information is need to find a marker that predicts ASD outcomes later in life.” The team’s longer term research goals include determining how this risk marker in combination with other neural, behavioral, genetic and other risk markers may eventually lead to earlier diagnoses of ASD.