A team led by Harvard Medical School researchers at Brigham and Women’s Hospital, in collaboration with investigators at Massachusetts General Hospital and Beth Israel Deaconess Medical Center, has developed a powerful computational tool for understanding brain health and disease, providing an enhanced way of characterizing the activity of the brain during sleep.
The researchers devised a new method that extracts tens of thousands of electrical events from the brain waves of a sleeping person. Information from these waveforms is then used to create a picture of brain activity that seems to act like a fingerprint — unique for each person and consistent from one night to the next.
They then used their approach to identify new potential biomarkers in the brain activity of people with schizophrenia. Their findings are published in the journal Sleep.
First and senior author on the study, Michael Prerau, HMS assistant professor of medicine at Brigham and Women’s, said that “this work expands the way we can look at brain activity during sleep. By moving beyond traditional notions that break up the complex continuum of sleep into specific categories and waveform classes, we can reveal new types of signals and dynamics that may be important for understanding brain health and disease.”
Sleep spindles
Scientists typically study brain activity during sleep using the electroencephalogram, or EEG, which measures brain waves at the scalp.
Sleep EEG was first studied in the mid 1930s, by looking at the traces of brain waves drawn on a paper tape by a machine. Many important features of sleep are still based on what people almost a century ago could easily observe in the complex waveform traces.
Thought-provoking content
In-depth stories on science and medicine
Even the latest machine learning and signal processing algorithms for detecting sleep waveforms are judged against their ability to recreate human observation.
In this study, the researchers asked, What can we learn if we expand our notion of sleep brain waves beyond what was historically easy to identify by eye?
One particularly important set of sleep brain wave events are called sleep spindles. These spindles are short oscillation waveforms, usually lasting less than two seconds, that are linked to our ability to convert short-term memories to long-term memories. Changes in spindle activity have been linked with numerous disorders such as schizophrenia, autism, and Alzheimer’s disease, as well as with natural aging.
In this study, rather than looking for spindle activity according to the historical definition, the team developed a new approach to automatically extract tens of thousands of short spindlelike waveform events from EEG data collected throughout an entire night.
Next, instead of looking at the waveforms in terms of fixed sleep stages — wake, REM, and non-REM sleep stages 1 through 3 — as standard sleep studies do, the researchers characterized the full continuum of gradual changes that occur in the brain during sleep.
Using all these data, the team created graphical representations called slow oscillation power and phase histograms, which provide a powerful visualization of the activity of all the waveforms as a function of continuous sleep depth and synchronized activity in the cortex.
“This further demonstrates the richness of the information that traditional, manual scoring leaves on the table,” said co-author Shaun Purcell, HMS associate professor of psychiatry at Brigham and Women’s.
Fingerprint patterns
When the team looked at data from two nights of sleep recordings from a group of healthy participants, the patterns observed appeared to be almost like fingerprints — highly specific to each person with strong consistency across nights. These results suggest new ways in which brain activity differs from person to person, even within groups of healthy people selected as control groups.
The researchers then compared the activity between the healthy subjects and a population of people with schizophrenia, a disorder that reduces spindle activity. Using their approach, the team not only saw the known differences in participants with schizophrenia, but found differences in other spindlelike waveforms occurring at other frequencies in the brain.
This suggests new potential EEG biomarkers of schizophrenia that could be useful in better understanding the mechanisms of the disorder and in the development of targeted treatments.
“This approach is really exciting,” said co-author Dara Manoach, HMS professor of psychology in the Department of Psychiatry at Mass General.
“We look forward to seeing how we can enhance our understanding, not only of schizophrenia, but also of other neurodevelopmental disorders characterized by differences in sleep, such as autism and pediatric epilepsy.”
“We are just starting to understand the scope of neurodiversity that exists within the general population,” said Prerau. “If we can more accurately characterize the individual differences observed in both neurological health and disease, we can work towards improved diagnostics and treatments.”
Co-authors were Patrick Stokes, Preetish Rath, Tom Possidente, Mingjian He, and Robert Stickgold. More information and an open-source code toolbox can be found on Prerau’s laboratory website.
This work was supported by the National Institute of Neurological Disorders and Stroke (R01NS096177) and the National Institute of Aging (R01AG054081 [MJP]).
