 |
Computational Neurobiology
Summary: Terrence Sejnowski's goal is to discover the principles linking brain mechanisms and behavior. His laboratory uses both experimental and modeling techniques to study the biophysical properties of synapses and neurons and the population dynamics of large networks of neurons.
Sleep
While we sleep we are immobile and unaware of our surroundings. Sleep takes up a significant fraction of our lives, and although we all need to sleep to maintain a normal daytime function, why we sleep remains a mystery. The average length of sleep varies greatly across species and generally decreases with the size of the animal and with age. Sleep deprivation leads to a reduction in efficiency with which we can detect signals and make decisions. Sleep deficits accumulate and lead eventually to sleep rebound, suggesting that the integration that occurs during sleep extends over days. Although sleep is a period of rest for the body, the brain is highly active, and we may be able to learn something about the function of sleep by studying the brains of sleeping animals.
Sleep EEG
Recordings of weak electrical signals from the scalp, called an electroencephalogram (EEG), give a picture of changing brain activity that is related to behavioral state. There is more power in the lower part of the EEG spectrum, and the scalp preferentially transmits lower frequency components. The awake EEG at rest contains strong alpha activity in the 10-Hz band. Alertness reduces the alpha power and increases power in the gamma band between 30 Hz and 80 Hz. When the brain falls asleep, the power in the lowest frequency bands increases, with the highest power in the delta band at 2–4 Hz occurring during slow-wave sleep (SWS). Periodically during the night, the brain shifts from SWS to a state called rapid eye movement (REM) sleep, in which the oscillations have a higher amplitude and resemble, by eye, the awake EEG. At the beginning of a night of sleep, the periods of SWS last for more than an hour, gradually shortening toward the morning, as the periods of REM sleep become longer. This general pattern is found in all mammals, though the relative amount of REM and SWS varies greatly between species.
Intermediate sleep (IS), which occurs during transitions between SWS and REM, is characterized by 10- to 14-Hz sleep spindles and takes up to 40 percent of the total sleep time. Sleep spindles, which are highly spatially and temporally coherent, last for around 1 second and recur every 5–10 seconds. They arise in the thalamus and recruit cortical circuits, which in turn project back to the thalamus, forming a feedback loop. The IS EEG is also characterized by K complexes, high-amplitude events that often are associated with sleep spindles. The neural mechanisms underlying the generation of sleep spindles in the thalamus have been identified and involve reciprocal interactions between the excitatory relay neurons and the inhibitory neurons in the reticular nucleus of the thalamus. Biophysical computational models of this circuit in my laboratory have led to predictions for the mechanisms underlying the generation and termination of spindles, which have been corroborated in recordings from brain slices by David McCormick (Yale University) and in vivo by Mircea Steriade (Laval University, Quebec).
Sleep Scoring
The different sleep stages are defined by characteristic features in the pen chart records. Because sleep disorders are often accompanied by aberrant sleep stages, it is common in sleep clinics to record the EEG together with signals from eye movements and muscle tone to score the amount of time spent in each stage during the night. The scoring is done manually in 20- to 30-second segments, using a standard reference introduced by Rechtschaffen and Kales in 1968, which classifies human sleep in three broad categories: SWS, IS, and REM. IS and SWS are further subdivided, respectively, into stages 1 and 2 and stages 3 and 4. This characterization of sleep and waking stages has been highly influential in guiding sleep research. Because of subjectivity in applying the rules, and transitions occurring on a timescale of a few seconds, the concordance between sleep experts is typically around 75 percent. Techniques such as supervised classifiers have not improved on scoring by humans, which remains the gold standard in the sleep field.
Automatic Staging of Human Sleep EEG
We have developed a new approach to automatically score sleep stages that does not depend on the rules used by human experts; rather, it is unsupervised in the sense that it finds features and creates clustered states independently of the human expert. The new method first normalizes the power in each frequency band across the night to look for information in the relative changes in power. Thus, although the power in the higher frequencies is much less than that in the lower frequencies, the higher frequencies contain information that critically distinguishes the different stages of sleep. Clustering in the space determined by independent component analysis and principal component analysis reveals four distinct stages that can be identified with awake, REM, SWS, and IS, with a time resolution of 1 second, using only a single EEG channel. The agreement with human experts is 85 percent.
The new method puts the sleep stages on a firmer footing and provides insights into the features that distinguish them. First, we find no evidence for sleep stage 1, which has been questioned on other grounds, nor is there evidence for distinguishing between stages 3 and 4 of SWS. Second, low frequencies can cleanly separate awake and REM sleep, without using eye movements or muscle tone. Thus, although the awake and REM EEG look the same to the eye on a chart record, our statistical analysis revealed strong differences. Third, SWS, which is generally considered to be the most stable of the sleep stages, is actually the most fragmented, with highly variable high-frequency events interspersed with the prominent high-amplitude delta oscillations. Fourth, the IS alternates between two states, one that is highly fragmented, like SWS, and another that is more stable, like REM. We are still analyzing the differences between these two substates, which may be a reflection of the slow oscillations (<1 Hz) recently discovered by Mircea Steriade using intracellular recordings. In particular, the sleep spindles, a hallmark of stage-2 sleep, appear to occur at the transition between the fragmented and stable periods during IS.
Avian Sleep Architecture
Birds have periods of inactivity that resemble mammalian sleep. Many of the features of the mammalian sleep architecture, however, are thought to be generated by the neocortex, which birds lack. In collaboration with Daniel Margoliash (University of Chicago), we applied the automated techniques developed for humans to score the sleep stages of zebra finches and found that not only do birds have the same sleep states as mammals—REM, IS, and SWS—but they also have the same architecture, with more SWS early in the early part of the night and more REM as the morning approaches. Is this a case of parallel evolution, or do birds have the same circuits that mammalian species have, but with a different morphological organization?
Enhanced Optics for Multiphoton Microscopy
Multiphoton microscopy allows fluorescent signals to be imaged deep in the cortex. We have developed a new class of hybrid objectives that has achieved nearly diffraction-limited performance of the imaging portion throughout a 200-μm field of view with a two- to fourfold improvement in epifluorescence collection efficiency over traditional objective designs, significantly improving the signal-to-noise ratio. We plan to use this new objective to record action potentials from multiple cells in vivo.
Last updated: May 1, 2007
|
|
|
|
