When an oscilloscope's audio monitor starts to screech rhythmically in a neurophysiology lab, its waves hint at one of the most puzzling patterns in biology. Sometimes multiple neurons will simply fire synchronously, multiplying the strength of a signal. But what has intrigued researchers since the 1920s, when Hans Berger first started describing EEG patterns, is that often these synchronous neurons produce oscillatory rhythms, detectable as waves of distinct frequencies. "Synchronous oscillations are conserved among widely different species, which suggests that they are important in neural processing," says Ole Paulsen of Oxford University, who will be chairing a symposium on Network Oscillations at this month's Society for Neuroscience meeting. "What role they play, we still do not know."
It could just be simple mechanics. "Every system which has opposing forces, such as excitation and inhibition, almost inevitably will generate oscillations," explains György Buzsáki of Rutgers University, who is writing a book titled
But according to Buzsáki and many others, synchronous oscillations (SOs) could be an important extra dimension of neural coding. "You have to look at something like macroeconomics, the laws of dynamics of very large numbers, and then you get a very different picture," says Walter J. Freeman of University of California, Berkley.
CATCHING THE WAVE
In the 1960s, neuroscientists infatuated with the results of then-new single-unit recording techniques developed a prejudice against recording from neural assemblies. Horace Barlow famously compared it to circling over a football stadium in a blimp and recording the crowd; you would get only meaningless noise. But in the late 1980s investigators began to rethink that idea, especially about neural synchrony. If you hear one side of a stadium cheer all at once, while the other side groans, that does tell you something.
Those who believe in a functional value for SOs would ascribe them a role in neural communication. "SOs do a certain job, call it a computation, by enabling the traffic of some messages from neural population A to neural population B," says Gilles Laurent of California Institute of Technology.*
There are a few notable holdouts. "I certainly am not yet convinced," writes William Newsome of Stanford in an E-mail. "In my lab, we have a very difficult time even detecting oscillatory activity in the cortex, to say nothing of viewing it as a functionally important feature."
The challenge is to develop an experimental assay for something that could be nothing more than an artifact in electrophysiologic data. "Ascribing a function to a feature of the data is much harder than ascribing a function to an ion channel or a receptor," explains Maurice Chacron, an instructor in a course on neural systems and behavior at the Marine Biology Laboratory in Woodshole, Mass., "You can not manipulate an SO directly; you can only manipulate it indirectly by blocking the ion channels or receptors that contribute to generating that SO."
SOs ON THE NOSE
Laurent's lab has performed these manipulations to support a functional role. Terry Sejnowski of the Salk Institute, says, "You have to show [that SOs] will have some impact on the next relay station and ultimately on behavior. The only experiment I know in that regard is Stopfer '97." In 1997, Mark Stopfer, then a postdoc with Laurent and now at the National Institutes of Health, used picrotoxin (a GABA antagonist) to disrupt SOs in the olfactory system of honeybees.1 It affected their ability to make fine odor discriminations: They could distinguish citrus from chocolate, but not lemon from lime.
READING THE CROWD:
© 2005 Nature Publishing Group/Debbie Maizels
Using several recording sites separated in three-dimensional space allows for triangulation of neurons in volume, such as with this four-wire tetrode. Spike amplitudes of neurons within a 50 μm radius are large enough for separation by current clustering methods. Although the signal decreases rapidly with distance, electrodes can 'hear' pyramidal cells as far away as 140 μm accounting for more than 1000 neurons. Ideal recording electrodes would have a large number of recording sites, but smaller volumes such that they don't damage neurons. Micro-Electro-Mechanical System (MEMS)-based recording devices have allowed for multi-shank probes that can record further separated neurons and recordings across various cortical layers. (From G. Buzsáki, Nat Neurosci, 7:446–51, 2005.)
Laurent has expanded his original interpretation to include a role for memory. "I think discrimination is, de facto, a consequence of memory," he explains. "Synchronization causes sparseness, which in turn generates clean, easily separated memory templates to be stored, making later fine discrimination possible."
The work from Laurent's lab is often praised as a technical tour de force, but many still disagree with it. Donald Wilson of the University of Oklahoma says, "You have to be cautious in interpreting those experiments." With his graduate student Max Fletcher, Wilson found that 7-day-old rats who have very few GABAergic interneurons don't produce SOs, and yet they had no problems making odor discriminations as fine as one carbon in a series of ethyl esters. New work,2 says Wilson, shows that "at low concentrations, the 7-day-olds are actually better at discriminating odors than the older ones who have developed interneurons. At the same concentration, we see oscillations at one age and don't detect them at the other age."
Despite his results, Wilson says SOs may still serve a purpose, albeit one closer to expectation than fine-odor discrimination. "Oscillations in a vertebrate olfactory bulb are less about 'this is lemon or lime,' and more about, 'this is what I'm expecting to smell' in this situation," according to Wilson. And he doesn't discount that there may be undetectable SOs in the young rats.
In another twist to the story, Leslie Kay of the University of Chicago in collaboration with Zoltan Nusser, now of the Hungarian Academy of Sciences, flipped the Stopfer '97 experiment by using GABA(A) receptor β3 subunit deficient mice (β3-/-), which produced enhanced oscillations.3 "These mice were better than normal at discriminating closely related odorants," says Kay. Since such an ability could be just as much a detriment to survival as a benefit, Kay asks, "Is there circuitry in the intact mammalian brain to modulate synchrony dependent on the task demands?" She is collaborating with computational modeler Nancy Kopell of Boston University to find an answer.
"Leslie Kay has done some of the most insightful analysis of how the behavioral state affects what oscillations you get," says Alan Gelperin, a pioneer in invertebrate oscillations, "She's shown that the γ band (40–80 Hz) is not a unitary band; it's two quite distinct kinds of oscillation that occur in two distinct kinds of behavior in the mouse."
Some researchers are convinced that SOs have a role in human thought. Michael Kahana of the University of Pennsylvania is well past the maybe phase. "Based on the data that has been collected in humans, there's no question that oscillations are relevant to human cognition," he says. "It is extraordinarily difficult to imagine a trivial reason for why one observes the diversity of oscillatory patterns and their modulation by behavioral or cognitive states in humans or monkeys."
The difficulty, he says, is that all the simple hypotheses for what they may do in humans are consistent with only a small part of the data and have not provided a unified theoretical framework. "The kinds of experiments one would like to do, to prove causality, are difficult to justify in humans," he says. "There are no knockout humans."
Buzsáki argues that oscillations are an "emergent property of interactive neurons," and the underlying assumption of pursuing knockouts needs to be questioned. "It's a hopeless quest to ask for an experiment that will remove the oscillator and leave everything intact," he says, "because emergent properties do not have receptors that can be targeted; only the constituent neurons do."
Kahana's own work in humans is done in conjunction with neurosurgeons who work on patients with drug-resistant epilepsy. As part of the preoperative diagnostic procedure, the neurosurgeons implant electrodes in the patients' brains. With the patients' permission, Kahana has conducted experiments that correlate SOs with memory and cognition, by having the patient's navigate a taxi within a virtual environment.4
Kahana and others warn against an early idea that SOs might solve the "binding problem," by tying together different attributes into one percept. He says it is premature to correlate a particular function or region with a particular frequency, even in the most rigorous, scientific sense. "There isn't a simple frequencies chart," he says. "There is no clear, one-to-one mapping between behavioral states and oscillations and between brain regions and oscillations."
Nevertheless, several papers have notably linked γ oscillations and functional magnetic resonance imaging (fMRI)-measured hemodynamic activity in sensory systems, which could prove to skeptics that fMRI actually does relate to a neural event. Or, the findings might suggest the question of whether a different method of fMRI could measure other frequencies.5
We know a great deal about which neurons in which areas respond to which stimuli, says Sejnowski; what we do not yet know is how the brain interprets and uses that information. That may be where SOs play their part, he suggests. "SOs are telling us something, something that we don't really get yet about how the information is organized – some more complex spatiotemporal pattern for communication."
*The author of this article worked for Gilles Laurent in 2002.