Brain oscillations compress odor representations as signals pass through olfactory networks
The percepts that such complex blends evoke in us are, however, astonishingly singular: ground coffee smells like coffee, not like a hopeless mess of hundreds of ingredients; Gio or Allure also have unique signatures (often associated with other memories). This contrast between a physical object's complexity and the uniqueness of how we perceive it is the expression of what brains do best: "bind" features together into highly recognizable patterns.
This is as true of smell as it is for the other senses: a person can immediately recognize his mother's face or voice. How this useful and effortless compression of information is accomplished by the brain is one of the deep mysteries of neuroscience. In addition, understanding the mechanisms of compression would help in the design of computerized pattern recognizers (e.g., face recognition devices), a very difficult task with many important applications.
This issue is the subject of new research from neurobiologist Gilles Laurent and his team in the California Institute of Technology's computation and neural systems program. In a paper appearing in the July 19 issue of the journal Science, the Laurent team reports that the complicated wiring in grasshoppers between the antennal lobe (the insect analog of the olfactory bulb in humans) and the mushroom body (the insect analog of the olfactory cortex) is arranged and functions in such a way that highly detailed information in the former is bound or compressed for future memory use in the latter.
To better explain the details of their discovery, it's probably best to first explain the organs of smell in grasshoppers. The first area associated with smell are the receptor cells that are the front line of cells coming into contact with the chemical elements of a smell. Information from the receptor neurons converge in the antenna lobe, which, in grasshoppers, comprises about 1,000 individual neurons. The signal then goes to the mushroom body, which comprises about 50,000 neurons.
By intricately wiring glass, silicon, and platinum wire electrode arrays into the brains of hundreds of grasshoppers to record activity from their neurons, then exposing the insects to a variety of smells, the Laurent team has demonstrated some of the fine details of this wiring and its consequences for odor encoding.
When a specific odor is detected by a grasshopper, the antennal lobe neurons, wired to the peripheral detector array, start a complicated "dance" that engages about half of its 1,000 neurons. Each individual odor evokes a different dance or spatio-temporal pattern that involves partially overlapping subsets of neurons activated at varying times. Hence, determining from these patterns the odor's identity is a very difficult task; it requires that an observer decode the details of the dance, identify the correlations between the activities of all the neurons, and put all this back together into a coherent whole. Said differently, the informative value of any antennal lobe neuron in isolation is close to zero: valuable information comes only from deciphering the message carried by the population.
Population decoding is precisely what is done by the downstream neurons (called Kenyon cells, in the mushroom body). Those neurons, using a complicated combination of wiring, biophysical properties, brain oscillations, and loops of inhibition, manage to compress the information carried by many antennal lobe neurons into highly specific and sparse signals. Thus, individual Kenyon cells are silent most of the time and produce a signal only in response to very specific odors. The signals from these neurons, when given out, are thus highly informative, Laurent says.
"If you observe Kenyon cell No. 2,976 and see that it produced one single pulse, you can be pretty confident that the animal has just detected a certain odor mixture and not another," he explains. Each Kenyon cell thus has a very limited, but highly specific repertoire of "preferred stimuli."
At the same time, this compression eliminates much of the information about the individual chemical elements that make up an odor. "Knowing that Kenyon cell No. 2,976 fired may tell me that the (grasshopper) just smelled a cherry blend, but it tells me nothing about the chemical composition of that smell."
This may explain why these individual elements cannot be perceived; the encoding and decoding of an odor as a whole (cherry or Gio) is done at a cost: detail is lost. The advantage, however, is that the storage and retrieval of this odor's representation has become very simple, fast, and manageable: Each odor, however complex, is now represented by very few, highly specific neurons. Because the mushroom body has many neurons (and our olfactory cortex has even more), a huge number of such memories can be stored.
"There are many reasons to think that odor perception may work in similar ways in vertebrates, including humans," Laurent says, explaining that the antennal lobe in insects, including flies, is very similar to the mammalian olfactory bulb, except that it possesses many fewer cells; and that the mushroom body is likewise similar to the human olfactory cortex.
"In the case of humans as well as animals, the brain is not doing analytical chemistry by pulling out individual components," he says. "Instead, you have a very good memory for odors, however complex, even though you lose information about details. After all, throwing away information is one of the most important things that brains do, but it must be done carefully.
"In olfaction as well as in vision and the other senses, the brain must represent and memorize a huge number of complicated patterns. One should expect that evolution has found an optimal way of solving this task. Our work provides the beginnings of a solution, although whether it applies to other senses remains to be seen."
In addition to Laurent, the other authors of the study, all members of Caltech's Division of Biology, are Javier Perez-Orive, Ofer Mazor, Glenn C. Turner, Stijn Cassenaer, and Rachel I. Wilson.
The paper is available online at http://www.sciencemag.org.
Contact: Robert Tindol (626) 395-3631
