Walk by a bakery, and you’ll smell fresh-baked bread. But would you smell it, if you’d never learned what bread was? “Not necessarily,” says Thomas A. Cleland, Psychology.
“The very concept of having odors that you recognize and identify, even being able to pick them out of the messy olfactory world out there, absolutely depends on learning,” says Cleland. “I argue that learning is fundamental to even being able to smell at all.”
It seems counterintuitive—if the molecules exist, shouldn’t we be able to detect them? “But your brain lies to you for your benefit,” Cleland says. “It doesn’t give you a literal, accurate representation of the world but an improved one for your purposes, which your evolutionary history and brain have decided for you.”
In the visual system, for example, mechanisms within the retina sharpen images, allowing us to better distinguish edges and objects. “Your visual sense will also embed these priors—ways of thinking or expectations—into visual perception,” Cleland says. In other words, your visual system is making things easier for you, by making reality seem clearer than it actually is.
Cleland is proving the same thing to be true in the olfactory system—that these mechanistic enhancements and learned expectations are essential to identifying odors. For a sensory system that was once seen as rather simple, many big, new, and complex questions follow: How and where does this learning happen? What are the olfactory mechanisms of learning and memory? Can those mechanisms tell us anything about learning and memory elsewhere in the brain?
How the Olfactory System Constructs Odors
The basic mechanism of olfaction may seem straightforward. Molecules bind to receptors, and those receptors send signals to the olfactory bulb, which sends signals to other regions of the brain. But the molecules that make up an odor bind to many different receptors, sometimes weakly or strongly, sometimes changing the configuration of the receptor to block other molecules from binding. With about 400 receptors in humans, and over 1,000 in mice and dogs, the signatures for certain odors become complex and high-dimensional.
Even more importantly, Cleland says, these signatures are almost never going to be in isolation from the continuous flow of molecules in the environment. Therefore, the patterns of receptor binding are extremely messy—patterns disrupt and get in the way of each other and do not resemble the characteristic signature of any particular odor. “I think this is the most important problem in olfaction,” says Cleland. “How can the system do this impossible thing of prioritizing and interpreting? But the very early circuitry in the system is giving us clues.
“What we think that the first couple of layers of the olfactory system do is to build odors and define their sort of fuzzy boundaries,” Cleland continues. “You get this messy input, and the perceptual system in your brain tries to match it with what you know already, and based on what you expect the smell to be. So the system will suggest that the smell is X and will deliver inhibition back, making it more like X to see if it works. Then we think there are a few loops where it cleans up the signal to say, ‘Yes, we’re confident it’s X.’”
This system allows animals to identify important odors and ignore unimportant ones. It enables clear distinction between almost identical odors when it matters most. Mice, for instance, can identify the dominant among them by their individual odors, even though they all smell like mice. “The olfactory bulb can take things that are very similar and move them apart by accentuating those differences and by identifying the really important distinctions,” Cleland says.
Therefore, what we end up smelling is a complex conversation between the chemical properties of a smell, what we might have been expecting, and what we already recognize. In other words, the discrete odor that we perceive is a product of our ability to categorize and to learn the categories.
The Chemistry of Memory
If learning is so essential to identifying odors, on a molecular level, how and where is that learning achieved? One way Cleland approaches this question is by tracking the effect and the location, within the brain, of molecules and proteins that turn up when learning happens.
“We’ve long known that learning causes certain molecules to be secreted,” says Cleland. “With short-term memory, synapses change strength and embed some kind of learning. And permanency, long-term memory, requires protein synthesis.”
The secretions and syntheses associated with learning have been observed in many areas of the brain and have been studied most often using fear conditioning—issuing shocks to teach mice models to avoid particular locations, for example. Cleland’s lab is interested in the more gradual learning that happens day to day. In an ongoing collaboration with the lab of Jeffrey A. Pleiss, Molecular Biology and Genetics, the two teams are looking at the brains of dozens of mice that have been trained with rewards to learn about odors over different periods of time, with more selectivity for specific odors or less. “We want to find out, in a big data kind of way, what molecules we should be paying attention to,” Cleland says. “Which RNAs, the things that make proteins, are being overexpressed; what’s the time course for over- or under-expression; and what changes in expression levels are a result of this learning? We want to know the patterns.”
This data, Cleland hopes, will shed light not only on the olfactory system but also may provide insight on how memory and learning works in other areas of the brain.
A Computational Model to Match Biology
Cleland’s team tackles these questions—of how learning really works in animals—using many different methods: by studying behavior and learning in mice models; with biological experiments, sending electrical signals through slices of brain; by optically clearing the brain with the CLARITY method, a new technique that allows Cleland to see a three-dimensional picture of stained, fluorescent molecules in a brain, which is otherwise rendered transparent. One of the most important projects in Cleland’s lab is building computer models of mechanisms within the olfactory bulb as well as a model of the entire olfactory bulb itself.
“What the models do is force us to put all of these things together to say that only one story, albeit a flexible one, can actually match all of these things that we study. So everything has the chance to falsify our story,” says Cleland.
To get an accurate biological picture, Cleland’s models must be as complex as the cells and interactions themselves. One major stumbling block has been that the available computer technology doesn’t allow for quick enough communication between machines to efficiently mimic biology. Cleland and his team have overcome this by building Myriad, a massively parallel simulator using video cards to model large-scale systems with dense analogue connections. They’ll release their simulator to the community for use in 2016-2017.
“I think our work will make this kind of modeling much more doable,” Cleland says. “You can only show what a system is capable of up to the limits of how well you’ve modeled it.”
The uniquely complex modeling that Cleland does also serves as a window into why he ended up studying the olfactory system. “It was my inability to decide between doing something that was high-level interesting but tough to study, like cognition, and something that’s physical, measurable—cold, hard science,” he says. “The olfactory system bridges these two levels. It’s high-dimensional and complex, but you can access it and measure it. Then I knew what I wanted to do. I could see a way that I could have an impact.”