From left to right: Roman Shusterman from University of Oregon, Alexei Koulakov from Cold Spring Harbor Laboratory, Massimo Vergassola from UC San Diego, and Dima Rinberg from New York University School of Medicine.
Olfaction is the final frontier of our senses, the one that is still mysterious to us. Despite large amounts of genetic, physiological, and perceptual data, many fundamental questions about our sense of smell remain unresolved. For example, unlike the analogous cases of vision and hearing, to a large degree, we do not understand what properties of molecules lead to olfactory signals. Attempts to predict a smell quality based on molecular structure do not produce reliable results. This is not to say that olfaction is not important to humans. Our sense of smell contributes much to of our perception of the taste of food, is involved in mechanisms of sexual imprinting, and otherwise strongly enhances our quality of life.
One of the possible reasons for the elusive nature of our sense of smell is its enormous complexity. The olfactory epithelium of our nose can be viewed as a single pixel with about 350 types of smell “color” sensors, often called odorant receptors. Each odorant receptor is a protein that can recognize odorant molecules floating in the air. Imagine an LED display with a single pixel and 350 types of color diodes, and you will get an idea of how rich and complex smell can be. Olfactory epithelia in mice and dogs contain a stunning 1000 different types of odorant receptors (humans have lost a large fraction of their olfactory receptor repertoire since the acquisition of three-color vision). If odorant receptors were all independent and evolved without awareness of each other, we would need to untangle the structure of a 1000-dimensional phase space – a task intractable with current experimental methods. Another possibility is that the odorant receptors sample a lower-dimensional subspace of the 1000-dimensional space. Some recent human perceptual data suggests that olfactory perceptual space, i.e. the phase space sampled by the olfactory receptors, is less than ten dimensional. Some of these dimensions are easy to comprehend, such as a smell’s pleasantness or complexity. Other dimensions represent molecular properties, such as hydrophobicity. Most of the ten dimensions are identified only by computational correlations found in the data, and are thus far unrelated to a particular human experience. Ten dimensions is a lot, but it is much better than 1000! The 10D odor universe, or Smellyverse, as we have called it, has the potential to be understood, in the same way that the 3D red-green-blue color universe has been understood by visual scientists.
This April, a group of theorists got together at KITP in a Follow-on Program to the program “Deconstructing the Sense of Smell” held last year. The main focus of the Follow-on Program was to design a combination of experimental and computational methods to probe the low-dimensional structure of the Smellyverse. We also attempted to speculate about the reason why this space may have evolved to be low-dimensional. The reason why color space was hypothesized to be only 3D is because of the limited space in each of the retinal pixels: increasing dimensionality of color space would mean putting more types of color sensors in each pixel and would lead to worsening of the eye’s spatial resolution. Limits to the dimensionality of sensory space are usually produced by the biological constraints imposed on the sensors. What types of constraints can exist in the olfactory system that restrict the dimensionality of the Smellyverse? We speculate that the constraints imposed on the olfactory system are behavioral: we have to sense and identify smells at minute concentrations to be able to differentiate hazards from attractants before other individuals. Being able to tell friend from foe at the smallest odor concentrations can differentiate a successful hunter/gatherer from some predator’s dinner. The need to balance sensitivity with complexity of odor signals limits the dimension of the odor space, in the same way that the balance between the diversity of colors and the spatial resolution of the eye limits the dimensionality of color space to three.
This hypothesis was fleshed out by data obtained by one of the Follow-on program participants, Dima Rinberg, an Associate Professor at the New York University. Dr. Rinberg is both a theorist and an experimentalist. As an experimental neuroscientist, he designs methods for stimulating mouse odorants receptors by light, turning a mouse’s nose into an extra retina. Using these methods, Dr. Rinberg can both create new unexpected odor signals and interfere with existing ones. Drawing new smells in the nose with light gives researchers a unique opportunity to explore the sense of smell. By interfering with the responses of the odor receptors, Dr. Rinberg and his collaborators managed to deduce that, in identifying smells, mice rely on the receptors with strongest affinity to odorants, i.e. those most sensitive to molecules at smallest concentrations. These receptors are the first activated after the odorant reaches the nose, and, as such, allow the fastest recognition of the identity of odors present at minute concentrations. Says Dr. Rinberg: “Using the most sensitive receptors to recognize odors is similar to aiming your eye’s most sensitive part at important objects in the environment -- we can force ourselves to recognize objects without directly looking, but we tend to aim our eyes at something that is important to us at the moment”.
Another way of thinking about the space of odors is to imagine them placed in physical space. Several researchers, including Massimo Vergassola from UC San Diego and Lucia Jacobs from UC Berkeley, propose that the olfactory system, perhaps on a very high level, computes the map of odorants in the physical space. Thus the olfactory system answers not only the “what” but also the “where” question. Evaluating the “where” information from elusive plumes of smells carried by the turbulent flow of air involves a very special type of computation. Understanding these neural algorithms will shape our thinking about the brain’s processing of odor information. Indeed, perhaps, some of the many dimensions of the Smellyverse are real space dimensions.
One of the foci of the KITP program, supported in part by the National Institutes of Health (NIH) and the Gordon and Betty Moore Foundation, was to understand the type of computational algorithms used by olfactory networks. Some of these algorithms are known, such as compressed sensing or Bayesian inference, but others are only beginning to be understood. The most exciting possibility, however, is that our KITP program will inspire the creation of new mathematical frameworks that will apply generally to olfactory, visual, auditory networks and beyond, It is our hope that the brain is up to the challenge to create theories enabling it to understand itself. A single brain, however magnificent, will not be sufficient. Combining several brains is necessary to achieve this worthy goal, which is something that the KITP program accomplished.
- Alexei Koulakov, Deconstructing the Sense of Smell Program Coordinator
KITP Newsletter, Fall 2016