Two of the most fundamental questions of sensory neuroscience are: 1) how is stimulus information represented by neuronal activity? and 2) what features of this activity are read out to guide behavior? The first question has been the subject of a large body of work across different sensory modalities. The second question remains a significant challenge, since one needs to establish a causal link between neuronal activity and behavior.

In olfaction, it has been proposed that information about odors is encoded in spatial distribution of receptor activation and the next level mitral/tufted cells, as well as in their relative timing and synchrony. However, the role of different features of neural activity in guiding behavior remains unknown. Using mouse olfaction as a model system, we developed both technological and conceptual approaches to study sensory coding by perturbing neural activity at different levels of information processing during sensory driven behavioral tasks. We developed methods for both one-photon spatiotemporal pattern stimulation using digital mirror devices at the glomerulus level in the olfactory bulb, and two-photon holographic pattern stimulation deeper in the brain, at the level of mitral/tufted cells. Using these techniques, we performed quantitative behavioral experiments to, first, measure psychophysical limits of the readability of different features of the neural code, and, second, to quantify their behavioral relevance. Based on these results, we built a detailed mathematical model of the behavioral relevance of the different features of spatiotemporal neural activity. Our approach can be potentially generalized to other sensory systems.