Temporal sequence geometry enables odor recognition and generalization

Animals encode sensory stimuli with precisely timed neural activity across modalities. For example, mice can rapidly recognize odors, independent of their concentration, using information relayed from their olfactory bulbs by complex spatiotemporal patterns of mitral and tufted cell (MTC) activity. Yet, it remains unknown how MTC activity patterns are structured, and what role precisely timed activity sequences play in guiding perception. Here we investigated the geometry of MTC odor tuning and the structure of odor response sequences by performing fast 2-photon calcium imaging of hundreds of MTCs with sub-sniff resolution. We constructed a space of MTC tuning using the pairwise correlations between MTC odor responses averaged over a sniff and discovered that odor specific sequential activity traversed this space in smooth, continuous trajectories. Further, we found that the early but not the later part of sequences carried concentration invariant information about odor identity. Finally, inspired by observation that neighboring MTCs in the tuning space are activated together in time independently of the odors presented, we propose a role of activity sequences in training the piriform cortex to learn perceptually generalizable odor representations. To test these ideas, we constructed and analyzed a computational model for sequence-based unsupervised training of synapses from MTCs to the piriform cortex, which revealed that sequential activity across the entire sniff permits perceptual generalization for novel odors, acting as a scaffold for learning relevant activity manifolds between networks.