If mitral cells are the answer, what is the question?

Why does the olfactory bulb contain two projection neuron populations? Called mitral and tufted cells (MCs and TCs), both sample receptor input at glomeruli but lie in different bulb layers, project to different regions of piriform cortex and form parallel circuits wherein TCs interact mainly with the anterior olfactory nucleus (AON), and MCs with piriform cortex (PC) proper. Experiments have revealed that (1) MCs are primarily driven by TCs, perhaps why (2) MC responses come later in the sniff cycle than TCs; (3) MCs receive excitatory cortical feedback from the AON; and (4) MC responses are harder to decode for odour properties than TCs. We connect these observations and propose that parallel TC-AON and MC-PC circuits provide parallel explanations of glomerular input as caused by high dimensional latent odour features and implemented in AON/PC pyramidal activity. However, the more peripheral TC-AON circuit uses a simple prior to quickly sketch the olfactory environment, while the deeper MC-PC circuit uses a more complex prior to construct a more accurate explanation of the odour input. Steady-state MC/TC activity reflects the error between glomerular input and its explanations, so the better explanation by the MC-PC circuit yields lower activity, potentially explaining observation (4) above. The similarity of the two computations makes it natural to express MC drive in terms of TC activity, and doing so in circuitry readily reproduces observations (1-3). Thus we suggest MCs are the answer to how the olfactory system can explain glomerular inputs using a more complex prior than that of the TC-driven circuitry, while still using that drive to avoid unnecessary computation.