Decorrelation by gain control in the mouse olfactory bulb

Abstract

Classifying sensory stimuli requires neural circuits to transform overlapping inputs into representations that are more distinct and decodable. Odour representations in the olfactory bulb (OB) have been reported to decorrelate from sensory input to output, a transformation thought to enhance odour separability. What has been missing is a systematic quantification of this transformation across odours, and a mechanistic account of how it can be implemented by OB circuitry. Using in vivo two-photon calcium imaging, we recorded glomerular input and output responses to a chemically diverse panel of 47 odours in mice. We observed a robust decorrelation of odour-evoked activity patterns from input to output: correlations between odour representations were consistently reduced, while overall response variance was preserved. This transformation increased the dimensionality of the population code and improved the separability of odour representations. Consistent with this, classification analyses showed that odour identity could be decoded with significantly higher accuracy at the output. To understand the mechanisms underlying this transformation, we developed a linear model of the OB with different levels of connectivity. Models in which each channel underwent only self-gain modulation reproduced the observed decorrelation nearly as well as models with fully unconstrained lateral connectivity. Analysis revealed that channels contributing unique information were selectively amplified, while redundant channels were attenuated. Together, our results show that the OB implements a measurable input-output transformation that enhances odour separability. By combining systematic recordings with computational modelling, we found that feedforward gain modulation serves as a simple and scalable mechanism capable of explaining the observed decorrelation, reframing how we understand the functional architecture of early olfactory processing.