Fig. 6: A model of the OB suggests a mechanism by which abGCs increase information in MCs.

a Schematics of MCs and GCs/abGCs in the olfactory bulb. b A schematic illustration of the model. Blue arrows denote inhibition and red excitation. c Connection strength between MCs and GCs/abGCs as a function of spatial proximity. Horizontal lines display the average connectivity based on connectivity width and peak. abGCs have broader, more promiscuous input connectivity than mature GCs. “abGCs input broadness” is the ratio between abGCs and GCs average connectivity. d Input–output functions for both GCs and abGCs. “abGCs excitability” is the ratio between abGCs and GCs input–output slope. e Silencing young abGCs (2.5% of total GCs) quenches MC responses (red curve, cf. Fig. 3e). Silencing the same amount of mature GCs had no effect (green curve). f Sharpening of MC tuning due to entire GC network (\({\mathrm{Sharpening}}_{{\mathrm{allGCs}}}\), i.e., the ratio of MC output tuning to input tuning) as a function of both GC-mediated MC lateral inhibition and abGC efficacy. Efficacy is defined as the product of abGCs excitability (d) and their input broadness (c). Solid line represents the parameter set estimated by constraining the model to fit our data. Dashed lines—confidence borders (see “Methods”). g \({\mathrm{Sharpening}}_{{\mathrm{abGCs}}}\) as measured in our data. Dashed lines—confidence borders (see “Methods”). Black dot represents the parameters used in our model. h Number of MCs needed to maintain a fixed level of discrimination between odors (Chernoff distance; see “Methods”) as a function of the proportion of young abGCs. Red/blue gradients represent increased/decreased discriminability, respectively. The red dot represents our baseline model network (1000 MCs and 2.5% (2500) abGCs). Black line—fixed value of discrimination in our model network (≈30 D_c), for different MCs/abGCs ratios. Arrow—proportion of abGCs (7.5%) that would achieve half the potential reduction in the number of MCs that an all-abGC network would achieve. a was created with BioRender.com.