Fig. 3: Olfactory-association-learning network expresses monotonic RP and prediction error components.

a Illustration of the MRI-compatible behavioral setup with MRI coil (left) comprising an odor port, a lick port, respective supply tubes, an infrared beam breaker, a head-fixation system, and a back cover (middle and right). b Anatomical illustration of olfactory, striatal and higher-order regions, mapped onto an MRI template brain (location from Bregma indicated in mm). aIC agranular insular cortex, AON anterior olfactory nucleus, aPC anterior piriform cortex, dStr dorsal striatum, M1/M2 motor cortex, mPFC medial prefrontal cortex, NAc nucleus accumbens, OB olfactory bulb, OFC orbitofrontal cortex, S somatosensory cortex, Tu olfactory tubercle. c Group-level Z-statistical maps showing BOLD correlates of RP values V(CS) from the TD model (n = 51 sessions in 18 animals). Statistical threshold was set to P < 0.025 false discovery rate (FDR)-corrected, for two-sided testing (as for the other maps unless otherwise indicated). Red color spectrum indicates areas with positive correlations to RP values, while blue colors indicate negative correlations. The regression revealed recruitment of olfactory and striatal brain regions with opposing effects on BOLD. d Within the regions associated with V(CS) from (c), monotonic RP (CS100 > CS50 > CS0 in red or CS100 < CS50 < CS0 in blue) was represented mainly in prefrontal, insular, striatal regions, and olfactory bulb. Maps were created by intersecting contrast maps CS100 > CS50 and CS50 > CS0, each thresholded at P < 0.025 FDR-corrected; and vice-versa: CS100 < CS50 and CS50 < CS0 (see the respective contrast maps in Supplementary Fig. 3e, f). e Recent outcome history modulating the BOLD response to CS50. Within the V(CS) regions from (c), we tested for a CS50 response enhancement by the reward in the previous CS50 trial. In olfactory and striatal regions, the CS50-associated response was indeed strengthened by a positive outcome history. f–h In the TD model, the prediction error at US is r-V(CS). We identified brain regions correlating with both prediction error components by intersecting the regressor maps of (f) r and (g) -V(CS) at US, each thresholded at P < 0.05 FDR-corrected. The intersection, shown in (h), was largely confined to the lateral NAc, the posterior Tu, and aIC.