Supplementary Figure 5: Prediction of odor preference across experimental groups (related to Fig. 5).

a. To confirm that coding structures undergo consistent changes along an axis representing valence we tested whether experience-dependent changes in coding structures can predict odor preference across experimental groups. We defined two subsets of fish: (1) the reference group, which consisted of naïve fish and one experimental group (for example, NAV (N = 13 animals) and ALA (N = 12)), and (2) the test group, which contained all other fish (for example, TRP (N = 16), HIS (N = 15), UNC (N = 12)). We then projected coding structures of the test group onto the PC 1 extracted from the reference group and asked whether the projections (‘test scores’) were correlated to behavioral odor preference in the test group. In all cases, correlations between test scores and behavioral odor preference were high and statistically significant. Hence, PC 1 extracted from all animals in any experimental group (and all NAV fish) defined a direction in coding space that reliably predicted behavioral odor preferences in the other, remaining experimental groups. These results directly demonstrate that different odor-reward assignments, as well as uncoupled odor exposure, resulted in modifications of coding structures along an axis that consistently represented valence (attractiveness). Pearson correlation coefficient (r) is reported (P values were determined using a t test of the null hypothesis of r = 0, two-sided, with d.f. = N–2). b. Same analysis as in (a) but the reference group contained NAV fish and two experimental groups (for example, NAV, HIS, UNC) while the test group contained the other two experimental groups (for example, ALA, TRP). As in (a), statistically significant correlations between test scores and behavioral odor preference were observed in all cases (correlations and statistical tests as in (a)). Number of animals in each group as in (a). c. Same analysis as in (a) but ALA fish were excluded from all analyses. The influence of ALA fish on the results was examined because Ala is an innately attractive odor. Excluding ALA modified correlations only minimally (compare to corresponding plots in b). Hence, results can be generalized in this dataset independently of the innate valence of the CS⁺. Number of animals in each group, correlations, and statistical tests as in (a). d. Correlation between PC 1 and behavioral odor preference when odor responses to Ala were excluded from the analysis in all experimental groups. Excluding Ala as an odor stimulus reduced the number of coding structure dimensions from six to three. Nevertheless, the correlation between PC 1 and behavioral odor preference remained highly significant (left) and the mapping of coding structures onto the first two PCs (right) was similar to the mapping under control conditions (Fig. 5). Hence, a consistent axis representing valence was identified by PCA even when responses to the innately attractive odor Ala were not considered. Number of animals in each group, correlations, and statistical tests as in (a).