Figure 4. Neural firing of the common population of cells during the compound probe day.

In all, 26 cells showed decline in neural firing across trials to both overexpectation and omission. This number of cells was greater than chance: a Monte Carlo simulation with one million repetitions reported that on average seven cells (full range: 0–22) were reward-responsive and showed a decline in neural firing to both the overexpectation and omission compounds from the early to the late trials (t(99999)=−7.7 × 10³, P<0.05). This was confirmed by a χ²-test of independence (χ²=13.5, P<0.05). (a) Correlation of the change in neural firing (early-late trials) between the overexpectation and omission compounds. This relationship was trial-specific and dependent on the true overexpectation–omission pairs. Randomly shuffling the trials in an independent manner disrupted this relationship: The true correlation was different to the correlation distribution obtained from a million permutations of the data (P=0.0103). Randomly shuffling the order of the neurons for Omission while maintaining the order of the neurons for overexpectation also disrupted the correlation. The correlation coefficient obtained by the true unshuffled data were different from the distribution of correlation coefficients obtained following one million randomized permutations of the data (P=0.0053). (b) Distribution of difference indices (from control) for overexpectation and omission during the first trial (insets), and during the early (blue bars) and late (yellow bars) trials of the compound probe. The distribution of these indices for the overexpectation compound was positively shifted during the early trials (μ=0.54, t(25)=4.2, P<0.05) with this shift being greatest on the very first trial (μ=0.99, t(25)=2.7, P<0.05, see inset), while it was centred on zero during the late trials (μ=0.02, t(25)=0.3 P>0.05). The shift in firing to the overexpectation compound relative to the control compound was significant (t(25)=3.9, P<0.05). The distribution of the indices for the omission compound was negatively shifted during the early trials (μ=−0.41, t(25)=−3.8, P<0.05) and even more during the late trials (μ=−0.72, t(25)=−5.5, P<0.05, with a significant difference between the two distributions (t(103)=2.5, P<0.05). The difference in firing between the omission and control compounds on the first trial was centred on zero (μ=−0.33, t(25)=−1.3, P>0.05, see inset).