Fig. 2: Elevated noradrenaline during learning increases overgeneralisation in behaviour.

a Schematic of example trial in the ‘Explicit memory test’ performed on Day 5, where participants were tested on the bird-bird associations learned on day 1. b Schematic of example trial in the ‘Implicit memory test’ performed on Day 5, where participants were required to recall which contextual cue (sofa) was incidentally paired with each bird stimulus. c-d, f, g Upper: Bootstrap-coupled estimation (DABEST) plots. Black dot, mean; black ticks, 95% confidence interval; filled curve, sampling error distribution. Lower left: memory accuracy (mean +/- SEM). Lower right: null distribution of the group differences generated by permuting subject labels, purple dot: true group difference. c Average memory accuracy in the ‘Explicit memory test’ was high. There was no significant group difference in overall accuracy (ATX – PLC (n = 22:22): permutation test: p = 0.421). d Average memory accuracy in the ‘Implicit memory test’ was relatively low: on average participants made errors on 75.4% of trials. There was no significant group difference in overall accuracy (mean: ATX: 75.3%; PLC: 75.6%; ATX – PLC (n = 22:22): permutation test: p = 0.498). e Schematic illustrating the predicted distortion in the cognitive map, where ‘spread of association’ leads to overgeneralisation in participants’ implicit memory. f ‘Overgeneralisation errors’ in the ‘Implicit memory test’ were defined as the percentage of error trials where, given the probe, the chosen contextual stimulus was neighbouring to the correct contextual stimulus. The ATX group made significantly more ‘Overgeneralisation errors’ than expected by chance (p = 0.014, effect size computed from 10,000 bias-corrected bootstrapped resamples²⁷), and significantly more than the PLC group (ATX – PLC (n = 22:22): 2-sided permutation test p = 0.048). g ‘Mean rank proximity’ in the ‘Implicit memory test’ was defined as the average link distance between the probe and erroneously chosen contextual stimuli. For instance, a choice for sofa ‘B’ with probe ‘1’ would be ranked as ‘2’ whilst a choice for sofa ‘D’ would be ranked as ‘4’. ‘Mean rank proximity’ across all error trials in the implicit memory test in ATX was significantly lower than expected by chance (ATX: n = 21, p < 0.001, effect size computed from 10,000 bias-corrected bootstrapped resamples²⁷) and significantly lower than in PLC (ATX – PLC (n = 22:22): 2-sided permutation test: p = 0.031).