In contrast to our pre-registered hypothesis, we found no group differences regarding the step analysis, indicating that nicotine administration had no influence on effort-based effects.

However, the effort-based model also aimed to test several effects that confirm paradigm validity, such as an effect of pathlength differences (Plevel), phases of the experiment and differences of aversive outcomes (AOlevel) (see Table 2 and Table S1/S2 for detailed results). The model-based ANOVA supports these effects and revealed a main effect of Plevel. A higher Plevel indicates a higher effort difference between the two paths. We find that number of steps increase from Plevel1 to Plevel3, indicating that even if the effort to avoid increases over Plevels, participants use the safer, higher effort path. The ANOVA further reveals a main effect of phase (Fig. 2a), indicating that participants took the highest amount of steps (highest effort) during the acquisition phase, when compared to the extinction or the forced extinction. During the forced extinction, the effort to avoid was also significantly lower, when compared to the extinction. Hence, instructing the participants that the snake-symbols are no longer followed by an aversive outcome, reduced the need to avoid and to take the longer paths.

Furthermore, the model-based ANOVA indicated an interaction between AOlevel by Plevel, showing that the lowest Plevel has the lowest number of steps for all three AOlevel, while the highest Plevel has the highest number of steps for all three AOlevel (all corrected p-values < 0.001; Table S2). Further and counter-intuitively, participants at the medium Plevel took more steps at the lowest AOlevel compared to the medium and highest AOlevels. However, this is not the case for either the low or high Plevels, where we found no effect of AOlevel dependence on Plevels (all corrected p-values > 0.05; Table S2). The fact that most steps in the medium path level were taken when faced with the lowest aversive outcome level is inconsistent with the expectation of balanced behaviour between threat and cost. Therefore we complement the analysis of the effort (number of steps) with a binary decision model as an analysis of the avoidance decision in the next section to describe participants’ behaviour. The model-based ANOVA also revealed a Plevel by phase interaction (Fig. 2b; Table 2). Looking at each Plevel individually, the number of steps decrease from ACQ to fEXT (all corrected p-values < 0.001, except low Plevel/ACQ– low Plevel/EXT; Table 2/S2). Also, in each phase the number of steps is increasing from low pathlevel to highest pathlevel (all corrected p-values < 0.001, except medium Plevel/fEXT– high Plevel/fEXT; Table 2/S2).

a) Effort level over phases. Highest effort for avoidance was found during acquisition. During extinction participants learned that it was not necessary to avoid and decreased their effort. Finally, after being instructed for safety, the effort level to avoid drastically decreased during forced extinction. b) Effort level depends on the pathlevel by phase.

Decisions over time per group. Sorted from least avoidance (always the short path) to most avoidance (always the long path), this figure depicts the decision per group over time. The decreased avoidance during extinction phase is visible in the nicotine group (left)

To follow-up on the difference in the decisions to take the shorter vs. the longer path, a binomial test was used to ensure that path decisions were not at chance level, but depended on the threat context of each phase. In all three phases, there was a difference between the path decisions and the chance level of 0.5, meaning that the decisions between paths were not taken randomly. During the acquisition phase, participants more often chose the longer path (87.5%, p < 0.001) than chance. The same effect was found in extinction, showing stronger avoidance by more frequently deciding for the longer path (77.1%, p < 0.001), than chance. During forced Extinction, this effect was the other way round. Now that both paths were instructed to be equally safe, more participants decided to use the shorter path, which takes less effort (79%, p > 0.001), as compared to chance level.

The decisions to take the shorter, more dangerous path, as compared to the longer, safer path per group is illustrated in Fig. 3. This figure illustrates that the placebo group might show a stronger preference for the longer path, i.e. avoidance of the dangerous path during extinction. This effect would go against our pre-registered hypothesis. In order to examine group differences in the decision for the avoidance of the shorter, more dangerous path, we constructed a decision-based general linear model and, as before, ran an ANOVA based on this model. The results of this ANOVA revealed a main effect of AOlevel (F(2,2916) = 6.08, p = 0.002), indicating that the higher the chance to receive an aversive outcome (higher AOlevel), the more likely participants decided to avoid the shorter and chose the longer, safer path (AOlow– AOmedium: z=-3.76, p < 0.001, AOmedium– AOhigh: z=-2.27,p = 0.023). A main effect of phase (F(2,2916) = 15.55, p < 0.001) mirrored the results from the effort-based analysis, indicating that participants showed the highest avoidance of the short path during the acquisition phase and then this avoidance decreased during extinction, while very little avoidance was observed during forced extinction (ACQ– EXT: z = 6.24,p < 0.001, EXT– fEXT: z = 20.75,p < 0.001). Furthermore, we found a trend towards a phase by group interaction (F(2,2916) = 2.33, p = 0.098; Fig. 4a). Even though, this interaction does not meet our statistical alpha-level, this trend goes against our pre-registered hypothesis that nicotine would increase avoidance of the shorter, more dangerous path. Post-hoc tests revealed a group difference during extinction, in which the nicotine group showed decreased avoidance of the short path, as compared to placebo controls (EXTnicotine - EXTplacebo: z=-4.791, pcorr<0.001). We found no influence of nicotine administration on the path decision during acquisition (ACQnicotine - ACQplacebo: z=-0.65, pcorr=0.516) and forced EXT (fEXTnicotine - fEXTplacebo: z = 1.73, pcorr=0.168). In contrast to the effort-based analysis, we did not find an interaction between AOlevel and Plevel (F(4,2916) = 0.29, p = 0.884).

a) Path decision. Nicotine administration leads to a decrease in avoidance behaviour during extinction, compared to placebo controls. b) Prediction error per group during extinction. The steeper slope in the prediction error in the placebo group compared to the nicotine group indicates that nicotine administration might impair to integrate unexpected omissions of aversive outcomes to learn that previously dangerous situations are now safe. Please note that no aversive outcomes were administered during extinction (i.e., unexpected omissions of aversive outcomes)

When including the prediction error score into the decision model, a steeper slope of the prediction error indicates that when more aversive outcomes were experienced in the previous trial, the decision in the present trial would be more likely the longer, but safer path.

Further, we found a phase by group by prediction error interaction (F(2,2951) = 7.34, p < 0.001), indicating a steeper prediction error slope in the placebo group during EXT, compared to the nicotine group (z=-4.07, p < 0.001). This means that the nicotine group decided to take the longer path more often, even when they crossed multiple snakes that did not elicit an aversive outcome in the prior trial, compared to the placebo group. The placebo group was more likely to choose the shorter path, when the previous trial indicated safety (Fig. 4b). Combined with the other results, this means that nicotine generally led to the shorter path being chosen more often (less avoidance), but not because of an enhanced prediction error-based safety learning.

We recorded the intensity of typical side effects, such as a dry mouth, dry skin, vertigo, weariness, blurred vision, nausea or headache on a visual analogue scale of one to seven (1 = no side effect, 7 = extreme side effect). When comparing the rated mean intensity between groups, we found higher intensity of side effects in the nicotine as compared to the placebo group (nicotine– placebo: t(64) = 3.6, p < 0.001). When we looked at each side effect separately, we found that not all of the side effects recorded were more severe in the nicotine group. The Mann-Whitney u test showed that only vertigo, weariness and nausea were significantly stronger in the nicotine group (p < 0.05).

However, the mean of side effects in both groups was smaller than a rated 2 on the scale (2 = very little side effects; nicotine group = 1.62, sd = 0.79; placebo group = 1.11, sd = 0.22). This indicates that although the nicotine group showed increased side effects, compared to the placebo group, these side effects are overall very weak. At the end of the experiment, participants were asked to guess their group assignment and 45.45% guessed correctly. This is in line with expectations of a successful blinding process, as randomly guessing between the two group options would result in 50% correct guesses.