Smokers show increased fear responses towards safety signals during fear generalization, independent from acute smoking

Smoking is highly prevalent among patients with anxiety disorders. Previous studies suggest that smokers show altered fear learning as compared to non-smokers. To test the effect of acute smoking on fear learning and generalization, we conducted a fear learning experiment online. 202 healthy subjects learned to differentiate a danger and a safe cue on day 1 and were tested for generalization of threat responses 24 h later. To see if the timing of smoking impacts fear learning, we formed three smoker groups with manipulations of acute smoking and withdrawal at different time-points (each group: n = 46) and one non-smoker control group (n = 64). Smoking manipulations contained a 6 h withdrawal after fear learning, smoking directly before or after fear learning. We found no group differences between smoker manipulation groups for fear learning or generalization. However, we found differences in fear generalization between smokers and non-smokers. Smokers showed increased fear ratings towards the stimulus that has been learned as safe and higher US expectancy to stimuli similar to the safe stimulus, when compared to non-smokers. Smoking might constitute a risk factor for impaired discrimination between danger and safety and smoking restrictions could be an effective way to reduce the risks of development or maintenance of anxiety disorders.


Statistics Fagerström
We checked both US expectancy and fear ratings for an effect of nicotine dependence with the Fagerström test for nicotine dependence (FTND). The score of the FTND was included into the model (lmer (RatingResults~(1|participants) +stimulus*time*group+fagerström)). The group factor that is included in the model includes the three groups with smokers. We found no main effect of the Fagerström score for US expectancy rating on day 1 (F(9,125.12)=0.409,p=0.928) or day 2 (F(1,133.91)=1.213,p=0.273). Also we found no main effect of the Fagerström score for fear rating on day 1 (F(1,133.38)=0.958,p=0.3295) or day 2 (F(1,134)=1.837,p=0.178). When checking for interactions of the FTND with our fixed effects in the model, we found a trend towards a stimulus by group by FTND interaction (F(2,2964.54)=2.532,p=0.08). For further analysis, we calculated correlation coefficients between the differential US expectancy rating on day 1 and the FTND for each group separately. As the data shows no normal distribution, we used spearman's rho for correlation analysis. Only group 4 showed a correlation between US expectancy rating and FTND (rho=0.163,p<0.011). No correlation between US expectancy on day 1 and the FTND was found for group 2 and group 3 (group 2:rho=-0.049,p=0.282; group 3:rho=-0.016,p=0.722).

Statistic Withdrawal
We checked both US expectancy and fear ratings for an effect of withdrawal. The score of withdrawal symptoms was included into the model (lmer (RatingResults~(1|participants) +stimuli*time*group+withdrawal)). The group factor that is included in the model includes the participants of group 2 and the regrouped participants into group 3. Statistics has only been calculated for the Generalization test, as participants were only asked to take a smoking break between day 1 and day 2 and there should be no effect of this smoking manipulation on day 1. We found no main effect of the withdrawal for US expectancy rating (F(1,66.99)=0.945,p=0.334) or fear rating (F(1,67)=0.121,p=0.729) on day 2.

Statistic sex
We checked both US expectancy and fear ratings for an effect of sex of participants. Sex was included into the model (lmer (RatingResults~(1|participants) +stimuli*time*group*sex)). The group factor that is included in the model is smoker vs. non-smoker. We found a main effect of sex for US expectancy rating for day 1 (F(1,1139)=6.389, p=0.012), but follow up post-hoc tests did not reveal any differences. Additionally, we found a stimulus by sex interaction (F(1,4376.1)=5.895, p = 0.015), which consisted of an increased US expectancy of females towards the CS-(CS-/male -CS-/female: estimate=-0.416, SE=0.180, z-ratio=-2.306, pcorr=0.042). Also we found a trend towards a group by sex interaction (F(1,1145.3)=2.733, p = 0.099). The follow up post-hoc test revealed an increased US expectancy of smoking females, when compared to non-smoking males (non-smoker/malesmoker/female: estimate=-0.482, SE=0.243, z-ratio=-1.986, pcorr=0.047). We found no further main effect or interaction of sex on the US expectancy on day 2, or the fear rating on both days.

Statistics reaction time
We checked US expectancy ratings on both days in regard of the participant's reaction time. We included reaction time as dependent variable into our model. Stimulus, block and group (smoker vs. non-smoker) were included as fixed effects into the model (lmer (reactiontime~ stimulus*block*group)). On day 1 we found a trend towards a main effect of stimulus (F(1,4383.5)=3.10,p=0.078), but follow up post-hoc tests showed no difference of reaction time between CS+ and CS-. Additionally we found a main effect of block (F(2,4384.90)=11.667,p<0.001). Follow up post-hoc tests revealed that subjects were rating faster over time (block 1-block 2: estimate=0.316,SE=0.0346,z-ratio=9.118,pcorr<0.001; block 2block 3:estimate=0.063,SE=0.0346,z-ratio=1.805,pcorr=0.071). On day 2 we found no differences regarding the reaction time.

Statistic STAI-T
We checked trait anxiety with the State-Trait Anxiety Inventory (STAI-T). An ANOVA showed no effect of group for the smoker manipulation groups (F(3,198)=0.805,p=0.492). An independent sample t-test revealed also no differences for smokers vs. non-smokers (t(200)=1.541,p=0.125).

Statistics age
We checked both US expectancy and fear ratings for an effect of age of the participants. We added age to the model (lmer (RatingResults~(1|participants) +stimuli*time*group*age) for the analysis of US expectancy and we added age as between subject factor to the type 3 rmANOVA in jasp for the analysis of the fear ratings. The group factor that is included into the model is smoker vs. non-smoker. We found no main effect or interaction of age on either day for US expectancy and fear rating. Inclusion of this variable as separate covariates still yielded robust results of our group differences.

Statistics alcohol and coffee consumption
We checked both US expectancy and fear ratings for an effect of alcohol or coffee consumption of the participants. Subjects stated their alcohol consumption as number of glasses per week and coffee consumption as number of cups per day. Two subjects were excluded from the coffee consumption analysis, because of unrealistic declarations of their coffee consumption (14 cups/day). We added either alcohol or coffee to the model (lmer (RatingResults~(1|participants) +stimuli*time*group*consumption) for the analysis of US expectancy and we added alcohol or coffee consumption as between subject factors to the type 3 rmANOVA in jasp for the analysis of the fear ratings. The group factor that is included into the model is smoker vs. non-smoker. We found no main effect of either alcohol or coffee consumption on either day for US expectancy and fear rating. Nevertheless, we found for the US expectancy ratings on day 1 a stimulus by block by group by coffee consumption interaction (F(2,4347.4)=4.042, p=0.0176). The interaction reflects increasing differentiation between the CSs with increasing amount of coffee in smokers (r=0.1, p=0.028), whereas this relationship is not existing (or even the opposite) in non-smoking individuals (r=-0.076, p=0.251). This difference between groups is most pronounced within the first block during acquisition training (z = 2.212, p = 0.027). Importantly, we found no effect of coffee consumption on the retrieval or generalisation of CS-responses on day 2, where we observed the most pronounced difference between smoking and non-smoking participants. These control analyses that included alcohol consumption and coffee consumption as variables consistently revealed smoking status as the significant predictor across measurements. Figure S1: Main effect of group (smokers vs. non-smokers) regarding the fear rating on day1. We found an increased fear rating in smokers, when compared to non-smokers. [**] indicates p<0.01.