A single oral dose of citalopram increases interoceptive insight in healthy volunteers

Rationale Interoception is the signalling, perception, and interpretation of internal physiological states. Many mental disorders associated with changes of interoception, including depressive and anxiety disorders, are treated with selective serotonin reuptake inhibitors (SSRIs). However, the causative link between SSRIs and interoception is not yet clear. Objectives To ascertain the causal effect of acute changes of serotonin levels on cardiac interoception. Methods Using a within-participant placebo-controlled design, forty-seven healthy human volunteers (31 female, 16 male) were tested on and off a 20 mg oral dose of the commonly prescribed SSRI, citalopram. Participants made judgements on the synchrony between their heartbeat and auditory tones and then expressed confidence in each judgement. We measured three types of interoceptive cognition. Results Citalopram increased cardiac interoceptive insight, measured as correspondence of self-reported confidence to the likelihood that interoceptive judgements were actually correct. This effect was driven by enhanced confidence for correct interoceptive judgements and was independent of measured cardiac and reported subjective effects of the drug. Conclusions An acute change of serotonin levels can increase insight into the reliability of inferences made from cardiac interoceptive sensations. Supplementary Information The online version contains supplementary material available at 10.1007/s00213-022-06115-7.


Fig S1. Distribution of Interoceptive Insight in both Conditions
Somatic and psychological effects 5 Table S1 shows the difference between scores in drug and placebo conditions on subjective ratings at test times. In addition to reporting results with change of heart rate, anxiety and nausea as covariates in the main analysis, 5 we did further analysis to confirm the absence of influence of these factors on our findings. First, we tested for correlations of each change with the effect of citalopram on interoceptive insight (Table S2). No changes in cardiac or self-report variables between drug and placebo were significantly related to interoceptive insight. Next, we conducted a second analysis on restricted datasets, whereby cases were removed until a statistical comparison between citalopram and placebo exceeded p > 0.8. If data is restricted (N=31) to no difference of heart rate (F(1,30)=.01 , = 10 .92, − 0.08, SD 4.4 bpm reduction) between drug conditions, the effect of citalopram on interoceptive insight remains ( (1,30) = 5.14 = .03). If data is restricted (N=37) to no difference on nausea (F(1,36)=.027, p = .87, = .13 SD 4.98), the citalopram effect on interoceptive insight remains ( (1,36) = 5.0, = .03). If data is restricted (N=42) to no difference on anxiety (F(1,41) = .03, p =.87, = .33 SD 12.79), the effect of citalopram on interoceptive insight still remains (F(1,41) = 4.2, p =.046). As noted in the main text, analysis including all these 15 factors simultaneously as covariates also preserves the interoceptive insight effect. Finally, we conducted a mediation analysis ( Fig S2) to assess the possibility of indirect citalopram effects on interoceptive insight via other measured factors. We ran the mediation analyses using a within-subjects approach (MEMORE v2.1), which also includes both changes and average values of the mediator across conditions in the model 5 (Montoya and Hayes 2017). This was conducted for all potential mediators showing differences between drug and placebo i.e., path B, or correlations between drug-placebo differences and task variables (path C). Results are reported in Table S3. No significant mediations were found.

Supplementary Experiment
The same participants performed a visual metacognitive Insight task (VMI) after the interoception task. The results are not directly comparable to the interoception task due to differences in task structure, but are included here because they indicate that effects of citalopram do not extend to visual metacognition. With caution, this 5 demonstrates the potential for selectivity of citalopram effects on interoceptive metacognition.

Methods
VMI task: The visual task was taken from (Fleming et al. 2014). Participants were shown circles containing dots and instructed to indicate which contained more. Following each trial, they were asked to indicate their confidence 10 in the previous response on a Likert scale. 200 trials were conducted in 8 blocks, with a self-timed rest every 25 trials.
The difficulty was staircased over the course of the task, with the difference in numbers of dots (Δd), adjusted to target a mean rate of correct answers of 70%, to keep a consistent level of difficulty between participants. One randomly selected circle always held 50 dots. After two consecutive correct responses, Δd was decreased by one dot; after one incorrect response, Δd was increased by one dot. 15 We calculated Δd corresponding to the number of dots differing between the two circles necessary to maintain 70% accuracy as a measure of performance. Staircasing was successful: mean accuracy was .71 ( = .02) on placebo and .72 ( = .02) on citalopram. Confidence scores were recorded on each trial. Using performance and confidence ratings (1 to 6), we calculated visual metacognitive insight (VMI) as AUC, corresponding to the interoceptive insight measure. This common measure allowed a direct comparison of citalopram effects on cardiac 20 interoception and visual exteroception. We could then look specifically for changes of confidence on correct and incorrect judgements.
We also calculated visual metacognitive sensitivity (meta d') and efficiency (meta d' / d') for reference to previous research (Green and Swets 1966;Fleming and Lau 2014). Correlation of VMI and meta d' in the placebo condition was = .92, and correlation between VMI and meta d'/d' was = .86. We used GLMs including order, 25 nausea, heart rate and anxiety as regressors of no interest to model within-subject differences between drug and placebo. We also examined the effects on confidence independently in correct and incorrect choices.
For comparison of VMI with interoceptive insight controlling for the difference in accuracy and number of trials between tasks, we completed a supplementary analysis on VMI. 1000 random draws of 20 samples each were taken from each participant separately for drug and placebo sessions, weighted to include correct and incorrect trials at the same proportion as that person's interoception task accuracy, in the same session. Computations on insight for each sample were made and the average of these used in statistical comparison with interoceptive insight at the same 5 level of accuracy.
To test if citalopram effects on interoception were over and above a general effect on metacognition, we performed a post hoc analysis of interoceptive insight with the addition of change of visual metacognition as a covariate in our original analyses.
Given an observation of near significant interaction with order of treatment for some measures of visual metacognition, order was included as a between subject factor in all models in addition to change of nausea, heart rate 15 and anxiety.
A comparison of citalopram effects between interoceptive insight and VMI demonstrated a significant drug x task interaction ( (1,45) = 6.56, = .014), with a greater effect on interoception. There was no effect of citalopram on VMI (Table S4). This remains the case for subsets of visual perception data matching interoceptive task accuracy (20 visual task trials with same mean accuracy as interoceptive task performance). 20 Inclusion of visual metacognition changes in the original interoceptive insight analysis (in addition to anxiety, nausea and heart rate) as a covariate did not predict or change the effect of citalopram on interoceptive insight