Functional brain responses to emotional faces after three to five weeks of intake of escitalopram in healthy individuals: a double-blind, placebo-controlled randomised study

Short-term intake of selective serotonin reuptake inhibitors (SSRIs) modulates threat-related amygdala responses in healthy individuals. However, how SSRI intake over a clinically relevant time period modulates threat-related amygdala responses is less clear. In a semi-randomised, double-blind, placebo-controlled study of 64 healthy individuals (SSRI n = 32, placebo n = 32), we examined the effect of 3–5 weeks of SSRI escitalopram (20 mg daily) on brain response to angry, fearful and neutral faces using BOLD fMRI. Data was analysed using a whole-brain region-wise approach extracting standardised effects (i.e., Cohen’s D). The study was conducted at the Copenhagen University Hospital. A priori, we hypothesised that SSRI would attenuate amygdala responses to angry and fearful faces but not to neutral ones. Whether SSRI modulates correlations between amygdala responses to emotional faces and negative mood states was also explored. Compared to placebo, 3–5 weeks of SSRI intake did not significantly affect the amygdala response to angry, fearful, or neutral faces (|Cohen’s D|< 0.2, PFWER = 1). Whole-brain, region-wise analyses revealed significant differences in frontal (|Cohen’s D|< 0.6, PFWER < .01) and occipital regions (|Cohen’s D|< 0.5, PFWER < .01). SSRI did not modulate correlations between amygdala responses to emotional faces and negative mood states. Our findings indicate that a 3–5 week SSRI intake impacts cortical responses to emotional stimuli, an effect possibly involved in SSRI’s therapeutic efficacy. Trial registration Clinical Trials NCT04239339.


Participants
In total, 64 healthy volunteers (mean age in years (SD) [range]: 25 (6)  [19-45], 41 females) completed the study.The volunteers were recruited from a list of individuals who expressed interest in participating in brain research.See Fig. 1 for the CONSORT flow diagram of the inclusion of research volunteers and reasons for missing data, which is detailed in appendix 1 in the supplementary material.The participants underwent screening for somatic illness, including a medical examination, blood screening for somatic disease, an ECG, and the presence of psychiatric conditions.See appendix 2 in the supplementary material for the exclusion criteria.The primary outcome of the overall study was to assess effects of SSRI on cognition assessed behaviourally, which is published elsewhere 49 .The primary outcomes were used to determine the study's sample size, detailed in the aforementioned publication.

Experimental design and drug administration
In a double-blinded design, participants were semi-randomised stratified with respect to sex, age, and IQ to receive a clinically relevant dose of escitalopram (20 mg daily in capsules of 10 mg) 50 or placebo in identical capsules for 3-5 weeks, representing a clinically relevant time period of SSRI intake 32 .The randomisation was done by an administrative staff member uninvolved in inclusion, data collection or analysis.Capsules were manufactured and distributed by the Capital Region Pharmacy in Denmark.Participants received oral and written instructions for taking escitalopram, including possible side effects.They were instructed to take one capsule (10 mg escitalopram in the active group) each morning for 3 days and two capsules (20 mg escitalopram in the active group) daily on the fourth day and until the last day of examination after 3-5 weeks.The study was conducted at the Copenhagen University Hospital, Rigshospitalet, between May 2020 and October 2021.
Participants were asked to fill in a daily medication logbook, recording the time and number of capsules taken, serving as a daily reminder to take the capsules and providing additional information about participant compliance.A blood sample was taken halfway through the intervention period and again at follow-up, typically in the morning before MRI, to verify intervention compliance.Further, participants' capsule containers were inspected at each visit, and their medication logbook was verified at the follow-up visit.Participants were instructed to take the capsule after the blood sample on blood sampling days.During the intervention period, side effects were monitored weekly.If participants experienced unusual side effects or other complications during the study period, they could contact the project physician.
Following the intervention period, participants completed an MRI scan session.Before MRI scanning, participants completed self-report questionnaires to evaluate their psychological state.Participants and investigators involved in data acquisition and analysis were blinded to intervention type until completion of data analysis.Once participants had completed the study, we assessed the blinding efficacy by asking whether they believed they had received an active drug or placebo.

Psychometrics
Before participants commenced the intervention, we measured their intelligence quotient (IQ) using the Reynolds Intellectual Screening Test 51 , body mass index (BMI) and administered questionnaires to assess levels of depressive symptoms using the Beck's Depression Inventory II (BDI-II) (range: 0-63, where a score above 20 indicates a dysphoric or depressed mood) 52 , levels of state anxiety using the State-Trait Anxiety Inventory (STAI) 53,54 (range: 0-80, higher levels indicating more anxiety, but no cut-off adapted to indicate severe clinical anxiety) and stress level using the Cohen's Perceived Stress Scale (PSS) (range: 0-40, higher levels indicating more stress, but no cut-off adapted to indicate severe clinical stress) 55 .
Both at baseline and follow-up (i.e., on the same day as MRI scanning), and typically in the morning, we assessed participants' level of negative mood states using the validated self-report questionnaire, the Profile of Mood States (POMS) 56,57 .The questionnaire was developed to measure transient mood states from 65 items using a 5-point Likert scale ranging from 0 (not at all) to 4 (extremely), from which a total composite score indicates the level of negative mood states (i.e., tension/anxiety, depression/dejection, anger/hostility, fatigue/ inertia, and confusion/bewilderment versus vigour/activity).Total score ranges from − 32 to 200, with higher scores indicating a more negative or disturbed mood but no cut-off indicative of severe mood disturbance 57 .

Emotional faces fMRI paradigm
The emotional faces paradigm consisted of four blocks of emotional faces (i.e., fear, anger, surprise, and neutral blocks, presented in a randomised order) interleaved with five control blocks of geometric shapes (i.e., circles and vertical/horizontal ellipses) and lasted 6.5 min 35 .A trio of stimuli was presented, with two on the bottom of the screen and one on the top.Participants were instructed to identify which of the two stimuli on the bottom matched the one on top as quickly and accurately as possible.For the faces blocks, a trio of stimuli were presented, including two similar faces to match and one odd face, all expressing the same emotion.The matching task was intended to ensure that participants attended to the stimuli.Written instructions (i.e., "match faces") were displayed (2 s) at the beginning of each faces block.Within each face block, six face trios were presented for four seconds each, interleaved with a fixation cross (" + ") presented for two, four or six seconds, mean = 4 s, to minimise expectancy effects and habituation.The control block followed a similar procedure; written instructions (i.e., "match shapes") were displayed for two seconds, and the fixation cross was displayed for fixed 2-s periods.The paradigm and participants' accuracy and reaction times were recorded in E-prime (Psychological Software Tools, Pittsburgh, USA).See appendix 3 for details on MRI acquisition.As we were only interested in the effects of SSRI on amygdala response to negative and neutral faces, we only included and analysed data from the angry, fearful and neutral face conditions, as done in previous studies 23,24,29,58 .The emotional faces paradigm was a secondary outcome of the overall study.

Data analysis
Pre-processing and analyses of fMRI data See appendix 4 for a description of pre-processing of fMRI data.We analysed data from each participant in general linear models (GLMs) by employing a canonical hemodynamic response function to the smoothed functional images to estimate task-specific BOLD activity (i.e., beta images).A high-pass filter (128 s) was applied to control for slow-frequency fluctuations.Motion parameters and censored volumes were included as covariates.The GLMs were used to generate contrast images (linear combination of task-specific effects) for our effect of interest (i.e., fearful, angry, and neutral faces contrasted to geometric shapes).Next, group-level analyses were used to determine population-level brain responses to fear, anger, and neutral faces in the SSRI group compared to the placebo group.The amygdala, our primary ROI, and all other brain areas were defined using the WFU PickAtlas v3.0.3 59,60 .For each participant, we extracted mean signal values from the bilateral amygdala for the main analyses, and for hemisphere-specific, whole-brain regions for secondary analyses, which were evaluated further in R version 4.0.3(https:// cran.r-project.org/).
We also conducted exploratory post hoc voxel-wise analyses to probe possible sub-regional effects.Following pre-processing that involved warping data in Montreal Neurological Institute (MNI) space (see appendix 4 for details), we fit the same GLMs described above to estimate task-specific BOLD responses for each participant, which were used in a group-level design matrices to estimate intervention effects.

Quantification of serum escitalopram concentration
We analysed the concentration of serum escitalopram (nmol/L) using ultra-high performance liquid chromatography/tandem mass spectrometry (Filadelfia Epilepsy Hospital, Dianalund, Denmark).

Statistical analysis
We included data for angry, fearful, and neutral faces and geometric shapes from the emotional faces paradigm.The data was analysed using a block design.Blocks of geometric shapes were collapsed and used to contrast each facial expression (e.g., BOLD response to angry faces versus BOLD response to shapes).We used linear regression models to evaluate differences in the amygdala response to angry, fearful, and neutral faces between SSRI and placebo groups.Age and sex were included as covariates.Using the same model, we performed analyses of group differences in whole-brain regional responses to angry, fearful, and neutral faces.To visualise these effects, we display regional standardised effects (Cohen's D) of SSRI mapped onto the set of voxels belonging to each respective region for a brain image in standard MNI space.As post-hoc analyses, we used the same statistical model to identify group differences on a voxel-wise level.www.nature.com/scientificreports/ We used linear regression models to evaluate whether amygdala responses to each emotional face were correlated with negative mood states, as measured with POMS on the same day, and whether SSRI modulated such potential correlation using an interaction term between amygdala responses and group.Age and sex were included as covariates.We computed post-hoc group-wise partial correlations between emotional amygdala responses and negative mood states with age and sex as covariates.
Across all tests, P-values are reported uncorrected (P unc ) and corrected for family-wise error rate (P FWER ) using the Bonferroni correction method.For the a priori amygdala analyses, we controlled for a family of three tests and the whole-brain analyses for a family of 88 tests.P-values were considered significant at P FWER < 0.05.Effects sizes are reported as Cohen's D. For the voxel-wise analysis, we used the AFNI-based 3dClustSim and 3dFWHMx functions to determine a cluster size unlikely to occur by chance at an uncorrected voxel-level threshold of P < 0.001, that maintains a type-I error rate of 0.05.For whole-brain and amygdala ROI-specific analyses, this was determined to be k > 211 and 2 voxels, respectively.
We used one-sample t-tests, including the whole sample, to examine whether the emotional faces paradigm induced a significant response in the amygdala.We used chi-square or two-sample t-tests to evaluate any differences between the groups on baseline measures, including sex, age, IQ, BMI, BDI-II, STAI, PSS and POMS, as well as POMS, censored volumes during fMRI, and participants' blinding at follow-up.For the emotional faces paradigm, accuracy rates were positively skewed, so we evaluated group differences using a two-sample permutation test with 10,000 Monte Carlo replications for each emotion.Group differences in reaction time for each emotion were assessed using t-tests.P-values are reported uncorrected (P unc ).

Sample characteristics
Participant demographics, psychometrics, serum escitalopram levels, and emotional amygdala responses as assessed with the emotional faces paradigm and its behavioural outcomes are reported in Table 1.BDI scores confirmed the absence of dysphoric or depressed mood (i.e., < 20; score range: active group = 0-9, placebo group = 0-12) 52 .STAI and PSS levels across groups were aligned with typical scores in a healthy cohort (STAI, score range: active group = 20-39, placebo group = 20-36; 44 PSS score range: active group = 2-20, placebo group = 1-19) 52 .The participants' mean IQ was in the normal upper range, i.e., higher than the average 100 IQ point of the general population, across both the active group (mean ± SD = 110 ± 8) and the placebo group (mean ± SD = 108 ± 7).Overall, there were no significant differences in baseline characteristics between the escitalopram and placebo groups (see Table 1 for group-wise descriptive data and estimate and P-values from statistical tests of group differences).

Treatment compliance and assessment of blinding
The serum escitalopram concentrations, as measured approximately halfway through the intervention period (mean ± SD: 12 ± 2 days) and at follow-up (mean ± SD: 26 ± 3 days), confirmed that all participants in the escitalopram group complied with the drug intervention.Although the fraction of participants in the escitalopram group who guessed that they got escitalopram was at a chance level (53%), the number was significantly higher than in the placebo group (χ 2 = 4.13, df = 1, P unc = 0.04).See Table 1 for descriptive data and estimate and P-values from statistical tests of group differences.

Amygdala responses to the emotional faces paradigm
Across the placebo and escitalopram groups, we observed a significant response of the amygdala to angry (mean ± SD: 0.333 ± 0.772, P unc = 0.001), fearful (mean ± SD: 0.316 ± 0.479, P unc < 0.001), and neutral faces (mean ± SD: 0.332 ± 0.670, P unc < 0.001) relative to geometric shapes.There was no evidence of significant differences between groups in accuracy or reaction time (see Table 1).Further, there were no significant group differences in censored volumes (P unc = 0.81).
We observed no statistically significant difference between the SSRI and placebo group on amygdala response to angry, fearful or neutral faces (angry: mean difference = − 0.10, SE = 0.47, P FWER = 1.00; fear: mean difference = − 0.05, SE = 0.12, P FWER = 1.00; neural: mean difference = − 0.09, SE = 0.17, P FWER = 1.00) relative to geometric shapes.Evaluation of sex differences showed that, across treatment groups, males had a significantly higher amygdala response to fearful faces than females (mean difference = 0.33, SE = 0.12, P FWER = 0.02).Otherwise, no sex differences were observed (angry: mean difference = − 0.01, SE = 0.21, P FWER = 1.00; neutral: mean difference = − 3.0, SE = 0.17, P FWER = 0.26).Effects of age were not observed either (angry: mean difference = − 0.01, SE = 0.02, P FWER = 1.00; fear: mean difference = 0.00, SE = 0.01, P FWER = 1.00; neutral: mean difference = − 0.01, SE = 0.01, P FWER = 1.00).See Table 1 for group-wise descriptive data as well as Cohen's D and P-values from the statistical test of group differences, Table S1 in the supplementary material for detailed estimates and corrected P-values from the statistical models assessing the effect of escitalopram on amygdala response to emotional faces, and Fig. 2 for individual and average amygdala responses to each emotion adjusting for the effect of sex and age.

Whole-brain responses to the emotional faces paradigm
Analyses of the effects of 3-5 weeks of intake of escitalopram on whole-brain regional responses revealed that responses to angry faces were significantly lower in the left opercular part of the inferior frontal gyrus for an illustration of the whole-brain region-wise effects of SSRI.See Table S3, S4, and S5 in the supplementary material for region-wise estimates and P-values from the statistical models.

Exploratory voxel-wise responses to the emotional faces paradigm
Across the whole-brain and within the amygdala, we did not observe any clusters showing a statistically significant effect of the SSRI intervention.
Table 1.Descriptive information for participants in each group along with group differences.Group differences in age, IQ, BMI, BDI-II, STAI, PSS, and POMS both at baseline and follow-up as well as reaction times in the emotional faces paradigm, were assessed using two-sample t-tests with estimate representing Cohen's D. Group differences in sex and blinding were assessed using the chi-square test with the estimate representing χ2.As accuracy rates were positively skewed, group differences were assessed using a two-sample permutation test with www.nature.com/scientificreports/

Negative mood states
There was no significant group difference in negative mood states at baseline or follow-up (see Table 1).We found no significant correlations between amygdala responses to angry, fearful, or neutral faces and negative mood states in the placebo group (angry, P FWER = 0.33; fear, P FWER = 0.12; neutral, P FWER = 0.07) or for a modulatory effect of escitalopram (angry, P FWER = 0.49; fear, P FWER = 0.29; neutral, P FWER = 0.19).For an overview of the results, see Table 2. See Table S2 in the supplementary material for an overview of all group-wise partial correlations.

Discussion
In 64 healthy individuals, we evaluated functional brain responses to angry, fearful, and neutral faces following 3-5 weeks of intake of either the SSRI escitalopram or placebo with a rigorous compliance protocol.This constitutes the largest such study to date in healthy individuals.Data was analysed using a whole-brain regionwise approach extracting standardised effects (i.e., Cohen's D).Compared to placebo, SSRI did not significantly affect amygdala responses to faces (relative to geometric shapes), a finding inconsistent with our hypothesis.However, whole-brain, region-wise analyses identified frontal and occipital brain regions wherein the response to negative faces (i.e., anger and fear) was lower but higher to neutral faces following SSRI intake compared to placebo.Our findings contrast with evidence for acute and short-term SSRI intake changing the amygdala response to emotional faces in healthy individuals but implicate that SSRI intake over a clinically relevant period modulates a broader set of brain regions, including the frontal and occipital cortex.The valence-specific reduction of response to negative emotions in cortical brain regions following SSRI may be involved in the clinical efficacy of SSRI.A primary region of interest in our analyses was the amygdala considering previous related studies 24,47 .We observed a pronounced and significant amygdala response to the emotional faces paradigm and a high degree of accuracy across both the SSRI and placebo groups, comparable to previous reports 8,35 .As such, the lack of a significant effect of SSRI on amygdala reactivity during emotion processing does not appear to stem from a weak or atypical neural response to the emotional faces paradigm.Two previous studies evaluating SSRI effects over a similar time frame also report a limited impact on the amygdala response to threat-related or happy faces 24,47 .However, in a sub-analysis, one of the previous studies reported that SSRI reduced amygdala response to threat in treatment-compliant participants, as confirmed by urinary drug concentration 47 .The authors suggested that the lack of a significant group difference could be attributed to non-compliance with the treatment.In our study, all participants were compliant with the treatment verified via serum SSRI concentration and self-report daily medication logbook, which suggests that non-compliance is not the cause of the non-significant results observed in our current study.Taken together, current evidence does not support that SSRI intake over a clinically relevant time period significantly modulates amygdala response to emotional faces across healthy populations.Future questions to be resolved include whether SSRI intake over a clinically relevant time period impacts other emotional processes involved in MDD, such as reinforcement sensitivity 62 .
Our whole-brain, region-wise analyses showed significantly lower brain responses to angry and fearful faces following SSRI intake compared to placebo in cortical regions, including frontal and occipital regions.Previous studies did not find a significant effect of 3 weeks' SSRI on emotion processing in cortical regions 24,47 , perhaps explained by their relatively smaller sample sizes.The frontal cortex has been identified as critical for face processing 5 and argued to exert top-down cognitive control of the amygdala 63,64 .This fronto-amygdala functional brain network is involved in regulating emotions 65 , and found to be compromised in MDD 66 .The observed SSRIinduced modulation of emotion processing in frontal regions may relate to the clinical efficacy of SSRIs.Regionspecific effects of sub-chronic SSRIs on emotion processing could stem from region-specific molecular changes in 5-HT tonus; a previous study in healthy individuals from our lab reported increased 5-HT tonus in some, but not all, regions following sub-chronic SSRI intake 67 .Future studies using positron emission tomography (PET) could identify the serotonergic receptors involved in this neuro-modulation of emotional face processing.
In contrast to our finding of a decreased brain response to angry and fearful faces, we observed significantly higher responses to neutral faces in frontal regions compared to placebo.This finding indicates that SSRI intake over a clinically relevant time period does not reduce brain responses to facial expressions per se but rather specifically to negative emotions (i.e., fear and anger), a valence-specific effect that may be associated with SSRIs' antidepressant properties 34 .As the modulatory effect of 5-HT seems to depend on emotional valence, we recommend future studies investigating serotonergic modulation of face processing to explore emotional face valence separately instead of collapsing them into an overall outcome of face processing, as done in some previous studies 38,39,47 .
In sum, our findings suggest that 3-5 weeks of SSRI intake valence specifically reduces brain response to negative faces, but in brain regions outside the amygdala, particularly in frontal regions.These results provide unique insight into possible mechanisms behind the clinical efficacy of SSRIs.However, we did not evaluate brain In post hoc analyses, we observed that effects on a voxel-wise level showed a similar pattern to those observed on a regional level, e.g., lower response to negative faces in frontal and occipital brain regions.However, no clusters remained statistically significant evaluating whole-brain effects nor focusing on the amygdala after correcting for multiple comparisons.We suspect that the apparent discrepancy between the effects of SSRI on a voxel-wise compared to regional level may stem from relatively limited statistical power for the voxel-wise analyses.

Methodological consideration
This study is not without limitations.The emotional faces paradigms have been shown to have low withinsubject test-retest reliability 68,69 .Although this mitigates the lack of a baseline scan in estimating drug effects, it nevertheless represents a noise source that may challenge our capacity to detect group differences.
Five participants dropped out due to side effects, of whom four received SSRI.Assuming that this group had an overall greater psychoactive response to SSRI, these missing data may have biased our results towards underestimating SSRI effects on amygdala reactivity.
In this study, we examined only brain responses per se.Future studies should investigate alternative modelling of these data, e.g., brain connectivity, which may capture relevant neural correlates associated with SSRI intervention 65 .
Participants receiving SSRI guessed their group assignment correctly at chance level, but the frequency of guessing SSRI group assignment was significantly higher than those in the placebo group (i.e., 16%).This contrasts with our previous study in healthy males with the SSRI fluoxetine, where nearly all participants guessed that they received a placebo 70 .The reason for the observed group discrepancy and whether it reflects an influential confound is not apparent, but its presence is notable.It is possible that perceptual differences influencing participants' guesses could be similarly associated with our observed exploratory group differences in frontal and occipital brain responses.
Our study population consisted of a North European sample of younger age, which limits our results' generalisability to other populations.Whether the results can be replicated in more diverse populations should be investigated in future studies.
Table 2.The models explore whether emotional amygdala responses is correlated with negative mood states and whether this correlation is modulated by 3-5 weeks' SSRI intake, controlled for covariates age and sex.Sex is the marginal effect of men relative to women.The amygdala response to angry, fearful and neutral faces (contrasted with geometric shapes) are assesed using BOLD fMRI.Mood disturbances are measured with the self-report questionnaire Profile of Mood States (POMS), with higher values indicating more disturbance.Estimate is the unstandardised beta.Abbreviations SE B, standard error for unstandardised beta; Punc, unadjusted significance level; SSRI, selective serotonin reuptake inhibitors.All P-values were corrected for family-wise multiple testing using the Bonferroni method (i.e.all P-values multiplied by three), displayed under PFWER.www.nature.com/scientificreports/

Estimate
In this study, we evaluated brain responses and related mood changes at a single time point.Future studies evaluating brain responses at multiple time points during SSRI intake could probe the trajectory of the drug's effects on brain response to emotional material, which may provide a more nuanced perspective on associated brain effects.

Conclusion
We did not find evidence that 3-5 weeks of intake of SSRI significantly affected amygdala response to threatrelated or neutral faces in a large cohort of healthy individuals.However, whole-brain analyses revealed significant effects in the frontal and occipital regions.These results implicate serotonergic modulation of top-down emotional processing, a modulation that may be involved in SSRIs' clinical efficacy.

Figure 3 .
Figure 3. Illustration of standardised effect sizes (i.e., Cohen's D) for regional effects of SSRI vs placebo on brain responses to emotional faces (i.e., angry, fearful and neutral faces constrasted with geometric shapes) in 64 healthy individuals.The regional Cohen's D is displayed for each voxel within each of the 90 AAL regions, i.e., cortical and subcortical regions.Blue colours indicate negative values, i.e., lower response in the SSRI group, and red/yellow colours indicate positive values, i.e., higher response in the SSRI group.FSLeyes coordinates of brain slices across emotions: y = − 8.3, x = − 3.6, z = 0.7.
10,000 Monte Carlo replications for each emotion, with the estimate representing Cohen's D. Group differences in amygdala response to angry, fearful and neutral faces were assessed using linear regression models with age and sex as covariates with estimate representing Cohen's D.