Journal Pre-proof Single-dose effects of Citalopram on neural responses to affective stimuli in borderline personality disorder: A randomized clinical trial

1 Background: Psychiatric medication that has a soothing effect on the limbic 2 responses to affective stimuli could improve affective instability symptoms as observed 3 in Borderline Personality Disorder (BPD). The objective of this study was to investigate 4 whether Citalopram vs. Placebo reduces the response of the affective neural circuitry 5 during emotional challenge. 6 Methods: N=30 femaie individuals with BPD diagnosis participated in a 7 Placebo-controlled, double-blind crossover trial design. Three hours after oral drug 8 intake, individuals with BPD viewed affective pictures while undergoing functional 9 Magnetic Resonance Imaging. Blood Oxygenation Level Dependent responses to 10 images of negative affective scenes and faces showing negative emotional expressions 11 were assessed in regions-of-interest (amygdala, anterior cingulate cortex, anterior 12 insula, dorsolateral prefrontal cortex). Blood perfusion at rest was assessed with arterial 13 spin labeling. Results: The neural response to pictures showing negative affective scenes was 15 not significantly affected by Citalopram (N=23). Citalopram significantly reduced the 16 amygdala response to pictures of faces with negative affective expressions (N=25, 17 treatment difference left hemisphere: -0.06 ±0.16, P<0.05, right hemisphere: - 18 0.06±0.17, P<0.05). We observed no significant effects of Citalopram on the other 19 regions. The drug did not significantly alter blood perfusion at rest. Conclusions: Citalopram can alter the amygdala response to affective stimuli in 21 BPD, which is characterized by overly responsive affective neural circuitry.


Introduction
1 Pharmaceutical compounds that engage affective brain circuits are promising candidates 2 for treating affective instability in Borderline Personality Disorder (BPD) (1). 3 Hyperreactivity of the amygdala and hyporeactivity of dorsolateral prefrontal cortex 4 (DLPFC) characterize neural emotion processing in BPD(2). Individuals with BPD 5 often use damaging self-regulation strategies such as non-suicidal self-injury and 6 dissociation to soothe highly aversive emotional states -an effect that may be mediated 7 by downregulation of the amygdala(3-6). A previous literature review identified brain 8 regions such as the amygdala, insula, dorsal anterior cingulate cortex as well as 9 prefrontal areas as promising neural targets for treatment of emotion dysregulation in 10 BPD(7). Treatment of choice for this disorder is psychotherapy(8), and clinical trials 11 found decreased responding of the amygdala after effective psychotherapy(9,10). These 12 and other studies(4,5,11,12) suggest a link between affective instability symptoms and 13 dysregulation of prefrontal-limbic brain circuits. Assuming a causal pathway from the 14 brain to behavior, the question is pressing whether medication can alter the neural 15 circuits of affective processing in BPD. Thereby, it could be possible to ameliorate 16 symptoms of affective instability. Evidence is currently lacking to show effective neural 17 modulation in BPD with existing pharmaceuticals. 18 antidepressants in the treatment of patients with BPD, although randomized controlled 1 trials are lacking to support this choice(20,21). Meta-analyses found little evidence for 2 effectiveness of antidepressants on BPD symptoms and no significant effect could be 3 detected for SSRIs (22). However, only little evidence exists and trials with Citalopram 4 are currently missing. 5 Neuroimaging studies in healthy participants show that Citalopram can reduce 6 the amygdala response to affective material(23-26). If it would do the same in 7 individuals with mental disorder, Citalopram could act upon limbic hyperreactivity and 8 consequently on affective instability symptoms in BPD. It is the hope that first evidence 9 of neural target engagement can inform future clinical trials to assess 10 psychopharmaceutical compounds in the treatment of BPD. 11 We conducted a pharmacological (ph) functional magnetic resonance imaging 12 (fMRI) study to investigate neural responses to pictures with negative affective content 13 in female patients with BPD within a double-blind, randomized crossover placebo-14 controlled design. The main purpose of this study was to assess the immediate effect of 15 The two tasks were administered in fixed order, as introduced below, and were 2 presented with Presentation software (Neurobehavioral Systems, Inc, Berkeley, USA) 3 via a 40" monitor located in the back of the scanner, which was visible for subjects 4 through a mirror placed on top of the head coil. The patient operated a button box with 5 the right hand to record behavioral responses. 6

Faces task 7
Participants viewed faces with emotional expression (disgust, sadness, and fear were 8 chosen based on meta-analyses(36)) from the Warsaw Set of Emotional Facial 9 Expression Pictures (WSEFEP, http://www.emotional-face.org/). A block design of 12 10 blocks with 6 faces each (aversive condition, AC; negative emotional expressions were 11 randomly mixed within blocks) and 12 blocks with scrambled faces (neutral condition, 12 NC) was used. Scrambled faces were chosen as control because of two reasons: First, 13 we wanted to match the faces task with the scenes task in terms of the analyzed contrast. 14 Second, previous work suggested altered responding in BPD not only to emotional 15 expressions but also to faces with neutral expression (37), which would compromise the 16 sensitivity of our design to detect drug-induced changes. In sum, 72 negative faces of 24 17 actors (12 female, 12 male) were shown for 3 seconds each. The inter-trial interval was 18 jittered to nine, 10, or 11 seconds and contained a white fixation cross on a black 19 background. To ensure attention, participants were asked to press a button to indicate 20 for every picture whether the person was male or female (AC), or whether color of the

Scenes task 1
The task was adapted from Paret et al. (38). We presented 42 pictures from the OASIS 2 picture set (39) to induce negative affect. We used pictures with negative affective 3 valence and high arousal (AC) in a block-design. During each of 14 blocks, lasting 18 4 seconds, three picture stimuli were presented for 6 seconds each, resulting in a set of 42 5 negative pictures in total. Due to the within-subject design, we used two picture sets 6 with similar characteristics concerning affective valence and arousal to avoid 7 habituation to picture content. These two sets were randomized between treatment visits 8 to avoid undesired effects of systematic presentation order. Scrambled pictures were 9 used in a non-affective control condition (NC) with the same number of trials and 10 presentation time. During the intertrial interval (jittered to nine, 10, or 11 seconds), 11 participants viewed a white fixation cross on a black background. To ensure attention, 12 participants were asked to press a button to indicate for every picture whether it showed 13 a person or not (AC), or whether the color of the bounding box around scrambled 14 pictures was blue or green (NC All imaging preprocessing and first-level analyses were carried out using FEAT (FMRI 9 Expert Analysis Tool) Version 6.00, part of FSL (FMRIB's Software 10 Library, www.fmrib.ox.ac.uk/fsl)(41). The following preprocessing steps were 11 performed: volume realignment to correct for participant head motion, B0 unwarping 12 using field map data to correct for EPI distortions, grand-mean scaling, and 13 spatial smoothing with a 5mm FWHM kernel. Next, FSL's melodic was applied to 14 extract independent data components and ICA-AROMA(42) was applied to identify and 15 remove secondary effects of head motion. Finally, a temporal high-pass filter with 0.01 16 Hz cut-off was applied to remove scanner drifts. We obtained the transformation of the 17 fMRI data to the participant's high resolution T1 anatomical space using FSL's 18 Boundary-Based Registration tool. A transformation from the participant's T1 space to 19 MNI152 standard space was obtained using linear alignment via FSL FLIRT with 12 20 degrees of freedom, and subsequently refined using non-linear steps as implemented 21 in FSL FNIRT. Data were screened for quality and excluded from further analysis in 22 case of superthreshold movement during a scan (>4mm, N=2 patients) and heavy 23 BOLD-signal dropout in one scan (N=1). More information on the composition of the 1 sample to be analyzed can be obtained from Figure 1. 2 After preprocessing, we conducted first-level statistical analyses for both the 3 Faces and the Scenes task separately. For each task, we included two regressors 4 respectively modelling the onset times of the faces/scenes (AC) and scrambled stimuli 5 (NC), convolved with a double-gamma HRF. The onset regressors consisted of 18-6 second blocks. The contrast of interest compared BOLD activity between the 7 scenes/faces and the scrambled control stimuli. 8 To show target engagement by the tasks, we prepared whole-brain maps from a 9 mass-univariate whole-brain analysis implemented in Statistical Parametric Mapping 10  Figure S1). To 1 test for the effect of Citalopram versus Placebo we derived p-values based on 2 permutation analyses. Specifically, we compared the average within-participant 3 difference between compounds (Citalopram -Placebo) with a distribution of 4 randomized within-participant differences. The effect of compound was deemed 5 significant if the true compound difference was smaller than 5% of the randomly 6 calculated differences. Random differences were obtained by within-participant 7 compound randomization, randomly re-assigning the compound to the two sessions and 8 calculating the difference score. This was repeated 10.000 times per participant, 9 yielding a distribution of 10.000 average differences across participants which was used 10 to assess the significance of the true difference (alpha=p<0.05). Complementary voxel-11 wise analysis within ROIs was conducted using FWE-correction with Small Volume 12 Correction in accordance with the original study protocol (Supplementary Material 2). 13 To follow up the results, we explored whether the neural response to Citalopram 14 would correlate with baseline symptom severity (i.e., BSL-23 score). 15 Analysis of ASL data is reported in the Supplement section 3. 16 stimuli (Supplement , Table S3). We did not observe any significant differences in 1 response accuracy between treatments in the scenes task (difference per condition 2

Functional neuroimaging
10 Whole-brain analyses of activated voxels showed robust activation in all four 11 ROIs in the scenes-task (Supplement, Figure S4 for illustration). In the faces-task, we 12 observed activation of the amygdala and the DLPFC, too, while no activation was 13 observed in the insula and the ACC (Supplement, Figure S5). 14 Testing our hypothesis that a single dose of 20 mg Citalopram as compared to 15 Placebo results in changes in brain activity during both fMRI tasks, we observed no 16 baseline and neural response were consistently negative, although modest and not 1 significant (Supplement , Table S4). Finally, we did not find differences between 2 Citalopram and Placebo treatment in blood perfusion as assessed with ASL 3 (Supplement, Figure S6, Table S5). 4

5
A single dose of Citalopram vs. Placebo reduced the amygdala response to emotional 6 faces as compared to scrambled images in individuals diagnosed with BPD. Neural 7 modulation by the compound was restricted to the amygdala, whereas Citalopram did 8 not significantly affect other ROIs involved in emotion and emotion regulation. 9 Differently than expected, we did not find reduced neural response to affective scenes 10 (primary endpoint), and we did not observe altered amygdala blood perfusion at rest. 11 These findings partially corroborate immediate alteration of limbic responding by 12 Citalopram as reported previously(23-26). For the first time, we could demonstrate that 13 this effect extends to BPD. The signifcant effects are limited to analyses of secondary 14 introspection. The study was not designed to detect potential effects of Citalopram on 1 BPD symptoms. 2 We found reduced amygdala response to faces with negative affective 3 expression, but not in response to pictures with scenes of negative affective content. 4 Most previous Citalopram phfMRI studies used face stimuli to probe modulation of 5 affective response (see references below). We selected neural responses to scene images 6 as primary endpoint, because this type of stimuli, like face stimuli, was also widely 7 applied in BPD-fMRI work(2). It is interesting that we did not observe similar 8 Citalopram effects with two frequently used affective stimulation paradigms from 9 psychiatric neuroscience. 10 An unplanned follow-up analysis was done to elucidate potential influences 11 from BPD symptom severity on the treatment response. Although not significant, across 12 all a-priori ROIs, patients with higher symptom severity differentiated less between 13 Citalopram and Placebo. Future investigations with larger sample sizes are necessary to 14 study how parameters of interest such as symptom severity moderate the Citalopram 15 response. 16 In comparison to the face stimuli, the scenes were more diverse and complex. 17 Furthermore, six faces were presented during a trial, three seconds each, which was 18 twice the number and duration of scene-images. That is, the frequency with which 19 salient picture information changed was higher in the faces-task compared to the scenes-20 task. Descriptively, we found overall smaller effects for amygdala responses in the main 21 effect for stimulus material (AC versus NC) when directly comparing the scenes-task 22 vs. faces-task. Consequently, smaller effects within the scenes-task might have reduced 23 the likelihood to detect differences between the Citalopram condition and the Placebo 24 condition in the scenes-task. Not all studies used placebo control(24) and only one other study reported double-18 blinding(46). Critically, methods for significance testing greatly differ between trials, 19 and some studies assess response to different stimulus categories separately such as 20 angry and fearful emotional expression, and in several ROIs, while they do not report 21 control for type one error(23,24,26). In light of this, critical interpretation of our 22 findings is demanded, because our study suffers from similar shortcomings, given the 23 number of statistical tests conducted for two tasks and several ROIs. The literature can 24 gain from future trials who also preregister their hypothesis and analysis plan.
We used ASL to compare absolute blood perfusion after Citalopram vs. Placebo 1 treatment and did not find significant differences. This finding is in accordance with a 2 previous study in healthy participants(48). The ASL sequence was acquired to quantify 3 overall perfusion in absence of an ongoing task. Of note, the goal of this was to 4 investigate the effect of Citalopram on perfusion in the amygdala and not to investigate 5 the effect of Citalopram on perfusion during task performance. 6 Due to our sample composition, conclusions are limited to female participants 7 only. Furthermore, the small sample size precluded detection of small/moderate effects 8 of Citalopram. 9 In conclusion, Citalopram can immediately act on amygdala processing of 10 emotion in BPD, but corroboration by future studies is needed.    Table 2. Comparison of brain response to A) scene pictures vs. scrambled pictures 1 (primary endpoint) and B) face pictures vs. scrambled pictures (secondary endpoint). 2 Percent signal change is reported per region of interest (M±SD). Statistically significant 3 results (P<0.05) are marked with an ansterisk. ACC=anterior cingulate cortex, A 4 insula=anterior insula, DLPFC=dorsolateral prefrontal cortex.