Randomized clinical trial shows no substantial modulation of empathy-related neural activation by intranasal oxytocin in autism

Evidence suggests that intranasal application of oxytocin facilitates empathy and modulates its underlying neural processes, which are often impaired in individuals with autism spectrum disorders (ASD). Oxytocin has therefore been considered a promising candidate for the treatment of social difficulties in ASD. However, evidence linking oxytocin treatment to social behavior and brain function in ASD is limited and heterogeneous effects might depend on variations in the oxytocin-receptor gene (OXTR). We examined 25 male ASD patients without intellectual disability in a double-blind, cross-over, placebo-controlled fMRI-protocol, in which a single dose of oxytocin or placebo was applied intranasally. Patients performed three experiments in the MRI examining empathy for other’s physical pain, basic emotions, and social pain. All participants were genotyped for the rs53576 single-nucleotide polymorphism of the OXTR. Oxytocin increased bilateral amygdala responsiveness during the physical pain task for both painful and neutral stimuli. Other than that, there were no effects of oxytocin treatment. OXTR genotype did not significantly interact with oxytocin treatment. Our results contribute to the growing body of empirical literature suggesting heterogenous effects of oxytocin administration in ASD. To draw clinically relevant conclusions regarding the usefulness of oxytocin treatment, however, empirical studies need to consider methods of delivery, dose, and moderating individual factors more carefully in larger samples.

months. Eventually, of a total of 41 patients screened, 28 were randomized and data from 25 patients could be analyzed.

Patient recruitment and inclusion/exclusion criteria
Patients were recruited from January 2015 to June 2017 at the Outpatient Clinic for Autism Spectrum Disorders at Marburg University Hospital in Marburg, Germany, where they had received a diagnosis of ASD prior to study participation. Eligible patients and their families were contacted through an invitation letter including brief information on the study. We invited male patients aged 15 to 35 years who had previously partaken in a molecular genetic study at the same study site, during which they had been genotyped for the rs53576 SNP of the OXTR. All were German native speakers. Concurrent psychopharmacological treatment was not an exclusion criterion, however, patients receiving medication (antidepressants, n = 1) were required to keep the dosage constant for the entire period of data acquisition. Principal exclusion criteria were a body mass index less than 18 or higher than 30, verbal IQ below 70 as measured with the Wechsler Intelligence Scale for Children 3 , traumatic lesions of the brain, severe neurological disorders (e.g. epilepsy), metal implants (MRI contraindication), acute suicidal tendency according to the clinical impression, known metabolic or endocrinological disorders, cardiac disorders (assessed through anamnesis, electrocardiogram, blood pressure and heart rate measurements), hypersensitivity to nasal sprays or other drugs, and comorbid drug, alcohol or nicotine (more than 15 cigarettes per day) abuse or dependence.

Randomization and blinding
Randomization (simple randomization) was carried out independently from the recruitment process and data acquisition by the Coordination Center for Clinical Studies at Philipps-University (KKS) Marburg, which created a list of subject numbers ("randomization numbers") for each genotype subgroup (carriers vs. non-carriers of the risk allele). Each number was thereby randomly assigned to an administration sequence (oxytocin first, placebo first). Blinding was carried out by an independent pharmacist at the Clinic Pharmacy of Heidelberg University Hospital, who received the complete list of randomization numbers and corresponding administration sequence from the KKS Marburg. Oxytocin and placebo were filled into spray bottles of identical appearance. Two bottles (one with oxytocin and one with placebo) were sealed together in a secondary packaging (foil). Secondary packaging and spray bottles were each provided with a specific randomization number. The bottles were additionally labeled with the sequence of administration (1 or 2). The packaged study drugs were sent to the trial center at the Department of Child-and Adolescent Psychiatry of Philips-University Marburg, where the patients were enrolled. After a study physician had obtained the patient's informed written consent, the KKS was contacted, which then assigned a randomization number to the patient. All investigators engaged in recruitment, data acquisition (including study nurses, study physicians and research assistants), and data analysis, as well as the patients, were blinded to the allocation arm, i.e. the administration sequence.

Procedure during study visits
MRI scanning took place at the brain-imaging unit at the Department of Psychiatry at Phillips-University in Marburg, Germany. Patients were invited to an initial visit that included study information, a physical examination, and confirmatory diagnostic procedures regarding the ASD diagnosis. Patients who met all inclusion criteria were then invited to two MRI sessions. These sessions were planned to be 7 days apart, which was achieved for most patients (n = 17). If this was not possible, the second session took place within 21 days after the first session. Patients were instructed not to eat, smoke or drink excessive amounts of water two hours before administration of the nasal spray. In the beginning of the first MRI session, contraindications for oxytocin (cardiac arrhythmia) were excluded by performing an auscultation of the heart and an ECG. During both MRI sessions, each participant's medical history, current health status, heart rate and blood pressure were assessed. Following the medical examination, a study physician administered the nasal spray to the patients. Six puffs per nostril (24 IU) were administered to patients aged 18 years or over, and four and five puffs (18 IU) were administered to the right and left nostril, respectively, in patients aged 15-17 years. The nasal spray contained either oxytocin (Syntocinon-Spray®; each puff containing 2 IU of oxytocin) or placebo (containing all ingredients except for the peptide). The placebo nasal spray was identical to the oxytocin nasal spray in appearance and smell.
Possible adverse effects were assessed using the Multidimensional Mood Questionnaire 4 , the Self-Assessment Manikin 5 , and a questionnaire asking about any subjectively experienced side-effects, as well as by monitoring heart rate and blood pressure three times before and once after the MRI measurement. Patients did not report any side effects during this trial. For analyses of treatment effects on mood and affective states please see Supplementary Tables S7 and S8.

fMRI paradigms
In the current study, we used three experiments examining the neural correlates of sharing others' mental and affective states. Two of these experiments, the physical and social pain experiment, have been previously described and applied in samples of autistic males without intellectual disabilities [6][7][8][9][10] .
In the physical pain experiment, 28 digital color photographs were presented, depicting another person's left or right hand or foot from a first-person perspective in either painful (physical pain, PP) or nonpainful (physical neutral, PN) situations 11 . Painful and neutral stimuli were matched in number as well as for semantic content and luminance. In each trial, patients were instructed to look at the photograph for 4.5 seconds and to rate the intensity of pain that the depicted person would experience in the respective scenario (1 = no pain, 5 = very strong pain). A fixation cross followed the rating phase of 3 seconds for an average of 6.1 seconds.
In the basic emotions experiment, 30 digital color photographs were presented 12 , showing a male or female face with a happy (HAP), sad (SAD) or emotionally neutral (EN) expression (10 photographs per stimulus category). Patients were instructed to look at each stimulus for 4.5 seconds and to rate how the depicted person felt (1 = very sad, 5 = very happy). A fixation cross followed the rating phase of 3 seconds for an average of 6.1 seconds.
In the social pain experiment, overall 36 validated hand-drawn sketches were presented, each displaying a marked protagonist in an either potentially embarrassing (social pain, SP, 12 sketches) or neutral public scenario (social neutral, SN, 12 sketches; another 12 sketches displaying a protagonist in an inappropriate scenario that he/she is not aware of 13 were not included in the current analysis).
The sketches were presented together with a brief description of the situation. Participants were instructed to look at each sketch for 12 seconds and to subsequently rate the intensity of the embarrassment the protagonist would feel in the respective scenario (1 = no embarrassment, 5 = very strong embarrassment). A blank screen with a fixation cross (1 second) was interleaved between the presentation of the sketch and the rating period (3 seconds), which was followed by an 8-second lowlevel baseline separating the trials.

Physical Pain: second-level analysis
For each subject and session, we included three regressors modeling the hemodynamic responses to physical pain (PP) and the corresponding neutral (PN) stimuli, as well as the rating period, with durations as described above. Six parameters modeling head motion on a scan-to-scan basis were added as regressors to account for noise. The resulting individual β-maps of activation during PP and PN were used in the analysis on the group level. Here, we defined a random-effects GLM with two factors (2x2) to investigate effects of treatment (oxytocin, placebo as a within-subject factor) and stimulus category (PP, PN as a within-subject factor) on activation differences. We included the order of treatment (oxytocin first, placebo first) as a between-subject factor to control for possible order effects.

Basic Emotions: second-level analysis
On the individual subject level, the GLM included four regressors for the sad (SAD), happy (HAP) and neutral (EN) facial stimuli, as well as the rating period, with durations described above. Again, six parameters modeling head motion on a scan-to-scan basis served as additional regressors to account for noise. The resulting β-maps of activation during SAD, HAP and EN were used in the analysis on the group level. Here, we defined a random-effects GLM with two factors (2x3) to investigate effects of treatment (oxytocin, placebo as a within-subject factor) and stimulus category (SAD, HAP, EN as a within-subject factor) on activation differences. Again, the order of treatment was included to control for possible order effects.

Social Pain: second-level analysis
On the individual subject level, the GLM included three regressors for social pain (SP) and neutral (SN), as well as the rating period, with durations described above. Again, six parameters modeling head motion on a scan-to-scan basis served as additional regressors to account for noise. The resulting β-maps of activation during SP and SN were used in the analysis on the group level. Again, we defined a random-effects GLM with two factors (2x2) to investigate effects of treatment (oxytocin, placebo as a within-subject factor) and stimulus category (SP, SN as a within-subject factor) on activation differences. Again, the order of treatment was included to control for possible order effects.

Additional analyses on genotype-treatment interactions
While effects of genotype and treatment were examined on absolute stimulus ratings and brain activation compared to a low-level baseline, we were also interested in potential genotype-treatment interactions on differential responses to the shown stimuli. That is, genotype might moderate the influence of treatment on the difference in ratings between neutral and painful/emotional stimuli, and the difference in corresponding brain responses. To increase sensitivity, we restricted our analyses to predefined regions of interest (ROI).
To examine these potential effects on the brain level, we computed β-images contrasting the painful or emotional stimuli to the respective neutral stimuli for each experiment, i.e. physical pain > physical neutral, happy faces > neutral faces, sad faces > neutral faces, and social pain > social neutral. For each contrast, we defined a random-effects GLM including treatment as a within-subject factor, resulting in four separate general linear models examining activation differences as a function of treatment. In a next step, we extracted averaged contrast estimates within the ACC, anterior insula, and amygdala to obtain an estimate of global activation of the entire ROIs. This was done separately for the oxytocin and placebo session within each of the four contrasts. The number of risk alleles was coded in a linear fashion (AA<GA<GG) and correlated with contrast estimates after oxytocin and placebo administration, using Spearman's rho as a nonparametric measure. The correlation coefficients for oxytocin and placebo were then tested for significant differences 14 . Significantly different correlations of genotype and brain activation depending on the treatment condition would suggest a significant genotype × treatment interaction.
We used a parallel approach to examine these effects in stimulus ratings. First, we calculated individual differences in ratings for neutral and painful/emotional stimuli separately for the oxytocin and placebo session. Again, the number of risk alleles was coded in a linear fashion (AA<GA<GG) and correlated (Spearman's rho) with rating differences after oxytocin and placebo administration. For each rating difference, the correlation coefficients after oxytocin and placebo administration were then tested for differences 14 .
In an exploratory approach, we also used this method to examine potential interactions of symptom severity (ADOS severity score) and treatment. Results of all analyses are reported in Supplementary Tables S1-S4.       Note. Results of the contrast Physical Pain > Physical Neutral in the whole brain, model accounting for genotype. All statistics are family-wise error (FWE) corrected for whole-brain analyses at the voxel level. The "Cyto Area"-column indicates the cytoarchitectonical area as assigned by the SPM Anatomy toolbox v2.2b 20 if available. Anatomical labels were derived respectively.    Table S9. Mean blood pressure and heart rate before and after substance administration

SUPPLEMENTARY TABLES
Note. Means and standard deviations (in brackets) for blood pressure and heart rate data. Blood pressure and heart rate were measured four times during study visits. T0 = baseline measure at the beginning of the visit, T1 = immediately after substance administration, T2 = 10-20 minutes after substance administration, prior to MRI scanning, T3 = ~90 minutes after substance administration, immediately after MRI scanning. Effects of treatment and time were examined using two-  T0  T1  T2  T3  T0  T1  T2     did not change the magnitude of these task and treatment effects, as can be seen from the large overlap in the statistical maps ("overlap", purple). "No change in effects" was defined as (a) no change in significance (no new significant clusters with k > 5, no loss of significant clusters with k > 5 compared to Figure S3. Basic Emotions: Comparison of significant effects across different models of brain activation controlling for age, dose, and treatment order. In our main analyses, we used a general linear model that included treatment (oxytocin, placebo) and stimulus category as within-subject factors, and genotype (AA<GA<GG) as a covariate ("original model", red). Faces induced widespread activity within occipital regions, including the fusiform and occipital face area, as well as inferior frontal regions, anterior insula, and medial cingulate cortex. There were no statistically significant effects of oxytocin treatment. Controlling for age (green), dose (blue), and order of treatment (pink) did not change the magnitude of the task effect, as can be seen from the large overlap in the statistical maps ("overlap", purple). "No change in effects" was defined as (a) no change in significance (no new significant clusters with k > 5, no loss of significant clusters with k > 5 compared to original model); (b) no change in location and cluster size (significant clusters have roughly same size [±10% of total voxels] and location, as indicated by visual inspection). Results are displayed at voxellevel p(FWE) < .05 across the whole brain. Brain images were created with MRIcroGL (https://www.nitrc.org/projects/mricrogl/). Figure S4. Social Pain: Comparison of significant effects across different models of brain activation controlling for age, dose, and treatment order.
In our main analyses, we used a general linear model that included treatment (oxytocin, placebo) and stimulus category as within-subject factors, and genotype (AA<GA<GG) as a covariate ("original model", red). Socially painful situations, compared to neutral situations, elicited significant activation in posterior temporal regions, posterior medial frontal gyrus, thalamus, inferior frontal gyrus, and supramarginal gyrus. There were no statistically significant effects of oxytocin treatment. Controlling for age (green), dose (blue), and order of treatment (pink) did not change the magnitude of the task effect, as can be seen from the large overlap in the statistical maps ("overlap", purple). "No change in effects" was defined as (a) no change in significance (no new significant clusters with k > 5, no loss of significant clusters with k > 5 compared to original model); (b) no change in location and cluster size (significant clusters have roughly same size [±10% of total voxels] and location, as indicated by visual inspection). Results are displayed at voxel-level p(FWE) < .05 across the whole brain. Brain images were created with MRIcroGL (https://www.nitrc.org/projects/mricrogl/).