Participants
Nineteen male, right-handed adults with ASD and 19 typically developed (TD) control participants were included in the study (age: ASD mean= 30, SD= 11, TD mean= 27, SD= 9). Two ASD cases and 2 TD controls were excluded from the Go/No-Go task due to significant head movement, leaving a sample of 17 ASD cases and 17 TD controls for the Go/No-Go task, and 19 ASD cases and 19 TD controls for the sustained attention task. The sample size was chosen based on results from our prior experiments targeting serotonin modulation using acute tryptophan depletion(9, 26), which were successful in detecting group differences in BOLD response with sample sizes of n=14. This implies an effect size (expressed in Cohen’s d) in excess of 1.2(9, 26). Exclusion criteria included medical disorders that could influence cognitive performance, major mental illnesses other than ASD, genetic disorders associated with ASD, alcohol or substance dependence or taking any medication affecting the serotonergic system (e.g. antidepressants, antipsychotics, benzodiazepines or mood stabilizers). The ASD diagnoses were made by a consultant psychiatrists using ICD-10 research criteria(27) and confirmed using the Autism Diagnostic Interview–Revised (ADI-R)(28) if an informant was available. Current autistic symptoms were measured by the Autism Diagnostic Observation Schedule (ADOS)(29). Intelligence was measured by the Wechsler Abbreviated Scale of Intelligence test (WASI)(30). All participants completed baseline self-reported questionnaires of autistic traits (Autism-Spectrum Quotient)(31), obsessionality (Obsessive-Compulsive Inventory-Revised)(32), and current symptoms of ADHD (Barkley Adult ADHD Rating Scale—IV)(33). Symptoms of anxiety and depression were assessed using The Hamilton Rating Scales for Depression(34) and Anxiety(35). All participants gave written, informed consent after receiving a complete description of the study. The study had National Research Ethics approval following review by the Stanmore Ethics Committee, London, United Kingdom.
Tianeptine administration procedure
Participants were required to complete two scanning sessions: one after receiving a single dose of 12,5mg of encapsulated tianeptine and one after receiving a dose of encapsulated placebo (ascorbic acid), in a randomized, double-blind, crossover design. A list of blinding numbers were produced independently and passed directly to the pharmacy in the outpatient department of the Maudsley Hospital, South London & Maudsley NHS Trust, London UK, using a computerised random number generator with blocked randomisation. The pharmacy used these numbers to blind each dose (placebo; tianeptine) as they were encapsulated. Both subject and researcher(s) were blind to dosing throughout data acquisition. The randomisation and encapsulation was conducted according to Good Medical Practice and in accordance with CONSORT & SPIRIT guidelines. Each dose was given to the participant 1h prior to scanning, as tianeptine reaches its peak plasma level after approximately 1h(36). There was a minimum of eight days between the scans to allow for complete washout of the drug (t½=3h; washout=5*t½=15h). All participants received a screening by a medical doctor before and after the administration of both doses.
Visual analogue scale
All participants completed self-report visual analogue scale (VAS) questionnaires prior to drug administration and after the MRI scan. Side-effects potentially associated with tianeptine were measured, including palpitations, nausea, dizziness, attentiveness, anxiety and irritability.
Go/No-Go inhibition fMRI task
In order to probe the brain’s response inhibition system, participants engaged in a Go/No-Go task (GNG) during each scanning session(8, 37). During this task, participants made either a motor response on a button box to Go signals or inhibited this response to No-Go signals. In this task, arrows appear pointing to either the left or right side of the screen. The participant responds by pressing the left or right button as fast as possible on a diamond-shaped keypad. Infrequently (12%), arrows pointing to the top (No-Go signals) appear. Subjects have to inhibit any motor response to these stimuli. In 12% of trials, slightly slanted (45 degrees) arrows pointing left or right (oddballs) appear and subjects have to respond as fast as they can, in the same way as for Go signals. No-Go responses were compared to successful Oddball trials(8, 37). There are two reasons we used the oddball instead of Go trials for the comparison. Firstly, this was done in order to control for the oddball effect of the No-Go trials. The No-Go trials are different from the Go trials and appear with less frequency, eliciting the so-called “oddball” attention effect. Participants pay more attention to rare stimuli than to high frequent stimuli. Hence, the No-Go trials in addition to measuring inhibition, also measure attention allocation to oddball stimuli. Furthermore, in order to control for this effect we added the oddball stimuli and contrasted No-Go with these oddball trials. Secondly, the Go trials appear with higher frequency than the No-Go trials. Hence, the oddball trials furthermore allow us to compare the same amount of No-Go vs oddball ('Go') stimuli.
Sustained Attention fMRI task
In order to probe the brain’s sustained attention network system, the Sustained Attention task (SAT) was performed during each scanning session(7, 12, 13). In this task, participants need to respond via a right hand button response as quickly as possible (i.e. within 1s) to the appearance of a visual timer counting up in milliseconds. When they press the button the counter show their reaction time in milliseconds. The visual stimuli appear either after short, predictable consecutive delays of 0.5s (260 stimuli in total), in series of 3–5 consecutive stimuli or after unpredictable time delays of 2, 5 or 8s (20 each), which are pseudo-randomly interspersed into the blocks of 3–5 delays of 0.5s. The long, infrequent, unpredictable delays place a higher load on sustained attention, as participants have to wait for them to occur and they do not know the exact time when they will occur (2s, 5s or 8s) - whereas the short, predictable 0.5s delays appearing in a row are typically anticipated. Participants learn to estimate the 0.5s and know that there will be several stimuli appearing in a row(38), placing a higher demand on sensorimotor synchronization(12).
We have previously consistently shown with this task that sustained attention networks are activated during the long relative to the short delays with progressively increasing activation in these networks from 2s to 8s (7, 12, 13). Here, we only report on the longest delay that elicits the strongest sustained attention activation, i.e. 8s vs 0.5s delays.
Baseline characteristics and task performance statistical analyses
Statistical tests were performed using the SPSS software (v23.0)(39). T-tests were used to compare baseline characteristics between groups and multivariate analysis of variance (MANOVA) determined any differences in performance and visual analogue scale outcome measures between group and drug conditions. Analysis of variance (ANOVA) was used to compare largest displacement in head movement between group and drug conditions.
For the GNG task the performance measures included: probability of inhibition (main inhibitory measure), mean reaction time to the Go signal (motor execution measure) and mean reaction time to the oddball signal. For the SAT task the performance measures included: coefficient of variation (variation in reaction time during performance of the task adjusted for reaction time, i.e. standard deviation of reaction time divided by reaction time), mean reaction time, premature responses and omission errors.
fMRI image acquisition
All participants were scanned at the Centre for Neuroimaging Sciences, King’s College London, on a 3-Tesla General Electric Signa HD x Twinspeed scanner (Milwaukee, Wisc.), fitted with a quadrature birdcage head coil. For the fMRI, we acquired T2*-weighted volumes (GNG=260; SAT=480) on non-adjacent slices (GNG=37;SAT=31) parallel to the anterior-posterior commissure. For GNG, imaging parameters were: TE=30ms, TR=1.8s, flip angle= 73° , slice thickness=3.0mm, in-plane voxel-size=3.75mm2, slice gap =0.7mm and matrix size=64x64 voxels. For SAT they were: TE=30ms, TR=1.5s, flip angle= 68°, slice thickness=3.0mm, in-plane voxel-size=3.75mm2, slice gap =1.4mm, and matrix size=64x64 voxels.
Also, a high resolution gradient echo structural scan was sagitally acquired to be used during normalization of the fMRI data into Talairach space. Imaging parameters were: TE=30ms, TR=3s, flip angle=90°, 43 slices, slice gap =0.3 mm, slice thickness=3.0 mm, matrix size=128x128 voxels.
fMRI image analysis
The fMRI data were analyzed using the XBAM (version 4) software developed at the King’s College London’s Institute of Psychiatry, Psychology and Neuroscience(40). The associated methods are described in brief in this section and in more detail in the supplementary material section. This non-parametric approach minimizes assumptions involved in image processing and has been previously described(26). Within each run, every volume was realigned to the mean of all the images in the run and then smoothed (in native space) using a Gaussian filter (full-width at half-maximum 8.8mm). Using a wavelet-based resampling method, a time series analysis was conducted on each individual subject, in order to compute a sum of squares (SSQ) ratio reflecting the BOLD effect. SSQ ratio maps were transformed into standard stereotactic space(41) using a two-stage warping procedure(40). First, an average image intensity map for each individual was computed, then warped onto their structural scan. A second stage process then transformed each of these maps from structural space to Talairach space by maximizing the correlation between the images at each stage. The SSQ ratio maps were then transformed into Talairach space using these same two transformations. Group brain activation maps were computed for each drug condition with hypothesis testing performed at both the voxel and the cluster level. Using data-driven, permutation-based methods, with minimal distributional assumptions, time series analyses were perfomed for group maps and inter-group random permutation for within/between-group ANOVAs to compute the distribution of the SSQ ratio under the relevant null distribution hypothesis. Thresholding to the required level of significance was then performed using a two-stage process: first at a voxel-wise p-value of 0.05, followed by grouping the supra-threshold voxels into 3D clusters and testing their significance against a null distribution of clusters occurring by chance in the permuted data. The cluster-wise p-value can thus be set in such a way as to yield less than one false positive 3D cluster per map. For GNG, brain activation during No-Go responses were compared to brain activation during successful oddball trials. For SAT, brain activation during 8s delays were compared to brain activation during 0.5s delays. A group brain activation map was produced for each group (TD, ASD) and medication (placebo, tianeptine) status. Finally, all ANOVA analyses were conducted with voxel level p<0.05 and a cluster level p<0.02 determined as described above.
Between group analysis of variance
A main effect of group (ASD, TD) analysis was conducted for the placebo condition for both GNG and SAT.
To investigate whether brain activation differences in the ASD group relative to the control group under placebo changed after tianeptine dose in ASD, a main effect of group analysis was conducted in regions showing a main effect of group under placebo, but now comparing the control group on placebo with the ASD group on tianeptine, to test whether tianeptine would abolish the baseline differences.
Furthermore, a within ASD effect of drug analysis was conducted, in regions showing a main effect of group, to investigate whether the degree of change in activation in ASD following tianeptine was significant.
Group x drug status interaction analysis of variance
A two-group (ASD, TD) by two-drug status (placebo, tianeptine) factorial repeated-measures ANOVA was conducted for each task. This analysis investigates how the BOLD response changes in brain regions in each group depending on drug status. The cluster-level threshold was adjusted to p<0.02, resulting in less than one false-positive cluster per map.
Correlations between symptomatology and change in functional activations
Pearson’s correlations were conducted in XBAM to investigate any associations between core symptoms (as measured by the ADI-R and ADOS, 5 symtoms in total) and differences in BOLD response between tianeptine and placebo conditions (tianeptine–placebo) within ASD, in regions showing a main effect of group during placebo, during both tasks (8 regions in total). The SSQ ratio was extracted for each cluster showing a correlation and plotted versus symptomatology. A False Discovery Rate analysis was conducted to account and correct for multiple comparisons (5 * 8 = 40 comparisons in total).