JPCB Journal

Introduction

Autism spectrum disorder (ASD) is a heterogeneous neurodevelopmental lifetime condition. The main characteristics are restrictive and repetitive behaviours and interests and social and communication difficulties [1]. In particular, processing of facial expressions, a skill critical to successful peer interactions, is impaired in ASD [2,3] and may be a major contributor to the social deficits [4]. Individuals with ASD can also have cognitive difficulties in executive functioning, including working memory [5,6], which involves the ability to encode, store and/or manipulate information in the mind over a short delay (maintenance) and to retrieve or recognize information [7]. Both emotional face processing and working memory abilities play a central role in perceiving and adapting to social cues in one’s environment. However, despite widespread difficulties in these domains in those with ASD [8,9], relatively little is known about the underlying functional brain networks supporting emotional face processing and working memory in ASD and how they may be affected by pharmacological intervention.

Individuals with ASD display global alterations in the activation and connectivity of task-based neural networks [10-13]. While functional MRI has been traditionally used in this work, magnetoencephalography (MEG) is the optimal non-invasive neuroimaging approach to investigate functional connectivity, as it provides a combination of high spatial, temporal, and spectral (frequency) resolution of neural oscillatory activity and connectivity [14,15]. There have been some studies employing MEG that examined functional connectivity in children and adults with ASD, encompassing a range of task-based paradigms to better understand the neural underpinnings of social cognitive difficulties. Two oftenused protocols in the field include the n-back working memory and emotional faces tasks [13,16-18].

Atypical functional brain connectivity in adults with ASD has been found in theta and alpha frequency bands during an n-back working memory task using MEG [19,20]. For instance, Yuk et al. found decreased theta-band connectivity between inhibition and working memory brain areas in ASD during recognition of repeated visual stimuli (1-back) despite similar task performance to typically developed (TD) adults. As theta-band connectivity has been linked to the integration of brain regions during working memory [21], these results suggest difficulty inhibiting task-irrelevant stimuli during working memory in adults with ASD, which may be exacerbated at higher cognitive loads.

Similarly, atypical functional connectivity in several frequency bands has been observed in those with ASD to emotional face stimuli [13,16,22]. Leung et al. showed decreased beta-band connectivity to angry faces in adolescents with ASD, also shown in adults by Mennella et al., which could be related to deficits in top-down control, while Safar et al. showed that when a wide age range was studied, significant age-and group-related changes were seen particularly in the gamma band.

An intervention that is currently being explored for improving social skills in ASD is the administration of oxytocin (OXT), a nine-amino-acid peptide. Central release of OXT and its closely related peptide, arginine vasopressin (AVP), have been found to be involved in social functioning, including social recognition, social memory, and affiliative behaviours (e.g., [23]). A few studies have examined the effect of intranasal OXT (IN-OXT) in TD adults and showed that it improved trust and prosocial behaviour [24-26] and improved recognition memory for faces but not objects [27, 28].

Due to its impact on social behaviours, OXT has been trialled over the years as a potential treatment of social cognitive symptoms in autism. A single dose of oxytocin was reported to improve a number of social behaviours, such as affective speech in adults with ASD [29] and facial emotion recognition in youth [30]. Longerterm interventions have also shown improvements in core autism symptoms such as social reciprocity [31] or in secondary measures such as repetitive behaviour, social cognition and attachment [3234]. The literature, however, is not unanimous on these effects, with other studies showing few or no improvements with IN-OXT administration [35,36]. It is argued that although OXT has a role in modulating social-cognitive behaviour and has potential therapeutic value, more needs to be understood about its neurobiological mechanisms, including its influence on brain connectivity underlying emotion processing and working memory [37,38].

Using MEG, the current study is the first to examine whether INOXT, administered as part of a 12-week randomized clinical trial, enhanced social cognitive functioning related to brain-behaviour signatures in adults with ASD. We hypothesized that IN-OXT would effect changes in functional brain connectivity during emotional faces and working memory tasks that would be associated with improved social cognitive behaviour.

Methods

This study was part of a 12-week randomized, double-blind, parallel design clinical trial (NCT01788072), approved by the Research Ethics Boards of Holland Bloorview Kids Rehabilitation Hospital and the Hospital for Sick Children in Toronto, Canada, as a secondary, exploratory outcome. All participants had a primary diagnosis of ASD assigned by an experienced clinician and confirmed via the Autism Diagnostic Observation Schedule (ADOS-G or ADOS-2 [39,40]). Adults with ASD took 24 International Units of IN-OXT (Syntocinon® made by Novartis) or a placebo (manufactured by the Toronto Institute of Pharmaceutical Technology), twice daily (morning and afternoon), for a 12-week period. From the behavioural cohort of 42 adults with ASD (33 males), 18 participants agreed to participate in the neuroimaging component, to be scanned twice, pre- (Time 1) and post-treatment (Time 2).

Adults with ASD were eligible if they were between 18-45 years of age, had a primary diagnosis of ASD, a clinical global impressionseverity score ≥ 4, and a verbal scale IQ ≥ 70. Participants were ineligible if they were born before 28 weeks’ gestational age or had a medical history of neurological disease. All 18 participants underwent scanning; however, three did not return for the post-trial, Time 2 visit, one had poor quality MEG data due to too much movement for both tasks, and another had poor data for only the working memory task. Thus, there were 14 participants (seven in the oxytocin group, seven in the placebo group) in the emotional faces task, and 12 participants (six per group) in the working memory task (see Table 1 for demographics; there were no significant differences between the groups). Therefore, we had 52 MEG datasets over the two time points and tasks.

Table 1: Participant demographics; ADOS scores range from 3 to 20 with greater symptom severity reflected by higher scores; (S.D.).

Social cognitive tasks

In the emotional faces task, participants were shown happy or angry faces beside a scrambled version of the face (target stimuli) in the MEG [13,16]. Happy and angry face stimuli (25 faces per emotion, 12 male faces) with a minimum validity rating of 0.8 were chosen from the NimStim Facial Expressions database [41]. Participants saw 50 trials per emotion condition, presented randomly, twice per hemifield, yielding a total of 200 face trials. Stimuli were presented for 80ms with an inter-stimulus interval that ranged from 1300-1500ms. They were instructed to fixate on the central cross and make a left or right button press as quickly as possible, corresponding to the side of the scrambled face.

To assess working memory, a 1-back task with complex multicoloured abstract images was used [19,42]. The protocol included 285 trials; 190 unique images were presented (‘New’), with 95 of these repeated on the subsequent trial (‘Repeat’). Stimuli were presented serially for 200ms each on a black background with an interstimulus interval with a white fixation cross randomly varying between 1050-1300ms. Participants were instructed to press a button as quickly as possible when they identified the repetition of an image presented one trial earlier. Behavioural accuracy and reaction time for Repeat trials were recorded. Participants were included in subsequent MEG analyses if they had a recognition accuracy of >55%; only correct trials were analyzed.

For both tasks, stimuli were presented using Presentation Software

(Neurobehavioral Systems Inc., https://www.neurobs.com/presentation) on a rear projection screen positioned 80cm from the participants’ eyes.

Neuroimaging acquisition and analyses

A 151-sensor CTF MEG system (Coquitlam, BC, Canada) recorded data with a 600 Hz sampling rate during the tasks, while participants lay supine. Head position was continuously tracked via three fiducial coils on the left and right pre-auricular points and nasion. An online anti-aliasing low-pass filter (0-150 Hz) and a 3rd order spatial gradient were applied to attenuate background environmental noise.

For co-registration, a structural T1-weighted MRI was obtained in all participants with a 12-channel head coil in a 3T MRI scanner (MAGNETOM, Siemens AG, Erlangen, Germany) using a sagittal 3D MPRAGE sequence (TR/TE =2300/2.96ms, FA =9º, FOV =192x240x256 mm, voxel size =1.0 mm isotropic). To allow for co-registration of MEG data, MRI radio-opaque markers were placed in the same positions as the MEG fiducial points.

MEG preprocessing and analyses

MEG data were preprocessed using the FieldTrip toolbox [43] in MATLAB. The continuous MEG data were band-pass filtered from 1 to 150 Hz using a 4th order Butterworth filter, with a notch filter at 60 and 120 Hz. For the working memory task, data were epoched from -500 to 600ms relative to the Repeat stimulus onset. For the emotional faces task, data were epoched from -600 to 600ms relative to stimulus onset (happy or angry faces).

Independent component analysis (ICA; ‘fastica’ function) was applied to identify ocular and cardiac artefacts, which were removed via manual inspection. Trials were further excluded if the signal exceeded +/-2000fT, and if initial median head position shifted greater than 5mm. one participant had to be removed from both analyses and another the memory task could not be used, as there were insufficient trials remaining (<25) after excluding trials with movement. For both the emotional faces and working memory tasks, there were no significant differences in head motion or number of trials analyzed between groups or sessions (Time 1 or Time 2), nor were there any significant group-by-session interactions (all p>0.05, see Supplementary Table 1 for descriptive summary).

A single shell head model was generated for each participant based on their anatomical MRI [44]. The centroids of the 90 cortical and subcortical coordinates of the Automated Anatomical Labelling (AAL; [45]) atlas were computed in standard template space (ICBM 152; [46] and non-linearly transformed onto analogous head space coordinates for each individual subject. A linearly constrained minimum variance beamformer [47] with 5% Tikonov regularization was used to estimate the broadband time series at each of the 90 AAL sources.

Following source reconstruction, the broadband MEG time series, at each AAL source location, were filtered into the following frequency bands: theta (4-7Hz), alpha (8-14Hz), beta (15-29Hz) and gamma (30-55Hz). To compute the instantaneous phase values for each source location and frequency band, the Hilbert transform was used. The cross-trial phase lag index (PLI; based on [48]) was computed to measure pair-wise functional connectivity between AAL source locations. For the emotional faces task, functional connectivity analyses were focused on the 0-200ms time window following stimulus onset using eight a priori regions of interest related to whole brain connectivity [13,49]: the bilateral fusiform gyri, amygdalae, insulae and anterior cingulate cortices (ACC). For the working memory task, we also took an a priori approach, focusing on the 100-400ms time window and the major hubs involved in working memory [19, 20]: the bilateral middle frontal gyri (MFG), the inferior frontal gyri (IFG), the inferior parietal lobes (IPL), the precuneus (PCUN) and the anterior cingulate cortices (ACC). Thus, for each subject (and frequency band) this yielded an 8-by-90 adjacency matrix and a 10- by-90 adjacency matrix for the emotional faces and working memory tasks, respectively. A baseline window of equal length to each of the active windows was selected (-200 to 0ms for the emotional faces and -300 to 0ms for the working memory task). The PLI values at each time point during the active window were z-scored relative to the baseline window.

The Network Based Statistic (NBS), a non-parametric technique for the statistical analysis of large networks, was used [50,51]. We ran between-group (Contrast: Placebo vs IN- OXT) and withingroup (Contrast: Time 1 [baseline] vs Time 2 [post-treatment]) ttests in NBS. A conservative primary component-forming threshold based on 5% of total possible connections (~40 connections) was used; 5000 permutations were run.

 Results Behavioural results

Table 2: Summary of behavioural performance; Values are described as mean ± SD when normally distributed and median [interquartile range] when skewed.

Emotional faces task

Findings revealed significant differences in connectivity at baseline between the Placebo and IN-OXT groups (IN-OXT > Placebo, pcorr = 0.02) in the gamma frequency band (30-55 Hz) to angry faces. This network involved connections predominantly among the bilateral limbic and frontal regions (Figure 1). The most highly connected areas (i.e., nodes with highest degree) included the right amygdala, right ACC and left insula. Within the IN-OXT group, we observed a significant difference in connectivity between baseline and posttreatment (Time 2 > Time 1; pcorr = 0.01) in the beta frequency-band (15-29 Hz) to angry faces. This network was anchored in the bilateral fusiform gyri and amygdalae, with connections predominantly among the limbic, occipital and frontal regions, largely in the left hemisphere (Figure 2). No significant changes over time were observed in the Placebo group within any frequency band, to angry or happy faces (all pcorr>0.05).

Figure 1: Significant between-group gamma network at baseline (Time 1);significant differences in connectivity were observed at baseline between the Placebo and IN-OXT groups in the gamma band to angry faces. The network involved connections predominantly among the bilateral limbic and frontal regions. Most highly connected areas included right amygdala, right ACC, and left insula. The node sizes in the gamma network (on the left) are scaled by degree (i.e., the number of connections). On the right, the mean network connectivity strengthw as extracted and plotted for both groups.

Figure 2: Significant within-group beta network in the IN-OXT group;significant differences in beta connectivity were observed in the IN-OXT group between the baseline and post-treatment to angry faces. The network involved connections predominantly among the limbic, occipital, and frontal regions, largely in the left hemisphere. This network was anchored in the bilateral fusiform gyri and amygdalae. The node sizes in the brain network (on the left) are scaled by degree and mean connectivity strength for this network at both time points are plotted on the right-side graph.

Working memory task

Similar to the emotional faces task, we ran between-group and within-group t-tests in NBS for the 1-back task. Significant differences in connectivity were observed at baseline between groups (Figure 3), with the Placebo group showing higher theta connectivity compared to the IN-OXT group (pcorr=0.039). Within the IN-OXT group, higher theta connectivity was found at Time 2 compared to Time 1 (pcorr=0.029; Figure 4). The two theta-band networks had a similar distribution, involving fronto-parietal connections with major hubs in the left MFG, the left IFG and in the right ACC. There were no significant differences in connectivity between baseline and posttreatment in the Placebo group (all pcorr>0.05).

Figure 3: Significant between-group gamma network at baseline (Time 1); The Placebo group showed higher theta connectivity compared to the IN-OXT group, anchored in bilateral frontal regions. Node sizes are scaled by degree. The mean connectivity strength for this network for both groups are plotted on the right-side graph.

Figure 4: Significant within-group theta network in the IN-OXT group;Higher theta connectivity was found at Time 2 compared to Time 1 only in the IN-OXT group, involving fronto-parietal connections with major hubs in the left MFG, the left IFG and in the right ACC. Node sizes are scaled by degree. The mean connectivity strength for this network at both time points are plotted on the right-side graph.