Atypical resting synchrony in autism spectrum disorder

Participants

Data were recorded from a total of 20 adolescents with ASD and 15 typically developing (TD) controls. Four ASD participants were removed from the analysis due to excessive movement during recording, resulting in a final sample size of 16 ASD participants (14 males, range = 12–15 years, 14.4 ± 1.1 years, all right‐handed) and 15 TD participants (13 males, range = 12–15 years, 14.9 ± 0.9 years, all right‐handed). Participants were recruited through the Hospital for Sick Children's Research4kids database and flyers posted in the Greater Toronto Area. Individuals with metal implants, a history of neurological or neurodevelopmental disorders (other than ASD for participants in the clinical group), or IQ ≤ 65 were excluded. All ASD participants had a prior clinical diagnosis, which was confirmed using expert clinical judgment and/or the Autism Diagnostic Observation Schedule‐General [Lord et al. 2000]. Full scale IQ was estimated for all participants using the Wechsler Abbreviated Scale of Intelligence, consisting of the Vocabulary and Matrix Reasoning subtests [Wechsler, 1999]. All participants and their legal guardians provided written consent for the protocol approved by the Hospital for Sick Children Research Ethics Board and the Declaration of Helsinki.

Data Acquisition

Participants were instructed to maintain visual fixation on a centrally presented gray cross inside a white circle on the screen while remaining still, relaxed and awake in supine position inside a magnetically shielded room. Five minutes of MEG activity was acquired at 600 Hz sampling rate with a band‐pass filter of 1–150 Hz and third‐order spatial gradient noise cancellation, using a CTF Omega 151 channel whole head system (CTF Systems Inc., Port Coquitlam, Canada). Throughout data acquisition, head position was recorded continuously by measuring the position of the three fiducial coils, located at the nasion and left and right preauricular points. Fiduciary head coils were energized at 1,470 Hz, 1,530 Hz, and 1,590 Hz. Participants with head movement less than 10 mm for 90% of the recording time were considered acceptable for further analysis. This standard of movement tolerance is typical in MEG studies of pediatric populations, allowing collection of MEG data from a clinical child population without creating a biased sample [Taylor et al. 2011]. Median head displacement during the 5‐minute recording did not differ between groups (P > 0.05).

Following the MEG recording, anatomical MRI scans were acquired for each participant using a 3T MR scanner (MAGNETOM Tim Trio, Siemens AG, Erlangen, Germany). T1‐weighted magnetic resonance images were obtained using a high resolution 3D MPRAGE sequence (TR/TE/flip angle = 2,300/2.96 ms/9°, FOV = 28.8 × 19.2 cm, 256 × 256 matrix, 192 slices, slice thickness = 1.0 mm³ isotropic voxels) with a 12‐channel head coil. To support source reconstruction of MEG activity, the anatomical MRI and MEG data were coregistered using fiducial references.

Data Analysis

A flowchart presenting an overview of data processing and analysis pipeline is depicted in Supporting Information Figure S1.

Frequency‐ and Region‐Dependent Alterations of Resting Synchrony in ASD

To facilitate characterization and interpretation of functional connectivity data from a large number of regions in the brain, we sorted connections represented in Figure 2 into regional clusters based on anatomical location (Fig. 1). This analysis approach enables the identification of alterations in resting synchrony which are expressed at levels of anatomical organization previously identified in the ASD literature (e.g. frontoparietal, frontofrontal, anterior–posterior). Group differences in within‐ and cross‐lobar functional connections were statistically significant in five out of the six frequency bands (P < 0.05, FDR‐corrected, Fig. 3). One of the primary observations from this large‐scale, lobar analysis was a frequency‐ and region‐specific pattern of overconnectivity and underconnectivity in the ASD group. In slower theta and alpha bands, occipital and parietal lobes expressed functional disconnection from widespread brain areas (including connections within these grouping themselves), with the most prominent effects observed in the alpha band. Conversely, faster gamma oscillations exhibited hyperconnectivity among frontal, temporal, and subcortical lobes in ASD adolescents. A more complex pattern of atypical intra‐ and interlobar connectivity was observed in the beta band, in which elements of the posterior disconnection were observed in concert with fronto‐tempo‐subcortical overconnection in the ASD group. Lastly, intralobar and interlobar connectivity in the delta band were similar between groups; however statistical differences between the two groups supported an anterior–posterior disconnection in the frontal cortex in ASD. No statistical differences in intralobe or interlobar functional connectivity were observed in the high gamma band.

For slow brain oscillations in the delta band, the parietal and subcortical lobes displayed significant decreases in functional connectivity with the frontal cortex (P < 0.05, uncorrected for frontoparietal, FDR‐corrected for frontosubcortical). Furthermore, intratemporal, intrasubcortical and intraparietal connections appeared hypoconnected in ASD. In the theta band, intralobar functional connectivity was lower in both parietal and occipital lobes and between these two lobes and the rest of the brain for the ASD group (P < 0.05, uncorrected for parietotemporal and parietosubcortical, FDR‐corrected for all other connections). A more striking trend of functional disconnection in the same direction was found in the alpha band (P < 0.05, FDR‐corrected). In the beta band, the ASD group had significantly higher functional temporotemporal, temporosubcortical, and subcorticosubcortical connectivity (P < 0.05, FDR‐corrected). In contrast, frontooccipital, parietooccipital, and occipitooccipital beta band connections were lower in ASD adolescents compared to TD adolescents (P < 0.05, FDR‐corrected). In the low gamma band, all intra‐ and interlobar connections between the frontal, temporal, and subcortical lobes were significantly higher in the ASD group (P < 0.05, uncorrected for frontofrontal connections, FDR‐corrected for all other connections). Finally, despite lack of statistical differences between groups in the high gamma band, the temporal lobes appeared overconnected to the frontal cortex, to subcortical regions and to themselves.

Increased Gamma‐Band Network Synchrony in ASD

NBS is a method to detect functionally interconnected (functionally integrated) nodes in a graph that cluster into a single component, and which are significantly different between two groups. In our study, NBS analysis revealed a single significant network (P = 0.012, see Fig. 4 for details). This network consisted of 15 nodes that were overconnected in the ASD group, in the low gamma band, consisting of frontal, temporal, and subcortical regions. Each region has been indexed by its region number, corresponding to Table S1 in the Supporting Information. Of note, the region with the greatest number of significantly different connections was the orbital part of the left middle frontal gyrus (MFGorb). Specifically, the left MFGorb was hyperconnected to the left and right rectal gyri, the opercular part of the right inferior frontal gyrus (IFG), the right hippocampus, and the left olfactory cortex. These NBS results are congruent with the large‐scale lobar connectivity findings noted above for the low gamma band, and indicate a specific network of hyperconnected nodes anchored in prefrontal cortex. No significantly connected components in the other five frequency bands were observed using the NBS approach.

Frequency‐Dependent Alterations in Global Network Topology in ASD

To investigate the hypothesis that network topology of spontaneous MEG activity is altered in adolescents with ASD, graph analyses were applied at the global (whole graph) and local (each node) levels to each subject's graph of resting phase synchrony for each of the six analyzed frequency bands. At the global level, we compared average strength, average clustering coefficient, and characteristic path length of the two groups. Figure 5 depicts group averages for each global measure for each frequency band. Significant differences between the groups were observed for low‐frequency oscillations (delta and theta) but not in the higher‐frequency bands (P < 0.05, FDR‐corrected).

Similar to our results from investigation of large‐scale anatomical organization within and between the lobes of the brain, alterations in network topology in ASD were also expressed in a frequency‐dependent manner. Overall, we found a general trend for increased “local connectedness”—increased global clustering coefficient coupled with decreased path length—of the functional network (theta, beta, low, and high gamma bands). By contrast, networks in the alpha frequency range were similar between groups, while delta band alterations in network topology showed effects in the opposite direction. Taken together, results at the ‘whole network' level suggest alterations in functional integration and segregation of regional clusters in adolescents with ASD. Of note, average clustering coefficient in delta and theta bands were significantly different between the groups (P < 0.05, FDR‐corrected), whereas characteristic path length was only significantly different in theta band (P < 0.05, FDR‐corrected). Average strength differences between groups were significant in the delta frequency range, where higher resting synchrony was observed in the TD adolescents (P < 0.05, FDR‐corrected).

Atypical Regional Topologies in Adolescents with ASD

Network topology was also investigated at each of the 90 nodes in order to capture alterations in brain network organization pertaining to the connectedness of particular brain regions in ASD. This analysis revealed significant between‐group differences in nodal strength, nodal clustering coefficient, and nodal path length in five of the six analyzed frequency ranges (P < 0.011 corrected, Table 1). Congruent with our findings using global network measures, nodal‐level measures “local connectedness” (clustering coefficient and path length) were higher in theta, beta, low, and high gamma band in ASD, but were decreased in the delta band, with no differences present in the alpha band. Overall, adolescents with ASD exhibited disrupted local segregation and integration of functional networks compared to TD adolescents. There were 13 nodes (regions) which were significantly different in local graph properties between the two groups; close to half of the regions belonged to the frontal lobes, including orbital and medial orbital parts of the superior frontal gyrus, the gyrus rectus, the orbital part of the middle frontal gyrus, and the pars triangularis of the inferior frontal gyrus (P < 0.011 corrected, Table 1).

Altered EEG and MEG Resting Connectivity in ASD

Functional disconnection of the parietal and occipital cortices from other cortical regions at lower frequencies (delta, theta, and alpha) was observed in ASD adolescents, in which the alpha band was the most striking. Decreased alpha coherence has been previously reported at the sensor/electrode level in resting‐state EEG studies in children [Coben et al. 2008] as well as adults with ASD [Murias et al. 2007]. We did not, however, observe a significant frontal [Murias et al. 2007] or temporal [Coben et al. 2008] disconnection from the rest of the cortex in theta or alpha bands albeit nonsignificant decreases in interlobar connectivity of frontal and temporal lobes in ASD in comparison to TD were noted. Although frontal and temporal connections to the occipital and parietal lobes were significantly hypoconnected in the present study, effects appeared to be driven by the low‐frequency disengagement of posterior regions from widespread brain areas. In delta and theta bands, lower coherence in ASD has been reported previously in parietal and occipital regions [Coben et al. 2008], in addition to temporal and central regions which we did not observe.

Interestingly, our results contrast those from a recent resting‐state MEG study that reported increased long‐range connectivity in the temporal, parietal, and occipital lobes in alpha band [Ghanbari et al. 2013]. The differences between this study and ours could be attributed to factors such as differences in age groups (children and adults versus adolescents), eye state (closed versus open), analysis level (sensor‐space versus source‐space), parcellation scheme (sensor‐groups versus atlas‐based seeds) and functional connectivity metric (synchronization likelihood versus wPLI). Moreover, the results presented here are likely to be more accurate with respect to specific brain regions and networks involved in altered oscillatory synchrony in ASD, as analysis of EEG/MEG connectivity among reconstructed sources is substantially more accurate than sensor‐based approaches in this regard [Schoffelen and Gross, 2009].

ASD Adolescents Express Decreased Anterior‐Posterior Connections at Slow Frequencies

Deficient long‐range connections along anterior–posterior brain pathways could disrupt experience‐dependent processes during development that are important for creating and maintaining neural connections [see Geschwind and Levitt, 2007 for review]. Atypical maturation of connectivity could hinder the development of social cognitive abilities such as joint attention and social responsiveness, which reliably discriminate young children with ASD from their TD peers [Baranek, 1999; Werner et al. 2000; Wetherby et al. 2004]. Evidence of reduced coherence between anterior and posterior brain regions on the basis of resting EEG data [Coben et al. 2008; Murias et al. 2007] supports this hypothesis. The view that frontoposterior connectivity is reduced in ASD is also supported by several lines of evidence from structural and functional MRI studies [see Just et al. 2012 for review]. Our findings using source‐space analysis of resting neurophysiological synchrony in adolescents with ASD are congruent with the above hypothesis at low frequency oscillations below 30 Hz. However, our results also suggest an extension to the disconnection theory: (1) that aberrant features of brain organization are expressed in a frequency‐dependent manner; and (2) both cortical and subcortical systems are affected.

The Hyperconnected Gamma Subnetwork May Reflect Socioemotional Deficits in ASD

Gamma oscillations are regarded as important as they are believed to coordinate precise synchronization in local circuits, and accordingly, play a vital role in information processing supporting cognition [Fries, 2009]. Converging evidence suggests that properties of gamma oscillations are altered in ASD during task performance [Orekhova et al. 2007; Rojas et al. 2011; Stroganova et al. 2012; Wilson et al. 2007]. Our finding of increased connectivity in the ASD group in a subnetwork of medial frontal, temporal and subcortical regions is of twofold interest; it lends support to the view that altered gamma‐oscillatory processes play a role in the pathophysiology of ASD, and demonstrates that, in particular, affected regions in this subnetwork include regions understood to be critical for social cognition.

The theory that excess high frequency oscillations reflect an imbalance in the excitation‐inhibition homeostasis in the cortex [Orekhova et al. 2007] may be relevant to findings of gamma‐band hyperconnectivity in ASD. Mutations in genes involved in the expression of excitatory and inhibitory neurotransmitters have been identified in ASD populations [Collins et al. 2006; Ramoz et al. 2004] as well as animal models of ASD [DeLorey et al. 2008]. This may provide the basis for a link between atypical oscillations and genetic mechanisms underlying ASD, as individual differences in the expression of gamma oscillations have been demonstrated to be under tight genetic control [van Pelt et al. 2012].

With respect to social cognition, there are several regions within the significantly different, hyperconnected gamma band subnetwork in ASD (identified using NBS) that are of particular interest. Such hyperconnected regions include the IFG and regions encompassing the “social brain”: the orbitofrontal areas, the amygdalae, the STG [Brothers, 1990], and the medial frontal cortex [Amodio and Frith, 2006]. Hyperconnectivity amongst regions critical for social cognition, in concert with hyperconnectivity to other regions of a different functional network may disrupt task‐relevant communication among brain regions underlying social cognitive abilities [Lynch et al. 2013]. Alternatively stated, atypical connectivity among these regions may limit dynamic interactions amongst the rest of the cortex [Uddin et al. 2013].

One region of particular interest is the IFG due to its role in language, social cognition, and emotion processing. The IFG is vital for speech and language processing, as it is responsible for both semantic and syntactic processing of linguistic inputs, in addition to the motor production of speech [Bookheimer, 2002]. The mirror neuron system, located in the IFG, is a neural mechanism by which others' actions and intentions can be understood [see Rizzolatti and Craighero, 2004 for review], and its dysfunction was proposed to give rise to the cascade of impairments that are characteristic of ASD [see Williams et al. 2001 for review]. In addition, atypical activation in fMRI has been demonstrated in the IFG during emotion processing tasks [Fusar‐Poli et al. 2009; Kesler‐West et al. 1999], suggesting aberrant connectivity of this region in ASD populations.

Another key observation from the NBS analysis relates to the functional role of two limbic regions: the right amygdala and the hippocampi. Converging evidence from studies in human and animals suggests that emotion perception and regulation is mediated by a brain circuit in which the amygdalae and hippocampi are central components. The amygdalae play a key role in the recognition and evaluation of emotionally salient stimuli and subsequent production of affective states [Phillips et al. 2003]. The right amygdala in particular, is linked to implicit emotional processing [Adolphs et al. 2005; Costafreda et al. 2008; Hung et al. 2012; Noesselt et al. 2005]. The hippocampi, apart from their general role in memory, modulate the production of contextually appropriate affective behavior that is elicited by emotionally salient stimuli. The hippocampi do this through inhibitory connections with the amygdalae and other structures involved in emotion perception [Phillips et al. 2003]. Interestingly, in childhood, several studies have supported the idea of enlarged amygdalae and hippocampi in ASD [Groen et al. 2010; Howard et al. 2000; Rojas et al. 2006; Schumann et al. 2004; Sparks et al. 2002], with volume losses that emerge in adulthood [Duerden et al. 2012; Nickl‐Jockschat et al. 2012; Yu et al. 2011]. Our findings of hyperconnectivity in the left and right hippocampi along with the right amygdala support the idea of a developmental abnormality in these regions in ASD, which persists into adolescence.

The last region of particular interest to highlight from the hyperconnected low gamma subnetwork revealed by NBS is the caudate nucleus. This nucleus thought to be part of an intricate system integrating multimodal information and regulating complex behavior [Alexander and Crutcher, 1990; Haber, 2003]. The caudate is also increasingly recognized to be implicated in affect processing [Arsalidou et al., 2013]. Given these associations, hyperconnectivity of the caudate nucleus in the current study could be associated with corticostriatal feedback and may therefore be related to behavioral problems in ASD. In line with this, stereotypic, repetitive behavior patterns related to the diagnostic criteria for ASD have been compared to those seen in obsessive‐compulsive disorders, which are known to be associated with structural and functional abnormalities in the basal ganglia [Langen et al. 2009]. Furthermore, in a voxel‐based morphometry longitudinal study characterizing the developmental trajectories of striatum in ASD, an increase was detected in caudate volumes in ASD, while it decreased in control subjects [Langen et al. 2009]. Taken together, our findings regarding the overconnected subnetwork in the low gamma frequency band can be interpreted considering the postulate that daily life, with its social demands and constantly changing situations, imposes additional social cognitive demands in adolescents with ASD resulting in abnormal information processing in regions related to social cognition, behavior, and emotion perception and regulation.

Differences in Global Network Topology Indicated Disrupted Functional Integration and Segregation in ASD

At the ‘whole network' level, our results indicated ASD adolescents exhibited disrupted functional integration and segregation of brain networks. In particular, we demonstrated differences in average strength, clustering coefficient and characteristic path length at both low and high frequencies, suggesting a less optimized topological organization in functional networks of adolescents with ASD. Furthermore, increased average clustering coefficient and shorter characteristic path length in theta, beta, low and high gamma bands indicate increased “cliquishness” properties of functional networks within these frequency ranges, supporting the idea of disrupted balance between global integration and local specialization. On the network level, this suggests a pattern of global overconnectivity and altered network topology favouring increased clusters of local connections in ASD. Moreover, stronger local connectivity has been previously reported in structural studies of autism using diffusion tensor imaging [Herbert et al., 2004; Li et al 2014].

It is possible that the overabundance of short functional connections could be related to redundant connectivity patterns and to proposed neurobiological mechanisms in ASD. In normal early development, overconnectivity is followed by a pruning of connections in the maturing brain [Supekar et al. 2009], suggesting network refinement [Hagmann et al. 2010]. Physiologically, in ASD, impaired pruning of connections in this dynamic process would result in the redundancy of connections. Hence, the remaining overconnected network may operate at different scales, resulting in a poor signal‐to‐noise ratio where the system is flooded with noise in the event of an incoming signal [Belmonte et al., 2004]. With poor signal‐to‐noise ratio, the output of the functional network may not be sufficiently distinct (too much integration and not enough segregation) to achieve the necessary information processing [Rippon et al. 2007]. In addition, recent work has demonstrated that brain maturation is reflected in a weakening of short‐range and a strengthening of long‐range connectivity [Dosenbach et al. 2010]. From this perspective, our findings may reflect a more immature connectivity pattern in adolescents with ASD.

There are some interesting nuances to this trend in globally altered network topologies that deserve mention. First, delta band measures of global network topology exhibit the opposite pattern of effects in ASD: networks have decreased average strength, decreased clustering coefficient and increased characteristic path length. This suggests a loss of functional segregation of networks in ASD. In the model of autism as a developmental disconnection syndrome, decreased functional segregation could be anticipated as it suggests that specific functional systems are less distinct or functionally segregated from one another. Second, although there is a general trend for global overconnectivity in ASD, alpha band functional networks appear to be relatively the same between the two groups. As average strength is a direct reflection of mean coherence (functional connectivity) within the network, our results suggest that there is no significant overall difference in phase synchrony at rest in alpha bands in adolescents with ASD compared to controls. This result is not surprising as it has been previously reported in resting EEG studies in ASD [Mathewson et al. 2012; Peters et al. 2013].

Focal Disruptions in ASD Involve the Default Mode Network, Frontal Cortex and the Temporoparietal Junction

When we looked at local graph measures to describe focal alterations in specific regions of the brain, we found three observations of interest in the ASD literature. First, several regions that displayed significant alterations in focal topology were implicated in the default mode network (DMN), such as the precuneus and the inferior parietal lobule. Second, five of the thirteen nodes that were significantly different in local graph properties in ASD belonged to the frontal cortex. Lastly, in the high gamma band, focal alterations were detected in theory of mind (ToM)‐relevant regions in the ASD group. One of the primary observations from the regional graph analysis relates to alterations in regions encompassing the DMN. The delta band group differences provide a particularly salient example of these alterations involving DMN regions. Decreased nodal strength, decreased clustering coefficient and increased path length in nodes of the DMN in delta band suggest loss of functional segregation within this network. In contrast, theta and high gamma bands reveal the opposite direction of effects in DMN regions; such regions appeared to be hyperconnected and demonstrate increased local functional connectedness. We propose that these findings—though seemingly opposite in effect—both suggest that regions involved in the DMN are functionally disrupted. In other words, adolescents with ASD may exhibit a less optimized topological organization in the DMN, leading to frequency‐dependent functional alterations in DMN regions. Accordingly, decreased clustering coefficient may relate to decreased functional segregation (i.e., delta band findings) of the DMN to the rest of the brain, while increased clustering coefficient may suggest lack functional specialization (i.e., theta, beta, low gamma band findings) due to an abundance of functional connections. Interestingly, evidence from structural and fMRI in ASD have reported both underconnectivity [Assaf et al. 2010; Stigler et al. 2011] and overconnectivity [Lynch et al. 2013; Uddin et al. 2013; Redcay et al. 2013] between DMN nodes.

The frontal lobes are responsible for numerous higher‐order cognitive functions, including planning, decision making and abstraction, and thus are a primary candidate for dysfunction in many neurodevelopmental and neuropsychiatric disorders. We report 5 out of 13 altered focal topologies in the frontal cortex, suggesting a less organized or more random distribution of functional networks involving frontal areas. Although our findings revealed altered local functional connectivity in the frontal lobes, we are limited in terms of the conclusions we can make regarding the anatomical proximity of these functional hyperconnections. In light of our large‐scale lobar connectivity results, it is plausible that short‐range connections involving the frontal cortex are underconnected at lower frequencies (e.g., delta band) and overconnected at higher frequencies (e.g., low gamma band). Even though current theory purports that there is local overconnectivity in the frontal cortex in ASD [Courchesne and Pierce, 2005], literature from multiple modalities involving task‐based and task‐free paradigms have been inconclusive in supporting this hypothesis [see Vissers et al. 2012, for review]. Nonetheless, the abnormal functional properties of frontal regions are interesting in light of both the association between ASD and the current frontal hyperconnectivity theory and evidence of frontal involvement in the DMN.

Another inference that may be drawn from observations from local graph theory results relates to regions implicated in ToM. The impairment of ToM (processing of mental states of others) in ASD has been linked to difficulties with the communications and interactions in everyday life [Kana et al. 2009]. The temporoparietal junction (TPJ), in particular, is thought to be central to the integration of social information and empathy, in addition to selective attention to salient stimuli (Decety and Lamm, 2007]. In the high gamma band, two regions surrounding the TPJ demonstrated altered focal topology in our study: the left superior and medial temporal gyri. In children, adolescents and adults with ASD, thinning of several regions in the TPJ region, particularly on the left side, has been reported [Greimel et al. 2013; McAlonan et al. 2005; Raznahan et al. 2010; Wallace et al. 2010]. Although our findings of increased local strength in TPJ regions are not immediately intuitive when considering neurological reports of volume reductions, it is suggestive of altered function within these regions that could interfere with ToM‐related information processing.

In general, we found alterations in global and local graph analyses supporting increased local connectedness in functional brain networks in ASD. We propose that increased local connectedness could imply decreased functional specialization of brain regions in ASD. Connections in the brain are formed at a high physical cost [Bullmore and Sporns, 2009; Sporns and Zwi, 2004] and the brain constantly negotiates the trade off between wiring costs and topological efficiency [Bullmore and Sporns, 2012]. Our findings of increased average strength in ASD and increased local connectedness suggest impaired network refinement in this population, particularly at a critical time of development in adolescence.

Limitations and Future Work

Some important caveats should be considered with regard to the present study. First, the limits of MEG in localizing and estimating the activity of deep brain sources remain an area of on ongoing research. Multiple studies have shown MEG to be effective at detecting weak signals emanating from deep brain structures such as the hippocampi during a variety of tasks [Cornwell et al. 2008a; Ioannides et al. 1995; Kirsch et al 2003; Nishitani et al. 1998], as well as the amygdala [Cornwell et al. 2007, 2008b, 2010; Hung et al. 2010; Ioannides et al. 1995; Liu et al. 1999; Luo et al. 2007; Moses et al. 2007; Streit et al. 1999] and thamalus [Bardouille and Ross, 2008; Bish et al. 2004; Tesche, 1996]. Moreover, there have been several realistic simulations that have shed light on parameters affecting our ability to detect such sources of activity [Balderston et al. 2014; Mills et al. 2012; Quraan et al. 2011]. Despite this empirical evidence, the debate over the ability of MEG to accurately detect and localize deep sources has persisted [for in‐depth discussion see Riggs et al. 2009; Stephen et al. 2005]. With this in mind, our findings regarding connectivity alterations involving deep brain sources in this study should be interpreted with a degree of caution.

Another issue is the effect of sorting brain regions into their anatomical, lobar subdivisions. Since each lobe is different in size, numbers of regions/nodes sorted into each were accordingly different. This results in an uneven number of regions for each lobe, so it is plausible that a small number of connections within a lobe may have exerted a disproportionate influence on the resulting depiction of lobar connectivity. To address this issue, we describe regional differences between the two groups using bivariate statistics (NBS) in addition to both global and nodal graph theory measures, which do not suffer from this limitation. We feel this level of description is nonetheless informative given prior theory and experimental reports describing altered connectivity in ASD at the level of large‐scale anatomical systems.

Third, while the atlas‐based whole brain reconstruction of MEG source activity has been shown to be a valuable approach for investigating connectivity in large‐scale brain networks, there are some limitations when using this method. For instance, the selection of sources is sparse and may not yield an accurate or adequate selection of brain structures that contribute to a more complete understanding of the brain in ASD. Such limitations might be ameliorated by adopting a whole‐brain parcellation scheme that is based on source sensor geometry to obtain a set of maximally independent patches [Palva et al 2010]. While such advances are excellent in addressing some experimental questions in normative studies, however, they introduce additional complications for our experimental question related to ASD. Specifically, it has been established that there are reliable structural brain differences in ASD, and as such, parcellation approaches which use structural information to define cortical “patches” for analyses may introduce potential confounds to the study (i.e., systematic group differences in brain morphology may introduce systematic differences in patch location, complicating interpretation of the results). Nonetheless, we feel the disadvantages of using a standard brain‐atlas are outweighed by the important advantage that it enables a more direct comparison between data from different modalities, especially in the context of heterogeneity in ASD findings.

Recently, there has been an emergence of evidence indicating brain overconnectivity in children with autism, contrasting multiple prior reports of underconnectivity in ASD [Di Martino et al. 2014; Keown et al. 2013; Supekar et al. 2013; Uddin et al. 2013]. These studies in resting‐state fMRI report hyperconnectivity in both long‐ and short‐range intrinsic connections across multiple brain regions in young children with ASD. Such findings add weight to the argument that hyperconnectivity outweighs hypoconnectivity in ASD, while also painting a more complicated picture regarding where disrupted brain connectivity in ASD may be dependent on altered age‐related trajectories. In view of these recent findings, a limitation to our study is the lack of longitudinal data in our cohort of ASD participants to comprehensively address this question. Furthermore, as adolescence is a period of heavy grey matter development within the brain, it will be interesting to characterize whether the functional alterations reported are directly related to structural alterations due to a difference in developmental trajectory in this population. Future research should elucidate the developmental trajectory of altered MEG connectivity in ASD, as well at the relations between atypical neurophysiological interactions and underlying brain structure.