Introduction
Autism spectrum disorders (ASD) are a group of neuro developmental disorders characterized by severe impairments of reciprocal social interaction, verbal and nonverbal communication, repetitive and stereotyped behaviors and abnormal sensory processes (American Psychiatric Association
2013; American Psychiatric Association
2000). The amygdala is thought to play a crucial role in the function of the ‘social brain’ in terms of being involved in social cognition, emotion recognition, socio-communicative perception and the regulation of emotional responses (Phelps and LeDoux
2005). The amygdala theory (AT) for ASD therefore hypothesizes that amygdala dysfunction underlies the social deficits seen in ASD (Baron-Cohen et al.
2000). In line with the AT’s predictions, the amygdala as a whole shows aberrant structural growth trajectories, exhibits abnormal functional connectivity (FC), and is involved in impaired emotion recognition and over-reactivity to aversive stimuli (Bellani et al.
2013; Green et al.
2013; Harms et al.
2010). The amygdala is, however, a composite structure, and its three major nuclei—the laterobasal, superficial and centromedial nuclei—have many connections with a wide variety of cortical areas. Yet, little is known about which amygdala pathways within the ‘social brain network’ (SBN) are compromised in ASD.
In earlier work, we used three anatomically defined amygdala subdivisions as seeds for a FC strength analysis and demonstrated that alterations within the amygdala network in ASD can be traced down to specific amygdala subdivisions. This approach, however, reveals functional connections that are associated with each amygdala subdivision throughout the whole brain and is therefore not system specific. In addition, anatomically defined subdivisions do not respect functional boundaries and vice versa, and each of the anatomically-defined amygdala subdivisions maintains connections along multiple pathways that are associated with various cognitive functions. In the present study, we therefore aimed to assess the functional architecture of the amygdala in adolescents with ASD by parcellating the amygdala based on its FC with three cortical seeds that are specifically anchored within the system of ‘social brain networks’ (SBNs).
Based on resting-state MRI scan data of healthy adults, Bickart et al. (
2010,
2012) characterized three major SBNs that involved the amygdala: a ‘social perception’ network, a ‘social affiliation’ network and a ‘social avoidance’ network. For the ‘social perception’ network, the amygdala exhibited the strongest FC with the lateral orbitofrontal cortex (lOFC). The primary focus of amygdala connectivity for the ‘social affiliation’ network was found in ventromedial prefrontal cortex (vmPFC). Caudal anterior cingulate cortex (cACC) showed the strongest connectivity with the amygdala within the ‘social avoidance’ network. Bickart et al. (
2010,
2012) used these cortical seed regions to parcellate the amygdala into three functional parcels: the ventrolateral, medial, and dorsal amygdala, respectively. Furthermore, the FC strength of these central nodes within the ‘social perception’ and ‘social affiliation’ networks correlated positively with the diversity and number of friends (Bickart et al.
2010,
2012).
Here, we first identified the three cortical seed regions found by Bickart et al. in our own adolescent healthy controls, and used these to parcellate the amygdala in both healthy controls and adolescents with ASD. We then compared the size of their associated parcels between the ASD group to the healthy controls. In posthoc analysis, we investigated these between-group differences further by delineating the relationship between the functional volumes and their FC strength to the three associated cortical seeds. Finally, we tested whether functional volume serves as a marker for social skills in ASD. Because the FC strength of the amygdala’s ‘vmPFC’ and ‘lOFC’ parcels have been reported as a good predictor of social network size (Bickart et al.
2010,
2012), we hypothesized that the volumes of the amygdala parcels predict the severity of social symptoms and impairment in ASD.
Discussion
We assessed the functional architecture of the amygdala in adolescents with ASD by parcellating the amygdala based on its FC with three cortical seeds (cACC, lOFC and vmPFC) that are anchored within known SBNs: the social avoidance network, the social perception network and the social affiliation network. Three functional parcels were created (AAC, AOF and APF) based on its FC with each of the three cortical seeds respectively. We found a significant enlargement of the AOF parcels in the ASD group, while there was a trend toward decreased volume of the other two parcels in ASD, especially of the APF parcel. We assessed the clinical relevance of our marker, and found that increased AOF parcel volume predicted impairments in social skills in the ASD group. In posthoc analysis, we found that the increase of the AOF parcel came at the cost of the APF parcel, as indicated by a decrease in FCAPF–vmPFC strength.
Our results align well with and extend earlier findings showing that especially the lOFC and vmPFC amygdala SBNs predicted social outcome in a healthy control sample (Bickart et al.
2012). All three affective networks are associated with generating appropriate adaptive social behavior and are known to work together closely (Bickart et al.
2014). We show for the first time that ASD is associated with FC abnormalities of the lOFC and vmPFC amygdala pathways, which are the central hubs of social perception and social affiliation network functionality. The AOF parcel roughly corresponds to the ventrolateral subregion containing the laterobasal nucleus of the amygdala and the APF parcel to the medial subregion containing the superficial nucleus according to the probabilistic cytoarchitectonically defined Jülich atlas (Amunts et al.
2005), which are associated with perceptual input processing of the amygdala (Rausch et al.
2016). Although the AAC parcel roughly corresponds to the dorsal amygdala containing the centromedial amygdala nucleus, Fig.
1c indicates hemispheric lateralization related to the AAC parcel. More specifically, the right AAC extends into the ventrolateral subregion, which has been assigned to the AOF parcel in the left hemisphere. The literature on lateralization effects on subregion level in the amygdala is however sparse, and it is therefore unclear what the functional implications of this lateralization effect are (Gläscher and Adolphs
2003; Gorka et al.
2017; McMenamin and Marsolek
2013). The AAC did not show any group effects in the FC strength analysis or parcel volume in our study. Our findings therefore may be in line with an earlier study investigating abnormalities in three anatomically defined amygdala subdivisions (the laterobasal, superficial and centromedial nuclei) that implicated amygdala under-connectivity between the superficial and laterobasal nuclei and cortex (Rausch et al.
2016).
Given the inhibitory relationship between the vmPFC and the amygdala in the literature (Motzkin et al.
2015), our findings of reduced FC
APF–vmPFC strength may point toward weaker inhibitory connections of the vmPFC amygdaloid circuit in ASD. Our posthoc FC strength analysis also demonstrates that a potential lack of inhibition from the vmPFC onto the amygdala in ASD is not linked to significantly increased FC strength of the lOFC amygdaloid network, because the FC
APF–vmPFC strength decrease was not accompanied by a significant increase in FC
AOF–lOFC strength. However, our results might indicate that a potential lack of inhibition from the vmPFC might be driving a weak—but not significant—increase of FC
AOF–lOFC strength in ASD, that is spanning a significantly larger area in the ASD group as compared to controls. Recent findings suggest that abnormalities in FC strength are rather characterized by a diffuse distribution of FC in ASD (Hahamy et al.
2015). In other words, the autistic brain may have idiosyncratic FC patterns, which cannot be identified in terms of a “common spatial locus” of abnormal FC strength changes. Instead, the areas in which the FC abnormalities occur, might be characterized by different FC between ASD subjects, which might partially explain diffuse and widespread FC changes in ASD. This suggests that ASD is not a disorder of unique abnormal loci per se, but rather a problem of the functional specialization as compared to controls. Therefore, because our functional volume measures provide a quantification of FC that is independent of a “common spatial locus” of abnormal activation within the amygdala (yet tied to specific functionality), our functional volume measure might indicate that alterations of the social perception lOFC amygdala network, are characterized by abnormal FC distribution.
We also assessed how well our FC markers predict social skills based on the AQ social subdomain in the ASD group. Since the increase of the AOF volume may be a consequence of decreased FCAPF–vmPFC strength, the correlation between decreased FCAPF–vmPFC strength and social skills was tested, but was not significant. However, we were able to relate our findings of increased AOF volume to social skills. Because the increase in AOF volume appears to be a consequence of decreased FCAPF–vmPFC strength, and because the AOF volume significantly predicted social skills in the ASD group, it might be surprising that we did not find significant negative relationships between social skills and the FC strength of the APF (and/or ACC) parcel(s). The most parsimonious explanation for this apparent inconsistency is that parcel volume and FC strength measure different things and that the first better probes the underlying pathology than the latter. A parcel’s volume is dependent on the number of voxels that exhibited maximum partial correlation with that parcel’s cortical target. Maximum partial correlation can be achieved by very small correlation differences, so for instance a relatively large parcel volume can be due to having many voxels with very small correlation differences. As such, differences in parcel volume can be great while the difference in average FC strength is very small. In other words, parcel volume, though derived from FC strength estimates, does not have to follow the same pattern as the average FC strength. The fact that parcel volume better predicts social skills than average FC strength can further be taken to imply that the underlying pathology can be attributed to a large number of voxels (neurons) that exhibit an abnormal balance in terms of its connectivity with the three cortical targets with only subtle alterations in the strength of these connections. Therefore, our functional parcel volume approach may provide a sensitive alternative to standard thresholding techniques for capturing subtle functional changes in the architecture of FC in ASD.
Reduced amygdaloid vmPFC strength link our functional volume abnormalities to results showing under-connectivity patterns in ASD populations. As the vmPFC is part of the mentalizing or the theory of mind network, person perception, self-knowledge (Amodio and Frith.
2006) and the processing of pleasant outcomes like social and monetary rewards (Rademacher et al.
2010) our results align well with the idea that FC along the amygdala-vmPFC pathway might be altered in ASD. One study showed that the dorsal medial PFC is activated rather than the ventral medial PFC in ASD during a self-referential task, which suggests under-connectivity of the vmPFC in ASD (Schulte-Rüther et al.
2011). Another study found amygdalo-vmPFC under-connectivity when viewing sad faces in ASD (Swartz et al.
2013). Yet another study investigated oxytocin-induced activation, i.e. a crucial hormone in affective processing, in the vmPFC and pointed to an oxytocin-induced activation increase in the vmPFC and that this effect furthermore improved socio-communication difficulties in ASD (Aoki et al.
2015). Thus, our findings of decreased FC
APF–vmPFC strength are consistent with the known abnormalities along the amygdala-vmPFC pathway in ASD.
The amygdala is known to be a complex subcortical structure with many efferent and afferent subcortical and cortical as well as intra amygdala connections. Therefore, investigating amygdala functional connections using the entire amygdala does not account for its complex underlying pathways. One more fine grained method to investigate abnormal amygdala functional connections is to parcellate the amygdala based on anatomical subregions within the amygdala (Rausch et al.
2016; Ball et al.
2007; Roy et al.
2009) to assess subregion specific abnormalities. This method is however restricted to predefined anatomical ROI’s based on healthy adult brains, which furthermore are associated to multiple functional pathways. In the current study, therefore we parcellated the amygdala into three functionally defined network parcels based on known amygdala SBNs within our own adolescent sample. This way, we established functionally meaningful amygdaloid subregions in our control sample, which were used to characterize the FC within three important amygdala SBNs in controls and our ASD group.
One potential limitation of our study is the use of a small homogenous sample. Our ASD sample does not include individuals with highly prevalent co-morbidities in ASD such as anxiety, depression or ADHD. No individuals with PDD–NOS were included in our sample. Therefore, our results only provide evidence for autistic core features. Although there was no trend towards group effects in FC
AOF–lOFC (
p = .682) or FC
AAC–cACC (
p = .162) strength, future work involving larger ASD samples may be able to stratify the FC of the SBNs according to age, gender, and symptoms (Murphy and Spooren
2012). In addition, in order to maximize sensitivity of social measures for predicting functional volume, AQ measures may be complemented with interview data (Vineland) and observational measurements (ADOS), which could not be included in the present work as they were deemed too demanding for the ASD group who had already been diagnosed at the time of the study.
Our results demonstrate that functional amygdala parcellation based on its FC with three major amygdala SBNs is a sensitive measure for capturing the functional architecture of dysfunctional amygdalocortical pathways in ASD. Within the three SBNs that were investigated within this study, our results suggest that underconnectivity between amygdala and prefrontal vmPFC is driving abnormal functional interactions between the amygdala and other amygdala networks. By parcellating the amygdala functionally into volumes pathophysiological mechanisms along the amygdalo-prefrontal pathway could be linked to increasing symptom severity in ASD.