Background
Brain lateralization occurs during typical development [
1]. Many reports exist of lateralized brain function underlying cognitive and behavioral processes, such as memory [
2] and emotional processing [
3]; however, the two most common reports of lateralized brain function are in relation to language and visuospatial processing [
4‐
6]. Most typically developing individuals have significant left lateralization in language regions [
7] and right lateralization in attentional regions [
4].
Two recent reports describe how lateralized brain function segregates into two broad networks - a right- and left-lateralized network - in typical development [
8,
9]. The left-lateralized network appears to participate more in intrahemispheric connections, while the right-lateralized network participates in connections between hubs of the network and brain regions in both hemispheres [
9]. In one report, the broad networks include 20 lateralization hubs, nine in the left-lateralized network and 11 in the right-lateralized network. The left-lateralized network includes core language regions (Broca and Wernicke areas) and regions of the default mode network (posterior cingulate cortex, medial prefrontal cortex, and lateral temporal parietal junction, among other areas) [
8]. The right-lateralized network includes regions from three networks associated with attention to external stimuli: the dorsal and ventral attention networks and the frontoparietal executive network.
Atypical lateralization in brain structure and function is associated with neuropsychiatric conditions and developmental disorders such as autism, schizophrenia, and specific language impairment [
10‐
16]. More specifically, autism is associated with reduced left lateralization or reversed lateralization of brain structure and function in core language regions and the white matter tracts that connect them. Abnormal brain lateralization in autism has been measured by multiple techniques, including magnetic resonance imaging (MRI) [
10,
17,
18], functional MRI [
11,
19,
20], diffusion imaging [
14,
15,
21], positron emission tomography [
22‐
24], and electroencephalography [
25‐
27]. It has been reported throughout the lifespan in infancy and childhood [
20,
24‐
29], adolescence [
20], and adulthood [
22,
23]. Lateralization of brain function correlates with language ability in individuals with autism [
25].
In contrast with reports of abnormal lateralization restricted to language-related regions, autism is more generally characterized by connectivity abnormalities across many large-scale brain networks. Abnormal connectivity observations in autism have been made using both functional and structural connectivity analyses [
30,
31]. Long-range connections between distributed connections are underconnected in autism [
32], although reports of overconnectivity also exist [
33]. The abnormal connections are found in default mode, motor, social, language, face processing, and salience networks, among others [
30,
34‐
48]. These core findings have been confirmed in a multisite dataset with over 1,000 subjects [
49,
50]. These studies suggest the pathophysiology of autism includes widespread deficits across structural and functional networks, rather than deficits confined to a single brain region.
The majority of reports on brain lateralization in autism focus on abnormal lateralization in language-related regions. It is unclear whether this is because language is typically associated with lateralized brain function and language impairment is a core feature of autism, or because abnormal lateralization in autism is truly most pronounced in language-related regions. To answer this question, Cardinale and colleagues (2013) characterized whether functional lateralization abnormalities in autism existed outside of language-specific regions. They found diffuse differences across many different functional networks [
51]. These widespread differences in functional lateralization existed in a small sample of children and adolescents (n = 20 for both groups), using independent component analysis to identify the functional networks. In light of Cardinale and colleagues’ findings and the widespread connectivity differences in autism, we hypothesized that lateralization abnormalities would be present across multiple networks.
In the present study, we investigated the 20 lateralization hubs that form the two lateralized networks reported previously [
8], using a multisite dataset with over 1,000 participants. We studied whether the lateralization of brain function differs between autism and typical development in a diffuse, network-wide manner or within isolated language-related brain regions. We also investigated whether lateralization of brain function correlates with clinical severity, age, and handedness.
Results
We investigated the lateralization patterns among the lateralization hubs of the left- and right-lateralized networks in typical development and autism, and then compared the lateralization patterns of the two groups. In the typically developing group, strong lateralization existed between the hubs of the left- and right-lateralized networks, respectively (Figure
2A). The hubs in the right hemisphere are part of right-lateralized connections that form a right-lateralized network. The hubs in the left hemisphere are part of left-lateralized connections that form a left-lateralized network. In the autism group, lateralization between the hubs also existed, although not as strongly as in the typically developing group (Figure
2B). When comparing the two groups, the majority of the differences existed in connections involving specific left-lateralized hubs (Figure
1 and Figure
2C). Only three of the connections survived multiple comparisons correction using a false discovery rate of
q <0.05. The three connections were in the left-lateralized network: the Wernicke area to the posterior cingulate cortex; the Wernicke area to the temporoparietal junction; and the Broca area to the posterior cingulate cortex. All three either lacked left lateralization or had greatly diminished left lateralization in the autism group compared to the typically developing group (Wernicke-posterior cingulate:
t(961) = 3.36,
P = 0.0008; Wernicke-temporoparietal:
t(962) = 3.30,
P = 0.001; Broca-posterior cingulate:
t(960) = 3.04,
P = 0.002).
We also repeated the analyses that identified the group differences in lateralized functional connections, using seven additional inclusion criteria to determine which subjects would be included in the analysis (Table
1). The connections that were most consistently abnormal in the autism group involved the Wernicke area and the posterior cingulate cortex (abnormal in criteria A to D, F, and G of Table
1) and the connection involving Wernicke area and temporoparietal junction (abnormal in criteria A to H of Table
1). Five other connections were abnormal in at least one of the seven inclusion criteria analyses, all involving core language regions and default mode regions in the left-lateralized network (Figure
1B).
Next, we compared the degree of lateralization for three groups of connections (that is, left-lateralized connections involving language regions, left-lateralized connections not involving language regions, and right-lateralized connections) in typical development and autism separately. In typical development, the left-lateralized connections (both connections involving language regions and connections not involving language regions) were more left-lateralized than the right-lateralized connections were right-lateralized (Language: t(514) = 3.97, P = 0.00008; Non-language: t(514) = 2.77, P = 0.006). The left-lateralized connections involving language regions were slightly more left-lateralized than the left-lateralized connections not involving language regions, although the difference was not significant (t(514) = 1.84, P = 0.07). In contrast, the autism group’s left-lateralized connections not involving language regions were more left-lateralized than the left-lateralized connections involving language regions (t(440) = 2.90, P = 0.004). Also, the left-lateralized connections involving language regions were as left-lateralized as the right-lateralized connections were right lateralized (t(440) = 0.39, P = 0.70); whereas, the left-lateralized connections not involving language regions were more left-lateralized than the right-lateralized connections were right-lateralized (t(440) = 3.35, P = 0.0009).
Finally, we investigated the relationship between lateralization in the three abnormal connections and autism severity, age, and handedness. We observed a trend toward less left lateralization in the connection between Wernicke area and the posterior cingulate cortex with increased autism severity (
r(314) = -0.10,
P = 0.07; Figure
3). If control subjects are included for whom ADOS scores were available, these results are more significant (
r(346) = -0.13,
P = 0.017). No significant relationships between lateralization and age or lateralization and handedness were found in either group.
Discussion
In this study, we tested brain lateralization in autism using functional connectivity MRI and found that abnormal lateralization of functional connectivity during rest in autism is most pronounced in specific left-lateralized connections that involve language regions (that is, Broca area and Wernicke area) and regions of the default mode network (that is, temporoparietal junction and posterior cingulate cortex), rather than diffusely affecting either the left- or right-lateralized functional networks. We also replicated previous results in the typically developing group that two interconnected lateralized networks exist in the brain, one in the left hemisphere, and one in the right hemisphere, with the left-lateralized network involving language and default mode regions, and the right-lateralized network involving brain attentional regions [
8].
Cardinale and colleagues found that abnormal lateralization in autism existed across many intrinsic networks, including primary sensory and higher-level association networks [
51]. We, too, found either a lack of left lateralization or greater right lateralization in the autism group; however, the regions or networks involved in abnormal lateralization differed. Rather than finding abnormalities throughout a number of networks as Cardinale and colleagues did, we only found significant differences after multiple comparison corrections in a handful of connections involving language regions and regions of the default mode network. Cardinale and colleagues did find lateralization in the default mode network in some of their supplemental analyses; however, they did not directly test lateralization between language regions and default mode regions.
The inconsistent results may reflect differences in the sample age, sample size, number of data acquisition sites, and/or data analysis methods. In Cardinale
et al., aggregate network measures were studied that pooled information across many ROI’s, whereas the present study used a more spatially localized approach tailored to study individual ‘connections’ between discrete brain regions. It is possible that subtle or subthreshold differences in lateralization in regions of the brain distinct from core language hubs, when pooled across entire functional networks, yield significant lateralization differences that may not survive rigorous statistical testing when evaluating small discrete ROI’s. In fact, we find this likely given the results of Figure
2, in which many more connections, including some that are not associated with language regions, exhibit decreased lateralization with
P <0.05. Virtually all of these show decreased lateralization in autism. Given the consistent direction of the effect, it seems probable that when pooled together, these connections may result in more widespread network differences in lateralization. Nevertheless, our results suggest the effect is much stronger in core language and default mode regions and our approach allows a more spatially localized assessment of effect size.
Neither our study nor the Cardinale
et al. study found a relationship between abnormal lateralization of intrinsic networks and social or communication impairments that survived multiple comparisons [
51]. This corresponds with variable relationships found between abnormal brain lateralization and functional connectivity in general. In individuals with autism, reduced functional connectivity within the default mode network relates to more social and communication impairments [
34,
41,
45‐
47]; however, other studies found no relationship between activation patterns or abnormal lateralization and autism severity or language ability [
9,
19].
The abnormal lateralization of connections involving regions of the default mode network and core language regions may represent an overall lack of specialization in brain regions that process language and social stimuli. Regions of the default mode network are involved in tasks that require language (for example, internal narrative and autobiographical memory) and theory of mind or understanding of another’s mental state [
60‐
62]. The temporoparietal junction and posterior cingulate cortex participate in the same component as core language regions during a language task [
63]. The temporoparietal junction participates in both semantic tasks and deactivates during cognitively taxing tasks (that is, has default mode characteristics) [
64]. The posterior cingulate cortex is more active in congruent and coherent language compared to incongruent or incoherent language [
65,
66]. The right inferior frontal gyrus is more active in autism compared to typical development during a language task, implying abnormal lateralization in a core language region that may have implications in its relationship with other brain regions (for example, as we found with the connection between Broca area and posterior cingulate cortex) [
67]. Together these observations suggest the abnormal lateralization between core language regions and default mode regions could account for some of the communication and social deficits experienced by individuals with autism. This possibility is also supported by findings that abnormal lateralization in language regions are correlated with decreased function on standardized testing [
9].
Reports of abnormal functional lateralization in specific language impairment correspond with previous reports in autism and the present study. Individuals with specific language impairment have less left-lateralized activation in Broca and Wernicke areas during speech tasks [
16,
68]. Individuals with developmental dyslexia also have less lateralization across the left hemisphere, as assessed by functional transcranial Doppler ultrasound [
69]. One study of note, however, found somewhat different results [
9]. It compared individuals with a history of specific language impairment but lacked a current diagnosis, individuals with a current diagnosis of specific language impairment, individuals with autism, and typically developing individuals. Over 80% of the individuals with a current diagnosis of specific language impairment showed right lateralization or bilateral activation during a language task, whereas over 90% of the individuals from the other three groups showed left lateralization. From this study, it appears abnormal lateralization is even more specific to individuals with a current diagnosis of specific language impairment.
The observation that abnormal functional lateralization in autism is most pronounced in connections between core language regions constrains hypotheses of developmental pathophysiology in autism. Our analysis suggests that abnormal language lateralization in autism may be due to abnormal language development rather than a deficit in hemispheric specialization of the entire brain, and would be more consistent with a search for mechanisms involving brain substrates for language acquisition rather than earlier potential mechanisms where hemispheric asymmetries emerge. This constraint is also supported by multimodal observations from DTI, functional MRI, structural MRI, and electrophysiologic studies that have all identified specific deficits in language-related lateralization but not differences in lateralization in other cognitive subsystems.
While the large sample size of the ABIDE dataset can be a tremendous advantage for improving statistical power and external generalizability of the results, it can also be a liability. The individual sites differ in many important data acquisition variables including inclusion criteria, demographics, pulse sequence, scanner type, and length of scan. Most of the included scans were very short, less than 10 minutes duration per subject. It is possible that the heterogeneity of the dataset may limit sensitivity for detecting small changes, and that in a more homogenous data sample, additional differences in lateralization would be found.
An additional limitation is that we did not attempt a discovery of all lateralization differences in an attempt to control the multiple comparison problem that would arise, but instead looked for lateralization differences only between a set of 20 regions that were previously identified as being hubs of lateralized networks in a control population (different from the control subjects used here). It is possible that systematic differences in lateralization are present in brain regions that are not necessarily hubs of lateralization networks in the brain, and which we could not detect. It is also possible that control and autism groups differ in precise spatial coordinates of some lateralization hubs, which we would not be able to detect.
Acknowledgements
The analysis described was supported by NIH grant numbers K08MH092697 (JSA), R01MH084795 (JEL, PTF, NL), R01MH080826 (JEL, ALA, NL, EDB), T32DC008553 (JAN), the Flamm Family Foundation (JSA), the Morrell Family Foundation (JSA). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Mental Health or the National Institutes of Health. Funding sources for the ABIDE dataset are listed at fcon_1000.projects.nitrcc.org/indi/abide. We thank Alyson Froehlich, Molly Prigge, Jason Cooperrider, Anna Cariello, and Celeste Knowles for their contributions to this project. We also sincerely thank the children, adolescents, and adults with autism, the individuals with typical development, and all the families who participated in this study. Finally, we would like to thank the anonymous reviewers whose thoughtful comments and questions greatly improved the manuscript.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
JN participated in the design of the study, performed the analyses, and wrote the manuscript. BZ helped in acquiring the data, participated in the design of the study, and helped to draft the manuscript. PF helped in acquiring the data, participated in the design of the study, and helped to draft the manuscript. AA helped in acquiring the data, participated in the design of the study, and helped to draft the manuscript. NL helped in acquiring the data, participated in the design of the study, and helped to draft the manuscript. EB helped in acquiring the data, participated in the design of the study, and helped to draft the manuscript. JL helped in acquiring the data, participated in the design of the study, and helped to draft the manuscript. JA participated in the design of the study, performed the analyses, and wrote the manuscript. All authors read and approved the final manuscript.