Language networks in Autism Spectrum Disorder: A systematic review of connectivity-based fMRI studies

Abstract

Autism Spectrum Disorder (ASD) is a heterogeneous condition associated with differences in functional neural connectivity relative to neurotypical (NT) peers. Language-based functional connectivity represents an ideal context in which to characterize connectivity because language is heterogeneous and linked to core features in ASD, and NT language networks are well-defined. We conducted a systematic review of language-related functional connectivity literature on individuals with ASD using PubMed, PsychInfo, Scopus, ProQuest, and Google Scholar, yielding 96 studies. Language-task studies indicated local over-connectivity within the language network and global under-connectivity of language with out-of-network regions in ASD. Resting-state studies showed mixed patterns, and connectivity was associated ASD symptomology and language skills. This evidence indicates language-task elicited local over-connectivity and global under-connectivity in ASD, but not a local versus global distinction of resting-state language-related connectivity. Associations with behavior suggest that local over-connectivity and global under-connectivity characterize ASD, and heightened language-related connectivity may support social function.

Introduction

Autism Spectrum Disorder (ASD) is a heterogeneous neurodevelopmental condition characterized by difficulty in social interaction and the presence of restricted and repetitive interests and behaviors (American Psychological Association, 2013). Mounting evidence suggests that ASD is associated with differences in functional neural connectivity relative to neurotypical (NT) peers (Di Martino et al., 2013; Muller et al., 2011; Vasa et al., 2016; Vertes & Bullmore, 2015). Functional connectivity is brain activity in two or more regions that shows a statistical relationship over time. This statistical relationship implies co-engagement or communication through anatomical connections that coordinate neuronal firing. In the Hebbeian ‘cells that fire together, wire together’ sense, connectivity is taken as a reflection of enduring and systematic relationships forming neural circuitry or networks (Vertes & Bullmore, 2015). Accordingly, functional connectivity quantification methods may provide a more nuanced and informative approach for understanding neurodevelopmental disorders compared to assessments of gross anatomy or functional activation (Insel et al., 2010; Koshino et al., 2005; Vasa et al., 2016; Vissers et al., 2012). Complex cognitive processes, such as language, involve distributed networks of numerous discrete brain regions (Just et al., 2004; Sporns et al., 2004; Vissers et al., 2012). Connectivity-based analyses can describe how these neural networks differ in typical versus clinical populations like ASD, building on decades of studies that focus on activation (Sporns et al., 2000; Vissers et al., 2012; Wang et al., 2021).

One of the more influential and theoretically impactful ideas to come from connectivity studies and prior reviews of this emerging scientific literature for ASD samples is a distinction between over-connectivity and under-connectivity. Researchers propose both over-connectivity (i.e., hyperconnectivity; Fishman et al., 2014; Yerys et al., 2015) and under-connectivity (i.e., hypoconnectivity; Just et al., 2007; Kana et al., 2007; Koshino et al., 2008; Muller et al., 2011; Verly et al., 2014) as the primary characterization of functional neural networks in ASD. Prior reviews have been relatively general in nature (e.g., Muller et al., 2011; Vissers et al., 2012), asking whether ASD is associated with over- versus under-connectivity, and have not yielded definitive evidence. The rapidly-growing literature on functional connectivity in ASD includes multiple new findings which call for an updated critical review. This systematic review examines functional connectivity, measured using functional magnetic resonance imaging (fMRI), as it pertains to language function in individuals with ASD. Language function is heterogeneous in ASD (Eigsti et al., 2011; Kjelgaard & Tager-Flusberg, 2001), and linked with core ASD features (e.g., social communication; Blume et al., 2021) and co-occurring conditions (e.g., developmental language disorder; Wittke et al., 2017). There is an extensive, albeit disjointed, literature examining language-related functional connectivity in ASD (Vissers et al., 2012; Herringshaw et al., 2016). Thus, language-related functional connectivity represents an ideal context in which to situate our focused characterization of over- versus under-connectivity in ASD.

Functional Connectivity and ASD

Connectivity studies have led to interesting and useful theories of neural connectivity differences in ASD relative to NT peers (Di Martino et al., 2013; Muller et al., 2011; Vasa et al., 2016; Vertes & Bullmore, 2015). Seminal work by Casanova et al. (2002) on cellular columnar structure and a subsequent fMRI study by Just et al. (2004) led to theories of local over-connectivity and global under-connectivity in ASD. Since this seminal work, many studies have documented relatively stronger statistical associations among brain activity time courses in ASD relative to NT peers within narrowly defined neural networks (local over-connectivity). These studies also document the opposite, relatively weaker connectivity of large-scale networks characterized by distal connections across the brain (global under-connectivity; Carper & Courchesne, 2005; Courchesne et al., 2001; Courchesne & Pierce, 2005; Just et al., 2004; Keown et al., 2017). Careful consideration of this literature, however, reveals inconsistency in how local versus global connectivity is operationalized (Herringshaw et al., 2016; Vasa et al., 2016; Vissers et al., 2012; Wilson et al., 2007).

Over- versus Under-Connectivity

Some studies define “local” connectivity among closely proximal subregions of similarly-organized cortex (e.g., within the inferior frontal gyrus [IFG]), while others define it for regions within an entire lobe or network (e.g., within the default mode network [DMN]). Likewise, some studies define “global” connectivity within a lobe, hemisphere, or network, whereas others define “global” connectivity as across the whole brain. Vissers et al.’s (2012) meta-analysis of structural and functional connectivity in ASD offered an organizational framework that defined local connectivity to be within one cubic centimeter, involving relatively short range association fibers, and global connectivity as connections greater than one cubic centimeter, involving relatively long range association fibers. Although useful for a general synthesis of connectivity studies, these spatial distance-based definitions do not readily map onto presumed neural networks or onto tests of connectivity as it pertains to cognitive processes, like social skills or language. Strongly inter-connected neural networks commonly engage distal brain regions, such as the IFG and posterior STG (pSTG) within the language network, along with proximal regions, such as the superior temporal gyrus (STG) and superior temporal sulcus (STS) within the language network.

In an effort to synthesize a more focused literature on connectivity, Wang et al. (2021) conducted a meta-analysis of resting-state connectivity in the DMN in ASD. Their synthesis indicated over-connectivity of visual, motor, and social-emotional regions within the DMN (e.g., right middle temporal gyrus [MTG], right supramarginal gyrus [SMG], cerebellum) and under-connectivity primarily associated with the angular gyrus (AG) within the DMN. These findings appear to simultaneously suggest spatially-distant global over-connectivity and spatially-proximal local under-connectivity. In contrast, Vissers et al. (2012) demonstrated robust evidence of spatially-distant global under-connectivity related to the DMN and among regions associated with social-emotional processing (e.g., the insula and somatosensory regions), but found little evidence of spatially-proximal local over-connectivity. Muller et al. (2011) also reviewed connectivity in ASD, distinguishing between studies showing under-connectivity versus studies showing over-connectivity, mixed over-/under-connectivity, or no evidence of under-connectivity. Similar to Vissers et al. (2012), Muller et al. reported that the majority of studies provided evidence of under-connectivity, and suggested that variance in findings reflects methodological differences, such as under-connectivity being associated with task-dependent function. It is possible that the predominant patterns of under-connectivity in Vissers et al. (2012) and Muller et al. (2011) reflect the more general nature of these reviews whereas the patterns of under- and over-connectivity in Wang et al. (2021) reflect the DMN-focused nature of their review.

Many individual studies of functional connectivity in ASD report over-connectivity within neurotypical networks and under-connectivity among regions outside of neurotypical networks (Fishman et al., 2015; Keown et al., 2017; Nebel et al., 2014; Rudie et al., 2012; Yerys et al., 2015), as well as atypical patterns of connectivity not observed in NT peers that do not readily lend themselves to an over- versus under-connectivity interpretation (Benkarim et al., 2020; Bolton et al., 2018). It is difficult to determine from the current body of literature whether patterns of local versus global connectivity in ASD reflect spatial-distance or functional-network definitions because individual studies do not clearly or consistently define local versus global. To our knowledge, the most recent synthesis of this literature that applies any local versus global definition is Vissers et al. (2012). In the current review, we operationalize local versus global connectivity in terms of a well-studied neural network. This approach aligns with Wang et al.’s (2021) focused meta-analysis on connectivity related to the DMN in ASD and allows us to describe Vissers et al. (2012) finding of spatially-distant global under-connectivity as within versus between networks of theoretical and clinical interest. This approach will clarify whether findings reflect local versus global connectivity, allowing us to draw stronger theoretical conclusions from the current literature base.

Language-related Functional Connectivity in ASD

To adopt a functional network-based definition of local versus global connectivity, we contextualize our review within the construct of language. Language is an area of theoretical and clinical significance in ASD because language skills vary widely in individuals with ASD (Eigsti et al., 2011; Kjelgaard & Tager-Flusberg, 2001; Lugnegard et al., 2011; Wittke et al., 2017). There is also emerging evidence that individual differences in functional neural patterns are associated with relative language level (Larson et al., 2022; Verly et al., 2014). In neurotypical individuals, the language network is well defined, involving a similar set of regions across studies. Core language regions include left hemisphere superior, middle, and inferior temporal regions, and the IFG, AG, SMG, and TPJ (Briggs et al., 2018; Price, 2010). While there are additional regions involved in language production, such as the supplementary motor area and cerebellum (Price, 2010), these regions are also involved in non-language motor function, such as bodily movements, and therefore not included in the core language network for the purposes of this review. The consistency of brain regions identified as core language regions allows us to examine the language network and its local connectivity (e.g., between the IFG and STG), as well as global connectivity between language network regions and out-of-network regions (e.g., IFG and occipital regions) across studies.

The Current Review

A systematic integration of the current literature on language-related functional connectivity in ASD is the next logical step in refining theories of local over-connectivity and global under-connectivity (Kana et al., 2006; Lee et al., 2017; Verly et al., 2014). In contrast to the great difficulty in conducting a quantitative analysis of widely varying methods, this qualitative synthesis draws on numerous studies to identify general patterns. This approach also allows us to leverage detailed information, such as task design and regions analyzed, to clarify inconsistencies and develop insights for the field. To achieve our goal of clarifying language-related functional connectivity in ASD, we focus on fMRI to reduce unmeaningful sources of variance (e.g., Vissers et al., 2012). Our pre-registered (https://osf.io/sgc7a) questions were: (1) Are language-related neural networks characterized by over-, under-, or mixed-functional connectivity in ASD relative to NT peers? (2) How does network connectivity reflect (a) ASD symptomology and (b) language-related behavior?

In our synthesis, the following topical organization became important for resolving lack of clarity in the evidence base. First, we separately review studies of task-based connectivity and studies of resting-state connectivity (i.e., no overt task demands), aligning with prior reviews that reported meaningful differences in connectivity between task and rest (Muller et al., 2011; Stevens, 2016; Vissers et al., 2012). Within the task and rest sections, we summarize local connectivity findings within the language network and global connectivity findings between language and out-of-network regions. For the task section, we also organize task-based studies according to task paradigms because findings varied depending on how task features elicited connectivity. Subsequently, we review studies examining functional connectivity lateralization, a special case of connectivity that tests the theory that neural specialization for language involves increasing left hemisphere dominance for language function (Olulade et al., 2020). This section is followed by a summary of associations between functional connectivity and behavior on measures of ASD symptomology and language. These studies test the degree to which neural networks are associated with complex cognitive processes, enhancing our theoretically and clinically informed understanding of neural networks in ASD. Two additional topics became relevant to the current synthesis: psychiatric comorbidities and developmental considerations. There was a surprising lack of studies examining psychiatric comorbidity in ASD, which is in contrast to the Research Domain Criteria framework (RDoC; Insel et al., 2010). This framework recommends examining dimensions of neurobiobehavioral functioning in psychopathology, such as social, cognitive, and affective functioning. RDoC also recommends examining neurobiobehavioral functioning across the age span, and there are several studies that provide cross-sectional information on connectivity at different stages of development. Lastly, we review current limitations in the evidence base, provide conclusions, and offer field-directing insights.

Methods

Our methodological approach aligned with PRISMA and Cochrane (2021) reporting standards. We performed literature searches on 10/27/2021 using the following databases: PubMed; PsychInfo; Scopus (international); ProQuest (grey literature, such as dissertations); and Google Scholar (grey literature, such as preprints). Broadly, our search terms were: Autism Spectrum Disorder; fMRI; and Language (see Supplementary Materials 1 for complete syntax). Our inclusion criteria were: participants must have Autism per DSM criteria or community diagnosis with validation using the Autism Diagnostic Observation Schedule (ADOS), Autism Diagnostic Interview (ADI), and/or Modified Checklist for Autism in Toddlers (MCHAT); methods must include functional magnetic resonance imaging (fMRI); methods must analyze connectivity of language-related brain regions or use a language task in the scanner (see Supplementary Materials 1 for complete inclusion criteria). Although we did not restrict year of publication, included articles were published in 2004 (Just et al., 2004) or later. Three screeners (not the authors) unfamiliar with research questions conducted preliminary title/abstract/keyword screening to ascertain that all studies met inclusion criteria. Following preliminary screening, two different raters (the first and second authors) conducted full-text screening and quality assessments as detailed below. Included studies were assessed for quality using the following parameters based on the GRADE Handbook (Cochrane, 2021): validity (e.g., ASD eligibility, group matching, methodological control); reliability (e.g., sample size, measurement precision, statistical approach); and applicability (e.g., relationship of sample, comparisons, and results to research questions).

Following the preliminary screening, the detailed assessment process included an initial training and consensus phase, followed by an assessment of reliability, with five articles per screener at each phase. The reliability for title/abstract/keyword and full-text screening was 100% and quality rating reliability was 84% overall (5 articles). Reliability reached 90% after removing one parameter (conflict of interest) where raters reached the same conclusion, yet marked a different level of quality. The quality raters (first and second authors) reached consensus on level assignment for the risk of bias parameter prior to continuing quality ratings. See Supplementary Materials 1 for complete methods, including quality rating forms, and extracted meta-data, and Supplementary Materials 2 for complete information on included articles. Note that all studies compare groups of individuals with ASD to NT individuals, except as indicated, and we refer to over- versus- under-connectivity as the difference in ASD relative to NT group functional connectivity. The term ASD encompasses any diagnosis on the autism spectrum (e.g., Asperger’s Syndrome). See Figure 1 for PRISMA flow chart of literature search and results, and Tables 1 (task-based) and 2 (resting-state) for included articles most relevant to language network connectivity.

Figure 1.

Flowchart of screening and inclusion.

*Some studies reported matching race/ethnicity between groups but did not report race/ethnicity makeup of their sample.

Table 1.

Language-task based studies.

Citation	N (mean years)	Task	Analysis	Results – language-related brain regions and networks	Results – task performance and association with behavioral measures	Quality Rating
Mash et al., 2021	14 ASD 6 NT	lexical decision	Seed; thalamus with within-hemisphere regions	Over-connectivity: during language task thalamus collapsed across right AG, SMG, and occipital pole, but no ROI-ROI pairwise differences Network: greater variance in connectivity during resting-state in ASD than NT peers	ASD 88.0% < NT 94.7% accuracy	3.82
Sridhar et al., 2021 preprint (Sridhar, 2021 dissertation)	30 ASD (15) 23 NT (15)	word category	Matrix; Reconfiguration	Over-connectivity: during task and resting-state; greater reconfiguration in ASD than NT peers and in ASD subgroups with higher versus lower in-scanner task performance Network: greater variance in resting-state and task-based connectivity in ASD than NT peers	ASD 80–91% < NT 90–97% accuracy; examined subgroups matched on accuracy and participants with <60% accuracy excluded Reconfiguration of connectivity with higher language scores on CELF word classes	3.91
Lai, 2011 dissertation	12 ASD with LI (non-sedated) & 27 ASD-sedated (12) 12 NT (12)	auditory stimuli presentation: speech, song, music	Seed; IFG	Over-connectivity: left IFG with left STG and AG during song relative to speech combined-ASD vs. NT Network: broader connectivity in left temporal regions (pSTG) in NT than ASD for speech relative to song.		3.09
Shen et al., 2012	14 ASD (24) 14 NT (24)	word category	Seed; significantly active regions; Structural equation modeling with IFG, MTG, extrastriate cortex	Over-connectivity: left IFG with several regions, including left temporal (STG), right frontal, and occipital regions; extrastriate with frontal regions; subregion information in supplemental Network: SEM greater involvement of extrastriate cortex in semantic network in ASD	ASD 79.7% < NT 89.6% accuracy Connectivity of left IFG and extrastriate with lower performance	3.73
Hegarty, 2015 Dissertation	13 ASD (22) 13 NT (23)	fluency language task	Graph theory; whole-brain and networks; DMN and semantic network	Over-connectivity: within fronto-parietal network and semantic association network	ASD M words produced = 28 < NT M words produced = 44	3.45
Just et al., 2004	17 ASD 17 NT	sentence comprehension	Seed; IFG, STG, prefrontal	Under-connectivity: across ROI pairs, left IFG, and dorsolateral prefrontal cortex Network: lower synchrony in ASD than NT peers	ASD error rates 8/13%; NT 5,/7% error rates in active/passive conditions NS	3.73
Kana et al., 2006	12 ASD (23) 13 NT (20)	sentence truth value with high/low imagery manipulation	Seed; task-related activation; Networks	Under-connectivity: greatest between frontal and parietal regions, not temporal or occipital	NS in-scanner task error rate	3.73
Jones et al., 2010	17 ASD (16) 20 NT (17)	word generation with letter/category manipulation	Seed; significant condition difference seeds	Under-connectivity: left IFG with left fusiform gyrus; left IPL, right precentral gyrus, left precentral gyrus with left fusiform gyrus and right precentral gyrus; left STG with left fusiform gyrus and right precentral gyrus. Network: less variance in right pCC and medial prefrontal cortex in ASD than in NT peers	Accounted for behavioral performance in analyses	3.82
Bednarz et al., 2017	18 ASD (10) 13 NT (10)	word similarities	Seed; reading network	Under-connectivity: Left IFG with left inferior occipital gyrus	NS in-scanner task accuracy Connectivity of left IFG and left thalamus with higher reading comprehension performance	3.64
Fan et al., 2021	30 ASD (10) 30 ASD (15) 33 NT (10) 34 NT (15)	semantic decision language task	Seed; cuneus, IFG, whole-brain	Under-connectivity: left IFG with MTG adolescents and cuneus with MTG in children	ASD < NT accuracy Connectivity of MTG and cuneus with fewer ASD features on the ADI verbal scale.	3.73
Williams et al., 2013	15 ASD children (13) 14 NT children (13) 13 ASD adults (25) 12 NT adults (21)	narrative reading literal/irony manipulation	Seed; activation-based	Under-connectivity: within language network during irony condition No group differences: ROI-ROI pairwise connectivity or for theory of mind network	ASD 11.7/28%; NT 7.1/21.2% child error rates in literal/irony conditions, NS ASD 5.9,/21.2%; NT 3.7/5.5% adult error rates in literal/irony conditions, NS literal condition, significant group difference irony condition	3.91
Mason et al., 2008	18 ASD (27) 18 NT (27)	3-sentence story comprehension, inferencing manipulation	Seed; language network	Under-connectivity: within theory of mind network (left MFG, right TPJ); between language and theory of mind networks, particularly left MFG with language network (left IFG, MTG)		3.64
Jasmin et al., 2019	19 ASD (19) 20 NT (21)	conversation with experimenter	Seed; whole-brain connectedness seeds	Over-connectivity: right anterior temporal lobe with whole-brain; right IFG with anterior insula; task relative baseline for the left pSTS with left middle temporal cortex, and left middle temporal cortex with right pSTS; subcortical regions at rest Under-connectivity: resting-state relative to task-based, except subcortical regions	NS in-scanner task performance Connectivity of right IFG and precuneus and task relative to baseline for left pSTS and right extrastriate with more ASD features on the SRS	3.91
Chouinard et al., 2017	12 ASD (32) 12 NT (33)	sentence comprehension (metaphor contrast)	Seed; IFG, MTG, STG; Graph theory	Under-connectivity: among seeds, including left IFG, MTG, STG Over-connectivity: among subcortical regions for metaphors	NS in-scanner task performance	3.64
Sharda et al., 2015	22 ASD (11) 22 NT (10)	speech, sung-word, music stimuli	Seed; IFG	Under-connectivity: IFG with anterior insula for sung-word stimuli Over-connectivity: left IFG with left temporal region for speech stimuli; IFG with left cerebellum for sung-word stimuli		3.91
Murdaugh et al., 2012	13 ASD (21) 14 NT (23)	word-comprehension self/other	Seed; pSTG, DMN; Classification	Under-connectivity: left pSTG with visual regions and right MTG; medial prefrontal cortex Over-connectivity: left pSTG with bilateral precentral gyrus, right postcentral gyrus, and right IFG (operculum) Classification Accuracy: 78% Informative connections: left pSTG		3.73
Lombardo et al., 2018	81 ASD (30 mos) 37 NT (26 mos)	speech stimuli: complex, simple, backward	Matrix-gene covariance latent variables	Network: connectivity-gene covariance latent variable distributed across language (STS, IFG), social (DMN), and subcortical (striatum) regions		3.82
Baxter et al., 2019	24 ASD adults (53) 17 ASD adolescents (21) 20 NT adults (50) 14 NT young adults (21)	fluency/word generation	Component - independent	Network: IFG primary contributor; included left temporal lobe, right cerebellum, and supplementary speech area in young adults and left temporal lobe, bilateral cerebellum, and supplementary speech area in adults; AG and subcortical structures in NT but not ASD	NS in-scanner task performance	3.73

Note. NS = no significant group difference.

Table 2.

Language-focused resting-state fMRI studies.

Citation	N (mean years)	Analysis	Results – language-related brain regions and networks	Results – association with behavioral measures	Quality Rating
Dinstein et al., 2011	29 ASD (29 mos) 13 language delay (19 mos) 30 NT (29 mos)	Seed; IFG, pSTG; Classification	Under-connectivity: inter-hemispheric STG and IFG Accuracy: 72% using STG and IFG interhemispheric connectivity	Connectivity of interhemispheric IFG with higher expressive language and fewer ASD features on the ADOS Soc-Comm	3.73
Gabrielson et al., 2018	17 Low verbal/cognitive ASD (12) 20 High verbal/cognitive ASD (13) 19 NT (12)	Network; auditory, homologue	Under-connectivity: right temporo-occipital with the DMN ASD-low versus NT; right fronto-opercular with DMN ASD-low versus NT; fronto-parietal and auditory network ASD-low versus ASD-high; interhemispheric-STG in ASD-low versus ASD-high and ASD-low versus NT	Connectivity within auditory network and between right and left temporal regions with ASD subgroup status	3.82
Li et al., 2019	409 ASD 455 NT (average age approximately 20 years; ABIDE)	Seed; IFG, STG	Under-connectivity: within DMN, medial prefrontal cortex, precuneus, and pCC; regions in the salience, visual, and mirror networks; inter-hemispheric IFG, STG, pCC, precuneus, and in left, but not right hemisphere Accuracy 78–100% Informative connections: interhemispheric connectivity to a greater degree than intrahemispheric connectivity	Connectivity of interhemispheric STG with fewer ASD features on the ADOS Soc-Comm; connectivity of interhemispheric pCC and dorsal posterior insula associated with fewer ASD features on the ADOS Soc-Comm	3.82
Linke et al., 2018	40 ASD (14) 38 NT (14)	Seed; language regions, interhemispheric	Under-connectivity: interhemispheric left and right STG, but not mSTG/mSTS	Connectivity of interhemispheric auditory regions with higher verbal IQ and lower auditory sensory processing scores; connectivity of thalamocortical regions with fewer ASD features on the ADOS repetitive behavior and SRS communication	3.82
Verly et al., 2014	17 ASD (14) 25 NT (14)	Matrix; language network identified by verb generation, used for resting-state analysis	Under-connectivity: interhemispheric language regions (IFG, pSTG) and left IFG with left pSTG at rest	Connectivity of left IFG and left pSTG with higher verbal IQ; connectivity of right IFG and right pSTG with higher receptive vocabulary on the PPVT	3.64
Xu et al. 2020	49 ASD (8–17) 33 NT (8–17) 29 ASD (18–30) 29 NT (18–30)	Seed; MTG subregions	Under-connectivity: left pMTG with right m/pCC and right pMTG with right SFG in children with ASD; left pMTG with right AG and right precuneus, right maMTG with right superior parietal gyrus, and right pMTG with right precuneus and left orbital MFG in adults with ASD	Connectivity of left aMTG and left IFG with ASD features on the ADOS Soc-Comm; connectivity of left aMTG and left IFG increased with age in ASD, but decreased with age in NT peers; connectivity of right maMTG and left superior occipital gyrus increased with age in the ASD, but decreased with age in the NT group; right pMTG and left MFG decreased with age in both ASD and NT peers	4. 00
Gao et al., 2019 (Gao, 2020 dissertation)	52 ASD (14) 50 NT (14)	Network; language network; Lateralization; Effective connectivity (post-hoc)	Over-connectivity: within language network, among left and right IFG, AG, pCC, left SMG, and bilateral dorsal precuneus, and right IPL with left pericentral; inferior frontal seeds with pCC and of pCC with occipital region Effective connectivity: pCC mediated frontal language and visual regions in approximately half of ASD participants No group differences: lateralization or between-hemisphere connectivity	Connectivity of right IPL and left pericentral with fewer ASD features on the ADI social subscale; connectivity of pCC and visual regions and of left pCC and right lingual gyrus with lower CELF scores in ASD subgroup showing pCC mediation of frontal language and visual regions	4.09
Kaminer, 2021 master’s thesis	20 ASD (25) 20 NT (24) (ABIDE)	Seed; language network	Over-connectivity: left SMG	Connectivity of language network and whole-brain with more ASD features on the ADOS Soc-Comm	3.55
Levinson, 2021 dissertation	59 ASD (15) 61 NT (16) (ABIDE)	Seed; temporal regions	Over-connectivity: right ventral anterior temporal lobe with right aSTS, STG, and pCC		3.55
Shih et al., 2011	21 ASD (14) 26 NT (14)	Matrix; pSTS, subregions	Over-connectivity: within pSTS Network: less temporal and spatial differentiation of STS subregions in ASD than NT peers	Lower temporal differentiation of pSTS with more ASD features, but not IQ; similar association with age in ASD and NT peers	3.91
Lee et al., 2017	676 total, 6 groups (<12, 12–19, >20 ASD and NT; each group >100 n)	Graph theory; language network	Under-connectivity: decreased degree centrality (number of edges) of left IFG in children with ASD and of left pSTG in adults with ASD within the language network Over-connectivity: increased degree centrality of left MTG in adults with ASD within the language network	Connectivity of left IFG (degree centrality) associated with fewer ASD features on the ADOS communication scale in children; connectivity of left IFG (degree centrality) associated with fewer ASD features on the ADOS communication scale in adults	3.73
Murdaugh et al., 2015 (related to Maximo, 2016)	16 ASD intervention (10) 15 ASD control (11) 22 NT (10)	Component – independent; Seed-based; reading network, IFG, STG	Under-connectivity: left IFG and insula; left IFG with fusiform gyrus Over-connectivity: right IFG (operculum); left IFG with MFG and left middle occipital gyrus; left pSTG and calcarine sulcus	Connectivity of reading network and left IFG and left pSTG, left IFG and motor regions, and left pSTG and left IFG and MTG with post-intervention group status; connectivity of left pSTG and left IFG (orbitalis), right cerebellum, occipital gyrus/left IFG and bilateral postcentral gyrus and right superior parietal lobule with higher reading performance	3.73
Murdaugh et al., 2012	13 ASD (21) 14 NT (23)	rsfMRI and task-based fMRI word-comprehension language task	Seed-based connectivity using DMN regions: medial prefrontal cortex, posterior cingulate/precuneus, Wernicke’s; resting-state classification accuracy	Under-connectivity: left pSTG with visual regions and right MTG; medial prefrontal cortex Over-connectivity: left pSTG with bilateral precentral gyrus, right postcentral gyrus, and right operculum Accuracy: 78% Informative connections: left pSTG	3.73
Nielsen et al., 2014 (Nielsen, 2013 dissertation)	539 ASD (17) 573 NT (17) (ABIDE)	Lateralization; language regions	Lateralization: less left relative to right in ASD for participants exhibiting left hemisphere lateralization, particularly left pSTG with pCC and TPJ; left IFG to pCC; language with DMN regions; non-language regions more left lateralized in ASD	Left hemisphere lateralization connectivity of left pSTG and pCC with fewer ASD features on the ADOS Soc-Comm, only significant across ASD and NT groups	3.91
Hong et al., preprint 2021	155 ASD (18) 151 NT (18) (ABIDE)	Matrix; whole-brain; Component – principal	Network: altered between network connectivity for the language network; altered within network connectivity for the DMN; patterns of over- and under-connectivity within and between networks	Connectivity of salience network with greater verbal-nonverbal IQ imbalance; connectivity of cingulo-opercular and language/auditory networks associated less with verbal-nonverbal IQ imbalance in ASD relative to NT peers	3.91
Xiao et al., 2021 preprint	52 ASD (2) 34 NT (2)	Component – independent; language regions/homologues	No group differences in connectivity	Connectivity of left temporal and left cuneus regions with fewer ASD features on the VABS Comm scale in ASD; group differences in the association between connectivity of left temporal and bilateral fronto-parietal operculum with higher expressive language and connectivity of left fronto-parietal operculum with higher receptive language (on the Mullen), and connectivity of right temporal-right LP with expressive language; connectivity of right temporal-right LP with VABS Comm scores and of LP and cerebellum for VABS Soc scores	3.82

Results and Discussion

Task-based Connectivity Within the Language Network

Lexico-semantic Tasks

Nine studies probed lexico-semantic connectivity by asking participants to covertly categorize or overtly list words given a prompt, such as categorizing words as “animal” versus “food” (Shen et al., 2012) or generating words starting with the letter “s” (Sridhar et al., 2021 preprint). The majority of this work (4 of 7 studies that tested connectivity within the language network) provided strong evidence of over-connectivity within the language network in ASD in adolescents and adults. This over-connectivity was primarily related to the left IFG (i.e., Broca; ASD n = 24 adults, 17 adolescents, NT n = 20 adults, n = 14 adolescents, Baxter et al., 2019; ASD n = 30, NT n = 15 adolescents, driven by ASD subgroup matched to NT group on in-scanner task performance, Sridhar et al., 2021 preprint), such as with the left STG (ASD n = 14, NT n = 14 young adults, Shen et al., 2012; note that the ASD group had lower in-scanner task accuracy than NT peers) and within a semantic association network (ASD n = 13, NT n = 13 young adults, Hegarty, 2015). However, Fan et al. (2021) reported under-connectivity of their left IFG seed region with the left MTG in their autistic adolescent group, suggesting that within-network connectivity patterns related to the MTG may differ from other regions, like the STG (ASD n = 30 children, n = 30 adolescents, NT n = 33 children, n = 34 adolescents,). Given that the ASD group had significantly lower accuracy during the in-scanner semantic judgement task, connectivity findings in Fan et al. (2021) may reflect errored performance to a greater degree in the ASD than NT group, though performance was above chance on average across groups.

In contrast, Jones et al. (2010) showed no evidence of over-connectivity in ASD (ASD n = 17 NT n = 20 adolescents) when comparing a letter or semantic category word generation condition to a baseline condition where participants recited months of the year. Relatively small differences between conditions would be expected in Jones et al. because both of their task conditions involved generating words. Other paradigms that compare a lexical decision condition to a rest condition (e.g., Baxter et al., 2019) or to perceptual decision (e.g., nonlinguistic characters; Fan et al., 2021; Wilson et al., 2018) are more likely to yield condition effects. Jones et al. (2010) conducted additional analyses regressing out task-elicited variance in connectivity and found under-connectivity in ASD relative to NT peers that reflected slowly-fluctuating connectivity (i.e., connectivity “occurring on top of the average task-related responses” p. 10) rather than slowly-fluctuating connectivity plus transient changes induced by particular task demands (e.g., trial-by-trial stimuli on the screen). This evidence further suggests that a contrast between conditions with similar task demands is less likely to find task-dependent over-connectivity and more likely to find task-independent under-connectivity (see also Muller et al., 2011). Two additional studies did not examine connectivity within the language network (i.e., the reading network, Bednarz et al., 2017; only a thalamus seed region; Mash et al., 2012). Collectively, this evidence demonstrates over-connectivity within the language network in ASD for word-level tasks, with the caveat that the MTG may be under-connected with the IFG in adolescents with ASD.

Sentence Comprehension Tasks

Three studies deployed sentence comprehension or “truth value judgement” tasks, which involve presenting auditory or written sentences and asking participants to indicate via button press whether the sentence is true or false. In contrast to word-level findings, none of these studies demonstrated over-connectivity within the language network in ASD (Chouinard et al., 2017; Just et al., 2004; Kana et al., 2006). Two studies showed under-connectivity within the language network, related to the left IFG, STG, and MTG (ASD n = 12, NT n = 12 adults, Chouinard et al., 2017; ASD n = 17, NT n = 17, age not reported, Just et al., 2004). The exception is Kana et al. (2006), which showed no group differences in connectivity between the frontal and temporal lobe (ASD n = 12, NT n = 13 young adults), likely due to their relatively coarse comparison of entire lobes rather than smaller regions as in Chouinard et al. (2017) and Just et al. (2004).

Conversation and Speech Stimuli Tasks

Three studies presented conversations or narratives in the scanner, as a fully passive task or with a response during or after scanning (e.g., comprehension questions; Williams et al., 2013) and three studies presented speech stimuli in the scanner, such as spoken words versus sung words (Sharda et al., 2015). The majority of this work showed over-connectivity within the language network in ASD relative to NT peers, with the exception of two studies deploying tasks with relatively more social features (e.g., nonliteral language, inferencing). In autistic children and young adults, there was over-connectivity of left middle temporal regions with the left pSTS (i.e., engaging in conversation; ASD n = 19, NT n = 20 young adults, Jasmin et al., 2019) and of the left IFG with left temporal regions (i.e., sung stimuli; ASD n = 22, NT n = 22 children, Sharda et al., 2015). Lai (2011) also showed over-connectivity of the left IFG with left STG and AG during sung words relative to spoken words for a combined group of individuals with ASD with/without co-occurring language impairment and who were/were not sedated (ASD n = 39, NT n = 12 children). Williams et al. (2013) reported under-connectivity within the language network for children and young adults with ASD in a nonliteral narrative condition with relatively high social demands (i.e., irony interpretation), but not in a literal narrative condition with lower social demands (ASD n = 15 children, n = 13 young adults, NT n = 14 children, n = 12 young adults; note that adults with ASD had lower accuracy in the nonliteral condition than NT adults). Similarly, Mason et al. (2008) reported under-connectivity between the left IFG and MTG in a narrative condition with high social demands (e.g., emotional state inferences), but not in conditions with fewer social demands in adults with ASD (ASD n = 18, NT n = 18 adults). One additional study did not examine connectivity within the language network (Lombardo et al., 2018). In sum, the majority of evidence suggests over-connectivity within the language network, with the exception of two studies showing under-connectivity for tasks involving nonliteral language.

Task-based Connectivity Between Language and Other Presumed Networks

Lexico-semantic Tasks

The nine studies probing lexico-semantic function provided strong evidence of under-connectivity of language regions with regions outside the language network, and minimal evidence of over-connectivity. In adolescents and adults with ASD, there was under-connectivity of the left AG with subcortical structures (ASD n = 24 adults, 17 adolescents, NT n = 20 adults, n = 14 adolescents, Baxter et al., 2019), left IFG with the left inferior occipital gyrus (ASD n = 18, NT n = 13 children, Bednarz et al., 2017; ASD n = 14, NT n = 14 young adults, Shen et al., 2012), left IFG and STG with the left fusiform gyrus and motor regions (Jones et al., 2010), the MTG with the cuneus (ASD n = 17 NT n = 20 adolescents, Fan et al., 2021), left pSTG with visual regions and right MTG (ASD n = 13, NT n = 14 young adults, Murdaugh et al., 2012), in a fronto-parietal network (ASD n = 13, NT n = 13 young adults, Hegarty, 2015), and inter-hemispheric language and language-homologue regions (Shen et al., 2012; Verly et al., 2014), such as the left IFG and right pSTG (ASD n = 17, NT n = 25 adolescents, Verly et al., 2014). There was also evidence of broad differences in network engagement in ASD relative to NT peers, potentially reflecting greater reconfiguration (ASD n = 30, NT n = 15 adolescents, driven by ASD subgroup with poorer in-scanner task performance than NT group, Sridhar et al., 2021 preprint) and less disengagement of the DMN in response to task-demands in ASD (Baxter et al., 2019). One study showed over-connectivity of the left pSTG with the bilateral precentral gyrus, right postcentral gyrus, and right IFG (operculum) on a self-versus-other semantic decision task in young adults with ASD (Murdaugh et al., 2012). This study had a relatively small sample size (ASD n = 13; NT n = 14) and a task involving more social demands than other lexico-semantic tasks (see also Mash et al., 2021 for evidence of over-connectivity of the thalamus and right AG, SMG, and occipital pole, though the ASD group had significantly lower accuracy on the in-scanner task than the NT group). Over-connectivity of the language network during lexical tasks may extend to motor and language-homologue regions in the presence of social task demands, yet the majority of evidence indicated under-connectivity between language and out-of-network regions.

Sentence Comprehension Tasks

The three studies employing a sentence comprehension task showed under-connectivity of the IFG with prefrontal non-language regions (ASD n = 12, NT n = 12 adults, Chouinard et al., 2017; ASD n = 17, NT n = 17 age not reported, Just et al., 2004) and frontal with parietal regions (Just et al., 2004; ASD n = 12, NT n = 13 young adults, Kana et al., 2006) in ASD relative to NT peers. Kana et al. (2006) also reported no group differences in connectivity related to temporal or occipital lobes and Chouinard et al. (2017) reported over-connectivity among subcortical regions in response to sentences with nonliteral language.

Conversation and Speech Stimuli Tasks

For the six studies presenting conversation or speech stimuli tasks, there was mixed evidence of over- and under-connectivity between language and out-of-network regions in ASD. The majority of this evidence came from studies employing tasks with relatively high social or music features and studies with different age groups. Mason et al. (2008) found under-connectivity between language and theory of mind networks related to the IFG and MTG for a narrative condition with inferencing demands (ASD n = 18, NT n = 18 adults). Sharda et al. (2015) demonstrated under-connectivity of the left IFG with the right insula, but also over-connectivity of the left IFG with left cerebellum, for sung-word stimuli (ASD n = 22, NT n = 22 children). Jasmin et al. (2019) found over-connectivity between the left middle temporal cortex and right pSTS in response to engaging in conversation with an examiner during scanning (ASD n = 19, NT n = 20 young adults). Lai (2011) showed no group differences in connectivity of the left IFG with out-of-network regions in sung relative to spoken word conditions (ASD-language impairment n = 12 children, ASD-sedated n = 27 children, NT n = 12 children) and Lombardo et al. (2018) and Williams et al. (2013) did not examine connectivity between language and out-of-network regions. There is no clear interpretation from the current evidence on over- versus under-connectivity for conversation and speech stimuli tasks, due in part to differences in task paradigms and participant samples.

Task-based Connectivity Summary

There was a high degree of inconsistency in the definition of local versus global connectivity from these studies and half of studies did not frame their analysis in terms of local versus global connectivity. For instance, Kana et al. (2006) defined local as within a lobe and global as between lobes, whereas Jasmin et al. (2019) defined local as among cortical regions and global as across the whole brain. Shen et al. (2012) note that models of local over-connectivity from prior research imply that cortices (e.g., the auditory cortex and extrastriate cortex) function in isolation, when in fact their study demonstrated connectivity among cortices (e.g., extrastriate and frontal cortex). Some studies provided no clear definition of local versus global connectivity. Lai (2011) and Just et al. (2004) discuss short versus long-range white matter tracts as relevant to connectivity patterns, but do not describe how white matter tracts are associated with cortical connectivity. Fan et al. (2021) briefly discuss a spatial-distance definition of connectivity, but did not define short-range versus long-range connections. Studies that examined networks, such as the language (e.g., Chouinard et al., 2017; Mason et al., 2008; Williams et al., 2013) or reading network (e.g., Bednarz et al., 2017), primarily focus on network-related connectivity rather than discussing the local versus global nature of connectivity. Unsurprisingly, findings from this body of work do not map onto theories of local versus global connectivity defined functionally-engaged networks.

Our application of a network-based definition of local connectivity revealed strong evidence supporting the theory of local over-connectivity and global under-connectivity in response to a language task. These patterns may reflect individual differences in language skills in ASD, particularly complex language skills. Over-connectivity within the language network suggests heightened lower-level processing, such as of phonological features or individual words, and under-connectivity between language and out-of-network regions suggests diminished integrative processing across distributed networks, such as higher-order processing of complex syntax and narratives (Just et al., 2004; Williams et al., 2013). Language-related local over-connectivity and global under-connectivity may therefore account for heterogeneity in language skills in ASD and represent a feature of ASD status. For instance, these findings align with evidence of a subgroup of individuals with ASD who have structural language impairments in morphosyntax (Eigsti et al., 2011; Kjelgaard & Tager-Flusberg, 2001; Wittke et al., 2017). This subgroup also presents with different patterns of language-related neural function relative to ASD peers without structural language impairments (Larson et al., 2022; Verly et al., 2014).

One of set of findings that diverged from these overarching patterns was under-connectivity within the language network in response to nonliteral language (conversation and sentence comprehension) and to passive sentence comprehension (Chouinard et al., 2017; Mason et al., 2008; Williams et al., 2013). This under-connectivity involved connections between the IFG, a core language region, or the MTG, a core social region, with other regions in the language network. Interestingly, the MTG is a key region in the “social brain” (Bolton et al., 2018; Price, 2010) and the social brain is characterized by under-connectivity in ASD (Damarla et al., 2010; Deshpande et al., 2013; Kana et al., 2017; Lee et al, 2009; Periera Quezada, 2019). Mason et al. (2008) varied social features of a narrative task, comparing connectivity in three conditions: physical inferences (direct consequences), emotional inferences (character feelings), and intentional inferences (mental state/theory of mind). For the intentional inference condition, there was under-connectivity between the left IFG and MTG and between a theory of mind region (left MFG) and the MTG. There was no under-connectivity within the language network for the other inference conditions. Though there was some evidence that MTG connectivity may differ in adolescence relative to childhood (Fan et al., 2021) and relative to adulthood (Baxter et al., 2019; Hegarty, 2015), the majority of evidence suggests that MTG patterns reflect social demands of the in-scanner tasks rather developmental stages. It is possible to hypothesize from this evidence that IFG and MTG under-connectivity within the language network reflects social features of language tasks rather than exclusively linguistic features of the tasks.

Resting-state Connectivity Within the Language Network

There were 15 resting-state studies that examined language-relevant seed regions or connectivity of the language network in the absence of any overt task demands. Similar to task-based studies, the evidence indicated over-connectivity within the language network in ASD relative to NT peers, though only five studies found significant effects. There was over-connectivity of the left IFG with left AG (ASD n = 52, NT n = 50 adolescents, Gao et al., 2019), over-connectivity of the left MTG (i.e., degree centrality; total n = 676 children, adolescents, and adults, Lee et al., 2017) and left SMG (ASD = 20, NT n = 20 young adults, Kaminer, 2021) with the rest of the language network, and over-connectivity within STS subregions (ASD n = 21, NT n = 26 adolescents, Shih et al., 2011). One study using the large Autism Brain Imaging Data Exchange (ABIDE) showed under-connectivity (i.e., less degree centrality) of the left IFG in children with ASD and left pSTG in adults with ASD, but over-connectivity of the MTG in adults with ASD, with the rest of the language network relative to NT peers (Lee et al., 2017). However, there were 64 regions comprising Lee et al.’s (2017) language network, including regions associated with visual, motor, and theory of mind function. Specific connectivity of the IFG, pSTG, and MTG was not analyzed within this large set. One additional study showed no differences in connectivity for ASD subgroups with varied verbal and cognitive abilities relative to a NT group (ASD n = 37, NT n = 19 adolescents, Gabrielson et al., 2018) and another study showed under-connectivity within the language network, but only as an interaction effect between group and IQ (i.e., driven by individuals with ASD and relatively lower verbal than nonverbal IQ; ASD n = 155, NT n = 151 young adults, Hong et al., 2021 preprint). Taken together, there is preliminary evidence of resting-state over-connectivity within the language network in ASD.

Resting-state Connectivity Between Language and Other Presumed Networks

Resting-state studies that examined connectivity between language and out-of-network regions in ASD report mixed evidence of over- and under-connectivity. From early childhood through adulthood, language and language homologue regions were predominantly under-connected, including the bilateral STG (ASD n = 29, NT n = 30 toddlers, Dinstein et al., 2011; ASD n = 37, NT n = 20 children, Gabrielson et al., 2018; ASD n = 409, NT n = 455 children, adolescents, and adults, Li et al., 2019; ASD n = 40, NT n = 38 adolescents, Linke et al., 2018; ASD n = 13, NT n = 14 young adults, Murdaugh et al., 2012) and IFG connectivity (Dinstein et al., 2011; Li et al., 2019), consistent with task-based lexico-semantic studies (Shen et al., 2012; Verly et al., 2014). Only one study showed over-connectivity at rest between language and language homologue regions, the left IFG with the right STS (ASD n = 52, NT n = 50 adolescents, Gao et al., 2019), but this finding is consistent with Jasmin et al.’s (2019) finding of over-connectivity of the right STS, a region associated with theory of mind processing, during a conversation task (ASD n = 19, NT n = 20 young adults). Xu et al. (2020) examined MTG subregion connectivity, finding differential under-connectivity of MTG subregions with other cortical regions, such as with the pCC, right AG, and left orbital MFG (ASD n = 49, NT n = 33 children and adolescents, ASD n = 29, NT n = 29 young/adults). Hong et al. (2021 preprint) also found under-connectivity between the language network and other networks (e.g., the DMN and salience networks), but only through an interaction with IQ (i.e., individuals with ASD and relatively lower verbal than nonverbal IQ; ASD n = 155, NT n = 151 young adults). This evidence indicates resting-state under-connectivity between language and out-of-network regions, predominantly between language and language homologue regions. It also suggests that patterns may differ for brain regions implicated in social function, such as the right STS.

In autistic children and adolescents, there was also over-connectivity of the left IFG with the left precuneus and pCC (Gao et al., 2019; ASD n = 59, NT n = 61 adolescents, Levinson et al., 2021), the left IFG with MFG and left occipital gyrus (ASD n = 16 child intervention group, ASD n = 15 child control group, NT n = 22 children, Murdaugh et al., 2015), pSTS subregions with the whole brain (ASD n = 21, NT n = 26 adolescents, Shih et al., 2011), and the right IFG (operculum) with a reading network component (Murdaugh et al., 2015). Levinson et al. (2021) also reported over-connectivity among right hemisphere homologue regions (ASD n = 21, NT n = 26 adolescents). One additional resting-state study applied effective connectivity analysis and found that the pCC, a DMN hub, mediated over- versus under-connectivity between frontal language and visual regions in an ASD subgroup (Gao et al., 2019). While some resting-state findings are consistent with task-based findings (e.g., over-connectivity of the right pSTS with the left middle temporal cortex and the left IFG with the left cerebellum; Jasmin et al., 2019; Sharda et al., 2015), most resting-state findings of over-connectivity between language and out-of-network regions are inconsistent with task-based findings. These studies provide mixed evidence of over- and under-connectivity between language and out-of-network regions in ASD, with the exception of converging evidence of under-connectivity between language and language homologue regions. Patterns of over-connectivity were evident in autistic children and adolescents, but not autistic adults, suggesting that resting-state over-connectivity between language and out-of-network regions may not be observed in autistic adults. Yet, this evidence comes from only four studies. It is possible that inconsistent evidence reflects different patterns of mediation among brain regions based on Gao et al. (2019), yet no additional studies provide clarity on this issue.

Resting-state Connectivity Summary

Many studies examining resting-state functional connectivity in ASD draw upon ABIDE datasets (Di Martino et al., 2014) and many focus on the DMN (e.g., Wang et al., 2021). Resting-state studies included in the current review primarily focused on connectivity of the language network with out-of-network regions, rather than connectivity within the language network. This fMRI evidence converged with numerous examples of under-connectivity of left hemisphere language regions with right hemisphere language homologue regions, suggesting inter-hemispheric under-connectivity at rest in ASD. Several studies of children and adolescents indicated over-connectivity of language regions with DMN and visual regions, and one study indicated that this connectivity may be mediated by a key DMN hub, the pCC. Preliminary evidence of connectivity within the language network suggests over-connectivity related to the MTG, AG, and SMG, and under-connectivity of the left IFG and pSTG with language network regions when analyzed together with regions associated with visual, motor, and theory of mind function. This body of work does not consistently map onto theories of local over-connectivity and global under-connectivity in ASD.

Inter-hemispheric under-connectivity of language and language homologue regions aligns with the network view of global under-connectivity and with task-based findings. This under-connectivity further suggests diminished integrative processing, which is necessary for language processing, and may reflect relative difficulty with language. Over-connectivity between language and DMN regions does not align with the network view of global under-connectivity, and may reflect greater involvement of the DMN in resting-state connectivity, rather than language-specific connectivity (see also evidence of less DMN dis-engagement in response to task-demands in ASD; Baxter et al., 2019). Thus, over-connectivity between language and DMN regions likely reflects ASD status rather than individual differences in language ability.

Functional Connectivity Lateralization

Three resting-state studies tested functional connectivity lateralization. Cardinale et al. (2013) reported less left relative to right hemisphere connectivity for several networks, including an auditory network component (ASD n = 24, NT n = 26), in ASD relative to NT peers. Nielsen et al. (2014) found the language network demonstrated less left relative to right hemisphere lateralization of connectivity than a group of out-of-network regions in ASD and the reverse was found in NT peers. These out-of-network regions included the TPJ, insula, prefrontal cortex, cingulate cortex, intraparietal sulcus, and motor regions. Nielsen et al. (2014) also examined lateralization of ROI-ROI pairs and reported less connectivity within the left relative to right hemisphere for core language regions and DMN regions, including the pSTG with the pCC and TPJ, and of the IFG with the pCC, in ASD relative to NT peers (ASD n = 539, NT n = 573). In contrast, Gao et al. (2019) found no significant group differences in functional connectivity lateralization of regions in their language network (ASD n = 52; NT n = 50).

These studies tested different sets of regions. Cardinale et al. (2013) did not report on specific regions in their auditory network, but visual depictions suggest it was comprised of classic temporal and IFG language regions. Nielsen et al.’s (2014) language network reportedly included Broca and Wernicke’s areas (likely reflecting IFG and pSTG), but the 15 ROIs comprising Broca and Wernicke’s areas were not defined. Gao et al.’s (2019) language network extended beyond classic language regions to include DMN and motor regions. Examining classic language network regions tests lateralization more precisely because specific left hemisphere regions become functionally specialized for language across development (Olulade et al., 2020). In core language regions, the preliminary resting-state evidence on lateralization suggests a different developmental pattern of neural specialization for language in ASD relative to NT peers. This pattern may reflect greater connectivity among right hemisphere homologue regions across development in ASD. Further research examining lateralization of functional networks is needed.

Associations between Functional Connectivity and Behavior

ASD Features.

Two of 18 task-based studies and 11 of 15 resting-state studies analyzed associations between connectivity and measures of core ASD symptomatology, such as ADOS or ADI scores. Findings support the theory that local over-connectivity and global under-connectivity of the language network are associated with ASD status. Some of this work shows that greater connectivity within the language network is associated with relatively more ASD features, such as the left IFG and pSTG with the rest of the language network (ADOS communication scale; Lee et al., 2017) and within the language network when regions are considered together (ADI verbal scale; Ingalhalikar et al., 2021). There is also evidence of an association between over-connectivity of the SMG with all other language network regions (e.g., STG, IFG) and relatively more ASD features (ADOS social-communication scale; Kaminer, 2021). Other work shows that greater resting-state connectivity of language regions with regions outside the core language network in individuals with ASD was associated with relatively fewer ASD features, such as between inter-hemispheric IFG regions (ADOS social-communication scale; Dinstein et al., 2011), left lateralized connectivity of the pSTG with the pCC (ADOS social-communication scale; Nielsen et al., 2014), a left temporal component with left cuneus, and a right temporal component with right lateral parietal regions (VABS communication scores; Xiao et al., 2021 preprint). This association was also evident for language task-elicited connectivity of the left MTG with the cuneus for the ADI verbal scale (Fan et al., 2021), but not language task-elicited connectivity of the pSTG with the right extrastriate for the Social Responsiveness Scale (Jasmin et al., 2019). Yet, the evidence on associations between language task-elicited connectivity and ASD symptomology is limited.

Language Skills.

One of 18 task-based studies and four of 15 resting-state studies analyzed the strength of associations between connectivity and measures of language skills in ASD. This small body of work indicates that greater connectivity of the language regions within the language network and with out-of-network regions is associated with relatively higher language scores. These findings included greater connectivity of the right and left STG (verbal IQ; Linke et al., 2018), greater connectivity within the language network and between language region homologues (e.g., in low versus high verbal/cognitive skill ASD subgroups; Gabrielson et al., 2018), and inter-hemispheric IFG (expressive language; Dinstein et al., 2011). There are broad findings of greater connectivity of STG and AG regions with bilateral IFG/IPL regions (receptive language) and right STG and AG regions with IFG and IPL regions (expressive language) being associated with higher language scores (Xiao et al., 2021 preprint). Greater leftward lateralization of frontal and temporal region functional connectivity was associated with higher VIQ and fewer ASD features (Cardinale et al., 2013; ASD n = 24, NT n = 26), and Dryburgh et al. (2020) reported that greater out-of-network connectivity was associated with higher verbal IQ in ASD, such as left ITG and right IPL (ASD n = 202, NT n = 226; see also Mizuno et al., 2011).

Only one study reported conflicting results. Weng et al. (2010) demonstrated that greater connectivity of temporal regions within the DMN and the pCC was associated with lower receptive vocabulary scores (Weng et al., 2010; ASD n = 16, NT n = 15). These findings may reflect atypical connectivity related to the DMN rather than atypical connectivity related to language function (e.g., Li et al., 2019; Lombardo et al., 2019; Murdaugh et al., 2012; Redcay et al., 2013). Studies of reading, which is subserved by language and visual regions, lend additional support to the pattern of greater connectivity being associated with relatively higher language-related performance in ASD. This body of work reported associations between better reading performance and greater resting-state connectivity of left IFG (orbitalis) with left pSTG (Murdaugh et al., 2015), the bilateral AG and right SMG within the reading network (Maximo, 2016; Maximo et al., 2017), and reading intervention-related increased connectivity of the left IFG and pSTG with the reading network (Murdaugh et al., 2015; Murdaugh et al., 2016), as well as greater language task-elicited connectivity of the left MTG with the frontal network (Murdaugh et al., 2016) and the left IFG with the left thalamus (Bednarz et al., 2017).

Interestingly, emerging evidence suggests that relationships between connectivity and language behavior do not vary by fMRI language task paradigm (e.g., Bednarz et al., 2015, Narayanan et al., 2010; Sridhar et al., 2021 preprint, Verly et al., 2014) or task-free resting state context (e.g., Dinstein et al., 2011, Linke et al., 2018; Murdaugh et al., 2016, Hong et al., 2021 preprint). These studies suggest that greater connectivity of language regions with other language or non-language regions, regardless of whether the connectivity is local or global, is associated with relatively better language skills in ASD. Given that local over-connectivity and global under-connectivity currently is posited for individuals with ASD and not for individuals with core deficits in language, such as in developmental language disorder, language-related connectivity may compensate for communication difficulty associated with core ASD symptoms. Heightened connectivity of language regions, particularly language and out-of-network regions, may represent a means of overcoming social difficulty, given that language used in social interaction is likely to draw upon integrative processing across networks.

Other Considerations

Psychiatric Comorbidity and Connectivity in ASD

Little is known about the influence of psychiatric comorbidity on language-related functional connectivity in ASD. Only one study compared connectivity across clinical groups. Wan et al. (2019) examined resting-state connectivity of the whole-brain in ASD, NT, and in ASD subgroups with co-occurring Attention-Deficit/Hyperactivity Disorder (ADHD) or co-occurring anxiety. This study found greater connectivity of an MTG cluster in an ASD group with co-occurring anxiety relative to an ASD group without co-occurring anxiety, and less connectivity of an ASD group with co-occurring ADHD relative to an ASD group with anxiety. They showed no other group differences in other language-relevant regions, suggesting that ADHD status may only be reflected in MTG connectivity differences in ASD.

Developmental Considerations

Seven studies provide evidence of age-related differences in connectivity using cross-sectional designs, but there are no longitudinal studies to date. Cross-sectional resting-state studies indicated that connectivity of the amCC with precentral/MFG increased with age in ASD, but decreased with age in NT peers (ASD n = 131, NT n = 169 children, adolescents, and adults, Balsters et al., 2016), and connectivity of the left MTG with left IFG increased with age in ASD, but decreased with age in NT peers (ASD n = 49, NT n = 33 children and adolescents, ASD n = 29, NT n = 29 young/adults, Xu et al., 2020; see also Sariya & Adnand, 2017 for preliminary evidence of greater between-network connectivity of fronto-parietal networks in adults than children with ASD). During a word generation task, Baxter et al. (2019) reported similar connectivity of the left IFG within a language network component in younger and older adults with ASD, greater connectivity of the thalamus and speech production regions in young autistic and NT adults, and greater connectivity of the STG in older autistic and NT adults. Williams et al. (2013) showed under-connectivity within the language network in adults and children with ASD in an irony, but not literal, narrative task condition (ASD n = 15 children, n = 13 young adults, NT n = 14 children, n = 12 young adults). There are also similarities in the association between age and functional specialization of STS connectivity in ASD and NT peers (ASD n = 21, NT n = 26 adolescents, Shih et al., 2011). These studies suggest that language network connectivity increases with age in ASD, though there is not enough evidence to draw clear conclusions.

Across studies included in this review, task-based studies suggest that MTG connectivity differed in adolescence relative to childhood and adulthood (Baxter et al., 2019; Hegarty, 2015; Fan et al., 2021), though these patterns appeared to reflect social features of the task to a greater degree than age-related differences. There was evidence of over-connectivity between language and out-of-network regions in children and adolescents with ASD in four resting-state studies (Gao et al., 2019; Levinson et al., 2021; Murdaugh et al., 2015; Shih et al., 2011), but not in the six resting-state studies of adults with ASD included in this review. Considering the specific regions implicated in these studies, there is preliminary converging evidence that connectivity within the language network may increase across development and connectivity between language and out-of-network regions may decrease across development in ASD.

Limitations of the Existing Evidence

There were surprisingly few studies that examined subregions within broadly defined language regions, such as the IFG and MTG. Given the association of the pSTG with classically defined Wernicke’s area and the mSTG with the auditory cortex, there were several studies that examined these subregions, yet no studies compared subregional connectivity of the IFG and only one study compared subregional connectivity of the MTG. There were also limitations in this body of evidence on associations between connectivity and behavior. No studies examined associations between connectivity and processes that are relevant to language and ASD, such as executive function and attention, and two task-based studies included in this review did not report behavioral performance on in-scanner tasks (Mason et al., 2008; Murdaugh et al., 2012). Only one study examined psychiatric comorbidity and only one study examined co-occurring structural language impairment in ASD, even though co-occurring conditions are prevalent in ASD (e.g., anxiety, depression, ADHD, language impairment). Additionally, the evidence base on developmental patterns of connectivity in ASD is limited. Changes across neurotypical development indicated that functional neural networks are not mature until puberty, and young adulthood is characterized by changes in the connectivity of regions within networks, particularly in response to task demands (Stevens, 2016). It follows that network comparisons between individuals with ASD relative to NT peers may yield different findings from childhood to adulthood, and it is currently unknown how these findings will vary across development.

There are several groups of individuals with ASD who are not represented in the reviewed literature, indicating that this evidence may generalize only to a specific subgroup of individuals with ASD. First, individuals with ASD who are nonspeaking have been historically excluded from imaging research as most imaging research focuses on individuals who meet developmental expectations for language and other cognitive skills. We found that only one study examined such a group, codifying this issue of generalizability across the autism spectrum (Gabrielson et al., 2018). This study suggests that connectivity patterns differ depending on relative language and cognitive abilities and additional work is needed to understand the directionality of these differences across the autism spectrum. Second, Black, Indigenous, and People of Color (BIPOC) have been historically excluded from autism research (Girolamo et al., 2022; Jones & Mandell, 2020). We could not evaluate this issue in our review given that only two studies reported racial and ethnic characteristics of participant samples. It is impossible to know the racial and ethnic makeup of the other 94 studies. This finding represents a call to action for autism research. We must examine intersectional factors, including structures of oppression and inequality (e.g., racism, ableism) that give rise to differences in the lived experiences that shape neural development to ensure that neuroimaging findings are generalizable (Buchanan & Wiklund, 2021; Girolamo et al., 2022; Jones & Mandell, 2020).

Overall Summary and Conclusions

The aim of the current review was to synthesize literature on language-related functional connectivity in ASD relative to NT peers from fMRI studies. We were interested in learning whether the aggregate evidence supported theories of local over-connectivity and global under-connectivity and whether connectivity patterns were associated with behavioral measures of ASD symptomology and language. This review has provided much needed clarity on local over-connectivity and global under-connectivity in ASD by situating the review within a construct of theoretical and clinical interest and by presenting a topically organized review of the evidence base.

We operationalized local versus global connectivity in terms of the functionally specialized and well-defined language network, which allowed for a focused synthesis. This novel contribution builds on prior work that operationalized local versus global connectivity in terms of spatial distance (i.e., local as ≤ one cubic centimeter and global as ≥ one cubic centimeter; Vissers et al., 2012) and in terms of the DMN, a task negative network (i.e., local as within the DMN and global as between DMN and out-of-network regions; Wang et al., 2021). The mixed findings in more general reviews (e.g., Muller at al., 2011; Vissers et al., 2012) relative to the clearer patterns observed in Wang et al.’s (2021) more focused, network-based review motivated our use of a network-based approach in the current review. Indeed, the current review has revealed more consistent connectivity patterns than more general prior reviews. We also found more consistent connectivity patterns for task-elicited than resting-state connectivity, underscoring the benefits of examining functionally-specialized networks. Network-based approaches may, therefore, contribute to connectivity theory to a greater extent than spatial-definition approaches. This synthesis revealed several clear patterns of findings. First, the task-driven language network in ASD was characterized by local over-connectivity and global under-connectivity relative to NT peers. These patterns underscore language difficulty in ASD, particularly difficulty with complex language function that requires higher-order integration and distributed neural networks. The caveat to these overarching patterns was under-connectivity within the language network for IFG and MTG regions in response to social task demands. Second, the resting-state language network in the absence of overt task demands was characterized by under-connectivity of inter-hemispheric language regions and over-connectivity of language regions and DMN regions. Under-connectivity between language and language homologue regions also suggests diminished language-related integrative processes in ASD relative to NT peers, which may affect language learning. Over-connectivity of language and DMN regions suggests associations between regions linked with language function and regions linked with social function. This architecture appears to be associated with ASD status rather than individual differences in language ability. Third, language-related local over-connectivity and global under-connectivity was associated with relatively greater ASD symptomology, providing support for the theory that local over-connectivity and global under-connectivity characterize ASD. In contrast, greater connectivity of language with other language regions and non-language regions was associated with relatively better language skills. This evidence suggests that heightened language-related connectivity may support social function. Collectively, this review suggests the model depicted in Figure 2.

Figure 2.

Model of language task-based connectivity in ASD.

Note. Positive association (+) indicates that relatively greater connectivity is associated with relatively more ASD symptoms and higher language assessment scores; Negative association (−) indicates that relatively lower connectivity is associated with relatively more ASD symptoms and lower language assessment scores; The evidence does not support a clear model of language-related resting-state connectivity.

There are several aims for future work that will clarify unknowns in the current evidence base. More nuanced subregion differentiation is needed in connectivity analyses given recent advances in parcellations and in identifying unique subregion functions (e.g., Briggs et al., 2018; Kuhl, 2021), as well as emerging evidence of unique connectivity patterns associated with subregions (e.g., within the MTG; Xu et al., 2020). More nuanced analysis of social features within language tasks is also needed due to the close relationship between language and social function and the importance of social skills in ASD. Only one study systematically varied the level of social demands within their language task (Mason et al., 2008). This study may serve as a model for future work that carefully examines connectivity elicited by social versus language task demands. Additional associations between functional connectivity and behavior are needed to better characterize heterogeneity in ASD and to better align with the Research Domain Criteria framework which emphasizes cross-diagnostic and multi-dimensional analysis (RDoC; Insel et al., 2010). Specifically, there is little evidence on how connectivity is associated with processes related to language, like attention, and how connectivity differs depending on co-occurring conditions, like psychopathology and language impairment. Moreover, there is no longitudinal evidence on connectivity in ASD and only minimal information on connectivity patterns in nonspeaking individuals with ASD and BIPOC individuals with ASD.

There are three analytical issues that warrant attention in future work. First, only one study examined effective connectivity. This analytical approach tests a priori hypotheses of interest and characterizes the directionality of correlated brain activity. It has the potential to further resolve inconsistencies in the evidence base. Second, there is only preliminary evidence on left hemisphere lateralization, the key neural theory of language development, and this evidence suggests diminished specialization of core left hemisphere language regions for language in ASD. Future studies should examine task-elicited lateralization of connectivity given that the body of resting-state connectivity findings do not map onto well-defined language networks and several connections associated with diminished lateralization in ASD are between language and DMN regions (a task-negative network). Finally, relatively little is known about connectivity patterns associated with the slowly-fluctuating BOLD signal relative to transient task-elicited changes in the BOLD signal. There is preliminary evidence of unique connectivity patterns when regressing out all task-relevant variance (Jones et al., 2010; see also Muller et al., 2011) and future work would benefit from carefully considering this analytical issue.

Conclusions

The current evidence suggests that language-task elicited local over-connectivity and global under-connectivity during fMRI are features of ASD. The challenge to viewing these patterns as a feature of ASD is their apparent context-dependency, including inconsistencies in the presence of social task demands and in the absence of overt task demands. However, associations between ASD symptomology and language-related connectivity support the view that local over-connectivity and global under-connectivity are features of ASD, regardless of task versus rest context. Through the lens of language, we have identified a relatively consistent profile which may be used to support future studies that test theories of connectivity in ASD and future studies that seek to identify robust neurobiological markers of ASD.

List of Terms and Abbreviations

Functional connectivity

brain activity that shows a statistical relationship between two or more brain regions over time

Over-connectivity

relatively greater strength of connectivity in ASD versus NT group

Under-connectivity

relatively lesser strength of connectivity in ASD versus NT group

ASD

Autism Spectrum Disorder

NT

Neurotypical

ROI

Region of Interest

ADOS

Autism Diagnostic Observation Schedule

Soc

socialization subscale

Comm

communication subscale

ADI

Autism Diagnostic Interview

SRS

Social Responsiveness Scale

AQ

Autism Quotient

CELF

Clinical Evaluation of Language Fundamentals, a standardized structural language assessment

Cortical

cerebral regions in the frontal, temporal, occipital, and parietal lobes

Subcortical

deep gray matter regions including the hippocampus, amygdala, thalamus, and putamen

Striatum

caudate, putamen, and ventral nuclei

p

posterior

a

anterior

v

ventral

m

medial

IFG

inferior frontal gyrus, including pars opercularis, pars orbitalis, and pars triangularis; represents a set of regions classically referred to as Broca’s area

STG

superior temporal gyrus; the superior posterior subregion of the STG is often called Wernicke’s area; the medial subregion is often called primary auditory cortex or Heschl’s gyrus

IPL

inferior parietal lobule; represents a region relevant to Wernicke’s area

SPL

superior parietal lobule

TPJ

temporoparietal junction; represents a region relevant to Wernicke’s area

AG

angular gyrus; represents a region relevant to Wernicke’s area

MTG

middle temporal gyrus

DMN

default mode network, a task-negative network with the posterior cingulate cortex as a central node

CC

cingulate cortex; part of the limbic, or subcortical, system and the posterior portion of the cingulate is a central node of the DMN

Language network

IFG, TPJ, AG, SMG, and temporal regions

Component analysis

separates signals by their sources and then extracts underlying sources to identify underlying networks; includes principle and independent component analysis

Effective connectivity

tests a priori directional relationships between variables, adding paths until a best-fit model is found; includes structural equation modeling in a data-driven approach using Group Iterative Multiple Models Estimation and multivariate autoregressive modeling to estimate causal relationships

Graph theory

models of networks that involve a set of nodes which are interconnected by edges, such as brain regions representing nodes and functional connectivity at a given statistical threshold representing edges; specific measures may include degree centrality (a node’s number of edges), betweenness centrality (how often a node is the shortest path between two other nodes), eigenvector centrality (nodal influence within a graph/network), and small worldness (quantifies number of short-range relative to long-range edges)