Sex-Related Patterns of Intrinsic Functional Connectivity in Children and Adolescents with Autism Spectrum Disorders

Abstract

Background:

Although a growing literature highlights sex differences in ASD clinical presentation, less is known about female variants at the neural level. We investigated sex-related patterns of functional connectivity (FC) within and between functional networks in children and adolescents with ASDs, compared to typically developing (TD) peers.

Method:

Resting-state functional MRI data for 141 children and adolescents (7–17 years) selected from an in-house sample and four sites contributing to the Autism Brain Imaging Database Exchange (ABIDE I and II) were submitted to group independent component analysis (ICA) to generate resting-state functional networks (RFNs). FC was estimated by generating RFN-RFN correlation matrices, which were directly compared between males and females, and ASD and TD groups.

Results:

Results revealed greater connectivity within the default mode network (DMN) in TD girls as compared to TD boys, while no such sex effect was observed in the ASD group. Correlational analyses with clinical indices revealed a negative relationship between sensorimotor connectivity and history of early autism symptoms in girls, but not in boys with ASD.

Conclusions:

A lack of neurotypical sex differentiation in DMN functional connectivity observed in boys and girls with ASD suggests that sex-related differences in network integration may be altered in ASD.

Lay Summary:

We investigated whether children and adolescents with autism spectrum disorders show sex-specific patterns of brain function (using fMRI) that are well documented in typically developing males and females. We found, unexpectedly, that boys and girls with autism do not differ in their brain functional connectivity, whereas typically developing boys and girls showed differences in a brain network involved in thinking about self and others (the default mode network). Results suggest that autism may be characterized by a lack of brain sex differentiation.

Introduction

Autism Spectrum Disorders (ASDs) are a group of neurodevelopmental disorders characterized by social communication deficits and restrictive and repetitive behaviors and interests, with symptom onset within the first years of life. While ASDs are symptomatically heterogeneous, the sex imbalance in prevalence (approximately 3–4:1 male to female ratio) is a consistently striking characteristic of the disorder, dating back to its initial description by Leo Kanner in 1943 (Baio et al., 2018; Constantino, 2017). Indeed, the male bias in ASD is intrinsic to the features of the disorder itself, which may be associated with female protective effects (biological mechanisms that reduce its incidence or severity in females), a phenomenon that is observed, for instance, in Rett Syndrome, an X chromosome-linked (XL) disorder with lethal consequences for males (Cauvet et al., 2019; Chahil & Bollu, 2018; Gockley et al., 2015).

Given the substantial sex dimorphism in ASD rates, most research studies have been conducted with exclusively or predominantly male cohorts. As a result, less is known about female variants of autism, although mounting evidence suggests that girls with ASDs may be under-identified (Loomes, Hull, & Mandy, 2017) and that ASDs may affect girls differently than boys (see Werling & Geschwind, 2013). Indeed, the sex imbalance in prevalence also biases our understanding of the diagnostic features of ASDs, which in turn impacts the study of sex/gender-dependent and independent characteristics of the disorder, and the investigation of etiology and distinct liability thresholds for genetic load and for likelihood of developing the disorder in boys vs. girls (M. C. Lai, Lombardo, Auyeung, Chakrabarti, & Baron-Cohen, 2015). For instance, sex-specific phenotypes may include fewer restricted and repetitive behaviors (Szatmari et al., 2012), as well as fewer externalizing behaviors (Hiller, Young, & Weber, 2014), in girls. Further, girls with ASDs within the average-to-high IQ range are underrepresented (Hiller et al., 2014), which may contribute to their lower participation rates in neuroimaging studies, as adherence to instructions to remain still presents a barrier in those with lower IQ, thus biasing the current understanding of the neural bases of ASDs. Finally, females may be more protected from heritable and de novo ASD risk variants, as XL factors and sex hormones may modulate the effects of risk-conferring genetic variants on the presentation of ASD symptoms (Werling & Geschwind, 2013). Altogether, these reports highlight the need to improve understanding of sex-related neurobiological differences in ASDs, which will facilitate the development of novel therapeutic interventions optimized for each sex (Constantino, 2017). However, despite the growing evidence of sex-related phenotypic and genetic variability in ASDs, only a few studies to date have focused on sex differences in brain organization in ASDs (Alaerts, Swinnen, & Wenderoth, 2016; Kovalev, Kruggel, & von Cramon, 2003; Kurth, Thompson, & Luders, 2018).

In neurotypical individuals, females generally have more symmetrical brains than males, with respect to brain structure, function, network organization, and connectivity (Gong, Gaolang, Yong, He, Evans, 2011), and such patterns are present well before adulthood. A large study of neurotypical children, adolescents, and young adults (ages 9–22 years, N = 674) also demonstrated that males displayed greater between-network connectivity, whereas females displayed greater within-module connectivity (Satterthwaite et al., 2015). By contrast, others have failed to detect sex differences in functional connectivity (FC) among typically developing (TD) children and adolescents (Solé-Padullés et al., 2016), though sex differences in FC in adolescence may be age- and IQ-dependent (Schmithorst & Holland, 2006).

While these findings demonstrate that sex plays a role in brain developmental trajectories across the neurotypical lifespan, very few studies have examined differences in brain organization between males and females with ASDs. The available evidence suggests that boys and girls with ASDs have distinct patterns of early brain overgrowth, with enlargement of brain volume observed more prominently, or even exclusively in boys (Courchesne, Carper, & Akshoomoff, 2003; Nordahl et al., 2011). Atypical sex-related brain organization appears to persist into adulthood, as typical patterns of sex differentiation in brain white matter connectivity and volume (i.e., higher in neurotypical men than women) are absent (Beacher et al., 2012), or attenuated (Cauvet et al., 2019; M.-C. Lai et al., 2017), in adults with ASDs. Similarly, sex-related patterns of neurotypical cortical thickness may be absent in adults with ASDs (Ecker, 2019; Ecker et al., 2017).

Among theoretical explanations of sex effects in ASDs, including the higher incidence of ASDs in males, Baron-Cohen’s “extreme male brain theory of autism” (EMBT), which links ASD etiology to the effects of fetal testosterone (Baron-Cohen et al., 2015), is perhaps most well-known. EMBT proposes that autism represents enhanced ‘systematizing’ neurocognition, which is predominant among neurotypical males versus females, in contrast to ‘empathizing’ patterns prominently reflected in neurotypical females (although the author allows that the degree of this dimorphism is somewhat distributed in the general population; Baron-Cohen, 2002). However, the EMBT remains controversial (Alaerts et al., 2016; Bejerot et al., 2012).

An alternative theory, known as the sex incoherence theory, proposes that ASDs are associated with neural androgyny or sex incoherence, rather than an amplification of stereotypically male neurocognitive phenotypes (Bejerot et al., 2012). Indeed, a neuroimaging study (Alaerts et al., 2016), albeit in a sample spanning both childhood and adulthood (ages 7–30 years), reported similar FC patterns between males with ASDs and TD females, and between females with ASDs and TD males. In particular, males with ASDs exhibited predominantly weaker connectivity relative to TD males between multiple regions of interest (ROIs), reflecting neural feminization, while females with ASDs showed stronger connectivity relative to TD females between the majority of ROI pairs with the whole-brain coverage, reflecting a shift toward male phenotypes. On the other hand, hypoconnectivity within the default mode network (DMN) was recently reported in both males and females with ASDs (ages 12–18 years), as compared to TD males, who in turn had weaker DMN connectivity than TD females (Ypma et al., 2016), consistent with previous evidence of DMN hypoconnectivity as a male-dominant trait in neurotypical population (Biswal et al., 2010; Tomasi and Volkow, 2012; Jung et al., 2015; Ritchie et al., 2018), and supportive of the sex incoherence theory. Yet another study recently reported mixed results supportive of both EMBT and sex incoherence theory, in a network-specific manner (Floris et al., 2018): in their cohort, spanning a broad age range (6.5–58 years), an exaggerated neurotypical male brain pattern (or ‘shift-toward-maleness’) was observed in the DMN, whereas a predominantly female pattern (or ‘shift-toward-femaleness’) was detected in the somatomotor network.

Taken together, sex-related patterns of functional brain connectivity in ASDs are equivocal, with evidence supporting both sex incoherence or extreme male-like neural patterns. Disparate conclusions likely reflect methodological differences between studies, as well as methodological shortcomings, such as inclusion of cohorts with broad age ranges spanning in some cases 50 years, which effectively masks the unique neurodevelopmental trajectories observed in ASDs. The lack of consensus also likely stems from the substantial etiological and clinical heterogeneity, manifesting as diversity of deficits and associated features and markers, characterizing ASDs (Jeste & Geschwind, 2014). Motivated by these challenges and evidentiary ambiguities, the present study set out to examine sex-related patterns of whole brain FC patterns and their relation to ASD symptoms, using a data-driven analytic approach, Independent Component Analysis, applied to the aggregated MRI data from a large data-sharing initiative, while restricting the sample age to childhood and adolescence.

Methods

Participants

Resting-state functional MRI datasets for 141 children and adolescents (ages 7–17 years) were selected from the publicly available Autism Brain Imaging Database Exchange (ABIDE I and II; DiMartino et al., 2014 and 2017; n = 111) and an in-house sample (n = 31), with 69 datasets from children with ASDs (35 females and 34 males) and 72 datasets from typically developing (TD) children (36 females and 36 males). Eligibility for the inclusion of datasets was determined based on age (<18 years), eye-status during resting state scans (eyes open), and fMRI data quality (low head motion: datasets in which > 80% of volumes had <0.5mm frame-wise displacement). ABIDE datasets were included only from sites that contributed at least 6 female datasets: New York University Langone Medical Center (NYU_1 n = 32, NYU_2 n = 5), Georgetown University (GU n = 17), Kennedy Krieger Institute (KKI n = 32), Oregon Health and Science University (OHSU n = 24; see Table 1 for details and participant characteristics and supplementary Tables S1 and S1a for site-specific characteristic information).

Table 1.

Participant Characteristics

Characteristic	ASD-B, n = 34	ASD-G, n = 35	Boys vs. Girls, p
Handedness (R/L)	30/4	30/5	0.76
Age (years), mean ± SD (range)	11.5 ± 2.1 (7.2–17.7)	11.6 ± 2.9 (7.0–17.9)	0.72
Verbal IQ*, mean ± SD (range)	106.2 ± 17.8 (79–143)	103.0 ± 14.5 (79–139)	0.44
Nonverbal IQ*, mean ± SD (range)	105.0 ± 17.1 (81–142)	100.1 ± 19.9 (53–140)	0.34
Full-scale IQ, mean ± SD (range)	104.3 ± 17.9 (81–142)	101.2 ± 17.8 (66–132)	0.48
ADOS (total), mean ± SD (range)	11.7 ± 4.3 (3–23)	10.6 ± 2.7 (6–17)	0.45
ADI-R (total), mean ± SD (range)	40.7 ± 10.6 (20–61)	35.6 ± 12.8 (13–64)	0.09
RMSD, FC head motion, mean ± SD (range)	0.074 ± 0.04 (0.03–0.14)	0.086 ± 0.04 (0.03–0.16)	0.07
Proportion on Psychiatric Medication	10/34	12/35	0.67
Characteristic	TD-B, n = 36	TD-G, n = 36	Boys vs. Girls, p
Handedness (R/L)	31/5	32/3	0.48
Age (years), mean ± SD (range)	11.8 ± 2.3 (7.0–17.9)	11.4 ± 2.0 (8.7–17.3)	0.45
Verbal IQ*, mean ± SD (range)	112.4 ± 13.4 (86–141)	112.2 ± 12.8 (85–146)	0.37
Nonverbal IQ*, mean ± SD (range)	109.6 ± 16.5 (83–137)	108.4 ± 11.8 (79–126)	0.99
Full-scale IQ, mean ± SD (range)	112.4 ± 9.3 (91–137)	113.1 ± 11.6 (80–132)	0.58
RMSD, FC head motion, mean ± SD (range)	0.06 ± 0.02 (0.02–0.09)	0.06 ± 0.02 (0.02–0.11)	0.54

Notes. ASD-B: ASD Boys, ASD-G: ASD Girls, TD-B: TD Boys, TD-G: TD Girls, ADOS: Autism Diagnostic Observation Schedule, ADI-R: Autism Diagnostic Interview-Revised, RMSD: Root mean squared displacement.

P-values reflect results from independent samples t-tests or chi-square tests of independence; ASD vs. TD comparisons are presented in Table S2. Details on the participants’ psychotropic medication status and incidence of co-occurring psychiatric diagnoses in the ASD group are presented in Table S3.

^*Nonverbal and Verbal IQ data do not include data from OHSU, as this information was not available for the ABIDE OHSU collection.

ASD diagnoses were confirmed using the Autism Diagnostic Observation Schedule (Lord, Rutter, Dilavore, Risi, Gotham, Bishop, 2012; Lord et al., 2000), Autism Diagnostic Interview – Revised (Lord, Rutter, & Le Couteur, 1994), and expert clinical judgment based on DSM-5 criteria (American Psychiatric Association, 2013). Inclusion in the TD group required absence of any developmental disorder (including ASD) or intellectual disability. Other specific exclusion criteria used for selecting TD participants (e.g., absence of neurological or psychiatric history, or absence of family history of ASDs) varied across ABIDE sites. Within each diagnostic group, boys and girls were matched, at the group level, on age, handedness, overall cognitive functioning (using full scale IQ), in-scanner head motion, and study site. Given the challenges involved in matching across all of these variables, head motion in girls with ASDs (ASD-G) was marginally higher than in boys with ASDs (ASD-B; p = 0.07). As such, motion was included as a covariate in comparisons between these two groups. Boys and girls with ASDs were also matched on autism symptoms (assessed with the ADOS Total Score), incidence of co-occurring psychiatric conditions, and psychotropic medication usage, due to the documented effects of these medications on brain function and connectivity (cf., Linke et al., 2017) (see Table 1 and Table S3; for comparisons between TD and ASD groups combined across gender, see Table S2). ASD participants with co-occurring psychiatric diagnoses or those reporting psychotropic medications usage were not excluded because of the high prevalence of comorbid psychiatric conditions and psychiatric medication usage in ASDs (Spencer et al., 2013). Overall cognitive ability was assessed by either the Wechsler Abbreviated Scale of Intelligence, 2nd Edition (WASI-II; Wechsler, 2011) the Wechsler Intelligence Scales for Children, 4^th and 5^th editions (WISC-4: Wechsler, 2006; WISC-5: Weiss & Saklofske, 2016), or the Differential Ability Scales, 2nd Edition (DAS-II—Elliot, 2007).

MRI Data Acquisition and Imaging Parameters

All MRI data were acquired using 3T MRI scanners, although imaging sequence parameters for both anatomical T1 and functional T2* scans varied across sites (see details on scanning parameters in Di Martino et al. (2014) and Di Martino et al. (2017). All fMRI data were collected during resting state (scan duration varied between 5:14 and 6:40 minutes), with participants instructed to lie still with their eyes open.

MRI Data Preprocessing

Data were preprocessed using AFNI (Cox, 1996) and FSL 5.0 (Smith et al., 2004). FSL was used to normalize anatomical images to MNI152 template space and segment structural images into white matter, grey matter, and cerebrospinal fluid. Functional data were slice-time and motion-corrected, using the middle time point as the reference volume for re-alignment; field map corrected (when field maps were available) to minimize magnetic field inhomogeneity; aligned to anatomical data using FSL FLIRT with six degrees of freedom; resampled to 3.0 mm isotropic voxels using sinc interpolation; and standardized to the MNI152 template, using FSL nonlinear registration tool. Spatial smoothing with an isotropic Gaussian filter to an effective full-width at half-maximum of 6mm, and temporal filtering of the signal with a high-pass Butterworth filter (0.008Hz < f) were applied. Six motion parameters, mean white matter and CSF signals (extracted from participant-level masks created with FSL and eroded by 1 voxel), and their first temporal derivatives (all equally high-pass filtered) were modeled as nuisance regressors and removed from the data through linear regression, in preparation for independent component analysis (ICA).

Functional Connectivity Analyses

Processed functional data from all ASD and TD participants, combined, were submitted to group independent component analysis (ICA) using FSL’s Multivariate Exploratory Linear Optimized Decomposition into Independent Components (MELODIC) to generate maximally independent intrinsic functional networks (de Lacy, Doherty, King, Rachakonda, & Calhoun, 2017). The Group ICA yielded 71 independent components. Components were marked as artifacts if their spatial maps were confined to the boundaries of the brain or cerebral ventricles, or were distributed irregularly (Griffanti et al., 2017). The remaining components were labeled based on two criteria: 1) weighted spatial correlation (r>0.25) with the ten primary networks derived by Smith and colleagues (Smith et al., 2009), and 2) concordance with ICA-derived functional networks in previous literature. Ultimately, of the 71 independent components (ICs) derived from ICA, 17 were retained for further analyses (see Figure S1 and Table S4).

FSL MELODIC dual-regression was used to generate subject-specific resting functional networks (RFNs) and associated time courses. Individual RFN spatial maps were thresholded at 90% of z-score values to generate individualized RFN masks for all participants. After extracting the mean time series for each RFN mask, FC was estimated by generating intra-individual Pearson correlation coefficients across all RFN-RFN pairs (136 pairs: (17×16)/2). The resulting correlation coefficients were transformed into Fisher’s z scores, and individual connectivity matrices were averaged within each group (ASD Boys [ASD-B], ASD Girls [ASD-G], TD Boys [TD-B], and TD Girls [TD-G]). Connectivity for all 136 RFN-RFN pairs were compared using linear mixed effects models with Group (ASD, TD) and sex (Girls, Boys) as predictor variables, head motion, IQ, and age as nuisance regressors, and site as a random effect nuisance variable. Type-I error inflation was controlled using False Discovery Rate correction for multiple comparisons.

In an effort to reduce the overall number of comparisons, RFNs were categorized into three overarching domains: sensorimotor or unimodal (e.g., visual, auditory networks), default mode network, and higher-order supramodal domains (e.g., language, executive networks—see Figure S3 for domain maps across groups). FDR-correction was used for all network and domain-level analyses (sensorimotor/unimodal, DMN, and higher-order/supramodal) to control type-I error rate. T-scores for comparisons between boys and girls in each domain were generated using MATLAB’s ttest2 function.

Results

Sex, Diagnosis, and Sex by Diagnosis Interaction Effects on Network FC

Connectivity matrices of mean RFN-RFN connectivity (z values) within each diagnostic group and sex are presented in Figure 1a,b. Linear mixed effects models with diagnosis, sex, and sex by diagnosis interaction terms as predictor variables, age, motion, and IQ as nuisance regressors, and site modeled as a random effect revealed the main effects of sex, diagnosis, and sex by diagnosis interaction effects (all at uncorrected p < 0.05). For descriptions of the effect sizes of simple comparisons, see Supplementary Materials, Appendix S1 and Figure S2. Overall, with regard to the main effects of sex, boys displayed higher connectivity than girls between the following networks: lateral visual (VIS1), default mode (PCC; DMN4), and occipital visual (VIS4) networks; and between sensorimotor (SM1 and SM2) and default mode (DMN3 and DMN4; see Figure 2 and Table 2a).

Figure 1.

Connectivity Matrices (z-values) for RFN-RFN connectivity for a) ASD-B (lower left triangle) and ASD-G (upper right triangle), and b) TD-B (lower left triangle) and TD-G (upper right triangle).

Figure 2.

Upper triangles depict beta values, and red cells in lower triangles depict p-values < 0.05 (uncorrected) showing RFN-RFN connectivity for a) main effects of diagnosis b) main effects of sex, and c) sex by diagnosis interactions.

Table 2a.

LME Results for RFN-RFN Connectivity

	Effect	Beta	P-Value	RFN 1	RFN 2
Diagnosis Effect	ASD > TD	−0.09	0.03	VIS2	DAN
		−0.13	0.03	VIS3	DAN
		−0.12	0.04	AUD	DMN4
	TD > ASD	0.02	0.02	DMN1	DMN2
		0.20	<0.001	DMN1	DMN4
		0.04	0.01	DMN2	rFP
		0.11	0.02	DMN4	rFP
		0.17	0.01	DMN4	Lang
Sex Effect	Boys > Girls	0.08	0.02	VIS1	DMN4
		0.06	0.02	VIS2	VIS4
		0.13	0.02	SM1	DMN4
		0.05	0.02	SM2	DMN3
Diagnosis by Sex Interaction	TD-G > TD-B, ASD-G = ASD-B	0.27	0.04	SM2	lFP
		0.47	0.02	SM1	DAN
		0.49	0.03	DMN1	DMN3
		0.33	0.03	DMN2	Lang

Note: Results are shown for uncorrected p< 0.05 for completeness. Results did not survive FDR correction for multiple comparisons. Analyses included motion, age, and IQ as nuisance regressors, and site as a random effect.

Main effects of diagnosis were observed as greater connectivity (ASD>TD) between medial visual (VIS2 and VIS3) and dorsal attention (DAN) networks, and between auditory (AUD) and default mode (DMN4) networks. Weaker connectivity (ASD<TD) was observed between medial (DMN1) and PCC (DMN2 and DMN4) nodes of the default mode network, and between default mode PCC node (DMN2) and the right frontoparietal (rFP) network, PCC (DMN4) and right frontoparietal (rFP), and between PCC (DMN4) and language (Lang) networks (see Figure 2 and Table 2a).

Finally, sex by diagnosis interaction effects were identified between sensorimotor and higher-order supramodal networks, with TD girls displaying greater connectivity than TD boys between the sensorimotor (SM1) and dorsal attention (DAN) networks, and sensorimotor (SM2) and left executive (lFP) networks, while the opposite pattern (girls showing weaker connectivity than boys) observed in participants with ASDs within the same connections (see Figure 2 and Table 2a). Other interaction effects were observed within the default mode network (DMN1 and DMN2), with TD girls displaying greater connectivity than TD boys within the DMN; and between default mode and language networks (DMN3 and LANG), with TD girls displaying greater connectivity than TD boys, with no differences observed between ASD boys and girls (see Figure 2 and Table 2a).

Domain-level Connectivity

In order to reduce the number of comparisons and to examine overarching connectivity patterns, the dimensionality of the connectivity matrix was reduced by grouping the individual RFNs into functional domains, achieved by averaging all RFN-RFN z-values within the unimodal sensorimotor, default mode, and higher-order supramodal domains. The averaged within- and between-domain z-values were compared within sex and diagnostic group using linear mixed effects models, with site modeled as a random effect, and motion, age, and IQ as nuisance regressors (see Figure 3a,b). Analyses revealed a significant main effect of sex on DMN connectivity, such that girls showed higher DMN connectivity than boys (see Table 2b). The sex by diagnosis interaction was not significant, but there was a trend such that the sex effect was present in the TD group, but absent in the ASD group (IQ and site were included in the model as nuisance regressors, with site modeled as a random effect). Further, there was a significant main effect of diagnosis on sensorimotor connectivity, controlling for motion and site, such that TD participants showed higher within-sensorimotor-domain connectivity than ASD participants.

Figure 3.

Top panel: Domain-level connectivity matrices (averaged z-values within- and between-domains) for a) ASD-B (lower left triangle) and ASD-G (upper right triangle), and b) TD-B (lower left triangle) and TD-G (upper right triangle). Bottom panel: DMN within-network connectivity (averaged z-values for all DMN RFN components) plotted for c) ASD-B vs. ASD-G, and d) TD-B vs. TD-G. TD-G showed significantly higher within-DMN connectivity than TD-B (t_{(1, 70)} = −2.87, p = 0.005, FDR corrected p = 0.01). ASD-B and ASD-G did not differ in their within-DMN connectivity.

Table 2b.

LME Results for Domain-Level Connectivity

Functional Domain	Effect	Beta	P-Value
Sensorimotor	Diagnosis: ASD > TD	−0.13	0.04
DMN	Sex: Girls > Boys	−0.33	0.003
	Sex X Diagnosis	0.26	0.10

^*Note: showing results for uncorrected p< 0.05 for completeness. Only main effect of gender on DMN connectivity survived FDR correction for multiple comparisons.

Although no significant sex by diagnosis interaction effects were detected, a secondary analysis comparing connectivity within sex and diagnostic group was conducted within these three overarching functional domains. This analysis revealed significantly greater within-DMN connectivity in TD girls compared to TD boys, t₍₇₀₎ = −2.87, p = 0.01 (FDR-corrected p-value), with effect size of Hedges’ G = −0.52. By contrast, no significant differences in FC within DMN, or within or between any other functional domains were observed between girls and boys with ASDs (or between ASD and TD boys, or ASD and TD girls) (see Figure 3c,d).

Correlations with Autism Symptoms

Relationships between FC and autism symptoms measured with ADOS and ADI-R were examined in males and females with ASDs, separately, as an exploration into sex differences in brain-behavior relationships. A total of 12 correlation tests were performed (correlations between connectivity in unimodal sensorimotor, DMN, and higher-order supramodal domains with ADI-R Total, and ADOS-Total scores). A negative correlation was observed between sensorimotor domain connectivity (average of z-values of the 7 sensorimotor RFNs) and early history of autism symptoms measured with the ADI-R Total in girls (r = −0.53, uncorrected p = 0.003; Bonferroni corrected alpha = 0.004), but not boys with ASDs (see Figure 4), when controlling for the effect of motion (see also Figure S4, for effects of motion on connectivity values). Associations between connectivity and autism symptoms were not statistically significant for any other domains in boys or girls, after correcting for multiple comparisons.

Figure 4.

Correlation between sensorimotor domain connectivity (averaged z-values for all 7 sensorimotor RFNs) and early history of symptoms as measured with the ADI-R (Total Score), for ASD-B (blue) and ASD-G (red). The zero-order correlation is plotted with (uncorrected p-value).

Discussion

This study investigated sex-related patterns of network connectivity in children and adolescents with ASDs, using a data-driven method to characterize intrinsic whole brain functional connectivity. The main and most striking finding was an absence of sex effects on FC observed in the ASD group, compared to TD children and adolescents who displayed sex-related connectivity differences within the default mode network, with TD girls showing greater DMN connectivity than TD boys.

First, our findings on neurotypical sex differentiation in the DMN observed in the TD group align with those of previous studies of typical adults, showing higher DMN connectivity in females compared to males (Biswal et al., 2010; Jung et al., 2015; Ritchie et al., 2018; Tomasi & Volkow, 2012), suggesting that DMN network connectivity distinguishes neurotypical male vs. female brains (Zhang et al., 2016). Notably, Lombardo et al. (2018) showed that fetal testosterone levels predict DMN connectivity in adolescent males, but not females, and linked their findings to specific genes that are expressed in DMN-relevant circuitry, arguing that DMN connectivity is negatively associated with prenatal steroid hormones, and may account, in part, for the observed DMN sex differences (Lombardo et al., 2018).

While it is not completely understood how sex differences in DMN connectivity translate into differences in cognition or behavior, increased DMN connectivity in females aligns with evidence of females’ greater prosocial orientation and stronger social-evaluative skills in comparison to males (Gur et al., 2012; Rose & Rudolph, 2006). This is especially noteworthy because, in addition to being spontaneously engaged during rest or active by default, in the absence of overt or covert task, the DMN is also associated with social inference and is said to play a central role in many aspects of social cognition (Kennedy & Adolphs, 2012), including understanding others and mentalizing (Andrews-Hanna, Saxe, & Yarkoni, 2014), as well as thinking about the self, in the past, present, and the future (Spreng, Mar, & Kim, 2009). Indeed, with task-activation data acquired in adults (ages 18–45) with and without ASDs, Lai et al. (2019) demonstrated hypoactivity in the ventral medial prefrontal cortex (forming part of the DMN) in males with ASDs as compared to TD males, during self-referential thinking, while no such sex difference was observed in females with ASDs when compared to TD counterparts. Further, heightened activity in the ventral medial prefrontal cortex was correlated with enhanced “camouflaging” in females – but not males – with ASDs, suggesting that there may be sex-mediated links between the DMN / social brain function in autism and the ability to blend in socially, despite autism-associated social deficits.

Given its role in self- and other-cognition, it is unsurprising that alterations in DMN connectivity have been implicated in several studies on the neurobiology of ASDs (Mars et al., 2012; Spreng & Grady, 2010; Uddin, 2011). Alterations in DMN functional connectivity are one of the few robust findings in ASDs (Assaf et al., 2010; Weng et al., 2010; Lynch et al., 2013; von dem Hagen et al., 2013; Ypma et al., 2016). Specifically, DMN hypoconnectivity has been widely reported in people with ASDs, in both males and females alike (Floris et al., 2018; Ypma et al., 2016). While this pattern is potentially supportive of neural masculinization, it also may reflect neural gender incoherence. On the other hand, there is evidence that diagnosis effects on DMN connectivity do not equally apply to boys and girls with ASDs. Alaerts and colleagues (2016) observed DMN hyperconnectivity in girls with ASDs (particularly within the PCC and superior temporal sulcus), while males with ASDs showed reduced DMN connectivity compared to TD males. In light of their findings, the authors argued that girls with ASDs showed a neural shift toward maleness characterized by DMN hyperconnectivity (as TD boys in their sample showed greater connectivity compared to TD girls), whereas a shift toward femaleness was observed in ASD boys. Thus, the authors proposed that ASDs may be characterized by neural sex incoherence or androgynization, rather than neural masculinization (Alaerts et al., 2016).

In contrast to seed- and ROI-based connectivity analyses conducted by Alaerts and colleagues (2016), the present study used a data-driven method to derive resting state networks and to assess their functional connectivity. This critical difference in methodology may partially explain the divergence of our findings from those put forth by Alaerts and colleagues (despite the likely overlap in samples as both studies utilized data from ABIDE). Independent Component Analysis (ICA) utilized in the present study yields individualized, data-derived spatiotemporal maps of intrinsic functional networks, rather than one-size-fits all atlas-derived ROIs, which create arbitrary spatial divisions between brain regions. As such, ICA provides increased sensitivity to intra- and inter-subject differences not captured by atlas-based ROI approaches (de Lacy et al., 2017).

Findings from the present study cannot be fully explained by EMBT (shift toward maleness in both genders), as our results did not reveal an exaggeration of the typical male neural profile in either boys or girls with ASDs. However, the sex incoherence theory (shift toward androgynous neural profiles) also does not fully account for the findings, as neural profiles for ASD boys did not resemble those of TD girls. Our results did not reveal connectivity patterns characterized by exaggerations of either male or female neurobiological profiles. Rather, a neurotypical sex differentiation in DMN functional connectivity observed in our TD cohort was absent among ASD children and adolescents. The absence of sex effects in the ASD group suggests that autism may be characterized by a perturbation of the processes that give rise to typical neural sex differentiation. These processes involve a complex interplay between genes, gonadal hormones, and social and cognitive development as they relate to sex identity (Hines, 2011). Indeed, perturbations to some of these processes (e.g., elevated fetal androgen exposure, elevated male hormone levels in females with ASDs) have been observed in association with ASDs (Baron-Cohen et al., 2015; Bejerot et al., 2012)

Moreover, relationships between functional connectivity and autism symptoms were detected in ASD girls, but not in boys, with weaker connectivity in the sensorimotor networks associated with history of greater early autism symptomatology (measured with the ADI-R) among girls with ASDs. This pattern suggests that in 7 – 17 years old girls with ASDs, greater network integration is associated with decreased early symptomatology. However, no such relationship between functional network integration and ASD symptoms was observed in boys. While these analyses do not reveal underlying causes for the differing patterns of association between sensorimotor network integration and symptom presentation in boys and girls with ASDs, the distinct relationships observed in boys compared to girls highlight the importance of investigating these patterns separately in males and females on the spectrum, as such patterns may go unnoticed in mixed-sex samples.

These findings need to be interpreted in the context of several limitations. First, our dataset lacked measures on pubertal status and did not control for menarche/menstruation in girls, and therefore, we could not rule out the potential confound of these differential sex effects on brain organization and functional connectivity (Klapwijk et al., 2013; Lisofsky et al., 2015; Marceau, Ram, Houts, Grimm, & Susman, 2011; Tanner, 1962). Further, given that, on average, boys begin puberty later than girls (Tanner, 1962), boys and girls in this study were likely not matched on pubertal status (as the groups were matched on chronological age). Importantly, however, pubertal timing onset has been shown to be equivalent in children and adolescents with and without ASDs (May, Pang, O’Connell, & Williams, 2017). Additionally, while allowing for greater sample sizes of girls and boys with ASDs, the use of multisite data has inevitably contributed noise (due to distinct phenotypic inclusion / exclusion criteria, variability between scanners, acquisition parameters, etc.), although variability due to site was statistically controlled in analyses. Additionally, datasets from ABIDE provide limited behavioral and phenotypic data, which restricted our ability to examine whether the observed effects were related to clinical or phenotypic variability.

The current analyses of sex-related differences in resting-state functional connectivity in children and adolescents with and without ASDs revealed a lack of typical sex differentiation in DMN functional connectivity in boys and girls on the spectrum. These distinct brain-behavior relationships provide support for the notion that ASD may manifest differently in girls than boys. As such, it remains critically important to include females in studies of ASDs, as much of what is known about this heterogeneous set of disorders is based on males. Rather than excluding females, studies should account for the effects of sex in their analyses by testing and controlling for sex effects in their samples. Further, as female ASD datasets are scarce, data-sharing initiatives (e.g., ABIDE I and II) remain a very useful resource for learning about unique features of ASDs in girls and women on the spectrum, which, in turn, contributes to our understanding of ASDs as a complex set of neurodevelopmental disorders.