**Evaluation of Chromosome Microarray Analysis in a Large Cohort of Females with Autism Spectrum Disorders: A Single Center Italian Study**

**Sara Calderoni 1,2,\* ,**† **, Ivana Ricca 3,**† **, Giulia Balboni <sup>4</sup> , Romina Cagiano <sup>1</sup> , Denise Cassandrini <sup>3</sup> , Stefano Doccini <sup>3</sup> , Angela Cosenza <sup>1</sup> , Deborah Tolomeo 3,5, Ra**ff**aella Tancredi <sup>1</sup> , Filippo Maria Santorelli <sup>3</sup> and Filippo Muratori 1,2**


Received: 7 September 2020; Accepted: 21 September 2020; Published: 9 October 2020

**Abstract:** Autism spectrum disorders (ASD) encompass a heterogeneous group of neurodevelopmental disorders resulting from the complex interaction between genetic and environmental factors. Thanks to the chromosome microarray analysis (CMA) in clinical practice, the accurate identification and characterization of submicroscopic deletions/duplications (copy number variants, CNVs) associated with ASD was made possible. However, the widely acknowledged excess of males on the autism spectrum reflects on a paucity of CMA studies specifically focused on females with ASD (f-ASD). In this framework, we aim to evaluate the frequency of causative CNVs in a single-center cohort of idiopathic f-ASD. Among the 90 f-ASD analyzed, we found 20 patients with one or two potentially pathogenic CNVs, including those previously associated with ASD (located at 16*p*13.2 16*p*11.2, 15*q*11.2, and 22*q*11.21 regions). An exploratory genotype/phenotype analysis revealed that the f-ASD with causative CNVs had statistically significantly lower restrictive and repetitive behaviors than those without CNVs or with non-causative CNVs. Future work should focus on further understanding of f-ASD genetic underpinnings, taking advantage of next-generation sequencing technologies, with the ultimate goal of contributing to precision medicine in ASD.

**Keywords:** autism spectrum disorders; copy number variants; females; Array-Comparative Genomic Hybridization (Array-CGH)

#### **1. Introduction**

Autism spectrum disorders (ASD) are a heterogeneous group of neurodevelopmental pathologies characterized by early onset abnormalities in social communication and interaction, as well as atypically restricted and repetitive behaviors and interests [1]. Despite the exact pathogenesis of idiopathic ASD not yet being fully elucidated, recent evidences suggest an interaction between genetic liability and environmental influences in producing early alteration of brain development [2].

In particular, among environmental risk factors, several maternal factors (including age ≥ 35 years, chronic hypertension, preeclampsia, gestational hypertension, and overweight before or during pregnancy) were associated with ASD in an updated review of the literature [3]. Updated data on the prevalence of ASD in the US (Centers for Disease Control and Prevention, CDC [4]) identified 1 in 54 children as having ASD, while the estimated prevalence of ASD in Italian population is 1 in 87, according to a recent investigation [5].

Crucially, since the first descriptions of autism [6,7], a strong male bias in ASD prevalence has been consistently observed, which becomes even more pronounced in individuals without intellectual disability, according to data from the 80s [8,9]. More recent studies have challenged this assertion, suggesting that missed or wrong diagnoses of ASD females, especially of those with good intellectual and language abilities, contribute to the skewed sex ratio in ASD [10].

The exact mechanisms underlying male vulnerability or female protection in ASD remain complex and scarcely investigated. A multifactorial model has been proposed where a mixture of gene variants and environmental factors contribute to liability, possibly interacting with sex-specific pathways such as those related to hormones or immune function [11,12].

Genetic investigations in ASD revealed frequently sexually dimorphic results. For example, a greater number of de novo copy number variants (CNVs) [13–16] as well as a higher rate of de novo single nucleotide variants (SNVs) found in exome sequences [17,18] have been observed in females with ASD (f-ASD) than in male cases, especially non-sense and splice site [19,20]. Conversely, a more recent study pointed to sex-specific mutations, specifically on the X chromosome, that may contribute to male prevalence in ASD [21]. On the other hand, as far as sex differences in symptom profiles are concerned, some previous studies suggested different phenotypic features in females than in males with ASD [22] like lower IQ [23], more impaired social and/or communicative functioning [24], psychopathological problems [25] and milder restricted and repetitive behaviors [26–28]. However, this issue remains controversial [29–33]. Females with ASD displayed also a higher rate of co-occurring neurological conditions than ASD males, i.e., microcephaly, developmental regression in socialization, minor neurological and musculoskeletal deficits [34], and epilepsy [35], all pointing to sex differences in genetic backgrounds.

The advent of chromosome microarray analysis (CMA) in clinical practice [36] allows for fast detection and accurate characterization of submicroscopic deletions and duplications (CNVs) of genomic DNA associated with ASD [37,38]. Learning societies and ASD experts recommend CMA as part of the first-line evaluation for individuals with ASD [39–41]. However, CMA brings up a higher level of polymorphic genomic rearrangements and the process to attribute causality in complex conditions such as ASD is not easy and straightforward.

This study aims to investigate the frequency of causative CNVs in a single-center cross-sectional idiopathic f-ASD cohort to delineate possible genotype/phenotype associations.

#### **2. Methods**

We collected the clinical data of a group of 93 females referred consecutively to the Autism Spectrum Disorders Unit of our Children Neuropsychiatry Hospital between 2015 and 2016. The age at the last clinical evaluation ranged from 21 months to 17 years. All participants received a clinical diagnosis of ASD based on the criteria of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) [1]. All the patients were unrelated.

According to our ASD-screening protocol, neurometabolic conditions and hypoxic-ischemic injury were investigated. All participants were evaluated by an expert clinical geneticist in order to exclude recognizable monogenic syndromes. Prior to this study, each individual had also been tested for the expanded repeat sequences in 5<sup>0</sup> -UTR of the *FMR1* gene as previously reported [42].

Based on this screening, we excluded two females with a history of perinatal hypoxia and diffuse white matter disease detected on brain magnetic resonance imaging (MRI), and one patient with macrocephaly harboring a pathogenic mutation in *PTEN*. In a single case (patient P11) we analyzed CNVs in spite of her presentation of a low-level somatic mosaicism for a fully-mutated/pre-mutated *FMR1* allele, because the patient's phenotype could not be fully explained by this genetic condition.

Hence, we tested 90 ASD female individuals for CNVs. Participants were classified as clinically affected by "essential" autism, based on the absence of major congenital abnormalities and major dysmorphism [43,44].

Cognitive evaluation was performed in 87 participants with specific cognitive scales based on the age and the language level. According to the age, children were tested respectively with the Griffiths Mental Development Scale—Revised (GMDS-R) [45], Wechsler Preschool and Primary Scale of Intelligence—third edition (WPPSI, III) [46] or Wechsler Intelligence Scale for Children— IV (WISC, IV) [47]. The evaluation of non-verbal females was performed using the Leiter International Performance Scale-Revised (Leiter-R) [48]. In three participants, the cognitive assessment was not performed because of scarce compliance due to severe autism symptoms.

Clinical assessment of expressive language skills defined females with a complete absence of language (*n* = 27) and a group of "verbal" f-ASD (*n* = 63).

The semi-structured Autism Diagnostic Observation Schedule second edition (ADOS-2) evaluation [49], which provides a measure of autism severity, was available in 67 participants. We recorded the score on the Social Affect (SA) and the Restricted and Repetitive Behaviors (RRB) domains for each proband. Since we used different ADOS modules according to the non-echolalic expressive language level of each patient at the time of the evaluation, we converted the global ADOS scores and the sub-scores of the SA and RRB domains in the corresponding Calibrated Severity Score (CSS) [50,51].

This study was approved by the Pediatric Ethic Committee of Tuscany Region (Italy), and was performed according to the Declaration of Helsinki and its later amendments or comparable ethical standards. All parents or legal representatives signed an informed consent form before the inclusion of their child in the study. The identities of all individuals were omitted.

#### **3. Procedure**

#### *3.1. Genetic Analysis*

CMA analyses were performed using the Agilent 8 × 60 K Microarray oligonucleotide platform with a median resolution of 100 Kbp, according to the manufacture's protocol (Agilent Technologies, Santa Clara, CA, USA). CNV coordinates refer to the Genome Reference Consortium Human Build 37 (GRCh37/hg19).

In each proband, CNVs were confirmed by quantitative polymerase chain reaction (qPCR). Segregation analyses in parental DNA (whenever available) were performed by qPCR. Polymorphic CNVs, based on Database of Genomic Variants data (DGV) [52]), were filtered out.

Non-polymorphic CNVs were classified as "causative" (C-CNVs) or "non-causative" (N-CNVs) according to the American College of Medical Genetics and Genomics (ACMG) guidelines [53]. We considered as "causative": (i) CNVs encompassing genomic regions or genes associated with ASD or with other neuropsychiatric conditions (i.e., intellectual disability, epilepsy and schizophrenia) in the Online Mendelian Inheritance in Man (OMIM) database [54]; (ii) CNVs containing "high confidence" ASD-genes reported in the Simons Foundation Autism Research Initiative (SFARI) Gene database [55] with a score < 3 or in the Autism Knowledge Base version 2.0 (Autism KB 2.0) database [56] with a score > 16; (iii) CNVs involving "candidate-genes" for ASD either reported in association with autism in literature, or listed in the aforementioned databases and with a SFARI Gene score ≥ 3 or an Autism KB score ≤ 16 (suggestive or "low confidence" candidate-genes). Conversely, CNVs were considered non-causative (N-CNVs) if they have never been associated with ASD or other neurodevelopmental disorders (NDDs). Patients who tested negative for CNVs were classified as "without CNVs" (w-CNVs).

To recognize significantly enriched functional modules, ASD-candidate genes encompassed by C-CNVs were evaluated by bioinformatics tools. A Core analysis run in the Variant Effects Analysis

mode through the use of the Ingenuity Pathway Analysis (IPA) software [57] figured out cellular processes related to our gene dataset (21 genes). A functional network encompassing our ASD-candidate genes was generated. Bridging nodes were denoted evaluating both direct and indirect interactions with stringent level of confidence and only related to neurological diseases. Gene ontology (GO) categorization was carried out using ToppGene Suite [58]. The top three ontologies for Molecular Functions and Cellular Component were annotated and statistical significance of GO terms was reported as -log10 (*p*-value).

#### *3.2. Statistical Analyses*

We used a chi-square test to investigate the association between the CNVs subtype and the type of CNVs (duplication or deletion) and the pattern of inheritance (de novo or inherited, paternal or maternal). A Mann–Whitney test was used to verify if there were any differences in the CNVs burden of the different CNVs subtypes (excluding patient P23 who carried a whole X-chromosome duplication).

We also investigated the phenotype of the individuals with the different CNVs subtypes testing with the chi-square test the association between the CNVs subtype and cognitive (IQ ≤ 70 vs. >70) and language (non-verbal vs. verbal) levels. A Mann–Whitney test was used to ascertain that the groups with different CNVs subtype were matched on age and to verify if there were any differences in the CCS score obtained on the total ADOS and on its AS and RRB domains. In case of statistically significantly differences we compute for *r* score as effect size index. This was interpreted as negligible (*r* < 0.10), small (0.10 ≤ *r* < 0.30), medium (0.30 ≤ *r* < 0.50), or large (*r* ≥ 0.50).

#### **4. Results**

#### *4.1. Chromosome Microarray Analysis (CMA)*

We performed CMA in 90 females affected by idiopathic ASD, detecting 35 CNVs (17 duplications and 18 deletions) in 29 (32.2%). Twenty-three participants had one CNV and six carried 2 imbalances. Sixty-one f-ASD were considered w-CNVs (67.8% of the whole group).

Out of 35 CNVs, 25 were classified C-CNVS (71.4%) and 10 N-CNVs (28.6%). In the whole group of 90 f-ASD, 20 patients harbored at least one possible disease-causing CNV (diagnostic yield 22.2%) (Figure 1).

**Figure 1.** Graphical representation of chromosome microarray analysis (CMA) results in our group of 90 females affected by autism. In the pie chart is depicted the percentage of individuals with causative copy number variants (C-CNVs), non-causative copy number variants (N-CNVs) or without copy number variants (w-CNVs).

Table 1 illustrates the results of CMA investigations. There were not recurrent C-CNVs, with the exception of two unrelated subjects who harbored a 15q11-q13 microduplication. Ten CNVs involved genomic regions already associated with known contiguous gene-deletion/duplication syndromes

1

associated with ASD or NDDs, 5 CNVs encompassed "high-confident" ASD-genes and ten involved genes reported in literature or in the SFARI Gene/Autism KB databases as possible candidates for autism.

The function and evidence of possible disease-association of the reported candidate-genes are summarized in Table 2. Bioinformatic analysis showed that 11 out of 21 of the reported disease-associated and candidate genes are involved in synaptic structure and transmission (*ADARB1*, *ASIC2*, *CADM2*, *DMD*, *GRIN2A*, *GRM7*, *NEDD4, NRXN1*, *PCDH15*, *PTPRD*, *TRPM2*) (Figure 2).

In 24 f-ASD, carrying 29 CNVs, we assessed a de novo origin in 8 and a paternal in 12, whereas CNVs were maternally-inherited in 9 patients. In 5 children we could not assess segregation because of lack of parental DNA. Table 3 shows the proportion of duplications and deletions and the mode of inheritance in relation to the different subtypes of CNVs. Overall, the rate of de novo CNV was 9.4%. All de novo CNVs involved known NDDs-associated genes/chromosomal regions. CNVs encompassing suggestive or "low confidence" ASD-genes were all inherited; 6 out 9 disrupted more than one NDD-gene or were associated with an additional C-CNV. Seven out of 9 maternally inherited vs. 6 out of 12 paternally inherited CNVs were causative.


**Table 1.** Chromosomal microarray (CMA) results in the 29 participants carrying at least one Copy Number Variant (CNV). For each participant with positiveresults are reported the genomic location and breakpoints of each CNV, the CNV subtype (deletion or duplication), the size in base pairs, the inheritance theassociatedknowngeneticsyndromeorAutismSpectrumDisorders(ASD)candidategenesinvolvedintherearrangement,andtheCNVclassificationor

 CMA

status,


*NRXN1* *TSPAN7*

*ITGB2*

21q22.3 (P4)

 Integrin

 B2, adhesion

 molecule

 implicated

 in synaptic

 pruning

Xp11.4 (P1)

2p16.2 (P29)

and growth regulation

Cell adhesion

nervous

contacts.

Member

 of the

integrins.

 It may have a role in neurite outgrowth

tetraspanin

 family, encodes

Alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionate

*Suggestive*

 *or "low confidence"*

 *candidate*

 a cell surface

 and

 (AMPA)

 receptor

 tra

fficking

 *ASD-genes* glycoprotein

 that complex

 with 3/2

 NR/3

 system required

 molecule,

 form a complex

 for

 with

neurotransmission

 and involved

 in the formation

 of synaptic

1/68

neuroligins

 at synapses

 in the central



candidates for autism.

ten involved genes reported in literature or in the SFARI Gene/Autism KB databases as possible

The function and evidence of possible disease‐association of the reported candidate‐genes are summarized in Table 2. Bioinformatic analysis showed that 11 out of 21 of the reported disease‐associated and candidate genes are involved in synaptic structure and transmission

**Figure 2**. Bioinformatic analyses performed on ASD‐candidate genes encompassed by C‐CNVs. (**A**) A Core analysis run in Variant Effects Analysis mode using the Ingenuity Pathway Analysis software figured out cellular processes related to our gene dataset (21 genes) generating a functional network encompassing 11 genes (in red). *Synaptic transmission* resulted the most significant functional annotation (*p*‐value 6.05 × 10<sup>−</sup>9). Bridging nodes (in white) were denoted evaluating both direct and indirect interactions related only to neurological diseases and with stringent level of confidence (**B**). Gene ontology (GO) categorization was carried out using ToppGene Suite. Top three ontologies for *Molecular Function* (dark grey) and *Cellular Component* (light grey) were annotated; statistical significance of GO terms was reported as −log10 (*p‐*value). The number of genes belonging to each category was reported on the right of each bar. **Figure 2.** Bioinformatic analyses performed on ASD-candidate genes encompassed by C-CNVs. (**A**) A Core analysis run in Variant Effects Analysis mode using the Ingenuity Pathway Analysis software figured out cellular processes related to our gene dataset (21 genes) generating a functional network encompassing 11 genes (in red). *Synaptic transmission* resulted the most significant functional annotation (*p*-value 6.05 × 10−<sup>9</sup> ). Bridging nodes (in white) were denoted evaluating both direct and indirect interactions related only to neurological diseases and with stringent level of confidence (**B**). Gene ontology (GO) categorization was carried out using ToppGene Suite. Top three ontologies for *Molecular Function* (dark grey) and *Cellular Component* (light grey) were annotated; statistical significance of GO terms was reported as −log10 (*p-*value). The number of genes belonging to each category was reported on the right of each bar.

#### *4.2. Phenotypic Characterization*

Twenty-seven f-ASD had an absence of language whereas 63 were "verbal".

Cognitive evaluation was performed in 87 participants, being three participants unfit for psychometric testing. Forty-two of the tested individuals had IQ ≤ 70 and 45 had IQ ≥ 70.

The 67 participants tested with ADOS-2 had the following mean (SD) Total, SA and RRB ADOS CSS, respectively: 6.57 (2.36), 6.79 (2.34), and 7.22 (2.30).

Supplementary Table S1 recapitulates clinical data of the studied population.


**Table 3.** Proportion of deletions vs. duplications and pattern of inheritance of the reported CNVs according to their classification (causative vs. non-causative).

*Note:* Inheritance was assessed in 29 out of 35 CNVs. C-CNVs = causative CNVs; N-CNVs = non-causative CNVs; *n* = number of CNVs for each group.

#### *4.3. Statistical Analysis*

We observed a statistically significant association between the heritage (de novo vs. maternal and paternal) and the subtypes of CNVs (C-CNVs vs. N-CNVs) (*Chi*<sup>2</sup> (1) = 4.21, *p* = 0.04). Indeed, all N-CNVs were transmitted and never arose de novo while all de novo CNVs were causative (38% of C-CNVs); 7 out of 9 (77.8%) C-CNVs were maternal and 6 out of 12 (50%) were paternal.

Whilst the type of genomic micro-rearrangement (deletion vs. duplication) was not statistically correlated to causative/non-causative definition (*Chi*<sup>2</sup> (1) = 0.41, *p* = 0.52), not considering CNVs associated with contiguous-gene syndromes, most of the breakpoints of causative duplications lie within at least one NDD-candidate gene (*n* = 6/8). C-CNVs had a CNVs burden value statistical significantly higher than those of the N-CNVs subtypes (mean (SD) = 1.14 (1.43) vs. 0.19 (0.16); Mann-Whitney *U* = 52.50, *z* = 2.56, *p* = 0.01, *r* = 0.49).

Table 4 shows the age, the cognitive and linguistic level as well as the autism severity of the three groups of individuals according to different CNV subtypes (causative, non-causative and without CNVs).

To investigate whether there were significant differences in clinical features between groups, we regrouped participants with negative CMA results (N-CNVs and w-CNVs) and compared their characteristics with cases with C-CNVs. The two groups resulted matched for age [mean (SD) = 66.95 (38.55) vs. 56.74 (38.03); Mann–Whitney *U* = 523.00, *z* = 1.72, *p* = 0.09].

We found that there were no differences between the two groups on the cognitive level (IQ ≤ 70 vs. IQ > 70; *Chi*<sup>2</sup> (1) = 0.47, *p* = 0.49), language level (non-verbal vs. verbal; *Chi*<sup>2</sup> (1) = 0.31, *p* = 0.58), and on the CSS obtained on the total score and on the AS ADOS domain (Mann–Whitney *U* = 262.50, *z* = 1.42, *p* = 0.16; Mann–Whitney *U* = 303.00, *z* = 0.77, *p* = 0.44).

The relative frequencies of the phenotypic features were the following: in the group with C-CNVs, 55% (11/20) had IQ ≤ 70; 60% had a moderate-severe level of autism symptoms (9/15), 35% had absence of language (7/20); in the group with negative CMA, 46% (31/37) had IQ ≤ 70; 75% had a moderate-severe level of autism symptoms (47/62), 28% had absent language (20/70).

Conversely, we found that the f-ASD with C-CNVs had a statistically significantly lower CSS on the RRB ADOS domain that those without CNVs or with non-causative (mean (SD) = 6.08 (2.14) vs. 7.50 (2.27); Mann–Whitney *U* = 197, *z* = 2.48, *p* = 0.01, *r* = 0.30).

**Table 4.** Demographic features of participants grouped according to CMA results. For each group (with causative and non-causative CNVs, or without CNVs) are reported the mean age at the last examination (in months), the rate of patients with a IQ level > 70 vs. ≤70, the rate of verbal vs. non-verbal patients, and the mean calibrated severity scores (CSS) of the global Autism Diagnostic Observation Schedule (ADOS) scores and the sub-scores of the Social Affect (SA) and Restricted and Repetitive Behaviors (RRB) domains. The language level was assessed in all 90 participants, the IQ level and the ADOS scores were available for 87 and 67 of the 90 individuals, respectively.


C-CNVs = participants with causative CNVs; N-CNVs = participants with non-causative CNVs; w-CNVs = participants without; *n* = number of patients for each group; SD = standard deviation.

#### **5. Discussion**

Although a recent meta-analysis and multidisciplinary consensus statement proposes exome sequencing at the beginning of the evaluation of unexplained neurodevelopmental disorders [73], CMA is still the recommended first-tier genetic analysis in the evaluation of ASD subjects [40,74].

In the last few years, investigations of large cohorts of ASD individuals [13,37,75] have identified a high burden of CNVs with rare C-CNVs being found in 5–10% of idiopathic ASD [76]. However, these data are often affected by gender-bias due to the high M/F ratio in the vast majority of the studies and even more recent investigations addressing type and frequency of C-CNVs did not allow with few exceptions—for separate gender examinations due to relatively small sample size [77–80].

Herein, we focused exclusively on a cohort of f-ASD and we found clinically significant CNVs in about 22% of patients. Few investigations have considered CNVs and clinical features in f-ASD in comparison with ASD males. In one study, large CNVs (>400 kb) were more frequent in f-ASD than in males (29% vs. 16%), and this difference was even higher (F/M 3:1) if analyses were limited to regions containing genes involved in NDDs [81]. In a similar vein, Levy and colleagues (2011) [13] detected that f-ASD have a high frequency of de novo CNVs (11.7% vs. 7.4% in males), and Sanders et al. (2015) [15] identified a significant difference in the rate of de novo CNVs between boys (5.3%) and girls (8.7%). Our numbers in an only girl cross-sectional, monocentric study denote a similar sex effect with a high diagnostic yield and a 9.4% occurrence of de novo variants.

All de novo CNVs involved known NDDs-associated chromosomal regions whereas CNVs encompassing suggestive or "low confidence" ASD-genes were all inherited and mostly disrupting more than one NDD-gene or associated with an additional C-CNV. Among C-CNVs, there was an excess of maternally-inherited potentially pathogenic CNVs. These findings support the "two-hit model" suggested in previous studies in which the compound effect of a small number of rare variants may contribute to phenotypic heterogeneity of ASD [82].

While literature in the ASD field reported an excess of clinically-significant deletions, we did not find a correlation between the type of genomic rearrangement and causative/non causative definition. Haploinsufficiency for genes within a deletion is a well-recognized cause of genetic disease. Conversely, interpreting the phenotypic consequences of microduplications is often challenging because the pathogenicity of most duplications cannot be explained by triplosensitivity. Sequencing the

breakpoints of 119 duplications, Newman et al. (2015) demonstrated that, rather than an extra copy effect, the phenotype of microduplications can be related to the misregulation of genes that span the breakpoints, through loss-of-function mechanisms due to altered transcription or translation or to the creation of fusion proteins with unknown functions [83]. In our f-ASD cohort, most of the causative non-syndromic duplications breakpoints disrupted at least one NDD-candidate gene, hence we can suppose that the pathogenic phenotype could be caused by similar mechanisms.

Unlike previous literature results [78], we did not find any association between C-CNVs and IQ or language deficits. Analyzing the phenotypic features of females with C-CNVs versus those with negative CMA results, we only observed statistically significantly lower scores on the restricted repetitive behaviors (RRB) ADOS domain in f-ASD with clinically significant variants. Recently, Barone et al. reported more severe autistic symptoms in individuals with C-CNVs [79]. The discrepancies with our data could reflect the diverse characteristics of the studied population, indeed several studies suggested a sex effect on RRB scores, which are reported to be repeatedly lower in female than in male subgroups [28,84–86]. Crucially, several lines of evidence suggest that social-communication (SC) and RRB symptom domains are underpinned by different genetic mechanisms. For instance, a recent genome-wide association study demonstrated that the RRB trait "systemizing" is heritable and genetically correlated with autism in the general population and that the SC and RRB domains in autistic subjects show low shared genetics [87]. In particular, the contribution of genetic factors to the RRB domain is sustained by their significative presence on both parents [88] and siblings [89] of probands with ASD. Overall, the impact of C-CNVs on ASD symptoms is still unclear and a recent work highlighted the contribution of environmental factors (i.e., maternal infections during pregnancy) on RRB severity in individuals with CNVs [90]. We can only speculate that we registered lower RRB scores in our f-ASD with positive CMA results because this sample represents the mild-end of a genomic "simple" disorder, while those girls with negative results could reflect the group of f-ASD with "complex" multifactorial etiology, as the largest portion of idiopathic autistic males.

With the exception of two subjects with a 15q11-q13 microduplication, no overlapping CNVs were detected, confirming the high genetic heterogeneity of ASD. Fifteen CNVs involved ASD/NDDs-associated genes or genomic regions already identified, whereas 10 CNVs encompassed genes reported as possible candidates for ASD in literature or in ASD databases (Tables 1 and 2). The contribution of each CNV to the phenotype of our f-ASD patients is discussed in the Supplementary File S1. Out of this list, some cases appear worth discussing.

The known contiguous-gene deletion/duplication syndromes detected in our cases were associated with a diagnosis of "idiopathic" ASD because these patients did not display any of the additional non-neurodevelopmental features specific of these syndromes, as dysmorphisms or congenital defects which can be seen in Smith-Magenis (P8), 17q12 microdeletion (P10), 2p15p16 deletion (P19), 22q11 duplication (P28) and SHOX duplication (P29) syndromes. These patients could represent the mild-end of the phenotypic spectrum of these genomic disorders, due to the "NDDs-protective effect" reported in females [16].

In some cases, reverse phenotyping allowed the investigation and prevention of important comorbidities, as in P25, who carries a de novo partial duplication of the *DMD* gene, which in females could manifest with muscle weakness and cardiomyopathy, and in P20, who carries a 16p11.2 duplication widely reported in ASD studies which is associated with the risk of developing psychotic symptoms [91].

Among clinically relevant rearrangements, aneuploidy was identified in a single subject, who presented an X chromosome trisomy (47, XXX). Interestingly, data in the literature did not report a greater risk for autism in X chromosome trisomy [92], even if difficulties in social functioning and, more broadly, an increased vulnerability for autistic traits are described [68].

The de novo 16p13 duplication detected in one patient (P3) involves partially *UPS7*. Variants affecting this gene were recently reported in 23 individuals with syndromic Developmental Delay/Intellectual Disability [93], and about half of reported subjects had ASD. P3 presents mild

motor developmental delay, absent speech, behavioral anomalies and ASD, suggesting that *USP7* haploinsufficiency should be suspected in a case of ASD with absence of speech and behavioral disorders. CNVs detected in P3 spans also *GRIN2A* and *RBFOX1*, so we cannot exclude a possible additional role of these genes in the phenotype of the patient.

The deletions found in P11, P14 and P15 reinforce the evidence of a possible contribution of *PCDH15*, *GRM7*, *CADM2* and *IMMP2L* genes to ASD susceptibility.

Finally, five CNVs spanned some "low-confidence" ASD-genes, which can be investigated in future studies (i.e., *TRPM2*, *ADARB1*, *RFX7*, *NEDD4*, *ASIC2*, *PTPRD*, *ST6GAL2*).

When new and old genes pinpointed by CMA studies were combined in functional modules using IPA and ToppGene Suite, we observed an enrichment in genes involved in synaptic function and transmission, which are well-established biological processes involved in autism and NDDs [94].

In conclusion, this study provides a representative picture of the spectrum of CNV in f-ASD investigated in a clinical setting. As expected, no specific CNVs have been found to be required for developing ASD, supporting the heterogeneity of affected molecular pathways. However, genes in the C-CNVs of our sample of f-ASD code mainly for proteins that could be grouped in two different functional systems: synaptic function/structure, and mRNA/protein processing. Of note, environmental exposures during specific windows of vulnerabilities in prenatal and perinatal life critically interact with genetic susceptibility contributing to ASD pathogenesis [95]. Our study suggests that females with idiopathic ASD have a high rate of pathogenic CNVs encompassing both known and new candidate ASD genes. Hence, studies on large samples of f-ASD carefully assessed from a clinical point of view could help in unraveling the genetic determinants of autism. Moreover, f-ASD with normal-array comparative genomic hybridization analysis could benefit from whole exome or genome sequencing [96], paving the way for the implementation of personalized treatments based on genetic findings.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2075-4426/10/4/160/s1, Table S1: Phenotipic characteristics of participants; Supplementary File S1: Contribution of each CNV to the phenotype of f-ASD patients.

**Author Contributions:** Conceptualization of the article, S.C., I.R., G.B., F.M.S., F.M.; Major contributors in writing the manuscript, S.C., I.R.; Sample collection and clinical characterization of patients, A.C., R.T.; Data Acquisition, R.C., D.T.; Statistical Analyses, G.B.; Genetic Analyses, D.C., S.D.; Contributed to review and editing the manuscript, F.M.S., F.M. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work has been partially supported by grant from the IRCCS Fondazione Stella Maris (Ricerca Corrente, and the "5 × 1000" voluntary contributions, Italian Ministry of Health). S.C. was partially funded by AIMS-2-Trials.

**Acknowledgments:** We gratefully acknowledge all the subjects who have participated in the study.

**Conflicts of Interest:** The authors declare that they have no conflict of interest.

**Ethical Statement:** The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the Pediatric Ethic Committee of Tuscany Region.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

**Andrew R. Pines <sup>1</sup> , Bethany Sussman <sup>2</sup> , Sarah N. Wyckoff <sup>2</sup> , Patrick J. McCarty <sup>3</sup> , Raymond Bunch <sup>4</sup> , Richard E. Frye 5,\* and Varina L. Boerwinkle <sup>2</sup>**


**Abstract:** Resting-state functional magnetic resonance imaging (rs-fMRI) has the potential to investigate abnormalities in brain network structure and connectivity on an individual level in neurodevelopmental disorders, such as autism spectrum disorder (ASD), paving the way toward using this technology for a personalized, precision medicine approach to diagnosis and treatment. Using a case-control design, we compared five patients with severe regressive-type ASD to five patients with temporal lobe epilepsy (TLE) to examine the association between brain network characteristics and diagnosis. All children with ASD and TLE demonstrated intact motor, language, and frontoparietal (FP) networks. However, aberrant networks not usually seen in the typical brain were also found. These aberrant networks were located in the motor (40%), language (80%), and FP (100%) regions in children with ASD, while children with TLE only presented with aberrant networks in the motor (40%) and language (20%) regions, in addition to identified seizure onset zones. Fisher's exact test indicated a significant relationship between aberrant FP networks and diagnosis (*p =* 0.008), with ASD and atypical FP networks co-occurring more frequently than expected by chance. Despite severe cognitive delays, children with regressive-type ASD may demonstrate intact typical cortical network activation despite an inability to use these cognitive facilities. The functions of these intact cognitive networks may not be fully expressed, potentially because aberrant networks interfere with their long-range signaling, thus creating a unique "locked-in network" syndrome.

**Keywords:** autism spectrum disorder; locked-in network syndrome; resting-state functional magnetic resonance imaging; temporal lobe epilepsy

#### **1. Introduction**

Autism spectrum disorder (ASD) has a prevalence of 1 in 54 children in the United States and is characterized by deficits in social communication and interaction, and restricted and repetitive behaviors [1]. The neurological basis of ASD remains elusive in this highly heterogenous population. Resting-state functional magnetic resonance imaging (rs-fMRI) is a promising tool that has been used to analyze neural networks in patients with neurodevelopmental and neurological disorders. A review of prior group-level rs-fMRI investigations suggests hypo-, hyper-, and mixed-connectivity patterns among individuals with ASD, with contradictory findings attributed to differences in age, sex, comorbidities, and/or variations in the rs-fMRI scanning and analysis procedures [2]. Other researchers

**Citation:** Pines, A.R.; Sussman, B.; Wyckoff, S.N.; McCarty, P.J.; Bunch, R.; Frye, R.E.; Boerwinkle, V.L. Locked-In Intact Functional Networks in Children with Autism Spectrum Disorder: A Case-Control Study. *J. Pers. Med.* **2021**, *11*, 854. https:// doi.org/10.3390/jpm11090854

Academic Editor: Chiara Villa

Received: 22 June 2021 Accepted: 25 August 2021 Published: 28 August 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

demonstrated that traditional case-control methods that assume homogeneity within clinical populations lead to a loss of subject-specific features of ASD at the group level and disguise the interindividual variation crucial for precision medicine [3].

To investigate these inconsistencies, a large-scale database replication study that characterized and evaluated connectivity patterns was conducted. The study reports evidence of reproducible ASD-associated functional hyper- and hypo- connectivity linked to clinical symptoms [4]. The authors reported that overall global connectivity was preserved in individuals with ASD, and hyperconnectivity patterns observed in the parietal and prefrontal regions were associated with the severity of deficits in communication and adaptive behavior. Furthermore, they suggested that the connectivity findings support the idea that individuals with ASD are unable to engage or disengage specific networks to the same degree as healthy, typically developing controls. If network engagement capacity is a key mechanism of ASD, then individual-based analysis may elucidate the network engagement potential and allow a better understanding of the clinical heterogeneity of ASD [5].

To address these knowledge gaps in ASD, we used a data-driven rs-fMRI wholebrain-network analysis to identify individual network pathology [6,7], similar to our work in epilepsy [8–10]. Independent component analysis (ICA) of rs-fMRI allows for the characterization and visualization of individual resting-state networks (RSNs), both typical and atypical [11–14]. In epilepsy, we utilize this method clinically to both identify the aberrant RSNs associated with seizure foci and to visualize intact cognitive RSNs. This allows the confirmation of intact cognitive networks and provides information regarding the proximity of intact cognitive networks to aberrant networks [15].

Similarly, in individuals with ASD, we may use rs-fMRI clinically to visualize network patterns in relation to the patient's clinical phenotype. One feature that differentiates individuals with ASD from those with brain injury is that individuals with ASD can have normal or extraordinary skills despite their disability, and some children with ASD can make substantial improvement with therapies, losing many, if not most, of their ASD symptoms. One of the most enigmatic subsets of ASD includes those that undergo neurodevelopmental regression, suddenly losing normal skills and rapidly developing the ASD phenotype. This suggests that the brains of some individuals with ASD have the capacity to support typical cognitive networks, at least at some time in their life, but may be unable to express these cognitive networks. For children with epilepsy, aberrant networks resulting from ongoing subclinical interictal discharges originating from the seizure onset zone (SOZ) can interfere with the function of typical cognitive brain networks, disrupting their ability to function optimally.

Thus, on the basis of the existing literature and our clinical observations, we hypothesized that some children with ASD may have intact cognitive networks identified on the individual level that are not fully realized, similar to children with epilepsy, perhaps because aberrant networks interfere with their function. However, unlike in children with epilepsy, we hypothesized that these aberrant networks are not localized to a SOZ. To investigate this possibility, subject-level ICA of rs-fMRI was used to identify and characterize the RSNs of children with regressive-type ASD and temporal lobe epilepsy (TLE). The unique relationship between patient and network characteristics was examined.

#### **2. Materials and Methods**

Using a case-control design, this study examined an age- and sex-matched cohort of children severely affected with regressive-type ASD, and a cohort of children with TLE without ASD. A 2017 meta-analysis reported an ASD pooled prevalence of 6.3% in patients with epilepsy, with a 41.9% risk of ASD in patients with focal seizures (TLE) [16]. Thus, a pathological control group was selected to distinguish ASD-specific atypical rsfMRI biomarkers from known TLE-specific markers given the increased comorbidity of these conditions. For the ASD cohort, children with regressive-type ASD were chosen since these children had documented normal development early in life, suggesting that their

brain could previously support normal cognitive networks. ASD was diagnosed using the Diagnostic Statistical Manual of Mental Disorders (5th ed., DSM-5).

The study sample size was limited due to rs-fMRI data availability for this patient population. Five patients with regressive-type ASD who were not making sufficient progress with standard therapy and five patients with TLE undergoing pre-surgical planning for epilepsy surgery underwent clinically indicated rs-fMRI at Phoenix Children's Hospital (PCH) between November 2018 and December 2020. The patient cohorts were age- and sex-matched. The PCH Institutional Review Board approved the study, and caretakers provided informed consent to authorize secondary analysis of the rs-fMRI data and review of the medical records.

#### *2.1. Resting-State MRI*

The rs-fMRI images were acquired and analyzed per prior reported standards [15]. Acquisition was from a 3 T MRI scanner (Ingenuity; Philips Medical Systems, Best, The Netherlands) with a 32-channel head coil. Patients received conscious sedation by propofol, per hospital clinical standards. Acquisition of rs-fMRI consisted of two 10-min runs totaling 20 min. Parameters were 2000 millisecond repetition time (TR), 30 millisecond echo time (TE), 80 × 80 matrix size, 80◦ flip angle, 46 slices, 3.4 mm slice thickness with no gap, 3 × 3 mm in-plane resolution, interleaved acquisition, and 600 total volumes. For anatomical reference, a T1-weighted turbo field-echo whole-brain sequence was obtained with TR 9 milliseconds, TE 4 milliseconds, flip angle 80◦ , slice thickness 0.9 mm, and an in-plane resolution of 0.9 × 0.9 mm.

#### *2.2. Independent Component Analysis Approach*

ICA is driven by empirical data rather than a priori information. Briefly, rs-fMRI voxels are grouped together into components according to similarity of blood oxygen level dependent (BOLD) signal oscillation [6,7]. The resulting independent components (ICs) are independently fluctuating clusters of brain activity or sources of noise that require expert review and interpretation [17]. ICA procedures were completed via the Multivariate Exploratory Linear Optimized Decomposition into Independent Components (MELODIC) tool [7]. The following standard preprocessing steps were applied: (1) deletion of the first five volumes to remove T1 saturation effects, (2) high-pass filtering at 100 s, (3) inter-leaved slice time correction, (4) no spatial smoothing, and (5) motion correction with MCLFIRT [18], with non-brain structures removed. Individual functional scans were registered to the patient's corresponding anatomical scan using linear registration [19], and optimized using boundary-based registration [20]. All participants had <1 mm head-motion displacement in any direction. As ICA was applied in the subject space, no standardized templates or spatial normalization procedures was performed. The total number of detected ICs was determined for each patient from established automated dimensionality estimates using a Bayesian approach, and an ICA threshold (*p* < 0.05) for IC detection was set by the standard local false discovery rate [6].

#### *2.3. Component Categorization*

IC categorization followed the working paradigm previously published [15], separating patient ICs into four categories—noise, typical RSNs, SOZ, and atypical (aberrant) networks—using criteria modified from established norms [11–15,21]. Noise components arise from respirations, cerebrospinal fluid movement, and tissue–fluid junctions. A component was determined to be noise if: it was not primarily located in grey matter; it varied significantly in coordination with physiological cycles (i.e., respiratory-related frequency range, 0.1–0.5 Hz; cardiac-related, 0.6–1.2 Hz; regular but fast oscillation pattern); or it was in a spatial distribution consistent with a machine-generated artifact [11,12]. Components determined to be neural networks were categorized as either typical RSNs, SOZs, or atypical (aberrant) networks. Typical RSNs were determined by visually comparing spatial features to established RSNs (e.g., motor, language, and frontoparietal), and comparing

temporal features of frequency and frequency power spectra with a known low-frequency, regular, slow-oscillating time course, and low-frequency power-spectra features of RSN norms [12–14]. SOZs were distinguished by a spatial pattern (more asymmetrically unilateral than expected, alternating localized activation-deactivation patterns of gray matter, and with a tapered tail from the cortex extending toward the ventricles) not conforming with noise or typical RSNs, an irregular time course, or containing a frequency >0.4 Hz [15]. Aberrant networks were distinguished by spatial locations that may overlap with known RSN, but do not conform to the RSN spatial pattern, noise, or SOZ criteria, having a regular sinusoidal oscillation pattern that is overlaid with irregular faster frequency, and having an atypical BOLD oscillation frequency >0.039 Hz [13–15]. rs-fMRI data were interpreted by an rs-fMRI specialist (senior author), wherein typical RSNs were operationally defined as those meeting the spatial and temporal criteria above by expert visual inspection, as in prior publications [12,15].

#### *2.4. Statistical Analysis*

Categorical variables (e.g., typical RSN: present, absent; aberrant (non-SOZ) network: present, absent) were generated for the motor, language, and frontoparietal networks for each patient. Due to the small sample size, a two-tailed Fisher's exact test was used to examine the significance of the association between the two factors in the contingency tables. Since, for each set of tests, we compared three cortical regions, requiring three statistical tests, the Bonferroni correction for inflated alpha was used to set the significance threshold at *p* <= 0.017 (0.5/3). An independent-samples t-test (two-tailed, *p* < 0.05) was used for the analysis of continuous variables.

#### **3. Results**

Figure 1 visually summarizes the clinical characteristics and typical and aberrant rsfMRI ICA-based networks for children with ASD. Figure 2 demonstrates a detailed example of typical RSNs and aberrant networks for a single ASD case (Patient 1). A narrative summary of the clinical and network features is presented below for all patients, followed by a statistical analysis of patient characteristics and resting-state networks.

#### *3.1. Patient Summaries of ASD Patients*

#### 3.1.1. Patient 1

A 7-year-old (yo) male progressively lost social interactions and eye contact from 9 to 15 months of age. At the time of rs-fMRI, he was diagnosed with ASD, cognitive delay, severe language impairment (LI), and developmental coordination disorder (DCD). Despite these deficits, frontoparietal (FP), language, and motor RSNs were intact. Aberrant networks were found over sensory, FP, and contralateral non-dominant language regions (Figure 2).

#### 3.1.2. Patient 2

A 4 yo female suddenly lost normal speech, social, language, cognitive, and motor abilities at 3.5 yo. At the time of rs-fMRI, she was diagnosed with ASD, borderline intellectual disability (ID), severe LI, and delayed visual-motor skills. Despite these deficits, FP, language, and motor RSNs were normal. Aberrant networks were found over the right FP, bilateral temporal, and opercular regions.

#### 3.1.3. Patient 3

A 16 yo female lost social skills and ceased making eye contact with others at 3 yo. At the time of rs-fMRI, she was diagnosed with ASD, ID, severe LI, and DCD. Despite these deficits, FP, language, and motor RSNs were detected. Aberrant networks were found over the left FP-temporal regions.

*J. Pers. Med.* **2021**, *11*, x FOR PEER REVIEW 5 of 12

**Figure 1.** Comparison of clinical and rs-fMRI findings in ASD patients. Columns 1–3: ASD patient typical motor, language, and frontoparietal network images and interpretation with corresponding discordant phenotypic clinical impairments. Column 4: ASD participant atypical (aberrant) network (non-SOZ, overlapping typical RSN) images. B, bilateral; DCD, developmental coordination disorder; F, frontal; FPN, frontoparietal network; L, left; MT, mesial temporal; Opr, operculum; R, right; RSN, resting state network; vmPFC, ventromedial prefrontal cortex. **Figure 1.** Comparison of clinical and rs-fMRI findings in ASD patients. Columns 1–3: ASD patient typical motor, language, and frontoparietal network images and interpretation with corresponding discordant phenotypic clinical impairments. Column 4: ASD participant atypical (aberrant) network (non-SOZ, overlapping typical RSN) images. B, bilateral; DCD, developmental coordination disorder; F, frontal; FPN, frontoparietal network; L, left; MT, mesial temporal; Opr, operculum; R, right; RSN, resting state network; vmPFC, ventromedial prefrontal cortex. *J. Pers. Med.* **2021**, *11*, x FOR PEER REVIEW 6 of 12

**Figure 2.** Example of **r**ight-sided language, frontoparietal, and aberrant networks in ASD Patient 1. Image shown is the sagittal, axial, and coronal T1-weighted MRI of Patient 1 with overlayed language (blue), frontoparietal (green), and aberrant (red) networks. Plotted are the respective network blood oxygen dependent signal (BOLD) power versus frequency (Hz/100), wherein typical is less than 6 Hz/100. The frontoparietal and language network spatial distribution and the BOLD power spectrum are typical, whereas the atypical networks have abnormal spatial and widely distributed power spectra. *3.1. Patient Summaries of ASD Patients*  **Figure 2.** Example of right-sided language, frontoparietal, and aberrant networks in ASD Patient 1. Image shown is the sagittal, axial, and coronal T1-weighted MRI of Patient 1 with overlayed language (blue), frontoparietal (green), and aberrant (red) networks. Plotted are the respective network blood oxygen dependent signal (BOLD) power versus frequency (Hz/100), wherein typical is less than 6 Hz/100. The frontoparietal and language network spatial distribution and the BOLD power spectrum are typical, whereas the atypical networks have abnormal spatial and widely distributed power spectra.

the right FP, bilateral temporal, and opercular regions.

found over the left FP-temporal regions.

A 7-year-old (yo) male progressively lost social interactions and eye contact from 9 to 15 months of age. At the time of rs-fMRI, he was diagnosed with ASD, cognitive delay,

A 4 yo female suddenly lost normal speech, social, language, cognitive, and motor abilities at 3.5 yo. At the time of rs-fMRI, she was diagnosed with ASD, borderline intellectual disability (ID), severe LI, and delayed visual-motor skills. Despite these deficits, FP, language, and motor RSNs were normal. Aberrant networks were found over

A 16 yo female lost social skills and ceased making eye contact with others at 3 yo. At the time of rs-fMRI, she was diagnosed with ASD, ID, severe LI, and DCD. Despite these deficits, FP, language, and motor RSNs were detected. Aberrant networks were

Aberrant networks were found over sensory, FP, and contralateral non-dominant

3.1.1. Patient 1

3.1.2. Patient 2

3.1.3. Patient 3

language regions (Figure 2).

#### 3.1.4. Patient 4

A 12 yo male lost language skills (from saying short sentences to saying only single words) and motor skills (ceased walking and performing fine motor skills) at 2.5 yo. He regained some language and motor skills at 5 yo and began to walk again, but he did not regain fine motor skills. At the time of rs-fMRI, he was diagnosed with ASD, severe ID, moderate LI, DCD, and epilepsy. Despite these deficits, FP, motor and bilaterallydominant language RSNs were intact. Bilateral language dominance is a relatively mild atypical feature found in children with epilepsy and dominate-sided language region network pathology [22]. Aberrant networks were found over the FP and language regions. The network pathology is consistent with children with drug-resistant epilepsy [8,15] which, in comparison to other aberrant networks discussed, has markedly erratic, highfrequency BOLD time courses and generalized spatial distribution of the cingulate, lateral temporal, and bilateral frontal regions.

#### 3.1.5. Patient 5

A 12 yo male lost language (from knowing 10 words to non-verbal) and motor skills (normal development to stereotypic movements) at 2 yo. At the time of the rs-fMRI, he was diagnosed with ASD, global developmental delay, severe LI, and DCD. Despite these deficits, FP, motor, and bilateral language RSNs were intact. Aberrant networks were localized to the FP, insular, and mesial temporal regions.

#### *3.2. Patient Summaries of TLE Patients (Controls)*

#### 3.2.1. Patient 6

A 15 yo female was diagnosed with intractable localization-related symptomatic epilepsy without status epilepticus after unprovoked complex partial seizures at 10 yo. The seizure focus was associated with an inferior temporal lobe encephalocele. Medical and developmental history was otherwise normal. The expected whole-brain network profiles were well-detected, including the motor, language, and FP networks. The SOZ was detected in the right temporal and right frontal regions. Non-SOZ aberrant networks were detected in the hand-arm motor areas.

#### 3.2.2. Patient 7

An 11 yo female was diagnosed with intractable epilepsy without status epilepticus after complex partial seizures at 10 yo. The seizure focus was associated with a left temporal lobe tumor. Early medical history was notable for appendectomy. Early developmental language milestones were delayed. Expected whole-brain network profiles were welldetected, including the motor, language, and FP networks. The SOZ was detected in the anterior half of the anterior temporal lobe. Non-SOZ aberrant network features were detected in the facial motor and language areas.

#### 3.2.3. Patient 8

A 4 yo male had a history of complex febrile seizures starting at 20 months of age. The patient presented with typical motor and cognitive developmental milestones but had a diagnosis of expressive language disorder with poor articulation. Overall, the expected whole-brain network profiles were well-detected, including the motor, language, and FP networks. The SOZ was detected in the left mesial-anterior temporal regions (language). No non-SOZ aberrant networks were detected.

#### 3.2.4. Patient 9

A 9 yo male was diagnosed with intractable absence epilepsy without status epilepticus at 7 yo. Medical and developmental history was normal. The expected whole-brain network profiles were well-detected, including the motor, language, and FP networks. The SOZ was detected in the left mesial-temporal region. No non-SOZ aberrant networks were detected.

#### 3.2.5. Patient 10

A 13 yo male was diagnosed with intractable localization-related idiopathic epilepsy without status epilepticus at 12 yo. Medical and developmental history was normal. The expected whole-brain network profiles were well-detected, including the motor, language, and FP networks. The SOZ was detected in the right and left mesial-temporal regions. No non-SOZ aberrant networks were detected.

#### *3.3. Statistical Analysis*

Table 1 presents the clinical characteristics and rs-fMRI ICA-based network characteristics and the corresponding Fisher's exact tests' statistics.


**Table 1.** Clinical and resting-state network characteristics. Fisher's exact test (two-tailed) was used to test the difference in frequencies. Significant *p* values (*p* < 0.017) are indicated in bold and italics.

#### 3.3.1. Clinical Characteristics

Children with ASD were between 4 and 16 yo (*M =* 10.66, *SD =* 4.71) and predominantly male (60%). Similarly, children with TLE were between 4 and 15 yo (*M =* 10.98, *SD =* 4.57) and predominantly male (60%). There were no significant differences in age (*t*(8) = 3.4, *p = 0*.916) or sex (Fisher's exact test, *p =* 1.0) between the children with ASD and TLE.

All children with ASD presented with clinically significant motor (100%), language (100%), and cognitive (100%) dysfunction, while children with TLE only presented with language dysfunction (40%) in addition to the epilepsy-related symptoms. Fisher's exact test (two-tailed) indicated a significant relationship between motor and cognitive dysfunction and ASD diagnosis (Table 1).

#### 3.3.2. rs-fMRI Networks

ICA of rs-fMRI data yielded 91, 119, 49, 48, 58, 66, 105, 96, 95, and 111 ICs for patients 1–10, respectively. There was no significant difference (*t* (8) = 2.3, *p =* 0.212) in the mean number of ICs generated from subject-level ICA of rs-fMRI for the children with ASD (*M =* 73, *SD =* 31.09) and children with TLE (*M =* 94.6, *SD =* 17.3).

There was no significant association between typical networks and diagnosis, as all children with ASD and TLE presented with intact motor, language, and FP RSNs (Table 1). Children with ASD presented with non-SOZ aberrant motor (40%), language (80%), and FP (100%) networks, while children with TLE only presented with non-SOZ aberrant motor (40%) and language (20%) networks in addition to the identified SOZs. Thus, children with ASD were significantly more likely to manifest aberrant FP networks (Table 1).

#### 3.3.3. Aberrant rs-fMRI Networks and Clinical Symptomatology

The relationship between network characteristics and clinical dysfunction was assessed using the full study cohort (*n* = 10). There was a significant relationship between atypical (non-SOZ) FP networks and cognitive dysfunction (Table 2). Specifically, clinically significant cognitive dysfunction was documented in 100% of the children with detected atypical FP networks, while cognitive dysfunction was not reported (0%) in children without atypical (non-SOZ) FP networks.

**Table 2.** Relationship between clinical symptomatology and presence of aberrant resting-state network. Fisher's exact test (two-tailed) was used to test the difference in frequencies. Significant *p* values (*p* < 0.017) are indicated in bold and italics.


#### **4. Discussion**

For the first time, we demonstrate that children with regressive-type ASD have intact motor, language, and FP neural networks with relatively typical spatial and temporal features, despite having moderate to severe disability in the skills typically subserved by these networks. The finding of overall preserved connectivity is consistent with prior ASD research [4,23]. Interestingly, the case sample only included children with regressive-type ASD, suggesting that these intact typical cognitive networks did subserve their normal expression prior to the regression. Furthermore, the rs-fMRI ICA data-driven approach extracts typical and atypical neural circuitry on an individual basis, demonstrating that multiple widespread aberrant neural networks characterize regressive-type ASD. Given that typical RSNs are intact but not well-expressed, we think that the atypical aberrant networks disrupt the fidelity of signaling within these long-range typical RSNs, essentially creating a locked-in network syndrome.

The ASD participants were found to have typical RSNs and additional broad aberrant networks. In comparison, the TLE controls had fewer aberrant networks beyond those localized to regions disrupted by seizure activity (SOZ). The aberrant networks found in ASD patients were not orthogonal to any canonical network, and this finding, in addition to their spatial location, may provide hints to the pathophysiology of ASD symptoms. In these patients, aberrant networks traversing regions associated with the FP network could interfere with the fidelity of signals of typical RSNs as they are communicated between distal regions. Higher-order networks typically integrate into other brain networks at around 18–24 months [24,25], which is around the same time children with ASD begin to show core symptoms and regression may occur [26]. According to our theory, interference from aberrant networks could impair the acquisition of cognitive inhibitory control [27], social skills, and other complex behavior. We think that interference from aberrant networks hinders communication and the optimal functioning of long-range typical RSNs, essentially creating a locked-in network effect. Alternatively, or complementarily, aberrant networks could inappropriately activate cortical areas. For example, aberrant networks traversing the somatosensory area, as in Patient 1, may be a biomarker of the expressed sensory symptoms.

Interventions that inhibit atypical networks could effectively "unlock" the intact RSNs, leading to symptomatic recovery. Recovery from locked-in syndrome, with compromised capacity to demonstrate consciousness yet intact supratentorial network function on rsfMRI, was reported [28]. The patients presented in this study were all refractory to standard treatments, such as behavioral and speech therapy. Potentially, other network-targeted

treatments, including those used in epilepsy, such as surgical and neuromodulatory treatments, could improve recovery rates of these patients [9,28,29].

This study has several limitations. The small sample size limits the generalizability of our results. It is also possible that the aberrant networks seen on imaging are an epiphenomenon and do not affect ASD symptoms. However, a recent large study found differences in the FP network regions between individuals with ASD and typically developing controls [4], thus supporting the notion that these aberrant networks interfere with typical RSNs. It is also possible that the aberrant networks reflect activity that does not interact with RSNs in an awake brain as our patients were sedated during scanning. However, such findings are not reported in neurotypical individuals studied under lowdose conscious sedation [30–32]. Further studies will be needed to correlate the presence or absence of aberrant networks and their characteristics, such as location, with detailed measures of cognitive and language function, as well as ASD symptomology, in larger cohorts. Unfortunately, with the current sample size, such analysis would not be valid. Thus, we look forward to larger studies in the future.

#### **5. Conclusions**

From these data, we propose that analyzing individual patients may provide evidence that ASD symptoms may correlate with aberrant networks that interfere with the maturation of typical RSNs, effectively creating a locked-in network syndrome. Further, we propose that it may be clinically useful to perform rs-fMRI on select patients with ASD. We believe this case series warrants larger systematic studies of ASD patients with rs-fMRI before and after typical treatment. rs-fMRI could help personalize treatment strategies by categorizing a patient's aberrant networks based on treatment response and symptom profile. In this way, rs-fMRI can act as an integrative tool to support a personalized precision medicine approach to ASD diagnosis and treatment.

#### **6. Patents**

Nothing to report.

**Author Contributions:** Conceptualization, A.R.P., B.S., P.J.M., R.B., R.E.F., and V.L.B.; methodology, V.L.B.; formal analysis, B.S., S.N.W., R.E.F., and V.L.B.; data curation, A.R.P., B.S., P.J.M., R.E.F., and V.L.B.; writing—original draft preparation, A.R.P., B.S., P.J.M., R.E.F., and V.L.B.; writing—review and editing, A.R.P., B.S., P.J.M., S.N.W., R.E.F., and V.L.B.; visualization, A.R.P., B.S., P.J.M., R.E.F., and V.L.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board of Phoenix Children's Hospital, Phoenix, AZ, USA.

**Informed Consent Statement:** Informed consent was obtained from the parents of all participants involved in the study.

**Data Availability Statement:** Data are available upon request.

**Acknowledgments:** We would like to thank the patients and families who volunteered their time to be part of our research program.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

