Participants
We collected brain MRI scans, clinical and genetic data from 118 individuals with ASD (mean 13.1 years, SD 4.6; male 113, 95.8%), and 122 TDC (mean 21.0, SD 9.7; male 75, 61.5%). Individuals with ASD were recruited from National Taiwan University Hospital, Taipei, Taiwan. They were clinically diagnosed as ASD by senior board-certified child psychiatrists based on the Diagnostic and Statistical Manual of Mental Disorders-5th edition (DSM-5) diagnostic criteria of autism spectrum disorder; the diagnosis was further confirmed by a structural interview using the Chinese version of the ADI-R [45, 46]. Participants with major neuropsychiatric disorders or with a full-scale IQ lower than 70 were excluded. The TDC participants were recruited from schools in the same districts of the ASD participants. All the participants went on clinical evaluation and interviews; all their parents were also interviewed by using the Chinese version of the Kiddie epidemiologic version of the Schedule for Affective Disorders (K-SADS-E) interview [47, 48] to exclude any current or lifetime ASD and other major psychiatric disorders including attention-deficit/hyperactivity disorder, schizophrenia, mood disorders, anxiety disorders, or neurodevelopmental disorders. The details of the psychometric properties of the K-SADS-E and interview training have been described elsewhere [47-50].
Procedure
The Research Ethics Committee approved the study before its implementation (Approval number: 201201006RIB; ClinicalTrials.gov number, NCT01582256). After the purposes and procedures of the study were fully explained and confidentiality was assured, written informed consent was obtained from the participants and their parents. All the participants were then assessed with the brain MRI, and the Wechsler Intelligence Scale for Children (version III) or the Wechsler Adult Intelligence Scale (version IV) for IQ profile according to their ages. The parents (mainly mothers) completed the following clinical measures about the participants.
Measures
The ADI-R [46] is a standardized, comprehensive, semi-structured, investigator-based interview for the caregivers of children with a mental age of 18 months into adulthood. It covers most developmental and behavioral aspects of ASD, including qualitative abnormalities in reciprocal social interaction, communication, and restricted, repetitive and stereotyped patterns of behaviors. The Chinese version of the ADI-R was approved by the Western Psychological Services in 2007 and has been widely used in several studies to validate the clinical diagnosis of ASD (e.g., [51-53]). The details of psychometric studies and interviewers training of using the ADI-R in Chinese have been described elsewhere [54, 55].
The Social Responsiveness Scale (SRS) [56] is a widely-used quantitative measure of autistic traits in the general population. It includes 65 items to measure the severity of ASD symptoms in natural social settings over the past six months for children and adolescents aged 4–18 years. Items were rated by parents or caregivers on a 4-point Likert scale from “0” (not true) to “3” (almost always true). The SRS has been demonstrated to have good internal consistency, construct validity, inter-rater reliability, test-retest reliability, and discriminative validity in prior research [56]. Its Chinese version demonstrates a four-factor structure (i.e., social communication, stereotyped behaviors/interest, social awareness, and social emotion), but is better conceptualized as a one-factor model [54]. The Chinese SRS has been widely used in ASD research in Taiwan (e.g., [48, 57-59]. High internal consistency was found for the four subscales (Cronbach’s alpha, .94–.95) and the total scale (Cronbach’s alpha, .95). To test the association between cingulate structures and social cognition, we specifically targeted the social awareness deficits subscale in this study.
Genotyping
Single nucleotide polymorphism (SNP) selection and genotyping.
Genomic DNA was prepared from peripheral blood using the Puregene DNA purification system (Gentra Systems Inc. Minneapolis, MI) according to the manufacturer's instructions. Five SNPs of the CNTNAP2, located in the intron 2 (rs779475) [37, 39], intron 13 (rs759178, rs2710102, rs2538991) [28, 40, 41], and intron 15 (rs2710126) [40], were selected for genotyping based on previous imaging genetic studies.
The primers of each SNP were designed by the platform of National Center for Genome Medicine (http://ncgm.sinica.edu.tw/), using GenePipe (http://genepipe.ncgm.sinica.edu.tw/seqtool/pages/getSeq.jsp) to retrieve SNP flanking sequences. All SNP genotyping was performed by SEQUENOM MassARRAY® System using the method of matrix-assisted laser desorption/ionization-time of flight mass spectrometry. The genotyping technology platform “iPLEX ® Gold reaction” provides high throughput, high accuracy, and low cost SNP analysis (http://ncgm.sinica.edu.tw/ncgm_02/snp_platform_e.html). The success rate of genotyping of the five selected SNPs were 99–100%. Genotype frequency is summarized in Supplementary Table S1.
MRI Data Acquisition
Brain images were acquired on a 3T MRI system (Trio, Siemens, Erlangen, Germany). Head movement was restricted with expandable foam cushions and was assessed immediately after image acquisition. High-resolution T1-weighted MR images were acquired covering the whole head with a three-dimensional (3D) magnetization-prepared rapid gradient echo (MPRAGE) sequence, resulting in an isotropic spatial resolution of 1 mm3.
Whole brain segmentation and cortical thickness calculation.
FreeSurfer V5.2.0 (https://surfer.nmr.mgh.harvard.edu/) was used on a 64-bit Linux operating system to reconstruct the cortical surface from the MPRAGE images [60]. Whole brain segmentation [61, 62] was performed. The cortical parcellation units of the cortex were automatically identified and labeled according to the Desikan atlas [63] within the FreeSurfer automatic cortical parcellation routine. The cortical thickness was automatically calculated by computing the shortest distance between the WM boundary and the pial surface at each vertex [64]. The reliability of the cortical thickness calculated by FreeSurfer has been validated [65]. The automatic reconstruction and calculation were reprocessed after manually correcting the detected erroneous part. The automatic parcellation of cortical regions derived 74 brain regions in each hemisphere, and the thickness of each region was then calculated. This study focused on the cingulate structure, in which GM was divided into five subregions, i.e., ACC, anterior and posterior parts of MCC, and dorsal and ventral parts of PCC, while WM was divided into four subregions, i.e., the rostral ACC, caudal ACC, , PCC, and isthmus (Figure 1).
Statistical Analysis
We used SAS v. 9.4 (SAS Institute Inc, Cary NC, USA) to perform the statistical analyses. Age and IQ profiles were compared by analysis of variance, while SRS subscores were compared between the ASD and TDC groups by the generalized linear model controlling for sex and age. In an age- and sex-compatible subsample (88 ASD & 51 TDC), we compared the volume of the GM and WM, and cortical thickness of cingulate structures for each subregion between ASD and TDC, controlling for sex, age, full-scale IQ, and handedness. The relationships between cingulate structures and social awareness deficits were examined by Pearson’s correlation analyses in the ASD group, partial out the effects of age, sex, and full-scale IQ. The genetic effects of the CNTNAP2 variants on the cingulate structures were firstly examined by the main effect of each SNP on cingulate structures (i.e., GM and WM volumes, and cortical thickness of each subregion). Then, we tested the interactions between the CNTNAP2 variants and age or diagnosis on each cingulate structure, controlling for age, sex, and full-scale IQ, as well as the main effects of each SNP and diagnosis. False discovery rate (FDR) was applied to correct for multiple comparisons. FDR q-value < 0.05 was set as statistical significance. As for sensitivity power analysis, with a total sample size of 240, the statistical tests had a power 0.8 to detect a difference with the effect size of 0.27.