Altered white‐matter integrity in unaffected siblings of probands with autism spectrum disorders

Abstract

Despite the evidence of altered white‐matter tract property in individuals with autism spectrum disorder (ASD), little is known about their unaffected siblings. This study aimed to investigate white‐matter integrity in unaffected siblings of ASD probands. Thirty‐nine unaffected siblings (mean age 15.6 ± 6.0 years; 27 males, 69.2%) and 39 typically developing controls (TDC) (14.2 ± 5.6 years; 26 males, 66.7%) were assessed with diffusion spectrum images and neuropsychological tests. Using the tract‐based automatic analysis and the threshold‐free cluster weighted (TFCW) scores, we searched for the segments among 76 tracts with the largest difference over the entire brain compared to TDC. Tract integrity was quantified by calculating the mean generalized fractional anisotropy (mGFA) values of the segments with the largest difference in TFCW scores. Unaffected siblings showed reduced mGFA in the bilateral frontal aslant tracts, the right superior longitudinal fasciculus 2 (SLF2), the frontostriatal tracts from the right dorsolateral and left ventrolateral prefrontal cortices, the thalamic radiations of the left ventral and the right dorsal thalamus, the callosal fibers of the splenium, and the increased mGFA of the callosal fibers of the precuneus and the left inferior longitudinal fasciculus. Among these, reduced right SLF2 mGFA was associated with social awareness deficits; impaired frontostriatal tract was associated with internalizing problems, while right frontal aslant tract integrity was associated with visual memory deficits. In conclusion, unaffected siblings showed the aberrant integrity of several white‐matter tracts, which were correlated with clinical symptoms and neurocognitive dysfunction. The altered tract integrity could be further examined in the probands with ASD. Hum Brain Mapp 38:6053–6067, 2017. © 2017 Wiley Periodicals, Inc.

INTRODUCTION

Autism spectrum disorder (ASD), a neurodevelopmental disorder with a strong genetic component, is characterized by striking sociocommunication deficits and restricted/stereotyped behaviors [American Psychiatric Association, 2013]. Phenotypic heterogeneity and genetic variability are among the most scientifically challenging features of ASD. It has been postulated that the expression of ASD might be rooted in a circumscribed set of neuroanatomical structures [Pelphrey et al., 2004], which may serve as critical endophenotypes to facilitate the research of the underlying mechanisms of this disorder [Kaiser et al., 2010].

Recent studies have suggested the existence of an endophenotype of ASD identified through the data of unaffected siblings [Spencer et al., 2012a, 2012b]. Endophenotypes are defined as heritable biomarkers associated with a disease entity that individuals may have, regardless of whether they have the disease or not [Gottesman and Gould, 2003]. Extant literature suggests that brain structural abnormality is a heritable biomarker for ASD [Brun et al., 2009; Jansen et al., 2015]. To dissect the neural substrates of autistic brains, some studies have applied diffusion tensor imaging (DTI) to measure white‐matter (WM) microstructure property and have shown that variability in WM integrity has a strong genetic basis [Chiang et al., 2011; Kochunov et al., 2011] and may serve as an endophenotype for ASD [Clemm von Hohenberg et al., 2013].

Several studies have examined the differences in WM microstructure property between ASD and typically developing controls (TDC) [Alexander et al., 2007; Barnea‐Goraly et al., 2010; Billeci et al., 2012; Catani et al., 2008; Groen et al., 2011; Jou et al., 2011; Ke et al., 2009; Mengotti et al., 2011]. WM disconnectivity is considered one of the suggested neuropathologies in ASD [Belmonte et al., 2004a, 2004b; Geschwind and Levitt, 2007]. Compared to TDC, younger children with ASD showed increased integrity of the WM microstructure in the corpus callosum, cingulum, arcuate fasciculus, and external and internal capsules [Billeci et al., 2012; Walker et al., 2012]. In contrast, older children and adolescents with ASD display decreased integrity in the superior longitudinal fasciculi (SLF), internal and external capsule, uncinate fasciculus, cingulum, corona radiate, and thalamic radiation [Barnea‐Goraly et al., 2010; Groen et al., 2011; Jou et al., 2011]. These findings indicate that the WM microstructure may be altered in ASD.

The siblings of probands with ASD, similar to the probands, suffered from a volumetric reduction in the posterior vermis [Mitchell et al., 2009] and amygdala [Dalton et al., 2007] when compared to TDC. However, most of the imaging studies in the siblings did not exclude those with a broader autism phenotype who exhibited an attenuated form of social and communication difficulties [Gronborg et al., 2013; Sucksmith et al., 2011; Wheelwright et al., 2010] and restricted interests and stereotyped behaviors [Hurley et al., 2007], but did not fulfill the diagnostic criteria of ASD, resulting in a biased estimation. Only two recent DTI studies excluded siblings with a broad autism phenotype and demonstrated that both the probands with autism and their unaffected siblings demonstrated similar differences in WM microstructure in comparison with TDC [Barnea‐Goraly et al., 2010; Jou et al., 2016]. Barnea‐Goraly et al. [2010] found a pattern of widespread, reduced WM fractional anisotropy (FA, an indicator of fiber tract integrity) in the frontal, parietal, and temporal lobes, including regions that are important for social cognition, whereas Jou et al. [2016] found a bilateral reduction in FA with association, commissure, and projection fibers in probands with autism, and a similar, albeit less severe, pattern with fewer affected tracts in the unaffected siblings, suggesting the presence of a neuroimaging endophenotype for autism. Notably, their sample sizes were relatively small (autism probands, N = 13 and 19; unaffected siblings, N = 13 and 20; TDC, N = 11). Besides, most previous studies were based on a hypothesis‐driven design, which may largely depend on the validity of prior knowledge and may miss unknown yet important mechanisms. Moreover, most studies used the mean FA to represent the integrity of the whole tract, despite the fact that the microstructure of one tract from its origin to its destination may not be constant along the whole tract. In this study, we adopted a hypothesis‐free analysis to investigate the alterations of 76 WM tract bundles over the whole cerebrum. A stepwise analysis was applied to each tract bundle by using a threshold‐free clustering weighted method, as described in the method section. In this way, we were able to identify segments along the tract pathway with the most remarkable difference between the unaffected siblings and TDC.

The clinical correlates of the identified structural alterations are important yet less well‐addressed. The volume changes of the dorsolateral prefrontal cortex (DLPFC), amygdala, and posterior vermis have been shown to associate with the severity of autistic symptoms in probands with autism, measured by the Autism Diagnostic Observation Scale [Mitchell et al., 2009]. In contrast, the phenotype expression of WM changes in unaffected siblings has not been clear until now. Two studies with small sample sizes have revealed no correlations between autism symptomatology and WM FA or axial diffusivity [Barnea‐Goraly et al., 2010; Jou et al., 2016]; however, the findings may be subject to type II errors [Barnea‐Goraly et al., 2010]. Other than autistic symptoms, neurocognitive deficits have been reported in individuals with ASD, particularly in executive function [Rommelse et al., 2011; Tager‐Flusberg and Joseph, 2003], visual memory [Chien et al., 2015], and spatial working memory (SWM) [Kercood et al., 2014; Vogan et al., 2014]. Their unaffected siblings are also likely to share the impairment or show compensatory functioning [Hughes et al., 1999; Noland et al., 2010; Wong et al., 2006], but the underlying mechanism is poorly understood. Overall, working memory processes are largely subserved by the prefrontal cortex and the parietal cortex [Baddeley, 2003; Carlson et al., 1998; Fuster, 2000; Greene et al., 2008; Kwon et al., 2002; Owen et al., 2005], while visual memory involves a broader system of the prefrontal, premotor, dorsal cingulate, and posterior parietal activation [Owen et al., 2005]. These regions were implicated in ASD with atypical neural activity [Koshino et al., 2005; Luna et al., 2002; Silk et al., 2006]. Using a larger sample, this study also aimed to investigate the phenotype expression of the altered structural connectivity in terms of autistic traits, behavioral symptoms, and neurocognitive functioning focused on visual memory and SWM.

Based on previous findings [Barnea‐Goraly et al., 2010; Jou et al., 2016], we hypothesized that the unaffected siblings of children with ASD would have an aberrant WM structure. Moreover, if these WM structural aberrations were endophenotypes for ASD, the aberrations could be correlated with phenotype expression in the unaffected siblings regarding clinical manifestations and neurocognitive functioning, which are relevant to ASD. In this study, we used Diffusion Spectrum Imaging (DSI) and an automatic method for tract‐specific analysis, called tract‐based automatic analysis (TBAA) [Chen et al., 2015], to measure the WM integrity of the tracts over the entire brain. The DSI data have the advantage of attaining better registration with the DSI template provided by TBAA, leading to a more accurate sampling of WM integrity along fiber tracts than DTI data.

MATERIALS AND METHODS

Participants and Procedures

We recruited 41 unaffected siblings and 41 age‐ and sex‐matched TDC in this study. The datasets which failed the quality control (1 sibling and 2 TDC) were excluded from our analyses. Therefore, the sample comprised 40 unaffected siblings of probands with ASD (mean age 15.6 ± 6.0 years; ranged from 8 to 33 years; male N = 27, 69.2%) and 39 TDC (mean age 14.2 ± 5.6 years; ranged from 8 to 31 years; male N = 26, 66.7%). The siblings were full biological siblings of 40 probands with ASD from our ASD cohort. All probands had a diagnosis of either autistic disorder or Asperger's disorder based on the Diagnostic and Statistical Manual of Mental Disorders, fourth edition (DSM‐IV) [American Psychiatric Association, 1994]; all of them also met the DSM‐5 diagnostic criteria for ASD [American Psychiatric Association, 2013], confirmed by the corresponding author (SSG). To ensure that the siblings were unaffected by ASD, all siblings underwent a clinical diagnostic interview by the corresponding author (SSG), and their parents received the Autism Diagnostic Interview‐Revised (ADI‐R) to confirm the absence of a diagnosis of autistic disorder or Asperger's disorder. Later on, their parents completed the Social Responsiveness Scale (SRS) and Social Communication Questionnaire (SCQ); one sibling was excluded from the analysis because his total score of SRS was higher than the threshold. All the siblings (N = 39) remained in the study scored below the cutoff threshold of the ADI‐R and the total scores of SRS (male, 70; female, 65) [Constantino and Gruber, 2005; Constantino and Todd, 2003] and SCQ (score of 15). Both siblings and TDC recruited from the community or school undertook an interview using the Kiddie‐Schedule for Affective Disorder and Schizophrenia‐Epidemiological version [Gau and Soong, 1999] to confirm that they were free from any psychiatric disorder. Any TDC with a psychiatric diagnosis or with SRS or SCQ total scores above the cutoff threshold were excluded from the study. A detailed family history was collected for TDC to ensure that neither their first‐ nor second‐degree relatives had a diagnosis of autistic disorder or Asperger's disorder and that they were siblings of unaffected children.

Autistic traits were quantified with the SRS in both unaffected siblings and TDC. Behavior symptoms were assessed by the Child Behavior Checklist (CBCL). All participants completed two Cambridge Neurological Test Automated Battery (CANTABeclipse, Cambridge Cognition Ltd, Cambridge, UK) tasks for visual memory (Delayed Matching to Sample, DMS) and SWM (Spatial Working Memory, SWM), and then undertook diffusion spectrum imaging (DSI) assessment.

The study procedure is in compliance with the Code of Ethics of the World Medical Association (Declaration of Helsinki). The Research Ethics Committee approved this study before implementation (approval number, 201201006RIB; http://www.ClinicalTrials.gov number, NCT01582256). After the purposes and procedures of the study were fully explained and confidentiality was ensured, written informed consent and child assent were obtained from the participants and their parents.

Assessments

Autism Diagnostic Interview‐Revised

The ADI‐R is a standardized, comprehensive, semi‐structured, investigator‐based interview of the caregivers [Lord et al., 1994]. It covers most of the developmental and behavioral aspects of ASD, including reciprocal social interaction, communication, and repetitive behaviors and stereotyped patterns, for children with a mental age from about 18 months into adulthood. The Chinese version of the ADI‐R was approved by the World Psychological Association in 2007 [Gau et al., 2011] and has been extensively used in ASD research in Taiwan [Chien et al., 2016]. All the interviewers reached an agreement of over 90%, ranging from 98.25 ± 1.91 to 99.38 ± 1.06, against the rating of each item in the Chinese ADI‐R by a qualified ADI‐R cross‐site trainer (YY Wu) and in‐site trainer (SS Gau) before the implementation of this study.

Kiddie epidemiologic version of the schedule for affective disorders and schizophrenia (K‐SADS‐E)

K‐SADS‐E [Puig‐Antich and Chambers, 1978] is a standard structured tool to assess current and previous mental disorders for youths from 6 to 18 years old, while its original form (SADS) was developed for adults aged 18 and older. The Chinese version of K‐SADS‐E has been widely applied in several studies to screen for current and lifetime mental disorders in clinical, neuropsychological, and neuroimaging research [Gau et al., 2015; Lin and Gau, 2017; Shang et al., 2016]. In this study, we used the Chinese K‐SADS‐E for every participant to ensure that none of them had any current or previous mental disorders, including schizophrenic disorders, affective disorders, anxiety disorders, and childhood‐onset mental disorders such as ASD, attention‐deficit/hyperactivity disorder, oppositional disorder, and so on.

Social Responsiveness Scale

The SRS is a self‐ or caregiver‐report four‐point Likert‐type questionnaire for quantifying autistic traits based on the frequency of each behavior (“0” never true and “3” always true) [Constantino et al., 2003]. It consists of 65 items, covering the descriptions of social awareness, social cognition, social communication, social motivation, and autistic mannerisms. It has good reliability and concurrent validity relative to the ADI‐R [Bolte et al., 2008; Constantino et al., 2004]. The psychometric properties of its Chinese version have been established, with a validated four‐factor structure, that is, social awareness, social communication, social emotion, and stereotyped behaviors [Gau et al., 2013], and have been widely used in autism‐related research in Taiwan [Yin et al., 2016]. We summed up the scores of items within the same factor [Gau et al., 2013], and correlated the four subscores with neural tract integrity.

Child Behavior Checklist

The CBCL [Achenbach, 1991] is a parent‐report questionnaire to screen for broad spectrum behavior symptoms in youths aged 4–18, including 8 constructs, i.e., withdrawn, somatic complaints, anxious/depressed, social problems, thought problems, attention problems, delinquent behavior, and aggressive behavior. Items were rated on a 3‐point scale from 0 (not true) to 2 (very true or often true). The subscores were calculated by totaling the item responses of each construct described in the original manual [Achenbach, 1991]. The Chinese CBCL has been widely used to measure behavioral syndromes in Taiwanese children [Shang et al., 2006] and adolescent [Yang et al., 2001] populations.

Social Communication Questionnaire

The SCQ [Rutter et al., 2003] is a parent‐report questionnaire to screen for autistic symptoms in individuals above 4 years old. It contains 40 yes or no items that correspond to the three core symptoms (impairment in social development, communication, and stereotyped/repetitive behaviors), with good convergent validity to the ADI‐R. The Chinese version of the SCQ has satisfactory reliability and validity [Gau et al., 2011] and has been widely used [Lau et al., 2013]. We added up all items and used the cutoff total score of 15 to exclude any possible participants with autistic traits.

Cambridge Neurological Test Automated Battery

The CANTAB is a set of computerized tests to examine nonverbal neuropsychological functions. Two subtests of the CANTAB were used to assess visual memory and SWM.

Delayed Matching to Sample

The DMS task is designed to evaluate the ability to remember the visual features of a complex and abstract pattern in a four‐choice delayed recognition memory paradigm [Egerhazi et al., 2007]. At the outset of each trial, a sample pattern appeared in the screen center for 4.5 s. At the simultaneous matching condition, the sample pattern remained onscreen when four choice patterns appeared. At the delayed condition, a delay of 0, 4, or 12 s was introduced between the appearance of the sample pattern and the choice patterns. The participant was instructed to touch the pattern that matched the sample. If the first choice was incorrect, the participant had to make other choices until a correct choice was made. After three practice trials, there were 20 counterbalanced test trials in a pseudorandom order, including five simultaneous trials and five trials for each of the three delay intervals, to ensure comparable encoding across all task conditions. We used the total number of correct responses to represent visual memory performance.

Spatial Working Memory

Each box on the screen contains only one token per trial. Participants were instructed to search through boxes to find sufficient blue tokens hidden inside. To complete trials in minimum steps, participants should remember the boxes which have already shown blue tokens. The SWM task begins with 4 practice trials (3‐box problem), followed by formal trials (4‐, 6‐, and 8‐box problems). The test results consist of 3 types of errors: “within error” occurs when participants search the same box more than once in the same trial; “between error” represents a mistake of searching boxes which have already shown blue tokens; and “double error” can be categorized as both a within error and a between error. We used total errors to represent spatial working memory deficits.

Image Acquisition

All images were acquired on a 3 T MRI system (Trio, Siemens, Erlangen, Germany) with a 32‐channel phased array head coil. T1‐weighted (T1W) imaging was performed using a 3D magnetization‐prepared rapid gradient echo (MPRAGE) sequence: repetition time (TR)/echo time (TE) = 2,000/3 ms, flip angle = 9°, field of view (FOV) = 256 × 192 × 208 mm³, acquisition matrix = 256 × 192 × 208. DSI was acquired with a single‐shot spin‐echo echo‐planar imaging sequence: TR/TE = 9600/130 ms, FOV = 200 mm, slice thickness = 2.5 mm, slice number = 54, and matrix size = 80 × 80. The diffusion acquisition scheme was composed of 102 diffusion‐encoding directions corresponding to grid points located in a half sphere of diffusion‐encoding space (q‐space) within a radius of 3 units, which corresponds to a b _max of 4000 s mm⁻² [Kuo et al., 2008].

All DSI datasets underwent a quality control procedure by measuring signal dropouts caused by head motions. All the acquired DSI datasets (54 slices × (101 directions DWI + 1 null image) = 5508 images) were scrutinized by calculating the signals in the central square (20 × 20 pixels) of each image. If the average signal intensity of an image (after correcting for its b value) was lower than 2 standard deviations from the mean of all images, the image was considered a signal loss. We previously found that the DSI dataset with more than 90 images of signal loss caused a significant reduction of WM integrity values with a percentage error >6% and should be discarded. After this procedure, the two groups showed no statistical difference in signal dropout counts (unaffected siblings: mean 40.92 ± standard deviation 21.56, TDC: 38.15 ± 20.34, P = 0.659) and signal‐to‐noise ratio (unaffected siblings: 23.61 ± 8.92, TDC: 26.27 ± 5.81, P = 0.164).

DSI Reconstruction

The diffusion probability density function (PDF) at each voxel was reconstructed based on the Fourier relationship between the PDF and q‐space signal [Callaghan et al., 1991]. Three‐dimensional Fourier transform was performed on the q‐space signal to reconstruct the PDF applied with a Hanning filter with a width of 17 units. The orientation distribution function (ODF, ψ(u)) was determined by computing the second moment of the PDF along each of the 362 radial directions (sixfold tessellated icosahedron). The generalized fractional anisotropy (GFA) value at each voxel was determined with the formula: SD(ψ)/RMS(ψ), where SD is the standard deviation and RMS is the root mean square of the ODF [Tuch, 2004].

Tract‐Specific Analysis

The TBAA method was employed to enable efficient tract‐specific analysis for multiple major fiber tracts over the entire brain [Chen et al., 2014, 2015]. The TBAA method relies on two important components: a DSI template and a comprehensive list of white matter tracts reconstructed on the template. The DSI template, called NTU‐DSI‐122, was a DSI dataset averaged over 122 co‐registered DSI datasets of healthy adults and was built in the standard ICBM152 space [Chen et al., 2015]. A total of 76 tracts were predefined on the template by performing streamline‐based deterministic tractography with multiple regions of interests (mROIs) using DSI Studio (http://dsi-studio.labsolver.org). The coordinates of streamlines were aligned along the proceeding direction of each tract bundle, interpolated into 100 steps, and saved as the sampling coordinates of GFA.

We employed the TBAA method with the following procedures. (a) All subjects were registered to create a study‐specific template (SST) [Chen et al., 2014, 2015]. (b) The SST was registered to the NTU‐DSI‐122 template [Hsu, 2013]. (c) Sampling coordinates of the 76 predefined tracts were transformed from the NTU‐DSI‐122 template to individual DSI via the transformation between the NTU‐DSI‐122 and SST, and the transformation between SST and individual DSI. (d) GFA values were sampled in native DSI space using the transformed sampling coordinates from the NTU‐DSI‐122. A two‐dimensional (2D) array of GFA values, 76 tracts by 100 steps, was created for each subject. The 2D arrays, named connectograms, were used for group analysis.

Threshold‐Free Cluster Weighted (TFCW) Scores for Group Analysis

TFCW scores were calculated following Smith's approach [Smith and Nichols, 2009] to estimate the weighted scores that take into account the effect size (ES) of the group difference and the length of the difference along a tract. The weighted scores were used to select tract segments based on the assumption that the group difference appeared on segments within an entire tract. First, we calculated the ES of each step between groups:

$${\mathrm{E}\mathrm{S}}_{\mathrm{p}}=\frac{|{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}_{\mathrm{A}}-{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}_{\mathrm{B}}|}{\sqrt{\left({\mathrm{S}{\mathrm{t}\mathrm{d}}_{\mathrm{A}}}^2+{{\mathrm{S}\mathrm{t}\mathrm{d}}_{\mathrm{B}}}^2\right)/2}}$$

Where the mean_A and mean_B were the mean values of group A and group B at the step p, and the Std_A and Std_B were the standard deviations of group A and group B at the step p, respectively.

The TFCW score for each step p was defined as

$$S_{\mathrm{p}}={\sum }_{e=1}^n{C}_{\mathrm{p}}\left(e\right)\times {\mathrm{E}\mathrm{S}}_{\mathrm{p}}$$

Threshold levels (e) for ES were set from 0 to 1 in 20 aliquots according to the original distribution of ES between these two groups, while the C _p(e) was the clustered step size at the threshold level e.

Higher scores of S _p indicated the more pronounced difference between groups with the weighting from continuous segment sizes. After calculating the values of S _p for all the steps, a histogram of S _p was plotted; the steps showing the scores in the top two percentile (the proportion of steps having original ES > 0.5) of the histogram were selected as the segments with noticeable group differences. Noncontinuous segments within one tract were considered one segment. The mean GFA (mGFA) values of the selected segments were calculated by averaging the GFA values from each segment.

Statistical Analysis

We used SAS 9.2 software (SAS Institute Inc., Cary, NC) to perform data analyses. Age and IQ subscores were compared by Student's t test between the unaffected siblings and TDC, while sex distribution and handedness were compared by Chi‐square test. In normality checking, all the mGFA were normally distributed, while most of the clinical variables were not (except for the total correct on the DMS task and social awareness deficits on SRS). Therefore, we adopted a nonparametric test (Wilcoxon sum rank test) for two‐group comparison regarding symptom severity and neurocognitive function (Table 1), and compared the mGFA using the general linear model (Table 2). In the regression analysis, we performed log transformation for the clinical variables and neurocognitive function (except for the above two normally distributed variables) to ensure that the data are normally distributed before model selection.

Table 1.

Demographics, autistic symptoms, behavior problems, and neuropsychological functions in unaffected siblings (SIB) and typically developing controls (TDC)

	SIB (n = 39)		TDC (n = 39)
	Mean/n	SD/%	Mean/n	SD/%	t/chi‐square	P
Age	15.6	6.0	14.2	5.6	t = −1.05	0.296
Male	27	69.2%	26	66.7%	x 2 = 0.059	0.808
Handedness	39	100%	38	97.4%	x 2 = 1.013	0.314
Full‐scale IQ	110.8	13.2	111.7	11.8	t = 0.31	0.759
Verbal IQ	112.1	13.6	110.4	11.9	t = −0.59	0.555
Performance IQ	107.6	13.1	112.1	12.8	t = 1.53	0.129
Social Responsiveness Scale a T scores (raw score range)	Mean	SD	Mean	SD	Z a	P
Social communication deficits	48.48	8.21	48.64	7.30	0.292	0.770
	(0–24)		(1–25)
Stereotyped behaviors	47.03	9.56	47.26	8.44	0.568	0.570
	(0–17)		(0–17)
Social awareness	41.97	8.15	47.06	8.36	2.481	0.013b
	(0–21)		(1–26)
Social emotions	47.09	9.34	47.45	9.61	0.146	0.884
	(0–12)		(0–15)
Total scores	45.27	8.72	47.27	7.92	1.297	0.195
	(4–60)		(4–69)
Child Behavior Checklist: T scores
Aggressive behaviors	50.84	9.12	47.85	8.52	−1.862	0.063
Anxious/depressed	53.94	14.33	46.98	6.58	−2.319	0.020b
Attentional problem	48.97	7.57	46.40	7.32	−1.502	0.133
Delinquent behavior	48.50	6.02	47.78	6.76	−0.788	0.431
Social problem	46.26	6.48	46.15	7.54	−0.337	0.736
Somatization	49.80	12.51	47.96	5.84	−0.024	0.981
Thought problem	48.84	7.59	47.61	5.73	−0.379	0.705
Withdrawn	50.01	8.03	45.85	6.19	−2.499	0.012b
Internalizing problem	51.68	11.69	46.43	5.49	−2.193	0.028b
Externalizing problem	50.02	7.94	47.71	7.63	−1.556	0.120
Delayed Matching to Sample
Total correct responsea	36.13	2.99	35.28	4.04	−0.769	0.4417
Spatial Working Memory
Total errorsa	18.54	15.61	16.18	14.48	−0.530	0.596

^aWilcoxon sum rank test with normal approximation and two‐sided P.

^bSignificant at P < 0.05.

Table 2.

Comparison of segment mGFA and whole‐tract mGFA between unaffected siblings (SIB) and typically developing controls (TDC)

	Segment mGFA							Whole‐tract mGFA
	SIB (N = 39)		TDC (N = 39)					SIB (N = 39)		TDC (N = 39)
	Mean	SD	Mean	SD	F	P	Cohen's d	Mean	SD	Mean	SD	F	P	Cohen's d
Frontal aslant tract (L)	0.312	0.022	0.325	0.028	6.14	0.016a	−0.55	0.291	0.012	0.296	0.010	7.03	0.010a	−0.50
Frontal aslant tract (R)	0.280	0.016	0.297	0.025	11.39	0.001a	−0.79	0.237	0.013	0.238	0.012	0.44	0.509	−0.11
Superior longitudinal fasciculus 2 (R)	0.241	0.036	0.265	0.046	7.74	0.007a	−0.58	0.245	0.012	0.248	0.015	2.61	0.111	−0.23
Inferior longitudinal fasciculus (L)	0.270	0.041	0.156	0.034	6.51	0.013a	0.57	0.255	0.025	0.357	0.014	1.72	0.194	0.35
Frontostriatal tract‐VLPFC (L)	0.243	0.019	0.260	0.026	11.88	0.001a	−0.74	0.223	0.014	0.230	0.017	5.85	0.018a	−0.41
Frontostriatal tract‐DLPFC (R)	0.312	0.021	0.325	0.021	9.44	0.003a	−0.60	0.295	0.015	0.301	0.012	6.62	0.012a	−0.47
Thalamic radiation: ventral (L)	0.312	0.024	0.321	0.019	5.33	0.024a	−0.44	0.278	0.015	0.283	0.015	5.99	0.017a	−0.38
Thalamic radiation: dorsal (R)	0.314	0.014	0.320	0.015	5.47	0.022a	−0.45	0.314	0.010	0.317	0.010	3.82	0.055	−0.26
Callosal fibers: precuneus	0.274	0.036	0.251	0.029	9.62	0.003a	0.71	0.270	0.031	0.253	0.026	7.01	0.010a	0.61
Callosal fibers: splenium	0.338	0.024	0.353	0.028	6.83	0.011a	−0.54	0.293	0.015	0.296	0.019	1.26	0.266	−0.17

^aSignificant level at P < 0.05.

Using the 10 segments to predict clinical severity, we put all 10 mGFA in the model to compute the proportion of variance (i.e., R‐square) which could explain each phenotype parameter (i.e., SRS subscores and total scores; CBCL internalizing problems; and the variables of DMS and SWM). Similarly, the 10 whole‐tract mGFA were also examined altogether to determine how much variance they could explain for each phenotype.

As whole‐tract mGFA explained more variance than segment mGFA, we used stepwise selection to identify the specific whole tracts with statistical significance among the 10 fiber tracts to predict each phenotype (i.e., SRS subscores and total scores, CBCL internalizing problems, and the variables of DMS and SWM). As the behaviors measured by CBCL were not specific to ASD, we only modeled the subscores that were different between siblings and TDC (i.e., anxious/depressed, withdrawal, internalizing problem). As the three subscores were highly correlated with one another, only the summary score, internalizing problem, underwent model selection. Sex and age were also included in the model for stepwise selection. Significance was set at P < 0.05.

RESULTS

Sample Characteristics

Table 1 presents the demographics of the unaffected siblings and TDC. The age, sex ratio, and handedness were not different between the two groups. The IQ profiles were not statistically different between siblings and TDC. The siblings were genuinely unaffected, and their SRS subscores and total scores did not deviate from TDC (Table 1). However, siblings exhibited a higher level of anxious/depressed, withdrawal behaviors, and internalizing problems compared to TDC (Table 1 and Supporting Information, Table I), although both groups did not have any anxiety‐related psychiatric disorders. The numbers of participants who were scored within clinical or borderline ranges were summarized in Supporting Information, Table I.

Threshold‐free cluster weighted

We selected 10 segments of fiber tracts using the TFCW method, as they demonstrated the highest level of differences between unaffected siblings and TDC (Fig. 1). These 10 segments included bilateral frontal aslant tracts (Fig. 2a), the right SLF2 (Fig. 2b), the frontal‐striatal tracts from the left ventrolateral prefrontal cortex (VLPFC) (Fig. 2c), the right dorsolateral prefrontal cortex (DLPFC) (Fig. 2d), the thalamic radiation from the left ventral thalamus and the right dorsal thalamus (Fig. 2c,d), the callosal fibers of the precuneus and the splenium (Fig. 2d), and the left inferior longitudinal fasciculus (Fig. 2e).

Figure 1.

The 10 segments selected by threshold‐free clustering weighted method. T1: the left frontal aslant tract; T2: the right frontal aslant tract; T3: the left inferior longitudinal fasciculus; T4: the right superior longitudinal fasciculus part 2; T5: the left frontostriatal tract from the left ventrolateral prefrontal cortex; T6: the right frontostriatal tract from the right dorsolateral prefrontal cortex; T7: the left thalamic radiation of the left ventral thalamus; T8: the right thalamic radiation of the right dorsal thalamus; T9: the callosal fibers of the precuneus; T10: the callosal fibers of the splenium. [Color figure can be viewed at http://wileyonlinelibrary.com]

Figure 2.

The neural tracts of the 10 selected segments: (a) the bilateral frontal aslant tracts; (b) the right SLF2; (c) the left frontostriatal tract from the ventrolateral prefrontal cortex, the left thalamic radiation of the left ventral thalamus, and the callosal fibers of the precuneus and the splenium; (d) the right frontostriatal tract from the dorsolateral prefrontal cortex, the right thalamic radiation of the right dorsal thalamus, and the callosal fibers of the precuneus and the splenium; (e) the left inferior longitudinal tract. [Color figure can be viewed at http://wileyonlinelibrary.com]

Group Comparison

Table 2 shows the group differences of the mGFA of 10 segments selected by TFCW and their whole tracts using the general linear model. For the 10 selected segments, unaffected siblings showed substantially lower mGFA, except for the callosal fibers of the precuneus and the left inferior longitudinal fasciculus, which showed higher mGFA. As for the whole‐tract mGFA of the 10 segments, siblings showed the same direction of changes in all tracts but only 5 of them were significantly different between two groups. The effect sizes generally decreased in whole‐tract mGFA (|Cohen's d| = 0.17∼0.61) than in segment mGFA (|Cohen's d|= 0.44∼0.79) (Table 2). If internalizing problems were controlled, 6 of the 10 segments were still substantially different between the two groups (Supporting Information, Table II).

Neural Correlates

To examine the relationships between the 10 segments/whole‐tract mGFA and both the clinical manifestations and neurocognitive function in the siblings, we first investigated how the 10 segments can explain the variance of each phenotype by checking the R ² of the model (Table 3). Altogether, the 10 segments explained 15.1–28.8% of the variance in SRS subscores, 35.4% of the variance in CBCL internalizing problems, and 31.3% and 36.2% of the variance in DMS total correct and SWM total errors, respectively. On the other hand, 10 whole‐tract mGFA could explain 12.5–40.2% of the variance in SRS subscores, 22.9% of the variance in CBCL internalizing problem, and 46.0% and 58.1% of the variance in DMS total correct and SWM total errors. Whole‐tract mGFA accounts for higher variances than segment mGFA; therefore, whole‐tract mGFA were checked for which mGFA can predict phenotypes in siblings by stepwise selection.

Table 3.

R square for 10 segments and 10 whole‐tract mGFA values to predict autistic symptoms and behavior problems in unaffected siblings

R square	Whole‐tract mGFA	Segment mGFA
Social Responsiveness Scale
Social communication deficits	0.239	0.254
Stereotyped behaviors	0.125	0.151
Social awarenessa	0.341	0.258
Social emotions	0.402	0.288
Total scores	0.240	0.167
Child Behavior Checklist
Internalizing problem	0.229	0.354
Delay Matching to Sample
Total correct responsea	0.581	0.313
Spatial Working Memory
Total errors	0.460	0.362

^aPhenotype raw scores were normally distributed, so they were directly analyzed without log transformation. Other variables were analyzed after log transformation to ensure the distribution was normal.

As for the SRS subscores of the siblings, the right SLF2 mGFA remained in the model to predict social awareness deficits in siblings (Fig. 3). The mGFA of the frontostriatal tract from the right DLPFC was associated with internalizing problems (Fig. 4). Regarding neurocognitive function, the mGFA of the right frontal aslant tract was positively associated with the numbers of total correct in the DMS task (Fig. 5).

Figure 3.

The relation between the mGFA of the right superior longitudinal fasciculus 2 and social awareness deficits on Social Responsiveness Scale in unaffected siblings (F = 8.38, P = 0.007).

Figure 4.

The relation between the mGFA of the right frontostriatal tract from the DLPFC and internalizing problems (log transformation to ensure normal distribution) in unaffected siblings (F = 7.10, P = 0.012).

Figure 5.

The relation between the mGFA of the right frontal aslant tract and neurocognitive functioning (delayed matching to sample: total correct response) in unaffected siblings (F = 8.46, P = 0.006).

DISCUSSION

To the best of our knowledge, this is the first study to apply the TFCW method to detect altered segments of white‐matter tracts assessed by DSI. This study is also one of the few to examine the property of white‐matter tracts in a large sample of unaffected siblings of probands with ASD. We found that 10 segments in the unaffected siblings were substantially different from those in TDC, including several tracts within or connecting the frontal lobe (the frontal aslant tracts, the frontostriatal tracts, and SLF2), thalamic radiation and the callosal fibers of the precuneus and the splenium. Except for the callosal fibers of the precuneus and the left inferior longitudinal fasciculus, which revealed an increase in mGFA, all the other eight segments showed reduced mGFA compared to TDC. The WM alterations in the unaffected siblings were associated with their clinical manifestations (i.e., social awareness and internalizing problems) and neurocognitive function (i.e., visual memory performance).

Consistent with recent DTI studies showing that probands with ASD and their unaffected siblings share similar WM microstructure [Barnea‐Goraly et al., 2010; Jou et al., 2016], our findings also demonstrate widespread WM alterations in a larger matched sample of unaffected siblings using the new TFCW method, and highlight several tracts connecting to the frontal regions and the posterior part of the callosal fibers. The TFCW method helps to detect specific segments of long‐range fibers that cause the largest differences between the two groups. The results of initial screening by TFCW top 2% criteria were robust, based on the fact that these findings were confirmed by the validation procedure using the general linear model. It is therefore justified to state that using the top 2% criteria as a cutoff point can efficiently capture most of the altered WM integrity in the siblings. Combined with previous studies regarding volumetric changes [Dalton et al., 2007; Mitchell et al., 2009], the siblings of probands with ASD may share brain structural abnormalities with their probands [Alexander et al., 2007; Barnea‐Goraly et al., 2010; Billeci et al., 2012; Catani et al., 2008; Groen et al., 2011; Jou et al., 2011; Ke et al., 2009; Mengotti et al., 2011], not only in specific brain regions [Dalton et al., 2007; Mitchell et al., 2009], but also in WM microstructure [Barnea‐Goraly et al., 2010; Jou et al., 2016].

The identified tracts correspond well to previous findings in probands with ASD. Evidence has shown that probands with ASD display a less mature WM microstructure than the TDC in the ILF and SLF, internal and external capsules, uncinate fasciculus, cingulum, corona radiate, and thalamic radiation [Barnea‐Goraly et al., 2010; Groen et al., 2011; Jou et al., 2011, 2016]. According to the classification proposed by Kaiser et al. [2010], these long tract abnormalities observed both in the probands and the unaffected siblings may reflect neural signatures with “trait” or “compensatory activity” but not “state activity.” “Trait activity” refers to the shared areas of dysfunction between unaffected siblings and probands with ASD that are related to the condition of having ASD and characterize the nature of disruption in its brain circuitry, thereby providing a promising neuroimaging endophenotype with which to facilitate efforts to bridge the gap between genomic complexity and disorder heterogeneity. “Compensatory activity,” which is unique to unaffected siblings (and not in probands), suggests a neural system‐level mechanism by which unaffected siblings might compensate for an increased genetic risk of developing ASD. Using this classification, SLF, thalamic radiation, and callosal fibers of the splenium may serve as trait bundles. The other tracts that were not selected to be altered by TFCW in our study but were found to be impaired in probands in previous studies, such as the cingulum [Ameis et al., 2013] and the anterior callosal fibers [Prigge et al., 2013], are potentially state tracts for ASD. Meanwhile, the tracts that were substantially altered in our unaffected siblings but have not been previously reported in probands may be candidates of compensatory bundles. In this sense, callosal fibers of the precuneus may be one of the compensatory bundles because a decrease in FA was found across the three subregions of the corpus callosum (genu, body, and splenium) in the ASD probands [Travers et al., 2015] while siblings in our study showed enhanced integrity. However, these notions may be too speculative and need to be validated in a three‐group comparison between probands, unaffected siblings, and TDC.

Multiple sources of evidence suggest corpus callosum atypicality in ASD. The corpus callosum is one of the first brain structures observed to be abnormal in neuroimaging studies of autism [Belmonte et al., 1995; Egaas et al., 1995; Frazier and Hardan, 2009; Hardan et al., 2009; Piven et al., 1997]. The decreased mean size of the corpus callosum, especially the anterior part [Prigge et al., 2013], is one of the most replicated structural imaging findings of the case–control studies of ASD [Travers et al., 2015]. The anterior and posterior callosal fibers are among the most rapidly developing white‐matter structures in humans. Throughout the postnatal developmental stage, white‐matter development of the splenium precedes that of genu, with the posterior callosal fiber structure remaining more directionally organized than the anterior callosum into adulthood [Hofer and Frahm, 2006]. In this sense, our finding of the altered microstructure in unaffected siblings peculiar to the posterior callosum is of particular interest. Whether the development of the posterior part follows a different programming than typically developing individuals in the way of compensation has not yet been studied.

Although recent DTI studies showed no correlations between autism symptomatology and WM FA in the unaffected siblings [Barnea‐Goraly et al., 2010; Jou et al., 2016], we found that the mGFA of 10 altered segments together can explain around 15–30% of the variance in autistic traits, while 10 whole‐tract mGFA can even explain as high as 40% of the variance. To our surprise, the impaired segments did not explain larger variance than whole‐tract mGFA. One possible reason is that the whole tract contains both the impaired segment and the unimpaired part of the tract, and these two parts may need to work together as a single unit to accomplish a task. As a result, the whole tract (both impaired and unimpaired segments) may explain more variance than the impaired segment only. Although the whole tract may explain more variance of the behaviors than the impaired segment, searching these impaired segments is more sensitive for identifying the affected tracts than the traditional method that averages the FA of the whole tract because the impaired segment may exhibit a larger effect size than the whole tract. The localized areas of abnormality may be used as markers to better differentiate patients and healthy controls. However, further studies are needed to validate its clinical significance.

Among the tracts, the SLF2 tract integrity was associated with the overall severity of social awareness deficits in unaffected siblings. The latest study has shown that the overall autistic trait measured by the Autism Spectrum Questionnaire (AQ) was associated with an average whole white matter skeleton FA by the Tract‐based Spatial Statistics (TBSS) [Gibbard et al., 2013]. They found that the regression slopes were strongest for social, communication, and attention switching scores. Another study also showed that AQ scores were correlated with FA of WM by TBSS, in several regions such as the frontal, parietal, occipital, and temporal lobes (including the posterior superior temporal sulcus) [Hirose et al., 2014]. Using tractography, they confirmed that the inferior fronto‐occipital fasciculus (IFOF) is a key fiber that links to autistic traits in healthy adults, emphasizing the role of IFOF in processing socioemotional information. Our findings more specifically highlight that WM integrity may be associated with social awareness deficits rated on the SRS in the unaffected siblings, and provide evidence showing that another long‐association fiber tract, SLF2, also subserves autistic traits. SLF2 is the major component of SLF which originates in the caudal‐inferior parietal cortex and terminates in the DLPFC. Similar to IFOF, SLF2 is a long‐range fiber connecting the frontal lobe and posterior brain and has been reported to be altered in probands with ASD [Cherkassky et al., 2006; Dichter, 2012]. Collectively, these lines of evidence support the notion that the long‐range fibers (SLF2 and IFOF) that are impaired in unaffected siblings and healthy adults with autistic traits are potential trait bundles for ASD.

As the first study to clarify the neural correlates of behavior problems in the unaffected siblings of ASD probands, we found that the higher levels of internalizing problems in the siblings were associated with the altered integrity of the frontostriatal tract from the right DLPFC. Evidence suggests that depressed individuals are characterized by aberrant functional connectivity in the frontostriatal circuits, and that these circuits are posited to support affective and cognitive processing [Furman et al., 2011]. Besides, the circuits connecting the medial prefrontal cortex and the ventral striatum are also involved in self‐esteem; its structural connectivity was associated with the trait of long‐term stability of self‐esteem while its functional connectivity during positive self‐evaluation was related to current feelings of self‐esteem [Chavez and Heatherton, 2015]. This evidence helps to explain the role of the frontostriatal tract on internalizing problems characterized by anxious/depressed behaviors and withdrawal. However, more direct evidence is needed to understand the effect of the frontostriatal tract on internalizing problems observed in our unaffected siblings.

We found that the performance of visual memory was associated with the integrity of the right frontal aslant tract. This tract was newly found and mapped during intraoperative imaging [Vassal et al., 2014], but its function was not clearly studied [Kronfeld‐Duenias et al., 2016; Sierpowska et al., 2015]. It connected the inferior frontal gyrus with the supplementary motor area and pre‐supplementary motor area and was recently introduced as part of a “motor stream” that plays an important role in speech production [Kronfeld‐Duenias et al., 2016]. It has also been shown to correlate with verbal fluency in both semantic and phonemic modalities [Kinoshita et al., 2015]. However, the DMS task is independent of speech production and semantic function where the stimuli are meaningless shapes, but this task indeed involves motor planning when responding to the computer task on the screen. Our study expands the existing knowledge of the implication of the right frontal aslant tract on short‐term memory performance measured by the DMS task. The DMS task is sensitive to medial temporal damage, with some input from the frontal lobes [Moscovitch, 1994; Robbins et al., 1994; Sahakian et al., 1988]. How the frontal aslant tract involves working memory deficits and whether it can also explain working memory difficulties specific to ASD warrant further research.

Limitations of this study are threefold. First, we used TFCW to select segments with the top 2% of differences; other altered tracts with smaller effect sizes were not included and could therefore have been possibly missed. Nevertheless, this approach ensures the identification of segments with the most substantial difference between siblings and controls. Second, this study did not recruit probands with ASD. Although many of the impaired tracts identified by our method were consistent with previous findings in the individuals with ASD, the altered WM tracts still need to be validated in the probands of our sample to see whether the same alterations also exist. Besides, the relationship between the impaired tracts and clinical phenotypes observed in unaffected siblings also needs to be examined in the probands. These WM alterations can possibly be considered an intermediate phenotype of ASD, but only when the ASD individuals showed the same impairments. However, in the three‐group comparison between the ASD probands, unaffected siblings, and TDC, gender distribution is always a major concern because ASD is a male‐predominant condition and gender influences brain connectivity. Third, SRS and SCQ alone may not be sufficient to exclude all siblings with a broad autism phenotype, even though the siblings with higher SRS (male > 70, female > 65) and SCQ (>15) were excluded. Nevertheless, this study used a large sample of unaffected siblings, whose phenotypes were ensured by clinical interview, standard ADI‐R interview, SRS, and SCQ questionnaires. We used a new method, TFCW, to identify the altered segments independent of prior hypotheses, and provide a new insight into the relationship between the altered WM and comprehensive phenotypes, including symptoms/behaviors and neurocognitive function.

In conclusion, unaffected siblings of probands with ASD showed widespread WM alterations. These changes may correlate with their autistic traits and neurocognitive functioning such as visual memory and spatial working memory. Besides, the TFCW method can efficiently identify tracts with remarkable effect size among the whole brain connectome. Our findings highlight the potential role of the WM microstructure as a trait marker for ASD.

DECLARATION OF INTEREST

None.

Supporting information

Supporting Information