Local structural connectivity is associated with social cognition in autism spectrum disorder

Abstract

The current theory implying local, short-range overconnectivity in autism spectrum disorder, contrasting with long-range underconnectivity, is based on heterogeneous results, on limited data involving functional connectivity studies, on heterogeneous paediatric populations and non-specific methodologies. In this work, we studied short-distance structural connectivity in a homogeneous population of males with high-functioning autism spectrum disorder and used a novel methodology specifically suited for assessing U-shaped short-distance tracts, including a recently developed tractography-based atlas of the superficial white matter fibres. We acquired diffusion-weighted MRI for 58 males (27 subjects with high-functioning autism spectrum disorder and 31 control subjects) and extracted the mean generalized fractional anisotropy of 63 short-distance tracts. Neuropsychological evaluation included Wechsler Adult Intelligence Scale IV (WAIS-IV), Communication Checklist-Adult, Empathy Quotient, Social Responsiveness Scale and Behaviour Rating Inventory of Executive Function-Adult (BRIEF-A). In contradiction with the models of short-range over-connectivity in autism spectrum disorder, we found that patients with autism spectrum disorder had a significantly decreased anatomical connectivity in a component comprising 13 short tracts compared to controls. Specific short-tract atypicalities in temporal lobe and insula were significantly associated with clinical manifestations of autism spectrum disorder such as social awareness, language structure, pragmatic skills and empathy, emphasizing their importance in social dysfunction. Short-range decreased anatomical connectivity may thus be an important substrate of social deficits in autism spectrum disorder, in contrast with current models.

autism spectrum disorders, structural connectivity, local connectivity, social cognition

See Eyler (doi:10.1093/brain/awy293) for a scientific commentary on this article.

Introduction

Autism spectrum disorder (ASD) is a life-long neurodevelopmental condition, currently estimated to affect 1 in 68 children in the USA (Developmental Disabilities Monitoring Network Surveillance Year Principal et al., 2014) and 1 to 1.5% of children and adults worldwide. Defining symptoms are impairments in social communication, repetitive and restricted behaviours and interests (Diagnostic and Statistical Manual of Mental Disorders 5th edition, DSM-5) (Loth et al., 2017).

Several concurrent models have been theorized to account for the pathophysiology of ASD. Among these, following initial results of decreased connectivity in ASD (Horwitz et al., 1988), some authors suggested that ASD is characterized by long-range underconnectivity and short-range overconnectivity (Belmonte et al., 2004; Courchesne and Pierce, 2005; Solso et al., 2016). This model posits that atypicalities in brain connectivity are central to the pathophysiology and symptoms of ASD (Geschwind and Levitt, 2007). The more nuanced dysconnectivity theory has recently gained increasing attention, and describes that in view of recent conflicting results, there is modest evidence in support of the long-range underconnectivity/short-range overconnectivity hypothesis. Many functional MRI and diffusion tensor imaging (DTI) studies show that individuals with ASD exhibit reduced functional connectivity and/or fractional anisotropy, or overconnectivity or a mixed pattern concerning long-range or local connectivity (Rane et al., 2015). As a whole, the existing connectivity data in ASD are inconsistent but all in all support the general idea of dysconnectivity in ASD.

Interestingly, short-range connectivity plays a pivotal role in development of structural integrity of the brain, varying from birth to adulthood and across cortical regions (Phillips et al., 2013; Ouyang et al., 2017). These cortico-cortical tracts, which connect adjacent cortical gyri, have been shown to play a crucial role in cognitive function (Fornito et al., 2012). They provide cascading connections from primary unimodal to sensory association and multimodal areas and are thus vital for linking sensorimotor and higher cognitive brain functions (Mesulam, 1990); they also represent an essential component of superficial white matter, as first described in the 19th century (Meynert, 1872). Superficial white matter rests just beneath the cortex and is composed of U-shaped association fibres with small diameter (Phillips et al., 2016), intracortical axons that extend directly to white matter from the overlying grey matter, and of termination fibres from deep white matter pathways (Catani et al., 2012). The superficial white matter axons are the last to be myelinated—they may remain incompletely myelinated until the third decade of life—and with fewer wraps than in the deep white matter. This results in high plasticity but also less protection against damage, and high vulnerability to disease processes (Wu et al., 2016). Since ASD is a neurodevelopmental disorder and presents social cognitive deficits, it is of great interest to study short-range, also called local, structural connectivity.

In contrast with most studies using functional MRI, few studies (Sundaram et al., 2008; Shukla et al., 2011; Ouyang et al., 2017) have used structural connectivity methods to probe local connectivity in ASD; they used track-based spatial statistics (TBSS), which have been designed to study deep long fibres and not U-shaped superficial white matter. The anatomical arrangement of superficial white matter fibres poses severe challenges for tractography (Reveley et al., 2015) because of their smaller size, complexity, and high intersubject variability. In addition, these studies focused on paediatric populations.

To overcome these limitations, we investigated a homogeneous adult population and used the first developed tractography-based atlas of the superficial white matter fibres (Guevara et al., 2017); this atlas is composed of the most stable short association tracts, having a length between 20 and 80 mm, which allowed us to thoroughly explore those fibres for the first time in ASD.

Our aim was to study the short-distance anatomical structural connectivity using diffusion weighted imaging (DWI) in a sample of adult patients with high-functioning ASD and to relate it to the main clinical manifestations of ASD such as social cognition and executive functioning. To our knowledge, no such work has been performed to date.

Materials and methods

Participants

We recruited individuals with ASD without intellectual disability and controls from the Mondor University Hospital (Créteil, France). All patients were diagnosed with DSM-5 criteria, with the Autism Diagnostic Interview-Revised (ADI-R) and the Autism Diagnostic Observation Schedule (ADOS) rated by trained clinicians; seven patients had either an antipsychotic or an antidepressant treatment or both, the others were unmedicated. Controls were recruited from the general population through advertising. Non-inclusion criteria were: age below 18 or above 55, personal history of intellectual disability or neurological disorder, substance-use disorder except for tobacco and contra-indications for MRI. Controls had no DSM-5 axis-I psychiatric disorder and no first-degree family history of schizophrenia, schizo-affective or bipolar disorder, and no personal history of neurodevelopmental difficulties. To increase the within-group homogeneity, we restricted our recruitment to male subjects with IQ > 70.

The clinical and cognitive evaluation of patients and controls included WAIS-IV. Self-questionnaires used were Communication Checklist-Adult (CC-A), which includes three domains (language structure, pragmatic skills and social engagement), and Empathy Quotient. Hetero-questionnaires were Social Responsiveness Scale (SRS), which yields social subscores (social awareness, cognition, communication and motivation) and restricted interests and repetitive behaviour subscore. We also used the BRIEF-A.

Of the 70 male subjects recruited (30 patients with ASD and 40 controls), 12 subjects (three patients and nine controls) with significant MRI artefacts or movements were excluded from the analyses. We thus analysed the data of 58 subjects: 27 patients with ASD and 31 controls aged from 18 to 55 years. Demographic and clinical characteristics of participants are presented in Table 1. The local ethics committee (CPP Ile de France IX) approved the study and after participants received a complete description of the study, we obtained written informed consent.

Table 1

Demographic, clinical and framewise displacements characteristics of participants

	ASD (n = 27)		Controls (n = 31)		Student t-test value	Degree of freedom	P-value
	Mean	SD	Mean	SD
Age	28	9.4	30	9.1	0.859	56	0.394
Framewise displacements	2.8	0.6	2.7	0.5	−0.380	56	0.705
Cognitive scores
Full IQ (n = 24/14)	104	21.8	114	15.3	1.500	36	0.142
IRP	100	20.8	104	15.3	0.612	36	0.545
ICV	110	18.9	122	11.8	2.231	36	0.026
IMT	106	20.1	109	16.6	0.537	36	0.595
IVT	91	19.2	107	17.7	2.586	36	0.014
Diagnostic scale
ADOS total (n = 25)	10.8	3.9	N/A	–	–	–	–
ADOS communication (n = 25)	3.9	1.4	N/A	–	–	–	–
ADOS social (n = 23)	6.7	3.1	N/A	–	–	–	–
Social cognition scales
Social Responsiveness Scale
T-score social (n = 23/12)	66	9.7	39	2.5	−12.109	27	<0.001
T-score social awareness (n = 22/12)	61	10.2	38	4.4	−9.037	31	<0.001
T-score social cognition (n = 23/12)	63	11.8	40	3.7	−8.261	30	<0.001
T-score social communication (n = 23/12)	65	9.1	41	3.4	−11.344	31	<0.001
T-score social motivation (n = 23/12)	67	9.5	41	2.8	−12.401	29	<0.001
Empathy quotient
Empathy quotient score (n = 19/13)	26	6.8	43	10.0	5.642	30	<0.001
Communication Checklist-Adult
Communication – language structure score (n = 22/13)	6.9	4.8	1.6	2.3	−4.381	32	<0.001
Communication – pragmatic skills score (n = 23/12)	16	10.7	2.7	2.6	−5.841	27	<0.001
Communication – social engagement score (n = 24/13)	34	8.6	7.8	5.6	−9.776	35	<0.001
Executive functions scale
BRIEF-A total score (n = 23/12)	132	26.8	79	10.0	−8.315	31	<0.001

MRI acquisition

All subjects underwent an MRI acquisition on a Siemens 3 T Magnetom Tim Trio, with a standard 12-channel head coil, at the NeuroSpin neuroimaging centre (CEA Saclay, France). The T₁-weighted acquisition parameters were: repetition time = 2300 ms; echo time = 2.98 ms; field of view = 256 × 256 mm²; voxel size = 1 × 1 × 1.1 mm³, duration = 7 min. DWI data were acquired along 60 directions, with a spatial resolution = 2 × 2 × 2 mm³, a b-value of 1400 s/mm plus one image in which b = 0, duration 10 min and a field map to correct for susceptibility distortion artefacts.

Data processing and quality control

We performed whole-brain deterministic tractography using a validated processing pipeline previously described in a study that highlighted long-range decreased fractional anisotropy in an overlapping sample of subjects with ASD compared with controls (Katz et al., 2016). We used BrainVISA and Connectomist 2.0 software (http://brainvisa.info) for image processing (full description in the Supplementary material). To segment the whole brain into the main superficial white matter tracts, we used the recent atlas of the most stable tracts, built with 79 typical subjects (Guevara et al., 2017), which describes 63 bundles in total, 34 bundles in the left hemisphere and 29 in the right. Authors selected only the bundles that were present in most of the subjects, with moderate to low variability in shape and number of fibres, and having a length between 20 and 80 mm. Finally, we visually checked each step of the DWI analysis and tractography process for each subject. We extracted the mean generalized fractional anisotropy along each reconstructed superficial white matter bundle as an estimation of its integrity. We chose to focus on generalized fractional anisotropy as it is the most widely studied DWI variable in ASD. A decreased generalized fractional anisotropy along white matter tracts may correspond to alteration in fibre coherence, axonal diameter, axonal density, demyelinization or oedema. Using this processing pipeline, we were thus able to study the connectivity of the 63 reconstructed superficial white matter bundles.

Statistical analyses

We used Student t-tests to compare the clinical variables, the subjects’ framewise displacements (to assess motion) and the number of fascicles of our samples.

Principal components analysis (PCA) was used for reduction of data dimension: we ran PCA on the mean generalized fractional anisotropy values of the 63 short tracts. Composite scores were thus built and computed for each subject, reflecting a linear combination of mean generalized fractional anisotropy values. The suitability of PCA was assessed prior to analysis. Inspection of the correlation matrix showed that all variables had at least one correlation coefficient >0.3. When a short tract belonged to several components, we attributed it to the component having the highest absolute value. The scree plot with inflexion point was used to determine the number of components to retain for rotation and interpretation. The scree plot for our analyses is illustrated in Supplementary Fig. 1. Outliers were defined as having standardized residuals >3 standard deviations (SD) and were not included in the analysis. Missing variables were replaced by the median generalized fractional anisotropy values of both groups. Finally, component scores are the linear composite of the optimally weighed original variables (mean generalized fractional anisotropy values).

To compare the mean score of each component between patients and controls, we performed an ANCOVA, considering age as covariate and diagnosis (fixed effects) as cofactor. Results were corrected for multiple comparisons (three tests) using Bonferroni correction.

To study the relationship between clinical/cognitive functions and mean component scores in patients and controls, we performed linear regressions considering mean component score as the dependent variable and age as covariate. Analyses were carried out separately in patients and controls. Results were corrected for multiple comparisons (nine tests) using false discovery rate (FDR) Benjamini Hochberg correction.

Before performing pairwise comparisons between patients and controls, the following assumptions were checked: standardized residuals were normally distributed, as assessed by Shapiro-Wilk test (P > 0.05) and a QQ-plot; homoscedasticity and homogeneity of variances were assessed by visual inspection of a scatterplot and Levene’s test.

We conducted several additional analyses to test for confounding factors. First, to test the effect of ADOS score, full IQ and number of fascicles on mean generalized fractional anisotropy measures, we performed a linear regression considering mean component score as the dependent variable and age and ADOS, age and IQ and age and number of fascicles, respectively as covariates. Second, for those components that display a significant difference between patients and controls, we looked which tracts were related with those cognitive functions, which are correlated to the components. To study this relationship, we performed a linear regression considering mean generalized fractional anisotropy value as the dependent variable and age as covariate. Results were corrected for multiple comparisons using FDR correction.

We conducted all statistical analyses with SPSS version 20.0 (IBM).

Data availability

The data that support the findings of this study are available from the corresponding author, upon reasonable request.

Results

Clinical characteristics, framewise displacements and number of fascicles

Clinical characteristics of all subjects in this study are presented in Table 1. There was no significant difference between mean ages, full IQ, and framewise displacements between ASD and controls. Finally, there was no significant difference in mean number of fascicles for the large majority of tracts (58/63).

Principal component analysis

As our whole-brain tractography analysis led to a large amount of correlated data, we used a PCA to capture the most relevant components, which retained the variance of the data. This revealed three components that had eigenvalues >1, explaining 15.1%, 6.5%, and 5.4% of the total variance, respectively. Visual inspection of the scree plot also indicated three components. In addition a three-component solution met the interpretability criterion. We thus retained three components and then used a Varimax orthogonal rotation to aid interpretability. Component loading of the rotated solution is reported in Supplementary Table 1. Twenty-two tracts constituted component 1, 16 component 2, and 13 component 3 (Supplementary Table 2).

Short-distance anatomical connectivity in each component

There was no evidence of age × diagnosis interaction for any of the three components.

We found no significant difference between the controls and ASD patients for mean scores of component 1 (P = 0.762) and component 2 (P = 0.405).

By contrast, we found a significant difference for the mean score of component 3. This component was significantly lower in the ASD patients compared with controls, after correction for multiple comparisons (P = 0.003; P_corrected = 0.009; df = 1; eta-squared effect size = 0.147). Sagittal, axial and coronal views of these tracts are illustrated in Fig. 1. Component 3 included temporal, frontal and parietal tracts and is hereafter named ‘temporal-frontal-parietal connectivity’.

Figure 1

Sagittal, axial and coronal views of the 13 short distance tracts of component 3.

Clinical measures and short distance anatomical connectivity

A subsample of 39 subjects (26 patients with ASD and 13 controls) for whom we had both generalized fractional anisotropy values and at least one cognitive measure were included in this analysis. Raw scores of SRS social score and subscores were converted to T-scores.

Patients had significantly higher values for the four social subscores of SRS (awareness, cognition, communication and motivation), the three domains of CC-A (language structure, pragmatic skills and social engagements) and BRIEF-A total score compared to controls. For these three scales, higher scores indicate worse performance. Empathy quotient was lower for ASD compared to controls (Table 1). Higher scores for this scale indicate better performance.

Within the control group, we found no significant correlation between temporal-frontal-parietal connectivity and clinical and cognitive scores.

Within the ASD group, we found a significant negative correlation between temporal-frontal-parietal connectivity and T-score of social awareness [standardized regression coefficient (SRC) = −0.56; P_uncorrected = 0.006; P_corrected = 0.018], communication-language structure (SRC = −0.60; P_uncorrected = 0.002; P_corrected = 0.018) and pragmatic skills scores (SRC = −0.54; P_uncorrected = 0.005; P_corrected = 0.018). In addition, we found a significant positive correlation between temporal-frontal-parietal connectivity and empathy quotient score (SRC = 0.52; P_uncorrected = 0.019; P_corrected = 0.043). Results are reported in Table 2 and plots of the four significant correlations in Fig. 2.

Table 2

Linear regressions between component 3 scores and clinical scores in patients with ASD

Clinical measures	SRC	P-value		FDR corrected P-value Clinical measures
		Clinical measures	Age
Empathy quotient score	0.52	0.019	0.182	0.043
T-score social awareness	−0.56	0.006	0.021	0.018
T-score social cognition	−0.20	0.348	0.144	0.522
T-score social communication	−0.14	0.526	0.174	0.676
T-score social motivation	0.03	0.897	0.163	0.920
Communication – language structure score	−0.60	0.002	0.016	0.018
Communication – pragmatic skills score	−0.54	0.005	0.033	0.018
Communication – social engagement	−0.35	0.094	0.227	0.169
BRIEF-A total score	−0.35	0.092	0.136	0.920

Figure 2

Correlations between temporal-frontal-parietal connectivity (component 3) and four features of social cognition.

Supplementary analyses

We performed additional analyses to test for confounding factors. We did not find any significant correlation between the temporal-frontal-parietal connectivity and the ADOS total score, the number of fascicles or the IQ. In the ASD group, the correlations remained significant when including IQ as a covariate in our multiple linear regression models.

Within the ASD group, we examined which of the 13 tracks of temporal-frontal-parietal connectivity (component 3) were correlated with the clinical scores. The results are reported in Supplementary Table 3). After correction for multiple comparisons (52 tests), we found a significant positive correlation between empathy and mean generalized fractional anisotropy of left inferior temporal-middle temporal (SRC = 0.69; P_uncorrected = 0.0004; P_corrected = 0.010). We found negative correlations between language structure and mean generalized fractional anisotropy of right supramarginal-insula (SRC = −0.72; P_uncorrected = 0.0004; P_corrected = 0.010). We also found negative correlations between social awareness and mean generalized fractional anisotropy of right supramarginal-insula (SRC = −0.65; P_uncorrected = 0.002; P_corrected = 0.035). Plots of the significant correlations are reported Fig. 3.

Figure 3

Correlations between mean generalized fractional anisotropy of specific bundles and three features of social cognition.

Discussion

We have investigated whole-brain white matter abnormalities in short-distance tracts and examined the correlations between neuropsychological functions and local connectivity patterns in ASD patients. Using a PCA, we found that the tracts’ fractional anisotropy were distributed among three components. There were no significant anatomical connectivity differences between ASD and controls in components 1 and 2. By contrast, our results show in individuals with ASD compared to controls, a deficit of anatomical connectivity in component 3, which comprises 13 short distance tracts, mostly present in frontal, temporal and parietal white matter regions. These regions are known to be important hubs involved in social cognition, emotion and language processing and have been reported of abnormal volume or cortical thickness in ASD (Yang et al., 2016). To our knowledge, this is the first comprehensive exploration of whole-brain short-range connectivity in ASD, and the first to report that these patterns of hypo-connectivity are associated with clinical severity of the disorder.

It is important to note that we observed short-range underconnectivity in ASD, whereas previous work described either short-range overconnectivity or short-range underconnectivity in ASD. This may be linked to (i) the clinical heterogeneity of ASD (age, sex, symptoms and severity of ASD, IQ); and (ii) the methodologies used (MRI machines, acquisition sequences and preprocessing pipelines) in the previous studies. Further, ‘connectivity’ can be assessed with many different methods, including anatomical (DTI) or functional (EEG, functional MRI).

Investigating connectivity in children with ASD is challenging from a clinical perspective, since ASD includes heterogeneous populations with an unstable diagnosis, as suggested by a study that evidenced six different developmental trajectories in patients (Fountain et al., 2012), with some severely affected early in life whereas others experience clinical improvement in later adolescence. In addition, it has been reported that 20% of patients diagnosed during childhood no longer met the diagnostic criteria for ASD at an adult age (Helles et al., 2015). Structural connectivity of short tracts has been explored in children with ASD. These results cannot be easily extrapolated to an adult population. As an example, a study showed in typical subjects the effects of age on short distance tracts and a widespread inverse relationship of fractional anisotropy values in superficial white matter with age (Nazeri et al., 2015). These child studies revealed atypical white matter microstructure in ASD indicated by decreased fractional anisotropy (Sundaram et al., 2008; Shukla et al., 2011). Using TBSS (Smith et al., 2006), one research group (Shukla et al., 2011) found reduced fractional anisotropy, increased mean and radial diffusivity in 4–35 mm length fibres in frontal lobe, and increased mean and radial diffusivity in temporal and parietal lobes in 26 children with ASD without intellectual disabilities aged from 9 to 18 years. Other studies (Sundaram et al., 2008) reported reduced fractional anisotropy and increased mean diffusivity of short tracts (defined as intra-lobe fibres) in the frontal lobe of both hemispheres in 50 children with ASD with mean age of 4.8 years. In contrast, a structural overconnectivity in prefrontal and posterior-cingulate cortex for ASD children aged 2–7 years has been described (Ouyang et al., 2017). It is apparent that the definitions of short-range connectivity varied across studies. Without a shared definition for short tracts, it is complicated to compare findings, since they correspond to the tracts connecting adjacent gyri (Im et al., 2014; Ouyang et al., 2017), or to tracts of short lengths regardless of their location (Shukla et al., 2011; Guevara et al., 2017). We only included adult patients with a stable diagnosis; in addition, we took age into account as a covariate in our statistical analysis and found no age × diagnosis interaction. This raises the possibility that the connectivity changes we observed are a compensatory process and proceed from brain rewiring due to defects with an origin early in childhood. This has been suggested in other neurodevelopmental conditions such as schizophrenia (Palaniyappan, 2017).

The lack of a specific methodology to study short, U-shaped fibres was another challenge. TBSS is not really designed to study short-range tracts (Maier-Hein et al., 2017), as it is focused on a simplified skeleton of deep white matter not taking into account superficial white matter (Zhang et al., 2010; Reveley et al., 2015). In our study, we have used state-of-art tractography algorithms with Q-ball modelling and a recent atlas built by Guevara et al. (2017) dedicated to the study of structural connectivity of short-distance tracts. This atlas of superficial white matter has been computed from a high quality high angular resolution diffusion imaging database of 79 subjects, using the same tractography software (Connectomist 2.0) as in our study. The weakness of TBSS’s skeleton strategy for superficial white matter is overcome by multisubject representations of each U-fibre bundle of the atlas modelling intersubject variability. The atlas has been defined from a hybrid non-supervised strategy associating parcellations of the cortical surface using FreeSurfer software and fibre clustering using shape similarities. Each bundle of the atlas has been shown to be reproducible in several groups of subjects, including psychiatric patients. For our study, only the short fibres between 20 and 80 mm were selected (Guevara et al., 2017).

Comparing our results with those obtained by functional connectivity is difficult as there is a complex and not static relationship between structural and functional brain connectivity (Uddin, 2013). Many functional connectivity studies, especially in early childhood, have led to the current theory of local overconnectivity (Keown et al., 2013; Supekar et al., 2013), while other studies showed either an opposite result (Khan et al., 2013) or mixed results depending on the regions considered (Maximo et al., 2013).

Social cognition includes the ability to share the perspective or point of view with another person, to mentally represent someone else’s intention, beliefs or emotion, to perceive and interpret social cues, to interpret the meaning of others behaviour or language (Mundy, 2018). Deficits in social cognition are among the core symptoms of ASD. Social cognition implicates several regions of the brain including the temporal region (Saitovitch et al., 2012; Olson et al., 2013) and the insular cortex (Uddin and Menon, 2009; Yamada et al., 2016). In this work, we showed in ASD patients strong correlations between mean generalized fractional anisotropy of left inferior temporal-middle temporal and empathy, and between mean generalized fractional anisotropy right supramarginal-insula and social awareness and language structure. All these cortical regions (temporal, inferior parietal and insular regions) are critical to social cognition. Decreased connectivity and altered integration of information in these local networks may thus be implicated in deficits in social cognition.

Finally, several mechanisms may account for dysconnectivity in short distance tracts in ASD. Courchesne and Pierce (2005) suggested that a series of pathological processes (neuroinflammation, migration defects, defective apoptosis, excessive cerebral neurogenesis and atypical synaptic pruning) may affect local connections in ASD patients.

Our study presents several strengths. Restricting the inclusion criteria to males aged between 18 and 55 and ASD with IQ > 70, with no history of alcohol use disorder, we conducted our analysis on a highly comparable sample in terms of age, sex and IQ avoiding confounding factors such as intellectual disability, age and unbalanced sex ratio. We used state-of-the-art tractography algorithms with Q-ball modelling and an automated segmentation pipeline that reduces the biases typically associated with manual definition of regions of interest and we performed a tough quality check to prevent quality acquisition bias. Finally, we carefully inspected each short distance tract for each subject; this quality check is important because of the high intersubject variability in the short distance tracts.

Several limitations and perspectives for future work should be considered. Our study was cross-sectional and in a limited sample. However, we found large effect size for our difference (eta-squared effect size = 0.147). This suggests that future, larger samples should allow the test of multivariate classification algorithms of ASD with short-range structural connectivity as input features. Such larger studies would also be helpful to disentangle associations between the two components of empathy, affective and cognitive, which are differentially affected in ASD. Restricting the inclusion criteria to adult ASD males with IQ > 70 limits the interpretation of the results to this population. Specifically, these findings are not generalizable to lower functioning adults and to females with ASD. It would thus be valuable to conduct an analysis of short-range structural connectivity in paediatric samples, but also in longitudinal studies, to address short-range connectivity changes during development. We did not include medications as a covariate, and this could have influenced the results, even though none of the controls and only a few of the subjects with ASD were medicated.

In conclusion, our study is the first to show that ASD subjects had a decrease of short-range structural connectivity in 13 tracts associated with social cognitive deficits, a key dimension of ASD. Albeit exploratory, these findings might suggest new targets for the treatment of social cognition deficits by repetitive transcranial magnetic stimulation, as some studies are suggesting that it affects structural connectivity (Anderson et al., 2016), although none have specially explored plasticity of short white matter tracts. Studies assessing structural connectivity abnormalities using more advanced models of water diffusion such as neurite orientation diffusion and density imaging (NODDI) are warranted to explore the cellular underpinnings underlying local dysconnectivity in ASD.

Abbreviations

Abbreviations
 
• ASD

  autism spectrum disorder

 
• PCA

  principal components analysis

 
• SRC

  standardized regression coefficient

Acknowledgements

The authors would like to thank the participating personnel of the centres, the personnel of the UNIACT lab, and the subjects who participated in this study.

Funding

This work was supported via collaboration with the Roche Institute for Research and Translational Medicine, the Investissements d’Avenir programs managed by the ANR under references ANR-11-IDEX-004-02 (Labex BioPsy) and ANR-10-COHO-10-01 and CONICYT FONDECYT 1161427, CONICYT PIA/Anillo de Investigación en Ciencia y Tecnología ACT172121 and CONICYT BASAL FB0008. This project has received funding from the European Union’s Horizon 2020 Research and Innovation Programme under Grant Agreement No. 785907 (HBP SGA2), No. 720270 (HBP SGA1).

Competing interests

F.B., C.C., C.B., E.T. and S.H. are employees of F. Hoffmann-La Roche Ltd who supported a part of study. The other authors have no other financial relationships with commercial competing interests to declare.