Reduced Brain Connectivity in Clinical and Dimensional Autism Phenotypes Beyond Familial Confounding – A Twin Study

Previous studies on brain connectivity in clinical and dimensional autism have largely focused on selective connections and yielded inconsistent results. This study aimed to overcome these limitations. Global ber tracking allowed a more unbiased assessment of white matter connectivity and utilizing a within-twin pair design introduced implicit control for genetic and environmental factors shared by twins and allowed conclusions regarding their impact. The study examined the within-twin pair associations between structural brain connectivity of anatomically dened brain regions and both clinical autism spectrum diagnoses and dimensional autistic traits in 85 twin pairs (n=170; 56% monozygotic; 25 individuals with autism spectrum diagnosis). Structural connectivity was estimated using diffusion tensor imaging and linear regression models were t, adjusted for IQ, other neurodevelopmental and psychiatric conditions and multiple testing.


Background
Clinical autism, here de ned as ful lling diagnostic criteria for an autism spectrum disorder, is characterized by challenges in social and communication functioning along repetitive behaviors, restricted interests, and alterations in sensory processing (1). Clinical autism is associated with low employment rates, increased risk of anxiety and depressive disorders, and premature mortality (2). The heterogeneity of clinical autism and the dimensionality of autism-de ning symptoms makes it challenging to establish reliable biomarkers for assessment and intervention purposes. Research indicates that clinical autism is the extreme end of continuously distributed autistic traits (3), here referred to as dimensional autism. Studying biomarkers in association with both dimensional and clinical autism is a meaningful approach because dimensional markers assess the quantity of a phenomenon and provide hence important additional information in addition to diagnostic markers. In this study, we therefore investigated brain connectivity in association with both, clinical and dimensional autism.
It is generally assumed that atypical brain development leading to altered brain connectivity underlies the clinical autism phenotypes (4). The nature of these connectivity alterations is however still largely unknown since neuroimaging ndings have hitherto remained inconclusive. Early models of altered brain connectivity in clinical autism suggested overarching cerebral under-connectivity, particularly between distal cortical brain regions (see e.g. (6)). However, these models have been challenged by ndings indicative of functional over-connectivity in individuals diagnosed with clinical autism (6, 7). This inconsistency can partly be explained by phenotypic variation of clinical autism and methodological heterogeneity (6). Furthermore, age seems to modulate autism-related connectivity atypicalities: A developmental model brain connectivity in clinical autism suggests wide-spread over-connectivity in early infancy, mirroring ndings of early brain overgrowth, followed by altered neurodevelopmental trajectories with regional over-or under-connectivity later in life (7). Finally, environmental factors in uencing the etiology of clinical autism, such as parental age or preterm birth (8), might modulate the brain-structural and functional alterations associated with the condition.
Structural connectivity correlates of clinical and dimensional autism Structural brain connectivity is commonly studied using Diffusion Tensor Imaging (DTI), which takes advantage of the fact that water molecules diffuse along white matter bers (i.e. anisotropically). A recent meta-analysis of voxel-based morphometry across 14 DTI studies, including 297 individuals diagnosed with clinical autism and 302 neurotypical (NT) individuals, revealed decreased Fractional Anisotropy (FA), indicative of reduced white matter integrity, in the left splenium of the corpus callosum and the right cerebral peduncle, possibly re ecting sensorimotor impairments (9). In contrast, a review of 16 studies applying 'tract based spatial statistics' (TBSS), where FA data from DTI images are projected onto a prede ned skeleton of prominent ber tracts, identi ed more wide-spread reductions in white matter connectivity in older children, adolescents and adults diagnosed with clinical autsim compared to matched controls (10). The uncinate and arcuate fasciculi, the inferior longitudinal fasciculus, the inferior fronto-occipital fasciculus and the cingulum -tracts that are crucial for language and face/emotion processing, episodic memory, object recognition and attention control -were particularly affected (10). In a population-based sample of 604 6-to-10 year-old children, FA within the left superior longitudinal fasciculus and axial diffusion in the corpus callosum and the corticospinal tract were negatively associated with dimensional autism, suggesting that some autism-related changes in white matter microstructure might show a dose-like effect, increasing with increasing levels of dimensional autism (11).
Utility of twin designs for assessing autism brain biomarkers Given the genetic heterogeneity of clinical and dimensional autism and the di culties to control for all possible confounding variables, twin studies provide the unique opportunity to implicitly control for the genetic and environmental factors shared by twins. These familial factors might co-occur with (clinical) autism without being part of the condition's phenotype and might have biased the results of previous studies (12). Within-pair associations are adjusted for 100% of genetic factors in monozygotic (MZ) twins, except for post-twinning mutations, and on average half of the genetics in dizygotic (DZ) twins. Comparing associations within MZ vs DZ twins can thereby help to differentiate genetic and environmental in uences on brain structure and function (13). For instance, a meta-analysis of 48 brain imaging twin studies in NT twins indicated strong genetic impact on cortical morphometric measures and FA of most brain structures, and environmental in uence on cortical thickness of the uncus, left parahippocampal gyrus, and insula as well as FA in the callosal splenium (14). Interestingly, the latter structures have also been implicated in ASD (9,10,15).
Only a few studies assessed twin pairs discordant or concordant for clinical autism, showing, for instance, that twins ful lling diagnostic criteria for clinical autsim and their co-twins who did not showed similar reductions in gray and white matter volume and cortical folding (e.g. 22,23). This suggests that these autism-related brain alterations might be largely genetic. However, results have varied between studies, most samples have been small and partly overlapping, and no within-pair comparisons have been performed. One recent larger twin study reported that cortical thickness and cerebellar white matter volume were more in uenced by environmental factors in twins with clinical autism compared to NT twins, whereas the genetic and environmental in uence on other brain-structural measures was similar in both twins with and without clinical autism (18). A previous twin study from our group revealed autismrelated intrinsic functional brain connectivity alterations within twin pairs between core hubs of the salience network (19), while this study is the rst assessing autism-related structural brain connectivity in twins.

Methods
In this study, we applied a global tracking approach, allowing reconstructing the entire white-matter connectome without making predetermined anatomical assumptions (20), and twin design for a hypothesis-free assessment of autism-related structural brain connectivity alterations, unbiased by familial confounders. While within-twin pair analyses are perfectly controlled for age, within-pair association between autism and brain connectivity can be modulated by age (19). Hence, we investigated potential interaction effects between clinical and dimensional autism and age on brain connectivity in a follow-up analysis.  Table 1. Of originally 420 individuals in the complete RATSS cohort, 230 had to be excluded mainly because they or their co-twin had missing or too low quality data (for a detailed description see the Exclusion procedure section in the supplementary text, section 1.1). Note. *other diagnoses means here ful lling any diagnoses other than ASD; MZ = monozygotic, DZ = dizygotic, NDD = ful lling criteria for neurodevelopmental disorders other than ASD, psych. = ful lling criteria for psychiatric diagnoses (primarily anxiety disorders and depression); NT = neurotypical (de ned as not ful lling diagnostic criteria for any of the assessed neurodevelopmental or psychiatric diagnoses)

Diagnostic assessment
Twins underwent comprehensive assessment according to the RATSS protocol (21), including rst choice standardized diagnostic instruments for clinical autism, such as the 'Autism Diagnostic Interview -Revised' (ADI-R) and the 'Autism Diagnostic Observation Schedule' (ADOS or ADOS-2). Other NDD and psychiatric diagnoses were determined based on a multitude of sources, including the 'Kiddie Schedule for Affective Disorders and Schizophrenia', the 'Diagnostic Interview for ADHD in adults' and the 'Structured Clinical Interview for DSM-IV' (SCID, axis I). General intellectual ability was assessed with the Wechsler Intelligence Scale for Children or Adults, 4th Editions (WISC-IV/WAIS-IV) and the composite IQ score was calculated based on three verbal comprehension and three perceptual reasoning subtests.

Assessment of dimensional autism
Dimensional autism was assessed using the parent report SRS-2, see supplementary text, section 1.2. for psychometric characteristics), using total raw scores as recommended for research settings (23).

Image acquisition and processing
An approximately 50 minute MRI session in a 3 Tesla MR750 GE-scanner included a 5 minute T1- After quality control and pre-processing, streamline counts between 56 atlas regions (Supplementary

Statistical analyses
First we conducted explorative analyses on group differences and within-pair differences in dimensional autism and IQ, which are summarized in the Supplementary text, sections 2.1 and 2.2.
As the main analyses, we performed linear regressions in the generalized estimating equations framework with connectivity strength measured as streamline counts as dependent variable and clinical or dimensional autism as independent variables. Using an identity link function and conditioning on a unique twin pair id, we conducted within-pair analyses where each individuals were compared to their cotwins. Thus, confounding (and mediating) factors that are stable between the twins, i.e. all genetic and environmental factors shared by twins, were adjusted for by design (see e.g. 21). We tted models for 903 selected connections, adjusting multiple comparisons using false discovery rate (FDR, Benjamini-Hochberg method), as well as IQ, NDD diagnosis other than clinical autism, and psychiatric diagnosis. Twin pairs discordant for the main predictor contribute directly to the estimate (76 pairs were discordant for dimensional autism and 17 for clinical autism) while the remaining pairs in uenced the standard errors and affected the estimate indirectly if they were discordant for any of the covariates. Standardized estimates were calculated, which can be interpreted as effect size estimates. In a follow-up analysis, the interaction terms between clinical or dimensional autism and age were added to the models.
In order to complement our within-pair analysis, we also performed linear regressions across the cohort using the same statistical framework, treating twins as individuals but adjusting standard errors for twin clustering, and summarized these secondary results within the Supplementary text, section 2.5.

Associations of clinical and dimensional aspects of autism and structural connectivity
The statistics of all within-pair associations with clinical or dimensional autism surviving the FDR correction are summarized in Table 2 and visualized in Figure 1. Modulating effects of age on the association between ASD and structural connectivity The interaction between age and clinical autism was signi cant for 13 connections. Among these, however, were none of the connections observed in the main analysis for clinical autism, but primarily intra-hemispheric fronto-occipital connections (see Table 3). These interaction effects were all negative, and no interactions between age and dimensional aspects of autism survived correction for multiple comparisons. The uncorrected Z-values of these interaction effects are visualized in Supplementary  Figures 3 and 4. Note. Statistics of the Age*ASD diagnosis interaction effects surviving FDR correction. β = standardized regression coe cient, 95%CI = 95% con dence interval, SE = standard error, Z = zstatistic, p corr. = FDR-corrected p-value, p uncorr. = uncorrected p-value.

Discussion
In this study, we investigated changes in brain connectivity associated with clinical and dimensional autism, using global ber tracking and applying a within-twin pair design where familiar factors are implicitly controlled for. Both clinical and dimensional aspects of autism were associated with reduced connectivity beyond the in uence of familial factors. However, different connectivity alterations appear to be relevant for the clinical autism based on present diagnostic algorithms compared to dimensional autism. The results are discussed in more detail below.

Clinical autism and structural connectivity
Twins with ful lling diagnostic criteria for clinical autism had reduced white matter connectivity between the brainstem and the left cuneus compared to their co-twins without a diagnosis. While several lines of evidence suggest a brainstem involvement in clinical autism, the direct evidence for brainstem alterations from postmortem histological and in-vivo neuroimaging studies in humans remains limited (25), likely because the majority of brain imaging studies focused on predetermined cortical regions of interest. The brainstem hypothesis of autism suggests that atypical early brainstem development in clinical autism has cascading effects on cortical development, resulting in alterations in sensory processing that might be causal to other autism core symptoms (25). For instance, the superior colliculus of the brainstem and its interaction with cortical visual regions has been linked to visual exploration during visual search (26). Such changes in low-level visual processing could have a secondary effect on higher-order visual processing and cognition. The cuneus is an occipital brain region contributing to the dorsal visual stream, involved in form, motion and spatial processing and is a central, integrative hub within a functional visual brain network (27). This region has previously been implicated in clinical autism in a large study (394 individuals with clinical autism diagnosis and 473 controls), where lower effective (directed) connectivity of cuneus / precuneus to temporal brain regions involved in face processing to the was found in relation to both clinical autism diagnosis and autism symptom severity in the clinical group (28).
Our nding of reduced connectivity between the brainstem and cuneus in individuals clinical autism compared to their co-twins might indicate an atypical development of early aspects of the visual pathway in clinical autism, which in turn may in uence low-level perception and, in consequence, social information processing. These alterations might be considered a trait marker of autism, i.e. a marker of the dichotomous presence of clinical diagnosis, regardless of symptoms severity.

Dimensional autism and structural connectivity
Our results suggest that twins with more pronounced autistic traits tend to have reduced white matter connections from the left hippocampus to the left parahippocampal gyrus and to the left fusiform gyrus. The left hippocampus is crucial for (episodic) memory (29) and is, via the parahippocampal gyrus, connected to the brain's Default Mode Network (30), which is believed to be involved in self-referential thinking (31). Connectivity between the hippocampus and the fusiform gyrus is crucial for facial emotional processing (32), and reduced connectivity between these regions (33) and altered microstructure of the hippocampus-fusiform pathway (more densely packed but thinner ) (34) has been observed previously in individuals diagnosed with clinical autism compared to controls. The human fusiform gyrus contains the fusiform face area (FFA) which is crucial for face perception (35). Since challenges in facial emotional processing are a core feature of clinical autism, many functional brain imaging studies on autism investigated the FFA during face processing. A meta-analysis of 50 fMRI studies concluded that the left FFA and the left parahippocampal gyrus are more strongly activated in individuals with clinical autism during social cognition tasks, most of which involving face stimuli (36). However, since the fusiform gyrus function is not restricted to face processing but includes for instance also object recognition and space processing (37), its connectivity to the hippocampus is also relevant for non-social visual processing, such as memory-guided visual exploration in visual search (38). Therefore, the reduced connectivity between hippocampus and fusiform gyrus in association with dimensional autism observed in this study might re ect alterations in visual processing, including but not restricted to facial emotional processing. These might be regarded as a state marker of autism, a marker of quantitative autistic trait severity.
Modulating effect of age Some associations between clinical autism and connectivity between intra-hemispheric connections, involving primarily fronto-occipital connections, were signi cantly modulated by age (Table 3). These negative interactions indicate that the effect of clinical autism on brain connectivity between these regions decreased with increasing age, in line with developmental models of clinical autism (7,19). These interaction effects should be interpreted with caution, due to the limited number of twin pairs discordant for clinical autism. Still -if validated in further studies -this is a clinically interesting observation in that it might re ect compensatory mechanisms affecting structural network organization, potentially taking place during adolescence and early adulthood.

Genetic vs environmental in uences
In this study, genetic and environmental factors shared by twins are implicitly controlled. However, comparing associations between MZ and DZ sub-cohorts can allow conclusions regarding the in uence of non-shared genetic and environmental factors, because MZ twins are largely genetically identical, while DZ twins share on average half of their genes. When splitting the sample into sub-samples of 46 MZ and 37 DZ twin pairs, the associations between clinical and dimensional autism and brain connectivity remained signi cant within DZ but not MZ twins. However, the estimates were quite similar and their CIs overlapped between MZ and DZ sub-cohorts, allowing no rm conclusions with respect to genetic in uence on this association. For the latter, larger MZ and DZ samples might be necessary. Since the associations pointed into the same direction in both MZ and DZ twins but were statistically weaker in MZ twins, we speculate that both, genetic and non-shared environmental factors contributed to these within-pair associations.

Limitations
Focusing primarily on twin pairs with marked differences in autistic traits made this study more sensitive for detecting within-pair associations with autism phenotypes, but also prevented us from classic twin modeling of the quantitative genetic and shared vs non-shared environmental effects since these estimated would have been biased.
Power analysis for our twin analysis is not straight forward, due to the non-independence of observations and due to our recruitment strategy, however, we aimed to approximate the questions using G*Power (version 3.1.9.2), assuming N = 76 pairs differing in autistic traits by at least one point on the SRS.2. While the sample is comparatively large for a neuroimaging twin study on autism, our power calculation indicated a power of 85.6 in order to detect medium sized effects (β > .3) at α = .05.
A larger sample (N = 714 discordant twin pairs) would have been required to detect small effects (β = .1) at a power of 80%. Further, the FDR-correction for the relatively large number of tested connections (>900) might have increased the likelihood for type-II errors. For an overview over also sub-threshold effects (Zmaps), please see the supplementary material ( Supplementary Figures 1-4). These revealed visually relatively similar patterns of both increased and decreased connectivity in association with both, clinical and dimensional autism, indicating that rather than differing fundamentally in their effects on overall connectivity pattern, clinical and dimensional autism might only differ in respect to their strongest associations with structural connectivity.
Our study had a wide age range and an even distribution of males and females, increasing the generalizability across ages and sexes, but this variability might on the other hand prevented us from detecting sub-group speci c effects. In contrast, excluding individuals with an IQ<75 and individuals with insu cient brain imaging data quality has likely reduced the noise in the data, but restricted the sample largely to individuals in the normal IQ-range. Moreover, twin cohorts differ from non-twin cohorts in several ways. For instance, twins are more frequently born prematurely and suffer more often from growth restrictions (39). Hence, we cannot exclude the possibility that they also differ from non-twin samples in terms of brain connectivity.

Conclusions
Using a data-driven approach and a within-twin pair design, we found evidence for reduced connectivity between brainstem-occipital connectivity as a possible trait marker of autism, and reduced connectivity between regions involved in visual and especially face processing as a possible state marker of autism, beyond familial confounding and across sexes and ages. These associations were signi cant in DZ twins alone and attenuated in MZ twins despite pointing into the same direction, potentially indicating both genetic and environmental contributions. Negative interactions effects between clinical autism and age on brain connectivity might re ect compensatory processes.

Declarations
Ethics approval and consent to participate and consent for publication The regional ethical review board in Stockholm (Regionala etikprövningsnamden i Stockholm) approved the study protocols. Written informed consent was obtained from all participants and/or their caregivers for participating in the RATSS study and publishing the group results based on the thereby acquired data.

Authors' contributions
The rst author (JN) conducted the statistical analyses and produced the rst draft with contribution from MiR. The second author (SM) conducted the global ber tracking of the DTI images. The last author (SB) designed the overall study concept. JN, SM, MaR, RK-H, LTvE and SB contributed to the analysis plan and interpretation of the data and revised the article critically. All authors read and approved the nal manuscript. Figure 1 Within-pair associations that survived FDR-correction in the whole sample. ALL = within-pair association within the whole sample, DZ = within dizygotic twins, MZ = within monozygotic twins. Lower/upper = lower and upper bound of the 95% con dence interval. Note that although none of the associations survived correction within the MZ sub-sample, the con dence intervals do not cross the zero line for the association between autistic traits and the L hippocampus -L fusiform gyrus connection and for the association between ASD diagnosis and brainstem -L cuneus connection. The forest plots (left side) were created in RStudio3.5.1, using the package "forestplot". Examples of the according connections (right side) were created using functions within the NORA platform (http://www.nora-imaging.com).

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementaryMaterialMolAut.docx