Despite distinct diagnostic core symptoms between autism spectrum disorder (ASD) and attention deficit hyperactivity disorder (ADHD), emerging evidence suggests clinical and cognitive overlaps (
1). Inattention and impulsivity are common presentations in ASD (
1); autistic-like social deficits are frequently reported in ADHD (
1). Cognitively, both disorders are associated with executive dysfunction involving attention, working memory, and planning (
1). Both ASD and ADHD show strong familial tendencies with high heritability (
2,
3). Unaffected siblings, the at-risk populations, tend to show subtle levels in symptoms and executive and attention dysfunction (
2,
3). Neuroimaging features, such as brain endophenotypes and brain-behavior relationships, for individuals with ASD or ADHD, together with their unaffected siblings, will inform etiological insight, targeted intervention plans, and mental health promotion.
The behavioral and cognitive overlaps among ASD and ADHD could be a result of shared genetic vulnerability (
4), which may be mirrored by similar abnormal brain network organization. White matter fiber connections, as quantified by diffusion MRI, constitute the brain structural architecture (
5), which plays a vital role in facilitating functional network formation (
6) and gray matter structural coupling (
7), but few neuroimaging studies of overlapping/distinct brain alterations between ASD and ADHD have been focused on this area (
8). For example, atypical white matter organization in the corpus callosum (CC) and superior longitudinal fasciculus (SLF) have been identified in both disorders, based on separate comparisons with typically developing control subjects (
8). However, studies directly comparing white matter profiles in ASD, ADHD, and typically developing control subjects cannot replicate these findings (
9,
10). In addition to methodological variability, these diagnoses include widely heterogeneous phenotypes, extending from “discrete disorders” to “quantitative spectrums” (
2,
3). Combined with the emerging evidence of idiosyncratic brain alterations (
11–
13), “patients’ averages” adopted by conventional case-control approaches obscure within-group variations, leading to inconsistent results (
8–
10). Furthermore, some of this inconsistency might be explained by age and sex confounders. The case-control design tends to address this issue by including people within narrow age ranges to avoid within-ASD/ADHD heterogeneity (
8–
10), resulting in nonrepresentative samples. Normative modeling is a valuable paradigm shift to derive personalized brain metrics (
11–
13) that quantify atypicality in ASD or ADHD as individual deviations in developmental trajectories relative to typically developing control norms. Such an approach could yield individualized neurobiological fingerprints (
11) that would account for the inherent heterogeneity and the nonlinear age and sex confounders and permit statistical inference at the individual level. Previous studies have demonstrated that normative modeling parses heterogeneous alterations in brain morphometry in people with ASD (
14) and ADHD (
15), respectively. To our knowledge, no published study has investigated atypical whole-brain major white matter tracts in individuals with ASD or ADHD and their unaffected siblings as an extreme of typically developing control norms.
Discussion
To our knowledge, this is the first study to apply white matter normative models to derive personalized neural tract profiles, rather than group averages. We found that ASD and ADHD shared similar white matter tract deviations relative to the norm, as well as interindividual variability in the degree of deviation, providing the first evidence of white matter idiosyncrasy in neurodevelopmental disorders. Unaffected siblings of ASD probands, but not those of ADHD probands, shared these deviation patterns to some extent. Dimensionally, we discovered multivariate patterns of white matter deviations that are correlated with multifaceted neurodevelopmental psychopathology and cognition across categorical boundaries.
Unlike previous multigroup studies (
9,
10), we determined that both ASD and ADHD presented widespread alterations in the CC and association fibers, but the projection fibers, especially the corticospinal tracts and optic radiations, were less affected. The observed altered tracts are largely consistent with those jointly reported in a systematic review (
8). The inconsistency between previous large-scale studies (
9,
10) and ours may be partly explained by the lower sensitivity of group averages in the case-control context in detecting disorder-related subtle effects, because of interindividual diversity caused by age and sex confounders (
11). This issue could be addressed by normative modeling. The remaining interindividual variability may be reflected by white matter idiosyncrasy in neurodevelopmental disorders. This strength of the normative model may also account for a null result from child-adult comparisons, as the child and adult subgroups appear to deviate to a similar extent relative to the norm, consistent with findings from a recent morphometric study (
32). Interestingly, the white matter tracts exhibiting significant neurodevelopmental deviation effects relative to the age and sex norms are largely consistent with the white matter pathways involved in neurotypical sex differences (
33,
34). Given the male predominance in neurodevelopmental disorders, our findings indirectly support a notion that sex-differential genetic and hormonal factors may contribute to these neurodevelopmental phenotypes (
35).
Despite some distinct patterns based on comparisons with the typically developing control norm, direct between-group comparisons revealed only one tract with more pronounced alterations in ASD than ADHD, namely, the CC-prefrontal cortex. In addition, significant correlations of both mean deviation and idiosyncrasy profiles were found between ADHD and ASD, suggesting that when conceptualizing ASD and ADHD as extreme conditions of white matter tract deviations from the age and sex norms, they share similar patterns, with subtle distinctions.
Our novel findings of white matter tract “idiosyncrasy” suggest interindividual variation in tract alterations, contrary to the notion of the “average ADHD brain” or “average ASD brain.” Specifically, several association tracts connecting the prefrontal and parietal cortex showed statistically increased variability among individuals with ADHD. Unlike in ASD, the literature suggesting brain idiosyncrasy in ADHD is scarce (
15). Individuals with ASD had highly varying deviations in the cognitive control part of the frontal and limbic circuitries. Our results complement previous findings of idiosyncratic prefrontal functional activation (
13) and connectivity patterns (
12) in ASD. Our findings suggest that white matter tract alterations and idiosyncrasy are equally important in understanding both disorders. This scattered and variable pattern of white matter tract deviations from the norm may help explain why heterogeneity is the rule rather than the exception in these neurodevelopmental conditions (
2,
3).
With an unaffected sibling design, we not only replicate results from previous small-sample reports (
36–
38) but also extend the potential endophenotypic white matter candidates to tracts connecting prefrontal regions and tracts involved in sensorimotor integration. This finding supports the notion of indispensable involvement of the prefrontal and sensorimotor circuits in autism and autism symptoms (
2,
39). Interestingly, these patterns were largely driven by the sisters of ASD probands. This sex-specific result may reflect female protective effects in a differential liability model of ASD (
35). Unaffected sisters could carry more etiological loads, leading to more prominent brain alterations but have phenotypic expression similar to that of their unaffected brothers. Conversely, we did not identify significantly altered tracts in unaffected siblings of ADHD probands based on a categorical model. But together with the finding of a modest correlation in white matter deviation profiles between ADHD-sibling pairs (
Figure 2G), this still suggests some degree of familial similarity of this trait, but in a milder form. This inconsistency with previous categorical white matter reports on siblings of ADHD probands may be explained by low-powered studies (
40,
41), use of the average-patient approach (
11), and potential compensatory brain phenotypes in siblings (
42). Together, white matter tract deviation patterns were more alike between ASD probands and their siblings than between ADHD probands and their siblings, largely echoing a previous endophenotypic study investigating intrafamilial relationships of white matter in siblings discordant for ASD (
43). This may partially reflect mildly higher estimates of heritability in ASD (
2) than ADHD (
3). Interestingly, white matter idiosyncrasy is largely not shared between proband-sibling pairs, echoing the notion that brain idiosyncrasy is a “state” marker of neurodevelopmental disorders (
12,
13). The findings that unaffected siblings shared some extent of white matter alterations with their probands warrant the development of proactive mental health promotion strategies for these at-risk populations.
ASD and ADHD showed relatively similar patterns in both mean levels and variability of white matter tract deviations relative to the typically developing control norm in terms of the involved tract and high correlations of these individualized measures between both disorders. This somewhat blurred diagnostic boundary is consistent with previous reports of overlapping gray matter morphometric features (
44–
46), functional networks (
47), and white matter tract characteristics (
9,
10) between ASD and ADHD. Nonetheless, previous well-powered studies and meta-analysis suggest generally distinct patterns of structural alterations (
48–
50), resting-state functional brain organization (
51), and task-functional MRI activation during cognitive control between ADHD and ASD (
49). But no such evidence exists based on white matter diffusion MRI studies. The specific developmental impact on environmental-genetic vulnerabilities (
52) may account for this nonuniform picture. Notably, given the white matter fiber connections as the structural architecture of the human cerebral cortex (
5), future studies might benefit from mechanistic investigation of structural-functional coupling (
53) to advance the understanding of these seemingly inconsistent transdiagnostic traits across MRI modalities. Overall, our data suggest difficulties identifying a single brain hallmark underlying either disorder (
15,
16), given the prominence of brain idiosyncrasy. Thus, rather than suggesting disorder-specific tracts, elucidating the white matter tract correlates of neurodevelopmental psychopathology across diagnostic entities could align with the RDoC to add bottom-up brain-driven symptom/cognition dimensions to future nosology (
16).
Given the intertwined nature of brain-behavior relationships, it is impossible to find a one-to-one correspondence between tracts and symptom/cognition measures. Instead, we leveraged CCA to enable the disentanglement of correlations and demonstrate orthogonal modes that represent multidimensional relationships with diverse levels of neurodevelopmental psychopathology. These brain-driven dimensions incorporated symptoms across diagnostic categories while remaining congruent with common clinical pictures, that is, ASD and ADHD often coincide but have distinct relationships among core symptom domains. In both modes, the social-communication domain largely covaried inversely with inattention, repetitiveness, and inflexibility (
54). However, our data did not find neat convergence (
55) between two common tools used to estimate autistic traits, which may be explained by rater differences (
56), age ranges (
57), inconsistent dimension structures across studies (
58), and inclusion of at-risk cohorts here. The first mode showed that deviations of tracts connecting the prefrontal and temporal cortex (CC, uncinate fasciculus), cingulum (body), frontal aslant tract, and perpendicular fasciculus were associated with the social-communication domain (inversely covaried with inattention). The second mode indicated that association tracts and CC connecting temporal and parietal lobes were associated with inflexibility. (For an explanation of what clinical symptoms and cognitive functions may be related to the identified tracts, please refer to the Supplementary Discussion section in the
online supplement.) These results not only replicate the univariate correlation between the anterior CC and social impairment (
9,
39), especially in the ADHD and ASD samples (
9), but also provide neurobiological evidence underpinning neurodevelopmental spectrums encompassing different domains of ASD and ADHD symptoms.
Aligning with the RDoC’s cognitive systems, the CCA derived multiple orthogonal cognitive dimensions. The first dimension mainly involved a spectrum from visual memory, planning, processing speed, and visual memory, linking with the arcuate fasciculus, limbic tracts, and optic radiations. The second cognition dimension comprised generalized intelligence, planning, inhibition, and visuospatial (working) memory associated with the SLF and cingulum. The third and fourth dimensions comprised other aspects of attention as well as verbal or nonverbal intelligence, respectively. The last mode reflected the remaining cognitive functions, including set shifting and response variability. (For more detailed discussion of each mode, please refer to the Supplementary Discussion section in the
online supplement.) Each dimension was driven by deviations of unique sets of white matter tracts. Specific items selected by CCA that are linked to white matter deviation patterns are quite consistent with results from the previous univariate analysis on brain-cognition relationships (
59–
61). Nonetheless, the brain-driven multifaceted cognitive function observed here does not accord with the reported fractionated constructs in neurotypical populations (
62). This inconsistency may reflect distinct intercorrelations between individual cognitive tasks and symptoms in neurodevelopmental disorders. Besides, different white matter metrics, representing different histology changes (
28), may be variably sensitive in detecting different multidimensional brain-behavior relationships, as evidenced in the complementary findings using diffusivity measures (see Figures S24–S27 in the
online supplement). Nonetheless, our multivariate analysis lends new evidence for brain-defined clinical and cognitive spectrums across neurodevelopmental conditions. This study also partly responds to the long-standing question of whether outwardly similar features are underlain by similar or distinct mechanisms among neurodevelopmental disorders (
1).
This study has a number of caveats. First, the moving average was selected to construct normative models based on model-fitting procedures. We acknowledge that no optimal method exists, given dynamic trajectories of white matter development. Thorough methodological investigations are warranted. Second, although our sample comprised a wide spectrum and presentations of neurodevelopmental disorders, individuals with intellectual disabilities were not included. This limits the generalizability of our results. Third, the limited number of females with ASD limits the inference of sex-specific effects. Our male-predominant neurodevelopmental sample may also potentially lead to type II error in the sex-by-group interaction. Fourth, our cross-sectional design did not allow inferences of individual trajectories based on dynamic normative statistics. Further, the age range within our neurodevelopmental sample misses the middle-late adulthood and very early developmental stages. Lastly, the unique strength of this study design also limits the possibilities for instant replication of the findings. No available data sets (e.g., the Philadelphia Neurodevelopmental Cohort [
63], POND [
10], ENIGMA [
46,
48]) have ever provided such a cohort simultaneously with universally deep phenotyping including neuropsychological tests, an unaffected sibling design, typically developing control subjects of the broad age range for normative modeling construction, and single-scanner, high-quality advanced diffusion MRI data in one lab.
In summary, we used normative statistics in a large sample of ASD and ADHD probands and their unaffected siblings to model individual heterogeneity in whole-brain major white matter pathways. When neurodevelopmental disorders are conceptualized as white matter deviations from normative patterns, ASD and ADHD are more alike than different. In addition to being purely deviated from the norm, interindividual variability in white matter tract deviations is also remarkable in ASD and ADHD probands, but not in their siblings, suggesting that white matter idiosyncratic patterns are characteristic of neurodevelopmentally disordered states. Echoing the RDoC, our CCA findings demonstrate how specific individual deviation in white matter alterations may contribute to diverse arrays of psychopathology and cognitive functions along a continuum from subtle to strong neurodevelopmentally phenotypical expressions. The modestly shared white matter tract deviations and dimensional brain-behavior relationships in unaffected siblings may point to potential candidates for endophenotypes in these at-risk populations. Moving forward, the determination of ADHD and ASD subtypes may be achieved by data-driven clustering on multimodal personalized Z-score profiles. This endeavor could facilitate more valid brain-behavior estimations, thereby contributing to precision psychiatry.