Normative Modelling of Brain Morphometry in Conduct Disorder: A Mega-Analysis from the ENIGMA Antisocial Behavior Working Group.

Conduct Disorder (CD) is characterized by a pervasive behavioral pattern of disregard for societal norms and the rights of others (APA, 2013). Recent findings suggest that CD is the leading cause of burden amongst mental disorders in youth aged 0-14 years (GBD, 2022) while also incurring significant economic and social costs to society (Goulter et al., 2023; Rivenbark et al., 2018). However, youths with CD constitute a highly heterogeneous population comprising more than 32,000 theoretical combinations of symptoms (Nock et al., 2006). The clinical heterogeneity is currently addressed by the age-of-onset (i.e., childhood versus adolescent-onset) and the limited prosocial emotions (LPE) specifiers to CD, the latter of which overlaps strongly with callous-unemotional (CU) traits (APA, 2013). For instance, childhood-onset CD (i.e., emerging before 10 years old) is indicative of more severe aggressive and antisocial behaviors, as well as poorer psychosocial outcomes than adolescent-onset CD (Bevilacqua et al., 2018), although both subtypes are strongly linked to negative adult outcomes (Odgers et al., 2007). Additionally, up to 50% of children with CD will exhibit elevated levels of CU traits (Barker et al., 2011; Frick et al., 2014). Those with high CU traits typically exhibit conduct problems earlier in life than those with CD and low levels of CU traits (Déry et al., 2019; Neo & Kimonis, 2021). Normative modelling is a valuable tool to quantify participant-level deviations from the norm in brain-derived phenotypes (e.g., cortical thickness [CT], surface area [SA], subcortical volume [SV]). While traditional case-control analyses can be informative for identifying regional differences, quantifying the degree to which these deficits fall outside the normative range is necessary to better understand the heterogeneity in neuroanatomy in youths with psychopathologies such as CD. In other words, normative modelling allows to explore how heterogeneous youth with CD are on a given brain region, possibly identifying subgroups. or instance, a recent study from our Working Group showed that CD was associated with significant reductions in SA and SV that were widespread across the brain, and a localized difference in CT (Gao & Staginnus, 2024). It remains unknown to what extent the differences found in our previous mega-analysis fall outside the age-normative range of developmental brain trajectory. Indeed, these effects could be considered at the lowest or highest bounds of the normative range, referred to as infra-normal and supra-normal deviations, respectively. This is highly informative, as a significant deviation from normative brain development may be indicative of accelerated or delayed brain maturation. This has yet to be explored in youth with CD, which is of utmost importance for our understanding of the role of neural alterations in CD, and debates surrounding whether CD (or its subtypes) should be considered a neurodevelopmental condition (Fairchild et al., 2019; Raine, 2017). Normative modelling is also particularly useful for understanding the heterogeneity within a disorder by quantifying how many participants with a certain condition exhibit an extreme deviation from the normative range (see Figure 1). Consequently, regional morphometric measures that exhibit extreme deviations across many participants categorized by a shared diagnostic label (i.e., CD) may be identified as neurobiological markers of such a disorder. For example, in our recent mega-analysis, it is possible that only specific subgroups of youth with CD exhibit more substantial alterations in some regions, which may be obscured by traditional CD classifications and clinically defined subgroups (e.g., age of onset and callous-unemotional traits). Normative modelling approaches allow us to map how heterogeneous youth with CD are in their deviations from age-specific normative trajectories of neuromorphometric measures (CT, SA, SV). More precisely, such an approach quantifies the degree to which youth with CD share a neuromorphometric abnormality (i.e., extreme deviation from age-specific normative trend) in the same region. The small effect sizes (Cohen’s d ≤ 0.26) found in our previous case-control analyses (Gao & Staginnus, 2024) may be partly explained by the fact that only a small subset of youth with CD show regional changes outside the normative range. This hypothesis has never been tested. An empirically validated normative model of brain morphometry has recently been developed using 37,407 healthy controls (53% female; aged 3-90 years) (Ge et al., 2024). Using this valuable resource, Haas and colleagues (Haas et al., 2024) have shown that patients at clinically high risk for psychosis (CHR-P) did not differ significantly from healthy participants in their deviation from the normative range (either infra-normal or supra-normal deviations). Furthermore, they revealed that the percentage of individuals showing extreme deviations in any given region do not exceed 12%, either for healthy controls (9.30%), and CHR-P (11.42%). The application of normative modelling in the field of antisocial behaviors is still in its infancy. However, a recent study has demonstrated that male adults with a history of severe violent behaviors had a significantly higher frequency of extreme negative deviations, especially in SA (i.e., subcallosal, insula) compared to controls (Haukvik et al., 2023). These findings suggest that on average, these subjects may show less developed SA in subcallosal cortex and anterior insula. More importantly, only a small percentage of participants with a history of violent behaviors showed extreme deviations in these regions (max 8.62% versus max 6.63% in HC), highlighting the substantial heterogeneity in the structural brain correlates of violence across participants, but also the limitation of traditional case-control analyses (Haukvik et al., 2023). Overall, these studies highlight the importance of taking individual differences into consideration, as well as using normative modelling to capture the variability of neuroanatomical measures across the normative range. While normative modelling may (or may not) identify general patterns of variation within psychiatric disorders, it also highlights the heterogeneity in brain deviations between patients sharing the same diagnosis or psychopathology. From regional structures to networks: Normative network mapping Recently, researchers have attempted to identify how brain lesions in spatially distinct regions may map onto a common network. For example, damage to regions in both prefrontal and temporal cortices are known to temporally precede the emergence of ASB (Nakaji et al., 2003; Anderson et al., 1999; Mitchell et al., 2006). Yet, it is still unclear how these spatially heterogeneous lesions are both related to the development of ASB. Using a novel approach that leverages the normative functional connectome of a thousand healthy volunteers, these lesions were functionally connected to a common network, which overlapped with meta-analytic maps of brain activity during morality, value-based decision making, and theory of mind (Darby et al., 2018). In other words, measuring where information is propagated across the brain (distributed versus localized) may help identify how seemingly distinct lesions (or alterations) across the brain may be related to the same network. Since within-subject test-retest reliability of functional connectivity remains poor (Noble et al., 2019), using connectome data from large samples of healthy individuals to investigate how subject-specific alterations are connected to the rest of the brain—rather than relying on subject's own data—represents a promising approach for studying individual-specific brain signatures (Fox, 2018; Segal et al., 2025). This approach has been successfully demonstrated in recent studies on brain activation (Peng et al., 2022), transcranial magnetic stimulation (Siddiqi et al., 2020), and deep-brain stimulation (Horn et al., 2022). Normative network mapping has great potential to identify “neural equifinality” across participants sharing a similar psychiatric condition such as CD. For example, Haas and colleagues (Haas et al., 2024) showed that most CHR-P participants reported extreme deviations in at least 1 regional morphometric measure (CT=70.4-74.6%, SA=62.8-79.7%, SV= 6.64-40%). However, less than 12% had deficits in the same specific region with comparable rates to those found in healthy controls (Haas et al., 2024). Moreover, this low rate of overlapping extreme deviations has also been identified among patients with bipolar disorder and schizophrenia (Wolfers et al., 2021; Lv et al., 2021), suggesting that it may be a characteristic of psychiatric disorders more broadly. Consequently, it is conceivable that, while patients may show significant deviation from a normative trajectory in spatially heterogeneous brain regions, those regions may in fact map onto a common network. This normative network mapping approach (see Figure 2 for an overview of the workflow) has recently been extended to meta-analytic data, suggesting that while youth with CD may show deficits in spatially distinct regions, these regions may be functionally connected to a common network. Indeed, the heterogeneity of regional peak coordinates across fMRI studies on CD mapped onto a shared network that included the posterior middle cingulate cortex, thalamus, hippocampus, visual and temporal gyri (Dugré & Potvin, 2024). This approach was further applied to fMRI studies on psychopathy, which revealed strong overlap in neural circuitry across CD and Psychopathy (Dugré & De Brito, 2024). However, the use of meta-analytic data currently limits the generalizability of these findings, a limitation that can be addressed using participant-level data from the ENIGMA-ASB Working Group. The potential of this method has also been demonstr