Youth antisocial phenotypes comprise two primary components: conduct problems (CP; i.e., externalizing symptoms involving rule breaking and aggression; related to conduct disorder and oppositional defiant disorder) and callous-unemotional (CU) traits (i.e., low prosocial emotions of empathy, guilt, and remorse; related to affective dimension of adult psychopathy; for review: Blair et al., 2014). These phenotypes are often comorbid and associated with aggressive and criminal behavior. However, CU traits are distinguished by profound socio-affective impairments (Blair et al., 2014; Frick et al., 2014b), which are thought to drive persistence in antisocial behavior (Frick et al., 2014a). Likewise, these phenotypes demonstrate nuanced differences in executive functioning (EF) tasks (e.g., Dotterer et al., 2021; Fantozzi et al., 2022; Gluckman et al., 2016) and brain structures that support EF (e.g., Alegria et al., 2016; Tillem et al., 2023; D.E. Winters et al., 2023; Winters et al., 2021). EFs are a set of higher-order cognitive abilities that support goal-directed behavior (Smolker et al., 2015) to which impairments are transdiagnostic (Roye et al., 2022) and relate to violent criminal behavior (Meijers et al., 2017). However, EF's relationship with CU and CP remain unclear. One potential reason for this is that single EF tasks are prone to measurement error (Burgess & Rabbitt, 1997; Miyake and Friedman, 2012; Phillips and Rabbit, 1997), which, fortunately, can be redressed by modeling latent factors across EF tasks (e.g., Miyake and Friedman, 2012; Miyake et al., 2000). Another reason may be that seemingly equivalent behaviors can stem from distinct brain patterns that align more closely with specific phenotypes. These distinctions can be clarified by examining the relationships between phenotypes, cognition, and brain function (Gilman et al., 2015). Specifically, such examinations would benefit from leveraging contemporary EF brain models that emphasize cross-network communication (for review: O'Reilly, 2010). Thus, the present study tests how the presence of CU and CP changes associations between contemporary EF brain models and latent components of EF.

EF consists of multiple cognitive processes, but many models converge on three core components: inhibition, shifting, and fluency (Karr et al., 2019; Latzman and Markon, 2010; Roye et al., 2022, 2020). The healthy function of these EFs is critical for mental health (Friedman and Robbins, 2022), emotion regulation (Gyurak et al., 2012), and socio-emotional competence (Riggs et al., 2006). The impairment in these EFs, particularly top-down cognitive control such as inhibition, are ubiquitous in mental health conditions (Friedman and Robbins, 2022). It is known that EF deficits in top-down control increase potential for mental health disorders; and different mental health symptoms may be linked to distinct EF components and unique neural mechanisms (Friedman and Robbins, 2022). Thus, differences may be revealed via EF components and brain interactions underlying CU and CP.

CU and CP's association with EF and associated functional brain properties in youth is complex. For example, when examining individual tasks, Dotterer et al. (2021) revealed neither CU traits or CP in youth associated with performance on the no/no-go or stop signal tasks, but did indicate an interaction between CU and CP such that CP associated with higher reaction times at lower CU traits and lower reaction times as higher CU traits. The authors conclude that CP uniquely contributes to sustained attention deficits during inhibitory processing tasks. Initial neuroimaging evidence by Tillem et al. (2023) demonstrates that those low in CU and high in CP had less efficiency (proportion of short connections) within the frontoparietal network, and this was partially supported by a whole brain study that did not detect frontoparietal efficiency in relation to CU (D.E. Winters et al., 2023). On the other hand, Gluckman et al. (2016) demonstrated CU traits uniquely associated with decrement in cognitive control, suggesting that it is the dynamic adaptation to implement top-down control in processes such as inhibition is a unique impairment in CU traits. Neuroimaging support indicates cross-network regions involved in conflict adaptation are less efficiently integrated at higher CU traits (D.E. Winters et al., 2023), and that the density (number of direct connections) between conflict adaptation regions accounts for CU traits, independent of CP, via cognitive control (D.E. Winters et al., 2023). Importantly, Winters and Sakai (2023) ound that individuals with higher CU traits had difficulty inferring others' emotions when cognitive control demands increased. These findings indicate the potential for important distinctions and interactions between CU and CP as well as brain connectivity that account for executive functioning deficits and may contribute to severity differences in antisocial behavior.

Adult studies on psychopathy have implemented latent EF modeling that demonstrates distinct associations between specific psychopathy facets affective deficits (specific CU related) and antisocial behavior (specific CP related). For example, Baskin-Sommers et al. (2015) found unique variance of broader psychopathy scores (i.e., Factor 1 and Factor 2) did not associate with a common factor of EF; however, further examination revealed the specific affective deficit facet (i.e., specific CU related; psychopathy checklist-revised facet 2 – Affective: Hare et al., 1990) accounted for deficits in common factor EF above all other facets of this measure (Baskin-Sommers et al., 2015). Similarly, Fournier et al. (2021) found EF tasks with affective stimuli that revealed that the affective deficit facet (CU specific) uniquely associated with worse inhibition and affect impairments above other antisocial facets. Friedman et al. (2021) used latent modeling and found the broader antisocial factor associated with lower common EF but they were unable to separate the affective deficit facet (see the articles supplemental). This finding may tie to other work suggesting EF deficits are more widespread in relation to CP but more specific in CU traits (Baliousis et al., 2019), which is echoed by Baskin-Sommers et al. (2022) who proposes a model positing that affective processing influences EF. It is unclear if findings in Winters and Sakai (2023) or Fournier et al. (2021) support Baskin-Sommers et al. (2022) model because we do not know if the underlying effect is due to salience detection (supporting the model) or top-down control (alternative explanation). Overall, this adult literature highlights the importance of modeling specific facets of antisocial symptoms for understanding the relationship between EF and both CU and CP.

Opportunities to improve our understanding of EF impairments are highlighted by adult meta-analysis discrepancies. For example, one meta-analyses indicates broader psychopathy factors for affective deficits (broadly CU related) have moderate to small decrements in inhibition (Gillespie et al., 2022) whereas another indicates only the broader antisocial factor (broadly CP related) associate with EF deficits (Burghart et al., 2023). There is substantial heterogeneity across studies included in these meta-analyses that limit cross-study inferences because, as Morgan and Lilienfeld (2000) state in their meta-analysis of EF in psychopathy, the mean effect does not properly represent the data. Given this, cross-validation could plausibly improve meta-analysis estimates in this line of study (e.g., Willis and Riley, 2017) by using study heterogeneity to examine effect reliability instead of imposing assumptions of normal distributions (i.e., random-effect regression). Importantly, Gillespie et al. (2022) points out many of these studies compare groups with arbitrary and varied definitions. Problems as a result of these arbitrarily defined groups by symptom extremes include (1) dichotomize continuous variables that loses explanatory information (Nuzzo, 2019), (2) impose assumptions of homogeneity that are demonstrably inaccurate (e.g., Dotterer et al., 2020; D.E. Winters et al., 2023), and (3) are misrepresented by means to which additional steps are required to reliably compare statistically (e.g., Liu and Wang, 2021).

These meta-analytic discrepancies increase the potential for bias related to sample characteristics. For example, a recent meta-analysis examined facet-level effects where one of the four studies used a community sample and the other three used a forensic sample (Gillespie et al., 2022), where the three forensic sample studies did not detect and effect (Maurer et al., 2016; Steele et al., 2016; Weidacker et al., 2017) and the one community sample study did find that an affective deficit (specific CU related) uniquely contributed to inhibition deficits (Fournier et al., 2021). While this could reflect sample differences, it is plausible these results are influenced by incarcerated sample characteristics such as floor effects of neurocognitive capacity (lower IQ and variance; e.g., Andover et al., 2011), celling effects of complex mental health issues (complex trauma and confounding mental health issues: e.g., Wolff and Shi, 2012), or non-random demographic factors (socioeconomic origins: e.g., Massoglia et al., 2013) that are pronounced in those that are incarcerated. This highlights the need to leverage samples with adequate variance and modern methods to bolster inferences on EF.

EF inferences distinguishing CU and CP can be bolstered by examining brain communication patterns based on a contemporary understanding of EFs in the presence of CU and CP. Critical developmental features of adolescent brains involve communication between regions (Ernst et al., 2015; Uddin et al., 2011). Brain interactions supporting EFs are emphasized in contemporary models that cut across canonical networks (for review of models: O'Reilly, 2010), and between network interactions have been found to be important for understanding both psychopathy (Dotterer et al., 2020) and CU traits (Winters and Hyde, 2022; D.E. Winters et al., 2023; Winters et al., 2021). Characterizing these connections may include any number of network properties (e.g., Bassett and Bullmore, 2009);. However, a metric such as efficiency is thought to be unhelpful for explaining the brain (Poldrack, 2015) and, because it describes the shortness of connections, efficiency is unlikely to reliably capture the longer connections involved in cross network models of EF. Therefore, the metric of connection density is plausibly more relevant because it describes all connections without imposing assumptions on how those connections occur. Such a brain metric can help us understand performance differences between mental health symptoms despite similar performance (Gilman et al., 2015) that could clarify the complexity in the literature. Finally, latent factor modeling of EF explains the EF domain with variance across related tasks that are prone to measurement error alone (Miyake et al., 2000). Thus, incorporating these factors in adolescent research may help resolve discrepancies in the literature.

The present study examines connection density in modern EF brain models in relation to latent EF factors and changes in the presence of CU and CP. We hypothesize that CU and CP will be uniquely associated with different EF components (inhibition, shifting, and fluency) and will also interact in distinct ways. Consistent with Tillem et al. (2023) we hypothesize that a three-way interaction for connection density*CP by CU will be present for inhibition as well as a common factor for EF. Finally, consistent with assertions by Baskin-Sommers et al. (2022), we hypothesize that CP will be associated with a common EF factor (representing global EF impairments) while CU will be associated with specific EFs. This work has implications for understanding the complex relationship between the brain and EFs in CU and CP.