Dimensional Brain-behavior Relationships in Children with Attention-Deficit/Hyperactivity Disorder

Introduction

Category-based diagnostic classification systems for psychiatric illness (e.g., bipolar disorder vs. schizophrenia vs. obsessive compulsive disorder) are central to modern-day clinical practice. Despite providing a standardized nomenclature and common clinical framework, high degrees of symptom overlap among disorders and extensive patterns of comorbidity have raised questions about the adequacy of a purely categorical psychiatric nosology (1, 2). The recent identification of common neural and genetic substrates crossing current diagnostic categories has further intensified concerns about the biological validity of categorical boundaries of psychiatric disorders. In response, researchers and clinicians are increasingly shifting towards dimensional conceptualizations of psychiatric illness (3, 4).

This reconceptualization has implications for our understanding of both diagnostic assessment and pathophysiology of psychiatric illnesses. Over the past three decades, the psychiatric neuroimaging community has primarily compared diagnostic groups (i.e., clinical probands vs. controls) to identify the neural correlates of psychiatric disorder. Dimensional measures of illness are rarely examined, or have been limited to measures of disease severity among patients. In contrast, a dimensional approach implies that the behavioral characteristics of a given disorder can be examined across both healthy and clinical populations, with the assumption that psychiatric disorders represent extreme variants of typical behavior. From this perspective, neural dysfunction associated with psychopathology likely represents variation along a spectrum that includes healthy brain function (5).

“Resting state” fMRI is emerging as an effective means of mapping dimensional brain-behavior relationships, due to its moderate to high test-retest reliability and the relative simplicity of data acquisition. Initial studies have explored brain-behavior relationships using a variety of cognitive and behavioral measures (e.g., response time variability, working memory accuracy, reading competence) (6–8). Studies examining psychiatric phenotypes are also beginning to emerge. For example, Di Martino et al. (9) related a measure of autistic traits in healthy adults to the strength of resting state functional connectivity (RSFC) between the anterior cingulate cortex (ACC) and insula. Similarly, Assaf et al. (10) observed that more severe social and communication Autism Diagnostic Observation Schedule scores were correlated with weaker positive RSFC between precuneus and ACC in patients with autism.

Although potentially fruitful, dimensional brain-behavior relationships must be interpreted with caution. Specifically, the relationship between dimensional measures and brain functional measures may vary as a function of whether or not pathological processes are present (i.e., the relationship between connection X and dimension Y may vary as a function of whether or not disease Z is present). Accordingly, certain relationships between dimensional measures of psychopathology and brain function demonstrated in healthy individuals may not be observed in clinical populations (i.e., they may not simply represent the extreme end of the measured dimension).

Here, to account for both dimensional and categorical effects, we investigated brain-behavior relationships in typically developing children (TDC) and children with Attention-Deficit/Hyperactivity Disorder (ADHD). Beyond the classic triad of hyperactivity, inattention and impulsivity, the high rates of externalizing (i.e., conduct or oppositional defiant disorders in 42 to 93% of cases) (11) and internalizing disorders (i.e., anxiety or depression in 13 to 51% of cases) in children with ADHD suggests the utility of dimensional approaches. Moreover, inclusion of TDC, who exhibit a lower prevalence of such disorders (e.g., major depressive disorder between 4 and 8% (12)) permits an examination of the full distribution of dimensional symptoms as well as the exploration of their neural correlates according to the presence/absence of inferred psychopathological processes.

We focused our exploration of RSFC on the default network (DN) (13, 14), as abnormalities in this network are increasingly appreciated in ADHD (15–17) and in populations characterized by either increased externalizing (18, 19) or internalizing symptoms (20, 21). Accordingly, we predicted that Internalizing and Externalizing scores derived from the Child Behavior Checklist (CBCL) (19) would be associated with differential RSFC within the DN. Moreover, we predicted that DN modulations would be observed across the full range of scores (i.e., across both ADHD and TDC participants). Finally, we hypothesized that the presence of a pathological process (i.e., ADHD) would further mediate dimensional brain-behavior relationships.

Methods and Materials

Results

Discussion

By adopting a hybrid approach in which we investigated brain-behavior relationships both dimensionally and categorically, we identified novel neural correlates of Internalizing and Externalizing scores in children aged 7–13 years. We identified dimensional brain-behavior relationships that were common to the two groups (TDC, ADHD), as well as relationships that were specific to one of the diagnostic groups or distinct across diagnoses.

We observed several significant relationships between CBCL scores and RSFC within the DN across all children, independent of diagnosis. That both Internalizing and Externalizing symptoms were related to DN RSFC is not altogether surprising. The putative functions of the DN have been suggested by empirical studies demonstrating a common pattern of brain activation across tasks involving “internal mentation,” (14) including moral decision-making (35), autobiographical memory (36), making predictions about the future (36) or inferring mental states to others (37). These functions have been reported to be altered in psychopathologies associated with either higher Internalizing (e.g., depression (38)) as well as higher Externalizing symptoms, (e.g., ADHD (39)). For example, a recent report demonstrated that higher levels of externalizing symptoms were associated with impaired affective decision-making in children (40). Studies have also shown that psychopathologies associated with increased internalizing symptoms such as depression were associated with impairments in theory of mind (41, 42). In addition, the complexity of the DN is being increasingly appreciated. For example, Andrews-Hanna et al. (43) demonstrated that the MTL subsystem, implicated in remembering and formulating thoughts prospectively, can be dissociated from the dMPFC subsystem, which is implicated in the representation of mental states (self- or externally oriented). Taken together, these findings support the relevance of DN dysfunction to both internalizing (e.g., depression) and externalizing (e.g., impulsivity) symptoms.

Distinct brain-behavior relationships between DN RSFC and CBCL scores were observed, depending on whether Externalizing or Internalizing scores were examined. Regardless of diagnostic group, higher Externalizing scores were associated with decreased negative RSFC between the MTL subsystem (part of the DN) and dorsal midline regions (part of the ‘task-positive’ network) commonly implicated in cognitive control. Decreased negative RSFC between DN areas and so-called ‘task-positive’ regions (6, 7) has been reported in a) typical controls who exhibit increased response time variability (7) and b) individuals with an ADHD diagnosis (19), of which response time variability is the single strongest behavioral predictor (44–46). Interestingly, a recent report demonstrated a relationship between response time variability and task-related brain activity in ACC in TDC but in temporal pole in children with ADHD, thus suggesting differential involvement of these regions in response time variability, depending on diagnosis (47). Our findings suggest that further investigation into the links between functional connectivity, externalizing behaviors, ADHD and increased response time variability is warranted.

In considering why the brain-behavior relationships for Externalizing scores appeared to be specific to the MTL subsystem, we note recent work implicating this subsystem in simulation of the future, or prospection, using episodic processes (14). Because prospection involves using and recombining stored information to plan and predict future events, it is not surprising that dysfunction in prospection has been discerned in participants exhibiting elevated externalizing behaviors such as impulsivity and/or hyperactivity (39, 48, 49).

Our findings draw attention to a recurring question – namely, what is the functional significance of patterns of negative functional connectivity? Negative RSFC is thought to reflect functional segregation or differentiation between brain systems (50). Accordingly, we speculate that decreased negative RSFC between the MTL subsystem and dorsal midline regions (such as dACC) in individuals with higher Externalizing scores reflects reduced functional differentiation between these systems and increased tendency towards cross-talk or interference – factors which likely contribute to sub-optimal behavioral self-regulation. Similar observations have been noted in task-based studies. For example, Weissman et al. (51) suggested that suppression of activity within DN component regions is crucial to efficient task performance during attentionally demanding conditions. Consistent with this proposal, several studies have demonstrated that a failure to effectively suppress DN activity is associated with decreased activation within task-relevant processing systems and compromised behavioral performance (17, 51, 52).

The brain-behavior relationships observed for Internalizing scores differed markedly from those observed for Externalizing scores. Higher Internalizing scores were associated with stronger positive RSFC between the midline core DN regions (i.e., PCC, aMPFC), as well as between PCC and IPL (part of the MTL subsystem). Internalizing scores comprise anxious, inhibited, depressed symptoms related to self-referential thoughts that affect the self-psychological environment. As such, our findings can be understood in the context of recent work implicating DN regions in spontaneous self-relevant cognition (14, 43). The direction of the relationships we observed (i.e., stronger positive RSFC within the DN associated with higher Internalizing scores) is consistent with a recent report implicating DN hyperconnectivity in depression and rumination. Specifically, Berman et al. (30) reported that excessive RSFC between the PCC and the subgenual cingulate cortex in depressed participants compared to healthy individuals was related to higher rumination traits.

Finally, we found that children exhibiting higher Internalizing scores show increased RSFC between HF and the bilateral temporal poles. Consistent with this pattern, Gorno-Tempiri et al. (53) reported behavioral changes in a patient following temporal pole degeneration and described a clear shift from an extravert to an introvert personality. Other authors have also associated temporal pole dysfunction with both depression and anxiety (54). Together with these observations, our findings suggest that, although not yet well understood, temporal pole dysfunction could play a role in the pathophysiology of internalizing symptoms.

As expected, multiple DN components exhibited differences in RSFC related to Internalizing and Externalizing dimensions, regardless of diagnosis. Such findings support emerging dimensional perspectives of psychiatric illness (5). However, our results also suggest that neither categorical nor dimensional measures alone provide a complete characterization of the relationships between behavior/symptoms and brain function. For several functional connections, we found that the specific nature of the dimensional brain-behavior relationships varied depending on whether or not psychopathology was present. For these functional connections, dimensional relationships were nested within diagnostic classification. These findings suggest that psychiatric illnesses (e.g., ADHD) should not be oversimplified as extremes of brain function (i.e., too much or too little functional connectivity). Instead, the presence of a psychopathological process may signify a more profound disturbance in aspects of brain function, with some, but not all systems exhibiting qualitative differences. At the same time, the relevance of dimensional brain-behavior relationships to symptom severity also highlights the importance of not oversimplifying psychiatric diagnoses (i.e., by taking purely categorical approaches).

Our results should be considered in light of limitations. We selected the two broad scales of the CBCL as measures of symptom severity, reflecting a wide variety of behaviors and symptoms. Although each scale may be subdivided further into psychopathological syndromes (e.g., depression, anxiety disorders), our sample size did not provide sufficient power to investigate each syndrome scale in detail. Another limitation was our use of a priori seed regions limited to the DN, which constrained the brain regions and networks we investigated. Therefore, although we observed several significant brain-behavior relationships, our analyses were necessarily incomplete, with type II errors likely. One conceptual limitation may relate to evidence that children with ADHD exhibit a delayed developmental trajectory, relative to TDC (55). As such, the distinct dimensional brain/behavior findings observed in ADHD could in part reflect brain immaturity, which has been suggested to characterize various developmental disorders (e.g., (56)). Further developmental studies, ideally longitudinally, are required to determine the extent to which ADHD brain differences reflect differences in developmental status (i.e., brain maturity) vs. age-independent aberrations. Although Externalizing and Internalizing scores did not differ between the sexes within either group and while no diagnosis by sex interaction was found, we cannot exclude the possibility that the higher proportion of males in the ADHD group may have exerted a confounding effect, as sex effects have been previously reported (57, 58). Finally, although we followed current recommendations to minimize introducing biases in ROI analyses (59), the exploratory nature of our work necessitates future replication in independent samples.

In summary, our findings highlight that RSFC provides a powerful tool for examining dimensional brain-behavior relationships and demonstrate the utility of considering both categorical and dimensional approaches when conceptualizing psychopathology. As such, they supports the incorporation of dimensional scales in addition to the classical categorical approach in future diagnostic classifications (i.e., DSM-5).

Supplementary Material