Hierarchical Neurocognitive Model of Externalizing and Internalizing Comorbidity

Abstract

Mounting evidence suggests hierarchical psychopathology factors underlying psychiatric comorbidity. However, the exact neurobiological characterizations of these multilevel factors remain elusive. In this study, leveraging the brain-behavior predictive framework with a 10-year longitudinal imaging-genetic cohort (IMAGEN, ages 14, 19 and 23, N = 1,750), we constructed two neural factors underlying externalizing and internalizing symptoms, which were reproducible across six clinical and population-based datasets (ABCD, STRATIFY/ESTRA, ABIDE II, ADHD-200 and XiNan, from age 10 to age 36, N = 3,765). These two neural factors exhibit distinct neural configurations: hyperconnectivity in impulsivity-related circuits for the externalizing symptoms and hypoconnectivity in goal-directed circuits for the internalizing symptoms. Both factors also differ in their cognitive-behavior relevance, genetic substrates and developmental profiles. Together with previous studies, these findings propose a hierarchical neurocognitive spectral model of comorbid mental illnesses from preadolescence to adulthood: a general neuropsychopathological (NP) factor (manifested as inefficient executive control) and two stratified factors for externalizing (deficient inhibition control) and internalizing (impaired goal-directed function) symptoms, respectively. These holistic insights are crucial for the development of stratified therapeutic interventions for mental disorders.

Figure 1. Identification of the stratified neural factors.

a. The predictive performance of behavioral symptoms related to psychiatric symptoms with the task-based connectivity model. Task-based connectivity was estimated from the EFT (angry and neutral conditions), the MID task (reward anticipation, positive reward feedback and negative reward feedback conditions) and the SST (go-wrong, stop-success and stop-failure conditions). b. The correlation matrix of the behavioral symptoms. The externalizing and internalizing symptoms showed high intra-correlations within their respective psychiatric domain, but low correlations between each other. Externalizing symptoms consisted of ASD, ADHD, CD and ODD. Internalizing symptoms comprised GAD, Dep., ED and SP. c. The correlation matrix of the brain-predicted symptoms. d. With a two-step reliable analysis, we identified two stratified neural factors for externalizing and internalizing symptoms, respectively. We first identified task conditions with reliable stratified cross-disorder edges, which are defined as predictive edges that only predict externalizing but not internalizing symptoms and vice versa. We found that only conditions from the SST and MID task had significantly more stratified cross-disorder edges than a random observation. Then, we further identified which type of cross-disorder edges reliably predict externalizing or internalizing symptoms, which was termed the stratified factors. We discovered that the externalizing neural factor consisted of positive-positive cross-disorder edges (positively predicted at least two externalizing symptoms), while the internalizing neural factor comprised negative-negative cross-disorder edges (negatively predicted at least two internalizing symptoms). e. We checked the longitudinal predictive effects and developmental trajectories of the stratified factors across ages 14, 19 and 23. EFT, emotional face task; MID, monetary incentive delay task; SST, stop signal task; ADHD, attention-deficit/hyperactivity disorder; ASD, autism spectrum disorder; CD, conduct disorder; ODD, oppositional defiant disorder; GAD, general anxiety disorder; Dep., depression; ED, eating disorder; SP, specific phobia; 14-brain, brain at age 14; 19-brain, brain at age 19; 23-brain, brain at age 23; 14-exter., externalizing symptoms at age 14; 19-exter., externalizing symptoms at age 19; 23-exter., externalizing symptoms at age 23; 14-inter., internalizing symptoms at age 14; 19-inter., internalizing symptoms at age 19; 23-inter., internalizing symptoms at age 23;

Figure 2:

Multilevel neuroanatomical characterization of the two stratified neural factors. a and b. The top 10% nodes and hub node connections (with high regional connections) in the externalizing and internalizing factor. The color bar indicates the normalized node degree (that is, the number of connections with other nodes). c. and d. The functional connections of the externalizing and internalizing factors shared similar large-scale network configurations that both were mainly localized between the motor, frontoparietal and salience networks. The color bar indicates the strength of normalized inter- or intra-network connections, where the number of connections between or within networks was divided by the largest connection number observed.

Figure 3. Functional specificity and generalization of the stratified neural factors.

a. The externalizing factor was specifically associated with most cognitive functions (14 of 25). The internalizing factor was specifically correlated with personality traits (5/5), especially neuroticism and anxiety. Two executive function measurements (between errors in SWM and proportion bet in CGT) and two personality traits (extroversion of NEO and reserve of TCI) showed distinct associations with externalizing and internalizing symptoms. The polygenic risk score (PRS) of ADHD and MDD showed a specific association with externalizing and internalizing factors, respectively. b Generalization of the NP factor across multiple developmental periods from preadolescence to adulthood in both population and clinical case-control datasets (ABCD, N = 1799; ADHD-200, N = 520; ABIDE II, N = 564; IMAGEN N = 998; STRATIFY and ESTRA, N = 433; and XiNan, N = 449). The significance level (that is, the grey color) was given as a false discovery rate (fdr) of 0.05. The P values were reported as the original value and could survive the multiple testing correction with Benjamin–Hochberg procedure. AGN, Affective Go-No Go; BMI, body mass index; DD, Delay Discounting Task, which measured ‘waiting’ impulsivity; MidOcci, middle occipital cortex; MidPFC, middle prefrontal cortex; NEO, NEO Personality Inventory; RVP: A, Target Sensitivity from Rapid Visual Information Processing task; PRM, Pattern Recognition Memory task; SURPS, Substance Use Risk Personality Scale; SWM, Spatial Working Memory task; TCI, Temperament and Character Inventory–Revised. AN, anorexia nervosa; BN, bulimia nervosa; AUD, alcohol use disorder; MDD, major depressive disorder;

Figure 4. Characterizing specificity of the cross-disorder neural factors at the levels of brain regions and extended networks.

a. Specificity score comparison of the three cross-disorder networks using regional-level node degree as an example. We first normalized the node degree to facilitate subsequent cross-factor comparisons. Then, we used a weighted method for calculating the regional specificity score with the formula: specificity score of Factor 1 = Factor 1/Factor 2 + Factor 1/Factor 3. By weighting the contribution of the brain region in the other two cross-disorder factors (Factor 2 and Factor 3), the estimated specificity score provides a more robust measure of the unique contribution of each brain region within this cross-disorder factor (Factor 1). Building upon previous findings , we hypothesize that as the brain scale increases from region to network, the specificity between cross-disorder factors will decrease. b and c. For each brain region and network, specificity scores were estimated for all three cross-disorder factors. The maximum specificity score for each ROI is considered indicative of its specificity to that cross-disorder factor. Then, we estimated a specificity distance, which is computed by subtracting the minimum specificity score from the maximum specificity score. This distance is interpreted as the uniqueness of this cross-disorder factor compared to others.

Figure 5:

Summary of hierarchical cross-disorder neural networks.