Neuroticism Heterogeneity in Item-Level Associations with Resting-State Functional Connectivity

490 

Abstract Submission 

Masaya Misaki^1,2, Heekyeong Park^3,1, Fan Chun Chieh^4,1, Wesley Thompson^4,1, Martin Paulus¹

¹Laureate Institute for Brain Research, Tulsa, OK, ²Oxley College of Health Sciences, University of Tulsa, Tulsa, OK, ³University of North Texas at Dallas, Dallas, TX, ⁴University of California San Diego, San Diego, CA

Masaya Misaki, PhD  
Laureate Institute for Brain Research|Oxley College of Health Sciences, University of Tulsa
Tulsa, OK|Tulsa, OK

Heekyeong Park  
University of North Texas at Dallas|Laureate Institute for Brain Research
Dallas, TX|Tulsa, OK

Fan Chun Chieh  
University of California San Diego|Laureate Institute for Brain Research
San Diego, CA|Tulsa, OK

Wesley Thompson  
University of California San Diego|Laureate Institute for Brain Research
San Diego, CA|Tulsa, OK

Martin Paulus  
Laureate Institute for Brain Research
Tulsa, OK

Neuroticism, a personality trait marked by negative affect and dysregulation [1], is linked to poor mental health outcomes [2] and significantly overlaps with symptoms of mood and anxiety disorders [3]. A genetic study [4] reveals that neuroticism is not genetically uniform but comprises various distinct genetic correlates, with certain traits linked to depression and others to anxiety. This indicates a complex, multi-genetic structure of neuroticism. Yet, the relationship between this heterogeneity and individuals' neurofunctional traits is not well understood. To explore the heterogeneity of neuroticism in its neurofunctional underpinnings, the present study examines resting-state functional connectivity (RSFC) associations with neuroticism item scores using the UK Biobank data.

The UK Biobank data from the first imaging visit of 33,194 participants (17,274 females, mean age = 63.6 [SD = 7.7]) were analyzed. The RSFC matrix provided by the UK Biobank represents a partial correlation between 55 independent components, containing 1,485 values. Neuroticism was assessed using the Eysenck Personality Questionnaire (12 items). We employed a machine learning (ML) predictive modeling approach to associate RSFC with neuroticism scores. An ML model was trained on whole-brain FC values to predict neuroticism scores, using three-fold cross-validation and controlling for confounding factors such as scan location, sex, age, motion, signal-to-noise ratio, and T1-EPI discrepancies. We utilized an automated machine learning (AutoML) approach that addresses the Combined Algorithm Selection and Hyperparameter Optimization (CASH) problem by automatically searching for and optimizing feature selection, model algorithm, and hyperparameters. Specifically, the H2O AutoML software package [5], known for its comprehensive algorithm selection and optimization capabilities, was used.

Classification accuracies were significant for all items (p < 5x10-8). A previous GWAS [4] identified four items each in the depressed-affect and worry clusters. The present result showed that items within the worry cluster showed higher classification accuracy compared to those in the depressed-affect cluster. Figure 1A illustrates the similarity patterns of the items, as measured by the correlation between the RSFC weights of the ML models. Notably, items within the same genetic cluster exhibit tight grouping. Figure 1B presents the correlation between RSFC pattern similarities and genetic correlations for each item pair. A high concordance was observed between RSFC pattern correlations and genetic correlations, r = 0.834 (p < 0.0001), suggesting a genetic basis for RSFC neuroticism subtypes. The RSFC signatures were distinctly different between the 'depressed affect' and 'worry' clusters (Fig. 2), indicating separate neural circuits for these subcomponents.

This study highlights the intricate relationship between genetic traits and RSFC patterns in neuroticism, demonstrating the distinct neural circuits involved in its different facets. The contrasting RSFC signatures between the "depressed affect" and "worry" clusters reveal the complexity of the neurofunctional basis of neuroticism. Specifically, RSFCs associated with depressed affect are distributed throughout the brain, while those associated with worry are concentrated in the cerebellum and caudate. The cerebellum, with its connections to the amygdala and prefrontal cortex [6], plays a critical role in modulating fear and anxiety responses. In comparison, the caudate, which is involved in reward processing and stress responses, is integral to the pathophysiology of anxiety disorders. These findings indicate that self-report measures of neuroticism may not fully capture its biological complexity, suggesting the potential for more comprehensive approaches to personality research that integrate neurofunctional data.

Psychiatric (eg. Depression, Anxiety, Schizophrenia) ¹

Classification and Predictive Modeling ²

Connectivity (eg. functional, effective, structural)

Anxiety

Other - Depression; Resting state; Functional connectivity; UK Biobank