Shared and unique multimodal profiles of individual variation in executive function and intelligence

943 

Abstract Submission 

Andrew Reineberg¹

¹University of Pittsburgh, Pittsburgh, PA

Andrew Reineberg  
University of Pittsburgh
Pittsburgh, PA

Individual differences in higher-level cognition are important for success in school, the workplace, relationships, and maintenance of physical health (Diamond et al., 2013). Two cognitive constructs, executive function (EF) and intelligence, are phenotypically and genetically correlated (Friedman et al., 2008). However, correlations between these constructs are not one. Recent evidence shows EF may be a unique predictor of psychopathology outcomes such as major depression while intelligence may be a unique predictor of educational attainment outcomes (Hatoum et al., 2023). The pattern of brain structure and function that underlies these cognitive constructs are an important intermediate phenotype between genes and outcomes such as psychopathology and educational attainment. The goal of the current study is to determine how the brain-based predictive signal for EF and intelligence are similar and different. To do so, we used an established predictive modelling framework (Shen et al., 2015) extended to include all imaging modalities and cognitive measures available in the UK Biobank dataset (Sudlow et al., 2015).

The current study is an analysis of behavioral, resting-state fMRI, anatomical MRI, and diffusion MRI data of 38,100 participants from the UK Biobank. EF and intelligence scores were calculated for each participant via a factor analysis of behavioral data collected at three time points: initial recruitment, an at-home follow-up analysis, and at time of brain scan. Imaging data analysis pipelines have been described previously in Miller et al. (2016). Brain measures from each modality - 25 global cortical and subcortical volume measures, 48 diffusion tracts, and 441 functional connections – were reduced to factors using ICA and entered in ridge regression models with 10% held out for later validation testing. One ridge regression model was run for each of the intelligence and executive function behavioral factor scores. All analyses controlled for age, motion during the resting scan, income, gender, and socioeconomic status. Only participants within three standard deviations of the mean motion during resting scan were analyzed. Final sample size included in models was n = 28674 individuals with all imaging modalities and behavior (Mage = 61.29, sdage = 7.09).

Ridge regression models predicted 25.8% of variance in the EF and 17.6% of variance in intelligence in the hold-out sample. See Figure 1 for a visual representation of the brain components that most contributed to these predictions. For the sake of interpretation, brain factors are described as a summary of the individual phenotypes with the highest loadings on the factor. Overall, the signals were mostly overlapping with only a few notable differentiating phenotypes. Of 514 total brain phenotypes, 205 and 188 were substantial contributors to prediction of EF and intelligence, respectively. 157 features were shared predictors of both phenotypes, predominantly global brain volume measures and functional connectivity between higher order association areas such as parts of the dorsal attention, frontoparietal, and default networks. Features uniquely associated EF were global grey matter and cerebrospinal fluid volumes; mean fractional anisotropy of pontine crossing tract and tapetum; and 42 functional connectivity features. Features uniquely associated with intelligence were 32 functional connectivity features.

Although EF and intelligence are highly correlated both behaviorally and generically, this correlation is not perfect. Brain signals associated with the unshared variance between these two cognitive constructs could be important intermediate phenotypes between genes and psychopathology. Our results suggest global grey matter volume measures, two white matter tracts, and a subset of the functional connectome are uniquely associated with EF and thus are candidate endophenotypes for psychopathology.

Executive Function, Cognitive Control and Decision Making ¹

Classification and Predictive Modeling ²

Cognition

FUNCTIONAL MRI

NORMAL HUMAN

STRUCTURAL MRI

WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC