Gray Matter Morphometry is Related to Reading Abilities, but not Meaningfully

1023 

Abstract Submission 

Steven Meisler^1,2, John Gabrieli^2,3

¹Harvard University, Cambridge, MA, ²Massachusetts Institute of Technology, Cambridge, MA, ³McGovern Institute for Brain Research, Cambridge, MA

Steven Meisler  
Harvard University|Massachusetts Institute of Technology
Cambridge, MA|Cambridge, MA

John Gabrieli, PhD  
Massachusetts Institute of Technology|McGovern Institute for Brain Research
Cambridge, MA|Cambridge, MA

Reading is a skill that must be explicitly learned, being introduced too recently to be a product of evolutionary pressure. Reading is almost universally taught to young children, whose brains tend to be more plastic, adapting in response to development and experience. However, despite the amount of research that has already been done investigating gray matter morphometric correlates of reading skills, there has been little convergence of results outside of a relationship between global brain volume and reading skills [1]. This calls into question whether individual differences in reading skill are reflected by MRI features of particular brain structures (such as left-hemisphere regions that support reading), which one would expect if brain plasticity is domain-specific. Beyond the possibility that there is not a relationship between local brain morphometry and reading skills, several factors could be contributing to prior inconsistent findings, including small sample sizes, various MRI acquisition and processing techniques, and different cohort-specific phenotype characteristics. All of these limitations can be addressed by using high-quality publicly-available data. We rigorously examined how gray matter morphometry relates to individual differences in reading skills across childhood and adolescence using the Healthy Brain Network dataset [2].

Our final cohort consisted of 1943 participants aged 5-21. Participants had quality-controlled T1-weighted (T1w) images and phenotypic data including the Tests of Word Reading Efficiency (TOWRE; n=1810) and the Wechsler Individual Achievement Test (WIAT; n=1835) to gauge reading scores. All T1w images were run through FreeSurfer's recon-all pipeline, resulting in morphometric surface maps of gray matter volume (GMV), surface area (SA), and cortical thickness (CT). These maps were normalized to a standard "fsaverage" space, harmonized across sites using NeuroHarmonize [3], and finally parcellated using the Destrieux anatomical atlas. Intracranial volume (ICV) was also calculated and harmonized across sites.

Using the R package mgcv, we ran generalized additive models (GAM) to test for correlations between region-wise morphometric measures and raw reading scores. Linear covariates included sex, image quality (coefficient of joint variation), and for some models intracranial volume (if the morphometric measure was GMV or SA). Age was included as a smooth regressor, given the wide age range of participants. Separate models were run to test for associations between ICV and reading scores.

Several anatomical parcels across both hemispheres exhibited significant and positive associations between reading scores and both SA and GMV, (p < 0.05; FDR corrected across regions), but not CT (Figures 1 and 2). However, despite statistical significance, the effect sizes (unique adjusted R-squared coefficient attributed to the reading score in predicting the brain metric) were modest, never exceeding 0.01, and some effect sizes were negative. In models relating ICV to reading scores, there were similarly positive statistically significant associations (p < 0.05, even after controlling for non-verbal IQ), also with only modest effect sizes.

Despite statistical significant relationships between reading skills and brain morphometry on global and local scales, limited effect sizes confound interpretability of these associations. Our findings suggest variability in reading abilities may not meaningfully contribute to predicting brain morphometry when compared to other variables in the models. This study reinforces earlier findings indicating a statistical link between morphometry and reading performance [1] [4], and advances this understanding by using a large quality-controlled sample of children and adolescents. However, these findings also underscore concerns of the limitations of cross-sectional MRI models in capturing brain-behavior relationships in reading.

Reading and Writing ¹

Early life, Adolescence, Aging ²

Neuroanatomy Other

Learning

MRI

Open Data

PEDIATRIC

STRUCTURAL MRI