The subjects for this research were drawn from five imaging datasets: the BCP49, various groups from the Human Connectome Project (HCP) encompassing development, young adults, and older adults (HCP-D, HCP-YA, HCP-A), along with the Healthy Brain Network (HBN) dataset. The ages of the participants in the BCP varied from just 16 days to 6 years. Initially, 557 participants were scanned at 1,095 separate time points before undergoing gradient-based quality checks. Ultimately, the study analyzed data from 652 participants (ages 5.6 to 21.9) from HCP-D, 1,206 (ages 22 to 37) from HCP-YA, 725 (ages 36 to 100) from HCP-A, and 772 (ages 5.6 to 21.9) from HBN, totaling 3,912 individuals across all cohorts.
On the ethics front, all utilized data stemmed from publicly accessible studies and received necessary ethics approval from local review boards, along with appropriate consent from participants. The current analysis focused exclusively on de-identified data in compliance with data usage agreements.
Quality control measures led to the exclusion of data from infants that exhibited excessive motion, reducing the BCP cohort to 343 individuals across 760 time points. Additional data filtering procedures, employing visual inspections and clustering methods, ensured the integrity of the dataset. By the end, 3,972 unique gradient sets from 3,556 individuals formed the basis of our final analysis.
For preprocessing functional MRI (fMRI) data, we adhered to a pipeline aligned with HCP standards. This procedure encompassed various stages, including correction for head motion, distortion correction, and registration of images. After denoising, independent component analysis was employed to remove any residual motion artifacts.
Structural data underwent quality assessments, alongside segmentation into white matter, grey matter, and cerebrospinal fluid. Additionally, diffusion MRI data were corrected for various distortions and underwent co-registration with structural MRI. A detailed aging coordinate system was established using lifespan surface atlases, allowing for comprehensive analyses across different age groups.
Functional connectivity matrices were computed based on fMRI data, leading to the generation of several connectivity gradients. By aligning individual-specific gradient axes across the lifespan, we utilized weighted principal component analysis to create template gradients that accurately represented connectivity variations.
Structural gradients were also calculated to evaluate the coupling between structural and functional elements throughout development. This involved computing correlation matrices from various microstructural features, allowing for a deep dive into structure-function relationships.
Throughout this project, we employed generalized additive mixed models (GAMMs) to chart lifespan trajectories for cortical gradients. These models adeptly captured nonlinear age-related changes, providing stable fits across a broad age range. For the analysis of gradient values, we accounted for potential confounding factors, including arousal states and cohort influences.
To assess cognitive functions, the NIH Toolbox Cognition Battery was used, analyzing several unadjusted cognitive scores in relation to functional connectivity metrics. For different age groups, we conducted various statistical models to see how gradient metrics influenced cognitive performance, and these metrics were further examined across the lifespan.
Mullen scales of early learning were analyzed for infants to determine any variations explained by FC gradient metrics. Finally, we validated findings through transcriptomic enrichment analyses and various visual aids, aiming to present a cohesive picture of brain development across the lifespan.
