Population clustering of structural brain aging and its association with brain development

Abstract

Structural brain aging has demonstrated strong inter-individual heterogeneity and mirroring patterns with brain development. However, due to the lack of large-scale longitudinal neuroimaging studies, most of the existing research focused on the cross-sectional changes of brain aging. In this investigation, we present a data-driven approach that incorporate both cross-sectional changes and longitudinal trajectories of structural brain aging and identified two brain aging patterns among 37,013 healthy participants from UK Biobank. Participants with accelerated brain aging also demonstrated accelerated biological aging, cognitive decline and increased genetic susceptibilities to major neuropsychiatric disorders. Further, by integrating longitudinal neuroimaging studies from a multi-center adolescent cohort, we validated the 'last in, first out' mirroring hypothesis and identified brain regions with manifested mirroring patterns between brain aging and brain development. Genomic analyses revealed risk loci and genes contributing to accelerated brain aging and delayed brain development, providing molecular basis for elucidating the biological mechanisms underlying brain aging and related disorders.

Keywords: MRI; adolescence; genetics; genomics; human; longitudinal analysis; neuroscience; structural brain aging.

Figure 1.. Overview of the study workflow.

(a) Population cohorts (UK Biobank and IMAGEN) and data sources (brain imaging, biological aging biomarkers, cognitive functions, genomic data) involved in this study. (b) Brain aging patterns were identified using longitudinal trajectories of the whole brain GMV, which enabled the capturing of long-term and individualized variations compared to only use cross-sectional data, and associations between brain aging patterns and other measurements (biological aging, cognitive functions and PRS of major neuropsychiatric disorders) were investigated. (c) Mirroring patterns between brain aging and brain development was investigated using z-transformed brain volumetric change map and gene expression analysis.

Figure 2.. Global (a) and selected regional (b, c) cortical gray matter volume rate of change among participants with brain aging patterns 1 (red) and 2 (blue).

Rates of volumetric change for total gray matter and each ROI were estimated using GAMM, which incorporates both cross-sectional between-subject variation and longitudinal within-subject variation from 40,921 observations and 37,013 participants. Covariates include sex, assessment center, handedness, ethnic, and ICV. Shaded areas around the fit line denotes 95% CI.

Figure 3.. Distributions of biological aging biomarkers (leucocyte telomere length (LTL) and PhenoAge) among participants with brain aging patterns 1 and 2.

Boxes represent the interquartile range (IQR), lines within the boxes indicate the median. Two-sided p values were obtained by comparing LTL or PhenoAge Levine et al., 2018 between brain aging patterns using unadjusted multivariate linear regression models. Results remained significant when adjusting for sex, age, ethnic, BMI, smoking status and alcohol intake frequency in the LTL model Demanelis et al., 2020 and sex, age, ethnic, BMI, smoking status, alcohol frequency and education years in PhenoAge model. Stars indicate statistical significance after Bonferroni correction. ****: p ≤ 0.0001, *: p ≤ 0.05.

Figure 4.. Effect size (Cohen’s D or odds ratio) for comparing the cognitive functions between participants with brain aging patterns 1 and 2.

Results were adjusted such that negative Cohen’s D and Odds Ratio less than 1 indicate worse cognitive performances in brain aging pattern 2 compared to pattern 1. Width of the lines extending from the center point represent 95% confidence interval. Two-sided p values were obtained using both unadjusted and adjusted (for sex, age, and TDI, education and income) multivariate regression models. Stars indicate statistical significance after FDR correction for 11 comparisons. ****: p ≤ 0.0001, ***: p ≤ 0.001, **: p ≤ 0.01, ns: p>0.05.

Figure 5.. Participants with accelerated brain aging (brain aging pattern 2) had significantly increased genetic liability to ADHD and delayed brain development.

Polygenic risk score (PRS) for ADHD, ASD, AD, PD, BIP, MDD, SCZ and delayed brain development (unpublished GWAS) were calculated at different p-value thresholds from 0.005 to 0.5 at an interval of 0.005. Vertical axis represents negative logarithm of P values comparing PRS in brain aging pattern 2 relative to pattern 1. Red horizontal dashed line indicates FDR corrected p value of 0.05. Colors represent traits and dots within the same color represent different p value thresholds. The trigonometric symbol indicates the average PRS across all p-value thresholds for the same trait.

Figure 6.. Genome-wide association study (GWAS) identified 6 independent SNPs associated with accelerated brain aging.

Total GMV at 60 years old was estimated for each participant using mixed effect models allowing for individualized baseline GMV and GMV change rate, and was used as the phenotype in the GWAS. (a) At genome-wide significance level (p=5×10^–8, red dashed line), rs10835187 and rs7776725 loci were identified to be associated with accelerated brain aging. (b) Quantile–quantile plot showed that the most significant p values deviate from the null, suggesting that results are not unduly inflated.

Figure 7.. The ‘last in, first out’ mirroring patterns between brain development and brain aging.

(a) The annual percentage volume change (APC) was calculated for each ROI and standardized across the whole brain in adolescents (IMAGEN, left) and mid-to-late aged adults (UK Biobank, right), respectively. For adolescents, ROIs of in red indicate delayed structural brain development, while for mid-to-late aged adults, ROIs in blue indicate accelerated structural brain aging. (b) Estimated APC in brain development versus early aging (55 years old, left), and versus late aging (75 years old, right). ROIs in red indicate faster GMV decrease during brain aging and slower GMV decrease during brain development, that is stronger mirroring effects between brain development and brain aging. (c) Mirroring patterns between brain development and brain aging were more manifested in participants with accelerated aging (brain aging pattern 2). The arrows point to ROIs with more pronounced mirroring patterns in each subfigure.

Figure 8.. Functional enrichment of gene transcripts significantly associated with delayed brain development and accelerated brain aging.

(a) 990 genes were spatially correlated with the first PLS component of delayed structural brain development, and were enriched for trans-synaptic signal regulation, forebrain development, signal release and cAMP signaling pathway. (b) 2293 genes were spatially correlated the first PLS component of accelerated structural brain aging, and were enriched for macroautophagy, pathways of neurodegeneration, establishment of protein localization to organelle and histone modification. Size of the circle represents number of genes in each term and P values were corrected using FDR for multiple comparisons.

Appendix 1—figure 1.. The sample selection workflow.

Appendix 1—figure 2.. Estimated rates of change in regional volumes for 33 bilateral brain regions.

Appendix 1—figure 3.. Stratification of the identified brain aging patterns using linear and non-linear dimensionality reduction methods.

Appendix 1—figure 4.. Effect size for comparing each individual blood biochemical metric (used to calculate the PhenoAge) between participants with brain aging patterns 1 and 2.

Appendix 1—figure 5.. Gene set enrichment of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and gene ontology (GO) of biological processes.

Appendix 1—figure 6.. Optimal number of clusters was chosen using elbow method (a) and silhouette method (b).

Author response image 1.. Stratification of the identified brain aging patterns using linear and non-linear dimensionality reduction methods.

(a) The principal component space of PC1 and PC2, and (b) two-dimensional projected locally linear embedding space derived from brain volumetric measures. Points have been colored and shaped according to grouping labels of the brain aging patterns.

Author response image 2.. Total gray matter volume (TGMV) (a) and the estimated rate of change (b) for females (red) and males (blue).

Rates of volumetric change for total gray matter and each ROI were estimated using GAMM, which incorporates both cross-sectional between-subject variation and longitudinal withinsubject variation from 22,067 observations for 19,958 females, and 18,854 observations for 17,055 males. Covariates include assessment center, handedness, ethnic, and ICV. Shaded areas around the fit line denotes 95% CI.

Author response image 3.. Distributions of biological aging biomarkers (leucocyte telomere length (LTL) and PhenoAge) among participants with brain aging patterns 1 and 2.

Author response image 4.. The “last in, first out” mirroring patterns between brain development and brain aging.