Transcriptome-wide association analysis of brain structures yields insights into pleiotropy with complex neuropsychiatric traits

Abstract

Structural variations of the human brain are heritable and highly polygenic traits, with hundreds of associated genes identified in recent genome-wide association studies (GWAS). Transcriptome-wide association studies (TWAS) can both prioritize these GWAS findings and also identify additional gene-trait associations. Here we perform cross-tissue TWAS analysis of 211 structural neuroimaging and discover 278 associated genes exceeding Bonferroni significance threshold of 1.04 × 10^-8. The TWAS-significant genes for brain structures have been linked to a wide range of complex traits in different domains. Through TWAS gene-based polygenic risk scores (PRS) prediction, we find that TWAS PRS gains substantial power in association analysis compared to conventional variant-based GWAS PRS, and up to 6.97% of phenotypic variance (p-value = 7.56 × 10^-31) can be explained in independent testing data sets. In conclusion, our study illustrates that TWAS can be a powerful supplement to traditional GWAS in imaging genetics studies for gene discovery-validation, genetic co-architecture analysis, and polygenic risk prediction.

Fig. 1. Selected significant gene-trait associations discovered in UKB (UK Biobank) cross-tissue TWAS analysis of 211 neuroimaging traits (n = 19,629 subjects for ROI volumes and 17,706 for DTI parameters).

The gene-level associations were estimated and tested by the cross-tissue UTMOST approach (https://github.com/Joker-Jerome/UTMOST). We used the p value threshold of 1.04 × 10⁻⁸, corresponding to adjusting for testing 211 imaging phenotypes with the Bonferroni correction. The x axis provides the IDs of the neuroimaging traits, and the y axis lists the detected genes in TWAS. The additional (UTMOST new) and previously reported GWAS-significant associations (MAGMA, FUMA, and FUMA&MAGMA) were labeled with different colors (orange, purple, green, and red, respectively).

Fig. 2. Cross-tissue TWAS-significant genes of neuroimaging traits (n = 19,629 subjects for ROI volumes and 17,706 for DTI parameters) that have been linked to other complex traits in previous GWAS.

For each of the cross-tissue TWAS-significant genes listed in the x axis, we manually checked the previously reported associations on the NHGRI-EBI GWAS catalog (https://www.ebi.ac.uk/gwas/). The genes associated with DTI parameters (DTI), ROI volumes (volume), and both of them (Both) were labeled with three different colors (blue, orange, and green, respectively).

Fig. 3. Overlapping cross-tissue TWAS-significant genes between neuroimaging traits (n = 19,629 subjects for ROI volumes and 17,706 for DTI parameters) and other complex traits and clinical outcomes.

The gene-level associations were estimated and tested by the cross-tissue UTMOST approach (https://github.com/Joker-Jerome/UTMOST). We adjusted for testing 211 neuroimaging traits (p value threshold 1.04 × 10⁻⁸) and 16 other traits (p value threshold 1.37 × 10⁻⁷) with the Bonferroni correction, respectively. The x axis provides the IDs of the neuroimaging traits. The y axis lists the 16 other traits, and Supplementary Data 15 details the resources of their GWAS summary statistics and the sample sizes of corresponding studies.

Fig. 4. Prediction accuracy (incremental R²) of gene-based polygenic risk scores constructed by UKB TWAS results (n = 19,629 subjects) on the four independent data sets.

The x axis lists the four independent cohorts (ADNI, HCP, PING, and PNC) and the y axis lists the ROI volumes. The displayed numbers are the proportions of phenotypic variation that can be additionally explained by UKB TWAS-derived gene-based PRS.

Fig. 5. Prediction accuracy (incremental R²) of gene-based polygenic risk scores constructed by UKB-derived TWAS summary statistics (TWAS PRS), variant-based PRS constructed by UKB-derived GWAS summary statistics (GWAS PRS), and both of them (GWAS PRS + TWAS PRS) on the four independent data sets (n = 19,629 subjects).

The x axis lists 28 ROI volumes whose TWAS PRS are significant in all the four data sets after the Bonferroni correction and the y axis lists the proportions of phenotypic variation that can be additionally explained by PRS.