Introduction
Hi, Welcome to the NeuroimaGene project!
The goal of this project is to improve the interpretability of genetic studies relating to neurological and/or psychiatric traits. Analyses such as genome-wide association studies often highlight genetic variants that segregate with disease. Methods such as fine mapping, eQTL colocalization, and transcriptome-wide analyses (TWAS) have been developed to identify the gene level targets of these trait-associated variants. Once a set of gene targets has been identified from GWAS findings, it often remains challenging to identify the biological consequences of variation in these genes. This is the motivation behind neuroimaGene.
Figure 1: Visual abstract for the NeuroimaGene project
NeuroimaGene is structured as a searchable repository of associations between neuroimaging features and predicted gene expression measures. Given a list of genes (eg. those associated with Schizophrenia in a recent TWAS) the user can query NeuroimaGene to find neuroimaging measures that are significantly associated with the predicted expression of those genes. This allows the user to translate a list of gene transcripts that are associated with disease to a set of neuroimaging measures. Typically, this neuroimaging is more proximal to the symptoms that patients experience.
How does it work?
For a full explanation, please see our paper, “A Transcriptomic Atlas of the Human Brain Reveals Genetically Determined Aspects of Neuropsychiatric Health.” Bledsoe, 2024 A quick overview is as follows.
The UK bio bank conducted both genetic sequencing and neuroimaging studies of ~40,000 individuals. Relying on these scans, they characterized 3,935 neuroimaging features in each patient. We refer to these as neuroimaging derived phenotypes (NIDPs). NIDPs include structural measures (T1), connectivity measures (dMRI), coactivation measures (fMRI/tfMRI), and more. (see https://www.fmrib.ox.ac.uk/ukbiobank/index.html). Using the set of 33,000 unrelated individuals, Smith et al performed quantitative genome wide association studies for each of the 3,935 NIDPs. Of these, approximately 3,500 had significant SNP based heritability (Smith, 2021).
We then conduct Joint Tissue Imputation (JTI) informed Transcriptome Wide Association Studies (TWAS) via summary statistics based S-PrediXcan for the >3,500 NIDP GWAS studies (Zhou, 2020). We identify genetically regulated gene expression (GReX) associated with neurological measures observed on MRI imaging. GReX is predicted in 19 different tissue contexts using JTI expression models from 13 neurological 6 neuro-adjacent tissues. The patients comprising the UKB neuroimaging study are 40-69 and were screened for overt neurological pathology. They generally represent an adult population without neurological disease.
As such, NeuroimaGene catalogs the neurological consequences of lifelong exposure to increases or decreases in gene expression.
An example association is as follows:
| gene | gene_name | gwas_phenotype | training_model | zscore | pvalue |
|---|---|---|---|---|---|
| ENSG00000147894 | C9orf72 | AmygNuclei_lh_volume_Basal-nucleus | JTI_Brain_Amygdala | 2.547560 | 0.0108479252 |
For the association data described in the table above, the user would read:
Differential genetically regulated expression of C9orf72 in the amygdala is positively associated with the volume of the basal nucleus of the left amygdala (uncorrected pvalue = 0.011).
Outputs:
gene: the Ensmble Gene ID
gene_name: the HUGO gene name
gwas_phenotype: the Neuroimaging Derived Phenotype as detailed by the UKB neuroimaging GWAS
training_model: the JTI-enriched tissue specific eQTL model in which the association is found
zscore: The normalized effect size across all tested GRex-NIDP-tissue associations (most appropriate for comparison)
pvalue: the uncorrected nomiunal pvalue
mod_BHpval: the modality-specific Benjamini-Hochberg false-discovery rate corrected association pvalue
mod_BFpval: the modality-specific Bonferroni corrected association pvalue
atl_BHpval: the atlas-specific Benjamini-Hochberg false-discovery rate corrected association pvalue
atl_BFpval: the atlas-specific Bonferroni corrected association pvalue
