A transcriptomic atlas of the human brain reveals genetically determined aspects of neuropsychiatric health

Xavier Bledsoe; Eric R Gamazon

doi:10.1016/j.ajhg.2024.06.002

. 2024 Jun 25;111(8):1559–1572. doi: 10.1016/j.ajhg.2024.06.002

A transcriptomic atlas of the human brain reveals genetically determined aspects of neuropsychiatric health

Xavier Bledsoe ^1,^2,^∗, Eric R Gamazon ^2,^3,^∗∗

PMCID: PMC11339608 PMID: 38925120

Summary

Regulation of gene expression is a vital component of neurological homeostasis. Cataloging the consequences of endogenous gene expression on the physical structure and connectivity of the brain offers a means of unifying trait-associated genetic variation with trait-associated neurological features. We perform tissue-specific transcriptome-wide association studies (TWASs) on over 3,400 neuroimaging phenotypes in the UK Biobank (N = 33,224) using our joint-tissue imputation (JTI)-TWAS method. We identify highly significant associations between predicted expression for 7,192 genes and a wide variety of measures of the brain derived from magnetic resonance imaging (MRI). Our approach generates reproducible results in internal and external replication datasets. Genetically determined expression alone is sufficient for high-fidelity reconstruction of brain structure and organization. We demonstrate complementary benefits of cross-tissue and single-tissue analyses toward an integrated neurobiology and provide evidence that gene expression outside the central nervous system provides unique insights into brain health. As an application, we provide evidence suggesting that the genetically regulated expression of schizophrenia risk genes causally affects over 73% of neurological phenotypes that are altered in individuals with schizophrenia (as identified by neuroimaging studies). Imaging features associated with neuropsychiatric traits can provide valuable insights into underlying pathophysiology. By linking neuroimaging-derived phenotypes with expression levels of specific genes, this resource represents a powerful gene prioritization schema that can improve our understanding of brain function, development, and disease. The use of multiple different cortical and subcortical atlases in the resource facilitates direct integration of these data with findings from a diverse range of clinical neuroimaging studies.

Graphical abstract

Here, we present NeuroimaGene, a searchable atlas detailing associations between genetically predicted gene expression and over 3,400 measures of the human brain. Gene expression in a tissue is predicted according to genetic data. We demonstrate that NeuroimaGene can identify schizophrenia-associated brain regions from genetic data alone.

Introduction

Several transcriptome-wide association studies (TWASs) have identified associations between genetically regulated gene expression (GReX) and neuropsychiatric traits. The contribution of GReX variation to disease pathophysiology remains an active area of research, with critical implications for our understanding of disease mechanisms.¹^,²^,³ Toward this end, we develop a publicly available resource, NeuroimaGene, for identifying the neurophysiological consequences of variation in GReX. We link 2 well-validated intermediate phenotypes: endogenous gene expression under genetic control and neurophysiology as captured by a broad array of neuroimaging-derived phenotypes (NIDPs) from magnetic resonance imaging (MRI) (Figure 1A).⁴^,⁵^,⁶^,⁷^,⁸

NeuroimaGene resource overview

(A) Conceptual overview of the neuroimaging TWAS framework. To create the NeuroimaGene resource, we took as input data >3,400 UKB neuroimaging GWASs. Using S-PrediXcan applied to the JTI models and the GWAS summary statistics, we quantify the magnitude, direction, and significance of the association between GReX and the NIDP for each GWAS. This analysis results in tissue-specific associations between GReX and NIDPs.

(B) In relation to brain morphology (trait), GReX can be used to quantify the proportion of SNP-based variation that is mediated by endogenous gene expression. Note that GReX may explain only a proportion of the SNP-based heritability for any given trait.

(C) GReX and NIDPs are potential endophenotypes mediating the effect of genetic variation on neuropsychiatric risk.

(D) Distribution of significant (FDR <0.05) NeuroimaGene associations across all 19 JTI models. The total number of GReX-NIDP associations identified in each tissue is represented in pale colors on the left y axis. The right y axis shows the number of unique genes involved in those associations (which may involve multiple NIDPs). These gene counts are detailed in dark colors. Blue bars represent associations from JTI models derived from brain tissues, while green bars represent associations from JTI models derived from non-brain tissues.

(E) Distribution of significant (FDR <0.05) GReX-NIDP associations across all JTI models grouped by the imaging modality used to capture each NIDP. The proportion of significant associations for each modality is given in parentheses, while the bars themselves represent the raw number of associations belonging to each modality.

(F) Manhattan style plot of all associations between GReX and NIDPs across the genome from the application of JTI models in 19 tissues. The x axis represents the linearized genome, and the y axis represents the negative log₁₀ of the Benjamini-Hochberg adjusted p value. Each point represents a single GReX-NIDP association and is colored according to the JTI model in which the association was calculated. GReX, genetically regulated gene expression; SNP, single-nucleotide polymorphism; TWAS, transcriptome-wide association study; GWAS, genome-wide association study; GTEx, Genotype Tissue Expression consortium; NIDP, neuroimaging-derived phenotype; JTI, joint-tissue imputation.

GReX characterizes the component of endogenous gene expression that is driven by germline variation in nucleotide sequence. This measure of genetic influence can be used to illuminate the biological mechanisms of neuropsychiatric disease. GReX has been associated with a broad range of neuropsychiatric traits.⁹ By quantifying the component of gene expression under genetic control, this measure is to be distinguished from gene expression changes that result from environmental and trait-determined influences (Figure 1B).⁸ Because multiple SNPs often affect regulation of a gene, this transcription-targeted approach improves statistical power relative to single-SNP analyses by concentrating multiple SNP effects onto relatively fewer, biologically informative (gene) targets.¹⁰

In this study, we leverage the UK Biobank (UKB) neuroimaging data to identify associations between variation in GReX and over 3,400 NIDPs. These quantitative descriptors of the structure, connectivity, and functional activation of the brain have been associated with neuropsychiatric health (Figure 1C).¹¹^,¹²^,¹³^,¹⁴^,¹⁵ The UKB’s application of automated image-processing pipelines to MRI permits standardized measurements for an enormous number of cerebral data points.¹⁶^,¹⁷^,¹⁸^,¹⁹ These data capture structural information, such as the volume, thickness, and surface area of brain regions; organizational variation describing the white matter connections between these components; and functional information, such as networks of synaptic activity associated with specific tasks or resting states.¹⁵^,¹⁶^,¹⁷^,¹⁸^,²⁰

Through analysis of GReX in a population-based cohort, NeuroimaGene captures the expression-mediated genetic architecture of neurological structure and function, facilitating the identification of specific genes and anatomic measurements involved in neurocognitive homeostasis.²¹ In our application, we replicate and extend evidence characterizing the neurogenic pathophysiology of schizophrenia (SCZ).

Material and methods

Data acquisition

The UKB is a large-scale prospective cohort study, which has collected genotypic and phenotypic data from approximately 500,000 residents of the United Kingdom. The standardization of data collection methods across participants and the large sample size of this study provides enormous advantages for epidemiological analysis. In 2014, the UKB began collecting neuroimaging data from study participants in the form of functional, diffusion, and structural MRI.⁷ The study developed an automated image processing pipeline to standardize and quantify its findings, generating 3,935 “image-derived phenotypes.”²⁰ In 2016, Elliot et al. published a genome-wide association study (GWAS) for each of the 3,144 image-derived phenotypes using genetic information from 8,428 UKB individuals.¹⁷ In 2021, an additional release of neuroimaging GWAS data was made available detailing associations for 3,935 NIDPs as measured in 39,691 brain-imaged samples derived from 33,224 individuals of European ancestry.¹⁶ The summary statistics results for these GWASs are publicly available at the UKB website (https://open.win.ox.ac.uk/ukbiobank/big40/).

Statistical analysis

All analyses are conducted in R version 4.0.5, Python 3.8.6, and Linux.

Model building and JTI framework

We leveraged pre-built gene expression models trained on RNA data from postmortem samples in order to predict gene expression from SNP profiles. The construction of these models is described in Zhou et al.⁵ The GTEx consortium maintains a repository of RNA expression data and genetic data collected from 54 non-diseased tissue sites across approximately 1,000 individuals.²² Processed GTEx data (for example, gene expression and expression quantitative trait loci [eQTLs]) are available from the GTEx portal (https://gtexportal.org). As described in Zhou et al., each joint-tissue imputation (JTI) model quantifies the association between SNPs and gene expression levels across the genome for each tissue. The GReX is highly shared across tissues, and JTI exploits this shared regulatory architecture to substantially improve expression prediction. These models, trained on the GTEx data, are publicly available at Zenodo:3842289 (https://doi.org/10.5281/zenodo.3842289).

TWASs of neuroimaging phenotypes

We applied JTI models derived from 19 GTEx (v.8) tissues to UKB GWAS summary statistics for all 3,935 NIDPs published by Smith et al. to identify GReX associations with NIDPs. We used S-PrediXcan, which generates the TWAS summary statistics.⁴ S-PrediXcan is included in the MultiXcan package. Code for MultiXcan is available on the Im lab Github repository (https://github.com/hakyimlab/MetaXcan). The methodology incorporates variance and covariance data of SNPs according to a linkage disequilibrium (LD) reference panel.

Multiple-testing correction in discovery cohort

Our study analyzed 22,815 genes and 19 GTEx-derived JTI models across 3,935 NIDPs. We adjusted for the multiple-hypothesis testing using the Benjamini-Hochberg false discovery rate (FDR) threshold of 0.05. For all analyses involving significance testing, we considered only the subset of NIDPs with significant SNP-mediated heritability (3,546).¹⁶

Gene set analysis

We conducted Gene Ontology (GO) term enrichment analyses using the ClusterProfiler R tool.²³ For all GO analyses, we set the size range for gene sets as 15–500 and used a Benjamini Hochberg p value cutoff of 0.05 for filtering of pathway associations. The Cellular Components library of GO was used.

Hierarchical clustering of NIDPs using nominally significant GReX associations

We estimated the correlations in TWAS associations among the NIDPs. We first stratified the TWAS associations according to the 19 JTI models. Each JTI model was represented by a matrix detailing the normalized effect size (Z score) of each gene on each NIDP. Each matrix consists of all genes in the JTI model and 3,934 columns (NIDPs). Each data point is the normalized effect size (Z score) of GReX on the NIDP. Instances where specific genes were not associated with certain NIDPs were coded as “NA.” Concerning data completeness, we used a 99% completion threshold for inclusion of NIDPs in the clustering analyses. No phenotypes failed thresholding. Using the matrix for each JTI model, we calculated the pairwise Spearman correlation for all NIDP pairs according to the TWAS Z scores.

After identifying the Spearman correlation distance between all NIDPs according to the Z scores from the GReX association, we reproduced the data in the form of a heatmap with axes ordered by Ward’s agglomerative hierarchical clustering method. The NIDPs involved in this ordering can be categorized according to a variety of criteria, including imaging modality, atlas used, lobe of the brain (for cortical measures), direction of fibers (for white matter tractography), left or right hemisphere, etc. Accordingly, we annotated each NIDP according to the multi-dimensional phenotypic descriptors provided by the UKB (Table S1). These annotations allow for visual examination of cluster patterns. Using the R package complexheatmap, we created annotated, clustered heatmaps for all the phenotypes according to Z scores from each JTI model.²⁴

UMAP dimensionality reduction of T1 and dMRI phenotypes

We applied the uniform manifold approximation and projection (UMAP) for dimension reduction package in R to improve the visual interpretability of clustering analyses for 2 subsets of GReX-NIDP associations.²⁵ For each of the 19 JTI models, we conducted UMAP analysis on the GReX measures nominally associated with (1) T1-weighted cortical and subcortical surface area/volume measures and (2) diffusion MRI (dMRI) tractography measures. We optimized the parameter values for n_neighbors (number of approximate nearest neighbors used to construct the initial high-dimensional graph) and min_dist (minimum distance between points in low-dimensional space) in order to prioritize global relationships over local similarities. This was performed through iterative analyses of each data subset across multiple combinations of the 2 parameter values and visual evaluation of cluster projection. Final parameter sets were selected based on those that generated the most interpretable separation and clustering between NIDPs according to annotations of side, region, and type of measurement. For the T1 dataset, we selected n_neighbors = 150 and min_dist = 0.09. For the dMRI dataset, these values were n_neighbors = 60 and min_dist = 0.1. We set initial seed values for the R session as well as for the UMAP internal functionality to improve the reproducibility of the analysis.

PCA dimensionality reduction of T1 phenotypes

We conducted principal component analysis of the T1 structural NIDPs in R according to the normalized effect size of each GReX-NIDP association. Pooling data from all 19 tested JTI models, we displayed the first 2 principal components of the data. We annotated the NIDPs by color with variation representing the type of cortical measure (surface area, thickness, or volume.)

Internal replication analysis for neuroimaging TWASs

The UKB generated neuroimaging GWAS data from a cohort of approximately 33,000 participants. In addition to this large dataset, the UKB released 2 mirror versions following the same analytic pipeline but using 2 independent subsets of the populations. These subsets had sample sizes of 11,000 and 22,000. To test the replication of the neuroimaging TWAS methodology, we replicated our initial TWASs in these 2 smaller datasets for independent comparison. We applied the 19 JTI models to the 11K dataset and the 22K dataset independently. In order to compare these 2 TWAS datasets, we selected all associations from both studies that surpassed a Benjamini Hochberg p value threshold of 0.05. From these data, we selected the GReX-NIDP associations that are present in both the 11K and 22K cohorts (matched on JTI models). These data represent TWAS associations between GReX and NIDPs derived from independent cohorts curated by the UKB. We calculated the (non-parametric) Spearman correlation as effect sizes were not normally distributed.

Using the MRI modalities described in the neuroimaging GWAS supplement, we assessed the distribution of all shared associations as well as the mean Spearman correlation between the 11K cohort and the 22K cohort for each NIDP. We then took the mean Spearman correlation value of all NIDPs in each JTI model category and performed a linear regression against the number of genes analyzed in each JTI model. We repeated these analyses, using the subset of data that surpassed a Bonferroni corrected p value threshold of 0.05.

External replication analysis of neuroimaging TWAS in ENIGMA

The Enhancing Neuroimaging Genetics through Meta-Analysis (ENIGMA) consortium has published a GWAS meta-analysis of cortical measures derived from 51,665 individuals.¹⁹ Of note, 10,083 individuals from the UKB were included in the ENIGMA analysis. The phenotypic measures include the surface area and thickness of the whole cortex as well as 34 cortical regions as defined by the Desikan cortical atlas.²⁶ Summary statistics for the ENIGMA dataset are available upon request from their website (https://enigma.ini.usc.edu/research/download-enigma-gwas-results/).We accessed the version of the data describing regional surface area and thickness that had been adjusted for the respective global measures and undergone genomic control adjustment to account for population stratification across the many different participating sites in ENIGMA. We then applied S-PrediXcan to the subset of ENIGMA NIDPs that demonstrated significant SNP-based heritability from the initial GWAS. We used the same 19 tissue-specific JTI models described for the UKB neuroimaging TWAS. We filtered these GReX-NIDP tissue associations according to a minimum r² of 0.1.

The ENIGMA consortium consolidated the left and right hemispheric regions into single phenotypes prior to performing the GWAS analysis. To best match the phenotypes between the 2 studies, we first identified all the TWAS associations from the UKB neuroimaging study that matched the heritable cortical area and thickness measurements described in ENIGMA irrespective of hemisphere (Table S5). Then, for NIDPs that were captured bilaterally in the UKB but not in the ENIGMA study, we selected genes whose GReX was significantly associated with the NIDP of interest in both the right and left hemisphere. This resulted in a set of 1,736 hemisphere-agnostic, tissue-specific GReX-NIDP associations for the Desikan cortical regions derived from the UKB data. Having identified the UKB associations to be replicated, we used the Bonferroni multiple-testing correction to ensure family-wise error rate (FWER) control (at α level of 0.05). The resulting p value threshold used for the ENIGMA replication analysis was 0.05/1736 = 2.880184e−05.

SCZ-associated GReX curation

We identified GReX associated with SCZ using the most recent major SCZ GWAS meta-analysis from Trubetskoy et al.²⁷ This extended GWAS consists of the core Psychiatric Genomics Consortium (PGC) dataset, an African American dataset, a Latino dataset, and the deCODE SCZ cohorts. Sample details are provided in the supplementary information for the initial publication. Trubetskoy et al. performed secondary analyses on the findings from the extended GWAS using individuals from the core PGC cohort owing to accessibility of individual-level data. Of relevance to this work, they performed summary Mendelian randomization (SMR) on the extended GWAS findings to highlight 101 unique genes associated with SCZ through genetically regulated expression. For TWAS models, they leveraged those derived from PsychENCODE, eQTLGen, and fetal brain data. For the sake of uniformity in data analysis, we selected the subset of 88 genes identified via SMR using the PsychENCODE data. These data included GReX associations from SNP models trained on brain and whole blood. Only the GReX from the whole-blood analysis that were also replicated in brain models were reported by Trubetskoy et al. for the sake of prioritizing neurological relevance. Of these 88 GReX, 63 could be mapped to unique Ensembl IDs, 50 of which were present in the NeuroimaGene resource.

SCZ NIDP curation

SCZ NIDPs were derived from 2 studies published by the ENIGMA consortium, 1 highlighting cortical regions and the other subcortical regions.²⁸^,²⁹ In assessing cortical changes associated with SCZ, van Erp et al. analyzed participant data from 39 different centers across the globe. These data represent 4,474 individuals with SCZ and 5,098 healthy controls, all of whom were assessed using a standardized approach. Cohort details and analysis methods were described in the methods section of the initial publication. Of relevance to this work, all imaging data in the study represent T1-weighted structural brain scans that were processed using the Desikan-Killiany (DK) cortical atlas in Freesurfer.²⁶ The study used univariate linear regression to identify group differences in the thickness and cortical area of the measured DK-atlas regions. Random-effects meta-analyses of both Cohen’s d and the partial correlation effect sizes were performed for each DK atlas region. Multiple-testing correction was performed using the FDR with a threshold of 0.05. Additional details on confounding and covariate analyses were included in the original publication. Following meta-analysis and multiple-testing correction, 33 hemisphere-specific cortical surface area and thickness measures had statistically significant differences between individuals with SCZ and healthy controls. Of the NIDPs reported by the UKB, we matched all 33 of these according to Freesurfer parcellation of T1-weighted structural imaging using the DK atlas.

Maximally discriminatory GReX threshold determination

In the style of polygenic risk score analyses, we assessed the GReX-NIDP significance threshold at which the data in NeuroimaGene was maximally predictive of SCZ risk. We identified all NeuroimaGene GReX associations corresponding to the NIDPs and GReX from the SCZ studies carried out by the ENIGMA consortium and Trubetskoy et al., respectively. These represent associations between GReX measures and NIDPs that have each been associated with SCZ in independent studies. Beginning with a nominal p value threshold of 0.05, we assessed the total number of associations that satisfied the threshold as well as the number of unique NIDPs represented by the association set. These represent 2 observed statistics: the number of SCZ GReX-NIDP associations that surpassed the significance threshold and the number of unique NIDPs in the SCZ GReX-NIDP associations that surpassed the significance threshold.

We then performed permutation testing to establish null distributions against which these statistics can be tested. To generate the null distribution, we selected 50 random genes (matching by 50 SCZ GReX measures that entered the experimental analysis). We then identified all NeuroimaGene associations that involve any of the 50 GReX measures and SCZ NIDPs at a p value less than or equal to 0.05. Similarly to the experimental gene set, we then calculated null statistics detailing the total number of associations and the total number of unique NIDPs. We thus quantified the number of associations between SCZ NIDPs and random gene sets for comparison with associations between SCZ NIDP associations and SCZ GReX. We repeated this procedure 1,000 times to estimate the null distribution for each of the 2 test statistics.

Using each distribution, we then identified the value below which 95% of null statistics fall. We performed this same analysis for the total number of GReX-NIDP associations and for the unique NIDPs included in each set, in line with the different null statistics. We repeated the permutation analysis at 6 different p value thresholds to identify the threshold at which the association between SCZ GReX and SCZ NIDPs was maximally greater than for random genes. From these data, we selected 0.005 as the optimal discriminatory threshold.

Mendelian randomization JTI

TWAS associations are susceptible to LD contamination. As a Mendelian randomization (MR) causal inference approach, MR-JTI models variant-level heterogeneity to obtain an unbiased estimate the gene causal effect on phenotype. As detailed in Zhou et al., the MR aspect of MR-JTI is designed to be a post-hoc test to adjust for horizontal pleiotropy and LD contamination in the subset of significant TWAS findings.⁵

We used the LiftOver script provided by the University of California, Santa Cruz (UCSC) genome browser to convert the UKB GWAS variant identifiers from Human Genome build 19 to build 38.³⁰ The command line script and the chain files required for running the UCSC LiftOver script are accessible upon request from the UCSC genome store (https://genome-store.ucsc.edu/). We then retrieved the full set of unfiltered eQTL associations from the GTEx consortium Google cloud repository.²² We intersected the retrieved unfiltered eQTL data with each of the GWAS files from the UKB NIDPs prioritized by the initial TWAS. We harmonized SNPs between the 2 studies according to the chromosomal location and the effect/reference allele. We retrieved the LD scores obtained from the 1000 Genomes Project based on the Reference SNP Identification (RSID).³¹ LD scores from the 1000 Genomes Project were downloaded from the online repository maintained by the Alkes Price group at the Broad Institute (https://alkesgroup.broadinstitute.org/LDSCORE/) in the file labeled “1000G_Phase3_ldscores.tgz.” Using the MR-JTI script, we conducted MR causal inference on these findings. Code for the JTI model training and MR-JTI causal inference analysis is hosted at the Gamazon Github repository (https://github.com/gamazonlab/MR-JTI).

Results

NeuroimaGene comprises a searchable atlas of associations between brain measures and the genetically regulated transcriptome

To identify associations between GReX and neuroimaging measures of the brain, we applied the JTI TWAS framework to summary statistics from the most recent release of the UKB neuroimaging GWAS (Figure 1A; Table 1) (material and methods).⁵ We used in silico gene expression models (hereafter “JTI models”) trained on transcriptome data in 19 tissues from the GTEx consortium. Our analysis identified associations between 3,934 measures of the brain and 22,815 imputed gene expression traits in the context of 13 brain tissues and 6 neurologically relevant tissues from outside the central nervous system (CNS) (Figure 1D).³²^,³³ We retained only those NIDPs demonstrating statistically significant SNP-based heritability. The tissue correlation of the GReX as well as non-independence of the NIDPs suggests that the conventional Bonferroni correction procedure (to ensure strong FWER control) is very conservative. Consequently, we corrected for multiple testing through a study-wide Benjamini-Hochberg FDR threshold of 0.05. The majority of the significant associations represented white matter fibers identified via dMRI and structural partitioning of the brain according to T1 imaging (Figure 1E). We provide these data as a searchable atlas detailing highly significant, quantitative associations between endogenous gene expression of >16,000 genes and >3,400 neuroimaging-derived measures of the brain (Figure 1F) (data and code availability).

Table 1.

Characteristics of the UK Biobank neuroimaging dataset

	22K subset			11K subset			Joint cohort
	female	male	total	female	male	total	female	male	total
n	11,624	10,514	22,138	5,787	5,299	11,086	17,411	15,813	33,224
Mean age	63.6 ± 7.3	65.0 ± 7.6	64.3	63.7 ± 7.4	65.0 ± 7.6	64.3	63.6	65.0	64.3
Min. age	45.1	46.1	45.1	46.3	46.1	46.1	45.1	46.1	45.1
Max age	81.8	81.8	81.8	81.6	81.0	81.6	81.8	81.8	81.8

Open in a new tab

Each “association” represents a statistically significant relationship between 1 NIDP and the GReX of a single gene as predicted by a JTI model (Figure 1C). Each NIDP-GReX association can thus achieve study-wide significance in up to 19 different tissue contexts. Several genes stand out as having significant effects on neuromorphology (Figure 1F). One such gene, syntaxin 6 (STX6), encodes a protein involved in the movement of transport vesicles to target membranes. Previous research reports that expression silencing of STX6 leads to repression of axonal outgrowth and impaired neuronal polarization in the hippocampus.³⁴ We find highly significant associations of the GReX of this gene with several NIDPs in the hippocampus as well as with a broad variety of NIDPs (Table S2).

Hierarchical clustering and dimensionality reduction of NIDPs according to GReX effect sizes recapitulate the spatial patterning of the brain

Each NIDP has its own set of GReX associations comprised of genes and their effect sizes on the NIDP measure. We investigated the extent to which unsupervised phenotype clustering of all NIDPs based solely on these GReX associations reflected known neurobiology. To this end, we performed hierarchical clustering of all NIDPs based solely on the GReX-NIDP effect sizes (material and methods, Table S4; Figures S17–S35). Of note, identical brain regions measured via different approaches clustered together, and these clusters demonstrated close clustering with groups of physically proximal brain regions (Table S4).

To better visualize the spatial patterning of the NIDPs, we next applied UMAP to the dMRI data as well as cortical surface area and volume measurements.²⁵ Prior analyses have demonstrated that empirical gene expression, as measured by RNA sequencing, carries information about the spatial organization of the brain.³⁵^,³⁶ It is not immediately clear to what extent these findings are driven by the genetically determined component of gene expression vs. environmental, context-specific determinants of gene expression. In this analysis, we specifically examined this question. We projected the NIDPs into 2-dimensional space in accordance with the similarity of GReX effect size (material and methods). Spatial grouping of NIDPs according to their transcriptomic profiles not only revealed clusters of NIDPs from the same general brain regions (Figures S10 and S11) but also produced a global pattern that closely mimics the 2-dimensional, unfolded organization of the human cortex (Figures 2 and S10). Subcortical regions clustered together near the brainstem and NIDPs describing the surface area and volume of different cortical regions self-aligned in both the rostral and longitudinal axes. While the 3-dimensional cortex folds into itself at the longitudinal fissure, projecting the cortical surface into 2 dimensions should locate the deepest aspect of the fissure near the top of the image as is observed in the cingulate gyrus and corpus callosum. Of note, the JTI models used in this analysis are themselves derived from specific brain regions and non-brain tissues. Because the dimensionality reduction was performed for each JTI model separately, the UMAP findings are due entirely to within-model variation and occur independently of spatial differences between JTI models. Classifying NIDPs exclusively by GReX effect size thus provides sufficient data to identify a sample’s place of origin relative to other NIDPs evaluated in the same individuals. The recapitulation of 2-dimensional brain structure from predicted GReX effect size alone both underscores the validity of GReX as a neurologically informative measure and highlights the richness of information contained within the NeuroimaGene resource.

GReX effects mirror cortical architecture

The spatial projection of cortical NIDPs according to GReX effect-size profiles recapitulates brain structure and organization. This uniform manifold approximation and projection (UMAP) of NIDPs corresponds to the GReX effect-size similarities of cortical surface area and volume measures as predicted by the amygdala JTI model. Similar analyses were performed for all 19 JTI models (Figures S10 and S11). The points in the cartesian plane are derived exclusively from the GReX effect-size vectors for each NIDP. NIDPs are colored according to cortical region (Table S1). Certain regions including the corpus callosum and the insula present as disparate clusters in 2-dimensional space. In both cases, these clusters appear more geographically contiguous when considering the 3-dimensional folded topology of the cortex. Shaded areas were delineated manually.

TWAS methodology for identifying neuroimaging GReX associations demonstrates high reproducibility in internal and external validation cohorts

The UKB neuroimaging GWAS release includes summary statistics of the full 33K subjects as well as 2 smaller GWASs completed on populations subsets comprised of 22K and 11K non-overlapping sets of individuals. As methodological validation, we applied the TWAS pipeline to these smaller independent datasets (material and methods). Using a Benjamini-Hochberg FDR of 0.05 cutoff, we observed 6,761 distinct GReX-NIDP associations meeting the significance threshold in both the cohorts of 11K and 22K participants. The GReX-NIDP associations showed 100% concordance of direction of effect between the cohorts with a GReX effect-size Spearman correlation of 0.98 (Figure 3A). The mean effect size of the 11K cohort was approximately 0.050, while the mean effect size in the 22K cohort was approximately 0.042, illustrating a small but upward bias in the 11K effect-size estimates (Welch’s t test p value of 0.025), likely due to the well-documented winner’s curse. The raw number of nominally significant replicating NIDPs across all JTI models in each modality category was largely skewed in favor of quality control, T1, and dMRI modalities (Figure 3B). For each NIDP modality, the mean Spearman correlation between the cohorts was consistently greater than 0.87 (Figure 3C). The mean Spearman correlation demonstrated similar consistency across all 19 JTI models (minimum mean Spearman ρ of 0.80) (Figure 3D). The number of imputable genes varied with the JTI model, owing in part to tissue sample size difference during model training. We assessed the impact of model size on effect-size correlation coefficient between the cohorts through linear regression. The resulting p value of 0.2 precluded attributing the variation in the effect-size correlation to variation in model size (Figure 3E). In summary, our pipeline identified highly replicable associations between GReX and NIDPs across 2 independent, moderately powered datasets.

NeuroimaGene methodology validation

(A) The correlation between effect sizes of tissue-matched GReX-NIDP associations shared across the neuroimaging TWAS data from 11K subjects and the neuroimaging TWAS data from 22K subjects. Each point represents a study-wide significant finding (Benjamini Hochberg false discovery rate <0.05). Data are plotted according to a pseudolog10 function with point color representing degree of overlap at each graphical coordinate. The trend line reflects linear regression. Spearman’s correlation = 0.98 with p value <2.2e−16.

(B) Two-facet plot detailing statistical characteristics of all associations (FDR <0.05) that were significant in both 11K and 22K cohorts. The left facet shows the distribution of associations grouped by the imaging modality used to capture each NIDP. The raw number of associations for each modality is included in parentheses, while the bars themselves represent the counts normalized according to the total number of NIDPs belonging to each modality. The right panel details the mean Spearman correlation across all data subsets grouped again by the imaging modality used to capture the NIDP. Error bars represent the 95% confidence interval.

(C) Vertical bar plot detailing the mean Spearman correlation across all data subsets grouped by the JTI model of the NIDP-GReX association. Error bars represent the 95% confidence interval.

(D) Scatterplot illustrating the association between mean tissue-specific correlation coefficient and the number of genes captured by each JTI model. The y axis is a linear representation of the mean Spearman correlation coefficient within each JTI model of all association effect sizes between the 11K and 22K cohorts. The x axis represents the number of genes assessed in each JTI model. Trendline reflects a linear regression of correlation coefficient against JTI model size with the adjusted r-squared and p value documented on the plot. All data in this plot are filtered according to the Benjamini Hochberg false discovery rate threshold of 0.05.

For replication of our findings in an external dataset, we applied the same TWAS pipeline used in NeuroimaGene to summary statistics from a cortical GWAS meta-analysis (N = 51,665) published by the ENIGMA consortium.¹⁹ The ENIGMA cortical GWAS identified SNP-level associations with surface area and thickness measures of 34 different cortical regions. The consortium utilized raw MRI data from over 60 cohorts and performed cortical parcellation according to the DK atlas. We selected for replication all significant associations (FDR <0.05) from the NeuroimaGene resource for the NIDPs that best corresponded to the phenotypes used in ENIGMA (material and methods). Matching these associations on JTI model, gene, and NIDP, we replicated 12.4% of associations in the ENIGMA cohort using a strict Bonferroni significance threshold (Figures S13 and S14, material and methods). This level of concordance was achieved in the presence of both diverse MRI acquisition protocols and differential in silico phenotyping of the NIDPs. Nevertheless, these findings should be interpreted with caution owing to sample overlap between the two studies; 20% of participants in the ENIGMA study are included in the UKB. Exclusion of such individuals was not possible here owing to the reliance of our study on summary statistics.

Tissue specificity of neuroimaging GReX associations provides insight into the molecular pathways underlying neurobiology

We investigated the extent to which tissue context breadth provides meaningful information about GReX mechanisms. First, to characterize the distribution of GReX-NIDP associations across the 19 different JTI models, we annotated each gene according to the total number of JTI models in which the gene was significantly associated with any brain measure (Figure 4A). Here, we relied on the FDR-significant subset of the data, as is appropriate for discovery analyses. A plurality of genes demonstrated a statistically significant neurological association in the context of only 1 single JTI model. Gene counts decreased as replication categories increased until reaching a nadir at 14. A similar pattern was observed when considering joint GReX-NIDP associations (Figure 4B). Expression of some genes was neurologically relevant in only a few tissues, while for others, relevance for genetically predicted expression was ubiquitous across tissue contexts. To explore the functional significance of this distribution, we conducted a comparative GO enrichment analysis of tissue-specific genes against tissue-shared genes (Figure 4D).²³ GReX in both sets independently clustered into multiple ontology categories, indicating the non-random nature of the identified associations.

Analysis of tissue context for NIDP-GReX associations

(A) Distribution of NIDP-associated GReX annotated on a scale according to the total number of JTI models. Genes associated with NIDPs in a single model are shown in the column labeled 1, while those that have significant associations (FDR <0.05) with NIDPs in all 19 tissue models are captured in the column labeled 19. The colors of each column represent the proportion of specific tissue contexts in each.

(B) Distribution of GReX-NIDP associations annotated on a scale according to the total number of JTI models as in Figure 3A. The actual number of GReX-NIDP associations is shown rather than the number of unique genes across potentially multiple NIDPs as is done in Figure 3A.

(C) Median Z-score magnitude (absolute value of normalized effect size) of GReX-NIDP associations according to the total number of JTI models in which the association was significant.

(D) Gene Ontology enrichment analysis of NIDP-associated GReX according to JTI model specificity. We use the Cellular Components library of Gene Ontology in this analysis.

(E) Mean standard deviation of replicated GReX-NIDP associations across JTI models according to the number of JTI models. Error bars represent the 95% confidence interval.

We then assessed the relationship between the identified neurological impact of gene expression and the number of tissues in which that expression is neuro-relevant through analysis of the mean effect-size magnitudes across each category. We identified a positive relationship such that GReX-NIDP associations shared across all tested tissues showed relatively higher magnitude changes in neural architecture (Spearman rank correlation p = 7.73e−06, coefficient ρ = 0.968) (Figure 4C). We calculated the standard deviation of effect-size magnitude for each gene in each column and identified a negative relationship as tissue count replication increased (p value <5.33e−7, ρ = −0.938) (Figure 4E). In summary, GReX-NIDP associations that were captured across multiple JTI models demonstrated greater effect-size magnitude with less variance across tissue models.

These data in conjunction with the previous GO suggested that tissue-shared GReX associations likely represent general biological processes that occur ubiquitously, while those GReX-NIDP associations identified in a few tissues demonstrated more moderate, context-dependent effect sizes.

GReX-NIDP associations identified in non-brain JTI models are highly relevant and provide information inaccessible from brain-only analyses

We assessed the significance of JTI model type (brain vs. non-brain) with regard to the GReX-NIDP associations. As in the previous section, we annotated each gene according to the number of JTI models in which it was associated with a neurological trait. Then we annotated the NIDP-associated GReX in each category according to the type of JTI model. The first tissue category, labeled as “non-brain,” consists of the adrenal gland, sigmoid colon, transverse colon, pituitary, liver, and whole blood. The rest of the brain phenotypes and the spinal cord are classified as “brain.”

Genes with significant associations exclusively in brain JTI models were classified as “unique brain,” while those that surpassed significance only in non-brain JTI models were classified as “unique non-brain.” Those with associations in both were labeled as “shared” (Figure 5A).

GReX in both brain and non-brain tissues provides complementary information reflecting the integrated nature of neurobiology

(A) Bar plot showing the counts of all 7,193 GReX measures classified according to their presence in brain-derived JTI models, non-brain models, or both.

(B) Boxplot showing the median Z-score magnitude (absolute value of normalized effect size) across all NeuroimaGene NIDPs at an FDR <0.05. The boxplot shows the interquartile range for all findings, grouped according to their presence in brain-derived JTI models, non-brain models, or both.

(C) Density plot showing the distribution of median-normalized GReX-NIDP effect sizes as measured in brain JTI models vs. the same measure as captured in non-brain tissues. Only GReX-NIDP associations shared across both types of JTI models are represented here (N = 1,262,333, FDR-corrected p value <0.05). Trend line represents a linear regression.

(D) Relative counts of tissue-concordant GReX-NIDP associations that share the same direction of effect across tissue types vs. those that show differing directions of effect.

(E) The number of GReX associations per JTI model plotted against the number of genes in the GTEx model, grouped by model type (brain vs. non-brain). Trend lines represent linear regression.

(F) Gene Ontology enrichment analysis of NIDP-associated GReX measures according to JTI model type. We use the Cellular Components library of Gene Ontology in this analysis.

As expected, the greatest number of informative GReX-NIDP associations were identified by JTI models derived from brain tissue. Still, nearly 50% of GReX associations detected in brain tissues were also detected in non-brain tissues, underscoring the potential for non-brain tissues to serve as proxy models for the hard-to-access CNS. Accordingly, for each shared GReX-NIDP association, we assessed the similarity in estimated effect size across brain and non-brain groups by comparing the median effect sizes of the 2 tissue groups (Figure 5C). GReX-NIDP associations captured across both brain and non-brain models demonstrated high consistency in cross tissue-type effect-size directions (Figures 5C and 5D).

We noted with interest that there were over 1,700 genes for which GReX exclusively outside of the brain was significantly associated with neurophysiological traits (Figure 5A). This seemed to suggest that transcriptomic variation outside the CNS was highly relevant to brain physiology. The median effect-size magnitude of GReX associations in non-brain tissues was significantly lower than that of brain tissues (p value <2.2e−16); however, the difference was small, with the median effect size falling within the interquartile range for the brain data (Figure 5B). One example of a non-brain gene is GALK1, which encodes the metabolic enzyme galactokinase 1. This gene is responsible for metabolizing galactose, a function that occurs almost exclusively in the liver.³⁷ Mutations in this gene have been associated with brain damage,³⁸ and similarly, we observe novel associations between predicted expression of GALK1 and a variety of white matter measurements (Table S3). In accordance with the liver-specific functioning of galactokinase, all NIDP associations for GALK1 expression are derived exclusively from the liver JTI model.

Each JTI model involves a variable number of genes. The ratio of total neurophysiological associations derived from each JTI model to the number of genes in each tissue model thus reflects a measure of marginal information gained per additional gene. We compared these ratios among the JTI models (Figure 5E) and found that the amount of information gained by increasing the size of non-brain tissue models was not statistically different than the information gained by increasing the size of brain JTI models (p value = 0.949). While brain tissue retained an absolute advantage over non-brain tissues, the marginal gain in neurophysiological information from each new gene was the same across brain and non-brain models.

To characterize the functional ramifications of this geographic partitioning, we conducted comparative GO analysis, contrasting the unique brain NIDP-associated GReX with the unique non-brain NIDP-associated GReX.²³ For genes exclusively associated with NIDPs via brain JTI models, there was again heavy enrichment in cell projection pathways, as is involved in axonogenesis and dendrite formation. In non-brain tissues, there was an enrichment in pathways reflecting cellular metabolism and cell signaling (Figure 5F).

NeuroimaGene application to SCZ recapitulates and extends upon previously documented observations

The most recent GWAS meta-analysis of SCZ identified 263 trait-associated loci. The mapped variants support a narrow sense heritability of 60%–80%.²⁷ Independently, clinical neuroimaging studies demonstrate statistically significant differences of brain morphology in the context of SCZ relative to healthy control participants. Interpretation of these findings is complicated as some brain changes may be secondary to disease progression while others may reflect the biology driving disease. We hypothesized that a subset of SCZ-associated neuroimaging features are influenced by the genetically determined expression of SCZ-associated genes. To test this hypothesis, we first identified SCZ-associated NIDPs from the large standardized clinical neuroimaging study. Performed by the ENIGMA consortium, this study identified 33 cortical NIDPs and 12 subcortical NIDPs associated with SCZ. We then identified corresponding NIDPs from the UKB neuroimaging study, matching on atlas parcellation, region name, and measurement type (material and methods).²⁸^,²⁹ We next sought to identify SCZ trait-associated genes. A TWAS of the most recent and largest SCZ GWAS meta-analysis to date identified 63 SCZ-associated genes (material and methods).²⁷ Of these genes, 50 were tested in the NeuroimaGene resource. These data represent 3 intersecting aspects of biology: GReX-NIDP associations from NeuroimaGene, NIDP-SCZ associations from the ENIGMA study, and SCZ-GReX associations from the GWAS/TWAS analysis (Figure 6A).

NeuroimaGene application to SCZ informs clinical neuroimaging findings

(A) Schematic overview of the application of the pipeline to identified tri-fold associations between genetically regulated gene expression (GReX), SCZ, and neuroimaging-derived phenotypes (NIDPs).

(B) Derivation of maximally discriminatory significance threshold. Comparison of total associations between SCZ GReX and SCZ NIDPs (scz_ct) and total associations between random GReX and SCZ NIDPs (null_ct) at different p value thresholds. The teal data represent the number of significant associations between random gene sets and SCZ NIDPs at each p value threshold. The pink data represent the actual number of significant associations between SCZ genes and SCZ NIDPs at each p value threshold.

(C) Quantification of SCZ GReX associations with SCZ NIDPs compared to a null distribution at p value ≤ 0.005. The experimental gene set represents the set of GReX measures associated with SCZ according to Trubetskoy et al. The histogram represents the distribution of associations between 1,000 random equally sized gene sets and SCZ NIDPs. The line in red represents the limit of association counts exceeded by only 0.05 of the random gene sets. The blue line represents the number of total associations between SCZ-associated NIDPs.

(D) Visualization of the surface area, thickness, and volumetric consequences of GReX for SCZ-associated genes as detected in the Desikan Atlas using fsbrain.³⁹ The effect size represents the mean GReX-NIDP effect size corrected for LD contamination via Mendelian randomization (MR-JTI). The intensity of the shading represents the magnitude of the effect size, and the color represents the direction of effect of increased GReX on the size of the region in question (blue = negative, red = positive).

(E) Distribution of MR-JTI significant associations between GReX of SCZ GReX and SCZ-associated NIDPs with significance values ≤0.005. NIDPs are largely polygenic with expression of multiple genes contributing to morphology. Measures of polygenicity for individual NIDPs within the context of SCZ can be observed across the x axis. Similarly, GReX of a gene can affect multiple NIDPs. The pleiotropic effects of SCZ GReX on SCZ NIDPs can be observed along the y axis.

SCZ NIDPs are more strongly associated with GReX of SCZ-associated genes relative to the full imputed transcriptome

We used the NeuroimaGene resource to identify all detected associations between SCZ GReX and SCZ NIDPs (Table S6). We then derived the maximally discriminatory p value through polygenic risk score-style significance thresholding (material and methods; Figures 6B, 6C, and S15). At p ≤ 0.005 and after adjustment for horizontal pleiotropy through MR-JTI (an MR framework),⁵ we identified 147 causal associations between GReX of SCZ-associated genes and clinical SCZ-associated NIDPs (Figures 6D, 6E, and S16; Table S7). These associations highlighted 33 NIDPs reflecting cortical and subcortical morphology and involved 16 genes.

The largest effect-size magnitudes were observed for thickness of the right parahippocampal gyrus and volume of the right hippocampus—regions that have been repeatedly associated with SCZ risk and severity.⁴⁰ Our analysis extends upon this work by establishing that genetically determined transcription of the SCZ risk genes INO80E and MAPK3 are associated with morphology of these NIDPs, independent of disease.

We wish to highlight how the integration of NIDPs with GReX analyses represents a synergistic approach to examining disease pathophysiology. First, this approach provides regional and spatial context to gene transcription. While the immunoglobulin glycoprotein EMB is already associated with SCZ, we provide evidence that its expression level in the blood causally impacts the volume of the right amygdala. This region is altered in individuals with SCZ and plays a role in modulating fear and anxiety—processes that are related to the paranoia commonly experienced in this population.⁴¹^,⁴² Second, this approach provides molecular context to the macrostructural changes observed in the disease. While the right amygdala is typically smaller in individuals with SCZ,²⁹ our data build upon this knowledge by providing evidence that the region is influenced by expression of INO80E, MAPK3, and EMB (all of which are implicated in mediating SCZ risk).²⁷ In this work, we thus highlight a subset of SCZ neuroimaging findings that are affected by SCZ genetics independently of disease process and partition these NIDPs according to their transcriptomic effectors. While we present individual associations, we do not yet have sufficient evidence to make claims about the contribution of these genes to specific symptoms associated with amygdala function in the context of SCZ.

Discussion

In NeuroimaGene, we provide a publicly available resource for examining neurological correlates of genetic variation. The number of studies identifying associations between genetic variation and neuropsychiatric traits is constantly increasing and so too does the importance of enriching these observational studies with data on a wide array of intermediate phenotypes. By characterizing the impact of GReX on diverse measures of brain structure and function, we provide a tool that is broadly applicable to questions of development, neural maintenance, neurodegeneration, cognition, and psychiatric health.

The resource leverages JTI-enriched models that have been demonstrated to outperform traditional single-tissue models.⁵ By analyzing multiple tissues rather than performing a single cross-tissue analysis, we retain information on the geographic partitioning of predictive and causal GReX variation. This is enhanced by the inclusion of non-brain tissues. The breadth of NIDPs characterized by the UKB and subsequently described here represent a major advantage of this resource as well. There is substantial variation in how different cortical, subcortical, and diffusion tensor imaging atlases parcellate the brain. Because NeuroimaGene includes a wide variety of such atlases, conversion between atlases is rarely necessary when searching for GReX associated with an NIDP of interest.

The utility of GReX as an informative endophenotype for neurological features is strongly underscored by the clustering analyses wherein we recapitulate the 2-dimensional structure of the human cortex and subcortex, using the effect size of the genetically predicted transcriptome alone. The highly significant replication of associations across the 2 independent population subsets and a third external data source further demonstrates the strength of our approach. Additionally, stratifying the neuro-relevant genes according to tissue specificity followed by GO analysis demonstrated that tissue specificity approximates differentiation. The genes with tissue-shared GReX associations clustered into pathways that reflect housekeeping functions such as cell membrane integrity and mitochondrial function. For example, the GReX of TRIT1 has statistically significant associations with NIDPs in all 19 tested tissues. This gene encodes a protein targeted to the mitochondria that modifies transfer ribonucleic acid (tRNA) to maintain correct reading frame during translation, a process that is ubiquitously important to cell survival (Table S3).⁴³

Conversely, the genes with tissue-specific GReX associations included pathways reflecting cell projections such as those required for dendrite and axon formation and cell junctions. One such gene is CACNB2, which is translated into the calcium voltage-gated channel auxiliary subunit beta 2, which participates in the specialized synaptic cell junction of neurons. GReX of CACNB2 demonstrated statistically significant associations with NIDPs only in putamen basal ganglia (Table S3). These categories illustrate the relevance of both general and highly cell-specific processes in the healthy development of the human brain. While associations shared across tissues, as are prioritized by cross-tissue analyses, capture a high-confidence set of associations with larger and more uniform effect sizes, they miss some associations captured in single-tissue analyses that are reflective of greater differentiation/specificity.

Clinically, the link between bodily health and neuropsychiatric health is well established, with conditions such as hepatic encephalopathy and pheochromocytoma-induced panic disorder representing high-visibility instances. Using 6 non-brain tissues with a priori evidence for neurological relevance, we present data suggesting that transcriptomic variation in tissues outside the brain has important effects on neurological homeostasis that are large enough to be captured by neuroimaging methods.

By prioritizing SCZ-associated GReX, these data support a genetic basis for cortical morphology changes observed in SCZ. Furthermore, we provide specific gene candidates for which expression variation may be driving these cortical changes. This approach is generalizable to conditions with significant heritability and neuroimaging correlates.

This work is limited by several factors: first, regulatory genomic architecture and LD scores translate poorly across populations of different ancestral backgrounds. As the NeuroimaGene resource was generated from samples of primarily European ancestry, the portability of the findings to other populations remains unclear. In future studies, we hope to leverage a broader range of data as reference transcriptome resources become available for less-well-characterized populations. Second, while MR adjusts for horizontal pleiotropy at the level of SNPs, we do not control for confounding that is mediated through genetic co-expression.⁴⁴ Though mitigated by using JTI models trained on SNPs in cis with their gene targets, the potential of co-expression limits the rigor of causal inference and necessitates the use of functional studies to validate findings. The presence of some sample overlap between the ENIGMA and UKB data is a limitation of our external replication analyses, potentially inflating the replication rate.

SNP variation can affect a phenotype through many different mechanisms. By measuring the impact of SNPs only on gene expression, our approach does not capture the proportion of narrow-sense heritability mediated through other mechanisms.⁹ This loss in breadth is redeemed by the increase in mechanistic specificity. Because our source data come from healthy individuals, neuroanatomical changes exclusive to disease-causing variants will be poorly covered.⁷ Because GReX is a continuous exposure, we anticipate that some neurological predisposition will be present in the brains of individuals with non-realized genetic risk.

Data and code availability

The datasets and code generated during this study are available at Zenodo (https://zenodo.org/records/8030632 and https://zenodo.org/records/10994978). The NeuroimaGene command line interface is available on Github (https://github.com/xbledsoe/NeuroimaGene).

Acknowledgments

We acknowledge support from the following National Institutes of Health (NIH) grants: NHGRI R35HG010718, NHGRI R01HG011138, NIMH R01MH126459, NIGMS R01GM140287, NIH RF1MH125933, NIH Office of the Director U24OD035523, and NIA AG068026. We thank Vanderbilt’s Advanced Computing Center for Research and Education (ACCRE) for infrastructure support.

Declaration of interests

The authors declare no competing interests.

Published: June 25, 2024

Footnotes

Supplemental information can be found online at https://doi.org/10.1016/j.ajhg.2024.06.002.

Contributor Information

Xavier Bledsoe, Email: xavier.bledsoe@vanderbilt.edu.

Eric R. Gamazon, Email: eric.gamazon@vumc.org.

Supplemental information

Document S1. Figures S1–S35

mmc1.pdf^{(28.2MB, pdf)}

Tables S1–S7

mmc2.xlsx^{(661.7KB, xlsx)}

Document S2. Article plus supplemental information

mmc3.pdf^{(31.9MB, pdf)}

References

1.Rees E., Owen M.J. Translating insights from neuropsychiatric genetics and genomics for precision psychiatry. Genome Med. 2020;12:43. doi: 10.1186/s13073-020-00734-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
2.Bray N.J., O'Donovan M.C. The genetics of neuropsychiatric disorders. Brain Neurosci. Adv. 2019;2 doi: 10.1177/2398212818799271. [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Need A.C., Goldstein D.B. Neuropsychiatric genomics in precision medicine: diagnostics, gene discovery, and translation. Dialogues Clin. Neurosci. 2016;18:237–252. doi: 10.31887/DCNS.2016.18.3/aneed. [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Barbeira A.N., Dickinson S.P., Bonazzola R., Zheng J., Wheeler H.E., Torres J.M., Torstenson E.S., Shah K.P., Garcia T., Edwards T.L., et al. Exploring the phenotypic consequences of tissue specific gene expression variation inferred from GWAS summary statistics. Nat. Commun. 2018;9:1825–1845. doi: 10.1038/s41467-018-03621-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Zhou D., Jiang Y., Zhong X., Cox N.J., Liu C., Gamazon E.R. A unified framework for joint-tissue transcriptome-wide association and Mendelian randomization analysis. Nat. Genet. 2020;52:1239–1246. doi: 10.1038/s41588-020-0706-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Liu S., Cai W., Liu S., Zhang F., Fulham M., Feng D., Pujol S., Kikinis R. Multimodal neuroimaging computing: a review of the applications in neuropsychiatric disorders. Brain Inform. 2015;2:167–180. doi: 10.1007/s40708-015-0019-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Miller K.L., Alfaro-Almagro F., Bangerter N.K., Thomas D.L., Yacoub E., Xu J., Bartsch A.J., Jbabdi S., Sotiropoulos S.N., Andersson J.L.R., et al. Multimodal population brain imaging in the UK Biobank prospective epidemiological study. Nat. Neurosci. 2016;19:1523–1536. doi: 10.1038/nn.4393. [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Gamazon E.R., Wheeler H.E., Shah K.P., Mozaffari S.V., Aquino-Michaels K., Carroll R.J., Eyler A.E., Denny J.C., GTEx Consortium. Nicolae D.L., et al. A gene-based association method for mapping traits using reference transcriptome data. Nat. Genet. 2015;47:1091–1098. doi: 10.1038/ng.3367. [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Li B., Ritchie M.D. From GWAS to Gene: Transcriptome-Wide Association Studies and Other Methods to Functionally Understand GWAS Discoveries. Front. Genet. 2021;12 doi: 10.3389/fgene.2021.713230. [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Chatzinakos C., Georgiadis F., Lee D., Cai N., Vladimirov V.I., Docherty A., Webb B.T., Riley B.P., Flint J., Kendler K.S., et al. TWAS pathway method greatly enhances the number of leads for uncovering the molecular underpinnings of psychiatric disorders. Am. J. Med. Genet. B Neuropsychiatr. Genet. 2020;183:454–463. doi: 10.1002/ajmg.b.32823. [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Reijnders J.S.A.M., Scholtissen B., Weber W.E.J., Aalten P., Verhey F.R.J., Leentjens A.F.G. Neuroanatomical correlates of apathy in Parkinson's disease: A magnetic resonance imaging study using voxel-based morphometry. Mov. Disord. 2010;25:2318–2325. doi: 10.1002/mds.23268. [DOI] [PubMed] [Google Scholar]
12.Fox A.S., Shackman A.J. The central extended amygdala in fear and anxiety: Closing the gap between mechanistic and neuroimaging research. Neurosci. Lett. 2019;693:58–67. doi: 10.1016/j.neulet.2017.11.056. [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Sasikumar S., Strafella A.P. Imaging Mild Cognitive Impairment and Dementia in Parkinson's Disease. Front. Neurol. 2020;11:47. doi: 10.3389/fneur.2020.00047. [DOI] [PMC free article] [PubMed] [Google Scholar]
14.HR S., JH C., T S., VS M. From the genome to the phenome and back: linking genes with human brain function and structure using genetically informed neuroimaging. Neuroscience. 2009;16:1–6. doi: 10.1016/j.neuroscience.2009.09.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Thompson P.M., Jahanshad N., Ching C.R.K., Salminen L.E., Thomopoulos S.I., Bright J., Baune B.T., Bertolín S., Bralten J., Bruin W.B., et al. ENIGMA and global neuroscience: A decade of large-scale studies of the brain in health and disease across more than 40 countries. Transl. Psychiatry. 2020;10:100. doi: 10.1038/s41398-020-0705-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Smith S.M., Douaud G., Chen W., Hanayik T., Alfaro-Almagro F., Sharp K., Elliott L.T. An expanded set of genome-wide association studies of brain imaging phenotypes in UK Biobank. Nat. Neurosci. 2021;24:737–745. doi: 10.1038/s41593-021-00826-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Elliott L.T., Sharp K., Alfaro-Almagro F., Shi S., Miller K.L., Douaud G., Marchini J., Smith S.M. Genome-wide association studies of brain imaging phenotypes in UK Biobank. Nature. 2018;562:210–216. doi: 10.1038/s41586-018-0571-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Thompson P.M., Stein J.L., Medland S.E., Hibar D.P., Vasquez A.A., Renteria M.E., Toro R., Jahanshad N., Schumann G., Franke B., et al. The ENIGMA Consortium: large-scale collaborative analyses of neuroimaging and genetic data. Brain Imaging Behav. 2014;8:153–182. doi: 10.1007/s11682-013-9269-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Grasby K.L., Jahanshad N., Painter J.N., Colodro-Conde L., Bralten J., Hibar D.P., Lind P.A., Pizzagalli F., Ching C.R.K., McMahon M.A.B., et al. The genetic architecture of the human cerebral cortex. Science. 2020;367 [Google Scholar]
20.Alfaro-Almagro F., Jenkinson M., Bangerter N.K., Andersson J.L.R., Griffanti L., Douaud G., Sotiropoulos S.N., Jbabdi S., Hernandez-Fernandez M., Vallee E., et al. Image processing and Quality Control for the first 10,000 brain imaging datasets from UK Biobank. Neuroimage. 2018;166:400–424. doi: 10.1016/j.neuroimage.2017.10.034. [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Qin H., Niu T., Zhao J. Identifying Multi-Omics Causers and Causal Pathways for Complex Traits. Front. Genet. 2019;10 doi: 10.3389/fgene.2019.00110. [DOI] [PMC free article] [PubMed] [Google Scholar]
22.null n., Aguet F., Anand S., Ardlie K.G., Gabriel S., Getz G.A., Graubert A., Hadley K., Handsaker R.E., Huang K.H., et al. The GTEx Consortium atlas of genetic regulatory effects across human tissues. Science. 2020;369:1318–1330. doi: 10.1126/science.aaz1776. [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Yu G., Wang L.-G., Han Y., He Q.-Y. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS A J. Integr. Biol. 2012;16:284–287. doi: 10.1089/omi.2011.0118. [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Gu Z., Eils R., Schlesner M. Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics. 2016;32:2847–2849. doi: 10.1093/bioinformatics/btw313. [DOI] [PubMed] [Google Scholar]
25.McInnes L., Healy J., Melville J. Umap: Uniform manifold approximation and projection for dimension reduction. arXiv. 2018 doi: 10.48550/arXiv:1802.03426. Preprint at. [DOI] [Google Scholar]
26.Desikan R.S., Ségonne F., Fischl B., Quinn B.T., Dickerson B.C., Blacker D., Buckner R.L., Dale A.M., Maguire R.P., Hyman B.T., et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage. 2006;31:968–980. doi: 10.1016/j.neuroimage.2006.01.021. [DOI] [PubMed] [Google Scholar]
27.Trubetskoy V., Pardiñas A.F., Qi T., Panagiotaropoulou G., Awasthi S., Bigdeli T.B., Bryois J., Chen C.-Y., Dennison C.A., Hall L.S., et al. Mapping genomic loci implicates genes and synaptic biology in schizophrenia. Nature. 2022;604:502–508. doi: 10.1038/s41586-022-04434-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
28.van Erp T.G.M., Walton E., Hibar D.P., Schmaal L., Jiang W., Glahn D.C., Pearlson G.D., Yao N., Fukunaga M., Hashimoto R., et al. Cortical Brain Abnormalities in 4474 Individuals With Schizophrenia and 5098 Control Subjects via the Enhancing Neuro Imaging Genetics Through Meta Analysis (ENIGMA) Consortium. Biol. Psychiatr. 2018;84:644–654. doi: 10.1016/j.biopsych.2018.04.023. [DOI] [PMC free article] [PubMed] [Google Scholar]
29.van Erp T.G.M., Hibar D.P., Rasmussen J.M., Glahn D.C., Pearlson G.D., Andreassen O.A., Agartz I., Westlye L.T., Haukvik U.K., Dale A.M., et al. Subcortical brain volume abnormalities in 2028 individuals with schizophrenia and 2540 healthy controls via the ENIGMA consortium. Mol. Psychiatr. 2016;21:547–553. doi: 10.1038/mp.2015.63. [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Kent W.J., Sugnet C.W., Furey T.S., Roskin K.M., Pringle T.H., Zahler A.M., Haussler D. The human genome browser at UCSC. Genome Res. 2002;12:996–1006. doi: 10.1101/gr.229102. [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Fairley S., Lowy-Gallego E., Perry E., Flicek P. The International Genome Sample Resource (IGSR) collection of open human genomic variation resources. Nucleic Acids Res. 2020;48:D941–D947. doi: 10.1093/nar/gkz836. [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Uellendahl-Werth F., Maj C., Borisov O., Juzenas S., Wacker E.M., Jørgensen I.F., Steiert T.A., Bej S., Krawitz P., Hoffmann P., et al. Cross-tissue transcriptome-wide association studies identify susceptibility genes shared between schizophrenia and inflammatory bowel disease. Commun. Biol. 2022;5:80. doi: 10.1038/s42003-022-03031-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Gamazon E.R., Zwinderman A.H., Cox N.J., Denys D., Derks E.M. Multi-tissue transcriptome analyses identify genetic mechanisms underlying neuropsychiatric traits. Nat. Genet. 2019;51:933–940. doi: 10.1038/s41588-019-0409-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Grassi D., Plonka F.B., Oksdath M., Guil A.N., Sosa L.J., Quiroga S. Selected SNARE proteins are essential for the polarized membrane insertion of igf-1 receptor and the regulation of initial axonal outgrowth in neurons. Cell Discov. 2015;1:1–15. doi: 10.1038/celldisc.2015.23. [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Mahfouz A., Huisman S.M.H., Lelieveldt B.P.F., Reinders M.J.T. Brain transcriptome atlases: a computational perspective. Brain Struct. Funct. 2017;222:1557–1580. doi: 10.1007/s00429-016-1338-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Mahfouz A., van de Giessen M., van der Maaten L., Huisman S., Reinders M., Hawrylycz M.J., Lelieveldt B.P.F. Visualizing the spatial gene expression organization in the brain through non-linear similarity embeddings. Methods. 2015;73:79–89. doi: 10.1016/j.ymeth.2014.10.004. [DOI] [PubMed] [Google Scholar]
37.Sørensen M., Mikkelsen K.S., Frisch K., Bass L., Bibby B.M., Keiding S. Hepatic galactose metabolism quantified in humans using 2-18F-fluoro-2-deoxy-D-galactose PET/CT. J. Nucl. Med. 2011;52:1566–1572. doi: 10.2967/jnumed.111.092924. [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Rubio-Gozalbo M.E., Derks B., Das A.M., Meyer U., Möslinger D., Couce M.L., Empain A., Ficicioglu C., Juliá Palacios N., De Los Santos De Pelegrin M.M., et al. Galactokinase deficiency: lessons from the GalNet registry. Genet. Med. 2021;23:202–210. doi: 10.1038/s41436-020-00942-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Schäfer T., Ecker C. fsbrain: an R package for the visualization of structural neuroimaging data. bioRxiv. 2020 doi: 10.1101/2020.09.18.302935. Preprint at. [DOI] [Google Scholar]
40.Sim K., DeWitt I., Ditman T., Zalesak M., Greenhouse I., Goff D., Weiss A.P., Heckers S. Hippocampal and parahippocampal volumes in schizophrenia: a structural MRI study. Schizophr. Bull. 2006;32:332–340. doi: 10.1093/schbul/sbj030. [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Pinkham A., Bass E., Klein H., Springfield C., Kent J., Aslan S. Inducing Paranoia is Linked to Increased Resting Amygdala Cerebral Blood Flow in Healthy Individuals. J. Exp. Psychopathol. 2022;13 [Google Scholar]
42.Fan L., Klein H., Bass E., Springfield C., Pinkham A. Amygdala hyperconnectivity in the paranoid state: A transdiagnostic study. J. Psychiatr. Res. 2021;138:117–124. doi: 10.1016/j.jpsychires.2021.03.049. [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Muylle E., Jiang H., Johnsen C., Byeon S.K., Ranatunga W., Garapati K., Zenka R.M., Preston G., Pandey A., Kozicz T., et al. TRIT1 defect leads to a recognizable phenotype of myoclonic epilepsy, speech delay, strabismus, progressive spasticity, and normal lactate levels. J. Inherit. Metab. Dis. 2022;45:1039–1047. doi: 10.1002/jimd.12550. [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Porcu E., Rüeger S., Lepik K., Santoni F.A., Reymond A., Kutalik Z. Mendelian randomization integrating GWAS and eQTL data reveals genetic determinants of complex and clinical traits. Nat. Commun. 2019;10:1–12. doi: 10.1038/s41467-019-10936-0. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Document S1. Figures S1–S35

mmc1.pdf^{(28.2MB, pdf)}

Tables S1–S7

mmc2.xlsx^{(661.7KB, xlsx)}

Document S2. Article plus supplemental information

mmc3.pdf^{(31.9MB, pdf)}

Data Availability Statement

[bib1] 1.Rees E., Owen M.J. Translating insights from neuropsychiatric genetics and genomics for precision psychiatry. Genome Med. 2020;12:43. doi: 10.1186/s13073-020-00734-5. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib2] 2.Bray N.J., O'Donovan M.C. The genetics of neuropsychiatric disorders. Brain Neurosci. Adv. 2019;2 doi: 10.1177/2398212818799271. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib3] 3.Need A.C., Goldstein D.B. Neuropsychiatric genomics in precision medicine: diagnostics, gene discovery, and translation. Dialogues Clin. Neurosci. 2016;18:237–252. doi: 10.31887/DCNS.2016.18.3/aneed. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib4] 4.Barbeira A.N., Dickinson S.P., Bonazzola R., Zheng J., Wheeler H.E., Torres J.M., Torstenson E.S., Shah K.P., Garcia T., Edwards T.L., et al. Exploring the phenotypic consequences of tissue specific gene expression variation inferred from GWAS summary statistics. Nat. Commun. 2018;9:1825–1845. doi: 10.1038/s41467-018-03621-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib5] 5.Zhou D., Jiang Y., Zhong X., Cox N.J., Liu C., Gamazon E.R. A unified framework for joint-tissue transcriptome-wide association and Mendelian randomization analysis. Nat. Genet. 2020;52:1239–1246. doi: 10.1038/s41588-020-0706-2. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib6] 6.Liu S., Cai W., Liu S., Zhang F., Fulham M., Feng D., Pujol S., Kikinis R. Multimodal neuroimaging computing: a review of the applications in neuropsychiatric disorders. Brain Inform. 2015;2:167–180. doi: 10.1007/s40708-015-0019-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib7] 7.Miller K.L., Alfaro-Almagro F., Bangerter N.K., Thomas D.L., Yacoub E., Xu J., Bartsch A.J., Jbabdi S., Sotiropoulos S.N., Andersson J.L.R., et al. Multimodal population brain imaging in the UK Biobank prospective epidemiological study. Nat. Neurosci. 2016;19:1523–1536. doi: 10.1038/nn.4393. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib8] 8.Gamazon E.R., Wheeler H.E., Shah K.P., Mozaffari S.V., Aquino-Michaels K., Carroll R.J., Eyler A.E., Denny J.C., GTEx Consortium. Nicolae D.L., et al. A gene-based association method for mapping traits using reference transcriptome data. Nat. Genet. 2015;47:1091–1098. doi: 10.1038/ng.3367. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib9] 9.Li B., Ritchie M.D. From GWAS to Gene: Transcriptome-Wide Association Studies and Other Methods to Functionally Understand GWAS Discoveries. Front. Genet. 2021;12 doi: 10.3389/fgene.2021.713230. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib10] 10.Chatzinakos C., Georgiadis F., Lee D., Cai N., Vladimirov V.I., Docherty A., Webb B.T., Riley B.P., Flint J., Kendler K.S., et al. TWAS pathway method greatly enhances the number of leads for uncovering the molecular underpinnings of psychiatric disorders. Am. J. Med. Genet. B Neuropsychiatr. Genet. 2020;183:454–463. doi: 10.1002/ajmg.b.32823. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib11] 11.Reijnders J.S.A.M., Scholtissen B., Weber W.E.J., Aalten P., Verhey F.R.J., Leentjens A.F.G. Neuroanatomical correlates of apathy in Parkinson's disease: A magnetic resonance imaging study using voxel-based morphometry. Mov. Disord. 2010;25:2318–2325. doi: 10.1002/mds.23268. [DOI] [PubMed] [Google Scholar]

[bib12] 12.Fox A.S., Shackman A.J. The central extended amygdala in fear and anxiety: Closing the gap between mechanistic and neuroimaging research. Neurosci. Lett. 2019;693:58–67. doi: 10.1016/j.neulet.2017.11.056. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib13] 13.Sasikumar S., Strafella A.P. Imaging Mild Cognitive Impairment and Dementia in Parkinson's Disease. Front. Neurol. 2020;11:47. doi: 10.3389/fneur.2020.00047. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib14] 14.HR S., JH C., T S., VS M. From the genome to the phenome and back: linking genes with human brain function and structure using genetically informed neuroimaging. Neuroscience. 2009;16:1–6. doi: 10.1016/j.neuroscience.2009.09.009. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib15] 15.Thompson P.M., Jahanshad N., Ching C.R.K., Salminen L.E., Thomopoulos S.I., Bright J., Baune B.T., Bertolín S., Bralten J., Bruin W.B., et al. ENIGMA and global neuroscience: A decade of large-scale studies of the brain in health and disease across more than 40 countries. Transl. Psychiatry. 2020;10:100. doi: 10.1038/s41398-020-0705-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib16] 16.Smith S.M., Douaud G., Chen W., Hanayik T., Alfaro-Almagro F., Sharp K., Elliott L.T. An expanded set of genome-wide association studies of brain imaging phenotypes in UK Biobank. Nat. Neurosci. 2021;24:737–745. doi: 10.1038/s41593-021-00826-4. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib17] 17.Elliott L.T., Sharp K., Alfaro-Almagro F., Shi S., Miller K.L., Douaud G., Marchini J., Smith S.M. Genome-wide association studies of brain imaging phenotypes in UK Biobank. Nature. 2018;562:210–216. doi: 10.1038/s41586-018-0571-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib18] 18.Thompson P.M., Stein J.L., Medland S.E., Hibar D.P., Vasquez A.A., Renteria M.E., Toro R., Jahanshad N., Schumann G., Franke B., et al. The ENIGMA Consortium: large-scale collaborative analyses of neuroimaging and genetic data. Brain Imaging Behav. 2014;8:153–182. doi: 10.1007/s11682-013-9269-5. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib19] 19.Grasby K.L., Jahanshad N., Painter J.N., Colodro-Conde L., Bralten J., Hibar D.P., Lind P.A., Pizzagalli F., Ching C.R.K., McMahon M.A.B., et al. The genetic architecture of the human cerebral cortex. Science. 2020;367 [Google Scholar]

[bib20] 20.Alfaro-Almagro F., Jenkinson M., Bangerter N.K., Andersson J.L.R., Griffanti L., Douaud G., Sotiropoulos S.N., Jbabdi S., Hernandez-Fernandez M., Vallee E., et al. Image processing and Quality Control for the first 10,000 brain imaging datasets from UK Biobank. Neuroimage. 2018;166:400–424. doi: 10.1016/j.neuroimage.2017.10.034. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib21] 21.Qin H., Niu T., Zhao J. Identifying Multi-Omics Causers and Causal Pathways for Complex Traits. Front. Genet. 2019;10 doi: 10.3389/fgene.2019.00110. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib22] 22.null n., Aguet F., Anand S., Ardlie K.G., Gabriel S., Getz G.A., Graubert A., Hadley K., Handsaker R.E., Huang K.H., et al. The GTEx Consortium atlas of genetic regulatory effects across human tissues. Science. 2020;369:1318–1330. doi: 10.1126/science.aaz1776. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib23] 23.Yu G., Wang L.-G., Han Y., He Q.-Y. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS A J. Integr. Biol. 2012;16:284–287. doi: 10.1089/omi.2011.0118. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib24] 24.Gu Z., Eils R., Schlesner M. Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics. 2016;32:2847–2849. doi: 10.1093/bioinformatics/btw313. [DOI] [PubMed] [Google Scholar]

[bib25] 25.McInnes L., Healy J., Melville J. Umap: Uniform manifold approximation and projection for dimension reduction. arXiv. 2018 doi: 10.48550/arXiv:1802.03426. Preprint at. [DOI] [Google Scholar]

[bib26] 26.Desikan R.S., Ségonne F., Fischl B., Quinn B.T., Dickerson B.C., Blacker D., Buckner R.L., Dale A.M., Maguire R.P., Hyman B.T., et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage. 2006;31:968–980. doi: 10.1016/j.neuroimage.2006.01.021. [DOI] [PubMed] [Google Scholar]

[bib27] 27.Trubetskoy V., Pardiñas A.F., Qi T., Panagiotaropoulou G., Awasthi S., Bigdeli T.B., Bryois J., Chen C.-Y., Dennison C.A., Hall L.S., et al. Mapping genomic loci implicates genes and synaptic biology in schizophrenia. Nature. 2022;604:502–508. doi: 10.1038/s41586-022-04434-5. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib28] 28.van Erp T.G.M., Walton E., Hibar D.P., Schmaal L., Jiang W., Glahn D.C., Pearlson G.D., Yao N., Fukunaga M., Hashimoto R., et al. Cortical Brain Abnormalities in 4474 Individuals With Schizophrenia and 5098 Control Subjects via the Enhancing Neuro Imaging Genetics Through Meta Analysis (ENIGMA) Consortium. Biol. Psychiatr. 2018;84:644–654. doi: 10.1016/j.biopsych.2018.04.023. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib29] 29.van Erp T.G.M., Hibar D.P., Rasmussen J.M., Glahn D.C., Pearlson G.D., Andreassen O.A., Agartz I., Westlye L.T., Haukvik U.K., Dale A.M., et al. Subcortical brain volume abnormalities in 2028 individuals with schizophrenia and 2540 healthy controls via the ENIGMA consortium. Mol. Psychiatr. 2016;21:547–553. doi: 10.1038/mp.2015.63. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib30] 30.Kent W.J., Sugnet C.W., Furey T.S., Roskin K.M., Pringle T.H., Zahler A.M., Haussler D. The human genome browser at UCSC. Genome Res. 2002;12:996–1006. doi: 10.1101/gr.229102. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib31] 31.Fairley S., Lowy-Gallego E., Perry E., Flicek P. The International Genome Sample Resource (IGSR) collection of open human genomic variation resources. Nucleic Acids Res. 2020;48:D941–D947. doi: 10.1093/nar/gkz836. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib32] 32.Uellendahl-Werth F., Maj C., Borisov O., Juzenas S., Wacker E.M., Jørgensen I.F., Steiert T.A., Bej S., Krawitz P., Hoffmann P., et al. Cross-tissue transcriptome-wide association studies identify susceptibility genes shared between schizophrenia and inflammatory bowel disease. Commun. Biol. 2022;5:80. doi: 10.1038/s42003-022-03031-6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib33] 33.Gamazon E.R., Zwinderman A.H., Cox N.J., Denys D., Derks E.M. Multi-tissue transcriptome analyses identify genetic mechanisms underlying neuropsychiatric traits. Nat. Genet. 2019;51:933–940. doi: 10.1038/s41588-019-0409-8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib34] 34.Grassi D., Plonka F.B., Oksdath M., Guil A.N., Sosa L.J., Quiroga S. Selected SNARE proteins are essential for the polarized membrane insertion of igf-1 receptor and the regulation of initial axonal outgrowth in neurons. Cell Discov. 2015;1:1–15. doi: 10.1038/celldisc.2015.23. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib35] 35.Mahfouz A., Huisman S.M.H., Lelieveldt B.P.F., Reinders M.J.T. Brain transcriptome atlases: a computational perspective. Brain Struct. Funct. 2017;222:1557–1580. doi: 10.1007/s00429-016-1338-2. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib36] 36.Mahfouz A., van de Giessen M., van der Maaten L., Huisman S., Reinders M., Hawrylycz M.J., Lelieveldt B.P.F. Visualizing the spatial gene expression organization in the brain through non-linear similarity embeddings. Methods. 2015;73:79–89. doi: 10.1016/j.ymeth.2014.10.004. [DOI] [PubMed] [Google Scholar]

[bib37] 37.Sørensen M., Mikkelsen K.S., Frisch K., Bass L., Bibby B.M., Keiding S. Hepatic galactose metabolism quantified in humans using 2-18F-fluoro-2-deoxy-D-galactose PET/CT. J. Nucl. Med. 2011;52:1566–1572. doi: 10.2967/jnumed.111.092924. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib38] 38.Rubio-Gozalbo M.E., Derks B., Das A.M., Meyer U., Möslinger D., Couce M.L., Empain A., Ficicioglu C., Juliá Palacios N., De Los Santos De Pelegrin M.M., et al. Galactokinase deficiency: lessons from the GalNet registry. Genet. Med. 2021;23:202–210. doi: 10.1038/s41436-020-00942-9. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib44] 39.Schäfer T., Ecker C. fsbrain: an R package for the visualization of structural neuroimaging data. bioRxiv. 2020 doi: 10.1101/2020.09.18.302935. Preprint at. [DOI] [Google Scholar]

[bib39] 40.Sim K., DeWitt I., Ditman T., Zalesak M., Greenhouse I., Goff D., Weiss A.P., Heckers S. Hippocampal and parahippocampal volumes in schizophrenia: a structural MRI study. Schizophr. Bull. 2006;32:332–340. doi: 10.1093/schbul/sbj030. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib40] 41.Pinkham A., Bass E., Klein H., Springfield C., Kent J., Aslan S. Inducing Paranoia is Linked to Increased Resting Amygdala Cerebral Blood Flow in Healthy Individuals. J. Exp. Psychopathol. 2022;13 [Google Scholar]

[bib41] 42.Fan L., Klein H., Bass E., Springfield C., Pinkham A. Amygdala hyperconnectivity in the paranoid state: A transdiagnostic study. J. Psychiatr. Res. 2021;138:117–124. doi: 10.1016/j.jpsychires.2021.03.049. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib42] 43.Muylle E., Jiang H., Johnsen C., Byeon S.K., Ranatunga W., Garapati K., Zenka R.M., Preston G., Pandey A., Kozicz T., et al. TRIT1 defect leads to a recognizable phenotype of myoclonic epilepsy, speech delay, strabismus, progressive spasticity, and normal lactate levels. J. Inherit. Metab. Dis. 2022;45:1039–1047. doi: 10.1002/jimd.12550. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib43] 44.Porcu E., Rüeger S., Lepik K., Santoni F.A., Reymond A., Kutalik Z. Mendelian randomization integrating GWAS and eQTL data reveals genetic determinants of complex and clinical traits. Nat. Commun. 2019;10:1–12. doi: 10.1038/s41467-019-10936-0. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

A transcriptomic atlas of the human brain reveals genetically determined aspects of neuropsychiatric health

Xavier Bledsoe

Eric R Gamazon

Summary

Graphical abstract

Introduction

Figure 1.

Material and methods

Data acquisition

Statistical analysis

Model building and JTI framework

TWASs of neuroimaging phenotypes

Multiple-testing correction in discovery cohort

Gene set analysis

Hierarchical clustering of NIDPs using nominally significant GReX associations

UMAP dimensionality reduction of T1 and dMRI phenotypes

PCA dimensionality reduction of T1 phenotypes

Internal replication analysis for neuroimaging TWASs

External replication analysis of neuroimaging TWAS in ENIGMA

SCZ-associated GReX curation

SCZ NIDP curation

Maximally discriminatory GReX threshold determination

Mendelian randomization JTI

Results

NeuroimaGene comprises a searchable atlas of associations between brain measures and the genetically regulated transcriptome

Table 1.

Hierarchical clustering and dimensionality reduction of NIDPs according to GReX effect sizes recapitulate the spatial patterning of the brain

Figure 2.

TWAS methodology for identifying neuroimaging GReX associations demonstrates high reproducibility in internal and external validation cohorts

Figure 3.

Tissue specificity of neuroimaging GReX associations provides insight into the molecular pathways underlying neurobiology

Figure 4.

GReX-NIDP associations identified in non-brain JTI models are highly relevant and provide information inaccessible from brain-only analyses

Figure 5.

NeuroimaGene application to SCZ recapitulates and extends upon previously documented observations

Figure 6.

SCZ NIDPs are more strongly associated with GReX of SCZ-associated genes relative to the full imputed transcriptome

Discussion

Data and code availability

Acknowledgments

Declaration of interests

Footnotes

Contributor Information

Supplemental information

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases