Brain-heart-eye axis revealed by multi-organ imaging, genetics and proteomics

Figures

• 

  Figure 1:. Study workflow a) The 2003 multi-scale patterns of structural covariance (PSCs), 82 heart imaging-derived phenotypes (IDPs) from magnetic resonance imaging (MRI), and 84 eye IDPs from optical coherence tomography (OCT). b) Three main sets of analyses were conducted to demonstrate the brain-heart-eye axis: i) phenotype-wide associations (PWAS) were performed between the 2003 brain PSCs, 82 heart IDPs, and 84 eye IDPs; ii) genome-wide association studies (GWAS) and post-GWAS analyses were conducted between the brain PSCs, heart IDPs, eye IDPs, and common genetic variants (SNPs), and iii) prediction ability of the brain PSCs, heart IDPs, eye IDPs, and their PRS were assessed with AI to predict 14 systemic disease categories and 8 cognitive scores. All analyses used the Genome Reference Consortium Human Build 37 (GRCh37). Readers can visualize the 2003 brain PSCs via the BRIDGEPORT knowledge portal: https://labs-laboratory.com//bridgeport. Detailed interpretations of the 84 eye IDPs and 82 heart IDPs are publicly available at https://labs-laboratory.com/medicine/eye and https://labs-laboratory.com/medicine/cardiovascular.

• 

  Figure 2:. Phenotypic associations of the 2003 brain PSCs, 82 heart IDPs, and 84 eye IDPs. a) The phenotypic associations between the 82 heart IDPs (left panel) and 2003 brain PSCs (middle panel) are shown after the Bonferroni correction (P-value<0.05/2003/82; effective sample size: 21,948<N<23,548; −0.23<r<0.52). For each PSC/IDP category, the percentage of the significant associations is displayed. In the right panel, we charted the significant PSC-IDP associations in the 3D brain space, providing the Pearson correlation coefficient (r) for representative brain PSCs with the heart IDP (LVM: left ventricle myocardial mass). b) The phenotypic associations between the 2003 brain PSCs (middle panel) and 84 eye IDPs (left panel) are shown. Significant results that survived the Bonferroni correction are presented (P-value<0.05/2003/84; effective sample size: 1284<N<4472; −0.09<r<0.24). We showed the high-level PSC/IDP categories for visualization purposes. Abbreviations: MtL: overall macular thickness (left); MoiL: macular thickness at the outer inferior subfield (left); MonL: macular thickness at the outer nasal subfield (left); MosL: macular thickness at the outer superior subfield (left);…

• 

  Figure 3:. Proteomics associations of the 2003 brain PSCs, 82 heart IDPs, and 84 eye IDPs. a) Proteome-wide associations between the 2003 brain PSCs (left panel) and 2923 plasma proteins (middle panel) are shown after the Bonferroni correction (P-value<0.05/2923/2003; effective sample size: 51<N<4520; −0.42<r<0.55). For each PSC or protein, the percentage of significant associations is proportional to the rectangle’s height. In the right panel, we annotated the protein expression profile in a specific tissue/organ by respective colors of the protein and organ; dotted lines show secondary expression across different organs. b) The proteomics associations between the 84 eye IDPs (left panel) and 2923 plasma proteins (middle panel) are shown after the Bonferroni correction (P-value<0.05/2923/84; effective sample size: 48<N<4250; −0.47<r<0.48). For each IDP or protein, the percentage of significant associations is proportional to the rectangle’s height. In the right panel, we annotated the protein expression in a specific tissue/organ by respective colors of the protein and organ; dotted lines show secondary expression across different organs. c) The proteomics associations between the 82 heart…

• 

  Figure 4:. Genome-wide associations of the 2003 brain PSCs, 82 heart IDPs, and 84 eye IDPs a) Cytogenetic regions where the genomic region was jointly linked to the heart IDPs and brain PSCs. Bonferroni correction was applied to denote significant genomic loci associated with the brain PSCs (P-value<5×10−8/2003) and heart IDPs (P-value<5×10−8/82). b) Cytogenetic regions where the genomic region was jointly linked to the eye IDPs and brain PSCs. Bonferroni correction was applied to denote significant genomic loci associated with the brain PSCs (P-value<5×10−8/2003) and eye IDPs (P-value<5×10−8/84). c) Cytogenetic regions where the genomic region was jointly linked to eye IDPs and heart IDPs. Bonferroni correction was applied to denote significant genomic loci associated with the heart IDPs (P-value<5×10−8/82) and eye IDPs (P-value<5×10−8/84). Readers can visualize the 2003 brain PSCs via the BRIDGEPORT knowledge portal: https://labs-laboratory.com//bridgeport. Detailed interpretations of the 84 eye IDPs and 82 heart IDPs are publicly available at https://labs-laboratory.com/medicine/eye and https://labs-laboratory.com/medicine/cardiovascular.

• 

  Figure 5:. Key genetic parameters of the 2003 brain PSCs, 82 heart IDPs, and 84 eye IDPs a-c) The SNP-based heritability (h2) was estimated using the GCTA software for the 2003 brain PSCs (C32–1024), 82 heart IDPs, and 84 eye IDPs. d-f) The selection signature (S) was estimated using the SBayesS software for the 2003 brain PSCs, 82 heart IDPs, and 84 eye IDPs. g-i) The polygenicity (Pi) was estimated using the SBayesS software for the 2003 brain PSCs, 82 heart IDPs, and 84 eye IDPs. j-l) The tissue-specific partitioned heritability using gene expression and chromatin data was estimated using the LDSC software for the 2003 brain PSCs, 82 heart IDPs, and 84 eye IDPs. We further categorized brain PSCs into different scales of the C parameters. The 82 heart IDPs are divided into 6 segmented heart regions: the left ventricle (LV), right ventricle (RV), left atrium (LA), right atrium (RA), descending aorta (DA), and ascending aorta (AA). The error bar denotes the standard error of the estimate of the mean. The 84 eye IDPs were classified into 11 high-level phenotype categories. Readers can visualize the 2003 brain PSCs via the BRIDGEPORT knowledge portal: https://labs-laboratory.com//bridgeport.…

• 

  Figure 6:. The gene-drug-disease network of the 2003 brain PSCs, 82 heart IDPs, and 84 eye IDPs The gene-drug-disease network reveals a broad spectrum of gene, drug, and disease interactions across the 2003 brain PSCs, 84 eye IDPs, 82 heart IDPs, and diseases beyond the three organs. The ICD code icons symbolize disease categories linked to the primary organ systems. All presented genes passed the FDR-corrected P-value threshold and were pharmaco-genetically associated with drug categories in the DrugBank database. Abbreviation: ICD: International Classification of Diseases. We showcased four representative drugs from the genes linked to the three organs.

• 

  Figure 7:. Genetic correlation and causality of the 2003 brain PSCs, 82 heart IDPs, and 84 eye IDPs a) The genetic correlation (gc) between the 2003 brain PSCs and 82 heart IDPs. Significant results after the Bonferroni correction (P-value<0.05/2003) are shown. With the application of a more stringent correction threshold (P-value < 0.05/2003/82), no statistical significance was observed. b) Significant genetic correlations (gc) between the 2003 brain PSCs and 84 eye IDPs after the Bonferroni correction (P-value<0.05/2003). With the application of a more stringent correction threshold (P-value < 0.05/2003/84), no statistical significance was observed. c) Genetic colocalization was evidenced at one genomic locus (2p22.2) between the mean myocardial wall thickness aha4 (WT_AHA_4) and C64_45 (PP.H4.ABF=0.98). d) Genetic colocalization was evidenced at one locus (3p21.31) between the right macular thickness at the inner temporal subfield (MitR) and C32_29. e) Genetic colocalization was evidenced at one locus (8p23.1) between the right inner nuclear layer thickness at the inner temporal subfield (InlR) and the LA stroke volume (LASV). f) The Brain2Heart causal relationship tested 2003 brain PSCs…

• 

  Figure 8:. Prediction ability of the 2003 brain PSCs and 82 heart IDPs for predicting 14 systemic diseases and 8 cognitive scores a) The classification for the 14 ICD-based disease categories was assessed using brain, heart, and eye features and their respective PRSs via support vector machines in a nested cross-validation (CV) procedure (i.e., training/validation/test datasets). We present the balanced accuracy from the CV results. b) Cognition prediction was assessed via support vector regression using brain, heart, and eye features, as well as their respective PRSs. Pearson’s r coefficient reflects the relationship between predicted values and the ground truth from the CV test datasets. Additional metrics, results using alternative machine learning methods, and outcomes from the independent test dataset can be found in Supplementary eFile 56-58. The size of the pie charts reflects the balanced accuracy achieved using specific feature sets. A more filled-in pie chart signifies higher accuracy.