Fig 1.
GWAS imputation of SNP variances in CF patients (n = 6,365) were used to impute genetically regulated gene expression, which were then tested for CF lung disease severity using either the PrediXcan platform (left arm), or TWAS (right arm). The association results from multiple tissues from each platform were combined through 2 different meta-analysis of multiple p-values from different tissues. GTEx: Genotype-Tissue Expression RNA-seq (n = 48 tissues); CF: LCL microarray (n = 753 samples), and nasal epithelial biopsy RNA-seq (n = 132 samples); DGN: Depression Genes and Networks RNA-seq from whole blood (n = 922 samples); HMP: harmonic mean p-value; EBM: empirical adaptation of Brown’s method; OMNIBUS: omnibus p-value from TWAS.
Fig 2.
Hierarchical clustering of genes whose imputed expression are associated with CF lung disease severity.
Consensus modifier genes (n = 52) were determined as p-value < 0.01 from all 4 meta-analyses of multiple tissue association testing described in methods, and the -log10(p-values) were clustered and represented as a heatmap with red-grey-blue color scale. The color represents direction of predicted expression change, with red indicates “protective”, or increased expression with increasing KNoRMA (milder lung disease), and blue, “harmful”, or increased expression with decreasing KNoRMA (more severe lung disease), and the intensity reflects the significance (p-values) of the association. White cells in heatmap indicate missing data, where the genes were not well predicted from the relevant tissues. The vertical color columns on the right indicate type of gene and chromosome near GWAS loci. The genes were clustered based on results from PrediXcan (left heatmap), and the order of the genes were kept the same for TWAS (right heatmap). Key patterns of negative and positive associations to KNoRMA across multiple tissues in the heatmap are highlighted by the dashed boxes. Arrows on top of the left heatmap identify the additional tissues over the 48 GTEx tissues common to both platforms, and arrows in the middle of the heatmaps show the results from whole blood tissues for TPPP.
Fig 3.
Manhattan plots of CF lung disease association p-values from gene expression imputation and GWAS.
Maximal p-values between 2 meta-analyses from imputed gene expression to KNoRMA by PrediXcan and TWAS were used in the Manhattan plots A and B respectively. The 28 consensus modifier genes within 1 Mb of 5 autosomal GWAS signals (red squares), and those not near GWAS signals (blue triangles) are labeled. Panel C represents GWAS p-values from the updated imputation [78] by fixed-effect meta-analysis performed according to the GWAS study [1]. The solid lines correspond to genome-wide significant p-value of 0.01 (for imputed expression, A and B) or 1.25x10-08 (for GWAS, C), while the dashed lines represent the suggestive p-value of 0.05 (for imputed expression) or 1x10-06 (for GWAS).
Table 1.
Consensus 52 CF lung disease modifier genes.
Fig 4.
Correlation of imputed gene expression association from PrediXcan and minimal GWAS association p-values.
Maximal p-values between HMP and EBM meta-analyses of CF lung disease associations from imputed gene expression (PrediXcan) for 26,750 genes from 48 GTEx tissues are plotted against minimal GWAS SNP p-values per gene among all cis-SNPs used in predictive models. The 52 consensus modifier genes are highlighted in red squares (near GWAS loci) and blue triangles (novel), while genes with minimal GWAS SNP p-values < x10-08 (dashed vertical line), but not among the 52, are highlighted in black diamonds. Solid line represents linear regression.
Fig 5.
Comparison of predictive model SNPs at chromosome 11 GWAS locus.
The -log10 p-values from GWAS analysis were retrieved for cis-SNPs in viable PrediXcan predictive models from 48 GTEx tissues for EHF, APIP, and PDHX. These p-values were formatted as bedGraph files and displayed through the UCSC genome browser (http://genome.ucsc.edu/) as custom annotation tracks, with vertical scales set between 0 and 10. The screenshot of the genome browser shows from top to bottom: GWAS SNP p-values, SNPs used in EHF gene expression imputation model, those for APIP, PDHX, and gene annotation from NCBI RefSeq genes.
Fig 6.
Gene set enrichment analyses (GSEA) were performed and enrichment plots were generated for selected gene sets using the Bioconductor R package, fgsea. For each enrichment plot, the horizontal black line at the bottom represent p-value ranks of protein-coding genes with most significant p-value rank on the left. The vertical bars represent individual genes in a gene set and their ranks. The green curves represent the cumulative enrichment score (ES), and the red horizontal dashed lines denote minimal (often 0) and maximal scores. Listed genes represent the leading edge with increasing ES, that contribute to the overall enrichment of the gene set. Panel A and C are GSEA results from PrediXcan platform, while B and D from TWAS. Particular gene sets shown are from GO biological process (A), and Biosystems (C–KEGG, B, D–Reactome).
Table 2.
Functional categories of significant genes (n = 149 out of 379) relevant to CF pathophysiology*.
Fig 7.
Effect of allele bias on gene expression quantification and disease phenotype association in CF nasal epithelial biopsy RNA-seq data set.
Comparison of CF lung disease (KNoRMA) association t statistics between different mapping protocols among 1,379 common imputable genes by respective predictive models among 5,634 unrelated CF patients are shown in A. HLA genes in A, are represented as red triangles, and x-axis represent standard and y-axis alternative mapping protocols. Panels B and C show gene expression quantifications by standard (x-axis) and alternative (y-axis) protocols in the format of FPKM for HLA-DQB1, and HLA-DRB1 genes. Each dot represents 1 sample (out of 132 total), with solid line denoting linear regression line, and dashed line representing equality.