Mendelian randomization and pathway analysis demonstrate shared genetic associations between lupus and coronary artery disease

Figures

• 

  Figure 1. MR demonstrates a positive association of effect sizes of SLE-associated non-HLA SNPs on SLE and CAD (A) Graphical depiction of the two-stage approach for an initial exploratory analysis using expanded groups of SNPs as IVs followed by a confirmatory analysis using highly curated IVs. (B and C) Forest plots of six MR causal estimates (β ± standard error). For results, gray indicates insignificant (p > 0.05) and red indicates positive causal estimates determined by each MR method. (B) Immunochip-derived SLE-associated non-HLA SNPs were used as IVs for SLE; summary statistics from both the SLE Immunochip study (left) and SLE GWAS (right) were used for the exposure; summary statistics from the CAD GWAS were used for the outcome. (C) Additional MR analyses using SNPs associated (p < 10−6) with SLE in the Immunochip and GWAS study (rows 1 and 2) or Phenoscanner reaching genome-wide significance (p < 5 × 10−8, row 3) were used as IVs; summary statistics for the exposure and outcome are indicated. MR analyses excluding the entire short arm of chromosome 6 and excluding only the extended HLA region (chr6: 27–34 Mb, left columns). Right columns show MR analyses using the same sets of SNPs…

• 

  Figure 2. MR demonstrates a net positive causal effect of SLE-associated non-HLA SNPs on CAD (A) MR diagram for testing the causal effects of SLE on CAD with respect to instrument relevance to the exposure, exclusion from the outcomes (i.e., CAD, MI, IS), and independence from confounding factors. LD clumping (R2 < 0.001) was used to obtain independent IVs. (B) Forest plots of MR causal estimates (β ± standard error) for SLE on CAD (CAD-a, CAD-b), MI (MI-a, MI-b), IS, cardiomyopathy (CM), and atrial fibrillation (AFib) GWAS using 16 MR methods. Missing PRESSO-OC (outlier correction) estimates indicate insignificant global tests for horizontal pleiotropy. For results, gray indicates insignificant (p > 0.05), red indicates positive (p < 0.05), and blue indicates negative (p < 0.05) causal estimates determined by each MR method. Numbers within forest plots indicate the SNPs used as IVs after harmonization. NOME, no measurement error. (C) Application of S-LDSC using summary statistics for SLE, CVD, and CAD GWAS to estimate the heritability (coefficient ± standard error) of the 284 SNP-predicted genes (top) and 160 SNP-predicted proteins from STRINGdb (bottom). Bar color indicates coefficient…

• 

  Figure 3. Analysis of SLE-associated SNP-predicted genes with causal effects on CAD by single-SNP MR (A and B) Forest plots (β ± standard error) of the top 25 (by absolute value of causal estimates) positive (A) and negative (B) causal non-HLA SNPs identified by single-SNP MR (SSMR) using the Wald ratio method. (C and E) Pie charts illustrating the chromosomal distribution of 80 positive (C) and 96 negative (E) causal SLE SNPs on CAD. (D and F) Cluster metastructures for the 200 (D) and 184 (F) predicted genes from positive and negative causal SNPs identified by single-SNP MR. Functional and cell-type enrichments for each cluster were determined using BIG-C (black labels) and I-scope (red labels), respectively. Bold black labels over colored shadings represent shared functional annotations for the clusters they surround. (G and H) S-LDSC using summary statistics for SLE, CVD, and CAD GWAS to estimate the heritability (coefficient ± standard error) of genes (open bars) and SNP-predicted proteins (hashed bars) predicted by positive (G) and negative (H) causal SNPs determined by SSMR. Bar color indicates coefficient significance.

• 

  Figure 4. SLE-derived gene network with causal implications for CAD and PPI-based MR (A) S-LDSC using summary statistics for SLE, CVD, and CAD GWAS to estimate the heritability (coefficient ± standard error) of the 2,336 genes (open bars) and 1,501 proteins (hashed bars) predicted by 838 Immunochip SNPs associated with SLE. Bar color indicates coefficient significance. (B) Functional and cell-type enrichments for cluster metastructures were determined using BIG-C (black labels) and I-scope (red labels), respectively. Bold black labels over colored shadings represent shared BIG-C functional annotations for the clusters they surround. Node size is proportional to the number of SNPs (height) mapping to the genes in each cluster (width). For node color, red and blue indicate significant positive or negative estimates, respectively for 14/16 MR methods used (tier 1); light red and light blue indicate significant positive or negative estimates by MR-IVW or at least 7/16 MR methods (tier 2); gray indicates insignificant. Thickness of the yellow border is roughly proportional to the negative log of the MR-IVW p value. Green border indicates clusters with −log(MR-IVW p value) > 3. (C) Forest plots…

• 

  Figure 5. Comprehensive PPI-based MR predicts sets of SLE-associated variants and pathways causal of CAD (A) S-LDSC using GWAS summary statistics for SLE, CVD, and CAD to estimate the heritability (coefficient ± standard error) of the 3,272 genes (open bars) and 1,972 protein-coding genes (hashed bars) predicted by 1,708 combined Immunochip- and Phenoscanner-derived SNPs. Bar color indicates coefficient significance. (B) Functional and cell-type enrichments for cluster metastructures were determined using BIG-C (black labels) and I-scope (red labels), respectively. Bold black labels over colored shadings represent shared BIG-C functional annotations for the clusters they surround. Node size is proportional to the number of SNPs (height) mapping to the genes in each cluster (width). For node color, red and blue indicate significant positive or negative estimates, respectively for 14/16 MR methods used (tier 1); light red and light blue indicate significant positive or negative estimates by MR-IVW or at least 7/16 MR methods (tier 2); purple indicates mixed estimates; gray indicates insignificant. Thickness of the yellow border is roughly proportional to the negative log of the MR-IVW p…

• 

  Figure 6. PPI-based MR identifies SLE SNPs with positive and negative causal effects on CAD (A) Workflow depicting PPI-based MR. (B and C) PPI-based MR validation. Forest plots (β ± standard error) from 16 MR methods using summary statistics from the SLE GWAS (B) or SLE Immunochip (C) as the exposure and CAD GWAS as the outcome. SLE-associated non-HLA SNPs mapping to positive and negative clusters, separately (by tier) and together (“All SNPs”) were used as IVs after excluding CVD and confounder-associated SNPs followed by stringent LD clumping (R2 = 0.001) and harmonization. The number of SNPs used as IVs for each SNP set are indicated in the plots. For results, gray indicates insignificant (p > 0.05), dark red indicates positive (p < 0.00075), red indicates positive (p < 0.05), dark blue indicates negative (p < 0.00075), and blue indicates negative (p < 0.05) by each MR method.

• 

  Figure 7. Genes and molecular pathways associated with positive causal clusters identify therapeutic interventions for managing CAD in SLE (A) All tier 1 and a selection of tier 2 clusters were functionally annotated using BIG-C, IPA, and the EnrichR database. Select drugs acting on direct gene targets or on any of the associated pathways (italics) are listed. (B) Venn diagram summarizing therapies that might uniquely impact SLE or CAD and those that may target pathways common to both diseases.