Mendelian Randomization of blood proteins in heart failure
We conducted a two-sample MR analysis to identify causally associated blood proteins with HF. Summary level data from INTERVAL[14], a study including pQTLs for 2,965 different blood proteins measured in 3,301 individuals, were leveraged to identify cis-acting gene variants as instrumental variables (IVs). A minimum of 3 independent (r2 < 0.1) gene variants within a window of 500kb were selected to identify IVs for the blood proteins (exposure). GWAS data from the Heart Failure Molecular Epidemiology for Therapeutic Targets Consortium (HERMES)[3], a meta-analysis including 47,309 HF cases and 930,014 controls of European ancestry, were assessed as the outcome. There were enough instruments to perform 822 cis-MR analyses. In inverse variance weighted (IVW) MR, we identified at a false discovery rate (FDR) of 5% 19 blood proteins that were significantly associated with HF (ABO, BAG3, FLT4, TDGF1, FUT3, FSTL1, ALDH3A1, GLCE, PTHLH, CDON, FCGR2A, RGMB, AMH, MIF, IL15RA, B3GAT3, CCDC126, ST3GAL6, APOA5) (Fig. 1) (Suppl. Table 1). Among those proteins, (OR per 1 SD) ABO (OR: 1.03, 95%CI: 1.02–1.04, PIVW=5.89E-13), BAG3 (OR: 0.79, 95%CI: 0.74–0.85, PIVW=2.59E-09) and FLT4 (OR: 1.08, 95%CI: 1.04–1.12, PIVW=3.34E-05) were significant after a Bonferroni correction (P < 6.08E-05, 0.05/822). We carried out the Cochran’s Q and Egger intercept tests to detect horizontal pleiotropy[15, 16]. Both the Cochran’s Q and intercept tests did not reveal heterogeneity or horizontal pleiotropy for the 19 blood proteins (Suppl. Table 1). Thus, the 19 blood proteins (FDR < 0.05) were considered as causal molecular candidates for downstream analyses. Among the causal candidates, 7 (ABO, FLT4, PTHLH, MIF, IL15RA, B3GAT3, CCDC126) were positively associated with HF, whereas 12 (BAG3, TDGF1, FUT3, FSTL1, ALDH3A1, GLCE, CDON, FCGR2A, RGMB, AMH, ST3GAL6, APOA5) were negatively associated with the risk of HF. The blood proteins with the largest negative and positive effect size on HF were BAG3 (OR: 0.79, 95%CI: 0.74–0.85, PIVW=2.59E-09) and MIF (OR: 1.19, 95%CI: 1.08–1.32, PIVW=5.53E-04), respectively.
As MR is subject to pleiotropy of the IVs, we performed sensitivity analysis. In weighted median MR, which is robust to invalid instruments[17] (variants with horizontal pleiotropy), 17 blood proteins (ABO, BAG3, CDON, APOA5, CCDC126, FLT4, IL15RA, ALDH3A1, PTHLH, RGMB, AMH, GLCE, TDGF1, FSTL1, FCGR2A, B3GAT3, MIF) remained significantly associated with HF (Suppl. Table 1). The directional effects were concordant between weighted median MR and IVW MR.
Enrichment and network pathway analysis
We aimed to identify the functional and pathway enrichments of candidate causal blood proteins. Among the causal candidates, 5 blood proteins (ABO, FUT3, GLCE, B3GAT3 and ST3GAL6) were classified as molecules involved in glycosylation in the Comprehensive GlycoEnzyme Database (GlycoEnzDB) (fold-enrichment 13.1, P = 1.40E-06, hypergeometric test). Of note, FUT3 (OR: 0.97, 95%CI: 0.96–0.98, PIVW=8.68E-05) and ST3GAL6 (OR: 0.97, 95%CI: 0.96–0.99, PIVW=1.06E-03) were both negatively associated with the risk of HF. FUT3 and ST3GAL6 are key enzymes leading to the generation of sialyl Lewis x, a glycan moiety decorating membrane and circulating proteins[18, 19] (Suppl. Figure 1). We next hypothesized that some of the causal candidate proteins may be involved in different ligand receptor interactions. By using a comprehensive repository of ligand-receptor interactions reported by Shao et al.[20], we found that blood proteins associated with HF in MR were enriched in ligand and receptors (fold-enrichment 5.7, P = 4.0E-07, hypergeometric test). These molecules may contribute to 43 different ligand-receptor pairs (Suppl. Table 2). The 43 ligand-receptor pairs were enriched in Gene Ontology (GO) (molecular function) for transmembrane receptor protein serine/threonine kinase activity (P = 3.68E-12), G protein-coupled receptor activity (P = 1.08E-10), patched binding (P = 5.43E-07), activin-activated receptor activity (P = 1.30E-06) and transforming growth factor beta-activated receptor activity (P = 3.38E-06) (Suppl. Figure 2) (Suppl. Table 3).
We next performed a pathway analysis by using a network approach. Protein interaction data from InnateDB, which includes more than 19,800 curated protein interactions, was leveraged to infer a blood protein network[21]. The 19 causal candidates proteins were used as seeds to generate a network including 155 nodes (proteins) and 160 edges (interactions) (Fig. 2A) (Suppl. Table 4). The causally associated blood proteins were overrepresented in the nodes (proteins) with the highest degree (≥ 90th percentile) (fold-enrichment 3.6, P = 7.26E-06, hypergeometric test). The top nodes (proteins) acting as hub molecules include MIF, BAG3, FSTL1, FCGR2A, TDGF1, FLT4, PTHLH, IL15RA, AMH and ALDH3A1. We interrogated the Kyoto Encyclopedia of Genes and Genomes (KEGG)[22] to perform a pathway enrichment analysis of the network. The highest enrichments were pathways in cancer (P = 1.03E-17), NF-kappa B signaling (P = 2.99E-13), TGF-beta signaling (P = 3.83E-13), lipid and atherogenesis (P = 9.53E-13) as well as fluid shear stress and atherosclerosis (P = 9.96E-12) (Fig. 2B) (Suppl. Table 5).
Cross-Phenotype analysis
Cross-phenotype association analysis were performed for the genetic association data of HF by using the interactive cross phenotype analysis of GWAS database (iCPAG)[23] which provides enrichment and similarity metrics between traits by using an exhaustive list of ancestry LD-specific association data from the NHGRI-EBI GWAS catalog[24]. After a Bonferroni correction, this analysis showed that 75 disorders and traits were significantly associated to HF (Suppl. Table 6). Figure 3 shows the highest enrichment between HF and traits-disorders. According to iCPAG, the highest enrichment were for beta blocking agent use (P = 1.36E-24), coronary artery disease (P = 2.70E-23), low-density cholesterol (P = 9.09E-21), myocardial infarction (P = 4.87E-20), apolipoprotein B (P = 1.24E-19) and parental longevity (P = 1.28E-19).
Multi-trait and multivariable MR analyses
Considering the cross-phenotype analysis showing a genetic overlap between HF and several cardiovascular disorder related traits, we performed a multi-traits MR analysis for the 19 causal blood candidate proteins. We leveraged 31 different GWAS covering 7 disease categories (atopic, autoimmune, cancer, cardiovascular, infectious, metabolic and neurologic) as outcomes. Figure 4 illustrates the MR analysis including the directional effect and the significance (-logP) between the 19 blood proteins (exposure) and the different disorder-traits (outcomes). Some blood proteins such as ABO and FCGR2A show significant association with several traits often with opposite directional effects (antagonist pleiotropy) with HF. However, some proteins such as BAG3, MIF and APOA5 show concordant associations between HF, the blood pressure (BAG3) and coronary artery disease (MIF and APOA5). BAG3 is significantly and negatively associated with systolic and diastolic blood pressure, whereas MIF and APOA5 are associated with coronary artery disease. We performed mediation analysis by using multivariable MR corrected for the exposure to cardiovascular traits. Following the correction for the diastolic blood pressure, the association between BAG3 and HF was not significant (Suppl. Table 7). Also, after correction for coronary artery disease the associations between MIF and APOA5 with HF were no longer significant (Suppl. Table 7). Taken together, these data suggest that, at least in part, the protective effect of circulating BAG3 is mediated by a reduction of the diastolic blood pressure. On the other hand, the impact of MIF and APOA5 on HF are likely mediated by the risk of coronary artery disease, a leading cause of HF[25, 26].
Drug target analysis
We leveraged several resources to carry out a drug target analysis. We investigated the blood proteins in order to document if they represent targets for licensed, in-development small molecules or biologics. In the Therapeutic Target Database (TTD)[27], 6 blood proteins (ABO, FLT4, FCGR2A, AMH, MIF, IL15RA) are reported as either clinical trial target or successful targets (Suppl. Table 8). In the Drug Gene Interaction Database (DGIdb)[28], a total of 111 drug-target pairs were reported for FLT4, ALDH3A1, PTHLH, FCGR2A, AMH, MIF, IL15RA and APOA5 (Suppl. Table 9). Several kinase inhibitors targeting FLT4, a blood protein positively associated with the risk of HF, are approved for the treatment of cancer. In the Open Targets database[29, 30], MIF, PTHLH, FLT4 and TDGF1 were reported as targets of approved and in-development drugs as well as antibodies (Suppl. Table 10). In the blood, MIF is positively associated with the risk of HF and is a target for Imalumab and Iguratimod, respectively an antibody and a small molecule inhibitor. Iguratimod is licensed in Japan for the treatment of rheumatoid arthritis[31], whereas Imalumab is a phase 1 monoclonal antibody destined for patients with solid tumors[32]. According to Open Targets, 15 blood proteins (MIF, CCDC126, IL15RA, FCGR2A, CDON, ALDH3A1, FSTL1, APOA5, AMH, BAG3, B3GAT3, ST3GAL6, FUT3, GLCE, RGMB) are deemed tractable for the development of antibodies (Suppl. Table 10). Taken together, these data suggest that according to the directional effect MIF, FLT4, PTHLH, ABO, CCDC126, IL15RA, and B3GAT3 (blood proteins positively associated with HF) are potential targets for HF as they are the object of approved, in-development inhibitors or deemed tractable for the development of novel inhibitors (antibodies).