Proteomic Associations of Adverse Outcomes in Human Heart Failure

Abstract

Background: Identifying novel molecular drivers of disease progression in heart failure (HF) is a high-priority goal that may provide new therapeutic targets to improve patient outcomes. The authors investigated the relationship between plasma proteins and adverse outcomes in HF and their putative causal role using Mendelian randomization.

Methods and results: The authors measured 4776 plasma proteins among 1964 participants with HF with a reduced left ventricular ejection fraction enrolled in PHFS (Penn Heart Failure Study). Assessed were the observational relationship between plasma proteins and (1) all-cause death or (2) death or HF-related hospital admission (DHFA). The authors replicated nominally significant associations in the Washington University HF registry (N=1080). Proteins significantly associated with outcomes were the subject of 2-sample Mendelian randomization and colocalization analyses. After correction for multiple testing, 243 and 126 proteins were found to be significantly associated with death and DHFA, respectively. These included small ubiquitin-like modifier 2 (standardized hazard ratio [sHR], 1.56; P<0.0001), growth differentiation factor-15 (sHR, 1.68; P<0.0001) for death, A disintegrin and metalloproteinase with thrombospondin motifs-like protein (sHR, 1.40; P<0.0001), and pulmonary-associated surfactant protein C (sHR, 1.24; P<0.0001) for DHFA. In pathway analyses, top canonical pathways associated with death and DHFA included fibrotic, inflammatory, and coagulation pathways. Genomic analyses provided evidence of nominally significant associations between levels of 6 genetically predicted proteins with DHFA and 11 genetically predicted proteins with death.

Conclusions: This study implicates multiple novel proteins in HF and provides preliminary evidence of associations between genetically predicted plasma levels of 17 candidate proteins and the risk for adverse outcomes in human HF.

Keywords: HFrEF; Mendelian randomization; heart failure; proteomics.

Figure 1. Proteome‐wide association study of 4776 proteins for death, and death or heart failure–related hospital admission (DHFA) outcomes: a Mendelian randomization framework.

PHFS indicates Penn Heart Failure Study; pQTL, protein quantitative trait loci; and WashU, Washington University.

Figure 2. Volcano plots demonstrating associations between all plasma proteins measured in PHFS (Penn Heart Failure Study) with death or heart failure–related hospital admission (DHFA) in the unadjusted Cox model (A), DHFA in the adjusted Cox model (B), all‐cause mortality in the unadjusted Cox model (C), and all‐cause mortality in the adjusted Cox model (D).

The plots show standardized hazard ratios against the principal components analysis–corrected log‐10 P value, to better visualize the importance of each biomarker in order of significance. The nominal significance level and the alpha‐corrected significance level are represented by dashed lines on the y axis. Only the top 100 proteins positively or negatively associated with death are labeled. Adjusted models include 1889 participants who had sufficient data for computation of the baseline Meta‐Analysis Global Group in Chronic Heart Failure (MAGGIC) risk score and had follow‐up data. Unadjusted models include 2209 for all‐cause mortality and 2234 for DHFA participants with baseline proteomics data and follow‐up data.

Figure 3. Concordance between standardized hazard ratios for proteome‐wide associations for death or heart failure–related hospital admission (DHFA) in the unadjusted Cox models (A), DHFA in the adjusted Cox models (B), all‐cause mortality in the unadjusted Cox models (C), and all‐cause mortality in the adjusted Cox models (D) for participants of the PHFS (Penn Heart Failure Study) and Washington University (WashU) heart failure registry

Figure 4. Canonical pathway analysis of proteins observed to be significantly associated with death or heart failure–related hospital admission (DHFA) in the unadjusted model (A), DHFA in the adjusted model (B), death in the unadjusted model (C), and death in the adjusted model (D) in PHFS (Penn Heart Failure Study).

Principal components analysis–corrected P value of 0.05 was used to determine significance. Numbers indicate the Z score corresponding to the direction and strength of each association. BMP indicates bone morphogenetic protein; cAMP, nicotinamide adenine dinucleotide; CLEAR, Coordinated Lysosomal Expression and Regulation; FXR, farnesoid X receptor; GABA, gamma‐aminobutyric acid; GP6, glycoprotein VI platelet; IL, interleukin; JAK2, Janus kinase 2; LPS, lipopolysaccharide; LXR, liver X receptor; NADH, reduced nicotinamide adenine dinucleotide; and RXR, retinoid X receptor.

Figure 5. Proteome‐wide cis‐Mendelian randomization (MR) study results for associations between genetically instrumented proteins from the deCODE and Fenland studies and death, and death or heart failure–related hospital admission (DHFA) in PHFS (Penn Heart Failure Study).

Log (odds ratio [OR]) regression estimates are plotted against −log (P value). The dashed line represents the 5% alpha threshold. ADH7 indicates alcohol dehydrogenase 7; ANG, angiogenin; ATOX1, antioxidant 1 copper chaperone; CCDC126, coiled‐coil domain containing 126; CD55, complement decay‐accelerating factor; CCL14, chemokine (C‐C motif) ligand 14; FCN2, ficolin 2; FGF23, fibroblast growth factor 23; HPGDS, hematopoietic prostaglandin D synthase; IGLL1, immunoglobulin lambda like polypeptide 1; SVEP1, sushi, von Willebrand factor type A, EGF and pentraxin domain containing 1; NEGR1, neuronal growth factor 1; STC1, stanniocalcin 1; and RSPO4, R‐spondin 4.

Figure 6. Two‐sample Mendelian randomization effect estimates per 1‐SD higher circulating plasma protein levels for death or heart failure–related hospital admission (DHFA) (A) and death (B).

Only associations between genetically predicted proteins and heart failure–related outcomes with statistically significant estimates (P<0.05) in either deCODE or Fenland cohorts are reported. ADH7 indicates alcohol dehydrogenase 7; ANG, angiogenin; ATOX1, antioxidant 1 copper chaperone; CCDC126, coiled‐coil domain containing 126; CD55, complement decay‐accelerating factor; CCL14, chemokine (C‐C motif) ligand 14; FCN2, ficolin 2; FGF23, fibroblast growth factor 23; HPGDS, hematopoietic prostaglandin D synthase; IGLL1, immunoglobulin lambda like polypeptide 1; SVEP1, sushi, von Willebrand factor type A, EGF and pentraxin domain containing 1; NEGR1, neuronal growth factor 1; OR, odds ratio; STC1, stanniocalcin 1; and RSPO4, R‐spondin 4.

Figure 7. Circos plot showing the relationships between heart failure–related outcomes and significantly associated proteins from cis‐Mendelian randomization.

ADH7 indicates alcohol dehydrogenase 7; ANG, angiogenin; ATOX1, antioxidant 1 copper chaperone; CCDC126, coiled‐coil domain containing 126; CD55, complement decay‐accelerating factor; CCL14, chemokine (C‐C motif) ligand 14; DHFA, death or heart failure–related hospital admission; FCN2, ficolin 2; FGF23, fibroblast growth factor 23; HPGDS, hematopoietic prostaglandin D synthase; IGLL1, immunoglobulin lambda like polypeptide 1; SVEP1, sushi, von Willebrand factor type A, EGF and pentraxin domain containing 1; NEGR1, neuronal growth factor 1; OR, odds ratio; STC1, stanniocalcin 1; and RSPO4, R‐spondin 4.

Figure 8. Summary of proteome‐wide Mendelian randomization (MR) analyses for heart failure adverse outcomes death, and death or heart failure–related hospital admission (DHFA).

Proteins that have detrimental effects are shown in bold and in red. ADH7 indicates alcohol dehydrogenase 7; ANG, angiogenin; ANTXR2, anthrax toxin receptor 2; ATOX1, antioxidant 1 copper chaperone; BAMBI, BMP and activin membrane‐bound inhibitor; CBR1, carbonyl reductase 1; CCDC126, coiled‐coil domain containing 126; CD55, complement decay‐accelerating factor; CCL14, chemokine (C‐C motif) ligand 14; CHST15, carbohydrate sulfotransferase 15; CNTF, ciliary neurotrophic factor; CRK, Crk proto‐oncogene adaptor protein; CT55, cancer/testis antigen 55; DIXDC1, Dix domain containing 1; DPYSL3, dihydropyrimidinase like 3; EDA, ectodysplasin A; EFEMP1, EGF‐containing fibulin‐like extracellular matrix protein 1; EFNB1, ephrin B1; EPHA4, erythropoietin‐producing hepatoma receptor A4; FABP3, fatty acid binding protein 3; FCN2, ficolin 2; FGF23, fibroblast growth factor 23; FBLN5, fibulin 5; HPGDS, hematopoietic prostaglandin D synthase; IGLL1, immunoglobulin lambda like polypeptide 1; IQCF3, IQ motif containing F3; KRT1, keratin 1; LILRA5, leukocyte immunoglobulin receptor A5; NEGR1, neuronal growth factor 1; OIT3, oncoprotein induced transcript 3; OR, odds ratio; PCDHGA1, protocadherin gamma subfamily A, 1; PLA2G12, phospholipase A2 group XIIA; PSD2, pleckstrin and Sec7 domain containing 2; RANBP3, Ras‐related nuclear protein binding protein; SET, SET nuclear proto‐oncogene; STC1, stanniocalcin 1; RNASE4, ribonuclease A family member 4; RSPO4, R‐spondin 4; SPON1, spondin 1; ST3GAL2, ST3 beta‐galactoside alpha‐2,3‐sialyltransferase 2; STX3, syntaxin 3; SVEP1, sushi, von Willebrand factor type A, EGF and pentraxin domain containing 1; TNNI3, troponin I3, cardiac type; and WFDC2, WAP four‐disulfide core domain 2.