Protein biomarkers of cardiac remodeling and inflammation associated with HFpEF and incident events

Figures

• 

  Figure 1. Methodologic approach to analysis of HFpEF-associated proteins. In step 1 univariate models with FDR adjustment were used to test associations between individual Olink proteins and HFpEF cases and non-HF controls in the CATHGEN discovery cohort. In step 2, significant proteins were tested in a multivariate adjusted model. In step 3, GSEA, validation in the TECOS cohort and LASSO were used to determine relevant HFpEF-associated proteins. In step 4, discriminative capabilities of multi-protein models were tested, as well as prognostic models for clinical outcomes. Abbreviations: Body Mass Index (BMI), Diabetes Mellitus (DM), False Discovery Rate (FDR), Gene Set Enrichment Analysis (GSEA), Heart Failure (HF), Heart Failure with Preserved Ejection Fraction (HFpEF), Hospitalization (hosp), Jackson Heart Study (JHS), Systolic Blood Pressure (SBP), Trial Evaluating Cardiovascular Outcomes with Sitagliptin (TECOS).

• 

  Figure 2. Proteins involved in cardiac remodeling, fibrosis and inflammation are associated with HFpEF. The volcano plot shows all 459 proteins analyzed in the study. Proteins with an odds ratio (OR) > 1.0 are upregulated in HFpEF cases compared to controls and proteins with an OR < 1.0 are downregulated. The 65 proteins significant in multivariate analyses in CATHGEN are shown in green. The 20 proteins that validated in the TECOS multivariate model are shown in orange.

• 

  Figure 3. Receiver Operator Characteristic Curve for discrimination of HFpEF in CATHGEN. Shown are four Receiver Operator Characteristic (ROC) curves for discrimination of HFpEF with NT-proBNP alone, the addition of the clinical adjusted clinical model (age, race, body mass index (BMI), systolic blood pressure (SBP), type 2 diabetes mellitus (DM) and creatinine) and the addition of both the 20 validating protein model and the LASSO with Monte Carlo cross-validation 13-protein biomarker model. The 20 validating proteins are: LCN2, U-PAR, IL-1ra, KIM1, CSTB, Gal-9, TIMP-1, MATN2, XCL1, NEMO, CRELD2, MMP7, SUMF2, TXNDC5, PRELP, TMSB10, CDH2, PARP-1, SPINT2, FCGR3B. The 13 LASSO selected proteins were AOC3, CA5A, CA6, CCL16, CLSTN2 , Gal-9, IGFBP3, KIM1, NEMO, PSGL-1, SERPINA12, SUMF2, VEGFD. The addition of either the 20 or 13-protein model to the adjusted model and NT-proBNP improved the discriminative capabilities of the model.

• 

  Figure 4. HFpEF-proteins associated with incident outcomes. The forest plots show the HR and 95% CI for the significant proteins associated with (A) incident HF hospitalization in CATHGEN and TECOs; (B) incident HFpEF hospitalization in JHS; (C) all-cause mortality in CATHGEN, TECOS and JHS.

• 

  Figure 5. Kaplan–Meier Curve for All-Cause Mortality by KIM1 level in in CATHGEN. The Kaplan–Meier curve shows overall survival for CATHGEN participants by KIM1 levels. KIM1 was the protein most strongly associated in the meta-analysis for all-cause mortality.

• 

  Figure 6. Circulating protein biomarkers associate with HFpEF, improve discriminative capabilities and predict incident events. Assessment of 459 unique Olink proteins in the CATHGEN discovery cohort followed by LASSO and validation in TECOS led to the identification of 29 HFpEF-associated proteins. These proteins were markers of cardiac remodeling, inflammation, metabolism, angiogenesis and fibrosis22. GSEA further identified pathways of inflammation and fatty-acid metabolism to be dysregulated. These proteins improved discriminative capabilities for HFpEF prediction and were associated with incident clinical outcomes.