Biobank-Scale Plasma Proteomics Identifies Novel Biomarkers in Hypertrophic Cardiomyopathy

Figures

• 

  Figure 1.. Overview of analyses and data sets involved. Hypertrophic cardiomyopathy (HCM) cases in UK Biobank, deCODE Genetics, and All of Us data sets were divided into prevalent cases (those diagnosed before the blood sample collection date or within 5 years) and incident cases (those diagnosed after the blood sample collection date +5 years). This represents an assumed potential lag between disease manifestation and clinical diagnosis. HCMR indicates Hypertrophic Cardiomyopathy Registry.

• 

  Figure 2.. Plasma proteomic association analyses in UK Biobank identified known and novel hypertrophic cardiomyopathy (HCM) biomarkers. A, Case-control differential protein expression analysis using prevalent HCM cases identified previously reported biomarkers of disease (eg, NT-proBNP [N-terminal pro-B-type natriuretic peptide]) and novel ones (ANGPT2 [angiopoietin-2] and LTBP2 [latent transforming growth factor-beta binding protein 2]). B, Time-to-event analysis of incident HCM diagnosis in the controls from above analysis (which includes incident HCM cases) provided orthogonal support for NT-proBNP’s association with disease. Multiple testing correction (MTC) applied via Benjamini-Hochberg procedure to control false discovery rate at 5%.

• 

  Figure 3.. Known and novel biomarkers were validated in external deCODE Genetics and All of Us data sets. Case-control differential protein expression analysis in external replication (A) DeCODE Genetics and (B) All of Us data sets validated known hypertrophic cardiomyopathy (HCM) associations of biomarkers NT-proBNP (N-terminal pro-B-type natriuretic peptide) and Troponin I (TNNI3), and novel biomarker ANGPT2 (angiopoietin-2). Plasma proteins tested in external replication deCODE Genetics data set included all those significant at the 5% FDR threshold (Figure 2A) if they were also measured in the SomaScan V4 study. Multiple testing correction (MTC) applied via Benjamini-Hochberg procedure to control false discovery rate at 5% (dashed horizontal line).

• 

  Figure 4.. Known and novel biomarkers showed potential prognostic utility. A, Multivariable Cox regression with incident adverse clinical outcomes in UK Biobank showed associations of known (NT-proBNP [N-terminal pro-B-type natriuretic peptide]) and novel (ANGPT2 (angiopoietin-2) and LTBP2 [latent transforming growth factor-beta binding protein 2]) biomarkers with composite outcomes after covariate adjustment. Error bars indicate 95% CIs. Multiple testing correction applied via Benjamini-Hochberg procedure to control false discovery rate at 5%. B through E, Kaplan-Meier event curves with stratification by (B and C) NT-proBNP (with overall and heart failure [HF] composites, respectively), (D) ANGPT2 and (E) LTBP2 levels (both with HF composite) at baseline (threshold at 75th percentile: 2.06, 1.2, and 1.23 SD, respectively) showed significant difference in event incidence between <75th percentile and ≥75th percentile subgroups for biomarker-composite combinations before covariate adjustment. Statistical test for difference in subgroups via log-rank test. VA indicates ventricular arrhythmia.

• 

  Figure 5.. Mendelian randomization analyses established causal associations between hypertrophic cardiomyopathy (HCM) and known biomarkers. A, Bidirectional, 2-sample Mendelian randomization analyses showed significant causality between HCM disease status and plasma proteins (X) prioritized by case-control association analyses. Multiple testing correction (MTC) applied via Benjamini-Hochberg procedure to control false discovery rate at 5%. Associations labeled if significant after MTC. B, Forest plot of significant conclusions indicated that common genetic risk of HCM causes increased NT-proBNP (N-terminal pro-B-type natriuretic peptide) and Troponin T (TNNT2) levels. Error bars represent 95% CIs. HCMR indicates Hypertrophic Cardiomyopathy Registry; and UKB, UK Biobank.

• 

  Figure 6.. Machine learning-based joint modeling of prioritized plasma proteins highlighted additive predictive contributions of known and novel biomarkers. A, Shapley additive explanation (SHAP) values quantify how much a specific protein’s measured level pushed the model’s prediction for an individual toward or away from a case prediction. Mean absolute SHAP values (unit=log[odds of hypertrophic cardiomyopathy]) for each plasma protein for prevalent hypertrophic cardiomyopathy (HCM) cases and controls showed the relative feature importance and impact on prediction by trained XGBoost classifier. NT-proBNP (N-terminal pro-B-type natriuretic peptide) has the greatest predictive contribution in both cases and controls. Error bars represent 95% CIs computed by bootstrapping the evaluated data set. B and C, Beeswarm plots of SHAP values for each plasma protein stratified for (B) cases and (C) controls indicates the distribution of SHAP values over all per-individual predictions in the data set. The extended positive tail for NT-proBNP reflects the strong impact of high NT-proBNP values for predicting cases whereas its truncated negative tail reflects a weak impact of low NT-proBNP values in…