A network medicine approach to study comorbidities in heart failure with preserved ejection fraction

Abstract

Background: Comorbidities are expected to impact the pathophysiology of heart failure (HF) with preserved ejection fraction (HFpEF). However, comorbidity profiles are usually reduced to a few comorbid disorders. Systems medicine approaches can model phenome-wide comorbidity profiles to improve our understanding of HFpEF and infer associated genetic profiles.

Methods: We retrospectively explored 569 comorbidities in 29,047 HF patients, including 8062 HFpEF and 6585 HF with reduced ejection fraction (HFrEF) patients from a German university hospital. We assessed differences in comorbidity profiles between HF subtypes via multiple correspondence analysis. Then, we used machine learning classifiers to identify distinctive comorbidity profiles of HFpEF and HFrEF patients. Moreover, we built a comorbidity network (HFnet) to identify the main disease clusters that summarized the phenome-wide comorbidity. Lastly, we predicted novel gene candidates for HFpEF by linking the HFnet to a multilayer gene network, integrating multiple databases. To corroborate HFpEF candidate genes, we collected transcriptomic data in a murine HFpEF model. We compared predicted genes with the murine disease signature as well as with the literature.

Results: We found a high degree of variance between the comorbidity profiles of HFpEF and HFrEF, while each was more similar to HFmrEF. The comorbidities present in HFpEF patients were more diverse than those in HFrEF and included neoplastic, osteologic and rheumatoid disorders. Disease communities in the HFnet captured important comorbidity concepts of HF patients which could be assigned to HF subtypes, age groups, and sex. Based on the HFpEF comorbidity profile, we predicted and recovered gene candidates, including genes involved in fibrosis (COL3A1, LOX, SMAD9, PTHL), hypertrophy (GATA5, MYH7), oxidative stress (NOS1, GSST1, XDH), and endoplasmic reticulum stress (ATF6). Finally, predicted genes were significantly overrepresented in the murine transcriptomic disease signature providing additional plausibility for their relevance.

Conclusions: We applied systems medicine concepts to analyze comorbidity profiles in a HF patient cohort. We were able to identify disease clusters that helped to characterize HF patients. We derived a distinct comorbidity profile for HFpEF, which was leveraged to suggest novel candidate genes via network propagation. The identification of distinctive comorbidity profiles and candidate genes from routine clinical data provides insights that may be leveraged to improve diagnosis and identify treatment targets for HFpEF patients.

Keywords: Comorbidities; Comorbidity network; Disease-gene prediction; HFpEF; Network medicine.

Fig. 1

Patient cohort description. Phenotyping algorithm to define HF cohorts. HF patients were selected with hospital visits over a time span of 13 years at the University Hospital Heidelberg. We defined a general HF cohort by selecting patients with either two or more HF relevant ICD-10 codes or one HF relevant ICD-10 code and one additional HF relevant clinical characteristic, yielding 29,047 HF patients. Based on LvEF, we subclassified HF patients to HFrEF, HFmrEF, or HFpEF. RWH Research Data Warehouse, HF heart failure, LvEF left ventricular ejection fraction; e/e’ is the ratio between early mitral inflow velocity and mitral annular early diastolic velocity on echocardiography

Fig. 2

Comparison of comorbidity profiles in heart failure subtypes. A Scheme of analysis. EH essential hypertension, CAD coronary artery disease, DMII diabetes mellitus type II, RA rheumatoid arthritis. B Multiple correspondence analysis of comorbidity profiles of HFpEF and HFrEF cohort. MCA dimensions were tested for association with clinical covariates and summed up to estimate total explained variance. C Proportions of the sum of parameter estimates of top 100 comorbidities of the patient classifier model, colored by disease categories. D Top 50 comorbidities of the patient classifier. The parameters are the absolute fitted values of the coefficients in the elastic net model for each comorbidity of the patient classifier separated by association to HFpEF (top) or HFrEF features (bottom). Colors indicate disease category using the same color legend as in B

Fig. 3

The heart failure comorbidity network (HFnet). A Scheme of comorbidity network analysis. EH essential hypertension, CAD coronary artery disease, DMII diabetes mellitus type II, RA rheumatoid arthritis. B Disease category composition of disease clusters (DCs) in the HFnet. Number of nodes per cluster in top barplot and number of diseases per category in side barplot. C Subgraphs of the HFnet visualized (left DC1, right DC3). Node size relates to prevalence, edge width to scaled phi-correlation, node color to disease category. Only edges with highest weights were plotted for visibility. D Comparison of patient cohorts based on DC similarity. Jaccard indices were calculated between each patient and each DC, then unpaired two-sided Wilcoxon rank test was applied to compare different patient cohorts. The log transformed p-value was multiplied by the sign of the test estimate for visualization purposes such that positive values indicate higher cluster similarity with the first cohort of the contrast label. Patient cohorts were selected by age stratification, sex, and HF subtype

Fig. 4

HFhetnet characterization. A Schematic overview of HFhetnet and its different layers built by including seven independent data sources. B Characterization of network layers by size (number of nodes and edges), edge density (percentage of possible edges), degree centrality, global transitivity (average probability of the neighbors of a node being connected), degree assortativity (preference of nodes to connect with nodes of similar degree), and literature bias (i.e., gene degree/PubMed score correlation). C Leave one out cross-validation results for all diseases with two or more DisGeNET links. We compared the performance of gene set recovery with different versions of the HFhetnet by modifying only the disease network. We compared HFnet + HPOnet (i.e., the original HFhetnet), only the HFnet (without HPOnet), and a rewired HFnet. Outliers are not plotted for visualization purposes. Paired, two-sided Wilcoxon test, *p < 0.001. AUC-PR area under the precision/recall curve, AUROC area under the receiver operator curve. GO Gene Ontology, HPO human phenotype ontology

Fig. 5

HFpEF gene prediction. A AUROC and AUC-PR for different HF-related gene sets in random walk probability vectors based on HFpEF and HFrEF comorbidity profiles. Prior knowledge gene sets are DisGeNET, Kegg pathway for dilated cardiomyopathy (DCM), cardiomyopathy (literature curated). Data-based gene sets are PheWAS, ReHeaT, and GWAS variants. B Prioritizing genes for HFpEF that are close to HFpEF comorbidity profiles in the HFhetnet and also display high ranking differences when compared to gene predictions based on HFrEF comorbidity profiles. C Scheme of experimental design for murine model of HFpEF by HFD/L-NAME diet. Cardiac ventricles were harvested after 9 weeks and bulk transcriptomics were collected. D Volcano plot displaying gene expression regulation in the murine HFpEF model compared to control. Labeled genes display HFpEF predicted genes from human comorbidity profiles. E Predicted HF genes from comorbidity analysis were enriched in gene-level t-statistics of murine differentially expression analysis comparing disease with control. Gene set enrichment p-value. ***p < 0.001. **p < 0.01