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

Lanzer, Jan D.; Valdeolivas, Alberto; Pepin, Mark; Hund, Hauke; Backs, Johannes; Frey, Norbert; Friederich, Hans-Christoph; Schultz, Jobst-Hendrik; Saez-Rodriguez, Julio; Levinson, Rebecca T.

doi:10.1186/s12916-023-02922-7

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

Research article
Open access
Published: 24 July 2023

Volume 21, article number 267, (2023)
Cite this article

Download PDF

You have full access to this open access article

BMC Medicine Aims and scope Submit manuscript

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

Download PDF

2699 Accesses
2 Citations
14 Altmetric
Explore all metrics

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.

Graphical Abstract

View this article's peer review reports

Biomarkers for the diagnosis and management of heart failure

Article Open access 14 April 2021

Functional and Metabolic Imaging in Heart Failure with Preserved Ejection Fraction: Promises, Challenges, and Clinical Utility

Article Open access 26 July 2022

Resolving the intertwining of inflammation and fibrosis in human heart failure at single-cell level

Article 03 October 2021

Background

Heart failure (HF) with preserved ejection fraction (HFpEF) represents an unmet public health concern with rising prevalence [1, 2]. Accumulating evidence indicates HFpEF is clinically and biologically distinct from HF with reduced ejection fraction (HFrEF), as reflected by missing therapy response in HFpEF patients to drugs effective in HFrEF [3]. HF patients suffer from a wide range of comorbidities, which are considered important for HF development and progression [4]. In the pathogenesis of HFpEF, comorbidities have been suggested as causal factors [3, 5] and could possibly be linked to genetic etiology. Treatment of comorbidity has also been shown to have beneficial effects of cardiac physiology [6], emphasizing the potential to address HF subtypes through their comorbidities.

Systems medicine attempts to model disease in a holistic manner. One facet of this, network medicine, is used to analyze complex systems such as patients, organs, or cells via network representation [7, 8]. Comorbidity networks represent diseases as nodes, connected via edges based on co-occurrence in patients. These networks can be used to define disease modules or explore topological changes between patient cohorts [9,10,11,12]. Previous work has shown that disease comorbidity is also often linked to shared disease genes that locate close together in gene-based networks like protein–protein interaction networks [12, 13]. This observation is often the basis of network-based gene prediction, where novel disease genes are predicted based on network proximity to known disease genes.

Cardiovascular diseases are particularly suited for system medicine approaches due to the typical multiorgan involvement [14] and multifactorial etiology [15]. To date, such approaches to study HFpEF have been limited, though the comorbidity-driven pathophysiology of HFpEF makes it a promising subject. In addition, despite the technological advances in multi-omics, knowledge of molecular characteristics of HFpEF remains limited, possibly due to difficulties of biopsy acquisition in HFpEF patients [16] and heterogeneity of HFpEF patients [17].

In this study, we applied a network medicine approach to describe comorbidity patterns in HFpEF and investigate a shared genetic background associated with these patterns. We first demonstrated that comorbidity profiles vary between HFpEF and HFrEF patients and derived distinct comorbidity profiles for each cohort. Then, we built a comorbidity network that contained disease clusters relevant for HF patients. The construction of a multilayer heterogeneous network by integration of prior knowledge resources allowed us to translate the comorbidity profiles into a gene signature for HFpEF. We corroborated this signature in the cardiac transcriptome of a murine HFpEF model. This network medicine approach allowed us to identify distinct comorbidity profiles and novel genetic patterns in HFpEF.

Methods

Study population

The study population was derived from a research data warehouse containing data from patients that visited the Department of Cardiology, Angiology, and Pneumology at Heidelberg University Hospital, Heidelberg, Germany [18]. Heidelberg University Hospital acts as a tertiary care center for the surrounding region, specializing in the treatment of cardiomyopathy. From this data warehouse, we identified patients with HF, visiting between 01.01.2008 and 01.01.2021. The study protocol was approved by the local ethics committee. HF was defined as two or more HF-relevant International Classification of Disease, version 10 (ICD-10) diagnosis codes (I50*, I11.0, I13.0, I13.2, I42.0, I42.5, I42.8, I42.9, I25.5) or at least one HF-relevant diagnosis and at least one of the following criteria: (i) elevated N-terminal pro b-type natriuretic peptide (NTproBNP) (> 120 ng/ml), (ii) recorded New York Heart Association functional class, (iii) echocardiography based E/e’ > 15 ( ratio of early diastolic mitral inflow velocity to early diastolic mitral annulus velocity), (iv) echocardiography or MRI-based left ventricular ejection fraction (LvEF) < 50%, and (v) documented loop diuretic. Patients with HF before age 40, those with a diagnosis of inheritable cardiomyopathy (I42.1-I42.4, I42.6, I42.7), and heart transplant patients (Z94.3) were excluded from the HF cohort. Within the HF cohort, HF subtypes were identified, based on echocardiographic or MRI-based LvEF. Patients with LvEF ≥ 50% were labeled HFpEF, LvEF 40–50% HFmrEF (HF with mid-range ejection fraction), and ≤ 40% HFrEF (Fig. 1). For all patients in the HF cohort, demographics, ICD-10 codes, operational and procedural codes, and targeted clinically relevant measurements were processed (Additional file 1: Fig. S1A, B) [11, 19,20,21,22,23,24,25,26,27,Full size image

Multiple correspondence analysis (MCA)

Disease profiles of HFpEF, HFrEF, and HFmrEF cohorts were captured as binary variables (0—patient has no record, 1—patient has a record of disease) of 569 phenome-wide association scan codes (PheCodes) (Additional file 1: Supplementary Methods). In this feature space (569 comorbidities × 17,665 HF patients), we performed MCA (R-package FactoMiner [39]). Each MCA dimension was then tested for association with clinical covariates with linear regression models (e.g., MCA-dimension 1 ~ age). For each covariate, we then summed the variance associated to all significantly associated dimensions (p-value < 0.05) as an estimate for the total associated variance.

Patient classifier

We trained random forest and regularized logistic regression (elastic net) models to predict HFpEF and HFrEF cohort labels on 569 PheCodes. The highest achieved mean area under the receiver operating characteristic (AUROC) in hyperparameter tuning was reported as an estimate for the model test error (Additional file 1: Supplementary Methods).

To derive the comorbidity profiles for HFpEF and HFrEF, we performed a forward selection with an L1-regularized logistic regression model of the 196 non-zero features from the elastic net model. Models were trained in R with R packages tidymodels using model engines from glmnet [40] and ranger [41].

HF comorbidity network (HFnet) construction

In disease comorbidity networks, nodes represent diseases while edges represent statistical association of co-occurrence, resulting in the graphical depiction of comorbidities as diseases that are statistically dependent. In detail, we selected edges using Fisher’s exact test for estimating statistical dependence and its Benjamini–Hochberg (BH) corrected p-value (< 0.0001) to discard non-significant disease pairs and keep a more sparse network structure. To determine strength of association, we calculated ɸ correlation, which can be interpreted as a Pearson correlation for binary variables. We selected all edges with positive correlation. To account for bias in ɸ correlations, we scaled the values by dividing by mean correlation values for every disease and assigned these values as edge weights [42].

Network node characteristics, such as betweenness, closeness, and degree centrality, and transitivity were calculated with the igraph R package. To calculate metrics based on graph distance, we replaced weights for edge i (W_i) with a new edge score S_i: \(S_i= \mathrm{max}(W)-W_i\). The constructed network was then compared to other comorbidity networks (Additional file 1: Supplementary Methods).

HF heterogeneous network (HFhetnet) construction

Disease-gene association

We used disease–gene associations provided by DisGeNet v7.0 [43, 44] and mapped the ICD-10 codes in DisGeNet to PheCodes (Additional file 1: Supplementary Methods). To ensure that the most frequent diseases in our cohort were mapped, we selected the most frequent 3-digit ICD-10 codes that were not mapped to DisGeNet and performed manual annotation via Unified Medical Language System (UMLS) IDs for 23 disease entities (Additional file 2: Supp. Table 1), e.g., PheCode 427.2 (atrial fibrillation) was manually mapped to the UMLS ID C000423. We only included disease–gene associations with a DisGeNet confidence score > 0.29. This cut-off was chosen, such that either one curated source or multiple experimental sources were necessary for disease–gene associations. Details on DisGeNET score calculation can be found at https://www.disgenet.org/dbinfo.

Gene–gene association

To consider multiple layers of gene organization, we constructed a multilayer gene network from different sources.

Omnipath [45, 46] is a meta resource of a multitude of biological knowledge databases, and we curated a network by connecting two genes if a resource provides a co-membership for a signaling pathway. We used the number of resources that reported a relationship as an estimate for the confidence in the relationship, which we introduced as edge weights in the Omnipath layer.

The protein–protein interaction (PPI) network was constructed based on the union of publicly available data from experimental and literature curated data (HuRi-union [47]) [48].

Gene Ontology (GO) gene networks have been constructed before, and we used the GO networks constructed by [49].

Each gene network was reduced to remove loops and multiple edges. To filter for genes relevant in cardiac tissue, gene networks were subset to genes expressed in the human heart on RNA or protein level. For protein expression, we used proteomic data [50, 51], where we selected all peptides that were detected in the human heart and used the leading gene associated with each peptide. For gene expression, we selected genes that were detected in the heart tissue in the Genotype-Tissue Expression (GTEx) Project v8 with a transcript per million value > 1 [52]. We chose this threshold to discard non-expressed genes but include lowly expressed genes. To ensure that gene programs only active in diseased hearts were also captured, we also included genes that were captured in a meta-analysis of HF transcriptomes [32, 33].

Disease–gene prediction and prioritization

To predict genes from diseases within the HFhetnet, we relied on a network propagation algorithm developed for multilayer networks (random walk with restart on multiplex heterogeneous networks; RWR-MH [53]). This algorithm is an extension of the random walk algorithm that tries to find a stationary distribution of probabilities that a node is visited when a random walk on the network is initiated in a set of seed nodes.

We assessed the performance of the link prediction task within the HFhetnet. For a given disease that was present in the HFhetnet and directly linked to two or more genes, we attempted to predict those genes after removing the direct links from the HFhetnet and running RWR-MH with the disease as seed node. The position of the target genes in the resulting probability ranking was then assessed with multiple metrics to estimate success of disease gene recovery (Additional file 1: Supplementary Methods).

For HFpEF and HFrEF gene prediction, we applied the RWR-MH, using the comorbidity profiles from the patient classifier as the seed nodes in the HFhetnet. This yielded two vectors of RW probabilities for each comorbidity profile. The top 500 genes yielded non-zero probability values for each profile. To select gene candidates that were differently ranked, we calculated a prioritization score for HFpEF and HFrEF. For this, we calculated \(G_i= {P_i}*|\Delta R_i|\).

G is the gene prioritization score, P is the RW-based probability, ΔR is the rank difference between HFpEF and HFrEF rankings for gene i.

Transcriptome analysis

We filtered lowly expressed genes and normalized samples using the Trimmed mean of M-values (edgeR [54]) and subsequent variance-stabilizing transformation (limma voom) and performed differential expression analysis (limma [55]. We performed principal component analysis and Gene Ontology enrichment with the enrichr [56] R package. For the overrepresentation analysis, we ranked genes by t-statistic and performed gene set enrichment analysis (fgsea R package [57]) of the top predicted HFpEF and HFrEF genes using different cut-offs.

Results

The study population

The study population consisted of 29,047 patients with HF (Fig. 1). Within this cohort, we identified three sub cohorts, HFpEF (8062 patients), HFrEF (6585 patients), and HFmrEF (3018 patients) based on LvEF. LvEF was not recorded in 11,382 HF patients, preventing subcohort labeling (i.e., unlabelled HF cohort). HFpEF patients were more often female compared to HFrEF patients (35 vs 25%, p < 0.01) (Table 1). However, we did not observe a significant difference in body mass index (median [IQR] = 26.8 [24.2, 30.0] vs 26.5 [24.1, 30.1] for HFpEF vs HFrEF, p = 0.9) or age (median [IQR] = 70 [61, 88] for HFpEF vs 70 [60, 70] for HFrEF, p = 0.5). When phenotypic data were available, cholesterol, LDL, HDL, and blood pressure values were higher in HFpEF patients compared to HFrEF, while NT-proBNP values were higher in HFrEF patients. Comorbidity burden measured by Elixhauser index was slightly lower in HFpEF than HFrEF patients, as previously reported [58]. HFpEF patients were intubated (8.5% vs 15%, p < 0.001) or received an implantable cardioverter-defibrillator (16% vs 26%, p < 0.001) less frequently than HFrEF patients, suggesting that the HFrEF cohort be a later stage of HF.

Table 1 Clinical characteristics of HFrEF, HFmrEF, and HFpEF cohorts. Descriptive statistics of HFrEF, HFmrEF and HFpEF cohorts. F female, m male, BMI body mass index, BP blood pressure, LDL low-density lipoprotein, HDL high-density lipoprotein, ICD implantable cardioverter defibrillator, PCI percutaneous coronary intervention, NT-BNP N-terminal pro b-type natriuretic peptide. All numerical values are median (IQR), Elixhauser index is mean (SD)

Full size table

High variation in comorbidity profiles is associated with HFpEF/HFrEF subtype

We expected differences in the composition of comorbidity profiles between HF subtype cohorts. To quantify this variance, we applied MCA and estimated the variance associated with sub-cohort labels and clinical features (Fig. 2A). Device implantation was the feature most strongly associated with variance in comorbidity profiles (Fig. 2B). When comparing HF subtypes, HFpEF and HFrEF cohort labels were associated with a high degree of explained variation (39.5%). HFmrEF patients seemed to be in an intermediate state, as they displayed lower variance when compared to HFpEF (25.2%) and HFrEF (18.6%). Sex and age were each associated with high variance (37.9% and 44.4%, respectively) as expected. In summary, this analysis approach identified a pronounced contrast between comorbidities in HFpEF and HFrEF patients.

Distinctive comorbidity profiles derived for HFpEF and HFrEF

Next, to explain and interpret the variance between HFpEF and HFrEF, we derived distinct comorbidity profiles for both cohorts. For this purpose, we fit random forest and elastic net classifier models with the 569 comorbidities as predictors to distinguish between HFpEF and HFrEF (Fig. 2A, Additional file 1: Fig. S2A,B). The highest achieved AUROCs were 0.778 for the random forest and 0.777 for the elastic net model, indicating that the random forest’s ability to model more complex interactions between comorbidities did not improve classifier performance substantially. The most important features were shared in both models (Additional file 1: Fig. S2C).

Next, because elastic net parameter estimates can provide both magnitude and direction, we selected the elastic net model to assign HFpEF and HFrEF a distinctive set of comorbidities. To select the most discriminant comorbidities, we performed forward selection. We found that the model with 100 comorbidities yielded a cross-validated AUROC of 0.780 (Additional file 1: Fig. S2D); 71 and 29 comorbidities from this model were assigned to HFpEF or HFrEF, respectively, which we will refer to as their comorbidity profiles.

These comorbidity profiles might be influenced by age, sex, time of visit, or time relative to HF diagnosis. We therefore investigated whether these factors influence the assignment of these 100 comorbidities to HF subtype by fitting a series of logistic regression models in different data subsets (Additional file 1: Supplementary Methods). We found that the derived comorbidity profiles of HFpEF and HFrEF yielded mostly consistent patterns independent of these factors (Additional file 1: Fig. S4).

The HFpEF profile (15 disease categories) was more diverse than the HFrEF profile (10 disease categories) and included comorbidities from the digestive disease, hematopoietic and neoplastic disease categories (Fig. 2C). Cardiovascular disease was the most important category in both profiles, accounting for 48.2% of the sum of parameter estimates in HFrEF and 38.3% in HFpEF. In HFpEF, important comorbidities included hypertensive and pulmonary heart disease, essential hypertension, inflammatory cardiac conditions (pericarditis, myocarditis), sleep apnea, osteopenia, neoplasms (multiple myeloma, breast cancer, metastasis in digestive systems), and rheumatoid disorders. The HFrEF comorbidity profile was characterized among others by myocardial infarction, ischemic heart disease, tobacco abuse, mitral valve disease, coma and cardiogenic shock, neurological disorders (vascular dementia, cerebral edema), chronic kidney disease, and diabetes type II (Fig. 2D, Additional file 1: Fig. S3).

In conclusion, the observed variation in comorbidity profiles between HFpEF and HFrEF was analyzed by interpreting patient classifiers. The derived features captured known subtype associations such as typical etiologies of HF including hypertensive heart disease (with HFpEF) and ischemic heart disease (with HFrEF) but also more novel and understudied comorbidities associated with HFpEF such as breast cancer or rheumatoid arthritis with HFpEF.

The HF comorbidity network (HFnet) captures HF specific disease relationships

To analyze patterns of disease co-occurrence in the HF-patient cohort, we constructed a comorbidity network as previously described [11, 12, 59,60,61]). This network was built by calculating pairwise disease correlations for the general HF-patient cohort (Fig. 3A) (Additional file 1: Fig. S6A,B). The resulting significant disease–disease relationships were assembled to form an undirected and weighted HF comorbidity network (HFnet) consisting of 569 nodes and 19,347 edges (Additional file 1: Fig. S6C), with edge weights defined by a statistical dependency of co-occurrence for each disease pair.

While certain comorbidities were distinctive for HFpEF or HFrEF, it was unclear whether the disease relationships that built the HFnet also depended on the HF subtype (Fig. 3A). When comparing odds ratios for each disease pair from both cohorts, we found a high concordance (Additional file 1: Fig. S5A,B). Only 33 disease pairs had significantly different odds ratios between HFpEF and HFrEF (Breslow–Dayes test with BH correction p < 0.01) (Additional file 1: Fig. S5C), suggesting that in the vast majority of cases, the co-occurrence of two diseases did not depend on whether it was assessed in HFpEF or HFrEF patients.

Whether the HFnet constituted a unique wiring of diseases or predominantly captured generalizable disease relationships was unclear. To investigate this, we analyzed and compared two additional disease networks: a human phenotype ontology network (HPOnet), where two diseases are connected if they are phenotypically similar, and Morbinet [11], another comorbidity network from a large patient cohort but without a cohort defining disease (Additional file 1: Supplementary Methods).

Jaccard index-based edge similarity of HFnet and Morbinet was 0.18 and of HFnet and HPOnet was 0.12 (Additional file 1: Fig. S7A,B,C). We then calculated network similarities with the DeltaCon algorithm to capture conserved node affinities between networks [1: Fig. S9H). No DC was positively associated with HFmrEF. Instead, HFmrEF patients were less similar to DC1 and DC6 than HFpEF patients and less similar to DC3, DC4, and DC5 than HFrEF patients.

In general, we found that aggregating comorbidity profiles (569 dimensions) to DC similarity (9 dimensions) allowed us to capture differences among patient cohorts in regard to sex, age, and HF subtype in meaningful disease groups.

Building the HF heterogeneous network (HFhetnet)

Biomedical research has yielded significant knowledge of disease gene associations, which can be harnessed to extrapolate novel disease gene relationships. HFpEF is a comorbidity-driven syndrome and we hypothesized that the identified HFpEF comorbidity profile could be translated to a genetic profile consisting of recurrent genetic associations to these comorbidities. In this part of our study, we first integrated multiple biomedical databases to construct a cardiac specific multilayer disease and gene network. We then estimated the success of this network to recover known disease–gene associations and, finally, used the HFpEF comorbidity profile to identify the most commonly associated genes.

To construct a gene network that reflected different hierarchies of gene function (i.e., pathway memberships, PPI, and ontological similarity), we integrated multiple databases and represented gene–gene relationships as networks (“Methods”) (Fig. 4A). To focus on genes relevant in cardiac tissue, we subset the resulting gene networks to protein coding genes expressed in the heart (Additional file 1: Fig. S10A). Next, we used DisGeNET, a resource containing disease–gene associations, to connect the HFnet with the gene network. We connected 400 diseases of the HFnet with a total of 4044 genes via 20,170 edges. As the HPOnet constructed earlier had a small intersection with the HFnet and captured a different type of disease relationship, it was included as an additional disease layer in our network.

The presented HFhetnet is an assembly of the data-driven comorbidity relationships (HFnet) and six biomedical databases resulting in a total of 13,572 nodes and 181,529 edges (Additional file 5: Supp. Table 4). Its main structure is set up by two biological networks (disease layer and gene layer) that each consist of two or four network layers, respectively (Fig. 4A). The two disease networks were the smallest when comparing node numbers (Fig. 4B). However, edge density was much higher resulting in centralization of these networks compared to the gene layers. Within gene layers, the ontological layers displayed the highest transitivity, as well as tendency to connect to hub genes (degree assortativity). To assess research bias in the gene networks, we calculated Pearson correlation between the number of abstracts in PubMed mentioning a gene and the gene’s network degree per layer and found that only the pathway layer (Omnipath) displayed significant correlation (p-value < 0.05). This is related to a biomedical research bias towards the investigation of a small number of genes [62]. Thus, the integration of experimental and ontological data can ameliorate the centrality of overstudied genes.

In summary, we constructed the HFhetnet by integrating various prior knowledge resources to incorporate genetic information. The different network layers of the HFhetnet captured unique node relationships and displayed particular network topologies.

Estimating the success of disease–gene prediction within the HFhetnet

To estimate the potential of the HFhetnet to predict disease–gene relationships, we estimated the success of predicting known disease genes. The rationale behind this approach is the guilt-by-association principle that assumes that functionally related genes are also associated in the network context. Extending this notion to heterogeneous networks, this principle can be interpreted as a disease being associated with relevant disease genes through its position in the network. To quantify this property, we applied a leave-one-out validation design to assess whether known disease genes can be recovered after removing the direct edges that connected them to a disease. After edge removal, the gene recovery was performed by applying the RWR-MH algorithm which considers each network layer and its topology ( Additional file 1: Supplementary Methods).

We performed this analysis by comparing the impact of three variations of the disease layer: (i) HFnet + HPOnet (original HFhetnet), (ii) only HFnet, and (iii) a rewired HFnet. Gene prediction worked best in the original HFhetnet (median AUROC 0.91, median AUC-PR 0.07, and median rank ratio 0.03) (Fig. 4C). This performance dropped for every metric when removing the HPO layer or when rewiring the HFnet (paired, two-sided Wilcoxon’s rank sum test p < e − 10). The rewired HFnet still performed better than random, which might be explained by (i) high edge density in the HFnet and (ii) the large size of the unaltered gene–gene and disease–gene network in comparison to the smaller HFnet.

Prediction success correlated weakly but significantly with gene set size (Additional file 1: Fig. S10C). In addition, neither disease prevalence nor DisGeNET confidence scores were significantly correlated with prediction success, suggesting that frequent diseases could not be predicted better than less frequent diseases. Prediction performance depended on disease category (Additional file 1: Fig. S10D) (Kruskal–Wallis test p-value < 0.01 for all metrics) with respiratory, neurological, genitourinary, and cardiovascular diseases performing best.

In summary, we found that within the HFhetnet, the disease genes remained close via the disease’s connection through its comorbidities or phenotypically similar neighbors. Thus, we concluded that HFnet and its extension, HFhetnet, captured meaningful disease–disease, disease–gene, and gene–gene relationships, which can be exploited for predicting a disease’s genetic profile through its comorbidities.

Predicting genes associated with comorbidity profiles of HFpEF and HFrEF

In the first part of our study, we found that HFpEF and HFrEF patients were distinguishable based on their comorbidity profiles. We then demonstrated that diseases within the HFhetnet were located in network proximity to their respective disease–genes. Leveraging both insights, we hypothesized that genes located close to the HFpEF and HFrEF comorbidity profiles could yield novel candidates for the respective HF subtype. In this section, we applied the RWR-MH algorithm with the HFpEF and HFrEF comorbidity profiles as seed nodes resulting in gene ranking based on network proximity (Additional file 1: Fig. S11A-C).

To assess whether the resulting gene rankings recapitulated known HF genes, we curated a set of HF-related genes from various prior knowledge sources and independent datasets (Additional file 1: Supplementary Methods), which had only little intersection (Additional file 1: Fig. S11D). We found that prior knowledge gene sets were well recovered within the HFpEF and HFrEF gene rankings (Fig. 5A, Additional file 1: Fig. S11E). Gene sets that were retrieved from experimental data (gene expression, PheWAS, GWAS) performed worse in the predictions. Next, we compared these prediction results with random comorbidity profiles and found that the HFrEF profile associated with Kegg’s dilated cardiomyopathy (DCM) (z-score AUROC 1.77; z-score PR-AUC 6.7) and DisGeNETs HF genes (z-score AUROC 1.76; z-score PR-AUC 2.46) (Additional file 1: Fig. S11F). This indicated that the HFrEF comorbidity profile which was more cardiac centered was closer to prior knowledge of HF genes within the HFhetnet. In general, well-known genes relevant for HF were recovered for both, HFpEF and HFrEF comorbidity profiles including NPPA, NPPB, TNFa, NOS2, NOS3, CCL2, IL1B, LMNA, and TTN (Additional file 1: Fig. S11D).

To emphasize genes that might be HFpEF specific, we prioritized genes that were close to the HFpEF but not to the HFrEF comorbidity profile within the HFhetnet (Fig. 5B, Additional file 6: Supp. Table 5, Additional file 1: Supplementary Methods). We found that MMP1, MHY7, and DAPK1 received the highest scores and other candidates included genes functionally involved in fibrosis (e.g., LOX), metabolism (MVK), transcriptional regulation (ATF6), coagulation (THBD), and oxidative stress (NOS1, XDH) (Table 3).

Table 3 Potential HFpEF candidates based on network proximity to comorbidity profiles. Genes are sorted by functional groups. Rank indicates prediction rank for HFpEF. TF transcription factor, CM cardiomyopathy

Full size table

Corroboration of HFpEF gene candidates in the transcriptome of a murine HFpEF model

After translating the comorbidity profile of HFpEF patients to an associated genetic profile, the functional relevance of this predicted profile remained unclear. We hypothesized that the relevance of the gene predictions could be suggested by transcriptional dysregulation in myocardial tissue of HFpEF. However, human molecular data of HFpEF is sparse and mechanistic insights are typically derived from mouse models [3]. Hence, we collected myocardial bulk transcriptomics from a murine HFpEF model, induced via high-fat diet and nitric oxide synthase inhibitor N[w]-nitro-l-arginine methyl ester (L-NAME) treatment [36] (Fig. 5C, Additional file 1: Supplementary Methods).

First, we confirmed important HFpEF phenotype characteristics including preserved ejection fraction, increased body weight, blood glucose levels, and blood pressure as well as diastolic dysfunction (increased E/e') (n_CT = 4, n_HFpEF = 4, Student’s t-test, p < 0.05, Additional file 1: Fig. S12).

Second, we assessed transcriptomic changes in HFpEF via differential gene expression analysis (Additional file 1: Fig. S13A-D). Upregulated genes confirmed processes involving fibrosis and metabolic stress (Additional file 1: Fig. S13C, D).

After confirming the phenotypic and molecular resemblance of the HFpEF model, we investigated gene expression dysregulation of the comorbidity-based gene profiles by performing enrichment analysis of the HFpEF and HFrEF gene predictions (Fig. 5D, Additional file 1: Fig. S13E). We found that the top 50 to 100 predicted HFpEF genes displayed significant enrichment in overexpressed genes in the murine HFpEF model, while the HFrEF predicted genes were not enriched (p-value < 0.05, Fig. 5E, Additional file 1: Fig. S13E). Fibrosis-related genes like LOX, SMAD9, and PTHL and hypertrophy-related genes like GATA5, GATA3, and MYH7 could be recovered together with XDH, among others. This suggested that the genetic profile derived from human HFpEF comorbidities associated with relevant gene expression dysregulation during myocardial remodeling in murine HFpEF.

Discussion

In this study, we provided a retrospective systems level analysis of comorbidities in a large cohort of HF patients. We derived clinically relevant insights by comparing comorbidity profiles between HFpEF and HFrEF patients and biological insights by defining genes associated with HFpEF and HFrEF comorbidity profiles.

Patient clustering has been previously shown to yield novel subgroups of HFpEF defined by multivariate similarity [109,110,111]. In contrast, the clustering of features (i.e., comorbidities) can inform about patterns of co-occurring disease groups. Our study demonstrated that this approach can be useful to interpret comorbidity profiles: the aggregation of co-occurrence patterns of diseases can help to organize illness into different levels of clinical concepts like organs (DC7—gastrointestinal tract), illness severity (DC5—intensive care), or disease categories (DC8—cancer). This aggregation via network clustering may also reduce multiple testing burdens and provide insights into the relevance of low prevalence diseases where comparisons for a single disease may be problematic.

In the patient classifier, HFpEF was characterized by a larger number of comorbidities with lesser emphasis on cardiac disorders. This supports the hypothesis of HFpEF as a comorbidity-driven systemic syndrome [112, 113]. We found that hypertensive heart disease was the most discriminant feature for HFpEF, which has been viewed as a major etiology for diastolic HF involving cardiac hypertrophy and myocardial stiffness [114, 115]. In contrast, ischemic etiologies including myocardial infarction characterized HFrEF consistent with other studies [116].

We identified more novel disease associations with HFpEF such as neoplastic diseases including breast cancer. HF related hospitalizations in breast cancer survivors have been recently associated more with HFpEF than with HFrEF [117], though the reasons for this remain incompletely elucidated [118]. The association to other cancerous diseases remains largely unexplored and should be addressed in future studies. Another interesting aspect of the HFpEF comorbidity profile was the high similarity to DC6, which contained rheumatic, osteologic, and mental diseases. Systemic inflammatory diseases could be a driving factor for HFpEF and rheumatic disease could constitute a pathophysiologic linkage [112, 119,120,121]. Bone mineralization also has been reported to be lowered in HFpEF patients [122] and is a symptomatic link to postmenopausal endocrinology [123]. While mental health has been studied in the context of HF extensively, differences between HFpEF and HFrEF are largely unexplored. The joint clustering of these disease complexes and their similarity to female patients provides a potential link between female sex and HFpEF. Future work should further explore these relationships.

HFpEF and HFrEF clearly displayed distinguishable comorbidity profiles. By contrast, HFmrEF, introduced as a unique form of HF in 2016 [124], appeared to be a combination of attributes from HFrEF and HFpEF. Thus, from the comorbidity perspective, it may be a transitional state instead of a unique syndrome as suggested before [125].

We predicted an associated genetic profile from data-driven HFpEF comorbidity profiles. This genetic profile indicates that HFpEF comorbidities are associated with recurrent patterns of genes involved in fibrosis, inflammation, cell differentiation, metabolism, and oxidative stress. As an example, the glutathione-S-transferases, NOS1 and Xanthine dehydrogenase (XDH), were identified by our network. XDH catalyzes the rate limiting step in purine metabolism producing uric acid [126] and previous literature supports both the role of serum uric acids in HF [104] and plasma XDH activity as relevant for adverse clinical outcomes in HFpEF [127]. Nitric oxide synthase (NOS) has been proposed to contribute to endothelial dysfunction in HFpEF [105, 106], and NOS1 inhibition was recently associated with recovery of diastolic dysfunction in a murine model resembling HFpEF [128]. Glutathione-S-transferases (GSTM1, GSTT1, GSTZ1) are antioxidant enzymes and polymorphisms of these genes have been reported as potentially relevant to HF and diastolic dysfunction [107, 108]. This group of genes could constitute crucial gene candidates involved in comorbidity-based HFpEF pathophysiology.

In general, HFpEF is likely to be a disease in which multiple genes and pathways contribute to the spectrum of phenotypes. Therefore, instead of using the disease–gene prediction to identify and validate individual genes, we have corroborated the overall effect of a spectrum of identified genes in murine gene expression data. While this provided additional evidence for the relevance of comorbidity-based gene prediction, further experimental validation is necessary to explore the functional role and reproducible validity of candidate genes. In real-world populations, it is likely that the genetic heterogeneity of the HFpEF syndrome will be influenced by the specific comorbidities that are well represented in each population. In previous disease–gene prediction studies, gene prediction was performed either by selecting multiple seed genes or single seed diseases [129, 130]. We propose that our approach for gene inference based on data-driven comorbidity profiles might be suitable for systemic syndromes where multimorbidity plays an important role like HF and especially HFpEF.

In addition, several data resources were generated in this study: (i) the HFpEF gene predictions, (ii) HFhetnet, and (iii) murine HFpEF transcriptome to help facilitate future efforts to understand HFpEF-related pathophysiology and benefit the research community.

This study had important strengths and was subject to several limitations. An important strength of this study is that we analyzed clinical care data, which is a real-world representation of patients and therefore allowed us to perform a data-driven analysis of comorbidities in this patient population. However, as a result, this analysis is limited to the information captured (i) in our hospital system and (ii) at the hospital visits of a patient. Therefore, obtained results could be subject to some common biases found in medical record-derived data, such as non-random interaction with the health care system resulting in some patient populations having more data than others [131], incomplete documentation [132], selection bias [133], missing data, and lack of documentation of potential confounders [134]. In addition, possible non-observed confounders like socioeconomic status or health-related behavior could not be taken into account due to lack of documentation in the medical record. We determined subtypes using LvEF, which can be error prone [135] and might not fully provide a sufficient criterion for the HFpEF diagnosis [136]. Patients with more serious conditions will tend to visit a tertiary health care provider more often and thus could be overrepresented. In our study, at a tertiary care center with a focus on cardiomyopathy, this seemed to affect the contrast between HFrEF and HFpEF, as HFrEF patients had higher intubation prevalence and DC 5 similarity. This may also have contributed to differences between this study population and other reports of HFpEF population characteristics. However, given the known heterogeneity of HFpEF and HFrEF [109,110,111], we believe these differences are plausible and a more granular approach to study HFpEF subtypes could be necessary to address inconsistent patient characteristics [137]. Another limitation of our study is the use of ICD-10 codes to capture comorbidities. Different ICD-10 codes are known to have different predictive value for disease, and therefore, it is likely that some diseases are over- or underrepresented in our data [138]. Moreover, we performed a cross-sectional analysis and therefore did not consider the timing and sequence of comorbidities when generating comorbidity profiles.

Given these limitations, future studies are necessary to address the generalizability of our findings to other HF populations and to delineate different disease trajectories by considering the time of events. Nevertheless, our study recapitulated known HF comorbidity patterns, as discussed above, that could substantiate more novel comorbidity patterns identified in this work.

Many open questions remain regarding HFpEF pathophysiology and genetics [16]. Interdisciplinary and translational approaches are needed to account for the cross-organ disease involvement that is suggested to be critical in HFpEF. The increasing abundance of routine clinical care data and novel approaches like network medicine can provide novel insights and guidance for future experimental approaches.

Conclusions

In our study, we found evidence for greater diversity of comorbidity profiles in patients with HFpEF compared to HFrEF. We further identified nine co-occurring disease groups which capture differences of disease prevalence regarding age, sex, and HF subtype. Here, we find that multimorbidity in HFpEF extends to disease clusters beyond typical HF comorbidities and includes rheumatoid, neoplastic, and gastrointestinal diseases. We further provided a biological interpretation of the HFpEF comorbidity profile, capturing overexpressed gene programs observed in murine HFpEF models. Oxidative stress, hypertrophy, cell differentiation, and fibrosis-related genes are recurrent patterns in genes associated to comorbidities of HFpEF and could constitute a link for the comorbid relationships of HFpEF resulting in a multiorgan disease state. Thus, our work highlights that comorbidity profiles are an important characteristic of HFpEF patients and should be incorporated into both clinical and genomic approaches to the study of HFpEF.

Availability of data and materials

All code from this analysis is available at github.com/janlanzer/hf_comorbidity. Clinical data could not be deposited for patient privacy reasons. Model outputs such edge tables of HFnet and HFhetnet are available in table format in Supplementary Tables. Murine RNA-sequencing data has been deposited in NCBI’s Gene Expression Omnibus (GEO) and is accessible through GEO Series accession number GSE225557.

Abbreviations

AUROC:: Area under the receiver operating characteristic
BH:: Benjamini–Hochberg
DC:: Disease clusters
GO :: Gene Ontology
HF:: Heart failure
HFnet:: Heart failure comorbidity network
HFhetnet:: Heart failure multilayer heterogeneous network
HFmrEF:: Heart failure with mid-range ejection fraction
HFpEF:: Heart failure with preserved ejection fraction
HFrEF:: Heart failure with reduced ejection fraction
ICD:: Implantable cardioverter defibrillator
ICD-10:: International classification of disease, version 10
LvEF:: Left ventricular ejection fraction
MCA:: Multiple correspondence analysis
NTproBNP:: N-terminal pro b-type natriuretic peptide
PheCodes:: Phenome-wide association scan codes
PPI:: Protein–protein interaction
UMLS:: Unified Medical Language System

References

Oktay AA, Rich JD, Shah SJ. The emerging epidemic of heart failure with preserved ejection fraction. Curr Heart Fail Rep. 2013;10:401–10.
Article PubMed Google Scholar
Dunlay SM, Roger VL, Redfield MM. Epidemiology of heart failure with preserved ejection fraction. Nat Rev Cardiol. 2017;14:591–602.
Article PubMed Google Scholar
Mishra S, Kass DA. Cellular and molecular pathobiology of heart failure with preserved ejection fraction. Nat Rev Cardiol. 2021;18:400–23.
Article PubMed PubMed Central Google Scholar
Pandey A, Vaduganathan M, Arora S, Qamar A, Mentz RJ, Shah SJ, et al. Temporal trends in prevalence and prognostic implications of comorbidities among patients with acute decompensated heart failure: the ARIC study community surveillance. Circulation. 2020;142:230–43.
Article PubMed PubMed Central Google Scholar
Simmonds SJ, Cuijpers I, Heymans S, Jones EAV. Cellular and molecular differences between HFpEF and HFrEF: a step ahead in an improved pathological understanding. Cells. 2020;9(1):242.
Palmiero G, Cesaro A, Vetrano E, Pafundi PC, Galiero R, Caturano A, et al. Impact of SGLT2 inhibitors on heart failure: from pathophysiology to clinical effects. Int J Mol Sci. 2021;22(11):5863.
Maron BA, Altucci L, Balligand J-L, Baumbach J, Ferdinandy P, Filetti S, et al. A global network for network medicine. NPJ Syst Biol Appl. 2020;6:29.
Article PubMed PubMed Central Google Scholar
Hu JX, Thomas CE, Brunak S. Network biology concepts in complex disease comorbidities. Nat Rev Genet. 2016;17:615–29.
Article CAS PubMed Google Scholar
Khan A, Uddin S, Srinivasan U. Comorbidity network for chronic disease: a novel approach to understand type 2 diabetes progression. Int J Med Inform. 2018;115:1–9.
Article PubMed Google Scholar
Guo M, Yu Y, Wen T, Zhang X, Liu B, Zhang J, et al. Analysis of disease comorbidity patterns in a large-scale China population. BMC Med Genomics. 2019;12(Suppl 12):177.
Article PubMed PubMed Central Google Scholar
Aguado A, Moratalla-Navarro F, López-Simarro F, Moreno V. MorbiNet: multimorbidity networks in adult general population. Analysis of type 2 diabetes mellitus comorbidity. Sci Rep. 2020;10:2416.
Article CAS PubMed PubMed Central Google Scholar
Hidalgo CA, Blumm N, Barabási A-L, Christakis NA. A dynamic network approach for the study of human phenotypes. PLoS Comput Biol. 2009;5: e1000353.
Article PubMed PubMed Central Google Scholar
Goh K-I, Cusick ME, Valle D, Childs B, Vidal M, Barabási A-L. The human disease network. Proc Natl Acad Sci USA. 2007;104:8685–90.
Article CAS PubMed PubMed Central Google Scholar
Lusis AJ, Weiss JN. Cardiovascular networks: systems-based approaches to cardiovascular disease. Circulation. 2010;121:157–70.
Article PubMed PubMed Central Google Scholar
Bousquet J, Anto JM, Sterk PJ, Adcock IM, Chung KF, Roca J, et al. Systems medicine and integrated care to combat chronic noncommunicable diseases. Genome Med. 2011;3:43.
Article PubMed PubMed Central Google Scholar
Shah SJ, Borlaug BA, Kitzman DW, McCulloch AD, Blaxall BC, Agarwal R, et al. Research priorities for heart failure with preserved ejection fraction: national heart, lung, and blood institute working group summary. Circulation. 2020;141:1001–26.
Article PubMed PubMed Central Google Scholar
Joseph J, Liu C, Hui Q, Aragam K, Wang Z, Charest B, et al. Genetic architecture of heart failure with preserved versus reduced ejection fraction. Nat Commun. 2022;13:7753.
Article CAS PubMed PubMed Central Google Scholar
GMS | GMDS 2012: 57. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e. V. (GMDS) | Webservice zur sicheren Pseudonymisierung durch Datentreuhänder. https://www.egms.de/static/en/meetings/gmds2012/12gmds121.shtml. Accessed 9 Dec 2022.
PheCode map. https://phewascatalog.org/phecodes_icd10. Accessed October 2021.
Wu P, Gifford A, Meng X, Li X, Campbell H, Varley T, et al. Map** ICD-10 and ICD-10-CM Codes to Phecodes: Workflow Development and Initial Evaluation JMIR Med Inform. 2019;7(4):e14325.
Gasparini A. comorbidity: an R package for computing comorbidity scores. JOSS. 2018;3:648.
Article Google Scholar
Traag VA, Waltman L, van Eck NJ. From Louvain to Leiden: guaranteeing well-connected communities. Sci Rep. 2019;9:5233.
Article CAS PubMed PubMed Central Google Scholar
Morbinet. https://shiny.odap-ico.org/morbinet/. Accessed October 2021.
HPO database. https://hpo.jax.org/app/. Accessed August 2021.
Köhler S, Gargano M, Matentzoglu N, Carmody LC, Lewis-Smith D, Vasilevsky NA, et al. The human phenotype ontology in 2021. Nucleic Acids Res. 2021;49:D1207–17.
Article PubMed Google Scholar
Greene D, Richardson S, Turro E. ontologyX: a suite of R packages for working with ontological data. Bioinformatics. 2017;33:1104–6.
Article CAS PubMed Google Scholar
Serrano MA, Boguñá M, Vespignani A. Extracting the multiscale backbone of complex weighted networks. Proc Natl Acad Sci USA. 2009;106:6483–8.
Article CAS PubMed PubMed Central Google Scholar
Koutra D, Vogelstein JT, Faloutsos C. DELTACON: a principled massive-graph similarity function. ar**v:1304.4657.
Czepluch FS, Wollnik B, Hasenfuß G. Genetic determinants of heart failure: facts and numbers. ESC Heart Fail. 2018;5:211–7.
Article PubMed PubMed Central Google Scholar
Kanehisa M, Furumichi M, Sato Y, Kawashima M, Ishiguro-Watanabe M. KEGG for taxonomy-based analysis of pathways and genomes. Nucleic Acids Res. 2023;51:D587–92.
Article CAS PubMed Google Scholar
Cardiovascular Disease Knowledge Portal. https://cvd.hugeamp.org/phenotype.html?phenotype=HF. Accessed 8 Aug 2022.
Ramirez Flores RO, Lanzer JD, Holland CH, Leuschner F, Most P, Schultz J-H, et al. The Reference of the Transcriptional Landscape of Human End-Stage Heart Failure. https://doi.org/10.5281/zenodo.3797044.
Ramirez Flores RO, Lanzer JD, Holland CH, Leuschner F, Most P, Schultz J-H, et al. Consensus transcriptional landscape of human end-stage heart failure. J Am Heart Assoc. 2021;10: e019667.
Article PubMed PubMed Central Google Scholar
PheWAS catalog. Nat Biotechnol. 2013. https://phewascatalog.org/#. Accessed 26 Sep 2022.
Denny JC, Bastarache L, Ritchie MD, Carroll RJ, Zink R, Mosley JD, et al. Systematic comparison of phenome-wide association study of electronic medical record data and genome-wide association study data. Nat Biotechnol. 2013;31:1102–10.
Article CAS PubMed PubMed Central Google Scholar
Schiattarella GG, Altamirano F, Tong D, French KM, Villalobos E, Kim SY, et al. Nitrosative stress drives heart failure with preserved ejection fraction. Nature. 2019;568:351–6.
Article CAS PubMed PubMed Central Google Scholar
Tong D, Schiattarella GG, Jiang N, May HI, Lavandero S, Gillette TG, et al. Female sex is protective in a preclinical model of heart failure with preserved ejection fraction. Circulation. 2019;140:1769–71.
Article PubMed PubMed Central Google Scholar
Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013;29:15–21.
Article CAS PubMed Google Scholar
Lê S, Josse J, Husson F. FactoMineR: An R package for multivariate analysis. J Stat Softw. 2008;25(1):1–18.
Friedman J, Hastie T, Tibshirani R. Regularization paths for generalized linear models via coordinate descent. J Stat Softw. 2010;33:1–22.
Article PubMed PubMed Central Google Scholar
Wright MN, Ziegler A. ranger: a fast implementation of random forests for high dimensional data inC++ and R. J Stat Softw. 2017;77:1–17.
Article Google Scholar
Chmiel A, Klimek P, Thurner S. Spreading of diseases through comorbidity networks across life and gender. New J Phys. 2014;16: 115013.
Article Google Scholar
Piñero J, Ramírez-Anguita JM, Saüch-Pitarch J, Ronzano F, Centeno E, Sanz F, et al. The DisGeNET knowledge platform for disease genomics: 2019 update. Nucleic Acids Res. 2020;48:D845–55.
PubMed Google Scholar
DisGeNET database. https://www.disgenet.org/downloads. Accessed August 2021.
Türei D, Valdeolivas A, Gul L, Palacio‐Escat N, Klein M, Ivanova O, et al. Integrated intra‐ and intercellular signaling knowledge for multicellular omics analysis. Mol Syst Biol. 2021;17:e9923.
Bioconductor - OmnipathR. https://www.bioconductor.org/packages/release/bioc/html/OmnipathR.html. Accessed August 2021.
HuRi. http://www.interactome-atlas.org/data/HI-union.tsv. Accessed August 2021
Luck K, Kim D-K, Lambourne L, Spirohn K, Begg BE, Bian W, et al. A reference map of the human binary protein interactome. Nature. 2020;580:402–8.
Article CAS PubMed PubMed Central Google Scholar
Buphamalai P, Kokotovic T, Nagy V, Menche J. Network analysis reveals rare disease signatures across multiple levels of biological organization. Nat Commun. 2021;12:6306.
Article CAS PubMed PubMed Central Google Scholar
Doll S, Dreßen M, Geyer PE, Itzhak DN, Braun C, Doppler SA, et al. Region and cell-type resolved quantitative proteomic map of the human heart. Nat Commun. 2017;8:1469.
Article PubMed PubMed Central Google Scholar
Heart Proteome Map. http://maxqb.biochem.mpg.de/. Access ID:PXD006675. Accessed August 2021.
GTEX v8 DB. https://www.gtexportal.org/home/datasets#datasetTitleDiv1. Accessed October 2021.
Valdeolivas A, Tichit L, Navarro C, Perrin S, Odelin G, Levy N, et al. Random walk with restart on multiplex and heterogeneous biological networks. Bioinformatics. 2019;35:497–505.
Article CAS PubMed Google Scholar
Robinson MD, McCarthy DJ, Smyth GK. edgeR: a bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics. 2010;26:139–40.
Article CAS PubMed Google Scholar
Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 2015;43: e47.
Article PubMed PubMed Central Google Scholar
Chen EY, Tan CM, Kou Y, Duan Q, Wang Z, Meirelles GV, et al. Enrichr: interactive and collaborative HTML5 gene list enrichment analysis tool. BMC Bioinformatics. 2013;14:128.
Article PubMed PubMed Central Google Scholar
Korotkevich G, Sukhov V, Budin N, Shpak B, Artyomov MN, Sergushichev A. Fast gene set enrichment analysis. BioRxiv. 2016.
Ergatoudes C, Schaufelberger M, Andersson B, Pivodic A, Dahlström U, Fu M. Non-cardiac comorbidities and mortality in patients with heart failure with reduced vs. preserved ejection fraction: a study using the Swedish Heart Failure Registry. Clin Res Cardiol. 2019;108:1025–33.
Article PubMed PubMed Central Google Scholar
Cruz-Ávila HA, Vallejo M, Martínez-García M, Hernández-Lemus E. Comorbidity networks in cardiovascular diseases. Front Physiol. 2020;11:1009.
Article PubMed PubMed Central Google Scholar
Klimek P, Aichberger S, Thurner S. Disentangling genetic and environmental risk factors for individual diseases from multiplex comorbidity networks. Sci Rep. 2016;6:39658.
Article CAS PubMed PubMed Central Google Scholar
Divo MJ, Casanova C, Marin JM, Pinto-Plata VM,de-Torres JP, Zulueta JJ, et al. COPD comorbidities network. Eur Respir J. 2015;46:640–50.
Article PubMed Google Scholar
Kustatscher G, Collins T, Gingras A-C, Guo T, Hermjakob H, Ideker T, et al. Understudied proteins: opportunities and challenges for functional proteomics. Nat Methods. 2022;19:774–9.
Article CAS PubMed Google Scholar
Wesseling M, de Poel JHC, de Jager SCA. Growth differentiation factor 15 in adverse cardiac remodelling: from biomarker to causal player. ESC Heart Fail. 2020;7:1488–501.
Article PubMed PubMed Central Google Scholar
Szumska D, Cioroch M, Keeling A, Prat A, Seidah NG, Bhattacharya S. Pcsk5 is required in the early cranio-cardiac mesoderm for heart development. BMC Dev Biol. 2017;17:6.
Article PubMed PubMed Central Google Scholar
Warren SA, Terada R, Briggs LE, Cole-Jeffrey CT, Chien W-M, Seki T, et al. Differential role of Nkx2-5 in activation of the atrial natriuretic factor gene in the develo** versus failing heart. Mol Cell Biol. 2011;31:4633–45.
Article CAS PubMed PubMed Central Google Scholar
Arasaratnam D, Bell KM, Sim CB, Koutsis K, Anderson DJ, Qian EL, et al. The role of cardiac transcription factor NKX2-5 in regulating the human cardiac miRNAome. Sci Rep. 2019;9:15928.
Article CAS PubMed PubMed Central Google Scholar
Tayal U, Prasad S, Cook SA. Genetics and genomics of dilated cardiomyopathy and systolic heart failure. Genome Med. 2017;9:20.
Article PubMed PubMed Central Google Scholar
Liu X, Shi G-P, Guo J. Innate immune cells in pressure overload-induced cardiac hypertrophy and remodeling. Front Cell Dev Biol. 2021;9: 659666.
Article PubMed PubMed Central Google Scholar
Blanton RM, Carrillo-Salinas FJ, Alcaide P. T-cell recruitment to the heart: friendly guests or unwelcome visitors? Am J Physiol Heart Circ Physiol. 2019;317:H124–40.
Article CAS PubMed PubMed Central Google Scholar
Croquelois A, Domenighetti AA, Nemir M, Lepore M, Rosenblatt-Velin N, Radtke F, et al. Control of the adaptive response of the heart to stress via the Notch1 receptor pathway. J Exp Med. 2008;205:3173–85.
Article CAS PubMed PubMed Central Google Scholar
Fernández-Ruiz I. Selective JAG1-NOTCH3 targeting shows potential for treating PAH. Nat Rev Cardiol. 2022;19:433.
Article PubMed Google Scholar
Zhao Q, Huang J, Wang D, Chen L, Sun D, Zhao C. Endothelium-specific CYP2J2 overexpression improves cardiac dysfunction by promoting angiogenesis via Jagged1/Notch1 signaling. J Mol Cell Cardiol. 2018;123:118–27.
Article CAS PubMed Google Scholar
Yuan T, Krishnan J. Non-coding RNAs in cardiac regeneration. Front Physiol. 2021;12: 650566.
Article PubMed PubMed Central Google Scholar
Wang J-K, Li Y, Zhao X-L, Liu Y-B, Tan J, **ng Y-Y, et al. Ablation of plasma prekallikrein decreases low-density lipoprotein cholesterol by stabilizing low-density lipoprotein receptor and protects against atherosclerosis. Circulation. 2022;145:675–87.
Article CAS PubMed Google Scholar
Dixit G, Blair J, Ozcan C. Plasma proteomic analysis of association between atrial fibrillation, coronary microvascular disease and heart failure. Am J Cardiovasc Dis. 2022;12:81–91.
CAS PubMed PubMed Central Google Scholar
Lau ES, Paniagua SM, Zarbafian S, Hoffman U, Long MT, Hwang S-J, et al. Cardiovascular biomarkers of obesity and overlap with cardiometabolic dysfunction. J Am Heart Assoc. 2021;10: e020215.
Article CAS PubMed PubMed Central Google Scholar
Eckenstaler R, Ripperger A, Hauke M, Petermann M, Hemkemeyer SA, Schwedhelm E, et al. A thromboxane A2 receptor-driven COX-2-dependent feedback loop that affects endothelial homeostasis and angiogenesis. Arterioscler Thromb Vasc Biol. 2022;42:444–61.
Article CAS PubMed PubMed Central Google Scholar
Butenas ALE, Rollins KS, Williams AC, Parr SK, Hammond ST, Ade CJ, et al. Thromboxane A2 receptors contribute to the exaggerated exercise pressor reflex in male rats with heart failure. Physiol Rep. 2021;9: e15052.
Article CAS PubMed PubMed Central Google Scholar
Hariri E, Kakouros N, Bunsick DA, Russell SD, Mudd JO, Laws K, et al. Nonplatelet thromboxane generation is associated with impaired cardiovascular performance and mortality in heart failure. Am J Physiol Heart Circ Physiol. 2022;323:H248–55.
Article CAS PubMed PubMed Central Google Scholar
Alonso F, Dong Y, Génot E. Thrombomodulin, an unexpected new player in endothelial cell invasion during angiogenesis. Arterioscler Thromb Vasc Biol. 2021;41:1672–4.
Article CAS PubMed Google Scholar
Eidizadeh A, Schnelle M, Leha A, Edelmann F, Nolte K, Werhahn SM, et al. Biomarker profiles in heart failure with preserved vs. reduced ejection fraction: results from the DIAST-CHF study. ESC Heart Fail. 2023;10(1):200–10.
Zhang J, Zhou HJ, Ji W, Min W. AIP1-mediated stress signaling in atherosclerosis and arteriosclerosis. Curr Atheroscler Rep. 2015;17:503.
Article PubMed PubMed Central Google Scholar
Huang Q, Qin L, Dai S, Zhang H, Pasula S, Zhou H, et al. AIP1 suppresses atherosclerosis by limiting hyperlipidemia-induced inflammation and vascular endothelial dysfunction. Arterioscler Thromb Vasc Biol. 2013;33:795–804.
Article CAS PubMed PubMed Central Google Scholar
Sickinghe AA, Korporaal SJA, den Ruijter HM, Kessler EL. Estrogen contributions to microvascular dysfunction evolving to heart failure with preserved ejection fraction. Front Endocrinol (Lausanne). 2019;10:442.
Article PubMed Google Scholar
Yoshida H, Matsui T, Yamamoto A, Okada T, Mori K. XBP1 mRNA is induced by ATF6 and spliced by IRE1 in response to ER stress to produce a highly active transcription factor. Cell. 2001;107:881–91.
Article CAS PubMed Google Scholar
Fu F, Doroudgar S. IRE1/XBP1 and endoplasmic reticulum signaling — from basic to translational research for cardiovascular disease. Curr Opin Physiol. 2022;28: 100552.
Article CAS PubMed PubMed Central Google Scholar
Schiattarella GG, Altamirano F, Kim SY, Tong D, Ferdous A, Piristine H, et al. Xbp1s-FoxO1 axis governs lipid accumulation and contractile performance in heart failure with preserved ejection fraction. Nat Commun. 2021;12:1684.
Article CAS PubMed PubMed Central Google Scholar
Li L, Zhao Q, Kong W. Extracellular matrix remodeling and cardiac fibrosis. Matrix Biol. 2018;68–69:490–506.
Article PubMed Google Scholar
Faul C. Cardiac actions of fibroblast growth factor 23. Bone. 2017;100:69–79.
Article CAS PubMed Google Scholar
Rodríguez C, Martínez-González J. The role of lysyl oxidase enzymes in cardiac function and remodeling. Cells. 2019;8(12):1483.
Agrawal V, Fortune N, Yu S, Fuentes J, Shi F, Nichols D, et al. Natriuretic peptide receptor C contributes to disproportionate right ventricular hypertrophy in a rodent model of obesity-induced heart failure with preserved ejection fraction with pulmonary hypertension. Pulm Circ. 2019;9:2045894019878599.
Article PubMed PubMed Central Google Scholar
Gupta MP. Factors controlling cardiac myosin-isoform shift during hypertrophy and heart failure. J Mol Cell Cardiol. 2007;43:388–403.
Article CAS PubMed PubMed Central Google Scholar
Yu J, Yang Y, Xu Z, Lan C, Chen C, Li C, et al. Long noncoding RNA ahit protects against cardiac hypertrophy through SUZ12 (suppressor of Zeste 12 protein homolog)-mediated downregulation of MEF2A (myocyte enhancer factor 2A). Circ Heart Fail. 2020;13: e006525.
Article CAS PubMed PubMed Central Google Scholar
Tanaka K, Valero-Muñoz M, Wilson RM, Essick EE, Fowler CT, Nakamura K, et al. Follistatin like 1 regulates hypertrophy in heart failure with preserved ejection fraction. JACC Basic Transl Sci. 2016;1:207–21.
Article PubMed PubMed Central Google Scholar
Seki M, Powers JC, Maruyama S, Zuriaga MA, Wu C-L, Kurishima C, et al. Acute and chronic increases of circulating FSTL1 normalize energy substrate metabolism in pacing-induced heart failure. Circ Heart Fail. 2018;11: e004486.
Article CAS PubMed PubMed Central Google Scholar
Qin M, Huang H, Wang T, Hu H, Liu Y, Cao H, et al. Absence of Rgs5 prolongs cardiac repolarization and predisposes to ventricular tachyarrhythmia in mice. J Mol Cell Cardiol. 2012;53:880–90.
Article CAS PubMed Google Scholar
Li H, He C, Feng J, Zhang Y, Tang Q, Bian Z, et al. Regulator of G protein signaling 5 protects against cardiac hypertrophy and fibrosis during biomechanical stress of pressure overload. Proc Natl Acad Sci USA. 2010;107:13818–23.
Article CAS PubMed PubMed Central Google Scholar
Al-Kindi SG, Buzkova P, Shitole SG, Reiner AP, Garg PK, Gottdiener JS, et al. Soluble CD14 and risk of heart failure and its subtypes in older adults. J Card Fail. 2020;26:410–9.
Article PubMed PubMed Central Google Scholar
De Franceschi N, Peuhu E, Parsons M, Rissanen S, Vattulainen I, Salmi M, et al. Mutually exclusive roles of SHARPIN in integrin inactivation and NF-κB signaling. PLoS ONE. 2015;10: e0143423.
Article PubMed PubMed Central Google Scholar
Lim S, Sala C, Yoon J, Park S, Kuroda S, Sheng M, et al. Sharpin, a novel postsynaptic density protein that directly interacts with the shank family of proteins. Mol Cell Neurosci. 2001;17:385–97.
Article CAS PubMed Google Scholar
Corker A, Neff LS, Broughton P, Bradshaw AD, DeLeon-Pennell KY. Organized chaos: deciphering immune cell heterogeneity’s role in inflammation in the heart. Biomolecules. 2021;12(1):11.
Essandoh K, Auchus RJ, Brody MJ. Cardiac decompensation and promiscuous prenylation of small GTPases in cardiomyocytes in response to local mevalonate pathway disruption. J Pathol. 2022;256:249–52.
Article CAS PubMed Google Scholar
Xu H, Shen Y, Liang C, Wang H, Huang J, Xue P, et al. Inhibition of the mevalonate pathway improves myocardial fibrosis. Exp Ther Med. 2021;21:224.
Article CAS PubMed PubMed Central Google Scholar
Nishizawa H, Maeda N, Shimomura I. Impact of hyperuricemia on chronic kidney disease and atherosclerotic cardiovascular disease. Hypertens Res. 2022;45:635–40.
Article PubMed Google Scholar
Mátyás C, Németh BT, Oláh A, Török M, Ruppert M, Kellermayer D, et al. Prevention of the development of heart failure with preserved ejection fraction by the phosphodiesterase-5A inhibitor vardenafil in rats with type 2 diabetes. Eur J Heart Fail. 2017;19:326–36.
Article PubMed Google Scholar
Cornuault L, Rouault P, Duplàa C, Couffinhal T, Renault M-A. Endothelial dysfunction in heart failure with preserved ejection fraction: what are the experimental proofs? Front Physiol. 2022;13: 906272.
Article PubMed PubMed Central Google Scholar
Simeunovic D, Odanovic N, Pljesa-Ercegovac M, Radic T, Radovanovic S, Coric V, et al. Glutathione transferase P1 polymorphism might be a risk determinant in heart failure. Dis Markers. 2019;2019:6984845.
Article PubMed PubMed Central Google Scholar
Singh MM, Kumar R, Tewari S, Agarwal S. Association of GSTT1/GSTM1 and ApoE variants with left ventricular diastolic dysfunction in thalassaemia major patients. Hematology. 2019;24:20–5.
Article CAS PubMed Google Scholar
Nagamine T, Gillette B, Pakhomov A, Kahoun J, Mayer H, Burghaus R, et al. Multiscale classification of heart failure phenotypes by unsupervised clustering of unstructured electronic medical record data. Sci Rep. 2020;10:21340.
Article CAS PubMed PubMed Central Google Scholar
Gulea C, Zakeri R, Quint JK. Model-based comorbidity clusters in patients with heart failure: association with clinical outcomes and healthcare utilization. BMC Med. 2021;19:9.
Article PubMed PubMed Central Google Scholar
Woolley RJ, Ceelen D, Ouwerkerk W, Tromp J, Figarska SM, Anker SD, et al. Machine learning based on biomarker profiles identifies distinct subgroups of heart failure with preserved ejection fraction. Eur J Heart Fail. 2021;23:983–91.
Article CAS PubMed Google Scholar
Franssen C, Chen S, Hamdani N, Paulus WJ. From comorbidities to heart failure with preserved ejection fraction: a story of oxidative stress. Heart. 2016;102:320–30.
Article CAS PubMed Google Scholar
Pfeffer MA, Shah AM, Borlaug BA. Heart failure with preserved ejection fraction in perspective. Circ Res. 2019;124:1598–617.
Article CAS PubMed PubMed Central Google Scholar
Tam MC, Lee R, Cascino TM, Konerman MC, Hummel SL. Current perspectives on systemic hypertension in heart failure with preserved ejection fraction. Curr Hypertens Rep. 2017;19:12.
Article PubMed PubMed Central Google Scholar
Hicklin HE, Gilbert ON, Ye F, Brooks JE, Upadhya B. Hypertension as a road to treatment of heart failure with preserved ejection fraction. Curr Hypertens Rep. 2020;22:82.
Article PubMed Google Scholar
Vedin O, Lam CSP, Koh AS, Benson L, Teng THK, Tay WT, et al. Significance of ischemic heart disease in patients with heart failure and preserved, midrange, and reduced ejection fraction: a nationwide cohort study. Circ Heart Fail. 2017;10(6):e003875.
Reding KW, Cheng RK, Vasbinder A, Ray RM, Barac A, Eaton CB, et al. Lifestyle and cardiovascular risk factors associated with heart failure subtypes in postmenopausal breast cancer survivors. JACC CardioOncol. 2022;4:53–65.
Article PubMed PubMed Central Google Scholar
Saiki H, Petersen IA, Scott CG, Bailey KR, Dunlay SM, Finley RR, et al. Risk of heart failure with preserved ejection fraction in older women after contemporary radiotherapy for breast cancer. Circulation. 2017;135:1388–96.
Article PubMed PubMed Central Google Scholar
Huang S, Cai T, Weber BN, He Z, Dahal KP, Hong C, et al. The association between inflammation, incident heart failure, and heart failure subtypes in patients with rheumatoid arthritis. Arthritis Care Res (Hoboken). 2021;75(5):1036–45.
Packer M. Link between synovial and myocardial inflammation: conceptual framework to explain the pathogenesis of heart failure with preserved ejection fraction in patients with systemic rheumatic diseases. Card Fail Rev. 2020;6:e10.
Gevaert AB, Boen JRA, Segers VF, Van Craenenbroeck EM. Heart failure with preserved ejection fraction: a review of cardiac and noncardiac pathophysiology. Front Physiol. 2019;10:638.
Article PubMed PubMed Central Google Scholar
Gao H, Patel S, Fohtung RB, Cawthon PM, Newman AB, Cauley JA, et al. Sex- and race-specific associations of bone mineral density with incident heart failure and its subtypes in older adults. J Am Geriatr Soc. 2023;71(3):742–55.
Sabbatini AR, Kararigas G. Menopause-related estrogen decrease and the pathogenesis of HFpEF: JACC review topic of the week. J Am Coll Cardiol. 2020;75:1074–82.
Article CAS PubMed Google Scholar
Ponikowski P, Voors AA, Anker SD, Bueno H, Cleland JGF, Coats AJS, et al. 2016 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure: The Task Force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC)Developed with the special contribution of the Heart Failure Association (HFA) of the ESC. Eur Heart J. 2016;37:2129–200.
Article PubMed Google Scholar
Savarese G, Stolfo D, Sinagra G, Lund LH. Heart failure with mid-range or mildly reduced ejection fraction. Nat Rev Cardiol. 2022;19:100–16.
Article PubMed Google Scholar
Bortolotti M, Polito L, Battelli MG, Bolognesi A. Xanthine oxidoreductase: One enzyme for multiple physiological tasks. Redox Biol. 2021;41: 101882.
Article CAS PubMed PubMed Central Google Scholar
Watanabe K, Watanabe T, Otaki Y, Shishido T, Murase T, Nakamura T, et al. Impact of plasma xanthine oxidoreductase activity in patients with heart failure with preserved ejection fraction. ESC Heart Fail. 2020;7:1735–43.
Article PubMed PubMed Central Google Scholar
Yoon S, Kim M, Lee H, Kang G, Bedi K, Margulies KB, et al. S-nitrosylation of histone deacetylase 2 by neuronal nitric oxide synthase as a mechanism of diastolic dysfunction. Circulation. 2021;143:1912–25.
Article CAS PubMed Google Scholar
van Dam S, Võsa U, van der Graaf A, Franke L, de Magalhães JP. Gene co-expression analysis for functional classification and gene-disease predictions. Brief Bioinformatics. 2018;19:575–92.
PubMed Google Scholar
Ata SK, Wu M, Fang Y, Ou-Yang L, Kwoh CK, Li X-L. Recent advances in network-based methods for disease gene prediction. Brief Bioinformatics. 2021;22(4):bbaa303.
Weiskopf NG, Rusanov A, Weng C. Sick patients have more data: the non-random completeness of electronic health records. AMIA Annu Symp Proc. 2013;2013:1472–7.
PubMed PubMed Central Google Scholar
Wells BJ, Chagin KM, Nowacki AS, Kattan MW. Strategies for handling missing data in electronic health record derived data. EGEMS (Wash DC). 2013;1:1035.
PubMed Google Scholar
Haneuse S, Daniels M. A general framework for considering selection bias in EHR-based studies: what data are observed and why? EGEMS (Wash DC). 2016;4:1203.
PubMed Google Scholar
Farmer R, Mathur R, Bhaskaran K, Eastwood SV, Chaturvedi N, Smeeth L. Promises and pitfalls of electronic health record analysis. Diabetologia. 2018;61:1241–8.
Article PubMed Google Scholar
Pellikka PA, She L, Holly TA, Lin G, Varadarajan P, Pai RG, et al. Variability in ejection fraction measured by echocardiography, gated single-photon emission computed tomography, and cardiac magnetic resonance in patients with coronary artery disease and left ventricular dysfunction. JAMA Netw Open. 2018;1: e181456.
Article PubMed PubMed Central Google Scholar
Pieske B, Tschöpe C, de Boer RA, Fraser AG, Anker SD, Donal E, et al. How to diagnose heart failure with preserved ejection fraction: the HFA-PEFF diagnostic algorithm: a consensus recommendation from the Heart Failure Association (HFA) of the European Society of Cardiology (ESC). Eur Heart J. 2019;40:3297–317.
Article PubMed Google Scholar
Triposkiadis F, Butler J, Abboud FM, Armstrong PW, Adamopoulos S, Atherton JJ, et al. The continuous heart failure spectrum: moving beyond an ejection fraction classification. Eur Heart J. 2019;40:2155–63.
Article PubMed PubMed Central Google Scholar
Thygesen SK, Christiansen CF, Christensen S, Lash TL, Sørensen HT. The predictive value of ICD-10 diagnostic coding used to assess Charlson comorbidity index conditions in the population-based Danish National Registry of Patients. BMC Med Res Methodol. 2011;11:83.
Article PubMed PubMed Central Google Scholar

Download references

Acknowledgements

We thank Ricardio Ramirez-Flores and Jovan Tanevski for helpful discussions that shaped the analysis. Laura Wienecke provided valuable feedback on the manuscript.

Funding

Open Access funding enabled and organized by Projekt DEAL. JDL, JHS, HCF, JSR, and RTL were funded by the Klaus-Tschira foundation through the Informatics for Life Consortium. JSR, RTL, and JB were supported by the Collaborative Research Center 1550 (SFB 1550, Molecular Circuits of Heart Disease) funded by the Deutsche Forschungsgemeinschaft (DFG).

Author information

Authors and Affiliations

Institute for Computational Biomedicine, Heidelberg University, Faculty of Medicine, and Heidelberg University Hospital, Bioquant, Heidelberg, Germany
Jan D. Lanzer, Julio Saez-Rodriguez & Rebecca T. Levinson
Department of General Internal Medicine and Psychosomatics, Heidelberg University Hospital, Heidelberg, Germany
Jan D. Lanzer, Hans-Christoph Friederich, Jobst-Hendrik Schultz & Rebecca T. Levinson
Faculty of Biosciences, Heidelberg University, Heidelberg, Germany
Jan D. Lanzer
Informatics for Life, Heidelberg, Germany
Jan D. Lanzer, Hans-Christoph Friederich, Jobst-Hendrik Schultz, Julio Saez-Rodriguez & Rebecca T. Levinson
Roche Pharma Research and Early Development, Pharmaceutical Sciences, Roche Innovation Center Basel, Basel, Switzerland
Alberto Valdeolivas
Institute of Experimental Cardiology, Medical Faculty Heidelberg, Heidelberg University, Heidelberg, Germany
Mark Pepin & Johannes Backs
DZHK (German Centre for Cardiovascular Research), Partner Site Heidelberg/Mannheim, Im Neuenheimer Feld 669, 69120, Heidelberg, Germany
Mark Pepin & Johannes Backs
Department of Cardiology, Internal Medicine III, Heidelberg University Hospital, Heidelberg, Germany
Hauke Hund & Norbert Frey

Authors

Jan D. Lanzer
View author publications
You can also search for this author in PubMed Google Scholar
Alberto Valdeolivas
View author publications
You can also search for this author in PubMed Google Scholar
Mark Pepin
View author publications
You can also search for this author in PubMed Google Scholar
Hauke Hund
View author publications
You can also search for this author in PubMed Google Scholar
Johannes Backs
View author publications
You can also search for this author in PubMed Google Scholar
Norbert Frey
View author publications
You can also search for this author in PubMed Google Scholar
Hans-Christoph Friederich
View author publications
You can also search for this author in PubMed Google Scholar
Jobst-Hendrik Schultz
View author publications
You can also search for this author in PubMed Google Scholar
Julio Saez-Rodriguez
View author publications
You can also search for this author in PubMed Google Scholar
Rebecca T. Levinson
View author publications
You can also search for this author in PubMed Google Scholar

Contributions

JDL and RTL conceived the analysis in this paper. JDL designed, and performed all computational analyses. AV assisted in experimental design. MP and JB performed mouse experiments. HH and JHS provided pre-processed clinical data or assisted in data acquisition. RTL and JDL interpreted the data. AV and JSR provided support in interpretation of the data. NF assisted in the interpretation of the patient cohort. HCF, JHS, and JSR obtained the funding to support this work. JDL drafted this manuscript. RTL, AV, MP, HH, JB, HCF, JHS, and JSR provided critical revision of the manuscript. JHS, JSR, and RTL supervised this work. All authors read and approved the final manuscript. RTL and JDL have full access to the data in the study and takes responsibility for the integrity of the data and accuracy of the analyses.

Corresponding authors

Correspondence to Jan D. Lanzer or Rebecca T. Levinson.

Ethics declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki. The ethics committee of the University Hospital Heidelberg approved this study (approval ID: S-571/2019).

Consent for publication

All authors provide consent for the publication of this manuscript.

Competing interests

JSR has received funding from GSK and Sanofi and consultant fees from Travere Therapeutics. AV is currently employed by F. Hoffmann-La Roche Ltd.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1: Supplementary Methods and Supplementary Figures S1-13. Methods:

Processing of Comorbidities and Other Clinical Predictors Patient classifier Comorbidity profile assignment compared for effects of age, sex, time to HF diagnosis and time of recording Comorbidity network clustering Disease network comparison Disease prediction metrics Heart failure gene sets. Fig. S1. ICD10 code map**. Fig. S2. Patient classifier training.Fig. S3. Parameter estimates from the patient classifier.Fig. S4. Time to HF and time of comorbidity profile assignment.Fig. S5. Comparison of comorbidities between HFpEF and HFrEF cohort.Fig. S6. HFnet overview.Fig. S7. Comparison of disease networks.Fig. S8. Comparison of centralities.Fig. S9. Comparison of clustering algorithms in the HFnet.Fig. S10. Leave-one-out cross validation of disease gene prediction.Fig. S11. HF subtype gene prediction.Fig. S12. L-NAME/HFD Phenotype data.Fig. S13. Myocardial gene expression in L-NAME/HFD. L-NAME/ HFD mouse model and RNA sequencing.

Additional file 2: Supplementary Table 1.

Manual disease code map**s

Additional file 3: Supplementary Table 2.

Centrality of nodes in HFnet

Additional file 4: Supplementary Table 3.

Disease Cluster Composition of the HFnet

Additional file 5: Supplementary Table 4.

Edge list of HFhetnet

Additional file 6: Supplementary Table 5.

Predicted HFpEF and HFrEF genes

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Cite this article

Lanzer, J.D., Valdeolivas, A., Pepin, M. et al. A network medicine approach to study comorbidities in heart failure with preserved ejection fraction. BMC Med 21, 267 (2023). https://doi.org/10.1186/s12916-023-02922-7

Download citation

Received: 30 December 2022
Accepted: 05 June 2023
Published: 24 July 2023
DOI: https://doi.org/10.1186/s12916-023-02922-7

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

Abstract

Background

Methods

Results

Conclusions

Graphical Abstract

Similar content being viewed by others

Background

Methods

Study population

Multiple correspondence analysis (MCA)

Patient classifier

HF comorbidity network (HFnet) construction

HF heterogeneous network (HFhetnet) construction

Disease-gene association

Gene–gene association

Disease–gene prediction and prioritization

Transcriptome analysis

Results

The study population

High variation in comorbidity profiles is associated with HFpEF/HFrEF subtype

Distinctive comorbidity profiles derived for HFpEF and HFrEF

The HF comorbidity network (HFnet) captures HF specific disease relationships

Building the HF heterogeneous network (HFhetnet)

Estimating the success of disease–gene prediction within the HFhetnet

Predicting genes associated with comorbidity profiles of HFpEF and HFrEF

Corroboration of HFpEF gene candidates in the transcriptome of a murine HFpEF model

Discussion

Conclusions

Availability of data and materials

Abbreviations

References

Acknowledgements

Funding

Author information

Authors and Affiliations

Contributions

Corresponding authors

Ethics declarations

Ethics approval and consent to participate

Consent for publication

Competing interests

Additional information

Publisher’s Note

Supplementary Information

Rights and permissions

About this article

Cite this article

Share this article

Keywords

Search

Navigation