Clinical epigenetics settings for cancer and cardiovascular diseases: real-life applications of network medicine at the bedside

Despite impressive efforts invested in epigenetic research in the last 50 years, clinical applications are still lacking. Only a few university hospital centers currently use epigenetic biomarkers at the bedside. Moreover, the overall concept of precision medicine is not widely recognized in routine medical practice and the reductionist approach remains predominant in treating patients affected by major diseases such as cancer and cardiovascular diseases. By its’ very nature, epigenetics is integrative of genetic networks. The study of epigenetic biomarkers has led to the identification of numerous drugs with an increasingly significant role in clinical therapy especially of cancer patients. Here, we provide an overview of clinical epigenetics within the context of network analysis. We illustrate achievements to date and discuss how we can move from traditional medicine into the era of network medicine (NM), where pathway-informed molecular diagnostics will allow treatment selection following the paradigm of precision medicine.


Introduction
Despite advances in early detection, therapeutic strategies, and supportive care, cancer and cardiovascular diseases (CVDs) remain the leading causes of morbidity and mortality worldwide [6,140], WHO reveals leading causes of death and disability worldwide: 2000-2019, Cancer (who.int)). In 2015, then US President Barak Obama launched a precision medicine initiative [105] with the goal of refining the current standard of care. To achieve this goal, noninvasive biomarkers are needed that can diagnose diseases on a mechanistic level, as well as novel or repurposed drugs that are able to target precisely these mechanisms. Integration of clinical with genomics or even multi-omics datasets such as transcriptomics and proteomics [146] has revealed novel disease-specific molecular pathways in cancer [12,43,102] and CVDs [76]. However, the connection of hereditary genetic determinants (which may allow patient stratification) and the individual risk for cancer or CVDs is lacking. This "missing heritability" eludes us due to the complex interplay of multiple factors in complex diseases, including epigenetics, which has already been shown to pro-Sarno et al. Clin Epigenet (2021) 13:66 fetal development, where epigenetic factors have been associated with an increased risk of CVDs in later life (transgenerational effect) [36,[98][99][100][101].
Epigenetics comprises regulatory mechanisms that do not affect the DNA sequence but alter chromatin compaction to modulate gene expression [46]. DNA methylation, regulation of chromatin accessibility, and histone tail modifications are key moderators of vital cellular processes, such as differentiation, survival, and response to external stimuli [46]. These mechanisms individually or in combination are often related to pathogenesis and offer an opportunity to improve disease diagnosis and to predict clinical outcomes [17]. As epigenetic modifications are partially reversible, chromatin-acting epi-drugs can be used to treat complex human diseases. DNA methyl transferase inhibitors (DNMTi), histone deacetylase inhibitors (HDACi), and enhancer of zeste 2 polycomb repressive complex 2 subunit inhibitors (EZH2i) have rapidly reached considerable clinical relevance. For the treatment of hematologic malignancies, one epigenetic biomarker [124] and eight epi-drugs Table 1 have been approved by the FDA and are currently in clinical trials for different solid malignancies Table 2. In contrast, there are, as yet, no approved epi-drugs for CVDs, however, more than twenty Phase 3-4 randomized trials are currently ongoing to evaluate the epigenetic efficacy of the so-called "repurposed" drugs against CVDs, including metformin, statins, and apabetalone, a bromodomain inhibitor with quinazoline structure [98,99] Table 2. The historically limited success in the discovery of epigenetic biomarkers and epi-drugs using a reductionist approach calls for a paradigm shift toward network medicine (NM), which combines big data, advanced bioinformatic tools, network science, systems biology, artificial intelligence, and clinical biometric data to investigate the pathogenesis of complex diseases such as cancer and CVDs. By considering the molecular perturbations of integrated biological pathways rather than a single molecular defect ( [15,75,92,64,120,123]), NM can leverage molecular interaction networks for advancing diagnosis, prognosis, and treatment Fig. 1 [15,75,120,123]. In this way, NM paves the way toward precision medicine and personalized therapy [54]. Focusing thus far mostly on genomics and transcriptomics, NM has recently been applied to the study epigenetic mechanisms such as DNA methylation changes in the pathogenesis of cancer (mostly) and CVDs Fig. 1.
The goal of this review is to present clinical applications of epigenetics in cancer and CVDs. Additionally, we discuss how NM can gradually be advanced from the bench to bedside via epigenetics. With this overview, we provide information and insights for both basic scientists and physicians who work at the interface of these new applications of clinical epigenetics and Network Medicine.

Approved epi-treatments against hematopoietic cancers
The main classification for epigenetic mediators divides them into writers, readers, and erasers [18]. DNMTs and HATs are "writers, " as they add a methyl or acetyl group on chromatin. These residues can be removed by the "erasers" such as HDACs or KDMs. Chromatin modifications may be interpreted by "readers, " such as the CBX (chromodomain) or BRD (bromodomain) family members. Writers, readers, and erasers comprise highly studied targets, some of which have led to the identification and development of FDA-approved drugs entering into the clinics.
DNMTi and HDACi are the epi-compounds most used for treatment. DNA hypermethylation is frequently associated with cancer development, and the two DNM-Tis, azacitidine (Vidaza) and decitabine (Dacogen), were approved for myelodysplastic syndromes (MDS) treatment in 2004 and 2006, respectively. These hypomethylating agents are currently the first-line therapy for MDS, when stem cell transplantation is unsuitable. Administered at a daily dose of 75 mg/m 2 for 7 days (azacitidine) and 5-20 mg/m 2 for 5/7 days (decitabine), these treatments are more effective than previously investigated drugs [110]. However, azacitidine has some limitations, including a low response rate, short duration of action, and aggravation of thrombocytopenia, leading to its co-administration with other therapies ( Table 2,  NCT01488565). Azacitidine is currently also under investigation as a treatment option in solid malignancies (see Sect. 2.4 for further information).
Decitabine is well tolerated and although the most significant effects have been observed in hematologic malignancies, it also displays a good activity against solid tumors. Different Phase I/II clinical trials are currently underway, including co-administration studies with HDACi for patients with advanced solid tumors (NCT01023737, NCT02453620, NCT03925428, NCT03590054, NCT01281176) or with cisplatin for resistant ovarian cancer [87]. Decitabine has shown numerous beneficial effects in solid tumors, leading to further investigation of methylation status as a prognostic marker in solid cancers. The FDA has approved second-generation HDACi, including vorinostat (Zolinza), belinostat (Beleodaq), romidepsin (Istodax), and panobinostat (Farydak) for treatment of cutaneous T-cell lymphoma (CTCL), peripheral T-cell lymphoma (PTCL), and multiple myeloma (MM).        Tavazzi et al. [128] Abbreviations Mutations in R132 and R172 in isocitrate dehydrogenase 1 and 2 (IDH1/2) have been found in approximately 20% of acute myeloid leukemias (patients with a devastating prognosis [109]. These mutations produce an abnormal conversion of α-ketoglutarate (D-2-kg) in α-hydroxyglutarate (2-HG), leading to a dysfunction of the enzymes using 2-kg as a cofactor, such as TETs and KDM enzymes, associated with DNA hypermethylation, "aberrant gene expression, " increase of proliferation and cell differentiation [95]. Recently, the FDA approved the first IDH1/2 inhibitors (NCT01915498,NCT02074839) against AML. Enasidenib (IDHIFA), targeting IDH2, and ivosidenib (TIBSOVO), IDH1, are administrated orally, blocking 2-HG in the blood of AML patients with IDH mutations. The patients treated with these agents showed a complete response (CR) or CR with partial hematological recovery after 8.2 months of treatment, response was superior with ivosidenib (32.8%) superior to the enasidenib treatment (23%). Ivosidenib has been approved in 2019 for AML patients older than 75 years or for patients in whom chemotherapy cannot be used.

Clinical breakthroughs after more than one decade of HDAC inhibitors use
Vorinostat, an HDACi used as an oral agent for CTCL since 2006 and now tested in clinical trials for solid tumors, displays low toxicity and high efficacy, making it a first-line drug for the treatment of this lymphoma. The Like vorinostat, romidepsin is one of the first drugs to be used for CTCL. Romidepsin is a potent class I HDACi and its ability to inhibit class II HDACs at higher concentrations suggests that it may act as a broad-spectrum HDACi [52]. Unlike other drugs, this molecule induces a change in electrocardiographic patterns, after administration, with flattening of ST and T waves and depression of the ST segment. Six patients with mild cardiovascular disorders died after romidepsin treatment in Phase II studies. This drug also shows significant activity in patients with PTCL and in CTCL patients with stage IIB or higher disease presentations [107]. To date, romidepsin has proved unsuccessful as monotherapy in squamous cell carcinoma of the head and neck (SCCHN) [57] and in metastatic castration-resistant prostate cancer (CRPC) [94] showing limited adverse effects such as fatigue, nausea, and vomiting.
Belinostat, approved in 2014 for PTCL treatment [74], generally displays no major adverse effects, but in rare cases induces liver damage. Belinostat is currently in several clinical trials Table 1.
Panobinostat was been approved in 2015 for multiple myeloma (MM) as a second-line therapy in patients not responding to bortezomib. The limitation of panobinostat has been the low efficacy when given as a single agent; thus, it is administered in combination with bortezomib and dexamethasone. MM patients treated with panobinostat in combination with bortezomib have a significant increase in survival (NCT01023308). Additionally, when studied in a Phase III trial in non-Hodgkin lymphoma patients, panobinostat-induced adverse events in 22% of the patients with severe vomiting and diarrhea (NCT01034163), and hepatic abnormalities in chronic myeloid leukemia (CML) (NCT00449761), although it shows promising anticancer effects, alone or in combination [88].
Although HDACi are excellent candidates for integrative network analysis due to their extensively studied molecular mechanisms, no network-based approaches were reported for repurposing this class of drugs to other cancer types. However, by using a meta-analytical approach Rafehi et al. integrated ENCODE data with microarray expression profiles showing HDACi-mediated suppression of EP300 target genes, including genes implicated in diabetes mellitus [108].

Epi-treatments being investigated for solid cancers
Mutations and overexpression of EZH2 have been associated with prostate, breast, liver, skin, lung, and gastric cancers, as well as with lymphoma and melanoma [1,7,8,58,60,114,129,139]. Tazemetostat (Tazverik, Epizyme, Inc), acting as a selective competitor of S-adenosyl-Lmethionine, is a first-in-class EZH2 inhibitor, administered orally against hematological and solid tumors. On June 18, 2020, the FDA granted approval to tazemetostat in adults with relapsed/ refractory (R/R) follicular lymphoma (FL) carrying an EZH2 mutation who have received at least two prior systemic therapies, or for those patients who have no alternative options. The cobas EZH2 Mutation Test (Roche Molecular Systems, Inc.) as a diagnostic test was also approved, based on two open-label, single-arm cohorts of a multicenter trial (E7438-G000-101, NCT01897571) in patients with FL after at least two prior systemic therapies.
Tazemetostat was also studied as a single agent in a Phase II clinical trial for patients with relapsed or refractory non-Hodgkin lymphoma [73], rare aggressive forms of lymphoma [85], follicular lymphoma in refractory patients in combination with atezolizumab (Tecentriq), diffuse large B-cell lymphoma (DLBCL), and is currently in the recruitment phase for prostate cancer in combination with enzalutamide or abitarerone/prednisol (NCT04179864). In combination with R-CHOP, a standard chemotherapy regimen, tazemetostat, is in ongoing studies as a first-line treatment for newly diagnosed highrisk elderly patients with DLBCL.

Ongoing clinical trials with epi treatments against cancer
Several chromatin-acting drugs are being investigated in clinical trials; see Table 2 for a complete overview. DNMTi and HDACi administration produces a synergistic effect on methylation state, with an increase in repression of pro-oncogenic and activation of apoptotic genes [96,106,145]. Clinical trials in different phases testing the combined administration of both classes of drugs are yielding excellent results [19]. Although toxicity appears more frequently in older patients, azacitidine proved effective in MDS and in AML (NCT01074047), where it represents an excellent replacement therapy for patients who are not candidates for more aggressive therapies (65-74 years) and for those with cytogenetics indicating an increased risk. The HDACi entinostat, a generally well-tolerated drug, is currently in Phase II trials for co-administration with azacitidine and nivolumab in patients with metastatic non-small lung cell cancer (NCT01928576), and in melanoma and lymphoma in co-administered with pembrolizumab (NCT03179930,NCT03765229). Activity of the EZH2 inhibitor CPI-1205 in combination with the antiandrogen enzalutamide (Xtandi) is being tested in Phase I/II trials as second-line treatment of metastatic castrationresistant prostate cancer patients (NCT03480646), and showing excellent tolerance and considerable anticancer activity, prompting further studies. The DOT1L inhibitor pinometostat (EPZ-5676) is part of an ongoing Phase I/II study for the treatment of AML and MLL (mixedlineage leukemia) (NCT03701295,NCT03724084). Tranylcypromine, an LSD1 inhibitor, is under investigation in at least 26 studies. A recent interventional study testing tranylcypromine with all-trans retinoic acid (ATRA) in AML and MDS found no serious side effects (NCT02273102), leading now to the more intensive evaluation study of its anticancer effects.
Osteosarcoma (OS) patients are generally treated with combination chemotherapy comprising cisplatin, doxorubicin, and high-dose methotrexate, with addition of ifosfamide [130]. However, a fundamental problem is the long duration of treatment, which leads to undesirable effects [118]. Several clinical trials in initial stages involve single or co-administration of epi-drugs [37]. EZH2 is a positive regulator of the growth of metastasis and prognosis of OS [38]. In January 2020, tazametostat (Tazverik) was quickly approved by the FDA for advanced metastatic and locally epithelioid sarcoma treatment in adults and pediatric patients (16 years), based on the successful results obtained in a phase 2 study (NCT02601950).
Additionally, tazametostat is in Phase 2 clinical trial against OS with mutations of EZH2, SMARCB1 and SMARCA2 (NCT03213665) in pediatric patients.
Despite the steady rise in the number of clinical studies that include epi-drugs as anticancer agents, only a few agents and combinations have proven useful for broad clinical use. Thus, a joint approach involving basic scientists and physicians for the development of clinically useful protocols is needed to improve our understanding of epigenetic mechanisms and how they can be targeted most effectively. This goal is particularly important for solid malignancies, where the use of chromatin-targeting drugs has not yet been shown to provide benefit over standard treatments. With respect to hematological malignancies, the clearly lower efficacy cannot only be attributed to limited knowledge of the oncogenic mechanism. Possible additional factors might include differences in solid tumor heterogeneity or in the 3D structures of solid malignancies where cancer cells in poorly vascularized tumor areas are difficult to reach with drugs. For all of these reasons and others to be uncovered, network medicine strategies may prove fundamental for an understanding of (epi)based treatments and for assessing responses, especially in treatment-resistant phenotypes.

Network medicine in the clinical setting of cancer prevention and diagnosis
Integrating drug discovery and pathology with network medicine analysis provides rational and efficient approach to identify novel treatments and allows repurposing clinically approved drugs for human diseases, with implications for personalized medicine. Recent examples illustrate how network medicine has been adapted in the field of epigenetics for diagnosis, precision medicine, and patient stratification in terms of prognosis Table 3.

Network medicine and biomarkers in cancer prevention, diagnosis, management and prognosis
DNA methylation is the most widely investigated epigenetic mechanism in network analysis Fig. 1. Changes in DNA methylation in promoter or other regulatory regions can reveal associations to cancer development and drug resistance [2,26,47,82] and allow predictions with high accuracy (reviewed in [111]). For instance, Capper et al. [23] trained a random forest classifier to distinguish between tumor entities of the central nervous system. The resulting epigenetic signatures typically consist of many features with unknown mechanistic explanation. Moreover, given that features tend to be highly correlated, they can often be replaced in a signature without a loss of accuracy. A consequence of this often-observed lack of robustness [133] is that we cannot distinguish between surrogates and features causally involved in the disease. For functional enrichment analysis, features (e.g., CpGs or regulatory regions such as promoters, repressors, or enhancers) are typically mapped to the closest gene using tools such as GREAT [89]. Such a gene-mapping is also necessary to leverage existing networks such as protein-protein interaction (PPI) or gene-regulatory networks in epigenomics data analysis, where the putative target gene with the closest transcription start site is considered. West et al. showed that such a mapping strategy can be used to extract subnetworks of cancer-related differentially methylated genes [135]. Subsequently, Jiao et al. proposed the functional epigenetic modules (FEM) method [67], which revealed HAND2 as a methylation hotspot in the endometrium and indicator of drug response in progesterone treatment [69]. Kim and Sun [71] showed that PPI networks are beneficial in network-regularized feature selection after dimension reduction. Li et al. used principal component analysis for feature reduction and sparse canonical correlation analysis to infer edge weights for gene pairs. A network-based pathway-extending approach using DNA methylation and gene expression data identified altered pathways [77]. Rather than relying on a PPI network as prior information, disease-specific regulatory networks can be inferred directly from epigenomic data. With their epigenetic module based on differential networks (EMDN) algorithm, Ma et al. used The Cancer Genome Atlas Program (TCGA) data to construct both a coexpression and a co-methylation network [84]. Bartlett Table 3 Examples of epigenetics and network-oriented analysis in cancer and CVD susceptibility Network- [143]. Using crosstalk between genetic modules, Cui et al. [35] created a comethylation network based on DNA methylation data. The K-shell algorithm, applied to three types of cancer, invasive breast carcinoma, skin cutaneous melanoma, and uterine corpus endometrial carcinoma, identified the main genes in the modules that are predictive of prognosis and classification. In invasive breast carcinoma, this network method identified ten genes responsible for metastasis and tumor progression. RCHY1 was found at the junction of two modules, closely related to the histone lysine demethylase KDM1B, which plays a key role in methylation and silencing. In addition to the methylation state, modification of the methylene group on DNA can be associated with cancer diagnosis. The conversion of 5-methylcytosine (5mC) to 5-hydroxymethylcytosine (5hmC) was studied in 30 glioblastomas using the OxyBS algorithm with an enrichment analysis via Genomic Regions Enrichment of Annotations Tool (GREAT). The results implicate the depletion of 5hmC in various cancer types, which is associated with an increase in proliferation markers [68]. Despite these promising early results in the network-based integration of DNA methylation in oncology, clinical translation of epigenomics-based network medicine methods is still lacking. In particular, the integration of epigenomics data in a multi-omics context remains challenging. A weighted correlation network analysis (WGCNA) of 201 patients in a TCGA prostate cancer dataset revealed hypermethylation of FOXD1 as an unfavorable marker for survival [141]. The combination of mRNA expression and DNA methylation datasets in WGCNA and downstream gene ontology (GO) enrichment using single sample gene set enrichment analysis (ssGSEA) yielded a 13-gene epigenetic signature associated with survival of breast cancer patients. This panel of genes considered upregulation of known cancerrelated pathways (e.g., mTOR signaling) to distinguish high-from low-risk cancer cases [9]. The genomes of 200 clinically annotated adult cases of de novo AML were studied using whole-genome or whole-exome sequencing as well as RNA-seq, microRNA-seq, and DNA methylation profiling. A potential driver mutation in DNA methylation and RNA-seq (e.g., in DNMT3A, NPM1, CEPBA, IDH1/2, and RUNX1) promoting AML pathogenesis in individual patients was found in all AML samples (Cancer Genome Atlas Research et al. [22]. The tool SWItchMiner (SWIM) was applied to RNA-sequencing data from TCGA [90] to characterize disease etiologies and to identify potential therapeutic targets [103]. In glioblastoma multiforme (GBM), SWIM revealed new insights into the molecular mechanism determining the stem-like phenotype of glioblastoma cells [49]. Here SWIM implicated FOSL1 as a putative master regulator of a core of four master neurodevelopmental transcription factors (i.e., SOX2, SALL2, OLIG2, POU3F2), whose induction was demonstrated to be sufficient to reprogram fully differentiated glioblastoma cells into stem-like cells [125]. More recently, SWIM was used in conjunction with network analysis to enhance disease module discovery [104]. Corces et al. [34] generated high-quality ATAC-seq data for 410 TCGA samples and identified cancer-and tissue-specific DNA regulatory elements in 23 cancer types. These chromatin accessibility profiles allowed the classification of cancer subtypes with a newly recognized prognostic importance. A gene co-expression network associated with ATAC-seq analysis was built by Dravis et al. [39] to analyze breast cancer biomarkers.
Here, SOX10 was identified as the major transcription factor that binds to genes responsible for neural activity.

Network medicine in cancer patient stratification using single-cell analyses
Network medicine has enabled the identification of molecular markers defining cancer subtypes [81]. Sinkala et al. stratified 185 patients with pancreatic cancer into two groups via proteomics data. Next, they built patientsimilarity networks for each of the molecular data types in TCGA which they then integrated into a joint network using similarity network fusion (SNF) [121]. SNF clustered the patients into two subtypes based on their joint molecular profiles. In a subsequent step, the K-nearest neighbor (KNN) algorithm and support vector machines (SVM) were used to identify representative biomarkers for these clusters in individual data types. While it has not yet entered clinical practice, singlecell epigenetics is an exciting new frontier that allows for studying small and previously unrecognized cell populations such as stem cells Table 4 [56,122], preimplantation embryos [144], and the heterogeneity of subpopulations of human tissues [53,83]. Single-cell epigenetics also has the potential to be applied in the diagnosis and prognosis in cancer treatment, where identifying the mutational profile might potentially impact disease evolution and response to treatment. Several sequencing protocols are developed in Table 4, including the single-cell bisulfite sequencing (scBS-seq) method to determine the DNA methylome of cell populations [122]. Recent single-cell approaches combine the methylome with RNA transcription profiling [63]. Moreover, single-cell chromatin accessibility profiling with scATAC-seq has provided the opportunity to recognize different populations within the same tumor and to discriminate cancer cell populations and their heterogeneity [21]. Recently, scTRIO-seq, which allows parallel profiling of genomic, transcriptomic, and DNA methylation in single cells, has led to the identification of two different hepatocellular carcinoma (HCC) subpopulations in the same patient. Based on the DNA and transcriptome of 25 HCC cells, Yu Hou et al. found two distinct populations differing in DNA copy numbers, DNA methylation, or RNA expression levels, of which the smaller tumors expressed more invasive cell markers [61]. scATAC-seq, used to identify co-varying transcription factor-cell surface marker pairs, was combined with scRNA-seq for cell surface marker expression to detect efficiently CD24-labeled chronic leukemia cells as co-varying with chromatin accessibility changes linked to GATA transcription factors [79]. A positive correlation was found between CD24 and GATA levels, characterized by high genetic and epigenetic variability, conferring resistance to imatinib mesylate treatment.

Network medicine for identification of cancer treatment response vs resistance
In the last 50 years, the identification of epigenetic molecular alterations acting as drivers of cancer development and progression have transformed the clinical practice of oncology from non-specific cancer cell elimination with nonspecific chemotherapies to a more cancer-selective approach that leverages molecular profiles [12]. Predicting drug resistance, and therefore, modifying treatment according to the phenotypic response of the tumor, remains a major challenge. Several resources, such as the Genome of Drug Sensitivity in Cancer (GDSC) database, provide information required to check tumor resistance to specific drugs (https:// www. cance rrxge ne. org/). Technologies for large-scale genomic and epigenomic profiling allowed the full (bulk) characterization of different tumors, revealing that they share similar driver mutations and enzymatic alterations [117]. For example, upregulation of SETDB1 and SETDB2 was found in resistant cells exhibiting a loss of H3K4me3 and H3K27me3 and an increase in H3K9me3 [3]. HDAC deregulation produced an aggressive phenotype in lung cancer cells resistant to doxorubicin [40]. The resistance to epigenetic therapies for a subpopulation of leukemia stem cells found in AML models [50] was not driven by genetic evolution, but was due to epigenetic plasticity [72]. Epigenetic modifications to centrosome proteins led to tumor development and drug resistance. The Manteia [127] gene ontology analysis data system reported a correlation between lysine acetyltransferase (KAT) 2A/B alteration and serine/ threonine-protein kinase PLK4 overexpression, producing resistance to tamoxifen and trastuzumab [51]. ChIPseq and DNA methylation profiles were used to study epigenetic profiles in drug-resistant melanoma, lung, and colon cancers. Thirteen genes associated with the interferon (IFN) pathway were found to be regulated by histone modifications, including the histone methyltransferase EZH2 [3]. Recent trials found that the response to a targeted drug depends on the anatomical cancer type. For example, BRAF-mutated melanoma, NSCLC, and hairy cell leukemia satisfactorily responded to vandetanib, a drug that targets the BRAF V600E mutation, while BRAF-mutated CRC does not [117]. Falcone et al. used a network approach based on the SWIM algorithm [44] to compare pairs of BRAF-mutated cancers and found a great number of switch genes suggesting that the cancer network of each tumor is different. A number of putative targetable kinases encoded by switch genes were reported in lung adenocarcinoma and thyroid cancer, while only one was found in colorectal cancer. Interestingly, the results were in accordance with clinical trial data showing a better response rate to vemurafenib in papillary thyroid cancer patients (overall response rate of 38.5%) than in colorectal cancer patients (ORR 4.8%) [44]. These results highlight the limit of the reductionist approach where typically one gene is implicated with one disease, as this strategy does not reflect the complexity of complex diseases such as cancer and CVDs in which many potential disease genes interact. Yildirim et al. [138] built a drug-target network, in which each drug was connected to its target proteins, and two complementary projections of it: a drug network, in which nodes are drugs and two drugs are connected to each other if they share the same targets, and a targetprotein network, in which the nodes are proteins and they are connected together if are targeted by the same drug. The authors observed that new drugs tend to target the already validated target proteins and that many clinically FDA-approved drugs do not target known disease-associated genes but are rather palliative drugs. A strategy aimed to increase the efficacy and to reduce the risk of adverse effects of monotherapy, is the therapy with a combination of multiple drugs. By analyzing the network-based relationship between drug pairs, their targets, and the proteins in the disease modules Cheng et al. revealed that drugs used in combination with synergistic effect have their drug-target modules overlapping with disease modules but well separated to each other in the human interactome [30,31]. Additionally, Tang et al. [126] developed a logic-based network pharmacology modeling approach, called TIMMA (Target Inhibition interaction using Minimization and Maximization Averaging), based on the integration of drug-target interaction profiles and single-drug sensitivities, to predict synergistic drug combinations. By applying TIMMA to the single-drug sensitivity profiles and to the kinomewide drug-target interaction of 41 kinase inhibitors in MDA-MB-231 cell line, the authors found a synergistic target interaction between inhibition of Aurora B, a key regulator of mitosis, and ZAK, a key regulator of p38 MAPK pathway, findings also confirmed in vitro.
Twin Convolutional Neural Network for Drugs in SMILES format (tCNNS) creates a drug correlation network using the SMILES chemical representation and information on the cell phenotype [80]. Pharmacological dose-response prediction can be also obtained by DeepDR based on mutation data from TCGA, a pretrained expression encoder, and a predictor network for drug response. From 9.059 tumor samples, DeepDR predicted cancer drug resistance and personalized therapy [32]. For resistant tumor treatment, one of the main unanswered questions is whether the cell acquires mutations during development, or whether there is a small group of cells within the same population that survives therapy. Single-cell analysis specifically allows for studying the complexity of a population [70]; however, the use of NM in epigenetics for the prediction of response versus treatment is only its inception [104].

Epi-therapy in cancer: can Network Medicine help physicians at the bedside?
Integrating network medicine with anticancer drug discovery programs would make a significant contribution to improving patient health. Recently, whole-genome bisulfite sequencing (WGBS) integrated with wholegenome sequencing (WGS) and RNA-seq was applied to 100 metastatic prostate biopsies to sequence the methylome whole genome, and transcriptome, respectively [142]. In 22% of resistant metastases, DNA methylation analysis identified an epigenetic subtype associated with hypermethylation and mutation of TET2, DNMT3B, IDH1, and BRAD, and regions where methylation is associated with AR, MYC, and ERG expression. NM has also been used in this context to predict potential implications of therapeutics. The administration of DNMTi, such as azacitidine or decitabine, which are used for myelodysplastic syndrome (MDS) treatment, may be advantageous in resistant metastases associated with hypermethylated gene promoters. Pursuing a strategy that allows for the identification of tumor subtype targets or a specific resistance mechanism may make it possible to repurpose drugs used in the treatment of other diseases.

Metabolic syndrome, diabetes, and foodome project
A notable expansion of the prevalence of metabolic syndrome (MS) has impact on the global risk of CVDs, in particular in aging Western world populations, which are prone to a high-fat diet and a sedentary lifestyle [42]. MS arises from complex crosstalk between genetic and epigenetic factors underlying obesity, in particular central abdominal obesity, that can strongly increase the risk for coronary heart disease (CHD), type 2 diabetes mellitus (T2D), and cancer [42,64,98,99]. In clinical practice, a panel of five well established cardiometabolic risk factors including abdominal obesity, increased triglycerides, decreased HDL cholesterol, hypertension, and hyperglycemia is assessed by physicians, and at least three of these must be present [42]. However, in precision medicine and personalized therapy, individual DNA methylation profiles might not only represent a promising tool for the prediction, diagnosis, and prognosis of obesity and MS, but also for improving treatments to lose body weight [113]. DNA methylation changes are key molecular drivers underlying the risk of MS upon detrimental exposures (e.g., nutritional patterns), especially during early development as well as during postnatal life, offering a possible framework by which to explain the "missing heritability" of MS [36,113]. Recently, a targeted DNA pyrosequencing and logistic regression analysis has revealed a significant positive relationship between DNA methylation levels at specific CpG islands in promoters of the PPARα and LPL genes and serum triglyceride levels (TG) in visceral adipose tissue samples from 53 MS patients in comparison with 55 healthy subjects [24]. In addition, a negative association linked methylation levels of the tumor necrosis factor gene with TG, glucose levels, HDL-c, and blood pressure suggesting a relevant factor potentially involved in preventing MS occurrence [24].
A study using the Illumina Methylation EPIC Beadchip on 1999 blood samples isolated in the Coronary Artery Risk Development in Young Adults (CARDIA) study has unveiled a strong positive association between accelerated epigenetic aging and both the MS severity score and the risk of developing MS, after adjusting for known risk factors [97]. This observation led to suggest that pharmacological and non-pharmacological interventions targeting the epigenetic aging process at the molecular level may potentially prevent MS.
Network analysis would be of paramount importance to understand the underlying molecular determinants and pathogenic processes that guide in the development of new predictive biomarkers and prevention tools. Thus far, network analysis has already been applied to separate MS components, mainly in T2D [48,98,99,119] and obese-T2D patients [62], and to construct gene regulatory networks, these analyses, however, did not consider epigenetic factors.
The epigenetic clock and nutritional epigenomics may contribute to aging as well as age-related diseases [4]. Thus, mapping food-related chemical profiles (diet) and molecular pathways is the main goal of the networkbased "Foodome" project (https:// www. barab asilab. com/ proje cts). Foodome emphasizes the use of digital eating patterns ("barcode") to define the personalized exposures to nutritional-chemical compounds [10]. Merging the barcode with genetic profiles and clinical data might fuel innovative platforms that can provide insights into the molecular basis for the "diet-mutation-disease risk" axis useful for precision medicine [10]. Since epigenetics plays a important role in the diet-genome crosstalk underlying vascular damage and CV risk, even during fetal development [36,98,99,101], the "individual foodome" could be integrated in longitudinal cohort studies investigating how epigenetic features change over time and in response to different nutritional exposures leading to the onset of CVDs.

Coronary heart disease (CHD)
Endothelial and systemic inflammation plays a relevant role in disease pathophysiology, and changes in DNA methylation levels of targeted genes may be causal or predispose to disease, contributing to destabilization and rupture of atherosclerotic plaques leading to acute cardiovascular events [64,65,101,115]. A recent study has emphasized the possible role of a DNA methylationbased risk score in optimizing the traditional predictors of CVD risk [136]. In addition, there is a possible correlation between blood-based methylation levels in the unique CGI regulating the human leukocyte antigen-G (HLA-G) gene, which encodes for an anti-inflammatory molecule with immunomodulatory properties, and cardiac computed tomography angiography (CCTA) features in patients with obstructive in comparison with non-obstructive CHD [116]. Hypomethylation of a specific fragment of CGI-associated HLA-G gene positively correlated with coronary calcium score and was predictive for disease severity suggesting that methylation might not only have a critical role in disease severity but also a role as noninvasive biomarker(s) improving the prognostic value of CCTA [116]. The WGCNA and Comb-p algorithms have been applied to the identification of blood-based differentially methylated regions (DMRs) and disease modules associated with incident CHD events in two independent cohorts (discovery set: 2129 women, replication set: 2726 subjects) [137]. This study has identified two modules highly enriched for development and immune-related processes. In addition, a multivariate analysis has revealed a positive correlation with BMI, highly sensitive C reactive protein (hsCRP), and TG [137]. Three DMRs annotated to the sodium/ hydrogen exchanger 1 (SLC9A1), solute carrier family 1 neutral amino acid transporter member 5 (SLC1A5), and trinucleotide repeat containing adaptor 6C (TNRC6C) genes significantly replicated across the two cohorts, providing possible useful predictive biomarkers [137]. However, the possible cause-effect relationship between methylation changes in these genes and CV risk needs to be determined.

Pulmonary arterial hypertension (PAH)
PAH is a rare and incurable disease characterized by vasoconstriction and consequent elevated pulmonary artery pressure owing to three main endophenotypes, endothelial dysfunction, cell proliferation/migration, and inflammation triggered by an interplay between genetic and epigenetic risk factors with exposure to detrimental environmental stimuli [25,86,100]. Data regarding the potential clinical relevance of differential epigenetic factors in PAH, mainly changes in DNA methylation, have been increasing in the past few years, and network-oriented approaches are helping to prioritize novel candidate genes and drug targets [100]. Recently, an integrated regulatory network has been constructed by integrating chromatin with transcriptomic and interaction profiling in pulmonary arterial endothelial cells (PAECs) obtained from end-stage PAH patients at the time of lung explant and control subjects [112]. As a result, an in-depth remodeling of active enhancers marked by H3K27ac and regulated by specific transcription factors may trigger perturbation of angiogenesis and endothelial-to-mesenchymal transition processes in PAECs in response to specific growth factor signals, as experimentally confirmed for target genes such as nitric oxide synthase 3 (NOS3) [112]. However, further studies will investigate a possible correlation between key gene regulatory networks and underlying PAH severity or responsiveness to vasodilatory therapy.

Heart failure (HF) and heart transplantation (HTx)
Heart failure (HF) affects approximately 20% of general population and contributes to 11% of deaths with an estimated incidence that will rise by 25% over the next 15 years [16]. HF can develop asymptomatically for years, and once diagnosed, the effectiveness of most drug therapy interventions are modest. Nonetheless, the recent tremendous impact, both of SGLT2 inhibitors and angiotensin-neprilysin inhibition, will likely reduce HF mortality [66,132]. However, the development of novel early biomarkers useful for stratification and/or as prognostic markers is a priority in the clinical setting of HF. Based on the ejection fraction (EF) value, we can classify three clinical phenotypes: HF with reduced EF (HFrEF), HF with mid-range EF (HFmrEF), and HF with preserved EF (HFpEF) [98]. An epigenetic-based phenotype mapping strategy of HFrEF, HFmrEF, and HFpEF patients seems to be a possible option to identify noninvasive biomarkers discriminating differential HF subgroups and/or novel drug targets to test in clinical trials aimed at establishing specific therapeutic strategies for each phenotypic profile [99]. However, the availability of cardiac biopsy from living patients and controls in humans is scarce; thus, most pathogenic studies are based on animal models. A pioneer multi-omics study integrated the methylome and transcriptome of left-ventricular biopsies and whole peripheral blood samples of 41 patients with HFmrEF caused by dilated cardiomyopathy and compared to 31 patients who underwent routine left-heart myocardial biopsy after receiving transplantation, as a control group. The promoter DNA hypomethylation of the natriuretic peptide A and B (NPPA and NPPB) genes regulates overexpression of these genes providing a novel putative class of HF biomarkers easily detectable in peripheral blood [91].
Graft surveillance after heart transplantation is a challenge in the management of transplanted patients, and current guidelines indicate that invasive endomyocardial biopsy is the gold standard to diagnose and monitor organ rejection. Ideally, graft rejection may be diagnosed and predicted by noninvasive biomarkers present in the peripheral blood or other biological fluids [99]. Specifically, an increasing interest in circulating epigenetic molecules is providing novel findings that may aid physicians in a more accurate risk stratification [99]. DNA methylation is a pivotal contributor to a balanced immune response toward the graft, due to regulation by both the innate and adaptive immune systems, with primarily T cells as the key players of alloreactivity and targets for immunosuppressive drugs [27,28]. Interestingly, FOXP3 gene expression was significantly higher in biopsy samples of rejectors collected before rejection in comparison with non-rejectors, and showed the tendency to predict rejection events [20]. Longitudinal studies could evaluate the possible role of this biomarker in the clinical setting of HTx.
Repurposed drugs and epitherapy in CVDs: network medicine can improve the current reductionist approach to reach the goal of personalized treatments As the discovery and development of novel drugs is a highly expensive time-consuming process, the repurposing of "old" drugs to treat CVDs is increasingly becoming an attractive goal. The repurposing of both metformin and statins has been widely evaluated in large controlled clinical trials for the prevention and treatment of CHD, PAH, HF, and complications after HTx Table 2. Metformin proved in 1994, is the first-line oral drug for treatment of T2D and obesity (www. ncbi. nlm. nih. gov/ books/ NBK40 9379). Metformin has glucose-lowering effects by acting on several molecular pathways and also represents an agonist of the SIRT1 enzyme [98,99]. Additionally, the first statin (lovastatin) received FDA approval in 1987 and now 6 statins, including simvastatin and pravastatin (semi-synthetic), as well as fluvastatin, atorvastatin, rosuvastatin, and pitavastatin (synthetic), have been introduced to the market providing first-line of oral drugs prescribed for dyslipidemias and prevention of atherosclerotic plaque development [41]. Basically, statins block the 3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase enzyme leading to lipid-lowering effects and also showing epigenetic interference (HDACi). The cardio-protective and anti-inflammatory effects of metformin and statins have also been widely demonstrated in patients affected by CVDs [98,99,101]; however, large controlled trials are needed to establish whether these beneficial effects are conferred by the glucose and lipidlowering effects, by interference with specific epigeneticsensitive pathways, such as inflammatory pathways or by both. Another example of drug repurposing includes tocilizumab (Actemra), a monoclonal anti-IL-6 antibody approved by FDA in 2010 for treatment of rheumatoid arthritis. In addition, there are a few examples of properly defined epi-drugs under clinical evaluation for treatment of CVDs. One of the most promising is apabetalone (RVX-208), inhibiting selectively the bromodomain and extra-terminal domain (BET) family of proteins binding to acetyl groups in order to normalize acetylation imbalance underlying cardiac dysfunction in T2D, CHD, PAH, and HF Table 2.

Ongoing and completed clinical trials
In contrast to cancer patients, epitherapy has not yet reached clinical practice in the management of CVDs. However, exploring the website http:// clini caltr ial. gov/, we retrieved a large number both of ongoing observational and interventional clinical trials (Phase 3 and 4) of these agents in CVDs Table 2 demonstrating the great interest in translating epigenetic findings from the bench to bedside to treat patients affected by CVDs.

Diabetes and CHD
A completed double-blinded, randomized, placebo-controlled Phase 3 trial (NCT02773927) compared the effect of metformin/agave inulin in comparison with agave inulin on adiponectin levels in patients with metabolic syndrome by grouping patients into in four clusters: group A, metformin plus agave inulin; group B, metformin plus placebo of agave inulin; group C, agave inulin plus placebo of metformin; group D, placebo of agave inulin plus placebo of metformin. To date, there are no published results from this trial. Metformin is also currently under investigation in a Phase 2, observational, prospective clinical trial (NCT01884051) enrolling 1.899 participants, in which the primary endpoints are measures of insulin resistance, urinary, and plasma oxidant stress markers, right ventricle lipid content, oxidative metabolism, and drug safety. As secondary endpoints, the quantification of lung metabolism, [18F]-fluorodeoxyglucose uptake, blood-based expression of the bone morphogenic protein receptor (BMPR) 2, right ventricle ejection fraction and volumes using magnetic resonance imaging, insulin resistance, and 6-min walking distance (6MWD) is evaluated. Additionally, the preliminary results from an ongoing Phase III trial (NCT02586155) have demonstrated possible efficacy in preventing myocardial damage in high-risk T2D-CHD patients under high-intensity statin therapy combined with RVX 208 (daily dose 100 mg capsule plus atorvastatin and rosuvastatin).

Pulmonary arterial hypertension
At the molecular level, trimetazidine can switch the metabolic cellular state from beta-oxidation toward glucose oxidation by inhibiting synthesis or carnitine-facilitated transport of fatty acids so that cardiomyocytes can obtain more energy [131]. Recently, it has been reported that trimetazidine may exert its cardio-protective role in women by affecting the DNA methylation profile of the cyclin dependent kinase inhibitor 2b (CDKN2B) gene [27,28]. Trimetazidine is being studied in a randomized controlled trial (NCT03273387) which aims to determine if 3 months of treatment (35 mg twice a day) combined with standard therapy can alter right ventricle function in PAH patients. The primary endpoint will be a change in right ventricular function, as quantified by cardiac magnetic resonance imaging (MRI). Secondary endpoints will contain modifications in cardiac fibrosis quantified by T1 cardiac MRI mapping, functional class, and plasma levels of lactate dehydrogenase (LDH). However, tocilizumab is being investigated in the "Therapeutic Open-Label Study of Tocilizumab in the Treatment of Pulmonary Arterial Hypertension study" (NCT02676947), a single-arm study aimed at investigating whether treatment with tocilizumab in a dose of 8 mg/kg monthly for 6 months may alter pulmonary vascular resistance in PAH patients [59]. The primary endpoints will be a change in pulmonary vascular resistance (PVR) and safety, with secondary endpoints including 6MWD, N-terminal pro-B-type natriuretic peptide (NT-pro-BNP), symptom burden, and quality of life. This trial is still ongoing Apabetalone is under investigation in a single-arm trial in a two-center study enrolling PAH patients combining the standard therapy with 100 mg of apabetalone twice a day for 16 weeks (NCT03655704).

Heart failure (HF) and heart transplantation (HTx)
An interesting example of drug repurposing arises from the randomized, double-blind, placebo-controlled crossover intervention DoPING-HFpEF trial (EU Clinical Trial Register: 2018-002170-52; NTR registration: NL7830), which will evaluate the possible cardio-protective effects of trimetazidine in patients affected by HFpEF [131]. A large randomized, double-blind study enrolling 6.975 HF patients with monitored EF provided evidence for which assumption of 1 g per day of long-chain omega-3 polyunsaturated fatty acids (PUFA) is associated with a small reduction (9%) in mortality and admission to the hospital for CV events in HF patients [128]. Dietary supplementation of PUFA may aid in normalizing circulating lipid levels, exerting beneficial systemic anti-inflammatory effects, preventing cardiac injury by affecting bloodbased global DNA methylation levels [98,101].

Which additional clinical benefits would the network medicine approach add?
The interactome may reveal differential pathogenic molecular drivers in each patient that are, in part, responsible for current limitations of the one-size-fits-all approach ( [15,64,76,86]). Indeed, CVD patients treated with optimized standard-of-care regimens can still show high residual morbidity and mortality risk [5]). Thus, interactome-based selection before enrolling subjects in a clinical trial may aid in selecting more homogeneous study populations in which to test a specific drug therapy. NM approaches would potentially resolve this heterogeneity by combining the interactome with biochemical and molecular assays and clinical information, potentially providing noninvasive biomarkers as well as drug targets ( [14,64,98,100,120]). A network-oriented paradigm of CVD pathogenesis would have the power of repurposing existing FDA-approved compounds by in silico prediction of targeted disease modules to speed up the discovery of putative novel personalized treatments [29]. Moreover, repurposing bioinformatic tools could offer new molecular mechanisms of existing drugs. As an example, the seed connector algorithm (SCA) has been applied to GWAS-derived CHD seed proteins and has suggested that the neuropilin-1 (NRP1) would be potentially a novel candidate disease gene [134]. Since NRP1 is the target of the anti-angiogenic agent pegaptanib, which is indicated for the treatment of neovascular age-related macular degeneration, the SCA algorithm approach suggested its possible repurposing for atherosclerotic diseases [134]. In addition, the use of network-based proximity measures would allow to quantify the relationship between CVD modules and drug targets helping to chart novel associations and discriminate if a candidate repurposed drug may be therapeutically effective or lead to unwanted side effects [55]. Similarly, another proximity score has highlighted a relevant role for dysregulation of the immune system in MS development, suggesting the repurposing of ibrutinib, a BTK inhibitor prescribed for hematological malignancies, to counteract the inflammatory state [93].

Concluding remarks network medicine's role in clinical medicine, challenges and a pathway forward to the new of precision medicine
The management of hematological malignancies has already seen a major benefit from epigenetics with nine epi-drugs currently approved by the FDA. Although clinical epigenetics is far less advanced in CVDs, large clinical trials show the promising results with respect to the effectiveness of drug repositioning and epitherapy in the treatment of MS, T2D, CHD, PAH, HF, and HTx. Therefore, in areas relevant to the cause of death, such as CVDs, there need to be more expeditious ways to repurpose drugs as we look toward the era of precision medicine.
The reasons for this gap may also be due to the different procedures for approval of clinical studies in cancer versus CVD. Several epigenome-wide association studies have revealed molecular pathways involved in the pathogenesis of cancer and CVDs that may offer robust biomarkers for precision medicine and personalized therapy. Novel technological developments as well as the application of NM may change our view of the role and analysis of epigenome deregulation in disease. The development and use of bioinformatics (and tools of artificial intelligence) hints at a deep change when dealing with human health, disease identification and handling. NM applications may improve our mechanistic understanding of tumorigenesis and the dynamics of driver and contributing (epi)mutations within the 3D structure of the cell and of tissues. The indication that a cytosine is first methylated and then hydroxymethylated suggests that the order of (some) epigenome changes should be possibly integrated mining the epigenome landscape in health and disease. Furthermore, different chromatin complexes (such as the readers and some erasers) use metabolic co-factors in their reactions, suggesting that, among the regulatory networks, the availability of those cofactors might, in turn, also regulate the prioritization of events. Indeed, the integration of genome and epigenome information together with the metabolic status of the cells might, in principle, be a necessary step forward to stratify better epi-biomarkers and epi-targets in disease.
Studying the role of spatiotemporal regulation of chromatin with integrative network approaches may represent the next step forward for personalized medicine, allowing drug-based fine-tuning of the epigenome. While a small number of computational tools already leverage epigenetic profiling data in a network context, most existing approaches are limited to promoter methylation, neglecting the influence of regulatory regions distal to promoters. Recently, the EpiRegio database established a link between cis-or trans-regulatory elements and their target genes [13]. In combination with disease-specific molecular profiling data, network enrichment [78] or disease module discovery [33] can be employed on such networks to reveal more distal and complex regulatory interactions for developing clinically relevant insights into epigenetic mechanisms in diseases in the future. However, these approaches are not yet well understood by physicians. Learning workshops on this groundbreaking field are necessary for the fields of cardiology and oncology. Interpreting the developments of epigenome deregulation to the clinic may also need a further understanding of the use of epi-drugs. For example, physicians should be aware that the reduced locus-selective specificity may lead to (epi)genome undesired effects. Novel drug discovery approaches targeting DNA mutations might be a more focused solution, as indicated for the WGCNA Blood samples isolated from 2627 subjects enrolled in FHS cohort Differences in circulating DNA methylation signatures located in regulatory regions of SLC9A1, SLC1A5, TNRC6C genes may be useful biomarkers to predict incident CHD Westerman et al. [137] EZH2 inhibitors discussed in the previous sections. In addition, methods such as "the proteolysis targeting chimera (PROTAC)" might produce specific drugs, perhaps able to (re)modulate the function of non-enzymatic chromatin complexes.
A future challenge will, thus, be to integrate such notions in a network context together with other modalities such as transcriptomics, proteomics, and miRNA expression as well as with clinical information. We expect that integrative network analysis will reveal regulatory patterns that can be exploited for diagnosis, prognosis, and treatment selection. We need to encourage the scientific integration of basic scientists and physicians in the field of clinical epigenetics. A relevant role is also played by innovative technology assets applied to the clinical setting and management of patient. None of NM approaches has yet reached clinical application, but some have been validated with excellent results in ex-vivo patients Table 5. Therefore, the final step will be to evaluate the potential of novel NM platforms in large clinical trials to test their reliability value in diagnosing, prognosticate, and treating cancer and CVDs, as two of the major causes of death worldwide.