Evolving systems biology approaches to understanding non‐coding RNAs in pulmonary hypertension

Abstract

Our appreciation of the roles of non‐coding RNAs, in particular microRNAs, in the manifestation of pulmonary hypertension (PH) has advanced considerably over the past decade. Comprised of small nucleotide sequences, microRNAs have demonstrated critical and broad regulatory roles in the pathogenesis of PH via the direct binding to messenger RNA transcripts for degradation or inhibition of translation, thereby exerting a profound influence on cellular activity. Yet, as inherently pleiotropic molecules, microRNAs have been difficult to study using traditional, reductionist approaches alone. With the advent of high‐throughput –omics technologies and more advanced computational modelling, the study of microRNAs and their multi‐faceted and complex functions in human disease serves as a fertile platform for the application of systems biology methodologies in combination with traditional experimental techniques. Here, we offer our viewpoint of past successes of systems biology in elucidating the otherwise hidden actions of microRNAs in PH, as well as areas for future development to integrate these strategies into the discovery of RNA pathobiology in this disease. We contend that such successful applications of systems biology in elucidating the functional architecture of microRNA regulation will further reveal the molecular mechanisms of disease, while simultaneously revealing potential diagnostic and therapeutic strategies in disease amelioration.

  

Introduction

Throughout the greater part of the last two centuries, the successful application of reductionist approaches to scientific discovery resulted in unprecedented advancements in the understanding and treatment of human disease. However, at the same time, reductionist studies have made it clear that molecular phenomena in disease states are unlikely to be explained by the aberration of a singular component. The more recent advent of high‐throughput, data‐gathering technologies has revolutionized our ability to interrogate multifaceted aspects of a cell and its functional state in incredible detail. Beyond characterizing the genomic landscape via the Human Genome Project, technological innovation and advancements in –omics‐based platforms, i.e. transcriptomic, proteomic and metabolomic approaches, have made multi‐dimensional, molecular profiling of biological samples more accessible, while simultaneously unleashing a deluge of data. Inherent to these datasets has been a growing appreciation of the complex and pleiotropic gene regulatory actions of non‐coding RNAs. Tens of thousands of RNAs have been discovered or predicted as encoded in the human genome, and their biological complexity presents a formidable challenge in the understanding of the systems‐based landscape of human physiology and disease.

In that context, pulmonary hypertension (PH) is a devastating vascular disorder characterized by complex pulmonary vascular remodelling that results in increased pulmonary vascular resistance and failure of the right ventricle (Simonneau et al. 2013). Moreover, PH represents a particularly daunting human disease, where non‐coding RNAs are known to control pathophenotypic manifestations. More specifically, microRNAs – a class of small non‐coding RNAs – have been the most avidly studied subset of non‐coding RNAs in this disease (Sessa & Hata, 2013; Boucherat et al. 2015; Huston & Ryan, 2016; Negi & Chan, 2017). MicroRNAs are evolutionarily conserved and negatively regulate gene expression at the post‐transcriptional level through targeting of the messenger RNA transcript (Nagano & Fraser, 2011). The study of microRNAs in PH can be challenging when using reductionist approaches, stemming from the facts that (1) these molecules exhibit pleiotropy in regulating multiple gene targets, and (2) there is a technical obstacle in identifying their direct functions in PH initiation and progression due to an inability to study diseased tissue until either patient death or lung transplantation. Thus, clinicians and scientists have been forced to analyse PH pathogenesis from a single snapshot in time via traditional experimental strategies that do not adequately account for the spatio‐temporal complexity of microRNA pathobiology. The recent availability of –omics‐based technologies has offered an opportunity to unravel at least some of the complexity of many human diseases, including PH. However, –omics profiling is not synonymous with systems biology, and when offered in isolation, –omics work can create more confusion than insight to disease pathogenesis. Inherent to interpreting –omics‐based datasets is the development and proper application of computational and mathematical tools in order to form testable predictions and to gain insight into the functional behaviour of a disease network. As such, we believe that these two inherently linked aspects that define systems biology – the acquisition of molecular profiles and computational simulation – when coupled with experimental validation will power and accelerate our understanding of the most influential aspects of non‐coding RNA pathobiology in the postgenomic era.

Understanding the organizational framework of a biological network

The application of in silico computational modelling of non‐coding RNAs in pulmonary vascular disease is an emerging tool utilized for the discovery of novel biological regulators in disease manifestation. Computational methodology can vary among modelling systems, ranging from static gene network mapping to more complicated neural network‐based machine learning. Yet, most methodologies will rely upon at least a fundamental concept of gene or pathway networks that are interconnected physically and/or functionally, across anatomic space and time. The acquisition and integration of –omics datasets with this conceptual framework can result in a series of highly interconnected and interdependent networks, in essence, a ‘network of networks’ through which the functional behaviour of a cell, tissue, organ system, individual or population is governed. It is this integration of in silico simulation and experimental validation of comprehensive molecular datasets that remains a formidable challenge in the successful application of systems biology in modelling cellular behaviour.

Fundamentally, each component of a molecular or cellular network can be understood as a series of nodes that are connected by links, serving as an abstract representation of an interaction; it is the sum of these nodes and links that establishes a network. The operation of a network by a power law is defined as a scale‐free network. Specifically, most biological and human disease networks are architecturally governed by a power law (i.e. organized in a scale‐free fashion), that is to say that the distribution of the network is highly non‐uniform: most of the nodes possess few links while a limited number of nodes contain many links, thereby acting as centralized hubs that hold the network together (Barabasi & Albert, 1999). This organizing principle is in contrast to a random network, where most nodes would have the same number of interactions. The discovery that most cellular systems are governed by a scale‐free architecture has substantial implications in our understanding of how the aberration of specific biological components can cause the disintegration or compensation of cellular behaviour, thereby producing a pathological or a subclinical phenotype, respectively (Jeong et al. 2000; Wagner & Fell, 2001; Barabasi & Oltvai, 2004). For example, the loss of a highly interconnected node (i.e. a hub) could result in the disintegration of cellular behaviour to manifest as a diseased state, whereas the loss of a peripherally and poorly connected node may produce a negligible phenotype.

Lastly, cellular function within the network is further organized by modules or clusters, where groups of functionally linked nodes act together for a common purpose (Hartwell et al. 1999). The identification of functional modules within a network is vitally important, as studies have demonstrated that a hierarchical modularity often operates within cellular networks (Ravasz et al. 2002; Ravasz & Barabasi, 2003). Thus, mapping the complex relationship among nodes, their organization into modules, and the hierarchical stratification of such nodes is crucial to understanding how cellular behaviour operates within a specific biological or clinical context.

Applications of systems biology to microRNAs in pulmonary hypertension

In the past few years, the principles of systems and network biology have just begun to be utilized to study the regulatory roles of microRNAs in PH. One of the first attempts to characterize and assemble –omics‐based microRNA datasets in PH examined an array analysis of whole lung tissue from chronic hypoxia‐induced and monocrotaline‐induced PH rodent models at several time points in disease pathogenesis. Of the 350 microRNAs they examined, only five were similarly dysregulated in both animal models, indicating a potential regulatory role in disease manifestation. However, they noted discrepancies between the two groups, perhaps indicating disease‐specific pathways for each model (Caruso et al. 2010). At the time, these array datasets were not subjected to higher order computational modelling, thereby limiting the ability to discern the complexity and, perhaps, interconnected nature of those five dysregulated microRNAs within the context of other RNA behaviour in PH.

Since then, our group has made attempts to discern the systems‐wide influence of microRNAs in PH by applying network theory to available –omics and curated datasets (Parikh et al. 2012; Bertero et al. 2014). To do so, we constructed in silico a PH disease network of genes and interactions based on curated seed genes with known relevance in the pathogenesis of PH from the scientific literature (Parikh et al. 2012). Interactions were derived from a master list of functional molecular associations containing all known human gene and molecular interactions (Barabasi et al. 2011), giving rise to our expanded PH disease network. The expanded PH disease network represented the functional interconnectivity among target genes. We ranked microRNAs using a spanning score based on size and intercluster spread of their target pools to identify miR‐130/301 as an important regulatory factor in PH (Bertero et al. 2014). Genes in the miR‐130/301 target pool were ranked by ‘hubness’ to determine top‐ranked genes involved in proliferative pathways within the lung microenvironment. We experimentally validated these predictions in vitro and in vivo to demonstrate that the miR‐130/301 cluster is upregulated within the pulmonary vasculature of PH patients, and that the miR‐130/301 cluster modulates proliferation with pulmonary arterial endothelial cells and smooth muscle cells. In addition, we defined a role for the miR‐130/301 cluster in regulating vasomotor tone (Bertero et al. 2014). Furthermore, guided by an observed overlap between a fibrosis network and our PH network, we found a mechanosensitive axis operating through the miR‐130/301 cluster to promote extracellular matrix remodelling and vascular stiffness in PH (Bertero et al. 2016). Notably, even beyond PH, our in silico modelling predicted that the miR‐130/301 cluster carries master regulatory actions in extracellular matrix biology in a variety of physiological and pathophysiological states through a shared signature of fibrosis‐relevant genes (Bertero et al. 2015 a). These findings signified an evolutionary conservation of the regulatory influence of the miR‐130/301 cluster across clinical contexts, connecting seemingly disparate diseases via a shared reliance on the same microRNA family. Thus, our initial successes combining computational –omics pursuits with experimental validation demonstrated that integration of systems biology with PH can lead to robust and often surprising insights into the root causes and effects of this disease.

Emerging challenges in developing further systems biology: insights of microRNAs in PH

Procuring diseased tissue and acquiring funding in the study of rare diseases

For future work in this field, there remain a plethora of challenges that must be addressed in order to improve the validity of –omics‐based, data‐driven predictions of microRNA function and advance the ease with which these approaches can be utilized by the contemporary scientist (Fig. 1). A major obstacle in the widespread application of systems biology approaches for discerning microRNA biology is the initial acquisition of high‐throughput, precise datasets, particularly in PH. As noted above, relevant pulmonary samples in PH can only be obtained for molecular profiling either at end‐stage disease upon patient death or after lung transplantation. This inherently limits our ability to obtain a molecular disease signature during critical time points, such as the disease initiation and its development over a temporal framework. The NIH Pulmonary Vascular Disease Phenomics (PVDOMICS) project has sought to address these barriers to comprehensive –omics profiling in PH – an initiative intending to define deep clinical phenotypes of 1500 participants coupled with various –omics‐based technologies in order to enhance the classification of PH via shared molecular features (Brittain & Chan, 2016; Hemnes et al. 2017; Newman et al. 2017). Thus, along with the already ongoing Pulmonary Hypertension Breakthrough Initiative (PHBI), the National Biologic Sample and Data Repository for PAH, and the UK National Cohort Study of Idiopathic and Heritable Pulmonary Arterial Hypertension, there exist substantial and ongoing agendas to address lapses in molecular profiling datasets in PAH (Tuder, 2009; Newman et al. 2017; Rhodes et al. 2017).

Figure 1. Application of systems biology in the elucidation of RNA biology in pulmonary hypertension.

The advent of –omics‐based technologies has offered a new‐found opportunity to elucidate the complexity of human diseases such as PH. Non‐coding RNAs, such as microRNAs, have been identified as central biological mediators in the root cause of this disease. Accordingly, the successful application of systems biology and network theory in the interpretation of these coding and non‐coding RNA datasets will strive to produce insight into the functional behaviour of non‐coding RNAs across networks of genes, tissues and diseases related to the PH pathophenotype(s). Computational simulation and subsequent experimental validation will provide the substrate from which a ‘microRNA‐specific, precision medicine’ approach may begin to be conceptualized, if specific challenges of implementation can be overcome.

Despite these efforts, the priority of discerning a direct molecular profile of non‐coding RNAs reflective of specific endophenotypes of PH may be unlikely in the near future. However, other avenues of attaining relevant microRNA disease signatures may be possible through single cell RNA sequencing of peripheral cells, namely circulating cells of the haematopoietic system. Inflammatory and immune dysregulatory components are increasingly being appreciated as driving mechanisms in the pathogenesis of various PH subtypes (Rabinovitch et al. 2014). As such, the acquisition of microRNA signatures of circulating inflammatory cells may provide insight into a hypothesized role for proinflammatory disease modules that underlie specific subtypes of PH, such as those associated with viral infection (i.e. HIV) or others stemming from autoimmune and connective tissue disorders (Farber & Loscalzo, 2004; Stenmark et al. 2005; Rabinovitch et al. 2014; D'Alessandro et al. 2018). In addition, single cell RNA sequencing of blood outgrowth endothelial cells (BOECs) – surrogates of the pulmonary endothelium – may also be an avenue through which unique microRNA disease signatures may be obtained. Particularly given the long‐standing controversy regarding the spatio‐temporal evolution of distinct endothelial cell populations in the diseased pulmonary vasculature, the interrogation of single cell niches within a given endothelial cell population could provide fundamental insight into the true kinetics of endothelial function during pathogenic initiation and progression (Michelakis, 2006). Already, a microarray analysis performed on BOECs isolated from control, heritable pulmonary arterial hypertension (PAH), and idiopathic PAH patients demonstrated differential expression of microRNAs, with four conserved between the two diseased states (Caruso et al. 2017). Of these four conserved species, microRNA‐15a has been independently predicted using network approaches to contribute to PH pathogenesis (Bertero et al. 2014). Although promising, it must be noted that the isolation and culturing of BOECs ex vivo may fundamentally alter the microRNA landscape in ways that are not indicative of PH pathophysiology. In the interim, currently available and extensive –omics datasets from cancer biology – such as The Cancer Genome Atlas – may be utilized, given mounting evidence demonstrating not only molecular parallels between cancer and PH but also that lung cancer predisposes to PH (Guignabert et al. 2013; Tomczak et al. 2015; Vencken et al. 2015; Pullamsetti et al. 2017 a, b ). As such, until larger datasets for PH have been accumulated, –omics databases in cancer represent an untapped resource to be mined for additional parallels with PH, where non‐coding RNA biology could serve as a central investigative point.

Developing innovative analytical methodologies

Beyond data acquisition, a second obstacle includes the computational and statistical processing of –omics data volume to obtain reliable hypotheses of biological mechanisms. Specifically, we continue to search for analytical pipelines that can prioritize relevant signals while still accounting for the propensity of a low signal to noise ratio. Of course, proper pre‐processing of datasets (i.e. quality control checks of raw reads, subsequent trimming and adaptor removal, and suitable read alignment) can serve as a mechanism to filter consistent and predictable sources of variability. But, there will always be a residual and inherent noise, given the sheer volume of –omics data points. As such, to minimize noise and the false discovery rate, iterative machine learning algorithms increasingly have been employed, given the ability to integrate and later interpret nuanced specificities observed in high‐throughput data due to experimental error and biological stochasticity (Ching et al. 2018). One such approach is to filter by novel attributes, such as clinical phenotypes, that are correlated to datasets for classification purposes (i.e. unsupervised learning). These results can then inform a new‐found schema for predictive and outcome‐based capacities (i.e. supervised learning). For example, machine learning has recently been employed to detect nuanced patterns in clinical data from patients with heart failure with preserved ejection fraction (HFpEF) to devise a better clinical classification schema of this heterogeneous disease state (Shah et al. 2015). The same principle can be applied to a disparate disease like PH, where multivariate datasets (i.e. clinical parameters and perhaps plasma microRNA biomarker profiles) may be incorporated and interpreted through a statistical learning algorithm to predict patient outcomes and response to treatments. Ultimately, however, the reliability of any initial systems‐based predictions relies on experimental validation, so that estimations of noise can be traced in their influence of the original hypothesis generated (De Keersmaecker et al. 2006).

After initial –omics data processing, ensuing computational modelling also serves as a significant rate‐limiting step. Currently, our in silico modelling of microRNAs in PH is based upon undirected networks, where an edge between two nodes a and b is unable to discern an exertion of functional control of one node over the other (i.e. the biological flow of information). Directed graphs are more suitable in the depiction of biological schemas since they can indicate the flow of information throughout a network, such as a signal transduction pathway; yet, the addition of directionality still fails to properly model critically important variables in cellular behaviour, namely the weight of an interaction (i.e. the confidence of interaction and its degree of influence) and the temporal framework in which the network and its nodes operate. Accordingly, if we were able to incorporate these variables into our computational modelling of a given microRNA perturbation, then we could be able to begin simulating the comprehensive effects of a variety of perturbations of a given microRNA on its interconnected gene network. In addition, since a given microRNA is capable of shifting its target gene pool based on microRNA:target gene stoichiometry and thus function without any change in its expression, advancing our models beyond simple differential expression analyses is crucial in order to not overlook microRNA functions embedded within large datasets (Bushati & Cohen, 2007). Along those lines, future models should consider higher complexity interactions, such as absolute concentrations coupled with relative stoichiometry as well as the presence of competing microRNAs, messenger RNAs or other non‐coding RNA species (i.e. long non‐coding (lnc)RNAs, etc.) and their relative affinities to a given transcript (Militello et al. 2017).

Notably, however, since only end‐stage diseased tissue is typically available for analysis, we are fundamentally limited in making such computational predictions currently. Potentially, these challenges could be addressed through the statistical mapping of differential dependency networks (DDNs), rather than differential expression networks, that can better attempt to delineate the nuances of gene regulation among various conditions (Zhang et al. 2009; Jung & Kim, 2014). In a DDN, interactions among nodes are assigned a weighted value, based on the degree and strength of a biological interaction, and it is the sum of these interactions and their probabilities that determine the predictive ability for a given DDN. Consequently, a high‐quality DDN, in theory, may have the predictive capacity to infer gene circuitry and thus statistically model any given biological state, such as initiating events or disease progression. Such DDNs could be generated from end‐stage disease tissue alone but would require a high volume of tissue specimens to attain the statistical power for accurately mapping high‐quality networks.

Integrating large and multivariate datasets in the prediction and tailoring of clinical treatment

Beyond the technological challenges of data acquisition and processing, another challenge is the incorporation of multiple layers of datasets relevant to microRNA biology across clinical spectra. Namely, machine learning predictive capacity is being used for this purpose through enhanced utilization of the electronic medical records (EMRs) and the subsequent integration across genomic data in other diseases. The linkage of EMRs to biological data (i.e. –omics‐based datasets) may provide insight into disease pathogenesis, as exemplified by the Electronic Medical Records and Genomic (eMERGE) network (Gottesman et al. 2013). The eMERGE network focuses specifically on developing methods to identify a clinical phenotype within the EMR and subsequently to integrate clinical data with genomic sequencing. The combination of these processes allows for the use of computational modelling to predict associations that can then be experimentally validated from DNA stored within a biorepository that corresponds to patient EMRs. The use of the eMERGE network makes it possible to assemble a cohort more quickly than traditional prospective cohort studies, especially when considering rare diseases like PH (Brittain & Chan, 2016). Incorporating clinical parameters with invasive haemodynamic parameters in patients with biological datasets already revealed genetic explanations for specific therapeutic responses. For example, pharmacogenomic studies have demonstrated a role for single nucleotide polymorphisms in the clinical efficacy of endothelin‐1 receptor antagonist therapy (Benza et al. 2015). In addition, certain serum biomarkers, such as the potent angiostatic factor endostatin, have not only been correlated with disease severity and patient outcomes in PH but have also been linked to heritable genetic variants (Damico et al. 2015). Moreover, metabolomic profiles of serum from a large cohort of idiopathic and heritable PH patients were found to be correlative of disease prognosis, indicating their importance in the clinical characterization of this disease (Rhodes et al. 2017). More basic studies regarding the metabolome in PH have been conducted, noting widespread metabolic reprogramming in endothelial cells with the causative bone morphogenetic protein receptor type 2 (BMPR2) mutation and in lung tissue from patients with PH (Fessel et al. 2012; Zhao et al. 2014). Transcriptomic analyses of the endothelium, utilizing cells isolated from patients with idiopathic PH and induced pluripotent stem cells (iPSCs) from carriers of the BMPR2 mutation, have been successfully employed in further understanding mechanisms of PH pathogenesis as well (Rhodes et al. 2015; Gu et al. 2017). However, how microRNA‐specific data can be overlaid with clinical information has yet to be explored. One possibility could include a comprehensive mining of existing genomic data for polymorphisms or epigenetic alterations in either microRNA genes or targeting sequences of target gene messenger RNA transcripts across PH patients that may correlate with a specific clinical classification, response to drug therapy, or survival.

Our previous work implicating the miR‐130/301 family in modulating fibrosis‐relevant gene programmes across various diseases suggests that there may be utility in exploring a more comprehensive restructuring of how we classify and treat diseases that share ‘network‐related’ dependences with PH via certain microRNAs (Bertero et al. 2015 b). Particularly, if specific microRNA activities exist as fundamental and evolutionarily conserved modulators of shared pathobiological pathways, we would then be able to pursue a more rational strategy for targeting those microRNAs for diagnostic and therapeutic benefits across multiple diseases. Not only would this change the fundamental way in which we understand disease but it may also sustain greater impetus for federal, industrial and academic investment in the development of microRNA‐based applications for multiple – rather than just single – disease contexts (Menche et al. 2015). As such, greater traction may be gained in the development of tissue‐specific and controlled‐release, microRNA‐specific therapies, which are major challenges in RNA‐based therapeutics in both PH and other chronic diseases. In addition, the diagnostic utility of circulating microRNAs in plasma may be more comprehensively explored in PH. Several circulating microRNAs have already been identified as correlative markers in disease severity in PH patients, suggestive of a potential and promising applicability in the identification of novel biomarkers (Wei et al. 2013). Currently, circulating microRNAs are presumed to be released by diseased tissue resulting from either tissue damage or a specifically regulated process of secretion, leading to uptake by neighbouring recipient cells (Baggish et al. 2014). These processes perhaps act as a form of paracrine signalling within the pulmonary vascular microenvironment or even acting with endocrine‐like functions to distant anatomic sites (Wei et al. 2013; Boucherat et al. 2015; Negi & Chan, 2017). Since plasma microRNAs have not been fully sequenced nor do we know their trajectories outside of the cell, they present an unexplored level of systems‐wide regulation across multiple cell types and organ systems in anatomic space.

Fostering fluency in computational and experimental biology

Finally, we contend that the utility and necessity of taking a systems biology perspective will be an essential skill for the contemporary scientist in PH and biomedical research in general. Consequently, there must exist a commitment by interdisciplinary training programmes to provide the next generation of scientists the preparation necessary to glide seamlessly among the computational, biological and medical sciences. In response to an era of –omics‐driven research, graduate programmes internationally have established training programmes in the fields of systems and computational biology. For the already established investigator, resources and education initiatives are available to gain a foundation in understanding, analysing and properly depicting –omics datasets (Fox & Ouellette, 2013). Despite these developments, there exist inherent challenges to appropriately train investigators in the realms of computational, experimental and clinical research. Nonetheless, we argue that many of these obstacles can be overcome in the future by motivating and providing incentives for scientists in these divergent disciplines to interact and collaborate.

Conclusions

The use of molecular profiling technologies has the potential to transform our understanding of non‐coding RNAs and their control of historically neglected diseases such as PH (Fig. 1). Yet, our ability to interpret such large datasets remains a formidable barrier in the application of such technologies. Additionally, we are still at the beginning of understanding how to model RNA and gene networks to explain complex pathobiological processes. While numerous challenges remain in this field, previous successes sustain our enthusiasm in developing these technologies further. Appropriate and deliberate incentives to integrate computational scientists, experimental biologists and practising clinicians into fully collaborative teams, as well as to train the next generation of multi‐disciplinary researchers, will be essential in revealing the next major insights into RNA biology and PH.

Additional information

Competing interests

S.Y.C. has served as a consultant for Actelion (significant), Gilead, Pfizer and Vivus (modest). The authors declare no other conflicts of interest.

Author contributions

L. D. H. and S. Y. C. both conceived and contributed to the writing of this manuscript. Both authors have approved the final manuscript version and have agreed to be accountable for all aspects of the work. Both persons designated as authors qualify for authorship, and all those who qualify for authorship are listed.

Funding

This work was supported by NIH grants R01 HL124021, HL 122596, HL 138437 and UH2 TR002073 as well as the American Heart Association Established Investigator Award 18EIA33900027 (S.Y.C.).

Biographies

Lloyd Harvey is a MD and PhD candidate at the University of Pittsburgh School of Medicine.

Stephen Chan is a physician‐scientist, Associate Professor of Medicine and Director of the Center for Pulmonary Vascular Biology and Medicine at the University of Pittsburgh School of Medicine and University of Pittsburgh Medical Center. Dr Chan devotes the majority of his time to running a basic science and translational research laboratory studying the molecular mechanisms of pulmonary vascular disease and pulmonary hypertension (PH). To capitalize on the emerging discipline of ‘network medicine’, the Chan laboratory utilizes a combination of network‐based bioinformatics with unique experimental reagents derived from genetically altered rodents and human subjects to accelerate systems‐wide discovery in PH. In doing so, Dr Chan's published work has contributed to the systems‐level understanding of functions of microRNAs as root causes of PH.