Non-Coding RNA Networks in Pulmonary Hypertension

Non-coding RNAs (ncRNAs) are involved in various cellular processes. There are several ncRNA classes, including microRNAs (miRNAs), long non-coding RNAs (lncRNAs), and circular RNAs (circRNAs). The detailed roles of these molecules in pulmonary hypertension (PH) remain unclear. We systematically collected and reviewed reports describing the functions of ncRNAs (miRNAs, lncRNAs, and circRNAs) in PH through database retrieval and manual literature reading. The characteristics of identified articles, especially the experimental methods, were carefully reviewed. Furthermore, regulatory networks were constructed using ncRNAs and their interacting RNAs or genes. These data were extracted from studies on pulmonary arterial smooth muscle cells, pulmonary artery endothelial cells, and pulmonary artery fibroblasts. We included 14 lncRNAs, 1 circRNA, 74 miRNAs, and 110 mRNAs in the constructed networks. Using these networks, herein, we describe the current knowledge on the role of ncRNAs in PH. Moreover, these networks actively provide an improved understanding of the roles of ncRNAs in PH. The results of this study are crucial for the clinical application of ncRNAs.


INTRODUCTION
Pulmonary hypertension (PH) is a serious disease characterized by progressively increased pulmonary vascular resistance and pulmonary artery pressure; the diagnostic criterion is mean pulmonary artery pressure ≥25 mmHg (Galiè et al., 2016;Weber et al., 2018). The increased pulmonary artery pressure in PH results from changes in the structure and function of the vessel wall, which is induced by abnormal pulmonary cell proliferation, apoptosis, and migration (Bourgeois et al., 2018a). Patients with PH may experience dyspnea, fatigue, syncope, chest pain, and/or edema of the legs and ankles. The causes of PH can be broadly classified as primary and secondary causes. To date, ion channels, vasoactive substances, immune factors, and genetic factors are known to be involved in the pathogenesis of PH (Chelladurai et al., 2016;Veith et al., 2016;Bourgeois et al., 2018b).
Recently, many non-coding RNAs (ncRNAs) have been recognized as important regulators in the development of PH. Most human genes (>95%) do not produce proteins but ncRNA molecules. Among them, microRNAs (miRNAs), long non-coding RNAs (lncRNAs), and circular RNAs (circRNAs) are the most widely studied. MiRNAs are small ncRNAs containing 21-22 nucleotides, which post-transcriptionally regulate gene expression (Wakiyama and Yokoyama. 2014). LncRNAs, which have more than 200 nucleotides, are transcribed from intergenic or intragenic regions. They can bind to proteins, RNA, or DNA to execute regulatory roles (Botti et al., 2017). CircRNAs are a novel class of ncRNAs with a closed loop structure, making them highly stable and capable of interacting with proteins or RNA (Di et al., 2019). NcRNAs have been identified to regulate multiple steps of gene expression. However, because of the large quantity and diverse mechanisms, it is difficult to comprehensively understand the roles of ncRNAs.
NcRNA-based therapeutics have emerged for several diseases, including PH. An effective ncRNA-based strategy demands a thorough understanding of the diverse and context-dependent regulatory relationships of ncRNAs. The regulation of gene expression by ncRNAs is frequently cell specific, suggesting that not only expression level, but also activity or bioavailability contribute to the biofunction of ncRNAs (Correia de Sousa et al., 2019). Thus, in this article, we reviewed the published literature to search for functional miRNAs, lncRNAs, and circRNAs in PH. Next, we constructed networks of validated ncRNAs and their interacting RNAs or genes to investigate the role of ncRNAs in PH.

Criteria for Study Selection
A literature search was performed in PubMed with the query listed in Table 1; we identified 602 articles. In addition, we also reviewed other public databases, including the Human microRNA Disease Database v3.2, miRWalk 2.0, and LncRNADisease v2.0, to identify validated functional ncRNAs in PH. Studies were selected when the following criteria were met: 1) the study reported pathogenic roles of miRNAs, lncRNAs, and/ or circRNAs in PH; 2) mechanistic studies were performed in pulmonary arterial smooth muscle cells (PASMCs), pulmonary artery endothelial cells (PAECs), and/or pulmonary artery fibroblasts (PAFs); and 3) the relationships between ncRNAs and their interacting RNAs or genes were experimentally identified via luciferase reporter assay, western blot, and/or qPCR. Using these criteria returned 140 qualified articles ( Figure 1).

General Characteristics of Qualified Articles
When sorted by publication date, we found that the number of eligible articles continuously increased year by year ( Figure 2A). The impact factors (IF) of the articles ranged from 0 to 36.13; articles with 3 ≤ IF < 5 accounted for the highest proportion ( Figure 2B). Of the 140 qualified articles, 32.14% were studies using human tissues or cells. In studies using experimental animals, rats were the most commonly used, accounting for 26.43% of the total studies ( Figure 2C). Moreover, when classified by cell type, 78.42, 15.83, 1.44, and 4.32% of studies were performed in PASMCs, PAECs, PAFs, and both PASMCs and PAECs, respectively ( Figure 2D).

NON-CODING RNA NETWORKS FOR PULMONARY HYPERTENSION 3.1 Construction of Non-coding RNA Regulatory Networks
Regulatory networks were constructed using ncRNAs and their interacting RNAs or genes in PASMCs, PAECs, and PAFs. Given ncRNA conservation among species, only human ncRNAs or ncRNAs that were conserved between human and experimental animals were included. If there were contradictory results, the results from higher-impact articles were selected. In addition, some crucial regulatory relationships between protein-coding genes and validated transcription factor-miRNA interactions from TransmiR v2.0 were also described in the networks to present an in-depth explanation on the roles of ncRNAs in PH. The nodes represented interacting molecules, and the edges represented the regulatory connections. Each edge indicated a publication supporting the connection. Square and

602
FIGURE 1 | Steps of the data analysis used to build the ncRNA networks. NcRNAs: non-coding RNAs. circular nodes represented ncRNAs and coding RNAs or genes, respectively. Node color was based on the type of molecule (lncRNAs and circRNAs are orange, miRNAs are blue, and coding RNAs or genes are empty). Node sizes represented their degrees (number of edges that directly link to the node). Edges represented the regulatory connections: red edges depicted links indicating repressive action (semicircular arrow heads), and black edges indicated activation (traditional arrow heads). The nodes in this network were involved in cell proliferation, apoptosis, migration, metabolism, endothelial-mesenchymal transition, and extracellular matrix remodeling. The steps used in our approach are shown in Figure 1.

General Characteristics of the Constructed Networks
In total, 140 articles describing 14 lncRNAs, 1 circRNA, 74 miRNAs, and 110 mRNAs, were included in our networks.

Functional Enrichment Analysis
We performed gene ontology (GO) biological process term analyses and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses using the database for Annotation, Visualization, and Integration Discovery (DAVID). The calculation process was dependent on a hypergeometric test, using a statistical significance threshold of p < 0.05 with a false discovery rate (FDR) correction. These analyses provided a general overview of the biological roles of the included ncRNAs. In addition, we performed cell type-specific functional enrichment analyses. However, owing to the lack of sufficient molecules, the enrichment analysis in PAFs could not be conducted. GO analysis and KEGG pathway enrichment in PASMCs and PAECs revealed several PH-associated terms, such as positive regulation of smooth muscle cell proliferation, positive regulation of endothelial cell proliferation, HIF-1 signaling pathway, and MAPK signaling pathway. The top 10 enriched GO biological process terms and KEGG pathways are shown in Figure 6.
FIGURE 3 | PH-associated network of ncRNAs and their interacting RNAs or genes in PASMCs. The square and circular nodes represent ncRNAs and coding RNAs or genes, respectively. Node color is based on the type of molecule (lncRNAs and circRNAs are orange, miRNAs are blue, and coding RNAs or genes are empty). Node sizes represent the degrees (the number of edges that directly link to the node). Edges represent the regulatory connections, and each edge indicates a publication. When multiple publications describe one interaction, multiple edges connect the same two nodes. Red edges depict links indicating repressive action (semicircular arrow heads), and black edges are those indicating activation (with traditional arrow heads). The nodes in this network are involved in cell proliferation, apoptosis, migration, and metabolism. PH: pulmonary hypertension; PASMCs: pulmonary artery smooth muscle cells; lncRNAs: long non-coding RNAs; circRNA: circular RNA; miRNA: microRNA.

Key Non-coding RNA Subnetworks
We built three networks according to the cell types. Here, we discuss several important subnetworks, along with their components and interactions, to improve understanding of the roles of ncRNAs in PH. Subnetworks with more than five nodes were regarded as key subnetworks.
3.4.1 The Hsa_circ_0016070/miR-942-5p/CCND1 Subnetwork CircRNAs are associated with various cardiovascular diseases. Hsa_circ_0016070 was the only circRNA included in our networks. This circRNA is located at chr1: 203595914-203702528, strand: +, promotes cell proliferation by mediating cell cycle progression, and is increased in PH patients . CCND1 is an important regulator of the cell cycle. It interacts with cyclin-dependent kinase 4 (CDK4) to form the cyclin D1-CKD4 complex, which then inactivates retinoblastoma (Rb) protein and induces G0 progression to S phase (Matsushime et al., 1991). The subnetwork showed that hsa_circ_0016070 overexpression induced CCND1 expression by buffering miR-942-5p . In addition, according to our network, CCND1 could induce the expression of miR-17-5p, miR-19a-3p, and miR-20a-5p, subsequently regulating the biological activities of PASMCs ( Figure 7A). Given the considerable number of identified circRNAs, there should be other PH-related circRNAs. A microarray expression profile in thromboembolic pulmonary hypertension patients indicated that hsa_circ_0002062 and hsa_circ_0022342 might be the key circRNAs for the development of chronic thromboembolic pulmonary hypertension (Miao et al., 2017). However, this finding has not been verified by more reliable experimental methods.

The CASC2/UCA1/miR-222-3p Subnetwork
LncRNA CASC2 is downregulated in hypoxia-induced PASMCs. As a ceRNA of miR-222-3p, CASC2 reduces the expression of ING5, which is a target gene of miR-222-3p, ultimately promoting PASMC proliferation and migration (Han et al., 2020). P27 and TIMP3 are two additional target genes of miR-222-3p . P27, a member of the Cip/Kip family of cyclin-dependent kinase inhibitors, negatively regulates cell proliferation (Toyoshima and Hunter, 1994). Meanwhile, TIMP3 is a member of the TIMP family, which regulates cell proliferation, apoptosis, and migration via both MMP-dependent or MMP-independent pathways . The present subnetwork links CACS2 to P27 and TIMP3 via miR-222-3p, further elaborating the mechanisms of PH ( Figure 7C).
UCA1 is the other lncRNA in this subnetwork and is highly expressed in hypoxia-induced PASMCs. Studies indicate that UCA1 does not interact with miR-222-3p, but directly inhibits ING5 by competing with ING5 mRNA for hnRNP I, which binds to ING5 mRNA and enhances its translation. Thus, UCA1 overexpression results in the downregulation of ING5 mRNA expression (Zhu T.-T. et al., 2019). The same regulatory pattern has been found between UCA1 and P27 in breast tumor studies (Huang et al., 2014). This interaction may also work in PH and partly contributes to P27 downregulation ( Figure 7C).

The MALAT1/miR-124-3p Subnetwork
LncRNA MALAT1, located at 11q13, is an 8.5-kb molecule that was identified by Ji et al. in a cancer study (Ji et al., 2003). Emerging evidence indicates that MALAT1 plays important roles in various diseases, including PH. Wang et al. reported that MALAT1 is highly expressed in pulmonary artery tissues and PASMCs from patients with PH. MALAT1 controls PASMC proliferation and migration by binding to miR-124-3p, which directly targets KLF5 (Wang D et al., 2019). Kang et al. showed that miR-124-3p also targets three regulators of the NFAT pathway, including NFATc1, CAMTA1, and PTBP1 (Kang B.-Y et al., 2013). The downregulation of miR-124-3p induces PASMC proliferation and reverses the differentiated PASMC phenotype by activating the NFAT pathway. In addition to its role in PASMCs, miR-124-3p also regulates the biological behaviors of PAH endothelial cells (PAH ECs) and PAFs. Studies have confirmed the role of the miR-124-3p/PTBP1 axis in PAH ECs and PAFs (Caruso et al., 2017;Wang et al., 2014;. Downregulating miR-124-3p activates PTBP1 expression, which promotes aerobic glycolysis by increasing the PKM2/PKM1 ratio, subsequently inducing PAH EC and PAF proliferation (Anastasiou et al., 2012). Li et al. reported another target of miR-124-3p, GRB2, which enhanced the proliferation of multiple human cells ; Figure 7D).

Subnetworks of the miR-130/301 Family
There are complicated relationships between the miR-130/ 301 family and other functional molecules associated with the pathogenesis of PH. In the present study, we found that FIGURE 5 | The PH-associated network of ncRNAs and their interacting RNAs or genes in PAFs. The square and circular nodes represent ncRNAs and coding RNAs or genes, respectively. Node color is based on the type of molecule (miRNAs are blue, coding RNAs or genes are empty). Node sizes represent the degrees (number of edges that directly link to the node). Edges represent regulatory connections. Each edge indicates a publication. When multiple publications describe one interaction, multiple edges connect the same two nodes. Red edges depict links indicating repressive action (semicircular arrow heads), and black edges are those indicating activation (traditional arrow heads). The nodes in this network were primarily involved in cell proliferation and extracellular matrix remodeling. PAFs: pulmonary artery fibroblasts.
Frontiers in Genetics | www.frontiersin.org November 2021 | Volume 12 | Article 703860 6   subnetworks of the miR-130/301 family were involved in multiple biological behaviors, such as proliferation, apoptosis, and migration in PASMCs, PAECs, and PAFs. In addition, these subnetworks also mediated the crosstalk of these pulmonary artery cells.
In PAFs, activation of the miR-130/301 family can induce cell proliferation and extracellular matrix remodeling by inhibiting PPARγ and LRP8. Meanwhile, matrix remodeling can activate POU5F1, which subsequently promotes miR-130/301 family expression (Bertero et al., 2015). Thus, a positive feedback circuit is activated that dramatically accelerates the development of PH ( Figure 8C).
The roles of the miR-130/301 family in different pulmonary artery cell types are not independent. Rather, the miR-130/301 family contributes to crosstalk between these cells. Extracellular matrix remodeling, which can be induced by overexpression of the miR-130/301 family, promotes proliferation and contraction of pulmonary artery cells viamiR-130/301 family-dependent and -independent pathways. The remodeled extracellular matrix can activate the POU5F1/miR-130/301 family/PPARγ axis in PASMCs, PAECs, and PAFs, subsequently regulating downstream molecules such as miR-204-5p, miR-424-5p, miR-503-5p, and FGF2 (Bertero et al., 2015). Upregulating miR-424-5p and miR-503-5p or inhibiting FGF2 in PAECs can repress PASMC and PAF proliferation induced by conditioned media from PAECs, indicating that these molecules are involved in the crosstalk among different pulmonary vascular cells. The remodeled extracellular matrix can also induce the expression of the proliferative miRNA, miR-27a/b-3p, in PACEs, and PASMCs, as well as the expression of the vasoconstrictor ET-1, and the inflammatory cytokine IL-6 in PACEs (Bertero et al., 2014;Bertero et al., 2015; Figure 9).

ENVIRONMENTAL FACTORS IN PULMONARY HYPERTENSION
Noncoding RNA interference is an important epigenetic mechanism. Recent evidence has identified the roles of epigenetic changes in the development of PH. These changes link the pathogenic genes of PH and environmental factors such as hypoxia, virus infection, and air pollution (Gamen et al., 2016). For example, BMPR2 is a transmembrane serine/threonine kinase receptor, which is essential for vascular homeostasis. Although mutations in the BMPR2 gene account for a considerable portion of patients with familial pulmonary artery hypertension (FPAH), only 20-30% of carriers with mutations in this gene suffer from PH, indicating that other factors contribute to the onset of the disease (Orriols et al., 2017;Zhao et al., 2019). According to our network, hypoxia can induce the expression of several miRNAs by such as,, and miR-27a-3p (Agrawal et al., 2014;Camps et al., 2014;el Azzouzi et al., 2013;Song et al., 2014;Zhao et al., 2019). Among them, miR-191-5p can increase cell proliferation, impair apoptosis, and induce phenotypic alteration through inhibiting BMPR2 Frontiers in Genetics | www.frontiersin.org November 2021 | Volume 12 | Article 703860 13 expression, subsequently contributing to vascular remodeling (Song et al., 2014). Therefore, the HIF-1α/miR-191-5p/BMPR2 axis reveals the connection between hypoxia and BMPR2 expression and partially explains the incomplete penetrance of BMPR2 mutations in FPAH.

POTENTIAL APPLICATIONS OF NON-CODING RNAS
Ultimately, studies on molecular mechanisms aim to inform clinical practices. NcRNAs are potential diagnostic biomarkers for PH. For example, circRNAs are not easily degraded, making them ideal serum biomarkers. Zhang et al. reported hsa_circ_0068481 overexpression in the serum from patients with IPAH. Furthermore, hsa_circ_0068481 expression is significantly correlated with 6-min walk distance, N-terminal pro-B-type natriuretic peptide, H2S, pulmonary hypertension risk stratification, right heart failure, and survival rate (Zhang et al., 2019a). However, because of the absence of an associated molecular mechanism, this circRNA was not included in our networks. NcRNAs may also act as potential therapeutic targets for PH. For example, Rothman et al. identified downregulation of miR-140-5p in a rat PH model. In vitro, miR-140-5p mimics suppressed PASMC proliferation and migration. In vivo, miR-140-5p mimics prevented the progression of established PH in rats (Rothman et al., 2016). The results are encouraging. However, ncRNA therapy is far from being applied in clinical settings, since a ncRNA may have diverse biofunctions. This means that when used as therapeutic agent, a ncRNA may cause adverse effects, some of which may even be life-threatening. In our opinion, carefully selected ncRNA targets and well-designed action sites can be helpful to avoid such adverse effects. These measures require a comprehensive and in-depth understanding of the mechanisms of ncRNAs in diseases. In this study, we constructed networks to demonstrate the current findings on ncRNAs from studies performed in PH patients and animal models. However, shortcomings of these studies, including the paucity of human data, sex bias, and heterogeneity of animal models, limit the translation of these findings into applications for human disease. Therefore, further studies should be performed to confirm these findings in different animal models and patient cohorts of PH. Additionally, large, welldesigned, and unbiased clinical studies are required to illuminate further application of ncRNAs.

CONCLUSION
The roles of ncRNAs in PH remained unclear. In this study, we performed an extensive literature search and adopted uniform and strict criteria for the selection of each article to avoid biased FIGURE 9 | Contribution of the miR-130/301 family to the crosstalk between various pulmonary artery cells. The remodeled extracellular matrix induces proliferation and contraction in pulmonary artery cells via miR-130/301 family-dependent and -independent pathways. Meanwhile, matrix remodeling can be induced by overexpression of the miR-130/301 family.
outcomes. The ncRNA networks were constructed by assembling ncRNAs and their interacting RNAs or genes from included articles. These networks provide a better understanding of the roles of ncRNAs in PH and can be helpful in elucidating the potential clinical applications of ncRNAs.

AUTHOR CONTRIBUTIONS
Conceptualization, HZ; writing-original draft preparation HZ and QZ; writing-review and editing, HZ and XL. All authors have read and agreed to the published version of the article.