Construction of a ceRNA coregulatory network and screening of hub biomarkers for salt‐sensitive hypertension

Abstract Salt‐sensitive hypertension (SSH) is an independent risk factor for cardiovascular disease. The regulation of long non‐coding RNAs, mRNAs and competing endogenous RNAs (ceRNAs) in the pathogenesis of SSH is uncertain. An RNA microarray was performed to discover SSH‐associated differentially expressed lncRNAs (DElncRNAs) and mRNAs (DEmRNAs), and 296 DElncRNAs and 44 DEmRNAs were identified, and 247 DElncRNAs and 44 DEmRNAs among these RNAs were included in the coexpression network. The coregulatory network included 23 ceRNA loops, and six hub RNAs (lnc‐ILK‐8:1, lnc‐OTX1‐7:1, lnc‐RCAN1‐6:1, GIMAP8, SUV420H1 and PIGV) were identified for further population validation. The ceRNA correlations among lnc‐OTX1‐7:1, hsa‐miR‐361‐5p and GIMAP8 were confirmed in SSH and SRH patients. A larger‐sample validation confirmed that GIMAP8, SUV420H1 and PIGV were differentially expressed between the SSH and SRH groups. In addition, SUV420H1 was included in the SSH screening model, and the area under the curve of the model was 0.720 (95% CI: 0.624‐0.816). Our study explored the transcriptome profiles of SSH and constructed a ceRNA network to help elucidate the mechanism of SSH. In addition, SUV420H1 was identified as a hub element that participates in SSH transcriptional regulation and as a potential biomarker for the early diagnosis of SSH.


| INTRODUC TI ON
Salt-sensitive hypertension (SSH) is an intermediate inheritance phenotype of essential hypertension associated with inter-individual differences and genetic predisposition. 1 Individuals who exhibit increases in blood pressure due to high-salt intake are defined as salt-sensitive (SS), whereas other individuals are salt-resistant (SR). 2 The prevalence of SSH is higher in aged individuals, black Americans and females groups, [3][4][5] and a high prevalence of salt sensitivity is reportedly a feature of hypertension in Asia. 6 Identification of the salt sensitivity among middle-to old-aged populations in China is important for providing advices on salt reduction and healthy habits.
It has been found that SSH is related to increased risks of age-related hypertension and cardiovascular events. 7,8 Additionally, SSH rats and patients exhibit obvious target organ damage, particularly in the renal and cardiovascular systems. 9,10 However, the pathogenesis of SSH is not completely understood and might be related to, for instance, the renin-angiotensin-aldosterone system (RAAS), ion and water channels, the endothelial system, the sympathetic nervous system and the natriuretic peptide system. 11 Therefore, it is necessary to further investigate the mechanism of SSH and discover the core elements involved in the pathways associated with the prevention, early identification and effective therapy of SSH.
Various studies have indicated that genomic variations are associated with SSH. [12][13][14] Because the transcriptome serves as the bridge between genomics and biological functions, it can also participate in the pathogenesis of SSH. MicroRNAs (miRNAs) have been identified as biomarkers for the diagnosis of SSH, 15 and this finding provides insights for the use of other non-coding RNAs as SSH biomarkers.
Long non-coding RNAs (lncRNAs) are endogenous non-coding RNAs with a length of more than 200 nucleotides that can regulate gene expression at the epigenetic, transcriptional and post-transcriptional levels. 16 However, the regulatory role of lncRNAs is not isolated but is associated with a complex of interacting miRNAs and messenger RNAs (mRNAs). An lncRNA can function as a competing endogenous RNA (ceRNA) to absorb available miRNAs and affect the binding of an miRNA to mRNAs through its own miRNA response element. 17,18 This RNA-RNA crosstalk has been widely studied in many chronic diseases, such as cancer, 19 coronary heart disease 20 and myocardial infarction. 21 However, the functions of ceRNA in essential hypertension, as well as SSH, are not well understood. Thus, the ceRNA coregulatory network of SSH is necessary to further explore the mechanism of SSH.
In addition, due to its features of high conservation and active functions, the dysregulation of lncRNAs has been found in and identified as a biomarker for many diseases, particularly cardiovascular disease. 22 Several in vitro studies have found that lncRNAs are differentially expressed in Dahl salt-sensitive rats. 23,24 The lncRNA NPPA antisense can influence the concentration of atrial natriuretic peptide in vivo through regulation of the alternative splicing of NPPA and contributes to the regulation of blood pressure. 25 Thus, lncRNAs might play roles in the pathogenesis of SSH and act as biomarkers for the identification of SSH. Similarly, various mRNAs, such as STK39, have been found to be involved in the pathogenesis of cardiovascular disease in hypertensive patients, 26 and mRNAs have also been associated with epithelial sodium channels 27 and renin 28 in SSH rats. Due to the lack of evidence regarding the roles of lncRNAs and mRNAs in SSH patients, we further screened the SSH-associated differentially expressed lncRNAs (DElncRNAs) and mRNAs (DEmRNAs) from the ceRNA network as hub biomarkers of SSH patients.
In summary, this study aimed to construct a ceRNA network of SSH using data obtained using RNA microarray and bioinformatics technologies, validate the coregulatory relationship among RNAs and screen the utility of the differentially expressed RNAs (DERNAs) that serve as hubs in the ceRNA network for the precise diagnosis of SSH. Additionally, functional enrichment and pathway analyses were performed to explore the potential functions of the hub RNAs, and these findings will provide insights on the mechanism of SSH at the transcriptomic level.

| Participants and sample collection
A total of 112 hypertensive patients (51 SSH and 61 SRH) were selected from the previously established EpiSS (System Epidemiology Study on Salt Sensitivity of Blood pressure) database. 29 The participants were enrolled via telephone notification based on the following inclusion criteria: (a) diagnosis of stage one essential hypertension; (b) age of 40-70 years; (c) living in Beijing for more than 5 years; and (d) Han ethnicity. Patients with severe coronary heart disease, heart failure, stroke, peripheral arterial disease, congenital heart disease, acute myocardial infarction, liver and kidney disease or cancer were excluded. The patients were divided into the SSH (case) and SRH groups (control) according to the modified Sullivan's acute oral saline and diuresis shrinkage test (MSAOSL-DST) results. 30,31 Demographic data and blood pressure and physical measurements were obtained by trained investigators using standard questionnaires. Blood collection was performed by professional nurses.
The blood collected in EDTA tubes was immediately transferred to RNA blood tubes (BioTeke Corporation) for further RNA experiments. This study was approved by the Ethical Committee of Capital Medical University. Prior to initiation of the study, all the participants were informed of the purpose of the study and signed informed consent forms.

| Total RNA extraction and lncRNA-mRNA microarray
A total RNA blood extraction kit (centrifugal type, BioTeke Corporation) was used to extract RNA from whole blood according to the manufacturer's instructions. An Agilent Bioanalyzer 2100 (Agilent Technologies) was then used to determine the concentration and purity of the RNA sample, and 100 ng of RNA was used for 1.5% agarose gel electrophoresis for preliminary quality control. The RNA samples satisfied the following criteria: concentration ≥80 ng/ μL, RIN ≥ 7 and A260/280 value between 1.9 and 2.2. The integrity of the agarose gel electrophoresis requires clearly visible 28S/18S bands without obvious degradation.
The Agilent SBC human (4*180 K) ceRNA array v1.0 was employed to detect the expression of lncRNAs and mRNAs. Ten participants were selected for the microarray test according to their SSH status, age, gender and body mass index (BMI). cRNA was amplified and labelled using the Low Input Quick Amp WT Labelling Kit (Agilent Technologies) following the manufacturer's instructions.
Labelled cRNA was purified using a RNeasy Mini Kit (QIAGEN, GmBH). Each slide was hybridized with 1.65 μg of Cy3-labelled cRNA using a Gene Expression Hybridization Kit in a hybridization oven. An Agilent Microarray Scanner was used to scan the slides.
The raw data were normalized using the Quantile algorithm with the 'limma' packages in R software. The microarray data are available from the Gene Expression Omnibus (GEO) database under the accession number GSE135111.

| Screening for DElncRNAs and DEmRNAs
The significance of DElncRNAs and DEmRNAs was identified based on four criteria: (a) the P values for differential expression were lower than 0.001; (b) the fold changes between the SSH and SRH groups were higher than 2 or lower than 0.5; (c) the RNA signal values were significantly different from the background noise; and (d) the expression range was higher than the median of the range of all RNAs or the mean expression signal was higher than the median of mean expression levels of all RNAs. 32

| LncRNA-mRNA coexpression
A Pearson linear correlation analysis was performed to estimate the lncRNA and mRNA coexpression relationships. The Pearson correlation coefficient (PCC) was calculated using the RNA expression levels. The lncRNA-mRNA pairs with a PCC ≥ 0.95 were selected for further target gene prediction and gene annotation.

| Target gene prediction
Two databases were used to predict the RNA target genes. First, we used miRDB (http://www.mirdb.org/custom.html) to predict the target miRNAs of DElncRNAs. Before the prediction, we obtained the whole sequences of DElncRNAs from LNCipedia.org (http:// Incip edia.org/db/search) and then inputted the sequences into miRDB. Second, miRmap (https://mirmap.ezlab.org/app/) was used to predict the target mRNAs of differentially expressed miRNAs (DEmiRNAs). The top ten listed mRNAs were selected as the prediction results.

| Construction of ceRNA network
We intersected the mRNAs obtained through target gene prediction and the coexpression analysis to obtain the target DEmRNAs. We

| Gene function annotation
The 'EnrichGO' (GO, gene ontology) and 'enrichKEGG' (KEGG, Kyoto Encyclopedia of Genes and Genomes) functions of the Bioconductor 'clusterProfiler' package of R 3.2.2 software were used for the function annotation and pathway analysis of DEmRNAs and for exploring the roles of RNAs in the pathogenesis of SSH. 34 P < .05 was considered to demonstrate significant gene enrichment results.

| Quantitative real-time polymerase chain reaction (qRT-PCR)
The difference and regulatory relationship among hub lncRNAs, mRNAs and miRNAs from the ceRNA network were validated by qRT-PCR. Whole blood samples from 51 SSH patients and 61 SRH patients were collected for the extraction of total RNAs for qRT-PCR using the SYBR Green method. The expression levels of hsa-miR-361-5p, which was the most significant miRNA identified in the previous validation study, and the hub lncRNAs and mRNAs were simultaneously investigated in 20 SSH patients and 19 SRH patients to explore the RNA interactions. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as a stable internal control. The details of the experimental process are provided in the instructions.
For each condition, three replicate experiments were conducted, and the mean cycle threshold (C t ) was calculated. The 2 −ΔCt method was used for the calculation of relative quantitative expression (ΔC t = Ct RNA − Ct GAPDH ).

| Statistical analyses
The sample size used for qRT-PCR validation was calculated using two independent t test formulas. SPSS 24.0 was used for the statistical analyses and hypothesis testing. The original microarray data were normalized using the 'limma' package of R software. If the data were normally distributed, two independent t tests were used to identify the differentially expressed RNAs between the SSH and SRH groups. The non-normally distributed data were normalized. The data that could not be normalized were analysed using a Wilcoxon rank-sum test. The qualitative variables were compared using Pearson chi-squared or Fisher's exact tests. Non-conditional logistic regression was applied to discover the factors that affect the risk of salt-sensitive hypertension. The regulatory relationship between DERNAs was analysed through linear and partial correlation analyses. The area under the curve (AUC) of the receptor operation characteristics curve (ROC) was used to evaluate the diagnostic effects of differentially expressed RNAs. Cytoscape 3.4.0 software was used to visualize the coregulatory network, and heatmaps of DERNAs were drawn using Cluster 3.0 and Java TreeView.

| Identification of differentially expressed lncRNA and mRNA profiles in SSH and SRH patients by microarray
Whole blood samples from five SSH patients and five SRH patients were collected for lncRNA and mRNA microarray analyses. These patients were six females and four males with an average age of 63.60 ± 1.58 years. The microarray chip contained 68 423 lncRNAs and 18 853 mRNAs. Figure 1 showed a flow chart of the study design. The analysis identified 296 lncRNAs and 44 mRNAs that were differentially expressed between the SSH and SRH groups, and these were regarded as DElncRNAs and DEmRNAs, respectively. As

| Coexpression between DElncRNAs and DEmRNAs
The potential interactions between DElncRNAs and DEmRNAs were investigated through Pearson linear correlation analysis. In this study, 247 DElncRNAs and 44 DEmRNAs with a PCC ≥ 0.95 were included in the coexpression network ( Figure 3). As shown in
Similarly, previously discovered DEmiRNAs were regarded as the cores to predict their target mRNAs using the miRmap database (

| GO enrichment and KEGG pathway analyses of DEmRNAs and target mRNAs of DEmiRNAs
Before validation, we integrated the DEmRNAs and target mRNAs of DEmiRNAs to investigate the potential functions of hub RNAs in the pathogenesis of SSH through KEGG pathway and GO enrichment analyses. The KEGG pathway analyses showed that the hub genes were involved in the cancer pathway, cAMP signalling pathway, mTOR signalling pathway, VEGF signalling pathway, aldosterone-regulated sodium reabsorption, WNT signalling pathway, Ras signalling pathway and HIF-1 signalling pathway. These results also indicated that the hub genes in the ceRNA network were enriched in 348 GO functional terms. The five most significant terms of each GO category were listed in Table S3, and these were primarily involved in basic biological processes.

| Baseline characteristics of the validation population and differences in hub lncRNAs and mRNAs between SSH and SRH
The baseline characteristics of 51 SSH and 61 SRH patients are described in  groups. Finally, we found that three mRNAs (PIGV, SUV420H1 and GIMAP8) were significantly differentially expressed between the two groups, whereas three lncRNAs exhibited no significant differences (Table S4).

| Establishing a diagnostic hub RNA model of SSH through logistic regression analysis
The associations of the six hub RNAs (lnc-ILK-8:1, lnc-OTX1-7:1, lnc-RCAN1-6:1, GIMAP8, SUV420H1 and PIGV) with SSH were analysed using a logistic regression model adjusting for dietary factors and history of coronary heart disease, which were found to be associated with SSH, although the differences of these two variables did not reach the statistically significant level in our study. First, we entered all six hub RNAs and adjusted factors into the model and found that SUV420H1 was a danger-

| D ISCUSS I ON
In the present study, we aimed to comprehensively integrate lncRNA-mRNA microarray data with previous miRNA sequencing and ischaemic stroke. 38 For example, the lncRNA CYTOR could modulate pathological cardiac hypertrophy by serving as a ceRNA for miR-155. 39 In addition, the alteration of lncRNA MALAT1 could up-regulate the expression of X box-binding protein by functioning as a ceRNA for miR-124, which contributes to pulmonary arterial hypertension. 40 However, the ceRNAs in SSH were poorly understood, which might be due to the difficulties associated with the definition of SSH. Rapid and chronic oral normal saline loading and depletion are commonly used to diagnose SSH, but all of these tests involve complex procedures and are associated with poor participant compliance. Our study investigated the transcriptome profiles of SSH based on a previously established EpiSS database and therefore fills in the blanks and provides a hypothesis for further validation of the mechanism of SSH.
The larger-sample validation revealed that three mRNAs were differentially expressed between the SSH and SRH groups (PIGV, GIIMAP8 and SUV420H1), and after adjusting for dietary factors, SUV420H1 was the only RNA included in the diagnostic model of SSH. Furthermore, SUV420H1 is the target gene of three SSHassociated miRNAs (hsa-miR-361-5p, hsa-miR-382-5p and hsa-miR-19a-3p) and is strongly correlated with 51 DElncRNAs in the SSH coexpression network. This gene is also involved in two ceRNA loops in SSH, and one of these loops has been verified in SSH patients. This evidence suggests that SUV420H1 might be a core element regulating the pathogenesis mechanism of SSH.
Chinenov et al found that this protein can affect glucocorticoid receptor (GR) target gene expression by participating in the transcriptional regulation of GR. 41 Evidently, glucocorticoid is an important regulator in renal Na + transport 42 that can improve Na + reabsorption and retention, which is stimulated by 11β-hydroxysteroid dehydrogenase (11βHSD). The positive regulatory relationships among 11βHSD, glucocorticoid and SSH have been widely elucidated in rats 43 and humans. 44,45 Thus, we hypothesize that SUV420H1 plays a role in the pathogenesis of SSH, probably by participating in the GR pathway, increasing Na + reabsorption and ultimately resulting in SSH. Additionally, hsa-miR-361-5p was found to be associated with insulin sensitivity through involvement in the WNT signalling pathway, 15 which might also be a potential pathway for hsa-miR-361-5p in SSH. 46 However, direct evidence on the relationships among hsa-miR-361-5p, lnc-OTX1-7:1 and SUV420H is lacking. We will explore the ceRNA regulatory relationships and their interactions on GR through cellular experiments to further illuminate the mechanism of SSH.
Similarly, GO enrichment and KEGG pathway analyses showed that the DEmRNAs of SSH are associated with aldosterone-regulated sodium reabsorption, which is an important mechanism underlying the pathogenesis of SSH. 47 Aldosterone is a steroid hormone that is produced by the zona glomerulosa of the adrenal cortex, which is responsible for homeostatic regulation. The hypersecretion of aldosterone will result in the elevation of blood pressure. 48 The renal function curve of SSH patients moves to the right, which demonstrates that increases in blood pressure result in decreases in sodium excretion. 49 The enrichment analysis illustrated that the hub biomarkers screened from the ceRNA network might result in glo- China, particularly the northern part, people prefer smoked and saltroasted, as well as stewed, chickens with soy sauce, which could increase invisible salt intake 52 and eventually result in an increase in blood pressure. 53 Thus, to better control their blood pressure, the population in northern China could change their dietary habits by reducing the use of paste and sauce, eating home-cooked meals rather than eating at restaurants, 54 using low-sodium salt substitutes 55 and potassium supplementation. 56 This study has some limitations. First, due to the difficulty of salt-sensitive determination, we validated the interactions among ln-cRNAs, miRNAs and mRNAs in 39 participants. Further larger-sample validation is needed to verify the ceRNA relationships. Second, due to the case-control study design, we could only observe the associations between hub RNAs and SSH. The causality needs to be further explored through loss-of-function and-gain-of-function experiments and animal models. Finally, considering the relatively high risks of SSH and the worse prognosis of cardiovascular events in the middle-to older-aged populations, we included essential hypertensive patients aged 40-70 years for the RNA microarray experiment to improve the efficiency of diagnosis and the effectiveness of treatment; thus, the results are only suitable for this population. In future research, we will collect samples from different age stages to explore the differences in ceRNA mechanisms among populations.
In conclusion, six hub DElncRNAs and DEmRNAs were identified from 23 ceRNA loops associated with SSH. The interactions of the six RNAs and hsa-miR-361-5p were validated and provide a hypothesis for the pathogenesis of SSH. In particular, SUV420H1 was not only involved in the ceRNA network but also included in the diagnostic panel of SSH, which indicates that this ceRNA might be a hub element that participates in the pathogenesis of SSH and an important biomarker for the early recognition of SSH.

ACK N OWLED G EM ENTS
This work was financially supported by grants from the National

CO N FLI C T O F I NTE R E S T
The authors confirm that there are no conflicts of interest. Wen-Juan Peng https://orcid.org/0000-0003-1273-6325