H19- and hsa-miR-338-3p-mediated NRP1 expression is an independent predictor of poor prognosis in glioblastoma

Glioblastoma multiforme (GBM) is the most common and also the most invasive brain cancer. GBM progression is rapid and its prognosis is poor. Therefore, finding molecular targets in GBM is a critical goal that could also play important roles in clinical diagnostics and treatments to improve patient prognosis. We jointly analyzed the GSE103227, GSE103229, and TCGA databases for differentially expressed RNA species, obtaining 52 long non-coding RNAs (lncRNAs), 31 microRNAs (miRNAs), and 186 mRNAs, which were used to build a competing endogenous RNA network. Kaplan–Meier and receiver operating characteristic (ROC) analyses revealed five survival-related lncRNAs: H19, LINC01574, LINC01614, RNF144A-AS1, and OSMR-AS1. With multiple optimization mRNAs, we found the H19-hsa-miR-338-3P-NRP1 regulatory pathway. Additionally, we noted high NRP1 expression in GBM patients, and Kaplan–Meier and ROC analyses showed that NRP1 expression was associated with GBM prognosis. Cox analysis indicated that NRP1 is an independent prognostic factor in GBM patients. In conclusion, H19 and hsa-miR-338-3P regulate NRP1 expression, and this pathway plays an important role in GBM.


Introduction
Glioblastoma multiforme (GBM) is the most malignant pathological type of glioma. GBMs grow rapidly, are highly malignant, and generally progress rapidly, leading to an average survival period of approximately 12 months [1,2]. Although comprehensive treatments including surgery, radiotherapy, and chemotherapy have recently improved, the survival time of GBM patients remains unsatisfactory [3,4]. Currently, tumor genomics provides a basis for discovering pathogenic alterations in malignant cells that can become therapeutic targets, offering hope for GBM patients [5]. Our early bioinformatics analysis also found molecules related to the prognosis of glioma patients and ubiquitinated molecules related to the prognosis of glioma [3,5]. Long non-coding RNAs (lncRNAs), which are an RNA species of greater than 200 nt in length that lack protein coding functions, account for the majority of transcribed sequences in the human genome [6][7][8]. LncRNAs act as competitive endogenous RNAs (ceRNAs), binding to miRNAs and "sponging" them from their targets, thereby regulating the protein levels of coding genes [3,[9][10][11]. Recent studies have shown that the dysregulation of lncRNAs, micro-RNAs (miRNAs), and their downstream target genes plays an important role in the occurrence and development of tumors [12][13][14]. However, the specific mechanisms of lncRNAs as ceR-NAs in individual tumor types remain to be discovered.
NRP1 is a non-tyrosine kinase transmembrane protein composed of 923 amino acids that is widely expressed in diverse tissue types [15]. NRP1 is also expressed in a variety of tumors including leukemia, prostate cancer, breast cancer, pancreatic cancer, and glioma [16,17]. Studies have shown that NRP1 regulates vascular endothelial factor (VEGF) to promote tumor angiogenesis [18,19]. NRP1 interacts with integrin beta-1 in pancreatic ductal adenocarcinoma to promote tumor growth and invasion [20]. Although we have conducted some studies on the role of NRP1 in tumors, we know very little about its role in GBM [21,22]. Therefore, we analyzed the relationship between NRP1 and the prognosis of GBM using genomics.
We obtained differentially expressed lncRNAs and mRNAs by analyzing GBM databases, predicted the miRNAs that could interact with them, and then constructed a lncRNA-miRNA-mRNA regulatory network. Kaplan-Meyer and receiver operating curve (ROC) analyses revealed five survival-related lncRNAs: H19, LINC01574, LINC01614, RNF144A-AS1, and OSMR-AS1. Additionally, we found that NRP1, HOXC6, and SHCBP1 mRNAs were differentially expressed in GBM and were related to patient prognosis. NRP1 is highly expressed in GBM, and it was confirmed in the Chinese Glioma Genome Atlas (CGGA) and GSE16011 databases that NRP1 expression is related to patient prognosis. Cox analysis confirmed that NRP1 was an independent prognostic factor for GBM patients. Ultimately, we propose that the H19-hsa-miR-338-3P-NRP1 pathway could be targeted for GBM treatment and/or used for clinical diagnostics.

Expression of the lncRNAs and mRNAs and ROC, Kaplan-Meyer, and Cox analyses
The beeswarm package of R software was used to analyze the differential expression of NRP1. The survival and survivalROC packages of R software were used to perform Kaplan-Meyer and ROC analyses, respectively, of the lncRNAs and mRNAs. The survival and survminer packages of R software were used to perform Cox analysis.

Statistical analysis
LncRNAs and mRNAs with |logFC| �1 and p-value <0.05 were considered statistically significant. ROC values >0.65 were defined as statistically significant. Univariate and multivariate Cox analyses were performed using R software, with P<0.05 considered statistically significant.

Constructing the ceRNA network in GBM patients
To obtain reliable lncRNAs, miRNAs, and mRNAs associated with GBM patient prognosis, we jointly analyzed the GSE103227, GSE103229, and TCGA databases. These three datasets revealed 1293, 1286, and 2497 differentially expressed lncRNAs, respectively, with 258 at the intersection of the groups (Fig 1A). From these databases, we also obtained 3732, 3564, and 6456 differentially expressed mRNAs, with 2353 common to all datasets (Fig 1B). MiRNAs that could potentially interact with these lncRNAs were obtained using miRcode. And then we retain the interacting lncRNAs, miRNAs and mRNAs. These data were used to demonstrate interactions, which were compiled into the ceRNA network. In total, the network comprised 52 lncRNAs, 31 miRNAs, and 186 mRNAs (Fig 2, S1 Table).

Cox analysis of using NRP1 to predict GBM patient prognosis
In TCGA and GSE16011 datasets, we found that NRP1 was highly expressed in GBM (Fig 6A and 6B). Both Kaplan-Meyer and ROC analyses of NRP1 were statistically significant Kaplan-Meyer and ROC analyzed the relationship between lncRNAs and patient survival. We show two other statistically significant lncRNAs: RNF144A-AS1 (Fig 4A and 4B) and OSMR-AS1 (Fig 4C and 4D). https://doi.org/10.1371/journal.pone.0260103.g004 ( Fig 6C-6F). Univariate Cox analysis found that age, IDH status, TERT status, ATRX status, and NRP1 were related to GBM patient prognosis, and the risk value increased by 1.052, 0.144, 0.582, 2.936, 0.180, and 1.294, respectively, for each additional unit (S2 Table). Multivariate Cox analysis found that age and NRP1 were independent prognostic factors for GBM patients, for which the patient risk value increased by 1.049 and 1.357 for each additional unit (Fig 7  and S2 Table). Using the newly constructed lncRNA-miRNA-mRNA regulatory network, we propose that NRP1 expression plays an important role in GBM and is regulated by H19 and has-miR-338-3p.

Discussion
Through joint analysis of GSE103227, TCGA, and GSE103229 data, we obtained 258 differentially expressed lncRNAs and 2353 differentially expressed mRNAs. MiRcode was used to predict miRNAs that could interact with the obtained lncRNAs; then mRNAs that could interact with these miRNAs were predicted. We constructed a ceRNA regulatory network using the list of interacting lncRNAs, miRNAs, and mRNAs, for which there were 52, 31, and 186, respectively. Kaplan-Meyer and ROC analyses revealed five survival-related lncRNAs: H19, LINC01574, LINC01614, RNF144A-AS1, and OSMR-AS1. To improve the accuracy of our data, we obtained three mRNAs through limited conditions: NRP1, HOXC6, and SHCBP1. Using these three mRNAs, we found the miRNAs that could potentially act on them, and then obtained the lncRNAs that act on these miRNAs. The intersection of these analyses revealed that only H19 satisfied all conditions. Therefore, we propose that NRP1 is regulated by a ceRNA network consisting of H19 and hsa-miR-338-3p. In TCGA and GSE16011, NRP1 was found to be highly expressed in GBM, and in CGGA and GSE16011, it was confirmed by both Kaplan-Meyer and ROC analyses that NRP1 was significantly associated with GBM outcomes. Cox analysis asserted that NRP1 was an independent prognostic factor for GBM. Previous studies have shown that H19 promotes cancer development through different mechanisms, as examples: in gastric cancer through Fas-related protein, in  cholangiocarcinoma through IL-6 and CXCR4, in colorectal cancer through HMGA1, and in multiple myeloma through P50/P65 [23][24][25][26]. Hsa-miR-338-3p is a tumor suppressor gene in glioma and colorectal cancer but an oncogene in salivary adenoid cystic carcinoma [27][28][29][30]. NRP1 is not only closely related to the occurrence and development of tumors and tumor immunity, but also related to vascular development [31][32][33][34][35][36]. However, we still do not know the lncRNAs and miRNAs that interact with NRP1 or whether NRP1 is an independent prognostic factors for GBM patients. On the basis of our analyses of a large number of GBM specimens from multiple databases, we have revealed the interactions of H19 and hsa-miR-338-3p with NRP1, and have shown that NRP1 is an independent factor for the prognosis of GBM patients. In addition, considering our early discovery of glioma, we found that the mechanism of glioblastoma has its own uniqueness [3]. However, how NRP1 regulated by H19 and hsa-miR-338-3p regulates the occurrence and development of GBM still needs further research.
In conclusion, we found that NRP1 is closely related to the prognosis of GBM patients and is an independent prognostic factor. NRP1 is highly expressed in GBM. In GBM, H19 and hsa- miR-338-3p regulate NRP1, and this signaling pathway may serve as an important molecular target for the diagnosis and treatment of GBM patients, thereby improving the prognosis of GBM.
Supporting information S1