LncRNA HOTAIR Promotes Tumorigenicity in Glioblastoma


 Background Recently, the 2021 WHO Classification of Tumors of the Central Nervous System (fifth edition) was published. WHO CNS5 incorporates numerous molecular changes to the accurate classification of CNS neoplasms. GBM, as a special type of glioma, is confined to IDH-wildtype in WHO CNS5, which has a short medical history and poor prognosis. LncRNA HOTAIR has been found to be responsible for the poor prognosis of several human cancers in previous studies. The clinical significance, the prognostic and treatment value of HOTAIR for GBM in the new WHO CNS5 classification remain unclear.Methods We created a Random Forest-based prediction model to assess the IDH1 mutational status of patients in REMBRANDT cohorts. Then we analyzed the role and the potential mechanism of HOTAIR in TCGA, CGGA, and REMBRANDT cohorts. GSVA and GSEA analysis were used to detect the potential biological functions and signaling pathways. The infiltration of immune cells was quantified by Cibersort. The effectiveness of targeted therapy in the clinical application was sought through drug response analysis. Furthermore, we verified that HOTAIR affected the proliferative activity of that glioma cells.Results We found that HOTAIR is highly expressed in all cohorts. The patients’ samples with high HOTAIR expression have poorer overall survival. The results of the functional analysis indicated that the cell cycle and proliferation-related processes were enriched in the high HOTAIR expression group. The infiltration of immunocytes is different in the two groups. In multiple immune checkpoints, the risk score showed a strong correlation. The analysis of drug treatment response present that the high HOTAIR expression group has a better treatment response and better curative effect on the treatment of temozolomide, sorafenib, lenalidomide, bexarotene, and axitinib.ConclusionIn conclusion, We found that HOTAIR could lead to poor prognosis of GBM mainly through regulation of cell cycle and apoptosis. And HOTAIR could be used as a marker to guide chemotherapy in GBM patients rather than immunocheckpoint inhibitor therapy.


Introduction
Gliomas are the most common primary malignant tumor in the brain that arise from glial cells. Glioblastoma (GBM), which is characterized by highly malignant and rapidly progressive, has a median overall survival (OS) of only 15months after being treated with radiation and adjuvant temozolomide chemotherapy (Stupp protocol) [1,2]. The fth edition of the World Health Organization (WHO) Classi cation of Tumors of the Central Nervous System (WHO CNS5) was recently released [3].
The distinction between adult and pediatric diffuse gliomas is being made for the rst time. Adult-type diffuse gliomas are classi ed as three types: astrocytoma, IDH-mutant; oligodendroglioma, IDH-mutant, and 1p/19q-codeleted; glioblastoma, IDH-wildtype [3]. These provide adequate conditions for selecting more homogeneous subjects for future clinical studies via reclassifying the pathologic types with the same molecular characteristics and similar prognosis into one type. With further research on the development of GBM, the role of new molecular markers in pathological diagnosis targeted therapy and immunotherapy and prognostic evaluation needed to be further explored. In this study, GBMs included were under the WHO CNS5 classi cation standard.
HOX transcript antisense RNA (HOTAIR) is a well-studied carcinogenic lncRNA located on human chromosome 12q13. It was the rst lncRNA to be discovered to in uence gene expression in trans fashion, and it is now suspected of playing a role in the carcinogenesis of several human cancers [4]. It has been reported that HOTAIR regulated the invasion and metastasis of breast carcinomas [5] and colorectal cancer[6] in a PCR2-dependent manner. Moreover, HOTAIR could block the expression of microRNA to promote bladder cancer [7], esophageal cancer [8], and gall bladder cancer [9] proliferation and invasion as a competing endogenous RNA. HOTAIR expression levels were found to be signi cantly positively correlated with tumor grade in a study of 295 glioma samples from the Chinese Glioma Genome Atlas (CGGA), Repository of Molecular Brain Neoplasia Data (REMBRANDT), and GSE4290 datasets. Glioma patients with high HOTAIR levels had shorter overall survival [10]. Concordantly, another study based on The Cancer Genome Atlas (TCGA), four datasets from the Oncomine database, and two independent Portuguese and French glioma series, con rmed that HOTAIR expression is far more frequent in grades III and IV than in grade II gliomas and normal brain [11]. They all demonstrated that HOTAIR could be a clinically-relevant biomarker of prognosis in GBM (2016 WHO CNS) [10,11]. Additionally, HOTAIR may be responsible for malignant progression and poor prognosis of glioma patients by regulating cell cycle [10].
However, the prognostic value of HOTAIR for GBM in the new WHO CNS5 classi cation and its molecular mechanisms underlying remain elusive. In this context, we aimed to clarify the clinical signi cance and function of HOTAIR in GBM samples by analyzing clinical and molecular pathology features.

Publicly available sample collection
All datasets used throughout this study are available to the public. Transcriptome data and clinical information of 205 patients of the CGGA cohort [12,13] were obtained from the CGGA website (http://www.cgga.org.cn). The clinical phenotype of 564 patients and expression data of normalized combined the TCGA brain lower-grade glioma (LGG) and glioblastoma multiforme (GBM) were obtained from UCSC Xena website [14] (http://xena.ucsc.edu). The CGGA and TCGA cohorts were the training cohorts. 179 patients from the REMBRANDT cohort [15] and associated clinical information were obtained from the GlioVis website (http://gliovis.bioinfo.cnio.es) as a validation cohort. The transcriptome data from GSE104722 was taken to construct the IDH1 mutant prediction model obtained from the GEO website (http://www.ncbi.nlm.nih.gov/geo). The details of these datasets are listed in Table  1.

IDH1 mutation status prediction
We created a Random Forest (RF)-based prediction model by R package "randomForest"[16] to assess the IDH1 mutational status of patients in REMBRANDT cohorts. The RNA-seq data of GSE104722 have been made to develop this prediction model, and samples with available IDH1 mutational status in TCGA was used as external validation for evaluating the model performance (Since the normalization method of transcriptome data of CGGA is different from that of GSE104722 and TCGA, CGGA is not taken as a validation set.). Differential expression analysis between IDH1 mutant and IDH1 wild-type samples was performed to select informative genes which served as input into the RF model (abs(Log2FC) > 1 & adjust P < 0.05). Speci cally, using the out-of-bag (OOB) error as a minimization criterion, at each iteration, the least important 20% of genes were removed until the OOB error rate reached its minimum. The nal RF prediction model was established with the genome with the lowest OOB error rate. Subject operating characteristic (ROC) curves were used to evaluate the predictability of the model in training and validation tests.

Pathway enrichment analysis
Based on the expression pro le of each sample between high and low HOTAIR expression groups, Gene Set Variation Analysis (GSVA) and Gene set enrichment analysis (GSEA) were conducted to estimate the score of gene-set in each group. GSVA was applied to enrich hallmark gene sets using the "GSVA" [17] R package, while the kegg gene sets were as the reference in GSEA using the "clusterPro ler"[18] R package. The enrichment score of each gene-set more than 1.2 and adjusted P-value less than 0.05 was regarded as signi cantly enriched.

The exploration of the immune microenvironment
To further explore the immune microenvironment in GBM, the relative abundance of 22 immune cells in each patient was computed by "Cell type Identi cation by Estimating Relative Subsets of RNA Transcripts (CIBERSORT)" [19]. Moreover, single-sample GSEA (ssGSEA) [20] in the "gsva" R package was performed to calculate the relative abundance of 28 immune cells. Sets of characteristic genes for each immune cell type were obtained from the previous studies [21]. We calculated the correlation between HOTAIR expression level and the abundance of 28 immune cells.
The response to immune checkpoint blockade prediction Based on the tumor that before immunocheckpoint inhibitor including anti-PD1 and anti-CTLA4 treatment expression pro les, the TIDE module can estimate patient response to immune checkpoint blockade [22] (http://tide.dfci.harvard.edu.).

Drug Sensitivity Analysis
Based on the GDSC2 dataset in Genomics of Drug Sensitivity in Cancer (GDSC) [23], the largest open pharmacogenomics database, a ridge regression model was constructed to predict the drug treatment response of each sample based on the transcriptome data using "pRRophetic" package. In the present study, we estimated the half-inhibitory concentrations (IC50) of temozolomide, sorafenib, lenalidomide, bexarotene, and axitinib of each individual.

Cell culture and cell transfection
The human glioma cell line U251 was purchased from the College of Life Science of Wuhan University (Wuhan, China). All cells were cultured in high-glucose Dulbecco's modi ed Eagle's medium (DMEM, Gibco, USA) containing 10% fetal bovine serum (FBS, HyClone, USA) with a standard humidi ed incubator under 5% CO2 at 37°C.
The full-length complementary cDNAs of human HOTAIR were synthesized and cloned into the pcDNA3.1 vector (Sangon Biotech). Lipofectamine 3000 (Invitrogen, Thermo Fisher Scienti c) was used to transfect the plasmids into U251 cells according to the manufacturer's protocol. The cells transfected for 48h were used for subsequent experiments.

Real-time PCR
RNA was extracted from U251 cells by using Trizol reagent (Invitrogen, Carlsbad, CA, USA). Isolated RNA was reverse transcribed to cDNA using the cDNA synthesis kit (Thermo Fisher, TOYOBO, Japan). The resulting cDNAs were subjected to quantitative real-time PCR with a Bio-Rad CFX Connect real-time PCR detection system (Bio-Rad, Hercules, CA, USA). The primer sequences are shown in Table 2. The relative quantitative value for each gene was calculated using the 2−ΔΔCt method.

CCK-8 and colony formation assays
Transfected U251 cells were seeded in 96-well plates and cell viability was assessed using the Cell Counting kit-8 (CCK-8; Vazyme, Nanjing, China) according to the manufacturer's protocol. The absorbance at 450nm was measured using a spectrophotometer.
In the colony formation assay, cells were seeded in six-well plates and cultured for 14 days after transfection. Then, we used PBS to wash the resulting colonies twice, and used 4% formaldehyde to x them for 15 min, nally used 0.1% crystal violet to stain for 15 min.

Statistical analysis
All statistical tests were conducted in R statistical software (Version 4.0.2) and GraphPad Prism 8.4.0 (GraphPad Software, San Diego, CA, USA) software. We used Student's t-test, Wilcoxon rank-sum test, and Permutation test in this study. The relationship between HOTAIR and other continuous variables was measured by the Spearman method. Survival analysis was carried out using Kaplan-Meier curve, and the log-rank test was used to determine the statistical signi cance of differences. Multivariate cox proportional hazard regression was used to explore the related independent predictors of the prognosis. All reported P-values were two-sided, and the statistical signi cance was set at 0.05.

Clinical characteristics of rearrangement of included populations
According to the latest classi cation criteria for gliomas from WHO [3], WHO grade 4 astrocytomas, IDH mutant in WHO CNS5 is mainly consistent with glioblastoma, IDH-mutant in the 2016 World Health Organization (WHO) Classi cation of Tumors of the Central Nervous System (2016 CNS WHO) [24]. Glioblastoma, IDH-wildtype in WHO CNS5 includes the glioblastoma, IDH-wildtype in 2016 CNS WHO. Therefore, IDH-wildtype glioblastoma patients in the previous classi cation were reserved for GBM in this study. IDH-mutant, WHO grade II-III gliomas were retained as IDH-mutant gliomas in the previous classi cation. While IDH-wildtype, WHO grade II-III gliomas combination of TERT promoter mutation or EGFR gene ampli cation or the combined gain of entire chromosome 7 and loss of entire chromosome 10 (+7/−10) were classi ed as GBM [25]. In this study, all gliomas included were reclassi ed according to WHO CNS5. A total of 948 glioma patients with prognostic information were included from three clinical cohorts (TCGA, CGGA, REMBRANDT). The detailed information of all clinical cohorts in this study is summarized in Table 1, and the ow chart of this study design is shown in Figure 1.

A robust model predicts IDH1 mutations based on transcriptome data
Utilizing the RF method, an effective prediction model of IDH1 mutation was constructed on the GSE104722 cohort. As shown in Figure 2a, this prediction model can accurately divide the 20 samples of GSE104722 into 2 groups (IDH1 wild-type and IDH1 mutant). The prediction accuracy of the model was 100% in the GSE104722 cohort and 95.7% in the TCGA cohort. The AUC of the model was 1.00 in the training cohort and 0.96 in the validation cohort, respectively, indicating that the model could effectively predict IDH1 mutation in other transcriptome cohorts (Figure 2b, 2c). Next, we applied the model to GSE104722 to identify the estimated IDH1 wild-type samples. For subsequent studies, GBM samples from TCGA and CGGA cohorts were invoked as a training set, and GBM samples estimated from the REMBRANDT cohort were used as an independent test set.
The expression of HOTAIR in GBM is higher and the higher the expression level, the worse the prognosis Firstly, lncRNAs with abnormal expression in GBM than other types of glioma were screened out from TCGA and CGGA cohorts respectively (Figure 3a). HOTAIR is highly expressed in both cohorts and veri ed in TEMBTANDT cohorts (Figure 3b, c). According to the median expression of HOTAIR, patients in the training and test sets were split into the high HOTAIR expression group and the low HOTAIR expression group. Kaplan-Meier survival analysis showed that GBM patients with low expression of HOTAIR had signi cantly higher OS than patients with high expression of HOTAIR (P < 0.05; Figure 3d). Multivariate analysis was performed on GBM patients in TCGA, CGGA, and REMBRANDT cohorts to assess whether HOTAIR was an independent prognostic factor. Variables with statistical signi cance (P <0.05) were used as independent prognostic factors in multivariate analysis. HOTAIR was found to be an independent prognostic factor in TCGA and CGGA cohorts (Figure 3e). Figure 3f illustrates the relationship between HOTAIR and routine clinical and molecular features in TCGA, CGGA, and REMBRANDT cohorts, respectively, there is a signi cant correlation between the HOTAIR subgroup and TCGA subgroups in CGGA.

Identi cation of HOTAIR-related biological processes
Gene Set Variation Analysis (GSVA) and Gene Set Enrichment Analysis (GSEA) were carried out to investigate which biological processes were associated with poor prognosis in GBM patients with high HOTAIR expression. Firstly, the expression pro les of patients with high and low expression of HOTAIR were analyzed by GSVA. We believe that the biological process of enrichment in the high HOTAIR expression group is related to the poor prognosis of patients. The results reported that cell cycle-related processes, such as G2M checkpoint and E2F targets, showed the highest correlation with poor prognosis. Proliferation-related processes, such as Myc targets V1 and Myc targets V2, autophagy-related mTORC1 signaling, and angiogenesis were also related to poor prognosis. Meanwhile, the apoptosis and metabolism-related processes, such as cholesterol homeostasis, fatty acid metabolism, and bile acid metabolism, were enriched in the group with low expression of HOTAIR (Figure 4a). The results of GSEA also suggested that the disorder of cycle-related processes may be a possible factor resulting in poor prognosis in patients with high HOTAIR expression (Figure 4b). It has been reported that the knockdown of HOTAIR induced the cell cycle arrest in G0/G1 phase in glioma cells [26]. Thus, we analyzed the relationship of the expression of HOTAIR and cell cycle markers (CCNB1(cyclin B1), CCND1(cyclin D1), CDC25A, and CDC25C) and apoptosis markers (CASP3, BAX, BCL2, MCL1). We found that the higher HOTAIR expression was positively correlated with these markers except BAX (Figure 4c, 4d). These results indicate that HOTAIR might promote the transformation of the cell cycle process and inhibit apoptosis, leading to a poor prognosis of GBM patients.
The relationship between HOTAIR and tumor immune microenvironment.
From the GSVA results of TCGA and CGGA cohorts, there is an interesting phenomenon that immunerelated pathways, such as in ammatory response and interferon-gamma response, are enriched in different groups (Figure 4a). In CGGA, the in ammatory response pathway was enriched in high HOTAIR expression patients, while the result of TCGA was the opposite. Thus, we estimated the composition of 22 immune cells in each patient of CGGA and TCGA cohorts by CIBERSORT. Comparing the composition of the immune cells of high and low HOTAIR expression groups, the patients in the high HOTAIR expression group had a lower proportion of CD8+ T cells and monocytes in TCGA, and activated natural killer cells in CGGA, but a higher proportion of follicular helper T cells in TCGA, and activated memory CD4+ T cells, and activated dendritic cells in CGGA (adjusted P < 0.05, Figure 5a). Only M0 macrophages were statistically different between the two risk groups in both the TCGA and the CGGA cohort. Combined with the results of ssGSEA, there was no correlation between the expression level of HOTAIR and the activated immune cells (adjusted P > 0.05, Figure 5b). Since the enrichment of immune cells and immune pathways in TCGA and CGGA was different, we speculated this might lead to differences in immunotherapy. So, the relationship of HOTAIR and immune checkpoint markers was analyzed in two cohorts. Some immune checkpoint genes, such as PDCD1(PD1), CTLA4, and LAG3 have a relatively higher expression in the high HOTAIR expression group in both two cohorts, especially LAG3 (P < 0.05 in both TCGA and CGGA cohorts). However, CD274(PDL1) and PDCD1LG2(PD-L2) were negatively correlated with HOTAIR expression in TCGA, but it was the opposite in CGGA, especially PDCD1LG2(PD-L2) (P < 0.05 in both TCGA and CGGA cohorts) (Figure 5c). Due to the intensity of intratumoral CD8+ T cell in ltrates and tumor programmed cell death ligand 1 (PDL1) expression have been proposed as distinct biomarkers of response to anti-PDL1 therapies [27,28]. In addition, in TCGA and CGGA cohorts, the density of CD8+ T cell and the expression of PDL1 were different in the HOTAIR subgroups, although it was not statistically signi cant. Hence, we evaluated the therapeutic response in two cohorts to immune checkpoint inhibitors via TIDE. The results showed that there was no signi cant correlation between the expression of HOTAIR and patients' response to immune checkpoint inhibitors. Interestingly, we found more patients responded to immune checkpoint blockade in CGGA than in TCGA ( Figure  5d). The IC50 of other commonly used antitumor drugs was predicted in different groups by using the pRRophetic algorithm. The high HOTAIR expression group had higher IC50 of the chemotherapeutic agents (Figure 5e). Based on our ndings and previous studies, HOTAIR promotes cell proliferation by regulating the cell cycle and apoptosis. The role of HOTAIR in GBM was con rmed by the cell experimental method. HOTAIR was overexpressed in glioma cell line U251, and the transduction e ciency was con rmed by qRT-PCR (Figure 6a). Cell proliferation assays, as well as colony formation assays, were performed to investigate the in uence of HOTAIR overexpression on tumor malignancy in GBM cells. The results indicated that HOTAIR overexpression promoted cell growth (Figure 6b), showed a marked increase in colony-forming ability (Figure 6c).
Compared to IDH-wildtype GBM, IDH-mutant GBM showed some molecular phenotypic changes, such as fewer EGFR ampli cation (rare vs 35%), PTEN mutations (rare vs 24%), and TERT promoter mutations (26% vs 72%), but more ATRX mutations (71% vs rare) and P53 mutations (81% vs 27%) [24]. These differences suggested that IDH-wildtype and IDH-mutant GBMs carried different molecular contexts. Given that they had different driving genes, molecular characteristics, and clinical prognosis, all IDHmutant diffuse astrocytic tumors are considered a single type in WHO CNS5, and the IDH-mutant GBM are classi ed as astrocytoma, IDH-mutant, CNS WHO grade 4, instead of GBM [3]. Gliomas exhibited pronounced heterogeneous are prone to confusion in clinical diagnosis and treatment [29]. The changes of classifying gliomas into more pathological types based on the combination of biological and molecular markers are more in line with the natural course of the disease. The new classi cation will enable clinicians to judge the prognosis of patients, choose the more suitable treatment and promote the exploration of novel treatments and evaluate the therapeutic effect of a new therapy. We aimed to explore the role and function of HOTAIR in GBM in WHO CNS5.
In previous studies, HOTAIR positively regulated an 18 genes cell cycle-related mRNA network in human glioma samples obtained from the CGGA and veri ed in GBM cells [30]. The results of our study showed that HOTAIR played a pro-oncogenic role also mainly by accelerating cell cycle conversion and inhibiting apoptosis in GBM. It manifested that there was no difference in the functional mechanism of HOTAIR with the changes of classi cation in glioma, and further highlighting the critical role of HOTAIR in regulating the cell cycle during gliomagenesis.
Furthermore, we also found that the immune-related pathways were enriched in different HOTAIR subgroups in different cohorts, and HOTAIR was signi cantly correlated with CD4+ T cells, monocyte cells, and dendritic cells, and some immune checkpoint genes (LAG3, PD-L2), suggesting that HOTAIR might be involved in the immune regulation of GBM patients. It was reported that HOTAIR was required for the secretion of various cytokines and in ammatory factors including IL-6, iNOS, TNFα, and MIP-1B induced by LPS-treatment macrophage, and played a central role in NF-κB activation upon stimulation with LPS [31]. Additionally, HOTAIR regulated glucose metabolism in macrophages potentially to meet the energy needs during the immune response [32]. Combined with our ndings, HOTAIR may be involved in immune signaling and in ammatory responses in GBM, which may be one of the reasons for the poor prognosis of patients. However, the predicted therapeutic response results in TIDE showed no difference between the two groups, manifesting that HOTIAR might not responsible for immune checkpoint inhibitors in GBM. But possibly due to racial differences, the response rate of CGGA patients to immunotherapy was signi cantly higher than that of TCGA patients.
Recently, HOTAIR was found to mediate the GBM chemoresistance through mediating the expression of hexokinase 2 by targeting miR-125 [33]. Exosome-mediated transfer of lncRNA HOTAIR regulated TMZ resistance by sponging miR-519a-3p in a study including 51 GBM patients receiving temozolomide treatment [34]. We also found signi cant differences in the sensitivity of temozolomide and sorafenib in different HOTAIR subgroups. In addition, HOTAIR was reported to be detected in GBM serum exosomes and observed a reduction of serum HOTAIR levels after surgery and a further reduction at the 2 weeks post-surgery follow-up in one recurrent GBM patient [35]. The si-HOTAIR has been shown to be successfully delivered via superparamagnetic iron oxide nanoparticles to promote the expression of PDCD4 at the transcriptional level, thereby reducing the proliferation, invasion, and tumorigenicity of human glioma stem cells [36]. These ndings manifested that HOTAIR could be a novel target for personalized treatment and monitoring the therapeutic effect of GBM.
The current study does, however, have certain limitations. To begin, we employed the machine learning method to predict IDH1 mutation status in some individuals due to inadequate molecular information. However, in these patients, there may be a little disparity between the estimated and actual IDH1 mutation status. Second, these patients were not included in GBM with de cient information on molecular signatures required for WHO grade II-III IDH-wildtype astrocytomas to be diagnosed as GBM. Third, the detection of the IC50 of drugs in the CCLE database was based on cell lines, and compared with cell lines, the transcriptome data of clinical tumor samples are contaminated by normal tissue and stroma components, which might reduce the accuracy of drug response prediction. Finally, although we veri ed that HOTAIR affected the proliferative activity of that glioma cells in vitro, further experimental and clinical validations are needed to support our hypothesis about the function of HOTAIR.

Conclusion
In conclusion, we analyzed the role of HOTAIR in GBM in WHO CNS5, and we found that HOTAIR might regulate the cell cycle process and inhibit apoptosis in GBM patients (no matter the fourth edition or the fth edition). In addition, HOTAIR may not be involved in the immune regulation of GBM. Overall, this study suggests that HOTAIR could be a novel insight on prognosis prediction and a novel approach for precision therapy but not immunotherapy.

Yes 130
No 430 +7/-10: Combined whole chromosome 7 gain and whole chromosome 10 loss Table 2 The gene primer sequences used in study.

Genes
Forward (  TCGA cohort (c.). The closer the AUC value is to 1, the higher the true positive rate is, and the lower the false positive rate is.