Production and Stabilization of Specific Upregulated Long Noncoding RNA HOXD-AS2 in Glioblastomas Are Mediated by TFE3 and miR-661, Respectively

Differential expression of long noncoding RNAs (lncRNA) plays a key role in the development of gliomas. Because gliomas are the most common primary central nervous system tumor and glioblastomas have poor prognosis, it is urgent to develop new diagnostic methods. We have previously reported that lncRNA HOXD-AS2, which is specifically up-regulated in gliomas, can activate cell cycle and promote the development of gliomas. It is expected to be a new marker for molecular diagnosis of gliomas, but little is known about HOXD-AS2. Here, we demonstrate that TFE3 and miR-661 maintain the high expression level of HOXD-AS2 by regulating its production and degradation. We found that TFE3 acted as a transcription factor binding to the HOXD-AS2 promoter region and raised H3K27ac to activate HOXD-AS2. As the cytoplasmic-located lncRNA, HOXD-AS2 could be degraded by miR-661. This process was inhibited in gliomas due to the low expression of miR-661. Our study explains why HOXD-AS2 was specifically up-regulated in gliomas, helps to understand the molecular characteristics of gliomas, and provids insights for the search for specific markers in gliomas.


Introduction
Glioma accounts for nearly 80% of primary tumors in the central nervous system. Glioblastoma multiforme (GBM, glioblastoma) is the most aggressive, invasive tumor of gliomas [1,2]. Patients with glioblastomas have a poor prognosis, and only have a mean survival of 14.6 months among adults [3,4]. However, some molecular features, such as isocitrate dehydrogenase 1 (IDH-1) mutation, confer a favorable prognosis [5]. IDH wild-type glioblastomas typically contain higher rates of epidermal growth factor receptor gene (EGFR) amplification and phosphatase and tensin homolog (PTEN) deletions [5][6][7]. O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation is initially identified as a prognostic and predictive marker within the diagnosis of GBM in patients treated with temozolomide [8]. IDH mutation is thought to be followed by other mutations, such as TP53, ATRX (astrocytoma) or co-deletion of 1p/19q (oligodendroglioma) [9,10]. To improve the glioma treatment, it is essential to get a better understanding of the molecular information of gliomas and develop new therapeutic strategies.
Long noncoding RNAs (lncRNA) are defined as a class of noncoding RNAs of more than 200 nucleotides (nt) in length [11]. Recent studies have demonstrated that the expression of lncRNAs is altered in tumors, and these changes can serve as markers for cancer diagnosis and potential drug targets [12]. LncRNAs are involved in cancer-related signaling pathways, affecting cell proliferation, angiogenesis, cell cycle, tumor invasion and a series of biological processes related to cancer genesis and development [13][14][15].
Some lncRNAs are highly expressed in gliomas, which have been reported to promote cell proliferation, migration, and invasion. The roles played by lncRNAs in tumors deserve in-depth study. We will have a more comprehensive understanding of lncRNAs and their corresponding molecular mechanisms that are involved in various key stages of cancer development and progression. On this basis, lncRNAs are expected to become new targets for tumor therapy [16].
We previously found that lncRNA HOXD Cluster Antisense RNA 2 (HOXD-AS2) was up-regulated in glioma cells by lncRNA microarray. Then, we confirmed that HOXD-AS2 promoted the proliferation of glioblastoma cells and the cell cycle progression; we also demonstrated that the growth of tumors was significantly constrained when silencing HOXD-AS2 in vivo [17]. It has also been found that the activation of the transforming growth factor beta (TGF-β) signaling pathway can up-regulate HOXD-AS2 in highexpressed MGMT glioma cells [18]. HOXD-AS2 can affect the invasion and apoptosis of gliomas by targeting miR-3681-5p and activating the metastasis associated in lung adenocarcinoma Transcript 1 (MALT1) through competing endogenous RNA (ceRNA) mechanism [19]. Moreover, HOXD-AS2 inhibits the PI3K/Akt signaling pathway by targeting the neighboring gene HOXD8 and acts as a tumor suppressor gene in gastric cancer [20]. Overall, there are few studies on HOXD-AS2, particularly the upstream mechanism of HOXD-AS2.
Interestingly, the pan-cancer analysis of HOXD-AS2 demonstrated that the overexpression was specific in gliomas. To explore the reason of this phenomenon, we analyzed the upstream mechanism of HOXD-AS2 in gliomas. We demonstrated that the transcription factor binding to IGHM enhancer 3 (TFE3) and hsa-miR-661 (miR-661) regulated HOXD-AS2 transcriptional activity and transcript stability. TFE3, located on chromosome Xp11.23, encodes a transcription factor containing the Basic Helix-Loop-Helix (BHLH) domain that binds to the MUE3 type E box sequence in the gene promoter, which coding protein promotes the expression of genes downstream of the TGF-β signal [21][22][23]. Fusion of TFE3 and several different partner genes occurs in several types of tumors, results in nuclear expression of the TFE3 protein. The main tumors associated with TFE3 gene fusion are renal cell carcinoma, alveolar soft tissue sarcoma, partial epithelioid hemangioendothelioma (EHE), some perivascular epithelioid cell tumors and rare ossifying fibromyxoid tumors [24][25][26]. miR-661 is frequently reported as a down-regulated gene in gliomas, which inhibits the proliferation, migration and invasion of glioma cells by targeting hTERT, and participates in ceRNA regulation [27,28].
Our studies reveal the molecular mechanism explaining why HOXD-AS2 was specifically overexpressed in gliomas and provide a theoretical basis for the future of HOXD-AS2 as a potential molecular diagnostic marker.

HOXD-AS2 Is Up-Regulated Specifically in Gliomas and Acts as a Risk Factor
We analyzed the expression level of HOXD-AS2 in different kinds of normal tissues via UCSC database. The data demonstrated that HOXD-AS2 was expressed at a low level in brain tissues ( Figure 1a). Then, we used The Cancer Genome Atlas (TCGA) [29] and The Genotype-Tissue Expression (GTEx) [30] data and found that HOXD-AS2 was specifically up-regulated in glioblastomas and low-grade gliomas (LGG) compared to other cancers ( Figure 1b). This phenomenon suggested that HOXD-AS2 might play an important role in the development of gliomas. According to the UCSC database, three transcripts of HOXD-AS2 had been reported. We designed three transcripts' specific primers. RT-qPCR showed that the transcript HOXD-AS2-927nt (nucleotides) was significantly up-regulated in six glioblastoma cell lines (LN229, A172, SNB19, T98G, U87MG and H4) compared with the astrocyte cell lines (HA and HA-sp) (Figure 1c). The result demonstrated that the expression of HOXD-AS2 was high in LN229 and A172; thus, we mainly chose these two glioblastoma cell lines for the subsequent study. U87MG, as a more common glioblastoma cell line, would also be used. Next, we identified the full length of HOXD-AS2 by 5 rapid amplification of cDNA ends (5 RACE). We used the gene-specific primer (GSP) for PCR, then, the sequencing result showed that 60 nucleotides at 5 end of HOXD-AS2 were extended compared with the UCSC database ( Figure 1d). To ascertain the localization of HOXD-AS2 in glioma cells, RT-qPCR showed that HOXD-AS2 was a kind of mainly cytoplasmic-localized lncRNAs (Figure 1e). In addition, we analyzed the expression of HOXD-AS2 in different progression status, subtypes and IDH mutation status in the Chinese Glioma Genome Atlas (CGGA) cohort [31] (Figure S1a-c), finding no significant differences. But the expression of HOXD-AS2 was lower in gliomas with 1p19q co-deletion [32] ( Figure S1d). Next, we investigated the relationship between HOXD-AS2 expression and glioma prognosis. Survival analysis using the Kaplan-Meier method revealed that high-expression HOXD-AS2 group (n = 153) had a shorter overall survival (OS) in glioma patients than the low-expression group (n = 157) ( Figure 1f). Overall, these results indicated that HOXD-AS2 upregulation might play a critical role in the development and progression of gliomas.

HOXD-AS2 Has no Abnormal DNA Copy Numbers in Gliomas and Its Expression Is Activated at the Transcriptional Level
As we all know, loss of heterozygosity (LOH) is the molecular genetics feature of oligodendrocyte. EGFR amplification and exon 2-7 deletion are the most common EGFR mutations in high-grade gliomas. Patients with EGFR amplification mutations have a poor prognosis [33]. Abnormal amplification or deletion of chromosomes is common in cancers. To test whether the high expression of HOXD-AS2 in gliomas is caused by copy number variations (CNVs), we observed CNVs at the HOXD-AS2 locus by the database cBioportal [34,35]. Compared with EGFR, as an amplification with a high incidence in gliomas, there was no significant copy number amplification at the HOXD-AS2 locus ( Figure  S1e). Then, we used absolute quantitative PCR to detect HOXD-AS2 expression in the astrocyte cell HA and glioblastoma cell lines, including LN229, A172, T98G and SNB19, at the level of gene and transcript. HOXD-AS2 did not significantly amplify in glioblastoma

HOXD-AS2 Has No Abnormal DNA Copy Numbers in Gliomas and Its Expression Is Activated at the Transcriptional Level
As we all know, loss of heterozygosity (LOH) is the molecular genetics feature of oligodendrocyte. EGFR amplification and exon 2-7 deletion are the most common EGFR mutations in high-grade gliomas. Patients with EGFR amplification mutations have a poor prognosis [33]. Abnormal amplification or deletion of chromosomes is common in cancers. To test whether the high expression of HOXD-AS2 in gliomas is caused by copy number variations (CNVs), we observed CNVs at the HOXD-AS2 locus by the database cBioportal [34,35]. Compared with EGFR, as an amplification with a high incidence in gliomas, there was no significant copy number amplification at the HOXD-AS2 locus ( Figure S1e). Then, we used absolute quantitative PCR to detect HOXD-AS2 expression in the astrocyte cell HA and glioblastoma cell lines, including LN229, A172, T98G and SNB19, at the level of gene and transcript. HOXD-AS2 did not significantly amplify in glioblastoma cells at the level of gene but was up-regulated at the transcription level (Figure 2a,b), suggesting that the upregulation of HOXD-AS2 was not associated with the abnormal amplification.
We thus presumed whether HOXD-AS2 is aberrantly transcriptionally activated in glioblastomas. According to the UCSC database, we selected a sequence with the length of 2270 nucleotides upstream of HOXD-AS2 transcription start site and inserted it into the pGL3-basic plasmid to detect the relative luciferase activity. Higher reporter gene activity was detected in glioblastoma cell lines, including LN229, A172 and U87MG, than the astrocyte cell HA-sp. The pGL3-basic empty vector was used as the negative control ( Figure 2c). Results indicated that the transcriptional activity of the upstream of HOXD-AS2 transcription start site was enhanced.
Then, we used JASPAR [36], ChIPBase [37] and promo (http://alggen.lsi.upc.es/ cgi-bin/promo_v3/promo/promoinit.cgi?dirDB=TF_8.3, accessed on 23 January 2022) to search transcription factors binding sites in HOXD-AS2 promoter, with 3000 nucleotides upstream of HOXD-AS2 transcription start site used as the sequence for prediction. Based on the results, we analyzed the correction of HOXD-AS2 and transcription factors in the TCGA and CGGA cohort. Supplementary Table S1 provided for all predicted genes. First, HOXD13 and HOXC9, which had the strongest correlation with HOXD-AS2, were selected to verify whether knockdown of HOXD13 or HOXC9 could down-regulate HOXD-AS2. RT-qPCR showed the expression of HOXD-AS2 was not decreased ( Figure S2a). Then, we used the same method to verify USF1, E2F1 and EBF1, which got the highest score in all genes, but the expression of HOXD-AS2 could not be down-regulated by knocking down any one of them ( Figure S2b). According to the previous reports, the expression of HOXD-AS2 is up-regulated through the activation of TGF-β signaling pathway in high-expression MGMT gliomas [18]. We analyzed the expression of MGMT in seven glioblastoma cell lines. The treatment of non-drug-resistant glioblastoma cell lines (LN229 and A172) with TGF-β1 did not affect the expression of HOXD-AS2 ( Figure S2c), and knocking down SMAD2, a key transcription factor in TGF-β signaling pathway, also did not inhibit the expression of HOXD-AS2 ( Figure S2d). These results suggested that there had other mechanisms regulating HOXD-AS2 expression in gliomas with MGMT low expression. Therefore, we chose glioblastoma cell lines with low expression of MGMT to explore the reason why HOXD-AS2 expression was up-regulated instead of T98G, a glioblastoma cell line with high expression of MGMT. It has been proven that STAT3 can regulate the expression of HOXD-AS1 in hepatocellular carcinoma [38]. HOXD-AS1 is located upstream of HOXD-AS2. We found that upstream of HOXD-AS2 transcription start site also had STAT3 binding sites by using JASPAR analysis. However, the expression of HOXD-AS2 could not be reduced by knocking down STAT3 ( Figure S2e).
To identify the potential transcription factors that interacted with the upstream of HOXD-AS2, we amplified the 5 biotin-labeled HOXD-AS2 promoter region located 2270 base pairs (bp) upstream of HOXD-AS2 and used the unlabeled sequence as the negative control (NC). The 5 biotin-labeled HOXD-AS2 promoter was subsequently used to pull-down its associated transcription factors in three glioblastoma cell lines (LN229, A172 and U87MG). After mass spectrometric (MS) detection, all the proteins obtained would be taken to intersect in three glioma cells. In total, six proteins were found only in all biotin groups and not in NC groups, suggesting that these proteins were specifically pulled down by biotin-labeled HOXD-AS2 promoter region. For further screening, we indicated four proteins, which expressed in biotin groups higher than in NC groups (biotin group/NC group, fold change and FC > 1.5) (Figure 2d).  (e,f) The expression of HOXD-AS2 was detected by RT-qPCR and Western blot after transfecting with TFE3/TEAD1/XRCC5 siRNAs in LN229 (e) and A172 (f). Data are presented as the mean ± SEM from three independent experiments. Significant results were presented as ** p < 0.01, *** p < 0.001. Two-tailed Student's t-test was used in (d).
(e,f) were analyzed by ANOVA.
Then, we used JASPAR [36], ChIPBase [37] and promo (http://alggen.lsi.upc.es/cgibin/promo_v3/promo/promoinit.cgi?dirDB=TF_8.3, accessed on 23 January 2022) to search transcription factors binding sites in HOXD-AS2 promoter, with 3000 nucleotides upstream of HOXD-AS2 transcription start site used as the sequence for prediction. Based on the results, we analyzed the correction of HOXD-AS2 and transcription factors in the TCGA and CGGA cohort. Supplementary Table S1 provided for all predicted genes. First, HOXD13 and HOXC9, which had the strongest correlation with HOXD-AS2, were selected to verify whether knockdown of HOXD13 or HOXC9 could down-regulate HOXD-AS2. RT-qPCR showed the expression of HOXD-AS2 was not decreased ( Figure S2a). Then, we used the same method to verify USF1, E2F1 and EBF1, which got the highest score in all genes, but the expression of HOXD-AS2 could not be down-regulated by knocking down any one of them ( Figure S2b). According to the previous reports, the expression of HOXD-AS2 is up-regulated through the activation of TGF-β signaling pathway in high-expression MGMT gliomas [18]. We analyzed the expression of MGMT in (e,f) The expression of HOXD-AS2 was detected by RT-qPCR and Western blot after transfecting with TFE3/TEAD1/XRCC5 siRNAs in LN229 (e) and A172 (f). Data are presented as the mean ± SEM from three independent experiments. Significant results were presented as ** p < 0.01, *** p < 0.001. Two-tailed Student's t-test was used in (d). (e,f) were analyzed by ANOVA.
Before that, we analyzed ChIP-seq data of the glioblastoma cell U87MG and astrocyte cells from Cistrome database [39]. The database showed that higher levels of H3K4me3 and H3K27ac were found in the upstream of HOXD-AS2 in U87MG ( Figure S2f). Of those ten proteins, we screened transcription factor binding to IGHM enhancer 3 (TFE3), TEA domain transcription factor 1 (TEAD1) and X-ray repair cross complementing 5 (XRCC5), as candidate transcription factors that could bind to the HOXD-AS2 promoter, which had been reported to participate in histone modifications.
TFE3 can act as an activated regulator and promote the assembly of an active enhancer repertoire, as evidenced by the deposition of H3K27ac, which then rapidly sequesters cohesion at inflammatory response enhancer-gene promoter clusters [40]. In colorectal cancer, KDM3A depletion causes H3K9me2 up-regulation mainly on TEAD1-binding enhancers rather than gene bodies, further resulting in H3K27ac decrease, less TEAD1 binding on enhancers and transcription impair [41]. H3K27ac is an important marker of activity enhancer with E1A-binding protein p300 and CREB-binding protein (CREBBP/CBP) as catalysts [42]. p300 interacts with XRCC5 by its HAT acetylating XRCC5 in colon cancers and cooperates with XRCC5 to regulate COX-2 expression through acetylating XRCC5 to promote colon cancer growth in vitro and in vivo [43].
It was found that only TFE3 could stably regulate the expression of HOXD-AS2 in two glioblastoma cell lines, LN229 and A172 (Figure 2e,f). Therefore, we hypothesized that TFE3 was the transcription regulator upstream of HOXD-AS2 in glioblastomas.

TFE3 Arises H3K27ac in Glioblastoma Cells and Accelerates the Expression of HOXD-AS2
Based on the TCGA and GTEx database, we obtained that TFE3 was specifically overexpressed in gliomas among different cancers (Figure 3a). In CGGA database, patients with high TFE3 expression had shorter survival ( Figure S3a). The expression of TFE3 was lower in gliomas with 1p19q co-deletion ( Figure S3b). These analyses found similar results for TFE3 and HOXD-AS2 (Figure 1f and Figure S1d). The expression of TFE3 in astrocyte cell lines and glioblastoma cell lines was also detected by RT-qPCR and Western blot. According to Western blot, TFE3 was up-regulated in glioblastoma cell lines, especially in LN229 and A172 ( Figure S3c,d). However, the RT-qPCR result ( Figure S3c) demonstrated that TFE3 expression was not significantly up-regulated on the RNA level. This might be related to post-translational modification. We hypothesized that in LN229 and A172, the mRNA translation efficiency of TFE3 was increased, leading to an increase in TFE3 expression on the protein level. Another possibility is that the TFE3 protein degradation was inhibited, resulting in increased accumulation of TFE3 on the protein level.
To verify whether TFE3 can bind to the HOXD-AS2 promoter, we found that TFE3 had two motifs and six binding sites at the 2270 nucleotides upstream of HOXD-AS2 by JASPAR. Then, we designed three pairs of primers for ChIP-qPCR. P1 represented the sequence located 1826-2021 bp upstream of HOXD-AS2, P2 represented the sequence located 1534-1684bp upstream of HOXD-AS2 and P3 represented the sequence located 396-587 bp (Figure 3b,c). It was found that TFE3 could be enriched in the region upstream of HOXD-AS2 by ChIP-qPCR (Figure 3d), which confirmed the interaction between the TFE3 and HOXD-AS2 promoter. Studies have demonstrated that TFE3 is associated with H3K27ac [40]. Based on the ChIP-seq data, we selected two pairs of primers with high peaks of H3K27ac for CHIP-qPCR. P1 represented the sequence located 1826-2021 bp upstream of HOXD-AS2 and P2 represented the sequence located transcription start site of HOXD-AS2 (chr2:176136890-176137010). ChIP-qPCR detected enrichment of H3K27ac upstream of HOXD-AS2 in glioblastoma cell lines, including LN229 and A172 ( Figure 3e). However, H3K27ac signal could not be detected in astrocyte cell HA by using the same primers ( Figure 3f). To investigate whether TFE3 affected the level of histone acetylation, ChIP-qPCR was performed after knocking down TFE3 in LN229. The results demonstrated that the knockdown of TFE3 inhibited the recruitment of H3K27ac (Figure 3g). The data above demonstrated that HOXD-AS2 was induced by TFE3 and H3K27ac in glioma cells. We also detected HOXD-AS2 promoter activity after the knockdown of TFE3 by transfecting pGL3 plasmid (Figure 2c) and detecting the relative luciferase activity. It was found that the knockdown of TFE3 decreased the activity of the HOXD-AS2 promoter (Figure 3h), indicating that TFE3 could activate the transcription of HOXD-AS2.

miR-661 Degrades HOXD-AS2 in Glioblastoma Cells
We had demonstrated that TFE3 could promote HOXD-AS2 production at the transcriptional level. Next, we considered whether the maintenance of overexpression of HOXD-AS2 in gliomas was related to the regulation of RNA stability.  (h) Luciferase activity assays were performed in astrocyte cell HA-sp and glioma cells included LN229, U87MG and A172, which were transfected with pGL3-basic vector or HOXD-AS2 promoter-containing pGL3 reporter vector and TFE3 siRNAs; firefly luciferase activity was detected and normalized by renilla luciferase activity. The data are shown as the means ± s.d. Two-tailed Student's t-test was used in (d-g). (h) was compared with control by ANOVA. Significant results were presented as * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001.

miR-661 Degrades HOXD-AS2 in Glioblastoma Cells
We had demonstrated that TFE3 could promote HOXD-AS2 production at the transcriptional level. Next, we considered whether the maintenance of overexpression of HOXD-AS2 in gliomas was related to the regulation of RNA stability.
We previously confirmed that HOXD-AS2 was preferentially localized in the cytoplasm (Figure 1e), so we hypothesized that microRNAs (miRNA) may degrade HOXD-AS2 at the post-transcriptional level. Five miRNAs were selected according to the following three conditions: (1) used the full-length sequence of HOXD-AS2 to predict miRNAs bound to HOXD-AS2 in starBase (https://starbase.sysu.edu.cn, accessed on 23 January 2022) [44], miRDB (http://mirdb.org, accessed on 23 January 2022) [45,46] and LNCedting (http:// bioinfo.life.hust.edu.cn/LNCediting/, accessed on 23 January 2022) [47]; (2) found miRNAs that were negatively correlated with HOXD-AS2 in TCGA and CGGA cohort; (3) searched for miRNAs that were down-regulated in gliomas via miRCancer (http://mircancer.ecu. edu/search.jsp, accessed on 23 January 2022) [48,49] (Figure 4a). Supplementary Table S2 provided for all predicted miRNAs. Firstly, we examined the expression of five miRNAs in astrocyte cell HA-sp and glioblastoma cell lines, including LN229, A172 and U87MG (Figure 4b). miR-3944-3p was excluded according to the degree of down-regulation. Then, miRNAs mimics were transfected into three glioblastoma cell lines and detected the expression of HOXD-AS2 by RT-qPCR. After overexpression of miR-454-3p, the expression of HOXD-AS2 did not obviously change, but the rest could down-regulate HOXD-AS2 (Figure 4c). We constructed the psi-check2.0 recombinant vector containing the full-length sequence of HOXD-AS2 and co-transfected with miRNAs mimics to detect relative lu-ciferase activity. We found that only miR-661 could degrade HOXD-AS2 in LN229 and U87MG cells (Figure 4d). In contrast, co-transfection with miR-661 inhibitors in astrocyte cell HA led to the increase of relative luciferase activity, suggesting that HOXD-AS2 became more stable when miR-661 was inhibited (Figure 4e). Based on the predicted binding sequences of miR-661 with HOXD-AS2 from the LNCedting, we designed the mutant sequences in the full-length of HOXD-AS2. The wild type (wt) and mutant (mut) plasmids were respectively co-transfected with miR-661 mimics in the LN229 cell and detected the relative luciferase activity. The results demonstrated that the mutant group did not cause the decrease of relative luciferase activity (Figure 4f).
In order to verify whether there was an endogenous combination between HOXD-AS2 and miR-661, we performed RNA immunoprecipitation (RIP) and biotin-coupled RNA-pull down technique. Ago2-RIP assays affirmed that HOXD-AS2 was enriched in Ago2 groups; anti-IgG was used as the negative control (Figure 4g), indicating that HOXD-AS2 existed in Ago2-related RNA-induced silencing complexes (RISCs). The RNA pull-down assay was performed to figure out whether miR-661 could combine with HOXD-AS2 in glioma cells. The results attested that miR-661 was significantly harvested in HOXD-AS2-sense probe groups in the U87MG cell; HOXD-AS2-antisense probe was used as the negative control ( Figure 4h). The above results indicated that miR-661 could endogenously bind to HOXD-AS2 in U87MG.   and miR-661 inhibitors, then, the relative luciferase activity and firefly luciferase activity was detected, with the latter normalized by renilla luciferase activity. (f) Dual luciferase reporter assays were conducted with wild type and mutant type (putative binding sites for miR-661 were mutated) luciferase reporter vectors (down). Up panel, sequence alignment of miR-661 and its predicted binding sites in HOXD-AS2. (g) RNA immunoprecipitation with anti-Ago2 was used to assess endogenous Ago2 binding to HOXD-AS2 in HA and U87MG cells, IgG was used as the negative control, and HOXD-AS2 expression was detected by RT-qPCR.

Overexpression of HOXD-AS2 Mediated by TFE3 Promotes Cell Cycle Progression
Based on our previous work, HOXD-AS2 could promote cell cycle in glioblastomas [17]. As expected, TFE3 silencing dramatically induced a G1-S arrest (Figure 5a). Meanwhile, we detected the expression of cell cycle-related proteins after knockdown of TFE3 by Western blot (Figure 5b). c-Myc stimulates proliferation in part by activating expression of CDK4 and several of the cyclins, which promote entry into S phase [50]; cyclin D1 is an essential regulator of the G1-S transition [51]; during cell-cycle progression, cyclin A (c) miRNAs mimics were transfected into glioblastoma cell lines LN229, A172 and U87MG to investigate the expression of HOXD-AS2 by RT-qPCR. (d) Luciferase activity assays were performed in LN229 and U87MG cells, which were transfected with psi-check2.0 vector and miRNAs mimics, psi-check-HOXD-AS2 and miRNAs mimics. Firefly luciferase activity was detected and normalized by renilla luciferase activity. (e) Astrocyte cell line HA was transfected with psi-check2.0 vector and miR-661 inhibitors, psi-check-HOXD-AS2 and miR-661 inhibitors, then, the relative luciferase activity and firefly luciferase activity was detected, with the latter normalized by renilla luciferase activity. (f) Dual luciferase reporter assays were conducted with wild type and mutant type (putative binding sites for miR-661 were mutated) luciferase reporter vectors (down). Up panel, sequence alignment of miR-661 and its predicted binding sites in HOXD-AS2. (g) RNA immunoprecipitation with anti-Ago2 was used to assess endogenous Ago2 binding to HOXD-AS2 in HA and U87MG cells, IgG was used as the negative control, and HOXD-AS2 expression was detected by RT-qPCR. (h) Biotin-coupled RNA pull-down was used to examine the interaction of HOXD-AS2 and miR661, and biotin-HOXD-AS2-antisense probe was used as the negative control. The data are shown as the means ± s.d. Two-tailed Student's t-test was used in (b-h). Significant results were presented as * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001.

Overexpression of HOXD-AS2 Mediated by TFE3 Promotes Cell Cycle Progression
Based on our previous work, HOXD-AS2 could promote cell cycle in glioblastomas [17]. As expected, TFE3 silencing dramatically induced a G1-S arrest (Figure 5a). Meanwhile, we detected the expression of cell cycle-related proteins after knockdown of TFE3 by Western blot (Figure 5b). c-Myc stimulates proliferation in part by activating expression of CDK4 and several of the cyclins, which promote entry into S phase [50]; cyclin D1 is an essential regulator of the G1-S transition [51]; during cell-cycle progression, cyclin A level increases at the onset of S-phase and contributes to the stimulation of DNA synthesis [52]; E2F1 belongs to the E2F family of transcription factors that regulate the expression of genes involved in cell cycle progression, especially at the G1 to S phase boundary [53]; p27 is able to inhibit cyclin-dependent kinases (CDK) activity and thus participates in cell cycle regulation. The results demonstrated that the expression of c-Myc, cyclin D1, cyclin A, and E2F1 decreased after knocking down TFE3, suggesting that the TFE3-mediated overexpression of HOXD-AS2 promoted the cell cycle progression in glioma cells. When TFE3 was knocked down, p27 was significantly up-regulated in LN229, but almost unchanged in A172. The relationship between p27 and TFE3 remains to be investigated. We then examined the effect of TFE3 on cell proliferation by MTS (3-(4,5-Dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium) assay and found that the proliferation of glioblastoma cell lines, A172 and LN229, also decreased after knocking down TFE3 (Figure 5c). level increases at the onset of S-phase and contributes to the stimulation of DNA synthesis [52]; E2F1 belongs to the E2F family of transcription factors that regulate the expression of genes involved in cell cycle progression, especially at the G1 to S phase boundary [53]; p27 is able to inhibit cyclin-dependent kinases (CDK) activity and thus participates in cell cycle regulation. The results demonstrated that the expression of c-Myc, cyclin D1, cyclin A, and E2F1 decreased after knocking down TFE3, suggesting that the TFE3-mediated overexpression of HOXD-AS2 promoted the cell cycle progression in glioma cells. When TFE3 was knocked down, p27 was significantly up-regulated in LN229, but almost unchanged in A172. The relationship between p27 and TFE3 remains to be investigated. We then examined the effect of TFE3 on cell proliferation by MTS (3-(4,5-Dimethylthiazol-2yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium) assay and found that the proliferation of glioblastoma cell lines, A172 and LN229, also decreased after knocking down TFE3 (Figure 5c).

Discussion
Further investigations of the molecular biological characteristics of gliomas, and clinical trials to identify more potential molecular markers, contribute to comprehending the pathogenesis of gliomas. HOXD-AS2, as a kind of lncRNA that is specifically upregulated in gliomas, has its own significance and value in research. We explored the mechanism of lncRNA after previously discovering its function in glioblastomas. It is worth mentioning that we did not prioritize the study of the downstream regulatory mechanism but focused on exploring the cause of the specific overexpression of HOXD-AS2 in glioblastomas.
In this study, we found that TFE3, as a transcriptional regulator, could bind to the HOXD-AS2 promoter region and recruit H3K27ac to promote HOXD-AS2 expression. After HOXD-AS2 was transported to the cytoplasm, miR-661 could degrade HOXD-AS2 through incomplete complementary binding. The expression of miR-661 was down-regulated in glioblastomas, so the degradation process was inhibited and HOXD-AS2 expression stayed at a high level in glioblastomas (Figure 5d). Furthermore, we found that the high expression of TFE3 in glioma was also specific and the level of H3K27ac upstream of HOXD-AS2 was low in astrocyte cells. The results confirmed it was specific that TFE3 mediated HOXD-AS2 in gliomas. There may be other pathways that co-regulate the high expression of HOXD-AS2, such as the previously mentioned TGF-β signaling pathway, which can activate HOXD-AS2 in drug-resistant glioma cells [18]. In earlier years, many studies report that TFE3 promotes the expression of genes downstream of the TGF-β signaling pathway [21][22][23]54,55]. In the future, we may be able to explore whether TFE3 and the TGF-β signaling pathway are related with regulating HOXD-AS2 expression in gliomas.
We had known from the previous report that TFE3 can act as an activator to promote the accumulation of H3K27ac in the inflammatory response [40]. We did not study how TFE3 recruited H3K27ac in gliomas, which could be a direction for future exploration. H3K27ac is associated with gene activation and is mainly enriched in promoter and enhancer regions [42]. The first step in gene transcription occurs when the chromatin changes from compact to loose, meaning that the chromatin becomes more open so that subsequent transcription factors can bind to the DNA [56,57]. Histone modification plays a key role in regulating the degree of chromatin compactness. There are many studies on histone modifications and gene regulation. Histone modifications can affect lncRNAs expression, and lncRNAs can also regulate downstream genes through histone modifications. It has been reported that H3K27ac can activate the lncRNA CCAT1, and CCAT1 affects cell proliferation and migration by regulating the expression of SPRY4 and HOXB13 in esophageal squamous carcinoma [58]. lncRNA GClnc1 promotes gastric carcinogenesis and progression by regulating SOD2 through histone acetylation [59].
In 2016, the WHO central nervous system added H3K27M mutant diffuse midline gliomas to diffuse astrocytic and oligodendrocyte tumors [60]. H3K27M mutations frequently occur in high-grade gliomas (HGG) in the midline of the brain, including those in the thalamus, brainstem and spinal cord in children and adolescents [61]. H3K27M mutation was found in 80% of children with diffuse intrinsic pontine glioma (DIPG) [62]. In recent years, it was found that H3K27ac increased in H3K27M HGG genome, and the enrichment of H3K27ac increased the expression of repeat elements in H3K27M HGG, which provided a theoretical basis for H3K27M HGG epigenetic therapy [63].
In the past decades, miRNAs have been demonstrated to be extensively deregulated in cancers, highlighting their important role in tumor development [64]. miR-661 is firstly found in breast cancer as a tumor suppressor [65], but miR-661 performs various functions in a cell-specific mode. It has been reported that miR-661 is up-regulated in non-small cell lung cancer (NSCLC) tissues and promotes cell proliferation, migration and invasion in NSCLC cells. miR-661 directly targets Rb1, which interacts with E2F1 and mediates the EMT process in NSCLC [66].
p53 is a well-known oncogene, and it has been reported that miR-661 targets both mdm2 and mdm4 mRNAs to enhance the oncogenic function of p53 [67]. The report, which shows the oncogenic effect of miR-661 in gliomas, is consistent with our studies and further confirms the negative regulation of HOXD-AS2 and miR-661. When people explored the relationship between lncRNAs and miRNAs, they often focused on ceRNA regulatory networks. However, the regulatory relationship was usually complex, and there were other regulation approaches besides ceRNA. miRNAs could also directly regulate the stability of lncRNAs. For example, EBV-miR-BART6-3p (BART6) inhibits the migratory infiltration of EBV-associated tumor cells by reversing epithelial mesenchymal transition (EMT), and BART6 directly binds to lncRNA LOC553103 and leads to the down-regulation of LOC553103 expression [68].
In summary, our work revealed the mechanism that TFE3 and miR-661 specially regulated HOXD-AS2 in glioma overexpression, and we hope that this work will pave the way for exploring gene specific expression. The differentially expressed lncRNAs, such as HOXD-AS2, will also be used as new diagnostic markers for gliomas in the future.

Cell Culture
Astrocyte cell lines HA and HA-sp were purchased from ScienCell Research Laboratories (Carlsbad, CA, USA) and cultured with astrocyte medium (catalog 1801). The human glioblastoma cell U87MG was purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA) and was maintained in MEM (modified Eagle's medium) (Gibco, Carlsbad, CA, USA) supplemented with 1 mM sodium pyruvate and 1% (vol/vol) non-essential amino acids (NEAA), besides 10% fetal bovine serum (FBS), 100 U/mL penicillin and 100 mg/mL streptomycin. The human glioblastoma cell lines, A172 and LN229, T98G and H4, were purchased from ATCC; SNB19 was obtained from the Beijing Tiantan Hospital of Capital Medical University, Beijing, China. These human glioblastoma cell lines were cultured in DMEM (Dulbecco modified Eagle medium)/high-glucose medium (Gibco, Carlsbad, CA, USA) and supplemented with 10% FBS, 100 U/mL penicillin and 100 mg/mL streptomycin. The isolation, culture and identification of GSC2, T2-4, T12-1 and U87-SLC cells were performed, as described previously [69,70]. All cells were maintained at 37 • C in a humidified atmosphere of 5% CO 2 .
TGF-β1 was dissolved as a liquid according to the manufacturer (R&D, 100-B-001). We added TGF-β1 to LN229 and A172 containing DMEM complete medium at the concentrations of 1 ng/mL, 2 ng/mL and 4 ng/mL and treated for 48 h.

RNA Preparation and Quantitative Reverse Transcription PCR (RT-qPCR)
Total RNA extraction from cells was performed using the RNeasy Mini Kit (QIAGEN, Hilden, Germany). Nucleoplasmic separation was performed by the NE-PER Nuclear and Cytoplasmic Extraction Reagents kit (Thermo, Waltham, MA, USA). After obtaining cytoplasmic and nuclear fractions, the RNA was purified using Trizol reagent (Invitrogen, Carlsbad, CA, USA). The RNA was reversely transcribed into first-strand complementary DNA (cDNA) using a reverse transcription system (Trans Gene, Beijing, China). High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, CA, USA) was used to miRNAs reverse transcription. Quantitative RT-PCR was carried out using SYBR Green (TAKARA, Otsu, Shiga, Japan). U6 and GAPDH were used as endogenous controls. The primers are shown in Supplementary Table S3.

Protein Extraction and Western Blot
Protein extraction and Western blot analysis were performed, as previously described [17]. Cells were lysed in TNTE buffer with protease inhibitors (Sigma Aldrich, St. Louis, MO, USA). After incubation for 30 min on ice, the homogenate was centrifuged at 12,000× g for 30 min. The supernatants were stored at −80 • C for a long time. The protein quantitation was performed by Pierce BCA protein assay kit (Thermo, 23227) according to the manufacturer's protocol. The Bovine Serum Albumin Standard (BSA) was diluted to 0.5 mg/mL for the preparation of standard curves. For Western blot analysis, the protein samples were separated on a 10% SDS-PAGE gel and transferred to nitrocellulose membranes (Amersham, Sweden). The membrane was blocked with 5% skim milk in TBST for 1 h. The primary antibodies that we used are listed in Supplementary Table S4, with anti-β-actin as an internal control. The primary antibodies were revealed using the appropriate secondary antibody conjugated to peroxidase and enhanced chemiluminescence. Anti-mouse IgG (ZB-2305, ZSGB-BIO, Beijing, China) and anti-rabbit IgG (ZB-2301, ZSGB-BIO, Beijing, China) were used as secondary antibodies. The concentration of the antibodies depends on the instruction by the manufacturer. All primary antibodies were incubated overnight at 4 • C on a shaker and all secondary antibodies were incubated at room temperature for 2-3 h.

Plasmid Construction and Cell Transfection
INTERFERin reagent (PolyPlus, Strasbourg, France) was used to transfect small interfering RNAs (siRNA) (20 nM), RNAiMAX (Invitrogen, Carlsbad, CA, USA) was used to transfect miRNA mimics (20 nM) and miRNA inhibitors (20 nM), FuGENE6 (Promega, Madison, WI, USA) was used to transfect the plasmids into the astrocyte cell lines and glioblastoma cell lines. All siRNAs' sequences designed for specific targets are listed in Supplementary Table S5, and purchased from Gene Pharma (Shanghai, China). We synthesized full-length complementary cDNAs of HOXD-AS2 and cloned it into the expression vector PCIG (a gift from Professor WeiMin Zhong). miRNAs mimics and inhibitors were purchased from Sangon (Shanghai, China). The sequences are also shown in Supplementary Table S5.

Flow Cytometry
Cell cycle was analyzed by flow cytometry and detected, as previously reported [71]. We used propidium iodide (PI)-stained cells. The data were analyzed by ModFit LT software.

Cell Proliferation Analysis
The MTS assay (Promega, Madison, WI, USA) was used to evaluate the effects of TFE3 on the proliferation of glioma cells. Cells were seeded at a density of 2500 cells per well and 100 µL of medium per well in 96-well plates. Twenty microliters of MTS (2 mg/mL DPBS)/PMS (0.92 mg/mL DPBS) detection solution was added to each well and incubated for 2 h. Absorbance was measured with a test wavelength of 490 nm and a reference wavelength of 630 nm to obtain sample signal.

Dual Luciferase Reporter Assay
To explore the transcriptional level of HOXD-AS2 promoter, the cells were transfected with the individual reporter plasmid pGL3-HOXD-AS2 promoter constructs (500 ng/well) and pRL-TK (50 ng/well) as an internal control; pGL3-basic was used as the control and pRL-TK (Promega, Madison, WI, USA) was used as the internal reference plasmid.
To confirm whether miRNAs could bind HOXD-AS2, cells were transfected with miRNA mimics or inhibitors (Sangon, Shanghai, China) (20 nM). After 20 h, the cell were transfected with the luciferase reporter vectors psi-CHECK 2.0 (Promega, Madison, WI, USA) containing the full-length sequences of HOXD-AS2 or mutant sequences purchased from Tsingke (Beijing, China) to detect the relative luciferase activity.
After transfection of plasmids for 48 h, the luciferase activity was measured at a dualluciferase reporter assay system (Promega, Madison, WI, USA). The experiments were performed in triplicate and the results were indicated as the means ± s.d. Firefly luciferase activity was normalized with renilla luciferase activity for each transfected well.

Chromatin Immunoprecipitation (ChIP)-qPCR
Pierce™ Agarose ChIP kit (Thermo, Waltham, MA, USA) was used to identify the enrichment of TFE3 and H3K27ac on the HOXD-AS2 promoter. Briefly, LN229 and A172 cells were incubated with formaldehyde for 10 min to obtain the DNA-protein crosslinks.
Then, cell lysates were subjected to sonicate to get the chromatin fragments and specific antibody or IgG (the negative control) were used for immunoprecipitation. Later, RT-qPCR analysis was employed to analyze the precipitated chromatin DNA. The PCR primers are listed in Supplementary Table S1. 4.9. DNA Pull-Down and Mass Spectrometry HOXD-AS2 promoter pull-down was performed by Dynabeads kilobaseBINDER kit (Thermo, Waltham, MA, USA) according to the manufacturer's protocol. Firstly, 5 Biotinlabeled probe was synthesized by PCR and 5 Biotin-labeled primers were purchased from Sangon (Shanghai, China). The DNA-magnetic beads mixture and the nuclear protein were incubated overnight, and the nuclear protein was extracted from the NE-PER Nuclear and Cytoplasmic Extraction Reagents kit (Thermo, Waltham, MA, USA). The supernatant was obtained for further mass spectrum (Applied Protein Technology, Shanghai, China). The protein peptide samples were digested by protein endonuclease (usually Trypsin) and the samples of enzymatic hydrolysis were analyzed by nanoLC-QE LC-MS/MS. Finally, the LC-MS/MS data were analyzed using mass spectrometry matching software, such as MASCOT, to obtain qualitative identification information of the target protein peptide molecules. The identified proteins are listed in Supplementary Table S6.

RNA Immunoprecipitation (RIP)-qPCR
RiboCluster Profiler RIP assay Kit (MBL, Beijing, China) was used for detecting the binding between HOXD-AS2 and Ago2. Briefly, cell lysates were incubated with pro-teinA/G (Thermo, 26162) beads conjugated with indicated antibody or IgG (the negative control). Followed RT-qPCR, analysis was performed to detect the enrichment of HOXD-AS2.

Biotin-Coupled RNA Pull-Down
For biotin-coupled RNA pull-down assay, Streptavidin Dyna beads (Dyna beads M-280 Streptavidin, Thermo) coated with the biotin-labeled HOXD-AS2 probe and antisense sequences as the negative probe were incubated with cell lysis and processed according to manufacturer's instructions. RNAs were eluted and reverse transcribed to cDNA ss in the previous method. The expression of miR-661 was quantified by RT-qPCR. U6 was used as the housekeeping gene.

5 Rapid Amplification of cDNA Ends
5 rapid amplification of cDNA ends was performed by smarter RACE 5 /3 kit (Takara, Otsu, Shiga, Japan) according to the manufacturer's protocol. The gene-specific primer and sequencing result are shown in Supplementary Table S1.

Statistical Analysis
Data are presented as mean ± SEM and were analyzed using Prism9. Overall survival (OS) was calculated using the Kaplan-Meier method. The significance of the differences between groups was estimated with Student's t-test, chi-square test or one-way analysis of variance (ANOVA), as appropriate. Statistical significance was set at p < 0.05. The correlations between variables were analyzed with the Pearson correlation coefficient.

Informed Consent Statement:
We exclude this statement because the study did not involve humans.
Data Availability Statement: Additional data are available upon reasonable request via e-mail.