The Long Non-Coding RNA HOXA-AS2 Promotes Proliferation of Glioma Stem Cells and Modulates Their Inflammation Pathway Mainly through Post-Transcriptional Regulation

Glioblastomas represent approximatively half of all gliomas and are the most deadly and aggressive form. Their therapeutic resistance and tumor relapse rely on a subpopulation of cells that are called Glioma Stem Cells (GSCs). Here, we investigated the role of the long non-coding RNA HOXA-AS2 in GSC biology using descriptive and functional analyses of glioma samples classified according to their isocitrate dehydrogenase (IDH) gene mutation status, and of GSC lines. We found that HOXA-AS2 is overexpressed only in aggressive (IDHwt) glioma and GSC lines. ShRNA-based depletion of HOXA-AS2 in GSCs decreased cell proliferation and altered the expression of several hundreds of genes. Integrative analysis revealed that these expression changes were not associated with changes in DNA methylation or chromatin signatures at the promoter of the majority of genes deregulated following HOXA-AS2 silencing in GSCs, suggesting a post-transcriptional regulation. In addition, transcription factor binding motif enrichment and correlation analyses indicated that HOXA-AS2 affects, directly or indirectly, the expression of key transcription factors implicated in GCS biology, including E2F8, E2F1, STAT1, and ATF3, thus contributing to GCS aggressiveness by promoting their proliferation and modulating the inflammation pathway.


Introduction
Glioma is the most common primary malignant brain tumor, affecting more than 200,000 individuals worldwide each year [1]. The 2007 World Health Organization (WHO)

HOXA-AS2 Is Specifically Expressed in IDHwt Glioma Samples and GSC Lines
As dozens of HOXA-AS2 isoforms are predicted by the Gencode gene project, we first assessed the expression of all HOXA-AS2 isoforms in three normal brain tissue samples (control), eight IDHwt and five IDHmut glioma samples, and two GSC lines using a strand-oriented RNA-seq approach. HOXA-AS2 was not expressed in control and IDHmut samples. Conversely, we observed a robust signal for some HOXA-AS2 isoforms in most, but not all, IDHwt samples and in the two GSC lines (Figure 1a). Refined analyses of one IDHwt glioma sample by 5 and 3 RACE-PCR (Supplementary Figure S1a), combined with analyses of the RNA-seq pattern, showed that the major HOXA-AS2 transcript in glioma corresponded to a longer form of the ENST0000522193.1 isoform that initiates from a CpG island promoter (CGI 46) and contains at least two exons (Figure 1a and Supplementary Figure S1a). The rest of the study focused on this major isoform.
with analyses of the RNA-seq pattern, showed that the major HOXA-AS2 transcript in glioma corresponded to a longer form of the ENST0000522193.1 isoform that initiates from a CpG island promoter (CGI 46) and contains at least two exons (Figure 1a and Supplementary Figure S1a). The rest of the study focused on this major isoform. For each sample, sense (in black) and antisense (in gray) transcription signals are shown in the lower and upper panels, respectively. The positions of 5′ and 3′ ends, identified using the RACE-PCR approach, in one IDHwt glioma sample are shown by a green and red vertical line, respectively. (b) and (c) Relative expression level of HOXA-AS2 in control, IDHmut and IDHwt glioma, and GSC samples from our cohort analyzed by microfluidic-based RT-qPCR (b) and from the TCGA cohort, analyzed by RNA-seq (c). In (b), values are the fold change relative to the geometrical mean of the expression of the housekeeping genes PPIA, TBP, and HPRT1.* p < 0.05, ** p < 0.01, *** p < 0.001 (Mann-Whitney U-test).
RT-qPCR analysis (n = 10 control samples, n = 8 IDHmut and n = 43 IDHwt glioma samples, and n = 6 GSC lines) confirmed that HOXA-AS2 was expressed only in IDHwt glioma samples and GSC lines (Figure 1b), while it was virtually undetectable in control and IDHmut glioma samples. To assess the reproducibility of these observations, we performed the same analyses in an independent cohort ("TCGA cohort") that included 5 control, 415 IDHmut, and 134 IDHwt glioma samples [27]. We could confirm that HOXA-AS2 was expressed in IDHwt samples, but not in control and IDHmut glioma samples ( Figure  1c). We observed a similar expression pattern for HOTAIRM1 but not for HOTAIR, two HOX lncRNAs involved in glioma biology. Specifically, HOTMAIR1 was expressed in , IDHmut (n = 5) and IDHwt (n = 8) glioma, and Glioma Stem Cell (GSC) (n = 2) samples. For each sample, sense (in black) and antisense (in gray) transcription signals are shown in the lower and upper panels, respectively. The positions of 5 and 3 ends, identified using the RACE-PCR approach, in one IDHwt glioma sample are shown by a green and red vertical line, respectively. (b) and (c) Relative expression level of HOXA-AS2 in control, IDHmut and IDHwt glioma, and GSC samples from our cohort analyzed by microfluidic-based RT-qPCR (b) and from the TCGA cohort, analyzed by RNA-seq (c). In (b), values are the fold change relative to the geometrical mean of the expression of the housekeeping genes PPIA, TBP, and HPRT1.* p < 0.05, ** p < 0.01, *** p < 0.001 (Mann-Whitney U-test).
Furthermore, Kaplan-Meier analyses of data from the R2: Genomics analysis and visualization platform (https://r2.amc.nl; accessed on 13 April 2022) and GDC data portal (https://portal.gdc.cancer.gov/ accessed on 13 April 2022) showed that IDHwt samples with a higher HOXA-AS2 expression level tend to have a poorer survival outcome (Supplementary Figure S2).
2.2. HOXA-AS2 Is a Nuclear RNA and Its Overexpression in IDHwt Glioma Is Associated with H3K27me3 Loss at Its Promoter To characterize HOXA-AS2 expression, we first determined the relative distribution of its mature spliced transcript in cell compartments. Expression analysis in three GSC lines indicated that it was localized in the cytoplasmic and nuclear compartments (Figure 2a). both IDHwt and GSCs, and HOTAIR only in IDHwt samples. HOTTIP, another lncRNA, was almost undetectable or extremely weakly expressed in all samples (Supplementary Figure S1b,c). Altogether, these analyses showed that HOXA-AS2 is overexpressed in aggressive IDHwt glioma samples and GSC lines. Furthermore, Kaplan-Meier analyses of data from the R2: Genomics analysis and visualization platform (https://r2.amc.nl; accessed on 13 April 2022) and GDC data portal (https://portal.gdc.cancer.gov/ accessed on 13 April 2022) showed that IDHwt samples with a higher HOXA-AS2 expression level tend to have a poorer survival outcome (Supplementary Figure S2).

HOXA-AS2 Is a Nuclear RNA and Its Overexpression in IDHwt Glioma Is Associated with H3K27me3 Loss at Its Promoter
To characterize HOXA-AS2 expression, we first determined the relative distribution of its mature spliced transcript in cell compartments. Expression analysis in three GSC lines indicated that it was localized in the cytoplasmic and nuclear compartments ( Figure  2a). Expression levels were normalized to U6 and 18S expression in the cytoplasmic and nuclear fractions, respectively. (b) DNA methylation changes (compared with control samples; n = 5) along the HOXA-AS2 genomic region in IDHwt (n = 55) glioma samples, detected with the HM450K array. Significantly hypermethylated regions are in red (no hypomethylated region was observed). .* p < 0.05, ** p < 0.01 (Mann-Whitney U-test). (c) ChIP analysis of H3K9ac, H3K4me3, and H3K27me3 at the HOXA-AS2 promoter region in control brain samples (n = 5), and IDHmut (n = 5) and IDHwt (n = 7) glioma samples. The values Figure 2. Subcellular distribution and molecular profile of HOXA-AS2. (a) Relative distribution of HOXA-AS2 transcripts in cytoplasm and nucleus of three GSC lines. Expression levels were normalized to U6 and 18S expression in the cytoplasmic and nuclear fractions, respectively. (b) DNA methylation changes (compared with control samples; n = 5) along the HOXA-AS2 genomic region in IDHwt (n = 55) glioma samples, detected with the HM450K array. Significantly hypermethylated regions are in red (no hypomethylated region was observed). * p < 0.05, ** p < 0.01 (Mann-Whitney U-test). (c) ChIP analysis of H3K9ac, H3K4me3, and H3K27me3 at the HOXA-AS2 promoter region in control brain samples (n = 5), and IDHmut (n = 5) and IDHwt (n = 7) glioma samples. The values obtained for each sample are shown. The precipitation level was normalized to that at the TBP promoter (for H3K4me3 and H3K9ac) and at the SP6 promoter (for H3K27me3). (d) Genome Browser integrative view at the HOXA-AS2 locus to show H3K4me3, H3K27ac, and H3K27me3 enrichment, DNA methylation (in the GSC-11 and H9-NSC lines), and the strand-oriented RNA-seq signal in GSC-6, GSC-11, and H9-NSC lines. The positions of the 5 and 3 ends identified by RACE-PCR are shown by a green and red line, respectively.
Next, we investigated the molecular bases of HOXA-AS2 overexpression in IDHwt glioma. We assessed the DNA methylation status of the HOXA-AS2 transcript in control (n = 8) and IDHwt (n = 55) samples by Infinium HumanMethylation450 (HM450K) BeadChip Arrays. In agreement with our previous observation made at the four HOX clusters [7], DNA methylation was increased at the HOXA-AS2 locus in IDHwt samples compared with the healthy control. However, this gain was localized mainly in the transcribed region, while methylation remained low at the promoter region. This indicated that HOXA-AS2 overexpression was not associated with changes in its promoter methylation status (Figure 2b). Conversely, ChIP analysis of the HOXA-AS2 promoter showed a marked decrease in the repressive H3K27me3 mark associated with a gain in the permissive H3K4me3 and H3K9ac marks in IDHwt glioma (n = 7) compared with control samples (n = 5) (Figure 2c).
To evaluate the relative contribution of histone modifications and DNA methylation changes to HOXA-AS2 overexpression also in GSCs, we analyzed two IDHwt GSC lines (GSC-6 and GSC-11) and a neural stem cell line (H9-NSC; control) using ChIP-seq, RNA-seq, and Infinium Methylation EPIC BeadChips. HOXA-AS2 expression in GSC-11, and, to a lesser extent, in GSC-6 cells, was associated with a gain in H3K4me3 in the promoter region and in DNA methylation (analyzed only in the GSC-11 line) in the rest of the locus, and with a marked decrease in H3K27me3 throughout the locus (compared with control H9-NSCs). This decrease was less important in GSG-6 than in GSC-11 cells, and mirrored the different HOXA-AS2 expression levels in these two cell lines (Figure 2d). This pattern is in agreement with our previous observation that the H3K27me3 status recapitulates the transcriptional activity at HOX clusters [7]. Altogether, these observations suggest that H3K27me3 loss at the HOXA-AS2 CGI/promoter is one the main mechanisms to explain its expression in GSC cells.

HOXA-AS2 Ectopic Overexpression Poorly Affects Human Neural Stem Cell Biology
To assess HOXA-AS2 function, we first investigated the consequence of overexpressing spliced HOXA-AS2 (longer form of the ENST0000522193.1 isoform) in human neural stem cells (H9-NSCs; two stable lines overexpressing HOXA-AS2 and two stable lines containing the empty pcDNA 3.1 vector; each experimental-control couple was generated in an independent transfection experiment). We confirmed HOXA-AS2 expression in the two HOXA-AS2-expressing H9-NSC lines and its absence in the two control lines (Supplementary Figure S3a).
We did not observe any difference in cell morphology, proliferation, and apoptosis in HOX-AS2-expressing and control H9-NSC lines (Figure 3a,b, Supplementary Figure S3b). Molecular characterization of the four cell lines did not highlight any significant DNA methylation difference (Infinium Methylation EPIC BeadChips) between HOX-AS2-expressing and control cell lines (Figure 3c). RNA-seq analysis revealed that the expression of only four genes was significantly altered in HOX-AS2-expressing cells, and HOXA-AS2: Fatty Acid Binding Protein 3 (FABP3) and Regulator of G Protein Signaling 16 (RGS16) were upregulated, while the endonuclease Schlafen Family Member 13 (SLFN13) and EBF transcription factor 2 (EBF2) were downregulated ( Figure 3d). Interestingly, FABP3 and RGS16 have been proposed as markers of invasive glioma [28,29], and EBF2 positively regulates neuronal migration [30].
These observations indicated that HOXA-AS2 overexpression in healthy human neural stem cells does not lead to detectable phenotypic alterations, but alters the expression of a few genes that may be relevant for glioma biology.
These observations indicated that HOXA-AS2 overexpression in healthy human neural stem cells does not lead to detectable phenotypic alterations, but alters the expression of a few genes that may be relevant for glioma biology. These cell lines were obtained from two independent transfections, referred to as TR1 and TR2, respectively. (b) Colorimetric assay (n = 5) to quantify cell proliferation in the two HOXA-AS2-expressing H9-NSC lines and their respective control line. (c) Infinium Methylation EPIC BeadChip data were used to produce scatter plots to correlate DNA methylation patterns of HOXA-AS2-expressing (n = 2) and control (n = 2) H9-NSC lines. (d) Volcano plot of differential gene expression in HOXA-AS2-expressing (n = 2) and control (n = 2) H9-NSC lines.

HOXA-AS2 Knockdown Affects GSC Morphology
To assess its function in GSCs, we silenced HOXA-AS2, using an shRNA approach, in the GSC-6 and -11 lines that overexpress HOXA-AS2, although at different levels (Figure 2d). The transfection of two independent shRNAs that target exon 2 led to a decrease in HOXA-AS2 expression by 50 to 70%, compared with control cells transfected with a nonsilencing shRNA (shNS) (Figure 4a,b). HOXA-AS2 silencing altered neurosphere formation (a marker of aggressiveness) in both cell lines (Figure 4c). The quantification of cell viability at 2, 4, 7, and 9 days after HOXA-AS2 silencing showed an important decrease in GSC proliferation compared with control cells (Figure 4d). and control (left panels, n = 2) H9-NSC lines. These cell lines were obtained from two independent transfections, referred to as TR1 and TR2, respectively. (b) Colorimetric assay (n = 5) to quantify cell proliferation in the two HOXA-AS2-expressing H9-NSC lines and their respective control line. (c) Infinium Methylation EPIC BeadChip data were used to produce scatter plots to correlate DNA methylation patterns of HOXA-AS2-expressing (n = 2) and control (n = 2) H9-NSC lines. (d) Volcano plot of differential gene expression in HOXA-AS2-expressing (n = 2) and control (n = 2) H9-NSC lines.

HOXA-AS2 Knockdown Affects GSC Morphology
To assess its function in GSCs, we silenced HOXA-AS2, using an shRNA approach, in the GSC-6 and -11 lines that overexpress HOXA-AS2, although at different levels ( Figure 2d). The transfection of two independent shRNAs that target exon 2 led to a decrease in HOXA-AS2 expression by 50 to 70%, compared with control cells transfected with a nonsilencing shRNA (shNS) (Figure 4a,b). HOXA-AS2 silencing altered neurosphere formation (a marker of aggressiveness) in both cell lines (Figure 4c). The quantification of cell viability at 2, 4, 7, and 9 days after HOXA-AS2 silencing showed an important decrease in GSC proliferation compared with control cells (Figure 4d).

E2F-8 and -1 Are Candidate Factors to Mediate HOXA-AS2 Function in GSCs
Detailed analysis of the transcriptional landscape showed widespread transcriptional alterations following HOXA-AS2 silencing. HOXA-AS2 silencing led to the deregulation of 1725 genes (975 up-and 750 downregulated, respectively) in GSC-6 cells and of 784 genes (457 up-and 327 downregulated, respectively) in GSC-11 cells (Supplementary  Table S2; Figure 5a). This difference between cell lines is consistent with the higher residual HOXA-AS2 expression level in silenced GSG-6 than in GSC-11 cells (Figure 4b). Despite this difference in the number of affected genes, gene ontology analyses showed that the same pathways were affected in both GSC lines. The group of downregulated genes following HOXA-AS2 silencing was enriched in genes involved in the cell cycle, specifically in cell and nuclear division. The upregulated group was enriched in genes involved in the inflammatory and immune response pathways (Figure 5b).   Table S2; Figure 5a). This difference between cell lines is consistent with the higher residual HOXA-AS2 expression level in silenced GSG-6 than in GSC-11 cells (Figure 4b). Despite this difference in the number of affected genes, gene ontology analyses showed that the same pathways were affected in both GSC lines. The group of downregulated genes following HOXA-AS2 silencing was enriched in genes involved in the cell cycle, specifically in cell and nuclear division. The upregulated group was enriched in genes involved in the inflammatory and immune response pathways (Figure 5b). To determine whether this transcriptional alteration could be due to the initial alteration of a few "master" transcription factors, we analyzed motif enrichment at the promoter of deregulated genes following HOXA-AS2 silencing. We found that the binding sites of 53 and 56 transcription factors were enriched at genes that were upregulated in GSC-6 and -11 cells, respectively. Similarly, downregulated genes were the putative targets of 23 and 34 transcription factors in GSC-6 and -11 cells, respectively (Supplementary To determine whether this transcriptional alteration could be due to the initial alteration of a few "master" transcription factors, we analyzed motif enrichment at the promoter of deregulated genes following HOXA-AS2 silencing. We found that the binding sites of 53 and 56 transcription factors were enriched at genes that were upregulated in GSC-6 and -11 cells, respectively. Similarly, downregulated genes were the putative targets of 23 and 34 transcription factors in GSC-6 and -11 cells, respectively (Supplementary Figure S4a). Considering the "top 20" transcription factors showing binding motif enrichment at deregulated genes (Figure 5c), we identified 13 transcription factors that were differentially expressed in at least one HOXA-AS2-silenced GSC line compared with the control (shNS). Specifically, E2F1, E2F2, and E2F8 (with binding motif enrichment in the promoter of downregulated genes) were downregulated in both HOXA-AS2-silenced GSC lines. Among the transcription factors with binding site enrichment in the promoter of upregulated genes, members of the Signal Transducer And Activator Of Transcription (STAT) and Interferon Regulatory Factor (IRF) families, and also Activating Transcription Factor 3 (ATF3) and MYC were upregulated following HOXA-AS2 silencing (Figure 5d). To determine whether these transcription factors were direct HOXA-AS2 targets, we analyzed the correlation between their expression level and that of HOXA-AS2 in glioma using the RNA-seq data of the 134 IDHwt glioma samples from the "TCGA cohort". We did not observe any significant negative correlation for the upregulated transcription factors with binding site enrichment at the promoter of upregulated genes. Conversely, the HOXA-AS2 expression level positively correlated with E2F8 and E2F1 expression, which, therefore, can be considered direct HOXA-AS2 targets (Figure 5d and Supplementary Figure S4b).

Changes in Gene Expression Are Not Associated with Major Changes in the Chromatin Signature at Deregulated Genes Following HOXA-AS2 Silencing in GSCs
Then, we investigated the molecular bases of the widespread transcriptional alteration observed following HOXA-AS2 silencing in the two GSC lines. We did not detect any significant DNA methylation change genome-wide and at the promoter of genes deregulated upon HOXA-AS2 silencing in GSCs as observed in GSC-11 (MethylationEPIC BeadChip) (Figure 6a, Supplementary Figure S5a). Next, we analyzed histone mark signatures relevant for promoter and/or enhancer activity (i.e., the permissive H3K4me3, active H3K27ac, and repressive H3K27me3 histone marks) by ChIP-seq in silenced GSC-6 cells due to the strongest effect of HOXA-AS2 silencing in this line (Figure 5a). Genome-wide changes were limited. We observed H3K4me3 changes (fold change > 2, FDR < 0.01) in 1133 regions that included 120 promoters and 55 enhancers. On the other hand, we detected H3K27ac and H3K27me3 changes only in 317 (43 promoters/15 enhancers) and 332 (21 promoters and 1 enhancer) regions, respectively (Supplementary Figure S5b). The limited number of putative regulatory regions affected (i.e., enhancers and promoters) can hardly account for the widespread gene deregulation observed following HOXA-AS2 silencing in GSC-6 cells (Figure 5a, Supplementary Table S2). To precisely determine the proportion of genes the deregulation of which was associated with chromatin signature alterations, we investigated the histone mark profile at the promoters of the 1725 genes with expression changes following HOXA-AS2 knockdown in GSC-6 cells. We detected concomitant changes in gene expression and promoter signature only for very few genes. We observed a gain in H3K4me3 and H3K27ac at the promoter of 15 and 2 genes (1.5% and 0.2% of all upregulated genes), respectively. We did not find any significant change in the repressive H3K27me3 mark (Figure 6b). Consistently, the two upregulated genes with a gain in H3K27ac, Neuralized E3 Ubiquitin Protein Ligase 3 (NEURL3) and Interferon Alpha Inducible Protein 27 (IFI27), also showed a H3K4me3 gain at their promoter (Figure 6c, Supplementary  Figure S5c). This analysis highlighted that for most genes, changes in gene expression following HOX-AS2 silencing was not associated with changes in the histone mark signature at their promoter. We obtained similar results (i.e., absence of change in histone mark signatures) also at the promoter of the putative direct targets of HOXA-AS2 (Figure 5c,d): the downregulated E2F-8 and E2F-1 genes and the upregulated STAT1 gene (Figure 6d,e, Supplementary Figure S5d).
Altogether, these observations suggest that HOXA-AS2 does not modulate the expression of its target genes through transcriptional regulation.

Discussion
Here, we investigated the regulation and role of the lncRNA HOXA-AS2 in GSCs. The Gencode gene project predicts ~12 HOXA-AS2 isoforms, but there is no consensus in the literature on what are the most relevant isoforms in different tumor types. Many studies focused on one isoform without providing a rationale for this choice. Therefore, first, we used RNA-seq, associated with Race approaches, to show that the longer variant of the ENST0000522193.1 isoform is the major HOXA-AS2 transcript in glioma samples. This variant contains at least two exons and is 1049bp in length after splicing. This isoform was included in the isoforms analyzed in the first studies on HOXA-AS2 in cancer (e.g., [20,23]).
Our data confirmed previous observations made in glioma samples classified according to the 2007 WHO criteria (i.e., GBM and lower-grade glioma) that HOXA-AS2 is upregulated in glioma and that its expression level is positively associated with advanced tumor stages [25,26]. By analyzing glioma samples classified according to the most recent

Discussion
Here, we investigated the regulation and role of the lncRNA HOXA-AS2 in GSCs. The Gencode gene project predicts~12 HOXA-AS2 isoforms, but there is no consensus in the literature on what are the most relevant isoforms in different tumor types. Many studies focused on one isoform without providing a rationale for this choice. Therefore, first, we used RNA-seq, associated with Race approaches, to show that the longer variant of the ENST0000522193.1 isoform is the major HOXA-AS2 transcript in glioma samples. This variant contains at least two exons and is 1049bp in length after splicing. This isoform was included in the isoforms analyzed in the first studies on HOXA-AS2 in cancer (e.g., [20,23]).
Our data confirmed previous observations made in glioma samples classified according to the 2007 WHO criteria (i.e., GBM and lower-grade glioma) that HOXA-AS2 is upregulated in glioma and that its expression level is positively associated with advanced tumor stages [25,26]. By analyzing glioma samples classified according to the most re-cent WHO recommendations [3], the present study refined these observations and also extended them to GSC lines. We found that HOXA-AS2 expression is a characteristic of IDHwt glioma samples and GSCs, while it is not expressed in IDHmut glioma samples. This is similar to our previous observation that widespread HOX gene overexpression is a molecular signature of both IDHwt glioma samples and GSCs [7]. However, the HOX genes expression pattern can differ between IDHwt glioma tissue and GSCs. Specifically, we observed that the expression of the lncRNA HOTAIR, which plays a critical oncogenic role in malignant glioma [31,32], was restricted to IDHwt glioma samples. Conversely, HOXA-AS2 was expressed in both IDHwt glioma samples and GSC lines, suggesting that it may contribute to GSC biology.
In all cancer types where it has been studied, including malignant glioma, HOXA-AS2 has been found to have oncogenic functions, mainly by promoting proliferation [25,33]. Similarly, our silencing experiments in GSC lines suggest that HOXA-AS2 influences cell proliferation, by acting primarily on E2F-8 and E2F-1, two main cell cycle regulators the deregulation of which promotes gliomagenesis [34,35]. In addition to promoting the expression of cell cycle genes, HOXA-AS2 also negatively regulated a subset of genes of the inflammatory pathway in GSCs, although probably in an indirect manner. Motif enrichment analysis suggested that this function could be mediated by the initial downregulation of a few members of the STAT and IRF families and also ATF3. This observation is consistent with the documented tumor suppressor role of STAT1 and ATF3 in glioma and GSC [36,37]. The inactivation of ATF3 is essential for the oncogenic potential of GSCs [37]. In addition, STAT1 downregulation allows GSCs to evade type I interferon suppression [38]. Altogether, our findings suggest that HOXA-AS2 can influence, directly or indirectly, several signaling pathways that are instrumental for the GSC oncogenic potential.
Our findings provide new insights into the HOXA-AS2 mechanism of action by suggesting that its effect in GSCs might rely mainly on post-transcriptional regulation. Its ectopic overexpression did not affect H9-NSC morphology and proliferation, and led to the deregulation of only four genes. Yet, three of them, FABP3, RGS16, and EBF2, are relevant for glioma biology [28][29][30]. This observation suggests that HOXA-AS2 overexpression, as detected in IDHwt glioma and in GSCs, is not sufficient on its own to promote a pathological phenotype. Given the HOX gene widespread reactivation and their functional role in glioma and GSC [7][8][9], it can be hypothesized that HOXA-AS2 collaborates with some of them in this process, such as the lncRNA HOTAIRM1 that is overexpressed in both IDHwt samples and GSCs, such as HOXA-AS2. Other/additional specific but yet-unknown partner(s) might be required for HOXA-AS2 oncogenic function in GSCs. Our findings in HOXA-AS2-silenced GSC lines suggest that such partner(s) may be implicated in posttranscriptional regulation. Indeed, the vast majority of genes transcriptionally deregulated following HOXA-AS2 silencing in GSCs did not show any change in relevant histone mark signatures at their promoter, including at genes we identified as direct HOXA-AS2 targets, suggesting a mechanism that does not rely on transcriptional regulation. It has been proposed that HOXA-AS2 might regulate gene expression by acting as a scaffold for epigenetic modifiers or by sponging miRNAs [26,39]. Our findings are in favor of an effect mediated through miRNA sponging in GSCs. However, it is also important to stress that the silencing strategy used here is not dedicated to reveal an effect mediated by epigenetic modifiers. If HOXA-AS2 regulates genes by acting as a scaffold, epigenetic marks, such histone modifications, deposited by modifiers could be maintained despite its silencing. Therefore, additional studies are needed to fully understand the relative contribution of HOXA-AS2 as a scaffold in GSCs, for instance, by evaluating whether HOXA-AS2 physically interacts with target genomic regions and by identifying its protein partners.
Altogether, our study revealed that HOXA-AS2 is a key actor of GSC biology. Our findings support a model in which its overexpression triggers a cascade of events that promote, through direct and indirect mechanisms, cell proliferation and immune tolerance. Additional studies are needed to test and validate this model in vivo. Nevertheless, HOXA-AS2 is a relevant candidate to support the GSC tumorigenic potential.

RNA Extraction
Total RNA was isolated from frozen tissue samples and frozen cell pellets as previously described [7]. Differential isolation of cytoplasmic and nuclear RNA from GSC samples was performed with the "Cytoplasmic and Nuclear RNA purification Kit" from Norgen Biotek (21000; Thorold, ON, Canada).

RACE-PCR
5 and 3 RACE-PCR amplifications were performed as previously described [46] with the GeneRacer Kit from Invitrogen (L150201, Illkrich, France). The primers used are described in Supplementary Table S1.

RT-qPCR
RT-qPCR data were from previously performed microfluidic-based qPCR assays using glioma and control brain samples and GSC lines (n = 10 control samples, n = 8 IDHmut and n = 43 IDHwt glioma samples, and n = 6 GSC lines) to assess the expression of 37 coding and 17 noncoding HOX transcripts [7]. The primers used for HOXA-AS2, HOTAIR, HOTAIRM1, and HOTIP amplification are described in Supplementary Table S1 [47,48]. These three genes display a similar expression level in IDHwt (n = 134) and IDHmut (n = 415) samples from the TCGA cohort.

DNA Extraction
DNA was isolated as previously described [7].

Array-Based DNA Methylation Analysis
Data for tumor, control, and GSC samples were previously obtained using the Human Methylation 450K (HM450K) BeadArray platform [40] (n = 55 IDHwt glioma, n = 8 control brain samples (GSE123678)), and Infinium Methylation EPIC BeadChips platform [7] (GSC-11 cells that express scramble shRNA) (GSE161175). For this work, DNA from H9-NSCs transfected with an empty vector or the HOXA-AS2 expression plasmid, and DNA from GSC-11 cells that express the two HOXA-AS2 shRNAs (shRNA-2 and shRNA-3) were also analyzed using Infinium Human Methylation EPIC BeadChips (Illumina). DNA bisulfite conversion and array hybridization were performed by Inte-graGen, SA (Evry, France) using the Illumina Infinium HD methylation protocol (Illumina, San Diego, CA, USA). Analyses were performed as previously described [40]. Briefly, β-values were computed using the GenomeStudio software from Illumina. Probes with poor-quality signal, missing signal, overlapping with common SNPs, or present on gonosomes were excluded. Differential analysis was performed using the limma package (https://bioconductor.org/packages/release/bioc/html/limma.html accessed on 11 April 2022) and probes were considered differentially methylated when the adjusted p value (FDR) was <0.05 and when the difference between groups was >0.1. In-house R scripts were used to produce bedgraph files to visualize signals on UCSC Genome Browser. Raw data are accessible at GSE199030.

ChIP qPCR
week and when spheres became large, they were enzymatically dissociated with Accutase (Merck-Millipore, Billerica, MO, USA). The molecular features of GSCs are described in [41][42][43], and the self-renewal, differentiation in vitro, and in vivo tumorigenicity (intracranial xenografts in immunodeficient mice) of GSC cultures were evaluated. SMART lentiviral vectors harboring nontargeting shRNA (VSC11713) or HOXA-AS2 shRNAs (V3SH11249; shRNA-2 V3SH11246-245316977, shRNA-3 V3SH11246-245298993) were purchased from GE Healthcare (Little Chalfont, UK). Sequences are in Supplementary Table S1. Viral particles were produced and concentrated by the Vectorology platform of Montpellier (Biocampus, Montpellier, France). GSCs were infected using a multiplicity of infection of 10 and processed 5 days later for RNA-seq, proteomic, DNA methylation, and ChIP-seq analyses. Data Availability Statement: Data supporting the reported results can be found at NCBI Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/ accessed on 15 April 2022) under the following accession numbers (see corresponding material and method section for details): GSE199030, GSE123892, GSE161438, and GSE161437 for RNA-seq data. GSE199032 and GSE123678 for the Epic and HM450K platforms DNA methylation data. GSE199032, GSE161436, GSM772736, and GSM772801 for ChIP-seq data.