Abstract

   Glioblastoma (GBM) is a lethal cancer characterized by hypervascularity
   and necrosis associated with hypoxia. Here, it is found that hypoxia
   preferentially induces the actin‐binding protein, Transgelin (TAGLN),
   in GBM stem cells (GSCs). Mechanistically, TAGLN regulates HIF1α
   transcription and stabilizes HDAC2 to deacetylate p53 and maintain GSC
   self‐renewal. To translate these findings into preclinical therapeutic
   paradigm, it is found that sodium valproate (VPA) is a specific
   inhibitor of TAGLN/HDAC2 function, with augmented efficacy when
   combined with natural borneol (NB) in vivo. Thus, TAGLN promotes cancer
   stem cell survival in hypoxia and informs a novel therapeutic paradigm.

   Keywords: glioblastoma stem cells, HDAC2, HIF1α Hypoxia, natural
   borneol, sodium valproate, transgelin
     __________________________________________________________________

   Here, it is found that TAGLN, which is preferentially expressed in GSC,
   supports the hypoxic response of GSC by binding to HIF1α to form a
   complex. Mechanistically, TAGLN stabilizes HDAC2 to deacetylate p53 and
   maintains GSC self‐renewal under hypoxia. Inhibition of TAGLN/HDAC2‐p53
   cell cycle axis by sodium valproate combined with natural borneol
   prolongs the survival of GBM experimental animal models.

   graphic file with name ADVS-11-2305620-g007.jpg

1. Introduction

   Glioblastoma (GBM) is the most prevalent primary intrinsic malignant
   brain tumor with conventional therapies offering only palliation.^[
   [80]^1 ^] GBM was one of the first cancers to be comprehensively
   genetically characterized with the p53 pathway among the core molecular
   pathways.^[ [81]^2 ^] p53 is a canonical tumor suppressor and
   transcription factor frequently mutated in tumor cells exposed to
   hypoxic microenvironments and regulates cell cycle progression and
   survival.^[ [82]^3 ^] Dysfunction of p53 and cell cycle regulators
   contributes to tumor cell proliferation, migration, and stemness.^[
   [83]^4 ^] Post‐translational modifications (PTMs) of p53 include
   ubiquitination, phosphorylation, and acetylation; PTMs direct p53
   protein stabilization and activation. Acetylation is essential for
   basic p53 function, and multiple sites of p53 (including lysines K370,
   K372, K373, K381, K382, and K386) are acetylated or deacetylated to
   define the overall level of p53 acetylation.^[ [84]^5 ^] p53
   acetylation is crucial for its tumor suppressive functions, and
   deacetylase inhibitors have shown preclinical efficacy in multiple
   cancer types, including GBM.^[ [85]^6 ^]

   The prior designation of GBM as glioblastoma multiforme belies its
   striking intratumoral diversity. Neoplastic cells within GBMs display
   hierarchies of cellular differentiation with stem‐like, cancer stem
   cells at the apex. GBM stem cells (GSCs) contribute to tumor
   initiation, maintenance, and therapeutic resistance.^[ [86]^7 ^] GBMs
   are defined by morphopathological features, including pseudopallisading
   necrosis, hypoxic and inflammatory regions bordered by tumor cells,
   including GSCs, which contribute to tumor recurrence and invasion into
   normal brain.^[ [87]^8 ^] Hypoxia and hypoxia‐inducible factors (HIFs)
   induce cancer stem cell markers and maintain GSC self‐renewal and
   proliferation.^[ [88]^9 ^] The HIF1α/STAT3 coactivation complex induces
   VASORIN to stabilize NOTCH1 and promote GSC growth.^[ [89]^10 ^] Based
   on the negative prognosis associated with hypoxia and its role in
   promoting GSC growth, we hypothesized that identification of novel
   hypoxia‐inducible molecular targets in GSCs may inform the development
   of novel treatments for this lethal cancer.

   TAGLN, also called SM22, is a focal actin‐binding protein and a key
   protein in smooth muscle cells. Like transforming growth factor‐β
   (TGF‐β), which induces TAGLN^[ [90]^11 ^], TAGLN serves
   context‐specific roles in cancer with both tumor‐suppressive and
   oncogenic functions in select cancer types. TAGLN is expressed at low
   levels in bladder carcinoma^[ [91]^12 ^] and prostate cancer^[ [92]^13
   ^], but potentiates the proliferation and growth of other cancer cells,
   which suggests that TAGLN might represent an early event in tumor
   progression. TAGLN is expressed at high levels in colorectal cancer^[
   [93]^14 ^] and ovarian cancer^[ [94]^15 ^], contributing to tumor
   infiltration and migration at advanced tumor stages. In bladder cancer
   cells, TAGLN is regulated by p53 and PTEN (phosphatase and tensin
   homolog) to inhibit proliferation.^[ [95]^12 ^]

   Here, we show that GSCs express high levels of TAGLN relative to the
   other tumor cells, and TAGLN informs poor prognosis for GBM patients.
   Therefore, we investigated the contribution of TAGLN in GSCs adaptation
   to hypoxia and its downstream molecular functions.

2. Results

2.1. Gliomas Express Increased TAGLN

   To identify critical hypoxia‐related molecular targets in GBM, we
   prioritized genes that were highly expressed in GBMs relative to normal
   brains, GSCs within tumors, and the hypoxic microenvironment.
   Therefore, we leveraged three single‐cell RNA‐seq databases for each
   element from GEO for gliomas ([96]GSE103224, [97]GSE139448, and
   [98]GSE141383) with TISCH2 (http://tisch.comp‐genomics.org/home/) and
   the overlapping differential genes (DEGs, n = 50, log[2]FC ≥ 1) (Figure
   [99]S1A, Supporting Information). Next, we generated a Venn diagram
   illustrating the intersection between: 1) 50 associated genes in the
   malignant cells of GBM from TISCH2; 2) DEGs (n = 1334, log[2]FC >2)
   from GEPIA GBM (http://gepia.cancer‐pku.cn/); 3) DEGs (n = 1650, ≥
   median) associated with hypoxia and GSCs in Genecards
   (https://www.genecards.org/); and 4) DEGs (n = 60, log[2]FC ≥ 1 and p ≤
   0.05) associated with HIF1A high expression in GlioVis Gravendeel_GBM.
   The intersection of the four datasets revealed two genes: TAGLN and
   EFEMP1 (Figure [100]1A ). Both genes have been reported to promote the
   migration and infiltration of cancer cells, with EFEMP1 shown to
   promote GBM invasion.^[ [101]^16 ^] To interrogate TAGLN expression in
   multiple tumors, we mapped TAGLN expression across The Cancer Genome
   Atlas (TCGA) databases for multiple tumor types. Compared to normal
   tissues, most tumors expressed relatively lower expression of TAGLN,
   whereas GBM patient samples showed increased expression
   (Figure [102]1B; Figure [103]S1B, Supporting Information). Mutations in
   isocitrate dehydrogenase (IDH) define a glioma subset with a favorable
   prognosis.^[ [104]^4 , [105]^17 ^] Gliomas with wild‐type IDH (IDH‐WT)
   expressed higher TAGLN levels than tumors harboring IDH mutations in
   the TCGA‐GBM/LGG (low‐grade glioma) RNA‐seq database (Figure [106]1C).
   Analysis of TAGLN expression in an independent glioma database, the
   Chinese Glioma Genome Atlas (CGGA), revealed that GBMs showed higher
   TAGLN expression than low‐grade gliomas (LGG), supporting an
   association with tumor grade (Figure [107]1D). To validate these
   findings at the protein level, we performed immunohistochemistry (IHC)
   for TAGLN on tissue arrays with glioma surgical specimens of various
   grades (grade 2 (n = 23), grade 3 (n = 13), and GBM (n = 45)) and
   normal brain tissues (n = 3). TAGLN was strongly expressed in glioma
   cells in 41 of 45 (91.1%) of GBMs, 11 of 13 (84.6%) of grade 3 gliomas,
   but in fewer than 30% of grade 2 gliomas, and was absent in normal
   brain tissues (Figure [108]1E,F). Consistently, immunoblotting (IB)
   validated high TAGLN expression in GBM patient samples (Figure
   [109]S1C, Supporting Information). TAGLN expression correlated with
   poor patient prognosis in the CGGA GBM (IDH‐WT) and the TCGA GBM/LGG
   (IDH‐WT) databases (Figure [110]1G; Figure [111]S1D, Supporting
   Information). Collectively, these data support TAGLN as a potential
   oncogenic factor in GBM.

Figure 1.

   Figure 1
   [112]Open in a new tab

   TAGLN is highly expressed in glioblastoma. A) Venn diagram showing the
   overlapping of four datasets, hypoxia and GSC‐related genes in
   Genecards, DEGs associated with high HIF1A expression,intersection of
   DEGs in three databases of TISCH2 and DGEs from GEPIA2 GBM, comprised
   two genes: TAGLN and EFEMP1. B) TAGLN expression in a panel of GBM
   (IDH‐WT) and normal brain samples from the Gliovis TCGA RNA‐seq
   dataset. Unpaired t‐tests were used to calculate statistical
   significance, p<0.01. C) Histogram of TAGLN expression in IDH wild‐type
   and mutant GBM samples from the Gliovis TCGA_GBM/LGG (IDH‐WT) RNA‐seq
   dataset. Unpaired t‐tests were used to calculate statistical
   significance, p<0.0001. D) Plot of TAGLN expression in patients from
   the CGGA (IDH‐WT) database (Student's t‐test). E) Representative images
   of glioma tissue microarray samples showing TAGLN expression. Scale
   bar, 50 µm. F) The proportion of TAGLN‐positive tissues in glioma or
   normal brain tissue was shown. G) Kaplan‐Meier survival analysis of
   patients with different TAGLN expression levels in the Gliovis CGGA
   (IDH‐WT) dataset (log‐rank test). Data are represented as mean ± SD
   (**p < 0.01, ***p < 0.001, and ****p < 0.0001).

2.2. Hypoxia Induces TAGLN in GSCs

   Next, we investigated hypoxia regulation of TAGLN expression. The Ivy
   Glioblastoma Atlas Project (Ivy GAP) mapped regional gene expression in
   GBM tissues. TAGLN was highly expressed in regions of pseudopalisading
   necrosis compared to peritumoral regions of GBM (Figure [113]2A ).
   TAGLN expression positively correlated with hypoxia‐related genes
   (HIF1A, CA9, LDHA, PDK1, PGK1, VEGFA, and EPAS1) in the TCGA GBM/LGG
   (IDH‐WT) and CGGA GBM (IDH‐WT) datasets (Figure [114]S2A,B, Supporting
   Information). IHC staining confirmed increased levels of hypoxic
   proteins (HIF1α and CA9) in pseudopalisades of GBM specimens,
   consistent with previous reports.^[ [115]^9 , [116]^18 ^] TAGLN was
   lowly expressed in other regions of GBM, while highly expressed in the
   pseudopalisades of GBM (Figure [117]2B; Figure [118]S2C, Supporting
   Information). Co‐immunofluorescence (IF) staining of TAGLN, CA9, and
   HIF1α revealed that hypoxic regions of GBM samples preferentially
   expressed TAGLN (Figure [119]2C; Figure [120]S2D, Supporting
   Information). To directly test the effects of hypoxia on TAGLN
   expression, we exposed a panel of patient‐derived GSCs to room air (21%
   O[2]) or hypoxia (1% O[2]) over a time course. Over 48 h, TAGLN
   expression increased in GSCs under hypoxia compared with normoxia
   (Figure [121]2D; Figure [122]S2E, Supporting Information), suggesting
   that TAGLN is induced by hypoxia.

Figure 2.

   Figure 2
   [123]Open in a new tab

   Preferential expression of TAGLN in hypoxia‐induced GSCs. A) Volcano
   plot of differentially expressed genes at peritumoral regions and
   pseudopalisades in GBM tissue. Red dots, genes upregulated in hypoxia;
   Blue dots, genes downregulated in hypoxia. B) Representative images of
   HE (Scale bar, 100 µm) and IHC (Scale bar, 50 µm) stained in
   pseudopalisades located around the necrosis (upper) and other areas
   (lower) of serial sections of hGBM samples. The red arrow points to the
   necrosis (N) in GBM and the black arrow represents the pseudopalisades
   (P) around the necrosis. The large black square in the lower‐left
   corner of the image is an enlarged view of the small black square in
   the image. C) Representative images of IF staining for TAGLN, CA9, and
   HIF1α in hGBM tissues. Scale bar, 50 µm. D) Expression levels of TAGLN
   and HIF1α were examined by IB in GSCs cultured in 1% O[2] for the
   indicated time. E) Representative images of IF staining for TAGLN and
   putative GSCs markers in hGBM sections. Scale bar, 25µm. F) Cell
   lysates were analyzed by IB for TAGLN, OLIG2, and GFAP expression. G)
   The expression of TAGLN in GSCs and glioma cells was tested by IB. H)
   IB analyzed TAGLN expression in GSCs and NPCs. I) Kaplan‐Meier curve
   showing patient survival based on HIF1A mRNA expression in the CGGA GBM
   (IDH‐WT) dataset. J) The Kaplan‐Meier survival curve was calculated to
   measure survival in the group of TAGLN ^hi HIF1A ^hi (n = 139), TAGLN
   ^hi HIF1A ^lo, (n = 73), TAGLN ^lo HIF1A ^hi, (n = 73), and TAGLN ^lo
   HIF1A ^lo (n = 139).

   To determine the specificity of TAGLN expression in GSCs, we
   interrogated TAGLN expression in the Gliovis data portal to reveal a
   positive correlation between TAGLN and GSC markers (Figure [124]S2F,
   Supporting Information). TAGLN co‐distributed with GSC markers CD133,
   CD15, SOX2, and OLIG2 in glioma cells (Figure [125]2E; Figure [126]S2G,
   Supporting Information), suggesting that TAGLN was preferentially
   expressed in GSCs in patient tissues. To directly test this, we
   generated differentiated GBM cells (DGCs) from patient‐derived GSCs
   cultured in 10% serum for 14 days to induce differentiation. Serum
   induced the astrocyte marker, glial fibrillary acidic protein (GFAP),
   and reduced the expression of GSC markers, OLIG2 and SOX2. Using these
   matched cells, TAGLN was preferentially expressed in GSCs but not in
   matched DGCs (Figure [127]2F; Figure [128]S2H, Supporting Information),
   bulk glioma cells (Figure [129]2G), or neural precursor cells (NPCs)
   (Figure [130]2H) under hypoxia. To further parse the impact of TAGLN
   and hypoxia, we segregated patients with GBMs based on levels of both
   TAGLN and HIF1A (high vs. low relative to the median) in GlioVis‐GBM
   (IDH‐WT). The median survival times (MST) of GBM patients with high
   HIF1A expression was 12.7 months, while that of patients with low HIF1A
   expression was 15.6 months, showing that GBM patients with high HIF1A
   expression had a poor prognosis, p = 0.001 (Figure [131]2I). Patients
   whose tumors had high expression levels of both TAGLN and HIF1A (MST =
   12.2 months) had worse survival than those with low expression levels
   of both genes (MST = 20 months), p<0.0001 (Figure [132]2J), indicating
   that the effect is not only linked to HIF1A but also to TAGLN.

2.3. TAGLN Combines with HIF1α to Form a Hypoxic Transcriptional Activator
Complex

   Given the central role of the HIF transcription factors in cellular
   responses to hypoxia, patient‐derived GSCs were transduced with
   lentiviruses encoding shRNAs with either a non‐targeting control
   sequence or targeting HIF1A or HIF2A (EPAS1) then cultured under
   hypoxia for 48 h. Upon HIF1α silencing, TAGLN was no longer induced by
   hypoxia (Figure [133]3A ), whereas targeting HIF2α did not alter TAGLN
   expression (Figure [134]3B). HIFs selectively bind to the core binding
   motif 5’‐A/GCGTG‐3’ located in gene promoters or intragenic regions to
   transcriptionally activate the expression of hypoxia‐inducible genes.^[
   [135]^19 ^] To confirm the role of HIF1A in TAGLN regulation, we
   analyzed the TAGLN promoter region, which covered a ≈3 kb tiled array
   surrounding the transcription start site, and predicted three possible
   hypoxic response element‐binding sites (designated as binding sites 1,
   2, and 3) by checking the UCSC Genome Browser Home (Figure [136]3C).
   Chromatin immunoprecipitation coupled with qRT‐PCR (ChIP‐PCR)
   identified HIF1α binding to site 2 in T387 GSCs and sites 2 and 3 in
   T3691 GSCs in the TAGLN promoter and induction of TAGLN expression
   under hypoxia (Figure [137]3D). Next, we explored the effect on HIF1α
   expression in GSCs following TAGLN knockdown or overexpression.
   Immunoblotting indicated that the expression of HIF1α was consistent
   with that of TAGLN (Figure [138]3E; Figure [139]S3A, Supporting
   Information). In GBM patient specimens, TAGLN co‐localized with HIF1α
   on IF (Figure [140]2C; Figure [141]S2D, Supporting Information).

Figure 3.

   Figure 3
   [142]Open in a new tab

   The hypoxic transcription factor‐TAGLN. A,B) IB showed the indicated
   protein levels in GSCs after targeting HIF1α or HIF2α. C) Schematic
   representation of the three predicted HRE‐binding sites in the TAGLN
   promoter. D) ChIP‐qPCR analysis of HRE site occupancy in TAGLN promoter
   (Student's t‐test). E) IB showed protein expression in GSCs after TAGLN
   targeting. F) Co‐IP analysis of endogenous TAGLN (left) or HIF1α
   (right) in GSCs, followed by determination of the expression of HIF1α
   and TAGLN by IB. G) mRNA expression of HIF1A and TAGLN was assessed by
   qRT‐PCR following HIF1A knockdown. H) Expression of TAGLN and HIF1A
   after targeting TAGLN in GSCs was examined by qRT‐PCR. I) TAGLN‐binding
   peak in HIF1A. J) HIF1A motif to which TAGLN binds. K) ChIP‐qPCR of the
   indicated genes in GSCs (Student's t‐test). Values in (D,G,H, and K)
   represent the mean ± SD from three independent experiments (*p < 0.05,
   **p < 0.01, ***p < 0.001, and ****p < 0.0001).

   TAGLN has been detected in the nucleus of colon cancer cells, where it
   interacts with transcription factors and also participates extensively
   in the transcription of genes.^[ [143]^20 ^] Therefore, we hypothesized
   that TAGLN may directly bind HIF1α. We performed reciprocal
   co‐immunoprecipitation (Co‐IP) experiments in GSCs, revealing binding
   between TAGLN and HIF1α (Figure [144]3F). Targeting HIF1A, but not
   HIF2A (EPAS1), reduced TAGLN transcript levels (Figure [145]3G; Figure
   [146]S3B, Supporting Information). Reciprocally, TAGLN regulated HIF1A
   transcription with a >70% reduction in HIF1A transcript levels after
   TAGLN targeting (Figure [147]3H). In gain‐of‐function studies, TAGLN
   overexpression induced elevated HIF1A mRNA levels (Figure [148]S3C,
   Supporting Information). ChIP‐seq showed that TAGLN is bound to the
   HIF1A promoter region in GSCs (Figure [149]3I). Multiple Em for motif
   elicitation analysis showed enrichment of elements within TAGLN‐binding
   sites (Figure [150]3J). We confirmed the presence of a unique TAGLN
   binding site in the upstream promoter of HIF1A (Figure [151]3K).
   Collectively, TAGLN forms a regulatory feedforward circuit with HIF1A
   to support the hypoxic responses of GSCs.

2.4. TAGLN Maintains GSC Self‐Renewal and Tumor Initiation With Hypoxia

   To elucidate TAGLN function in GSCs, we performed loss‐of‐function
   studies by transducing patient‐derived GSCs with either a non‐targeting
   control shRNA sequence (shCONT) or one of two distinct short hairpin
   RNAs (shTAGLN#739 and shTAGLN#752) to knock down TAGLN expression. We
   confirmed that both shTAGLNs reduced TAGLN expression compared with
   shCONT (Figure [152]S4A,B, Supporting Information). Targeting TAGLN
   attenuated the viability of patient‐derived GSCs (Figure [153]4A ;
   Figure [154]S4C, Supporting Information). Sphere formation is a
   surrogate marker of self‐renewal, albeit with caveats. Following TAGLN
   knockdown in GSCs, the tumorsphere formation frequency of GSCs was
   reduced, as assayed by in vitro limiting dilution (Figure [155]4B;
   Figure [156]S4D, Supporting Information). Stable TAGLN knockdown
   reduced the ability of GSCs to form neurospheres (Figure [157]4C,D;
   Figure [158]S4,F, Supporting Information). TAGLN silencing also
   inhibited DNA replication, as revealed by 5‐Ethynyl‐2‐deoxyuridine
   (EdU) incorporation (Figure [159]4E,F; Figure [160]S4G,H, Supporting
   Information). Decreased levels of GSC markers SOX2 and OLIG2 were
   observed in TAGLN‐targeted GSCs compared with controls, accompanied by
   increased expression of GFAP (Figure [161]4G; Figure [162]S4I,
   Supporting Information). Furthermore, TAGLN silencing induced apoptosis
   measured by elevated Cleaved Caspase3 and Cleaved polyadenosine
   5’‐diphosphate‐ribose polymerase (PARP) by IB (Figure [163]4H; Figure
   [164]S4J, Supporting Information). These data indicated that targeting
   TAGLN disrupted GSC maintenance.

Figure 4.

   Figure 4
   [165]Open in a new tab

   Targeting TAGLN impairs GSC self‐renewal and tumorigenicity. A) Cell
   viability assay of GSCs infected with the two shRNAs against TAGLN (Two
   Way ANOVA). B) In the limiting dilution assays, the sphere‐forming
   frequency was examined following TAGLN knockdown (ELDA). C,D)
   Representative images of neurospheres derived from GSCs expressing
   shCONT or C) shTAGLN and D) quantification. Scale bar, 100 µm. E)
   Representative images of the EdU assay after TAGLN knockdown and F) the
   ratio of EdU‐positive cells was calculated. Scale bar, 50 µm. G)
   Expression of the indicated proteins during TAGLN targeting was
   detected by IB. H) IB of apoptotic proteins in GSCs after TAGLN
   knockdown. I) Experimental design. J) Representative images of HE
   staining in the brains of tumor‐bearing mice. The scale bar is 10 mm.
   K) Kaplan‐Meier curves of mice were drawn to determine the burden of
   tumor progression on GSCs expressing shCONT or shTAGLN (log‐rank test,
   n = 9 for each group). L) Representative IHC images of the indicated
   proteins stained on frozen sections of the GBM xenografts. The bar
   represents 50 µm. Values in (A, B, and D) represent the mean ± SD from
   three independent experiments (**p < 0.01, ***p < 0.001, and ****p <
   0.0001).

   In vivo tumor initiation is the gold standard for cancer stem cells. To
   assess GSC dependency on TAGLN in vivo, 2 × 10^4 T387 GSCs transduced
   with lentiviruses encoding either shTAGLN or shCONT were implanted
   intracranially into the frontal lobes of immunocompromised mice.
   Tumor‐bearing brains were harvested after the presentation of the first
   neurological sign in any group (Figure [166]4I). Consistent with our
   findings in vitro, no tumor was found in animals bearing GSCs
   transduced with shTAGLN, which prolonged tumor latency and extended the
   survival of tumor‐bearing mice (Figure [167]4J,K). Analysis of
   tumor‐bearing brains revealed that targeting TAGLN expression in GSCs
   reduced the expression levels of TAGLN, the proliferation marker Ki67,
   and the GSC marker SOX2, but increased levels of the GFAP
   differentiation marker and Cleaved Caspase3 (Figure [168]4L; Figure
   [169]S4K, Supporting Information). Collectively, our data indicate that
   TAGLN is a critical GSC dependency.

2.5. TAGLN‐HDAC2 Promotes GSC Self‐Renewal by Regulating p53 Deacetylation

   To ascertain the mechanism by which TAGLN maintains GSC self‐renewal,
   we performed comparative RNA‐seq on GSCs transduced with shTAGLN or
   shCONT. TAGLN silencing induced extensive changes in GSC gene
   expression (Figure [170]S5A, Supporting Information). Concordant with
   phenotypic effects of targeting TAGLN, the largest subset of
   downregulated genes was primarily involved in cell cycle control and
   the p53 signaling pathway (Figure [171]5A ). Gene set enrichment
   analysis (GSEA) of data from TCGA and CGGA in Gliovis revealed that
   high TAGLN expression correlated with the p53 signaling pathway (Figure
   [172]S5B,C, Supporting Information). To determine the direct molecular
   connections, we immunoprecipitated TAGLN and resolved proteins that
   coprecipitated with TAGLN using mass spectrometry (MS). Two genes
   (HDAC2 and HCFC1) satisfied three criteria: 1) proteins detected in
   TAGLN‐IP MS but not control IP MS; 2) proteins predicted to interact
   with TAGLN with high confidence using HitPredict
   (http://www.hitpredict.org/); and 3) genes within Genecards
   (https://www.genecards.org/) associated with hypoxia, GSCs, p53
   signaling and cell cycle regulation, displayed on a Venn diagram
   (Figure [173]5B). We prioritized HDAC2 because of its strong
   association with the p53 signaling pathway and cell cycle in analysis
   of TCGA GBM/LGG (IDH‐WT) and CGGA GBM (IDH‐WT) datasets (Figure
   [174]S5D–G, Supporting Information), relative to HCFC1 (Figure
   [175]S5H–K, Supporting Information). HDAC2 contributes to tumor growth
   in part by controlling the deacetylation of p53 and transcriptional
   repression of TP53 target genes, while also mediating p53 stability.^[
   [176]^21 ^] We then assessed all HDAC family members that were elevated
   in GBM and functioned as potential oncogenes. Upon TAGLN knock down,
   HDAC2 was specifically decreased showing the relative specificity of
   TAGLN in affecting HDAC2 expression (Figure [177]S5L,M, Supporting
   Information). Genes commonly dysregulated in cancer in the p53/cell
   cycle signaling pathway and HDAC2 were displayed from the RNA‐seq data
   upon targeting TAGLN (Figure [178]5C). GSEA analysis validated the
   association with HDAC2 (Figure [179]S5D–G, Supporting Information),
   indicating that the activated TAGLN/HDAC2‐p53 axis may stimulate GSC
   proliferation under hypoxia. TAGLN and HDAC2 were strongly correlated
   with TP53 and the cell cycle‐related genes CCNE1, CDK2, and CDKN1A in
   the independent CGGA (IDH‐WT) and TCGA GBM/LGG (IDH‐WT) databases
   (Figure [180]S5N,O, Supporting Information). In GBM patient‐derived
   biopsies, TAGLN and HDAC2 each co‐localized with p53 on IF
   (Figure [181]5D,E; Figure [182]S5P, Supporting Information). TAGLN or
   HDAC2 knockdown increased acetylated p53 (ace‐p53) at K382 in GSCs as
   measured by IB, with no perceptible alteration in total p53 or
   phosphorated p53 protein levels (Figure [183]5F and G; Figure
   [184]S5Q–S, Supporting Information). Targeting TAGLN or HDAC2 in GSCs
   induced ace‐p53 and p21 (inhibitors of the CCNE/CDK2 complex) while
   reducing CCNE1 (G1 phase cyclin) and CDK2 (cyclin‐dependent kinase from
   G1 phase to S phase) (Figure [185]5F–H; Figure [186]S5T,U, Supporting
   Information), but did not cause major changes in the expression of p27,
   CCND1, CDK4, or CDK6 (Figure [187]S5R,S, Supporting Information). In
   vivo, histopathological analysis of xenografts derived from GSCs
   transduced with shTAGLN or shCONT revealed loss of TAGLN was associated
   with induction of ace‐p53 and p21 with reduced CCNE1 and CDK2
   (Figure [188]5I,J; Figure [189]S5V,W, Supporting Information). In
   gain‐of‐function studies, TAGLN overexpression in GSCs induced
   alterations of TAGLN/HDAC2‐deacetylated p53‐cell cycle target genes
   expression, opposite to that observed after TAGLN or HDAC2 knockdown,
   as expected (Figure [190]S5X,Y, Supporting Information). Thus,
   TAGLN‐HDAC2 regulates the GSC phenotype in association with p53
   regulation and cell cycle control.

Figure 5.

   Figure 5
   [191]Open in a new tab

   The preservation of the self‐renewal capacity of GSCs benefits from the
   activation of the TAGLN/HDAC2 deacetylated p53 signaling. A) Scatter
   plot of KEGG pathway enrichment analysis of differentially expressed
   genes in the RNA‐seq data after TAGLN knockdown. B) Proteins
   interacting with TAGLN were analyzed in HitPredict (high confidence, n
   = 185), proteins co‐precipitated with TAGLN were detected by mass
   spectrometry (n = 190), and proteins associated with glioma stem cells,
   hypoxia, p53 signaling pathway, and cell cycle in Genecards (> median,
   n = 1362), and overlapping proteins HDAC2 and HCFC1 were identified
   using Venn diagram. C) Heat map showing the differentially expressed
   genes after TAGLN knockdown. D) IF staining of TAGLN and p53 in hGBM
   tissues. Scale bar, 25 µm. E) Expression of HDAC2 and p53 in hGBM
   tissues was observed by IF. Scale bar, 25 µm. F,G) The indicated
   protein levels were examined in cell lysates of GSCs expressing
   shTAGLN, shHDAC2, or shCONT. H) mRNA levels of TAGLN and TAGLN target
   genes were assessed by qRT‐PCR following HDAC2 knockdown. I) IHC
   staining of TAGLN and p53 signaling elements in GBM xenografts with
   TAGLN knockdown. Scale bar, 50 µm. J) Quantification of the indicated
   protein intensities in IHC by measuring IOD (I). Values in (H and J)
   represent the mean ± SD from three independent experiments (*p < 0.05,
   **p < 0.01, ***p < 0.001, and ****p < 0.0001).

2.6. TAGLN Binds to HDAC2 and Mediates its Degradation Primarily Through the
Lysosome

   As the TAGLN interactome revealed binding to HDAC2 localized to the
   nucleus (Figure [192]5B; Figure [193]S5L,M, Supporting Information), we
   analyzed the CGGA GBM (IDH‐WT) dataset, revealing that HDAC2 associated
   with poor prognosis in GBM (Figure [194]S6A, Supporting Information).
   To define the interplay among TAGLN, HDAC2, and HIF1α, we first
   targeted TAGLN and detected a reduction in TAGLN, HDAC2, and HIF1α
   expression (Figures [195]6A and [196]3E; Figure [197]S6B, Supporting
   Information). Reciprocally, HDAC2 knockdown also reduced the levels of
   HDAC2, TAGLN, and HIF1α (Figure [198]6B). We selectively knocked down
   HIF1α expression and detected decreased TAGLN and HDAC2 by IB
   (Figure [199]6C), indicating an interdependent relationship among
   TAGLN, HDAC2, and HIF1α. Consistent with the in vitro observations,
   xenografts derived from GSCs transduced with shTAGLN expressed
   decreased HDAC2, CA9, and HIF1α protein levels measured by IHC relative
   to GSCs transduced with shCONT (Figure [200]6D,E).

Figure 6.

   Figure 6
   [201]Open in a new tab

   TAGLN binds to HDAC2 to form a co‐transcriptional activator complex. A)
   Expression of HDAC2 and TAGLN was detected by IB. B) Expression of
   TAGLN and HDAC2 in GSCs after HDAC2 knockdown was examined by IB. C)
   Expression of HDAC2, TAGLN, and HIF1α during HIF1α targeting was
   detected by IB under hypoxia. D) Representative IHC images of TAGLN,
   HDAC2, CA9, and HIF1α expression in the xenografts. Scale bar, 50 µm.
   E) IOD of HDAC2, TAGLN, CA9, and HIF1α were measured in (D). F) IB
   analysis of TAGLN and HDAC2 in different treatment groups. G,H) IF
   staining of HDAC2 (red) and LAMP1 (lysosomal marker; green) after DMSO
   or chloroquine treatment of TAGLN targeted GSCs. Scale bar, 10 µm. I)
   IHC staining revealed HDAC2, HIF1α, CA9, and TAGLN expression in
   pseudopalisades (upper) or other areas (lower) of GBM. Scale bar,50
   µm. J) Representative images of TAGLN and HDAC2 expression in GBM
   specimens. Scale bar, 25 µm. K) Co‐IP analysis of TAGLN and HDAC2 in
   three GSC cell lines. Values in (E) represent the mean ± SD from three
   independent experiments (*p < 0.05, **p < 0.01, and ***p < 0.001).

   The correlation between TAGLN and HDAC2 expression in GBM cells
   promoted the investigation of molecular mechanisms underlying the
   effects of TAGLN on HDAC2 expression. GSEA revealed that
   TAGLN‐correlated genes were enriched in the lysosomal pathway (Figure
   [202]S6C, Supporting Information). TAGLN‐knockdown GSCs were incubated
   with chloroquine (which has multiple mechanisms of action, including
   interference of lysosomal activity and autophagy) or MG132 (a
   proteasome inhibitor); chloroquine restored HDAC2 downregulation in
   GSCs (Figure [203]6F–H; Figure [204]S6D,E, Supporting Information),
   whereas treatment with MG132 did not affect HDAC2 stability in T387
   GSCs but had an effect on T3691 GSCs (Figure [205]S6F, Supporting
   Information), suggesting that lysosome‐mediated protein degradation
   involved in HDAC2 depletion upon TAGLN targeting. To investigate the
   mechanism by which HDAC2 regulates TAGLN expression, we treated
   HDAC2‐knockdown GSCs with the proteasome inhibitor MG132 or the
   lysosomal inhibitor chloroquine for 2 h and used the same dose of DMSO
   as a control. As a result, MG132 increased the expression of TAGLN,
   while chloroquine caused no noticeable change (Figure [206]S6G,H,
   Supporting Information). Furthermore, the silencing of HDAC2 by two
   independent shRNAs enhanced TAGLN ubiquitylation in T387 and T4121 GSCs
   (Figure [207]S6I, Supporting Information). Overall, these results
   suggest that TAGLN mediates HDAC2 degradation primarily through the
   lysosome, while silencing HDAC2 in GSCs disrupts the stabilization of
   TAGLN, leading to its proteasomal degradation.

   To explore TAGLN and HDAC2 expression in GBM patient surgical
   specimens, we performed IHC staining and observed that HDAC2 (which is
   a hypoxia‐related gene), TAGLN, CA9 (a hypoxic marker), and HIF1α were
   all located in GBM tumor cell nuclei and highly expressed in
   pseudopalisades (upper) (Figure [208]6I; Figure [209]S6J,K, Supporting
   Information). IF analysis confirmed TAGLN and HDAC2 (positively related
   to the GSC markers) localization (Figure [210]6J; Figure [211]S6L,M,
   Supporting Information). In the CGGA (IDH‐WT) and TCGA GBM/LGG (IDH WT)
   datasets, TAGLN and HDAC2 mRNA levels were positively correlated
   (Figure [212]S5N,O, Supporting Information) and patients whose tumors
   were TAGLN ^hi HDAC2 ^hi had a worse prognosis (Figure [213]S6N,O,
   Supporting Information). Based on the co‐expression, we hypothesized
   that TAGLN and HDAC2 directly interacted. Indeed, Co‐IP revealed that
   TAGLN coprecipitated with HDAC2. Reciprocally, HDAC2 coprecipitated
   with TAGLN in the same lysates, suggesting that TAGLN and HDAC2 formed
   a transcriptional regulator complex that sustained each other's
   stability and promoted GSC proliferation (Figure [214]6K). To further
   elucidate the interaction between HIF1α, TAGLN, and HDAC2, we employed
   Co‐IP in normoxia and hypoxia. Our data observed that HIF1α combined
   with the TAGLN/HDAC2 complex under hypoxia, while the binding of HIF1α
   or TAGLN to HDAC2 was not found under normoxia (Figure [215]S6P,
   Supporting Information). Taken together, our data suggest that HIF1α,
   TAGLN, and HDAC2 are mutually regulated, forming complexes that
   maintain the self‐renewal of GSCs under hypoxia.

2.7. Targeting TAGLN/HDAC2 with Sodium Valproate and Natural Borneol
Combination Therapy

   Based on the GSC dependency on TAGLN in vivo, we sought clinically
   efficacious small molecule inhibitors that selectively target
   TAGLN/HDAC2 activity for GBM therapy. Sodium valproate (VPA; structure
   shown in Figure [216]7A ) is an antiepileptic drug, which can
   selectively inhibit the activities of HDAC1 and HDAC2.^[ [217]^22 ^]
   Sodium valproate treatment inhibits GSC proliferation and survival, as
   well as promotes GSC differentiation.^[ [218]^23 ^] However, the exact
   mechanism of its killing effect on GSCs as an HDAC2 inhibitor remains
   to be elucidated. To confirm the therapeutic effects of sodium
   valproate on GSCs, we treated GSCs with varying concentrations of
   sodium valproate and found that the protein levels of HDAC2, but not
   HDAC1, was attenuated at 4 mM, which was accompanied by decreased TAGLN
   expression (Figure [219]7B; Figure [220]S7A, Supporting Information).
   Under hypoxia, we applied VPA to GSCs, and Co‐IP was performed to
   verify the protein interaction between TAGLN and HDAC2. Treatment of
   T387 and T4121 GSCs with 4mM VPA significantly weakened the ability of
   TAGLN to bind to HDAC2 (Figure [221]S7B, Supporting Information).
   Similar to results from genetic targeting of TAGLN, GSC cell viability
   was reduced with increasing concentrations of sodium valproate
   treatment (Figure [222]7C), with reduced GSC marker expression levels
   and induction of GSC differentiation and apoptosis (Figure [223]7D,E).
   Sodium valproate treatment inhibited GSC tumorsphere formation
   (Figure [224]7F,G). As valproate has been considered an HDAC2
   inhibitor, we investigated the role of TAGLN in tumor cell response by
   rescuing treatment effects with TAGLN overexpression (TAGLN‐OE) in
   GSCs. TAGLN‐OE partially rescued GSC self‐renewal following sodium
   valproate treatment (Figure [225]7H). Likewise, sodium valproate
   suppressed the activity of the TAGLN‐HDAC2‐p53 axis in GSCs
   (Figure [226]7I), which TAGLN‐OE partially rescued in (Figure [227]S7C,
   Supporting Information). Thus, valproate targets TAGLN/HDAC2 to inhibit
   GSCs, suggesting that blocking TAGLN/HDAC2 may result in a high
   therapeutic index for GBM treatment.

Figure 7.

   Figure 7
   [228]Open in a new tab

   Sodium valproate blocks TAGLN/HDAC2 to promote P53 acetylation and the
   self‐renewal of GSCs. A) Schematic representation of the structural
   formula of sodium valproate. B) T387 and T3691 GSCs were treated with 2
   or 4 mM sodium valproate for 48 h in hypoxia, and the levels of TAGLN,
   HDAC1, and HDAC2 were analyzed by IB. C) Viability of GSCs over a 3‐day
   period following sodium valproate treatment (Two Way ANOVA). D) GSCs’
   markers expression detected by IB in sodium valproate‐treated GSCs. E)
   Expression of apoptotic proteins in GSCs was analyzed following
   treatment with sodium valproate. F) Representative images and G)
   quantification of neurosphere formation in GSCs treated with sodium
   valproate (Student's t‐test). Scale bar, 100 µm. H) Cell viability
   assays after sodium valproate or normal saline treatment in GSCs with
   or without TAGLN overexpression (Student's t‐test). I) p53 signaling
   elements were assessed following sodium valproate treatment. J)
   Schematic of the structural formula of natural borneol. K) Experimental
   design. L) Kaplan‐Meier curves of tumor‐bearing mice treated with the
   indicated treatments (Log‐rank test, n = 5 for each group). M)
   Mechanistic model diagram of TAGLN regulating GSC stability. Under
   normoxia, TAGLN remained at a low level, and the transcriptional
   activity of HDAC2, which interacted with TAGLN, was reduced and
   degradation was enhanced, resulting in a decrease in the self‐renewal
   of GSC. Hypoxia can induce upregulation of TAGLN expression, by forming
   a transcriptional complex with HIF1α and upregulating the level of
   deacetylation of p53 after binding to HDAC2, thereby promoting cell
   cycle progression and maintaining the stability of GSC. Values in (C,
   G, and H) represent the mean ± SD from three independent experiments
   (*p < 0.05, **p < 0.01, and ****p < 0.0001).

   Many potentially effective agents against GBM in vitro fail in vivo due
   to constraints of the blood‐tumor barrier (BTB). To investigate whether
   sodium valproate could suppress xenograft growth in vivo, we randomized
   tumor‐bearing mice after 14 days into groups intraperitoneally injected
   with normal saline (NS) as the vehicle control or different
   concentrations of sodium valproate for an additional 14 days.
   Preliminary experiments did not yield differences in survival of
   tumor‐bearing mice and the lack of changes in HDAC2 and TAGLN
   expression among treatment groups as a pharmacodynamic study suggested
   that ineffective drug concentrations were achieved (Figure [229]S7D,E,
   Supporting Information). Considering the systemic toxicity and drug
   delivery efficiency of sodium valproate, we initiated treatment earlier
   (day 7) of tumor‐bearing mice with sodium valproate at 150 mg kg^−1 for
   14 days and found extension of tumor latency, but did not achieve an
   ideal therapeutic effect (Figure [230]S7F, Supporting Information).
   Based on the limitations in drug delivery as a limitation in brain
   tumor treatment, we considered options to improve valproate efficacy.
   Natural borneol improves the delivery efficiency of chemotherapeutic
   drugs by increasing the permeability of the BTB^[ [231]^24
   ^](Figure [232]7J). To assess the combinatorial antitumor effects of
   valproate and natural borneol in vivo, we initiated treatment of
   GSC‐derived xenografts one week after tumor formation with borneol (25
   mg kg^−1) or paraffin liquid (PL) control administered by gavage; 30
   minutes later, mice were treated with sodium valproate (150 mg kg^−1)
   or NS (Figure [233]7K). The combination of natural borneol and sodium
   valproate prolonged the survival of tumor‐bearing mice and increased
   the VPA levels in xenografts compared with sodium valproate monotherapy
   (Figure [234]7L; Figure [235]S7G, Supporting Information). Compared to
   NS control or sodium valproate monotherapy, the combination of natural
   borneol and sodium valproate decreased TAGLN, HDAC2, CCNE1, CDK2, and
   Ki67 while increasing ace‐p53, p21, and Cleaved Caspase3, but did not
   affect the expression of p53, CDK4, CCND1, CDK6, p27, and P‐p53 (S392)
   (Figure [236]S7H,I, Supporting Information). Collectively, sodium
   valproate represents a potentially useful GBM therapy due to targeting
   HDAC2‐TAGLN‐p53 with improved efficacy through combination with
   borneol.

3. Discussion

   GBM remains lethal and incurable despite maximal therapy. GBMs have
   been extensively characterized genetically, but precision oncology has
   offered little benefit in neuro‐oncology. One contributing factor to
   GBM treatment failure is likely cellular heterogeneity, including the
   presence of self‐renewing GSCs that commonly resist conventional
   therapy and augment angiogenesis, invasion, and immune escape. Here, we
   identified TAGLN as a novel target in GSCs that serves to bridge
   histone deacetylase function and p53 to reveal improved therapeutic
   targeting. TAGLN is preferentially expressed by GSCs, suggesting a
   potentially high therapeutic index. In contrast to previous reports
   that TAGLN serves as a suppressor during tumorigenesis initiation in
   other cancers^[ [237]^12 , [238]^13 , [239]^25 ^], TAGLN acts as a
   functional marker for tumor stemness in GBM.

   Hypoxia promotes tumor growth and therapeutic resistance.^[ [240]^8 ,
   [241]^26 ^] The expression of TAGLN, a gene involved in hypoxic
   immunity, predicts a poor prognosis of gastric cancer.^[ [242]^27 ^]
   HIFs promote GSC self‐renewal and tumor angiogenesis.^[ [243]^9 ,
   [244]^28 ^] HIF1α, but not HIF2α, regulated hypoxic induction of TAGLN
   expression (Figure [245]3A,B,G; Figure [246]S3B, Supporting
   Information). As inhibition of TAGLN reduced HIF1α transcription and
   translation (Figure [247]3H,E), TAGLN may also function as a
   transcriptional regulator. Although TAGLN does not have a nuclear
   localization sequence, it binds to transcription factors and enters the
   nucleus. As predicted, TAGLN bound at the binding motif predicted by
   ChIP‐seq on the HIF1α promoter (Figure [248]3I–K). To the best of our
   knowledge, the present study is the first to show a hypoxic
   transcription complex of TAGLN‐HIF1α in GSCs.

   Dysregulation of TAGLN contributes to tumor growth.^[ [249]^13 ,
   [250]^29 ^] In this study, we found that TAGLN was critical to maintain
   GSCs (Figure [251]4). We found interactions between TAGLN and the p53
   axis contributing to GSC proliferation and survival. p53 and PTEN
   regulate TAGLN to inhibit bladder cancer cell proliferation.^[ [252]^12
   ^] As a result, we now broaden the molecular interactome with TAGLN to
   add HDAC2 regulating p53 deacetylation at K382, in turn, mediating
   acceleration of cell cycle progression. HDAC2 is involved in p53
   deacetylation to prevent proliferation arrest in tumor cells.^[
   [253]^30 ^]

   Genetic targeting of TAGLN reduced in vivo tumor growth in GBM,
   resulting in an increase in survival of tumor‐bearing mice, supporting
   TAGLN as an effective target (Figure [254]4K). Direct TAGLN inhibitors
   have not been reported, whereas effective drugs that block the
   TAGLN/HDAC2‐p53 axis offer the possibility of therapy for GBM patients.
   Clinical retrospective studies have shown that the antiepileptic drug
   sodium valproate, alone or in combination with radiotherapy or
   chemotherapy, can suppress seizure activity and extend the survival of
   GBM patients^[ [255]^23 , [256]^31 ^], which may be explained by its
   additional pharmacodynamic properties, particularly its HDAC2
   inhibition. Although prior studies have shown that sodium valproate is
   an effective glioma treatment^[ [257]^32 ^], the mechanism by which
   sodium valproate inhibits the growth of GSCs remains unclear. Sodium
   valproate inhibited GSC proliferation, phenocopying silencing
   TAGLN/HDAC2, suggesting the specificity of sodium valproate inhibiting
   TAGLN/HDAC2.

   GBMs display regionally intact BTB^[ [258]^33 ^], which hampers drug
   penetration, accumulation, and uptake in GBM tumors, limiting
   chemotherapeutic efficacy. As an anti‐epileptic drug, valproate is
   brain penetrant but the concentrations may be insufficient to reach
   therapeutic concentrations in the tumor, but poses toxicity to normal
   tissues. Natural borneol can open the blood‐brain barrier and enhance
   the penetration of drugs into the brain.^[ [259]^24 , [260]^34 ^] Wang
   and co‐workers confirmed that natural borneol monotherapy induced GBM
   cell apoptosis by regulating HIF1α expression.^[ [261]^35 ^] Here, we
   demonstrate that sodium valproate combined with natural borneol
   prolonged the survival of tumor‐bearing mice and increased the VPA
   levels in xenografts (Figure [262]7L; Figure [263]S7G, Supporting
   Information).

   In conclusion, our study provides a novel insight into
   TAGLN/HDAC2‐promotion of cell cycle progression through deacetylation
   of p53 at K382 under hypoxia and offers a targeted approach to improve
   the therapeutic effect of GBM. TAGLN may be an independent predictor of
   overall survival in GBM and hypoxic transcriptional regulators. The
   combination of TAGLN/HDAC2 blocker and natural borneol for treating GBM
   patients reveals that combined traditional Chinese and Western medicine
   enhance the delivery of sodium valproate into GBM tumors and improve
   preclinical efficacy. These results support the development of
   TAGLN/HDAC2 blockers as anti‐GSC treatments and open the possibility of
   generalizing these findings to the therapy of other malignant tumors.

4. Experimental Section

Cell Lines and Cell Culture

   All cell lines were dissociated and sorted from GBM patient surgical
   samples and xenografts, as previously described.^[ [264]^36 ^] All GSC
   and NPC cell lines were generously gifted by Dr. Shideng Bao in the
   United States. GSCs and NPCs were cultured in Neurobasal‐A medium
   supplemented with 10% B27 (Cat# 354277, Gibco), 20 ng mL^−1 EGF (Cat#
   105‐04, PrimeGene), 20 ng mL^−1 bFGF (Cat# 104‐02, PrimeGene), 1 mM
   sodium pyruvate (Cat# 11360070, Gibco), 2 mM l‐glutamine (Cat#
   25030081, Gibco), and 100 U mL^−1 penicillin‐streptomycin (Cat#
   15140‐122, Gibco). Differentiated GSCs (DGCs), 293T (ATCC), and glioma
   cells were cultured in DMEM supplemented with 10% FBS (Cat# A3161001C,
   Gibco). Cells were cultured in a humidified 37 °C, 5%CO[2], 21%, or 1%
   O[2] incubator for no >2 months after recovery. Moreover, all cell
   lines were shown to be free of mycoplasma infection (Cat# C0297S,
   Beyotime Biotechnology).

Human GBM tissues

   This study was approved by the Medical Ethics Committee, Tongji Medical
   College, Huazhong University of Science and Technology, Wuhan, China
   (serial no. [2021] IEC (A244)). Samples were collected from each
   patient after obtaining their written informed consent. All the patient
   studies were conducted in accordance with the principles of the
   Declaration of Helsinki. The clinical information of the patients is
   detailed in Table [265]S1 (Supporting Information).

Intracranial Tumor Assay and Medicine Treatment

   Healthy male and female nude BALB/c mice were purchased from the
   Beijing Vital River Laboratory Animal Technology Co., Ltd (Beijing,
   China) and were housed in peculiar pathogen‐free cages. Five nude mice
   in each cage were checked daily by certified veterinarians and
   laboratory personnel and minimized all discomfort and pain in mice.
   Nude mice were randomly divided into groups for all experiments. GSCs
   were transduced with shCONT or shTAGLN by lentiviral infection for 48
   h. The GSC suspension (2 × 10^4 cells for transducing GSC T387) was
   intracranially injected into the brains of 4‐5‐week‐old nude mice as
   described previously. For the treatment experiments, tumor‐bearing mice
   were gavaged with natural borneol (25 mg kg^−1, i.g., China Institute
   for the Control of Pharmaceutical and Biological Products) or paraffin
   liquid (25 mg kg^−1, i.g., Cat# 8042‐47‐5, Aladdin) on day 7, half an
   hour later, intraperitoneally injected with sodium valproate (150 or
   300 mg kg^−1, i.p., Cat# HY‐10585A, MCE) or normal saline for 14 or 21
   days. To compare the growth of the xenografts between the different
   groups, the brain tissue of the mice in the experimental group was
   harvested simultaneously when the control group mice developed
   neurological symptoms and performed with H&E staining. Mice were
   anesthetized by intraperitoneal injection of a mixture of ketamine (100
   mg kg^−1) and xylazine (10 mg kg^−1) before all surgical procedures.
   After surgery, the mice were euthanized by cervical dislocation under
   deep anesthesia. All mouse studies were performed using protocols
   approved by the Experimental Animal Ethics Committee, Huazhong
   University of Science and Technology (serial no. S1819).

Plasmid Generation, Lentiviral Transfection, and Reagent Treatment

   TAGLN overexpression plasmid was generated by cloning the coding
   sequence (CDS) of the human TAGLN gene into the pLVX‐puro‐3 × FLAG
   vector (Cat# 632164, Clontech). The shRNA sequence was shown in Table
   [266]S2 (Supporting Information) and shRNAs for HIF1A, EPAS1, TAGLN,
   and HDAC2 were generated from pLKO.1‐puro (Cat# SHC002, Sigma‐Aldrich).
   Packaging vectors psPAX2 (5 µg) (RRID: Addgene_12260, Addgene), and
   p‐CMV‐VSV‐G (5µg) (Cat# 8454, Addgene) were co‐transfected with
   lentiviral vectors expressing non‐target control shRNA, specific shRNA,
   or full‐length TAGLN (5µg) into 293T cells for lentivirus production.
   Lentiviral particles were obtained 48 h post‐transfection by filtering
   with a 0.45‐µm filter and then used to treat GSCs. 48 h after the virus
   infection of GSCs, the medium was changed and GSCs were treated with
   puromycin (2 mg mL^−1, Cat# A1113803, Gibco) for 24 h.

   For cell reagent treatments, transfected GSCs were cultured for 48 h
   under hypoxic conditions and treated with lysosome or the proteasome
   inhibitor, chloroquine (10 mM, Cat# HY‐17589A, MCE) or MG132 (10 µM,
   Cat# HY‐13259, MCE) for 2 h. GSCs were co‐incubated with varying
   concentrations of sodium valproate for the indicated time under hypoxic
   conditions.

Hematoxylin‐eosin (HE), Immunohistochemistry (IHC), and Immunofluorescence
(IF)

   For HE and IHC staining, paraffin sections of GBM tissue or xenografts
   were first placed in a 62 °C oven for 1–2 h. Sections were then
   deparaffinized with xylene, followed by successive incubation with
   100%, 95%, 75%, 50% ethanol, and finally hydration. Sections were
   heat‐induced antigen retrieval with sodium citrate or EDTA buffer
   according to the antibody instructions. Sections were washed three
   times with PBS and incubated with an appropriate amount of 3% H[2]O[2]
   for 10 min at room temperature, washed three times with PBS, and
   blocked with 5% serum of the corresponding species for 1 h at room
   temperature. Next, the primary antibody was added and incubated
   overnight at 4 °C. The next day, sections were washed three times with
   PBS and incubated with secondary antibodies for 30 min at 37 °C,
   followed by DAB detection with horseradish peroxidase (HRP) conjugate
   and counterstaining of nuclei with hematoxylin. Quantitative analysis
   was performed by evaluating the mean integrated optical density using
   Image‐Pro Plus (IPP) software.

   For IF staining, frozen GBM tissue was embedded with OCT compound and
   sectioned. Frozen sections were first fixed with 4% PFA for 20 min,
   rinsed three times with PBS, and incubated with 0.5% Triton‐X100 (Cat#
   T8200, Solarbio) for 20 min. Sections were rinsed three times with PBS,
   blocked with 5% normal donkey serum (NDS) for 1 h, and then incubated
   with primary antibody overnight at 4°C. The next day, after three
   washes with PBS, samples were performed with the Donkey anti‐Rabbit IgG
   (H+L) Cross‐Adsorbed Secondary Antibody, Alexa Fluor 546 (Cat# A10040,
   RRID:AB_2534016, Thermo Fisher Scientific), and the Donkey anti‐Mouse
   IgG (H+L) Cross‐Adsorbed Secondary Antibody, Alexa Fluor 488 (Cat#
   A21202, RRID:AB_141607, Thermo Fisher Scientific) for 1 h. After three
   washes in PBS, the nuclei were counterstained with DAPI (Cat# D1306,
   Thermo Fisher Scientific) and sealed with antifade. Images were taken
   with an Olympus microscope (Olympus FV1000, Japan).

   IHC was performed using antibodies against anti‐TAGLN (Cat# ab14106,
   RRID:AB_443021, Abcam), anti‐SOX2 (Cat# ab171380, RRID:AB_2732072,
   Abcam), anti‐HIF1α (Cat# 66730‐1‐Ig, RRID: AB_288208, Proteintech),
   anti‐CA9 (Cat# NB100‐417, RRID:AB_10003398, NOVUS), anti‐GFAP (Cat#
   80788, RRID:AB_2799963, Cell Signaling Technology), anti‐Ki67(Cat#
   27309‐1‐AP, RRID:AB_2756525, Proteintech), anti‐Cleaved Caspase3 (Cat#
   9664, RRID:AB_2070042, Cell Signaling), anti‐HDAC2 (Cat# ab219053,
   Abcam), anti‐Acetyl‐p53(K382) (Cat# 2525, RRID:AB_330083, Cell
   Signaling Technology), anti‐p21 (Cat# 2947, RRID:AB_823586, Cell
   Signaling Technology), anti‐CCNE1 (Cat# ET1612‐16, HUABIO), anti‐CDK2
   (Cat# YT0832, Immunoway biotechnology), anti‐CCND1 (Cat# 26939‐1‐AP,
   RRID:AB_2880691, Proteintech), anti‐p27 (Cat# 3686, RRID:AB_2077850,
   Cell Signaling Technology), and anti‐CDK4 (Cat# 12790, RRID:AB_2631166,
   Cell Signaling Technology).

   IF antibodies against rabbit anti‐TAGLN (Cat# ab14106, RRID:AB_443021,
   Abcam), mouse anti‐SOX2 (Cat# ab171380, RRID:AB_2732072, Abcam), mouse
   anti‐HIF1α (Cat# 66730‐1‐Ig, RRID: AB_288208, Proteintech), mouse
   anti‐CA9 (Cat# YM3076, Immnuoway), mouse anti‐OLIG2 (Cat# MABN50,
   RRID:AB_10807410, Millipore), mouse anti‐CD15 (Cat# 4744,
   RRID:AB_1264258, Cell Signaling), mouse anti‐CD133 (Cat# BF0403,
   RRID:AB_2833933, Affinity Biosciences), rabbit anti‐HDAC2 (Cat#
   ab219053, Abcam), mouse anti‐HDAC2 (Cat# sc‐9959, RRID:AB_627704, Santa
   Cruz), mouse anti‐LAMP1 (Cat# ab25630, RRID:AB_470708, Abcam) and mouse
   anti‐p53 (Cat# sc‐126, RRID:AB_628082, Santa Cruz) were used.

Cell Viability Assay

   GSCs were seeded into 96‐well plates at a density of 1000 cells per
   well in four replicate wells. Plates were removed on days 0, 1, 3, and
   5 post‐inoculation under hypoxia. Cell viability was quantified using
   the CellTiter‐Glo Luminescent Assay (Cat# G9242, Promega) as previously
   described.^[ [267]^10 ^] All the data were normalized to day 0.

Sphere Formation

   Briefly, the indicated GSCs (1×10^3 cells per well) were seeded in
   triplicate in 24‐well plates under hypoxic conditions, and the total
   number of neurospheres in each well was quantified after 3 days.

Limiting Dilution Assay

   Digested GSCs were plated at an increasing number of 1, 10, 20, 30, 40,
   or 50 cells per well in 96‐well plates with 6 repetitions, and each
   well was checked and quantified the percentage of wells with
   tumorspheres after 5 days under hypoxia (described in^[ [268]^10 ^]).
   p‐values were computed by Extreme Limiting Dilution Analysis (ELDA)
   software available at http://bioinf.wehi.EdU.au/software/elda/.

EdU Incorporation Analysis

   EdU assays were performed by following the manufacturer's protocol as
   previously described.^[ [269]^10 ^] Spheroids were first incubated with
   EdU (Cat# [270]C10310‐1, RIBOBIO) for 2 h and then fixed with 4%
   paraformaldehyde. The spheroids were incubated with 0.5% Triton‐X100,
   treated with 1 × Apollo staining reaction solution, and stained with
   DAPI. At least 10 GSCs neurospheres per group were examined for
   EdU‐positive cells.

Quantitative Real‐Time PCR (qRT‐PCR)

   Total RNA was extracted using the TRIzol Total RNA Extraction Reagent
   (Cat# 10606ES60, YEASEN). 1 µg RNA was reverse transcribed into cDNA
   using the Hifair III 1st Strand cDNA Synthesis SuperMix (Cat#
   11137ES60, YEASEN). qRT‐PCR was performed using Hieff qPCR SYBR Green
   Master Mix (Cat# 11203ES03, YEASEN). All values were obtained in
   triplicate. The primers used for qRT‐PCR were designed using
   PrimerBank. The primers used for human samples are listed in Table
   [271]S3 (Supporting Information).

Immunoblotting (IB)

   Lysed cells on ice using RIPA buffer (50 mM Tris‐HCl, pH 8.0, 150 mM
   NaCl, 0.5% NaCl, 1% NP‐40, 0.1% SDS) containing protease/phosphatase
   inhibitors for 15 min, pipetting every 5 min. The mixture was then
   centrifuged at 12 000 rpm for 15 min at 4 °C, the supernatant was
   collected and protein lysates were quantified by BCA Protein
   Concentration Assay Kit (Cat# BL521A, Biosharp). Lysates were boiled in
   5 × protein loading buffer for 8 min and 40 µg of protein was separated
   on SDS‐PAGE Bis‐Tris gels and transferred to PVDF membrane (Cat#
   [272]SEQ00010, EMD Millipore). After blocking the membrane with 5% skim
   milk for 1 h at room temperature, the designated primary antibodies
   were incubated overnight on the corresponding molecular weight bands at
   4 °C. Specific antibodies against TAGLN (Cat# ab14106, RRID:AB_443021,
   Abcam), TAGLN (Cat# GTX101608, RRID: AB_1952134, GeneTEX), SOX2 (Cat#
   ab171380, RRID:AB_2732072, Abcam), HIF1α (Cat# 66730‐1‐Ig, RRID:
   AB_2882080, Proteintech), HIF2α (Cat# NB100‐122, RRID:AB_10002593,
   NOVUS), CA9 (Cat# NB100‐417, RRID:AB_10003398, NOVUS), GFAP (Cat#
   80788, RRID:AB_2799963, Cell Signaling Technology), Cleaved Caspase3
   (Cat# 9664, RRID:AB_2070042, Cell Signaling Technology), Cleaved PARP
   (Cat# 5625, RRID:AB_10699459, Cell Signaling Technology), HDAC2 (Cat#
   sc‐9959, RRID:AB_627704, Santa Cruz), HDAC1 (Cat# sc‐81598,
   RRID:AB_2118083, Santa Cruz), Ace‐p53(K382) (Cat# 2525, RRID:AB_330083,
   Cell Signaling Technology), p21 (Cat# 2947, RRID:AB_823586, Cell
   Signaling Technology), CCNE1 (Cat# 4129, RRID:AB_2071200, Cell
   Signaling Technology), CDK2 (Cat# 2546, RRID:AB_2276129, Cell Signaling
   Technology), CCND1 (Cat# 2978, RRID:AB_2259616, Cell Signaling
   Technology), p27 (Cat# 3686, RRID:AB_2077850, Cell Signaling
   Technology), CDK4 (Cat# 12790, RRID:AB_2077850, Cell Signaling
   Technology), CDK6(Cat# 3136, RRID:AB_2229289, Cell Signaling
   Technology), P‐p53(Ser392) (Cat# 9281, RRID:AB_331462, Cell Signaling
   Technology), p53 (Cat# sc‐126, RRID:AB_628082, Santa Cruz), DDDDK‐tag
   (Cat# AE005, RRID:AB_2770401, Abclonal), P‐p53(Ser15)(Cat# 28961‐1‐AP,
   RRID:AB_2881236, Proteintech), GAPDH (Cat# sc‐47724, RRID:AB_627678,
   Santa Cruz) and α‐Tubulin (Cat# AC012, RRID:AB_2768341, Abclonal) were
   used for the immunoblotting analysis.

   The next day, incubated with the peroxidase‐affiniPure goat anti‐rabbit
   IgG(H+L) (Cat# 111‐035‐003, RRID:AB_2313567, Jackson ImmunoResearch
   Laboratories) or the peroxidase‐conjugated goat anti‐mouse IgG(H+L)
   (Cat# 115‐035‐003, RRID:AB_10015289, Jackson ImmunoResearch
   Laboratories). ECL chemiluminescence detection reagent (Cat# PI34080,
   Thermo Fisher Scientific) was added for imaging, and band analysis was
   performed using Image Lab Software.

Co‐Immunoprecipitation Assay (Co‐IP)

   GSCs were collected after hypoxia and lysed in Pierce IP Lysis Buffer
   (Cat# 87788, Thermo Fisher Scientific) supplemented with HaltTM
   Protease/Phosphatase Single‐Use Inhibitor Cocktail (Cat#78442, Thermo
   Fisher Scientific). After lysis on ice for 15 min, the supernatant was
   obtained by centrifugation. 20 µL protein A/G agarose (Cat# sc‐2003,
   RRID:AB_10201400, Santa Cruz) was washed with 0.1% Triton‐X100, PBS,
   and lysis buffer to remove nonspecifically bound proteins from the
   beads. Total protein lysate (800 µg) was incubated with 2 µg primary
   antibody for 1‐2 h at 4 °C. Then it was mixed with the above
   supernatant‐removed beads and incubated overnight at 4 °C on a vertical
   shaker. After four washes with PBS the next day, the binding components
   were eluted by boiling with 2 × protein loading buffer and an
   immunoblotting assay was performed.

   Rabbit anti‐TAGLN (Cat# GTX101608, 1:500, GeneTEX), rabbit anti‐HDAC2
   (Cat# ab219053, 1:30, Abcam), mouse anti‐HIF1α (Cat# ab1,
   RRID:AB_296474, 1:200; Abcam), IgG rabbit, and IgG mouse were used for
   IP.

Mass Spectrometry Assay

   GSCs were lysed in IP lysis buffer and the supernatant was incubated
   with TAGLN (Cat# GTX101608, 1:500, GeneTEX) or IgG (Cat# ab218427,
   Abcam). Added 20 µL of the protein A/G agarose and incubated overnight
   at 4 °C. The proteins were separated by SDS‐PAGE and stained using
   InstaBlue Protein Staining Solution (Cat# B8226, APExBIO). The visible
   band was cut, treated with acetonitrile (ACN) until it turned white,
   and then trypsinized. Peptides were examined using a Q Exactive HF‐X
   mass spectrometer (Thermo Fisher). The top 40 most abundant precursors
   were subjected to high‐energy collisional dissociation (HCD) and
   examined using MS/MS.

Chromatin Immunoprecipitation (ChIP) Assay

   ChIP assay was performed according to the instructions for the EZ‐Magna
   ChIP A/G Chromatin Immunoprecipitation Kit (Cat# 17‐10086, Millipore
   Sigma). In brief, hypoxia‐treated GSCs suspensions were collected and
   fixed with a final concentration of 1% formaldehyde (Sigma‐Aldrich) for
   10 min at room temperature, terminated with glycine, and washed three
   times with cold PBS. The cell suspension was centrifuged and the
   supernatant was removed, and cell lysis buffer containing protease
   inhibitor cocktail II was added to lyse cytoplasmic proteins. After 15
   min of lysis on ice, the mixture was centrifuged at low speed to remove
   the supernatant, and the pellet was resuspended in nuclear lysis
   buffer. After incubation on ice for 5 min, ultrasonic fragmentation of
   genomic DNA into fragments of ≈500—1000 bp, the mixture was
   centrifuged, and the supernatant was collected.

   Two aliquots of the sonicated mixture were supplemented with ChIP
   dilution buffer to a volume of 500 µL and taken out 1% of the mixture
   as “input”. HIF1α or TAGLN antibody was added to one and IgG antibody
   to another and incubated at 4 °C for 1—2 h. Then, 20 µL of fully
   suspended magnetic protein A/G beads was added to each aliquot and
   incubated overnight on a vertical shaker at 4 °C. The next day after
   centrifugation, the supernatant was removed and the beads were washed
   sequentially with low‐salt immune complex washing buffer, high‐salt
   immune complex washing buffer, LiCl immune complex washing buffer, and
   TE buffer. After centrifugation, resuspend in 100 µL ChIP elution
   buffer and incubated with proteinase K for 2 h at 62 °C. After removing
   the magnetic beads with a magnetic stand and purifying the DNA
   fragments. The relative enrichment of the indicated genes was detected
   by qRT‐PCR.

ChIP‐Seq And Data Processing

   ChIP‐seq libraries were constructed with ChIP and sequenced using the
   Illumina platform with pair‐end reads of 100–300 bp. Removed the splice
   sequence in ChIP‐seq raw data with Cutadapt, the low‐quality clean data
   were taken out with Trimmomatic,^[ [273]^37 ^] and FastQC^[ [274]^38 ^]
   was used for quality control and filtering. Cleaned reads were mapped
   against the human reference genome hg38 using Bowtie2^[ [275]^39 ^]
   with default settings. Samtools software converted the SAM format to
   BAM. MACS2^[ [276]^40 ^] was performed to search the peak value of the
   ChIP‐seq, and input DNA was used as the control.

   The irreproducible discovery rate (IDR)^[ [277]^41 ^] software was used
   to select reproducible peaks with high confidence between two
   duplicates for subsequent analysis. R‐package ChIPseeker^[ [278]^42 ^]
   (ENSEMBL101) was used to annotate the peaks. MEME^[ [279]^43 ^]‐ChIP
   was used for motif discovery and identification of the combined peaks.
   Differential peak analysis of each group was performed using the
   MAnorm.^[ [280]^44 ^]

RNA‐Seq

   Digested GSCs (T387) were infected with shCONT or shTAGLN lentivirus
   for two days. Three biological replicates were used for each sample
   group. Total RNA was extracted from cells using TRIzol reagent (Cat#
   AM9738, Thermo Fisher Scientific). A small portion of each RNA sample
   was subjected to qRT‐PCR analysis to determine the knockdown efficiency
   before RNA‐seq analysis. Library preparation and next‐generation
   sequencing were performed using the NovoGene software. Data processing
   was performed on a public server (https://cn.novogene.com/).

LC‐MS Conditions

   In the VPA assay, LC‐MS was performed using an UltiMate 3000 RS
   chromatograph (Thermo Fisher, USA) connected to the Q Exactive mass
   spectrometer (Thermo Fisher, USA). The chromatographic elution was
   performed on an agela venusil C18 plus column (50 × 2.1 mm, 5 µm) with
   a column temperature of 40 °C. Aqueous phase A was 10mM ammonium
   acetate, and organic phase B was methanol. The elution gradient was
   0–2 min, 80% A; 2‐3.5 min, 80% A⟶5% A; 3.5‐5.8 min, 5% A; 3.5‐3.8 min,
   5% A⟶80% A; 3.8‐5 min and 80% A at a flow rate of 0.4 mL min^−1. The
   injection volume was 5 µL. The mass spectra was performed using an
   electrospray ionization source (ESI) with negative ion scanning
   selected reaction monitoring mode (SRM). The ESI‐MS operating
   parameters used in the method were set as follows: ion spray voltage of
   4000 V, sheath gas 50 arb, pilot gas 15 arb, temperature 320 °C, and
   collision gas. After weighing the sample, 1 mL of methanol was added,
   vortex mixed well, and grind with grinding beads for 5 min. Centrifuged
   at 13 000 rpm for 10 min and collected the supernatant with 0.22 µM
   filter membrane filtration and collected filtrate for analysis. The
   chromatogram collection and integration of VPA were processed by the
   software Xcalibur 4.0 (Thermo Fisher), and linear regression was
   performed with a weighting coefficient of 1/x ^2.

Bioinformatics Analyses

   The nucleotide sequence of TAGLN upstream 3000 was determined from UCSC
   (http://www.genome.ucsc.EdU/) and the binding motif of HIF1A was
   predicted by JASPAR (https://jaspar.genereg.net/). Expression,
   correlation, and survival of TAGLN, HDAC2, and related genes were
   analyzed using Timer 2.0 (http://timer.cistrome.org/), The Cancer
   Genome Atlas (TCGA) (https://portal.gdc.cancer.gov/), Chinese Glioma
   Genome Atlas (CGGA) (http://www.cgga.org.cn/), and GLIOVIS
   (http://gliovis.bioinfo.cnio.es/). RNA‐seq expression profiles from the
   GBM marginal zone and pseudopalisade region were obtained from the Ivy
   Glioblastoma Atlas Project (http://glioblastoma.alleninstitute.org/).

Quantification and Statistical Analysis

   All data were presented as the mean ± SD. Three biological replicates
   were subjected to all in vitro cell experiments. All statistical
   analyses were performed using GraphPad Prism software. Differential
   analysis of the expression profiles of the rim region and
   pseudopalisade necrosis obtained by RNA‐Seq was performed using the
   limma package in R. |Log[2]FC|>1 and/or p<0.05 were considered
   statistically significant. The significance of the differences between
   the two groups of data was assessed using the Mann‐Whitney U test for
   non‐parametric data and a two‐tailed unpaired Student's t‐test for
   parametric data. Pearson's correlation test was used to examine the
   relationship between variables in the tumor tissue. The significance of
   the Kaplan‐Meier curves was assessed using a log‐rank test comparing
   the different patient or mouse groups. Experimental sample numbers (n)
   and p‐values are shown in the figures, figure legends, and results
   section.

Conflict of Interest

   The authors declare no conflict of interest.

Author Contributions

   X.Y., J.N.R., and H.L. conceived and supervised the study. H.L., S.C.,
   Y.Z., Y.L., and X.M. designed and performed the experiments. H.L.,
   S.C., H.M., and Y.L. analyzed the data. G.L., Z.C., P.Z., L.C., P.P.,
   H.Z., Z.C., M.D., S.C., H.M., Q.X., B.W., S.Z., K.S., F.W., D.G., and
   W.Z. provided the patient samples. H.L. drafted the manuscript. H.L.,
   Y.Z., H.M., H.L., Q.W., J.N.R., and X.Y. reviewed and edited the
   manuscript. L.Z., F.M., J.N.R., and X.Y. revised the manuscript. All
   authors reviewed the manuscript.

Supporting information

   Supporting Information
   [281]Click here for additional data file.^ (4MB, pdf)

Acknowledgements