Therapeutic implications of altered cholesterol homeostasis mediated by loss of CYP46A1 in human glioblastoma

Abstract Dysregulated cholesterol metabolism is a hallmark of many cancers, including glioblastoma (GBM), but its role in disease progression is not well understood. Here, we identified cholesterol 24‐hydroxylase (CYP46A1), a brain‐specific enzyme responsible for the elimination of cholesterol through the conversion of cholesterol into 24(S)‐hydroxycholesterol (24OHC), as one of the most dramatically dysregulated cholesterol metabolism genes in GBM. CYP46A1 was significantly decreased in GBM samples compared with normal brain tissue. A reduction in CYP46A1 expression was associated with increasing tumour grade and poor prognosis in human gliomas. Ectopic expression of CYP46A1 suppressed cell proliferation and in vivo tumour growth by increasing 24OHC levels. RNA‐seq revealed that treatment of GBM cells with 24OHC suppressed tumour growth through regulation of LXR and SREBP signalling. Efavirenz, an activator of CYP46A1 that is known to penetrate the blood–brain barrier, inhibited GBM growth in vivo. Our findings demonstrate that CYP46A1 is a critical regulator of cellular cholesterol in GBM and that the CYP46A1/24OHC axis is a potential therapeutic target.


Introduction
Glioblastoma (GBM) is the most common primary malignant brain tumour in adults (Fack et al, 2017;Lhomond et al, 2018). Despite aggressive therapy, including maximal surgical resection followed by radiotherapy and temozolomide treatment, the median patient survival is 14.6 months from initial diagnosis (Stupp et al, 2005). New treatment strategies are therefore urgently needed.
Emerging evidence has linked disrupted cholesterol homeostasis to cancer development including GBM (Bovenga et al, 2015;Kuzu et al, 2016;Cheng et al, 2018). However, epidemiological data remain contradictory regarding the relationship between cancer risk and serum cholesterol levels, suggesting that circulating cholesterol levels alone have marginal effects on cancer development (Silvente-Poirot et al, 2018). Oxysterols are oxygenated derivatives of cholesterol that participate in the regulation of cholesterol metabolism. They are derived from the diet or generated through endogenous cholesterol metabolism in tissues. Oxysterols include 24-hydroxycholesterol (24OHC), 25-hydroxycholesterol (25OHC), 27-hydroxycholesterol (27OHC) and ring oxysterols (Janowski et al, 1996;Berrodin et al, 2010) that are known to, in addition to cholesterol metabolism regulation, modulate signalling pathways such as Hedgehog, Wnt and MAPK (Kloudova et al, 2017). In neurodegenerative disease and cancer, oxysterols interact with specific proteins and transcription factors, such as the liver X receptors (LXR) and oxysterol-binding protein and insulin-induced gene 1 (INSIG1) (Kloudova et al, 2017). Recently, it has been shown that 27OHC can inhibit prostate cancer growth through depletion of intracellular cholesterol levels (Alfaqih et al, 2017). Yet, others have shown, in breast cancer cells, that 27OHC induces an epithelial-to-mesenchymal transition (EMT) leading to increased tumour growth (Torres et al, 2011;Wu et al, 2013). It has also been shown in a subtype of early-stage hepatocellular carcinoma that disrupted cholesterol homeostasis is associated with poor prognosis (Jiang et al, 2019). In GBM, the notion that excessive accumulation of cholesterol can fuel the growth of tumour cells has been supported by the following evidence. First, de novo cholesterol synthesis is suppressed in GBM cells compared with normal human astrocytes leading to exogenous cholesterol uptake through up-regulation of the low-density lipoprotein receptor (LDLR) (Villa et al, 2016). Second, loss of endogenous oxysterols in GBM leads to decreased cholesterol efflux (Villa et al, 2016). At present, the mechanisms causing dysregulation of cholesterol homeostasis in GBM are not clear, especially with regard to oxysterol loss.
In mammals, cholesterol homeostasis is tightly regulated by a complex protein network centred around two groups of transcription factors, sterol-regulatory binding proteins (SREBPs) and liver X receptors (LXRs), that are involved in cholesterol import, export, synthesis, metabolism and esterification (Ikonen, 2008). SREBPs mainly promote the transcription of genes, such as HMGCR and LDLR, that are involved in cholesterol synthesis and uptake when intracellular cholesterol levels decrease. LXRs respond to excessive intracellular cholesterol by inducing the expression of genes, such as ABCA1 and ABCG1, that are responsible for cholesterol efflux and LDLR degradation through induction of the E3 ubiquitin ligase IDOL (Zelcer et al, 2009). Targeting these pathways has been shown to be an effective strategy for inhibiting growth in GBM animal models (An & Weiss, 2016).
Cancer cells have clearly developed mechanisms for an accumulation of excessive intracellular cholesterol in order to support their growth. Cholesterol metabolism might therefore represent a potential therapeutic target. However, the key factors contributing to the accumulation of cholesterol in GBMs remain unclear. Here, we used large-scale in silico analyses of whole-transcriptome databases to identify dysregulated genes in GBMs involved in cholesterol homeostasis. One of the most dramatically down-regulated genes was cholesterol 24-hydroxylase (CYP46A1), a brain-specific enzyme responsible for the elimination of cholesterol through conversion of cholesterol into 24(S)-hydroxycholesterol (24OHC) (Moutinho et al, 2016). CYP46A1 expression emerged as a prognostic marker in GBM patients, and in functional studies, overexpression or pharmacological activation of the CYP46A1/24OHC axis suppressed GBM cell growth in vitro and in vivo. Our results show that changes in CYP46A1 are critical for the dysregulation of cholesterol homeostasis in GBM and that targeting CYP46A1/24OHC may provide a new opportunity for GBM therapy.

Results
Cholesterol level promotes GBM cell growth First, we determined the difference in cholesterol regulation between GBM and normal brain. Gene set enrichment analysis (GSEA) based on the Rembrandt data revealed that GBM (n = 217) and normal brain tissues (n = 28) displayed a distinct enrichment of cholesterol metabolic processes compared with normal brain (Appendix Fig S1A). Cholesterol synthesis, however, was positively associated with normal brain (adjusted P < 0.05, respectively; Appendix Fig S1A), which is consistent with previous reports indicating that GBM cells mainly depend on exogenous cholesterol synthesized by normal human astrocytes for growth (Villa et al, 2016). To confirm that GBM cells depend on exogenous cholesterol, two GBM cell lines (LN229 and U251) and GSCs from a primary GBM (GBM#P3) were cultured in media with FBS or in lipoprotein-deficient serum (LPDS). Growth curves generated revealed a significantly slower growth of GBM cells in LPDS after 5 days (P < 0.01). However, the addition of 5 lg/ml LDL (the lipoprotein with the highest cholesterol content) to the LPDS medium rescued GBM cell growth compared with LPDS medium (Appendix Fig S1B-D). These data confirm the importance of exogenous cholesterol for GBM cell growth as reported previously (Guo et al, 2011;Geng et al, 2016).

CYP46A1 is a tumour suppressor candidate in GBM
To identify the most dysregulated cholesterol-related genes in GBM, we performed bioinformatic analysis on publicly available genomic datasets. First, we derived a gene signature of 176 genes involved in cholesterol biology based on Gene Ontologies (Alfaqih et al, 2017). Through differential analysis using the Rembrandt dataset, we uncovered a total of 13 differentially expressed genes between GBM (n = 217) and normal brain tissue (n = 28) (Fig 1A and B). The brain-specific cholesterol regulator, CYP46A1, emerged as the most differentially expressed gene with decreased expression in GBM (log 2 fold change = 2.335, adjusted P = 5.85E-25). Differential analysis based on the Chinese Glioma Genome Atlas (CGGA) dataset also revealed CYP46A1 as one of the most dysregulated transcripts (log 2 fold change = 1.966, adjusted P = 4.63E-09) between GBM (n = 128) and normal brain (n = 5; Fig EV1A-C). Univariate Cox regression analysis in GBM patients was also used to screen the cholesterol-related genes based on prognostic values. After ranking by P-value, CYP46A1 emerged among the top 3 genes (APOBR, CELA3A and CYP46A1) associated with GBM prognosis (Fig EV1D). By analysing the Cancer Genome Atlas (TCGA) pan-cancer data including 31 different cancer types, the expression of CYP46A1 was found to be significantly increased in normal brain compared with GBM and LGG (Appendix Fig S2A). Loss of CYP46A1 in GBM was further confirmed by analysing several public glioma datasets (over 1,500 samples were enrolled; P < 0.01, respectively; Appendix Fig  S2B).
The intra-tumoral expression pattern of CYP46A1 in GBMs was further determined using the IVY GBM RNA-seq data (http://gliobla stoma.alleninstitute.org/). CYP46A1 was highly expressed at the leading edge (which is mainly comprised of normal brain cells) compared with other tumour regions (Appendix Fig S2C). Single-cell RNA-seq data (Darmanis et al, 2017) further demonstrated that CYP46A1 is mainly expressed in neurons, astrocytes and oligodendrocyte precursor cells (OPCs) and to a lesser extent in tumour cells (Appendix Fig S2D). CYP46A1 protein levels were also examined in different cell lines (Appendix Fig S2E). Normal human astrocytes (NHAs) displayed abundant CYP46A1 protein levels, while GBM cells (GBM#P3, GBM#05, GBM#BG7, LN229, U251 and LN18) showed much lower expression. To confirm that CYP46A1 expression is reduced in GBMs at the protein level, we performed IHC staining for CYP46A1 on an independent cohort of glioma (n = 58) and normal brain tissue samples (n = 6). CYP46A1 expression was found to be cytoplasmic. Expression was also consistently high in normal brain tissue (5/6; 88.3%) compared with LGG (14/24; 58.3%) and GBM (6/34; 17.6%), and decreased with increasing glioma grade (Fig 1C and D). IHC staining in one specific GBM sample demonstrated that CYP46A1 was positively expressed in adjacent brain tissue while absent in GBM ( Fig 1E). IHC data from the Human Protein Atlas (n = 17) corroborated these results, confirming high expression of CYP46A1 in normal brain (Fig EV2A), which decreased with increasing tumour grade (Fig EV2B and C).
We subsequently assessed CYP46A1 levels based on the 2016 WHO classification of gliomas, using the TCGA data. CYP46A1 was higher in three LGG subtypes (LGG-Oligo, LGG-Astro and LGG-IDHwt) and lower in GBM-IDHmt and GBM-IDHwt ( Fig 1F). Although there is a general down-regulation of CYP46A1 in GBM compared with normal brain, higher expression of CYP46A1 was also observed in the Neural GBM molecular subtype (Fig 1G), which is associated with a more favourable prognosis, relative to the other subtypes based on the TCGA Verhaak-2010 molecular classification of GBM (Noushmehr et al, 2010).
To address why CYP46A1 is decreased in GBMs, we examined its promoter/enhancer regions using the combined chromatin immunoprecipitation sequencing (ChIP-seq) data from GEO and ENCODE databases. Here, H3K27ac peaks (marker of active promoters) within the promoter region of CYP46A1 were lower in GBMs compared with normal brain tissue (Appendix Fig S3A). We also examined the active enhancer landscape of CYP46A1 across three matched pairs of GSCs and differentiated glioma cells (DGCs). CYP46A1 enhancers and mRNA levels tended to decrease in GSC versus DGC, as measured by ChIP-seq (H3K27ac and H3K4me3 peak levels) and mRNA data (Appendix Fig S3A-C). These results were also validated through ChIP-qPCR and Western blot analysis (Appendix Fig S3D  and E). Taken together, abnormal histone modifications may partially explain reduced CYP46A1 expression in GBM.
Decreased CYP46A1 levels correlate with worse survival in glioma patients To determine the clinical significance of CYP46A1, Kaplan-Meier analysis was performed using the CGGA dataset. GBM patients with high CYP46A1 mRNA levels (based on the median value) exhibited significantly better overall survival (OS) as well as progression-free survival (PFS) (Fig 1H and I). CYP46A1 was also a prognostic indicator in LGG patients (Fig 1H and I). The prognostic value of CYP46A1 was further validated in TCGA, Rembrandt and Phillips datasets (Appendix Fig S4A-C). CYP46A1 was also validated as an independent prognostic indicator using univariate and multivariate Cox regression analysis of OS (HR = 0.390, 95% CI = 0.262 to 0.581, P < 0.001; Appendix Table S1) in CGGA patients and showed a prognostic trend in multivariate analysis of TCGA patients. Collectively, these results emphasize the prognostic relevance of CYP46A1 in GBM.
LN229 and GBM#P3-lenti-CYP46A1 or GBM#P3-lenti-control cells were injected intracranially into mice to evaluate tumour growth in vivo. For both cell types, CYP46A1 overexpression inhibited tumour growth, as observed in haematoxylin and eosin (HE) staining (Fig 2E), and prolonged overall survival in mice (P < 0.05; Fig 2F). Overexpression of CYP46A1 also led to reduced levels of the proliferation marker PCNA but increased expression of the apoptotic marker cleaved caspase-3 ( Fig 2G, Appendix Fig S5D and E). These data confirm a tumour-suppressive role for CYP46A1 in human GBM.

24OHC inhibits GBM growth and GSC maintenance
The CYP46A1 gene encodes a cytochrome P450 oxidase, sterol 24hydroxylase, the main function of which is to convert cholesterol into 24OHC (Fig 3A). We therefore characterized metabolic changes in 24OHC in human GBM. 24OHC levels were decreased in primary GBMs (n = 3) compared with normal brain tissue samples ◀ Figure 1. Expression of CYP46A1, a gene involved in brain cholesterol homeostasis, is lost in GBM and associated with increasing tumour grade.

A
Heatmap of the differentially expressed cholesterol-related genes between normal brain tissues (n = 28) and glioblastomas (n = 217) from the Rembrandt dataset.
Gene expression values are z-transformed and coloured red for high expression and blue for low expression, as indicated in the scale bar. B Volcano plot showing the fold change (log2) in cholesterol-related gene levels based on GBM versus normal brain tissue. Data were obtained from the Rembrandt dataset. C Representative images of IHC staining for CYP46A1 protein in normal brain and different pathological grades of gliomas (n = 64). Scale bar = 30 lm. D Quantification of CYP46A1 IHC staining in normal brain (n = 6) and different pathological grades of gliomas (n = 58). E Representative images of CYP46A1 IHC staining in GBM and adjacent brain tissues from one specific case. Scale bar = 30 lm. F CYP46A1 expression levels in tumours from the TCGA dataset using 2016 WHO classification. Data are shown as the mean AE the standard error of the mean (SEM; n = 667). ***P < 0.0001. Statistical significance was determined by one-way ANOVA. G CYP46A1 expression levels in different molecular subtypes from the Rembrandt GBM dataset. Shown are means and SEM (n = 245). ***P < 0.0001. Statistical significance was determined by one-way ANOVA. H-I Kaplan-Meier analysis for patient OS and PFS based on high versus low expression of CYP46A1 in LGG and GBM. Data were obtained from the CGGA dataset.
P-values were obtained from the log-rank test.
(P < 0.05; Appendix Fig S6A). These results were consistent with decreased CYP46A1 expression in GBM. Ectopic expression of CYP46A1 in GBM cells in vitro, however, led to a dramatic increase in 24OHC concentration in culture media ( Fig 3B) as well as in GBM cell pellets (P < 0.01; Appendix Fig S6B). Exogenous 24OHC inhibited LN229 growth in a dose-and time-dependent manner (Appendix Fig S6C and D). IC 50 s for 24OHC were 28.86 lM in LN229 and 33.81 lM in GBM#P3 cells ( Fig 3C).
To identify potential mechanisms for 24OHC inhibition of GBM proliferation, cell cycle parameters and apoptosis were determined A qRT-PCR analysis of CYP46A1 mRNA levels in LN229 and GBM#P3 cells transfected with lentivirus expressing CYP46A1 (lenti-CYP46A1) or control sequence (lenti-Ctrl). GAPDH was used as an internal control. Shown are means and SEM (n = 3). LN229: ***P < 0.0001; GBM#P3: ***P < 0.0001. Statistical significance was determined by two-sided Student's t-test. B Western blot analysis to confirm CYP46A1 overexpression in GBM cells. C Growth curves for GBM cells in vitro infected with lenti-Ctrl or lenti-CYP46A1 derived from trypan blue staining. Shown are means and SEM (n = 3). LN229: **P = 0.003; GBM#P3: **P = 0.002. Statistical significance was determined by two-sided Student's t-test. D Colony-forming assay for GBM cells infected with lenti-Ctrl or lenti-CYP46A1. Shown are means and SEM (n = 3). LN229: *P = 0.013; LN18: *P = 0.012. Statistical significance was determined by two-sided Student's t-test. E Representative H&E staining of orthotopic tumours derived from GBM cells infected with lenti-Ctrl or lenti-CYP46A1. Scale bar = 2 mm. F Kaplan-Meier survival curve of tumour-bearing mice injected with GBM cells infected with lenti-Ctrl or lenti-CYP46A1 (n = 5 per group). A log-rank test was used to assess statistical significance. G IHC for CYP46A1, PCNA and cleaved caspase-3 protein levels in the intracranial tumours. Scale bar = 100 lm.
ª 2019 The Authors EMBO Molecular Medicine 12: e10924 | 2020 using flow cytometry (analysis of Annexin V-FITC and PI staining). Treatment of LN229 and GBM#P3 with 24OHC increased apoptosis in a dose-dependent manner (Fig 3D and E). 24OHC also elevated caspase-3/7 activity in LN229 (Appendix Fig S6E and F) and dramatically inhibited colony formation in LN229 and LN18 compared with controls (P < 0.001, Fig 3F). Furthermore, Western blot analysis showed an increase in cleaved PARP and cleaved caspase-3, and a decrease in PCNA in a time-dependent manner ( Fig 3G and Appendix Fig S6G). Cell cycle analysis based on PI staining indicated only a slight increase in G2/M phase cells after 24OHC treatment (Appendix Fig S6H). Finally, treatment of GSCs, GBM#P3, GBM#BG7 and GBM#BG5, with 24OHC, led to a reduced tumorsphere formation (Fig 3H and I). In contrast, 24OHC was less toxic to normal human astrocytes (NHAs) and rat brain organoids ( Fig EV3A-C). In summary, these data show that 24OHC specifically inhibits GBM growth.

24OHC suppresses the growth of GBM cells by reducing cholesterol accumulation
It has been shown that neurons and astrocytes both produce oxysterols, including 24OHC, as a result of brain cholesterol metabolism. These molecules act as endogenous ligands for the liver X receptors (LXRs) to decrease excess cellular cholesterol levels (Repa et al, 2000;Venkateswaran et al, 2000;Zelcer et al, 2009;Chen et al, 2013). In GBM cells in vitro, 24OHC treatment resulted in a significant decrease in total intracellular cholesterol levels (P < 0.01, Fig 4A) and filipin staining which reflects free cholesterol levels ( Fig 4B). These results therefore show that 24OHC inhibits GBM cell growth by depleting intracellular cholesterol. To further validate 24OHC induced growth inhibition, rescue experiments with cholesterol were performed in LN229 cells. The addition of exogenous cholesterol significantly increased the growth of 24OHC treated cells (P < 0.0001; Fig 4C). Cholesterol also attenuated apoptosis caused by 24OHC in both LN229 and GBM#P3 ( Fig 4D) and decreased cleaved PARP ( Fig 4E). Cholesterol also compensated for the reduction in GSC sphere formation caused by 24OHC in both GBM#P3 and GBM#BG7 cells (P < 0.05; Fig 4F). Finally, exposure to methylb-cyclodextrin (MbCD), which depletes cellular cholesterol, also inhibited GBM cell survival (P < 0.001 at 10 mM; Fig 4G). In conclusion, these results demonstrated that 24OHC suppresses GBM growth by depleting cellular cholesterol.

24OHC inhibits GBM growth through regulation of LXR and SREBP1 activity
To determine the mechanisms underlying 24OHC inhibition of GBM growth, we performed RNA sequencing (RNA-seq) of GBM#P3 cells treated with or without 24OHC (heatmap Fig 5A). GO analysis indicated that differentially expressed genes were closely related to biological processes involved in negative regulation of lipid storage and positive regulation of cholesterol efflux, cholesterol homeostasis and cell apoptosis. In addition, inflammatory signalling pathways including IL-10 and NF-jB pathways were regulated by 24OHC ( Fig 5B). KEGG pathway analysis also indicated that these genes were strongly associated with the LDLR pathway, the FXR (farnesoid X receptor) and LXR regulation of cholesterol metabolism ( Fig 5C). Moreover, GSEA showed that 24OHC was associated with positive regulation of cholesterol efflux, but negatively associated with gene signatures linked to cholesterol homeostasis, SREBP targets and stem cell proliferation ( Fig 5D). A volcano plot of the RNA-seq results revealed significant changes in several target genes of two cholesterol homeostasis transcription factors, LXR and SREBPs. Interestingly, GSC stemness markers, including SOX2 and SOX4, decreased after exposure to 24OHC (Fig 5E), which was consistent with a decreased tumorsphere formation ability. RNA-seq results for these genes were further confirmed by qRT-PCR ( Fig 5F) and immunofluorescence staining of SOX2 in GBM#P3 tumorspheres treated with 24OHC (Appendix Fig S7A). 24OHC is an endogenous LXR agonist (Chen et al, 2013). We found that 24OHC triggered GBM cell death in the presence of LXR-623 (P < 0.05; Fig 5G), which is an effective synthetic LXR agonist (Villa et al, 2016). These results indicate that 24OHC might kill GBM cells not only through activation of LXR, but also through other mechanisms. Based on the RNA-seq data, 24OHC was also found to suppress SREBP signalling (Fig 5D and E). Analysis of the genomic datasets consistently showed that SREBP1 was the predominant SREBP isotype expressed in GBMs predicting poor overall survival (Appendix Fig S7B). This is consistent with prior reports showing that SREBP1, but not SREBP2, contributes to GBM progression and tumorigenesis (Geng et al, 2016). We thus reasoned that 24OHC influenced SREBP1 transcriptional activity in GBM. On Western blots, proteins levels were correspondingly altered; 24OHC caused a decrease in protein levels of nuclear SREBP1 (N-SREBP1), precursor SREBP1 (P-SREBP1) and LDLR as well as an induction of ▸ Figure 3. CYP46A1 inhibits the growth of GBM cells through the production of 24OHC.
These results show that inhibition of cholesterol leading to reduced cell growth by 24OHC is partly mediated by up-regulation of LXR target genes and down-regulation of SREBP1 activity.

Efavirenz inhibits GBM growth and prolongs survival of intracranial tumour-bearing mice
Based on the findings that CYP46A1 inhibits GBM growth by converting cholesterol to 24OHC, we investigated the effect of efavirenz (EFV) (Fig 6A), which is an anti-HIV medication known to activate CYP46A1 activity through binding to the enzyme's allosteric site (Mast et al, 2014(Mast et al, , 2017aAnderson et al, 2016). Bioinformatic chemical-protein interaction network analysis (STITCH) indicated a strong association between EFV and CYP46A1 protein, as well as PCNA (Appendix Fig S8A). EFV treatment reduced intracellular levels of cholesterol in GBM cells (Fig 6B and C). The IC 50 values for EFV were 16.81 lM in LN229 and 24.58 lM in GBM#P3 ( Fig 6D). Also, increased levels of 24OHC were observed in culture media from EFV-treated cells (P < 0.05, Fig 6E). Functional studies showed that EFV potently reduced LN229 and GBM#P3 tumour growth in a time-and dose-dependent manner (Appendix Fig S8B-D). Colony formation assays confirmed the toxicity of EFV in LN229 and LN18 (P < 0.01; Fig 6F). Flow cytometry analysis demonstrated that EFV induced apoptosis in GBM cells (P < 0.001, Fig 6G), while sparing normal brain organoids and NHAs (Appendix Fig S9A-C). In GSCs (GBM#P3, GBM#BG7 and GBM#BG5), EFV significantly suppressed tumorsphere formation (Fig 6H). At the molecular level, EFV led to increased levels of the LXR target ABCA1 and the apoptosis marker cleaved PARP, but to decreased levels of SREBP1 and LDLR, as well as PCNA and SOX2 ( Fig 6I). Finally, exogenous cholesterol treatment (Appendix Fig  S9D) or CYP46A1 knockdown (Appendix Fig S9E and F) partially restored EFV-induced growth inhibition of GBM cells. These results suggest that the anti-cancer effect of EFV is mediated through regulation of cholesterol levels.
The penetration of the blood-brain barrier (BBB) by EFV has been well-documented in several studies. Median levels of EFV concentration in the cerebrospinal fluid (CSF) were reported to be 11.1-30 lg/l (Tashima et al, 1999;Best et al, 2011;Tovar-y-Romo et al, 2012;Yilmaz et al, 2012;Calcagno et al, 2015;Winston et al, 2015). EFV (0.09 mg/kg/day) by oral gavage treatment in C57BL/6J mice, which is 6,667-fold lower than the dose given to HIV patients (600 mg/kg/day), was recently reported to be sufficient to activate CYP46A1 and to generate a 42% increase in 24OHC in mouse brain tissue (Mast et al, 2014). Therefore, we investigated the therapeutic impact of EFV in mice bearing LN229 and GBM#P3 orthotopic xenografts ( Fig 6J). Oral administration of EFV (0.1 mg/kg/day) was given to mice bearing tumours (n = 5 per group). Overall survival of mice was prolonged (median survival: 29 versus 35 days for GBM#P3; 27 versus 44 days for LN229; Fig 6K). For GBM#P3, bioluminescence imaging revealed a significant inhibition of tumour growth at day 21 (P < 0.01, Fig 6L and M). A trend for increasing body weight was also observed in EFV-treated animals (Fig 6N). No cytotoxic effect of EFV was observed during treatment. IHC staining of brain sections from tumour-bearing mice also demonstrated that EFV suppressed the expression of PCNA and induced protein levels of cleaved caspase-3 (Appendix Fig S10A-D). In conclusion, these findings demonstrate that EFV represents a promising drug repurposing strategy for GBM treatment.

Discussion
Despite varying genetic backgrounds, several solid tumours including GBM rely on lipid metabolism to fuel tumour growth (Cheng et al, 2018). In the present study, GSEA of genomic data revealed an accumulation of cholesterol in GBMs compared with normal brain tissue. Consistent with previous reports (Guo et al, 2011;Geng et al, 2016), we showed that GBM cells require extracellular cholesterol to fuel tumour growth, indicating that a metabolic reprogramming and cholesterol homeostasis dysfunction occurs in GBM. Using a bioinformatic approach based on publicly available genomic data, we identified the most dysregulated cholesterol-related gene, CYP46A1, and investigated its role in GBM. The expression of CYP46A1 was negatively correlated with WHO grade and malignant clinicopathological features of gliomas. Moreover, through large-scale survival analysis, we showed that low CYP46A1 expression in glioma tissues is associated with poor prognosis and can therefore serve as an independent prognostic indicator.
Human CYP46A1, which is located on chromosome 14q32, contains 15 exons and encodes for cholesterol 24-hydroxylase (Lund et al, 1999). CYP46A1 is specifically expressed in the brain and is ◀ Figure 5. The downstream mechanisms of 24OHC in GBM progression.

A
Heatmap of differentially expressed transcripts in RNA-seq data from GBM#P3 cells treated with DMSO or 20 lM 24OHC for 72 h. Gene expression data were mean z-transformed for display and coloured red for high expression and blue for low expression. B, C GO and KEGG enrichment analysis of differentially expressed genes based on RNA-seq data with adjusted P-value. D GSEA plot showing normalized enrichment scores (NESs) for cholesterol homeostasis, positive regulation of cholesterol efflux, SREBP targets and stem cell proliferation signatures using RNA-seq data from 24OHC and DMSO-treated cells. E Volcano plot highlighting up-regulated (red) and down-regulated (blue) genes based on RNA-seq data from GBM#P3 cells after 24OHC treatment. LXR targets, SREBP targets and stemness transcription factors are specifically included. **adjusted P < 0.01, ***adjusted P < 0.001. F qRT-PCR analysis of 24OHC-regulated genes in GBM#P3 cells after treatment with 20 lM 24OHC for 72 h. GAPDH was used as an internal control. Data are shown as the mean AE SEM (n = 3). ***P < 0.0001, ***P < 0.0001, *P = 0.0192, ***P = 0.0004, *P = 0.0118, *P = 0.0101, ***P < 0.0001. Statistical significance was determined by two-sided Student's t-test. responsible for the conversion of cholesterol into 24 (S)-hydroxycholesterol (24OHC), which crosses the BBB into the systemic circulation for metabolism (Lutjohann et al, 1996;Bjorkhem et al, 1997Bjorkhem et al, , 1998. The enzyme is responsible for at least 40-50% of brain cholesterol turnover (Lutjohann et al, 1996;Xie et al, 2003). Dysregulation of CYP46A1 expression has been shown to occur in several neurodegenerative diseases, including Parkinson's, Alzheimer's and Huntington diseases (Bjorkhem et al, 2013;Leoni et al, 2013;Soncini et al, 2016;Swan et al, 2016). However, its role in cancer progression is less understood (Soncini et al, 2016). In GBM, abnormal histone modifications such as acetylation and methylation are closely related to tumorigenesis and drug resistance (Romani et al, 2018). In the present study, we found lower levels of histone modification of H3K4me3 and H3K27ac sites at the promoter region of CYP46A1, which may partially account for the decreased expression of CYP46A1 in GBM relative to normal brain. In addition, it has been reported that both the demethylating agent 5 0 -aza-2 0 -deoxycytidine (DAC) and the histone deacetylase inhibitor (HDACi) trichostatin A (TSA) can induce CYP46A1 mRNA levels Nunes et al, 2010). Further studies of the upstream regulation of CYP46A1 are therefore warranted.
In functional assays, we observed that overexpression of CYP46A1 repressed GBM proliferation both in vitro and in vivo. through its enzyme activity. Further experiments indicated that 24OHC inhibits the growth of GBM through depletion of intracellular cholesterol. Therefore, the CYP46A1/24OHC axis suppresses proliferation of GBM by disturbing cholesterol homeostasis. Previous studies have identified 24OHC as an activator of LXR and inducer of several genes involved in cellular cholesterol efflux (Lehmann et al, 1997;Janowski et al, 1999), such as the ATPbinding cassette transporter A1 (ABCA1) in both neurons and glia (Fukumoto et al, 2002) and apoE in astrocytes (Liang et al, 2004). RNA-seq revealed that 24OHC caused an up-regulation of LXR targets, such as ABCA1, ABCG1 and IDOL (the E3 ligase of LDLR), and down-regulation of several SREBP target genes. In summary, these data show that 24OHC acts, at least partially, by regulating the activity of two essential transcription factors, LXR and SREBP, involved in cholesterol homeostasis. Although 24OHC likely modulates cell death also through other pathways (Noguchi et al, 2015), our results show that inhibition of CYP46A1/24OHC is a critical underlying regulator of cholesterol homeostasis in GBM. Thus, restoration of the CYP46A1/24OHC axis is a promising GBM treatment strategy. Structure analysis of CYP46A1 suggested that distinct compounds can bind the active site of the enzyme (Mast et al, 2008). EFV, a BBB penetrating drug that has been used for the treatment of HIV, was found to bind to the CYP46A1 allosteric site. The drug alters the shape of the active site of cholesterol-free CYP46A1, making cholesterol-binding tighter and enzyme catalysis more efficient (Mast et al, 2014). Supporting its CNS activity, oral administration has been shown to improve long-term spatial memory, decrease amyloid-b content and also prolong survival of Alzheimer's mice (Mast et al, 2014(Mast et al, , 2017bPetrov et al, 2019). More recently, EFV has been shown to reduce phosphorylated Tau (pTau) accumulation in a human iPSC-derived AD model (van der Kant et al, 2019). EFV has been reported to exert anti-tumour effects in some cancer types, including renal clear-cell carcinoma, prostate cancer and ovarian cancer (Landriscina et al, 2008;Houede et al, 2014;Perna et al, 2017). However, the efficacy of EFV in the treatment of GBM remains unknown. In the current study, we observed that EFV (20 uM) significantly suppressed GBM growth and induced tumour cell death. At the molecular level, EFV stimulated the activity of CYP46A1, which resulted in elevated levels of 24OHC leading to a ◀ Figure 6. Efavirenz, an activator of CYP46A1, attenuates the growth of GBM cells and prolongs survival of mice bearing intracranial tumours. The survival curves of tumour-bearing mice implanted with LN229 or GBM#P3 cells after EFV or PBS treatment (n = 5 per group). A log-rank test was used to assess statistical significance. L, M Bioluminescent images and the corresponding quantification of tumours in mice implanted with GBM#P3 cells at days 7 (n = 5 per group, P = 0.58) and 21 (n = 5 per group, P = 0.005). Data are shown as mean AE SEM. Statistical significance was determined by two-sided Student's t-test. N Body weight of tumour-bearing mice after 3 weeks of EFV or PBS treatment. Data are shown as the mean AE SEM (n = 5 per group). Statistical significance was determined by two-sided Student's t-test. (Left panel) GBM cells have an excessive cholesterol accumulation caused by CYP46A1 down-regulation, which leads to a reduced 24OHC production that in turn cause a suppression of LXR signalling and an induction of SREBP targets. (Right panel) Restoration of CYP46A1 activity by EFV impairs GBM tumorigenic properties by increasing 24OHC levels, which in turns decreases intracellular cholesterol levels. Mechanistically, EFV treatment leads to the up-regulation of LXR target genes including ABCA1, and down-regulation of SREBP1 activity and LDLR expression.
12 of 18 EMBO Molecular Medicine 12: e10924 | 2020 ª 2019 The Authors decreased accumulation of cholesterol, accompanied by increased levels of the LXR target ABCA1, and decreased levels of SREBP1 and LDLR. Notably, exogenous cholesterol treatment or CYP46A1 knockdown partially restored growth inhibition of GBM cells induced by EFV, suggesting that the anti-cancer effect of EFV is mediated at least partially, by regulating cholesterol levels. EFV has been frequently associated with neuropsychiatric adverse events (NPAE) when given to HIV patients at high doses (Apostolova et al, 2017). However, major NPAE were not observed in mice at lower doses of EFV, indicating an appropriate therapeutic window (Mast et al, 2017b). Finally, EFV showed promising anti-glioma effects both in vitro and in vivo. A schematic representation of our results is illustrated in Fig 7. We show that accumulation of cholesterol and dysregulated cholesterol homeostasis in GBM is mediated by loss of CYP46A1 and that CYP46A1 represents a viable therapeutic target in GBMs. An activator of CYP46A1 inhibited tumour growth in vitro and in vivo by increasing the production of the cholesterol metabolite 24OHC, thus disturbing cholesterol homeostasis in GBM. Moreover, restoration of the activity of the CYP46A1/24OHC signalling axis by EFV, a medication currently in widespread use for the treatment of AIDs, might be a promising GBM therapeutic strategy.

Ectopic expression
Lentivirus expressing CYP46A1 or a control sequence was obtained from GeneChem (Shanghai, China). A lentivirus with luciferase was obtained from Obio (Shanghai, China). Lentivirus expressing CYP46A1 was used to infect LN229 and GBM#P3 followed by puromycin selection (Sigma-Aldrich; St. Louis, MO) to generate cell lines stably expressing CYP46A1. The ORF of SREBP1 in pEnter and vector plasmid control were purchased from Vigene Biosciences (Jinan, Shandong, China). Transient transfection was carried out using Lipofectamine 3000 (Life Technologies; Gaithersburg, MD) according to the manufacturer's protocol.

RNA isolation and quantitative real-time PCR
Total RNA was extracted from cells using TRIzol reagent (Invitrogen; Carlsbad, CA, USA) and reverse-transcribed using the Rever Tra Ace qPCR RT Kit (Toyobo; Osaka, Japan). cDNA was amplified using SYBR Green on the Roche Light Cycler 480 for quantification. The relative expression levels of mRNA were normalized to glyceraldehyde-3-phosphate dehydrogenase (GAPDH). Sequences of the primers used are shown in Appendix Table S3.

Clinical specimens
Archived paraffin-embedded glioma tissues (WHO grades II-IV) were collected from patients (n = 58) who underwent surgery in the Department of Neurosurgery, Qilu Hospital of Shandong University. Normal brain tissue samples (n = 6) were taken from trauma patients who underwent partial resection of normal brain as decompression treatment for severe head injuries. The informed consent was obtained from all subjects and that the experiments conformed to the principles set out in the WMA Declaration of Helsinki and the Department of Health and Human Services Belmont Report.

Establishment of intracranial GBM xenografts and efavirenz treatment
Experimental procedures were approved by the Research Ethics Committee of Shandong University and the Ethics Committee of Qilu Hospital (Shandong, China). The Institutional Animal Care and Use Committee (IACUC) of Shandong University approved all animal procedures. Four-week-old male athymic nude mice (Foxn1 nu mut/ mut; SLAC Laboratory Animal Center; Shanghai, China) were bred under specific pathogen-free conditions at 24°C on a 12-h day-night cycle. For orthotopic transplantation, mice were grouped randomly (n = 5 per group) and intracranially injected with 1 × 10 6 LN229 or GBM#P3 cells resuspended in 10 ll PBS. At day 7 after cell implantation, mice were given drugs (efavirenz) or PBS by oral gavage at 0.1 mg/kg/day until sacrifice based on tumour burden defined by the observation endpoints of the experiments (neurological signs became apparent). GBM#P3 expressing luciferase xenograft growth was monitored with bioluminescent imaging using an In Vivo Imaging System (IVIS) Spectrum (Perkin-Elmer; Waltham, MA). Brains were collected and fixed in 4% formaldehyde for haematoxylin and eosin (HE) staining and IHC analysis.

LC-MS analysis
24OHC measurements were performed by Biotree Biotechnology Co., Ltd (Shanghai, China). Cell samples were prepared as follows: 1,000 ll of extract solvent (precooled at À20°C, acetonitrile-methanol-water, 2:2:1) was added to pelleted cells, and samples were vortexed for 30 s and homogenized at 45 Hz for 4 min. Media samples were prepared as follows: a 200 ll aliquot of each individual sample was precisely transferred to an Eppendorf tube. Methanol (400 ll) and acetonitrile (400 ll) were added to the aliquot, and the samples were vortexed for 30 s. Cell samples or media were sonicated for 15 min in an ice-water bath followed by incubation at À20°C for 1 h and centrifugation at 13,523 g at 4°C for 15 min. A 750 ll aliquot of the clear supernatant was transferred to a new Eppendorf tube and dried under a gentle nitrogen flow. The residual was reconstituted with 150 ll of solution (acetonitrilemethanol-water, 2:2:1), centrifuged at 13,523 g at 4°C for 15 min. An 80 ll aliquot of the clear supernatant was transferred to an autosampler vial for UHPLC-MS/MS analysis. Stock solutions were individually prepared by dissolving or diluting each standard substance to give a final concentration of 10 mmol/L. The UHPLC separation was carried out using an Agilent 1290 Infinity II Series UHPLC System (Agilent Technologies; Santa Clara, CA), equipped with an Agilent ZORBAX SB-C18 (50 × 2.1 mm, 1.8 lm, Agilent Technologies). Calibration solutions were subjected to UHPLC-MRM-MS/MS analysis using the methods described above. The calibration standard solution was diluted stepwise, with a dilution factor of 2. These standard solutions were subjected to UHPLC-MRM-MS analysis. The signal-to-noise ratios (S/N) were used to determine the lower limits of detection (LLODs) and lower limits of quantitation (LLOQs). The LLODs and LLOQs were defined as the analyte concentrations that led to peaks with signal-to-noise ratios (S/N) of 3 and 10, respectively, according to the US FDA guideline for bioanalytical method validation. The precision of the quantitation was measured as the relative standard deviation (RSD), determined by injecting analytical replicates of a QC sample. The accuracy of quantitation was measured as the analytical recovery of the QC sample determined. The per cent recovery was calculated as [(mean observed concentration)/(spiked concentration)] × 100%. Finally, 24OHC content for each sample was quantified and normalized to whole-cell protein or cell count. Values shown were normalized to controls.

Cholesterol measurements
Cholesterol content was measured using the Cholesterol Quantitation Kit (Sigma-Aldrich, MAK043-1KT), and filipin staining was performed according to the manufacturer's protocols. For filipin staining (free cholesterol marker), GBM cells were treated for 72 h with 24OHC, EFV (MedChem) or vehicle control. Cells were stained with 0.5 mg/ml filipin in 1% BSA/PBS for 1 h after fixation with 4% paraformaldehyde for 15 min and rinsed 3 × 5 min with PBS. Images were captured by confocal microscopy (Leica TCS SP8; Wetzlar, Germany).

RNA sequencing
GBM#P3 cells were treated with 20 lM 24OHC or DMSO for 72 h. Cells were harvested, and total RNA was isolated using TRIzol (Invitrogen). RNA sequencing (RNA-seq) was carried out by OE Biotech. Co., Ltd. (Qingdao, Shandong, China). The genome-wide transcriptomic analysis was performed on a set of 3 separate experiments (24OHC and vehicle control treatment groups). P-values were adjusted using Benjamini and Hochberg's multiple test correction procedures. Differential expression was determined based on fold change (FC) and FDR with |log2(FC)| > 1 and FDR < 0.01. DAVID (https://david.ncifcrf.gov/) was used to perform GO and KEGG analysis, and data visualization was accomplished using R software. Gene set enrichment analysis (GSEA; https://www.broadinstitute. org/gsea/index.jsp) was performed to find differential phenotypes between 24OHC and vehicle control treatment groups.

Flow cytometry
Cells were harvested, rinsed twice with PBS, fixed with 75% (v/v) ethanol overnight at 4°C and stained with propidium iodide (PI; BD Biosciences; San Jose, CA, USA) at room temperature for 20 min. To detect apoptosis, cells were rinsed with PBS, resuspended in 200 ll binding buffer and incubated with Annexin V-FITC and PI (BD Biosciences; San Jose, CA, USA) at room temperature for 15 min. Both cell cycle distribution and apoptosis were analysed on a C6 flow cytometer (BD Biosciences; San Jose, CA, USA). Data were postprocessed using NovoExpress Software (ACEA Biosciences; San Diego, California, USA).

Caspase-3/7 activity assay
In vitro caspase-3/7 activity was determined using CellEvent TM Caspase-3/7 Green Detection Reagent (Invitrogen; Carlsbad, CA, USA) according to the manufacturer's protocol. For the kinetic assay, LN229 cells were treated with 24OHC at different concentrations (0-20 lM) and 2 lM Caspase-3/7 Green Detection Reagent. IncuCyte â S3 Live-Cell Analysis System (Essen BioScience, Ltd.; Welwyn Garden City, Hertfordshire, UK) was used to visualize the progression of apoptosis using FITC/Alexa Fluor TM 488 filter settings at desired time points. Cells were monitored and imaged with the IncuCyte â S3 Live-Cell Analysis System (Essen BioScience, Ltd.; Welwyn Garden City, Hertfordshire, UK), and graphics were generated with IncuCyte software.

Tumorsphere formation assay
GSCs (1,000 cells/ml/well) were seeded in 6-well ultra-low adhesion plates (Corning Inc.; Corning, NY, USA) or general plates (Corning Inc.) and cultured for 10 days. Inverted phase-contrast microscopy (Nikon; Tokyo, Japan) was used to count and acquire images of the non-adherent tumorspheres.

Colony formation assay
LN18 or LN229 (1,000 cells/well) cells were seeded in 6-well culture plates in triplicate and incubated at 37°C in a humidified chamber of 5% CO 2 for 14 days. Colonies were washed with PBS, fixed in methanol for 15 min and stained with crystal violet for 15 min at room temperature. Images of colonies were acquired under brightfield microscopy (Nikon; Tokyo, Japan), and those with a diameter of ≥ 100 lm were counted.

Live-dead cell staining
Live-dead cell staining was performed using the Live/Dead Cell Double Staining Kit (Sigma-Aldrich; St. Louis, MO, USA) according to manufacturer's protocol. Briefly, brain organoids were treated with DMSO or efavirenz for 3 days. The medium was changed before live-dead cell staining to exclude the influence of DMSO and efavirenz on staining. Calcein-AM and PI working solutions were prepared in PBS at a proper dilution. The staining solutions were mixed into a working solution with the brain organoid suspension at a ratio of 1:2 (v/v) and incubated at 37°C for 15 min. Images for live-dead cells were captured with confocal microscopy (Leica TCS SP8; Leica Microsystems; Wetzlar, Germany).

Brain organoid cultures
The preparation and culture of brain organoids have been described previously (Bjerkvig et al, 1986). Briefly, rat foetal brains at the 18 th day of gestation were dissected out aseptically, cut into 0.5 × 0.5 mm pieces and rinsed three times with PBS. Single cells were obtained through serial trypsinization and seeded into agarcoated multi-well dishes for 21 days to form mature brain organoids. Images were captured, after treatment of brain organoids with 24OHC, EFV or DMSO (vehicle control) for 3 days, using bright-field microscopy (Nikon; Tokyo, Japan).

Statistical analysis
Kaplan-Meier survival curves were generated and compared using the log-rank test. The association between CYP46A1 expression and ª 2019 The Authors EMBO Molecular Medicine 12: e10924 | 2020 clinicopathological characteristics was determined in a two-tailed chi-square test. For animal studies, mice were randomly allocated to control or treated group. No blinding was done. The Kolmogorov-Smirnov test was used to assess the normal distribution of data. Statistical analysis was performed using an unpaired Student's t-test for two-group comparison and one-way analysis of variance (ANOVA) for multi-group comparisons using GraphPad Prism 7.0 (La Jolla, CA). All the experiments were repeated at least three times with triplicates unless stated otherwise. Data for each treatment group were represented as the mean AE SEM. All tests were two-sided, and P-values < 0.05 were considered to be statistically significant.
Expanded View for this article is available online.