The Epigenetic Drug 5-Azacytidine Interferes with Cholesterol and Lipid Metabolism*

Background: The anticancer drug 5-azacytidine acts through an incompletely understood mechanism. Results: 5-Azacytidine reprograms the glycerolipid biosynthesis pathway and prevents activation of master transcription factors that regulate lipid homeostasis. Conclusion: 5-Azacytidine deeply modifies how cells manage cholesterol and lipid synthesis. Significance: The findings unravel important insights into the mechanism of 5-azacytidine and highlight new potential cancer therapeutics. DNA methylation and histone acetylation inhibitors are widely used to study the role of epigenetic marks in the regulation of gene expression. In addition, several of these molecules are being tested in clinical trials or already in use in the clinic. Antimetabolites, such as the DNA-hypomethylating agent 5-azacytidine (5-AzaC), have been shown to lower malignant progression to acute myeloid leukemia and to prolong survival in patients with myelodysplastic syndromes. Here we examined the effects of DNA methylation inhibitors on the expression of lipid biosynthetic and uptake genes. Our data demonstrate that, independently of DNA methylation, 5-AzaC selectively and very potently reduces expression of key genes involved in cholesterol and lipid metabolism (e.g. PCSK9, HMGCR, and FASN) in all tested cell lines and in vivo in mouse liver. Treatment with 5-AzaC disturbed subcellular cholesterol homeostasis, thereby impeding activation of sterol regulatory element-binding proteins (key regulators of lipid metabolism). Through inhibition of UMP synthase, 5-AzaC also strongly induced expression of 1-acylglycerol-3-phosphate O-acyltransferase 9 (AGPAT9) and promoted triacylglycerol synthesis and cytosolic lipid droplet formation. Remarkably, complete reversal was obtained by the co-addition of either UMP or cytidine. Therefore, this study provides the first evidence that inhibition of the de novo pyrimidine synthesis by 5-AzaC disturbs cholesterol and lipid homeostasis, probably through the glycerolipid biosynthesis pathway, which may contribute mechanistically to its beneficial cytostatic properties.

Epigenetic marks, such as DNA methylation and histone acetylation, finely alter chromatin structure to precisely control gene expression in a time-, cell-, and tissue-specific manner (1). DNA cytosine methylation, catalyzed by DNA methyltransferases forming 5-methylcytosine at specific CpG dinucleotides, is responsible for the establishment of silent chromatin regions (2). Inhibitors of DNA methyltransferases, such as the unmethylable cytosine analogs 5-azacytidine (5-AzaC) 4 and 5-Aza-2Ј-deoxycytidine (DAC), generate hypomethylated DNA, allowing re-expression of silenced hypermethylated genes (2). Accordingly, based on the premise that they induce re-expression of tumor suppressor genes and because they lower malignant progression to acute myeloid leukemia and increase survival, both 5-AzaC and DAC are used as standards of care for patients with myelodysplastic syndromes (MDS) (3).
Comparative studies have revealed major mechanistic disparities between DAC and 5-AzaC. Although DAC is the most effective hypomethylating agent, 5-AzaC is more potent to reduce cell viability and proliferation in acute myeloid leu-kemia cell lines (4,5). Global DNA microarray analyses demonstrated that the effect of each drug on the cellular transcriptome was very distinct, with largely non-overlapping gene expression profiles. In order to be incorporated into DNA, 5-AzaC has to be converted to DAC by ribonucleotide reductase, which is an inefficient process (ϳ10 -20%) (2). Consequently, 5-AzaC can also be incorporated into different RNA subspecies and may affect nucleic acid and protein metabolism (6). Thus, 5-AzaC antineoplastic effects on abnormal hematopoietic cells may rely on methylation-independent mechanisms, which remain to be determined (4, 6 -8).
Sterol regulatory element-binding proteins (SREBPs) are a family of transcription factors that coordinates homeostatic gene expression of proteins and enzymes required for uptake and biosynthesis of cholesterol, fatty acids, triacylglycerols, and phospholipids (9). The proteolytic release of the N-terminal transcriptionally active fragment of SREBPs from their membrane-bound precursor is finely regulated by cholesterol through a negative feedback loop mediated by sterolsensing endoplasmic reticulum (ER)-resident proteins. Impairment of this mechanism can result in imbalance of cellular cholesterol and lipid content and is associated with severe clinical complications, such as non-alcoholic fatty liver disease, atherosclerosis, and cancer (10 -12). Complementary to genome-wide association studies that have identified loci associated with lipid production (13), epigenetics represents one of the most promising fields to study the impact of induced environmental reprogramming of gene expression in metabolic diseases (14). Indeed, mounting evidence indicates that gene promoter DNA methylation levels may be associated with lipid metabolism gene expression (15).
In the present study, we explored the effects of DNA methylation inhibitors 5-AzaC and DAC on cholesterogenic and lipid gene expression and defined a previously unrecognized mechanism regulating the activation of SREBPs. Treatment of various cell lines or injection of mice with 5-AzaC strongly and selectively reduced expression of SREBP target genes independently of DNA methylation. Our data show that 5-AzaC, unlike DAC, promotes triglyceride synthesis and accumulation of lipid droplets and impedes SREBPs activation. In sterol-resistant Chinese hamster ovary cells, the activation of SREBP-2 was insensitive to 5-AzaC, indicating that this antimetabolite alters the ER cholesterol content. Co-incubation with UMP or cytidine completely reversed the effects of 5-AzaC, demonstrating that inhibition of UMP synthase and CTP depletion is the underlying mechanism. Taken together, these data highlight a major DNA methylation-independent effect of 5-AzaC and the existing link between the de novo pyrimidine and glycerolipid biosynthesis pathways and SREBP signaling.
Animals-Wild-type C57BL/6 male mice were obtained from Charles River and maintained on a standard rodent diet for 3 days in a 12-h light/12-h dark cycle for acclimatization. Pcsk9-deficient male mice (Pcsk9 Ϫ/Ϫ ; Jackson Laboratories) were continuously backcrossed to C57BL/6 mice at least six generations prior to experimentations. 8 -10-week-old male mice (ϳ25 g) were injected intraperitoneally or subcutaneously with 0.9% NaCl (saline) or with 2.5, 5, or 10 mg/kg/day 5-AzaC. 24, 48, or 120 h postinjection, mice were anesthetized, and blood was collected by cardiac puncture, and dissected livers were snap-frozen in liquid nitrogen for further analyses. The Montreal Heart Institute Animal Care and Ethical Committee approved all animal studies.
Reverse Transcription and Quantitative Real-time PCR-The integrity of total RNA samples, isolated using TRIzol (catalog no. 15596026, Invitrogen), was verified by agarose gel electrophoresis or by an Agilent 2100 Bioanalyzer profile. Afterward, cDNA was prepared using SuperScript II reverse transcriptase according the manufacturer's instructions (catalog no. 18064-014, Invitrogen). Quantitative real-time PCR was performed with the MX3000p real-time thermal cycler (Agilent) using PerfeCTa SYBR Green SuperMix, UNG, Low ROX (catalog no. 95070 -100, Quanta Biosciences). For each gene of interest, dissociation curves and agarose gel electrophoresis were performed to ensure a unique PCR product. Arbitrary units were determined from PCR duplicates for each sample using the TATA box-binding protein (TBP), the ribosomal protein S14, or glyceraldehyde 3-phosphate dehydrogenase (GAPDH) as a normalizer. Oligonucleotide sequences are listed in Table 1.
DNA Microarrays and Data Analysis-Extracted total RNA was purified with the RNeasy MinElute cleanup kit (catalog no. 74204, Qiagen). The quality of the total RNA was evaluated on an Agilent 2100 Bioanalyzer system. The microarray experiment was performed using the GeneChip Human Gene 1.0 ST (catalog no. 901085, Affymetrix). For each sample, 100 ng of total RNA was converted into cDNA using the Ambion WT Expression Kit (catalog no. 4411974, Invitrogen). 6 g of the single-stranded cDNA was fragmented and labeled using the Affymetrix GeneChip WT Terminal Labeling Kit (catalog no. 900670), and 2 g of the resulting cDNA was hybridized onto the chip. The whole hybridization procedure was performed using the Affymetrix GeneChip system according to the protocol recommended by Affymetrix. The hybridization was evaluated with Affymetrix GeneChip Command Console Software (AGCC), and the quality of the chips was evaluated with Affymetrix Expression Console. Partek Genomics Suite was used for data analysis. First, the data were normalized by the RMA (robust multichip average) algorithm, which uses background adjustment, quantile normalization, and summarization. Then the transcripts found to be significantly differentially expressed between control (DMEM) and treatment (5-AzaC) groups by more than 2-fold were included in the gene enrichment and pathway analyses, which were performed using the Web-based DAVID functional enrichment algorithm (21). Complete microarray data can be found in supplemental File 1.
Gaussia Luciferase Assay-Human PCSK9 (Ϫ1000 bp), LDLR (Ϫ1020 bp), and TBP (Ϫ1000 bp) proximal promoter cDNAs were generated by PCR using genomic DNA from HepG2 cells as template. Sterol response element (SRE; bp Ϫ345 to Ϫ337) and HNF1 (hepatocyte nuclear factor 1) motifs (bp Ϫ386 to Ϫ374) were mutated within the 1000-bp PCSK9 proximal promoter by directed mutagenesis, as described (22). All amplified products were digested with SpeI and HindIII endonucleases and ligated into pCMV-GLuc vector (catalog no. N8081S, New England Biolabs) in order to replace the CMV promoter. Selected clones were verified by DNA sequencing. All oligonucleotides used are listed in Table 1. Before transfection, HepG2 cells were seeded in 24-well plates at a density of 1.5 ϫ 10 5 /well. 24 h later, cells were transfected in duplicate with the corresponding pGLuc construct. After overnight incubation, cells were washed twice with DMEM and incubated in 0.5 ml of DMEM without or with 10 M 5-AzaC for 24 h. 20 l of conditioned media was loaded into black 96-well plates, and relative activity of secreted Gaussia luciferase was assessed by luminescence measurements using the BioLux kit (catalog no. E3300L, New England Biolabs) and the BioTek Synergy 2 microplate reader.
Immunocytochemistry-24 h after treatment, HepG2 or CHO cells were washed three times with PBS, fixed with 4%
Analysis of SREBP Cleavage-Cell fractionation was carried out as described previously (25) with minor modifications. HepG2 or HEK293 cells from 60-mm dishes were washed three times with PBS and resuspended in 0.4 ml of 10 mM Hepes-KOH (pH 7.5), 10 mM KCl, 1.5 mM MgCl 2 , 5 mM sodium EDTA, 5 mM sodium EGTA, and 250 mM sucrose supplemented with complete protease inhibitors and 25 g/ml ALLN for 30 min on ice. Cells were then passed 25 times through a 22-gauge needle and centrifuged at 1000 ϫ g for 5 min at 4°C. Supernatants were centrifuged at 100,000 ϫ g for 30 min at 4°C. Membrane pellets were resuspended in 75 l of 10 mM Tris-HCl (pH 7.3), 100 mM NaCl, 1% (w/v) SDS, 1 mM EDTA-Na 2 , 1 mM sodium EGTA-Na 2 . Pellets from the initial 1000 ϫ g spin were resuspended in 0.1 ml of 20 mM Hepes-KOH (pH 7.6), 2.5% (v/v) glycerol, 0.42 M NaCl, 1.5 mM MgCl 2 , 1 mM sodium EDTA, and 1 mM sodium EGTA supplemented with protease inhibitor and ALLN, rotated for 1 h at 4°C, and centrifuged at 100,000 ϫ g for 20 min at 4°C. The resulting supernatant was designated as the nuclear extract. For liver subcellular fractionation, snap-frozen pieces were homogenized and processed as described above. After total protein concentration measurements, membrane and nuclear fractions were then subjected to Western blot analyses.
Enzymatic Assays and Incorporation of Glycerol 3-Phosphate into Lipids-For CTP-phosphocholine cytidylyltransferase activity, HepG2 cells were plated at a density of 2 ϫ 10 6 /60-mm dish. 48 h later, cells were preincubated with 10 M 5-AzaC for 2 h and then incubated for 15 min in cholinefree Hanks' balanced salt solution supplemented with 2 Ci of 3 H-labeled choline chloride per dish without or with 10 M 5-AzaC and subsequently incubated for up to 4 h in DMEM. Organic and aqueous fractions were isolated, aliquots of each were subjected to thin layer chromatography, and spots corresponding to phosphatidylcholine and p-choline were removed, and radioactivity was determined as described previously (26). Phosphatidic acid (PA)-CTP cytidylyltransferase (CDS) activity was measured as described previously (27) in microsomal fractions of HepG2 cells in the presence of 0 -50 M 5-AzaC. De novo glycerolipid biosynthesis was monitored by the addition of 3 Ci [1,3-3 H]glycerol without or with 10 M 5-AzaC or 0.1 mM cytidine or both for 24 h. Following incubation, phospholipids, diacylglycerol, and triacylglycerol (TG) were extracted and separated by thin layer chromatography, and incorporation of radioactivity was determined as described previously (27).
MS-based Nucleotide Analysis-Metabolites from 5-AzaCtreated and untreated cells (ϳ5 ϫ 10 6 ) were extracted as follows. Cells were washed three times with ice-cold 150 mM ammonium formate, pH 7.4, and scraped into 380 l of LC/MS grade 50% methanol water mixture. A volume of 320 l cold

5-Azacytidine Restricts SREBP Activation
acetonitrile was added before the cells were lysed by bead beating for 2 min at 30 Hz (TissueLyser, Qiagen). A volume of 300 l of ice-cold water and 600 l of ice-cold methylene chloride was added to the lysate. Samples were vortexed and then allowed to rest on ice for 10 min for phase separation followed by centrifugation at 4000 rpm for 5 min. The upper aqueous layer of the samples was dried by vacuum centrifugation operating at Ϫ4°C (Labconco) and resuspended in 50 l of LC/MS grade ice-cold water by sonication and vortex. Samples were then clarified by centrifugation for 15 min at 15,000 rpm at 1°C. For LC/MS-MS analysis, nucleotides were separated by ultrahigh performance liquid chromatography (1290 Infinity UPLC, Agilent Technologies) using a Scherzo SM-C18 (3 ϫ 150 mm) 3-m column and guard column (Imtakt) operating at 10°C. Solvent A consisted of 5 mM ammonium acetate in water, and solvent B consisted of 200 mM ammonium acetate in 80% water and 20% acetonitrile. Mono-, di-, and triphosphonucleotides were separated using a linear gradient from 0% to 100% B over a period of 5 min followed by 5 min at 100% B at a flow rate of 0.4 ml/min. Nucleotides were eluted into an electrospray ionization source and detected by multiple reaction monitoring using a triple quadrupole mass spectrometer (6430 QQQ, Agilent Technologies). Quantifying ion integrated intensities were compared with external calibration curves collected at the same time as sample mass spectrometric data acquisition.

5-AzaC Alters Cholesterol-regulated Gene Expression-
To study the role of DNA methylation in cholesterol and lipid homeostasis, we first incubated human liver-derived HepG2 cells with increasing concentrations of the unmethylable analog of cytidine 5-AzaC. We observed a very robust and dose-dependent decline in secreted proprotein convertase subtilisin/kexin type 9 (PCSK9) protein levels, the third locus associated with familial hypercholesterolemia (28), whereas albumin secretion remained constant for all dosages tested (Fig. 1A). Quantitative PCR (QPCR) analyses showed that 5-AzaC caused a strong reduction in the mRNA abundance of PCSK9 and HMG-CoA reductase (HMGCR), the rate-limiting enzyme for cholesterol biosynthesis, reaching the maximal effect at 10 M (Fig. 1B), without apparent cytotoxicity (data not shown). Whereas the expression of the cholesterogenic transcription factor SREBF2 (the gene encoding for SREBP-2) was not significantly modified, we noticed that 5-AzaC increased low density lipoprotein receptor (LDLR) gene expression (Fig. 1B), which is usually co-regulated with PCSK9 and HMGCR through SREBP-2 (29). Then, to define the time required for 5-AzaC to modify cholesterogenic gene expression, HepG2 cells were incubated with 10 M 5-AzaC for 0, 2, 4, 6, 8, and 24 h. QPCR analyses revealed that PCSK9 and HMGCR expression began to decrease, and that of LDLR increased, already after 6 h and culminated at 24 h postincubation (Fig. 1C). These results show that the effect of 5-AzaC on those genes occurred within 24 h and indicate that inhibition of DNA methyltransferases and de novo methylation during cell division may not be involved.
To verify if these effects are caused by DNA demethylation, HepG2 cells were incubated with cytidine or its nucleoside analogues 5-AzaC or DAC ( Fig. 2A). Remarkably, only 5-AzaC robustly reduced mRNA levels of HMGCR and PCSK9 and increased LDLR expression (Fig. 2B). Kinetic experiments demonstrated that the decrease in PCSK9 expression was not due to the accelerated decay of its mRNA (Fig. 2C). On the other hand, we noticed that the rapid mRNA turnover of LDLR and, to a lesser extent, of HMGCR was stabilized by the addition of 5-AzaC (Fig. 2C), similar to the effect of berberine, a cholesterol-lowering compound (30). Consequently, in human hepatoma HepG2 and Huh-7 cell lines, LDLR protein was strongly up-regulated by 5-AzaC concomitantly to the reduction of PCSK9 (Fig. 2, D and E), a natural inducer of LDLR degradation (reviewed in Ref. 31), but not in HEK293 cells that do not express PCSK9 (Fig. 2D).
To compare the impact of 5-AzaC and DAC on DNA methylation, we measured the expression of the large intergenic non-coding RNA H19, a maternally imprinted region hypermethylated in HepG2 cells (Fig. 2F). Under basal conditions, H19 was nearly undetectable in HepG2 cells, but it was strongly re-expressed only upon DAC treatment, ϳ125-fold higher than 5-AzaC (Fig. 2F). Because DAC did not significantly modulate expression of PCSK9, HMGCR, and LDLR (Fig. 2B), we conclude that the effect of 5-AzaC on SREBP-2 target genes does not primarily involve DNA methylation.

5-AzaC Promotes Lipid Droplet Formation and Disturbs
Lipid Homeostasis-Next, we examined the consequence of 5-AzaC on subcellular distribution of lipids and cholesterol by confocal microscopy. We noticed that 5-AzaC induced an intracellular accumulation of phospholipids and cholesterol only in the presence of lipoproteins (Fig. 3A), which correlates with increased LDLR levels (Fig. 2D). Intriguingly, 5-AzaC strongly induced the formation of lipid droplets (LDs) independently of exogenous lipoproteins (Fig. 3, A and B). Filipin staining also demonstrated a noticeable reduction of cholesterol in cells grown in lipoprotein-deficient serum and treated with 5-AzaC (Fig. 3B), probably resulting from the 5-AzaCinduced HMGCR knockdown (Fig. 2B).
To identify the target(s) of 5-AzaC in HepG2 cells, we used a DNA microarray approach and examined modifications in global gene expression. Whole genome expression analyses revealed that from a total coverage of 28,871 genes, 26,955 (93.4%) were not affected, 1,187 (4.1%) were down-regulated, and 729 (2.5%) were up-regulated more than 2-fold after treatment with 5-AzaC ( Fig. 4A and supplemental File 1). Gene ontology analyses highlighted that lipid and sterol biosynthetic, transport, localization, and organic acid and ketone metabolic processes were strongly down-regulated, with highly significant p values and -fold enrichments without major modification of other biological processes (Fig. 4A). Even more selectively, SREBPs target genes involved in the complete program of cholesterol (SREBP-2) and fatty acid biosynthesis (SREBP-1) (29) were significantly down-regulated by 5-AzaC (Fig. 4, B and  C). Thus, we hypothesized that 5-AzaC might have interrupted SREBP activation.
5-AzaC Reduces SREBP Nuclear Levels-Cellular sterol levels tightly regulate the activation of membrane-bound SREBPs.

5-Azacytidine Restricts SREBP Activation
When sterol levels are low, SREBPs are released from the ER retention protein INSIG and transported into COPII vesicles toward the Golgi apparatus with the help of SREBP cleavageactivating protein (SCAP) (9). SREBPs are then sequentially cleaved by site 1 (S1P) and site 2 (S2P) proteases to release their transcriptionally active N-terminal domain into the nucleus, which induces genes bearing a promoter-embedded SRE. SREBP-2 induces the expression of cholesterol metabolism genes (e.g. HMGCR, MVK, LDLR, and PCSK9), and SREBP-1 induces that of lipid metabolism genes (e.g. FASN, SCD, and FADS1) (9,29). Accordingly, to test the effect of 5-AzaC on SREBP-2 activity, we subcloned the proximal promoter of PCSK9, a SREBP-2-regulated gene, and that of TBP (TATA box-binding protein; invariable internal standard used for QPCR analyses) into Gaussia luciferase vector (Fig. 5A). As described previously (22), we mutated separately or in combination the SRE or hepatocyte nuclear factor 1 (HNF1) binding site motifs in PCSK9 promoter, which were shown to be crucial for its sterol-dependent gene expression. In HepG2 cells, relative secreted luciferase activities showed that 5-AzaC lowers exclusively the transcriptional activity of wild-type PCSK9 promoter (Fig. 5A), probably affecting nuclear SREBP-2 and/or HNF1 content. Consequently, HEK293 cells were transfected with a transcriptionally active nuclear form of SREBP-2, which bypasses the sterol-dependent regulation present in the ER (9). Under this condition, 5-AzaC did not alter the nuclear transport or content of truncated SREBP-2 and failed to repress its transcriptional activity (Fig. 5B). In sterol-deprived HepG2 cells, PCSK9, HMGCR, and LDLR mRNAs are markedly up-regulated, an effect that can be reversed by adding sterols (Fig. 5C) (32), which block the activation of SREBP-2 and reduce its nuclear form (Fig. 5D). Similarly to sterols, 5-AzaC decreased PCSK9 and HMGCR mRNA levels (Fig. 5C) resulting from reduced nuclear SREBP-2 protein levels (Fig. 5D). In addition, 5-AzaC also strongly lowered nuclear content of the lipogenic transcription factor SREBP-1 (Fig. 5D). Importantly, 5-AzaC did not affect SREBP-2 protein synthesis after 4 h of incubation as compared with CHX (Fig. 5E). Total SREBP-2 protein levels were decreased after a 24-h incubation with 5-AzaC, albeit less than with CHX, probably resulting from the autoregulatory feedback loop of SREBP-2 (33). Of note, protein levels of PCSK9, LDLR, or proprotein convertases PC5A and furin, all under the control of a CMV promoter  JULY 4, 2014 • VOLUME 289 • NUMBER 27 and overexpressed in HepG2 cells, were not regulated by 5-AzaC, indicating that protein synthesis is not affected (data not shown). Taken together, our data demonstrated that 5-AzaC most likely prevents SREBPs activation within the secretory pathway.

5-Azacytidine Restricts SREBP Activation
We next compared the direct effect of 5-AzaC on SREBP-2 synthesis and processing in M19, 25-RA, and AC29 CHO mutant cell lines. M19 cells are cholesterol auxotroph because they are deficient for S2P and unable to release membranebound SREBPs (19). 25-RA cells constitutively express active SREBPs due to the mutation D443N in the sterol-sensing domain of SCAP that prevents binding to INSIG and ER retention of SREBPs in the presence of sterols (18). Isolated from the 25-RA cell line, AC29 cells are also deficient in acyl-CoA:cholesterol acyltransferase 1 (ACAT1) and cannot esterify cholesterol within the ER (20). Upon treatment with 5-AzaC, we observed a reduction of both precursor and nuclear forms of SREBP-2 in parental CHO-K1 cells, but not in 25-RA and AC29 cells that are insensitive to sterols and constitutively activate SREBP-2 (Fig. 5F). In contrast, CHX blocked SREBP-2 protein synthesis in CHO-K1 as well as in 25-RA and AC29 mutant cell lines at levels comparable with S2P-deficient M19 cells (Fig.  5F). Therefore, the mRNA expression of the SREBP-2 target gene LDLR was found to be reduced by 5-AzaC only in CHO-K1 cells, suggesting impaired SREBP-2 activation and signaling (Fig. 5F, bottom). Of note, LDLR mRNA is not stabilized in 5-AzaC-treated CHO-K1 cells as compared with HepG2 cells (Fig. 2C). Importantly, we noticed that 5-AzaC induced the formation of LDs exclusively in CHO-K1 cells, emphasizing a direct involvement of both SREBP and ACAT1 (Fig. 5G). Therefore, we conclude that 5-AzaC directly interferes with cholesterogenic and lipid gene expression by promoting cholesterol accumulation in the ER and inhibition of SREBP processing.
Cross-talk between Pyrimidine and Glycerolipid Biosynthesis-GPAT3 (glycerol-3-phosphate acyltransferase 3) is an ER-associated enzyme that plays an important role during adipogenesis by catalyzing the initial step for de novo TG synthesis (Fig. 6A) (34). Our microarray and QPCR data revealed that AGPAT9 (1-acylglycerol-3-phosphate O-acyltransferase 9, the gene encoding for GPAT3) is among the highest up-regulated genes by 5-AzaC in HepG2 cells (ϳ20-fold; Fig. 6B and supplemental File 1). Moreover, 5-AzaC strongly reduced apolipoprotein B-100 (apoB100) secretion (Fig. 6C), which may be a result of cellular lipid diversion to LDs or disturbance of phospholipid homeostasis (35). Diacylglycerol kinase mRNA was also strongly decreased by 19-fold ( Fig. 6A and supplemental File 1). This prompted us to investigate in more detail the implication of the glycerolipid biosynthesis pathway in response to 5-AzaC. Quantification of the de novo glycerophospholipid biosynthesis confirmed a strong increase in TG and highlighted a significant decrease in cardiolipin synthesis (Fig. 6D). The addition of propranolol, a phosphatidic acid phosphohydrolase (PAP) inhibitor (Fig. 6A) (36), significantly tempered the effect of 5-AzaC on

5-AzaC Reduces Nuclear SREBP-2 and Expression of Its Target Genes in Mouse
Liver-Next, we tested the ability of 5-AzaC to decrease the expression of SREBP-2-regulated genes in vivo. C57BL/6 or hypocholesterolemic Pcsk9 Ϫ/Ϫ mice were injected intraperitoneally with a low dose of 5-AzaC (2.5, 5, or 10 mg/kg/ day; 10 mg/kg corresponds to one-third of the dose used in the clinic) (40). Similarly to HepG2 cells (Fig. 1, A and B), 5-AzaC strongly reduced in a dose-dependent manner Pcsk9 and Hmgcr expression and that of Hmgcr in the liver of C57BL/6 and Pcsk9 Ϫ/Ϫ mice, respectively (Fig. 8, A and B). In addition, circulating Pcsk9 was barely detectable in plasma of C57BL/6 mice injected with 10 mg/kg 5-AzaC (Ϫ95%; Fig. 8C). However, 5-AzaC did not significantly alter Ldlr gene expression, and even the strong reduction of Pcsk9 expression did not result in increased hepatic Ldlr protein levels for this short period of time (Fig. 8, A and D).
To assess the impact of multiple doses of 5-AzaC (similar to clinical dosage regimen for MDS patients) (3) on SREBP-2 tar-

5-Azacytidine Restricts SREBP Activation
get genes, C57BL/6 mice were injected intraperitoneally or subcutaneously with one, two, or five doses of 10 mg/kg/day 5-AzaC (Fig. 8E). Our QPCR data showed that at least two consecutive doses of 5-AzaC are needed to robustly decrease Pcsk9 (Ϫ77%) and Hmgcr (Ϫ68%) mRNA levels (saline versus two or five doses (2d or 5d); Fig. 8E), indicating that 5-AzaC could have been eliminated 48 h after a single injection (one dose (1d); Fig.  8E). Accordingly, circulating Pcsk9 was strongly reduced after two injections of 10 mg/kg 5-AzaC (Ϫ75%; Fig. 8F), which resulted in an increase in total Ldlr protein levels (ϳ1.7-fold; Fig. 8G). Based on our observations that 5-AzaC decreased nuclear SREBP-2 levels in HepG2 cells (Fig. 5), we next pro-ceeded to the subcellular fractionation of mouse liver cells. Western blot analyses of cell fractions indicated that, compared with saline, nuclear Srebp-2 protein levels were significantly reduced in livers of mice injected with two doses of 10 mg/kg 5-AzaC (Ϫ50%; Fig. 8H). These data provide further evidence that 5-AzaC severely interferes with the SREBP pathway and cholesterogenic and lipid gene expression both in vitro and in vivo.

DISCUSSION
5-AzaC and DAC can both inhibit DNA methylation but differ in terms of modulation of gene expression (5). Herein,  ). B and C, relative expression levels of SREBP-2 and SREBP-1 target genes extracted from microarray data (gray bars, DMEM; black bars, 5-AzaC). Listed are genes involved in cholesterol biosynthesis and regulation of low density lipoprotein uptake (LDLR and PCSK9) (B) and genes controlling fatty acid biosynthesis (C). Error bars, mean Ϯ S.D.  JULY 4, 2014 • VOLUME 289 • NUMBER 27 our data demonstrated that 5-AzaC, independently of DNA methylation, specifically down-regulates expression of SREBP-regulated genes in addition to increasing the formation of LDs. Despite the fact that ϳ95% of genes were unaffected by 5-AzaC, expression profiling analyses have revealed a highly significant enrichment for cholesterol and lipid metabolic processes among repressed transcripts ( Fig.  4A and supplemental File 1). Using subcellular fractionation, we showed that 5-AzaC strongly decreased nuclear levels of the transcription factors SREBP-1 and SREBP-2 (Figs. 5D and 8H). In addition, whether added before or after mevastatin treatment (condition 1 or 2, respectively; Fig. 5C),
Growing evidence shows that phospholipid and glycerolipid homeostasis is crucial for lipoprotein secretion, lipid storage, and sensitivity of SREBPs to ER cholesterol levels (35,42,43). Accordingly, we noticed that 5-AzaC promoted TG and LD formation and severely altered apoB100 secretion (Fig. 6, C-E). Our microarray data also revealed that MTTP (microsomal triglyceride transfer protein) and FABP1 (fatty-acid binding protein 1) were strongly downregulated by 5-AzaC in HepG2 cells (Ͼ8-fold; supplemental File 1), without globally affecting expression of apolipoproteins, lipoprotein receptors, cholesterol transporters, and other major regulators of lipid metabolism (e.g. ACAT1, ABCG5/8, MYLIP, and CH25H). MTTP and FABP1 are required to lipidate apoB100 and therefore to properly form nascent very low density lipoprotein particles (44,45). In case of interruption of very low density lipoprotein assembly or cholesterol and lipid overload within the ER, cells maintain homeostasis by promoting the formation of cytosolic LDs (46). Moreover, we noticed that AGPAT9 is among the highest up-regulated genes by 5-AzaC in HepG2 cells (supplemental File 1 and Fig. 6B). It has been shown that ectopic overexpression of AGPAT9 promotes lipogenesis by selectively increasing TG production without affecting phospholipid synthesis (34,47). Thus, we surmise that the combined dysregulation of MTTP, FABP1, and AGPAT9 gene expression by 5-AzaC contributes to the accumulation of cytosolic LDs and reduced SREBP processing. However, SREBP-2 is still processed in 25-RA CHO cells that spontaneously accumulate LDs regardless of the presence of 5-AzaC (Fig. 5, F  and G). Also, in 5-AzaC-treated HepG2 cells, the addition of propranolol prevented the formation of LDs (Fig. 6E) and partially rescued PCSK9, HMGCR, and AGAPT9 expression ( Fig. 6B) but had no obvious effect on apoB100 secretion (Fig. 6C). Therefore, the existence of a direct link between the effect of 5-AzaC on apoB100, LD formation, and activation of SREBPs remains to be determined.
As a ribose analog, 5-AzaC can also be incorporated into different RNA species and affect their functions (6). Nonetheless, our results indicate that the specific effects we observed on SREBP-2 targets, except for LDLR stabilization, do not involve mRNA degradation (Fig. 2C) or inhibition of protein synthesis (Fig. 5, E and F). In addition, 5-AzaC did not directly inhibit CTP-phosphocholine cytidylyltransferase or CDS activity. Hence, we reasoned that it might interrupt SREBP signaling through the direct inhibition of UMP synthase (6), which is crucial for the synthesis of CTP required for TG and glycerophospholipid biosynthesis pathways (Fig. 6A) (37,39). Indeed, co-incubation with either UMP or cytidine entirely reversed the effects of 5-AzaC (Fig.  6, B-D). Our data strongly suggest that, by reducing the de novo synthesis of CTP needed for CDP-diacylglycerol formation by CDS, 5-AzaC selectively reroutes the conversion of PA toward TG synthesis, which provokes an accumulation of intracellular LDs and a reduction of cardiolipin synthesis (Fig. 6). Interestingly, our microarray data also revealed that diacylglycerol kinase (Fig. 6A and supplemental File 1) mRNA is strongly reduced by 5-AzaC, suggesting that CTP depletion selectively targets the synthesis of glycerolipids. Although very little is known about AGPAT9 and DGK (diacylglycerol kinase) gene regulation, it is conceivable that by indirectly reducing CDS enzymatic activity, 5-AzaC may disturb key intermediate metabolites downstream of PA or may act via diacylglycerol, which was shown to regulate a number of intracellular signaling networks (Fig. 6A) (37).
One of the major hallmarks of cancer cells is their capacity to reprogram metabolic pathways to stimulate the biosynthesis of proteins, cholesterol, and lipids (48). Rapid cell proliferation requires increased uptake of lipids and de novo lipogenesis to continuously provide mevalonate pathway metabolites, cholesterol, and fatty acids needed for cell membrane synthesis, cell signaling, post-translational modifications of proteins, and energy supply (12,49). Increased  JULY 4, 2014 • VOLUME 289 • NUMBER 27 SREBP signaling in bone marrow and peripheral blood cells of patients with MDS has been associated with poor survival prognosis (50). In addition, it was demonstrated that the activation of the SREBP pathway protects cancer cells from lipotoxicity (51) and promotes growth of activated CD8 ϩ T lymphocytes (52), which expand in MDS patients (53). Through the mevalonate pathway, SREBPs also provide key intermediates required for the isoprenylation of small GTPases, such as farnesylation of Ras and geranylgeranylation of Rho, both involved in cancer progression and metas-tatic dissemination (54). Accordingly, recent studies revealed that statin therapy, which inhibits HMG-CoA reductase and blocks mevalonate and sterol synthesis, significantly improved overall survival in patients with MDS and acute myeloid leukemia (55,56).

5-Azacytidine Restricts SREBP Activation
SREBP signaling could contribute to the beneficial cytostatic effect of 5-AzaC in MDS patients. This work has revealed a previously unrecognized cross-talk between pyrimidine and glycerolipid synthesis and cholesterol-regulated activation of SREBPs and highlights new potential molecular targets with the aim of preventing hematological disorders.