Crosstalk between ATF4 and MTA1/HDAC1 promotes osteosarcoma progression

The stress response gene activating transcription factor 4 (ATF4) is involved in metastatic behavior and cellular protection. Here we show that ATF4 is upregulated in osteosarcoma (OS) cell lines and patient clinical samples as compared to matched non-tumor tissue. Overexpression of ATF4 in OS cells promoted cell proliferation, migration and lung metastasis. Furthermore, the expression of ATF4 was markedly reduced in metastasis associated protein (MTA1) or histone deacetylase 1 (HDAC1) knockdown OS cells, but MTA1 overexpression increased the stability and activity of ATF4 protein via ATF4 deacetylation by HDAC1. ATF4 in turn enhanced the expression of MTA1 and HDAC1 at the transcription level, suggesting a positive feedback loop between ATF4 and MTA1/HDAC1. Clinically, the level of ATF4 was positively correlated with that of MTA1 in OS. Mice injected with ATF4-overexpressing cells exhibited a higher rate of tumor growth, and the average weight of these tumors was ×90% greater than the controls. Taken together, these data establish a direct correlation between ATF4-induced OS progression and MTA1/HDAC1-associated metastasis, and support the potential therapeutic value of targeting ATF4 in the treatment of OS.


INTRODUCTION
Osteosarcoma (OS) is a highly malignant bone tumor in children and adolescents [1]. In 15-30% of OS cases, pulmonary metastases are observed at initial diagnosis [2]. Although several signaling pathways have been found to be involved in OS pathogenesis, the molecular mechanisms underlying OS progression have not yet been fully elucidated. It is necessary to identify new targets to enhance the efficacy of OS treatment and improve long-term survival of OS patients.
Activating transcription factor 4 (ATF4) belongs to the ATF/cyclic AMP response element binding protein (CREB) family [3]. ATF4 expression is increased in response to stresses in the tumor microenvironment [4]. Paradoxically, ATF4 was shown to activate both proliferation and apoptosis. For example, ATF4 increased VEGF-dependent tumor growth triggered by osteopontin, whereas excessive ATF4 activation was separately shown to promote tumor cell death [5,6]. Despite being considered a global regulator of osteoblast biology and bone metabolism [7], the role of ATF4 in OS is poorly understood.
Metastasis-associated protein (MTA), including MTA1, MTA2, and MTA3, is a component of the nucleosome remodeling and histone deacetylation (NuRD) complex [8]. MTA1 expression and its clinicopathologic relevance have been widely investigated in various cancers, including OS [9,10]. MTA1 functions as a transcriptional coactivator or corepressor of various genes. In most human cancers, overexpression of MTA1 promotes disease progression [11]. However, crosstalk between ATF4 and MTA1 remains largely unstudied.
The present study was aimed at exploring the function of ATF4 in OS. Our results revealed that MTA1/ HDAC1 promoted stabilization of ATF4 protein. ATF4 increased MTA1 and HDAC1 expression, suggesting a positive feedback loop between ATF4 and MTA1/HDAC1. In addition, ATF4 showed pro-tumor activity in vitro and in vivo, suggesting the potential therapeutic value of targeting ATF4 in the treatment of OS.

ATF4 overexpression promotes OS growth and metastasis
Immunohistochemical staining showed that ATF4 expression was significantly higher in clinical OS tissue than in control samples ( Figure 1A), suggesting that ATF4 plays an important role in OS development. Compared with the control, ATF4 overexpression promoted proliferation in U2OS cells ( Figure 1B and 1C), and ATF4-U2OS cells exhibited increased invasive potential to an extent greater than 80% ( Figure 1D), and higher migratory ability ( Figure 1E), when compared to vector-U2OS cells.
To examine ATF4 in vivo effects, control or ATF4-U2OS cells were subcutaneously injected into the right flanks of mice and tumor growth was evaluated. Figure 1F indicates that all mice injected with ATF4-U2OS cells exhibited a higher rate of tumor growth compared with control cells, and the average weight of ATF4-U2OS tumors was ~90% more than that of control tumors. ATF4-U2OS-injected mice also exhibited earlier tumor growth than controls.
As shown in Figure 1G (left panel), all mice in the ATF4 group had gross evidence of pulmonary metastatic lesions, whereas no lung nodules were visible in the control group, which was further confirmed with H&E staining (middle panel). The average number of pulmonary nodules in the ATF4 group was 9.2, as compared to 0.2 in the control group (right panel). Additionally, ATF4 knockdown inhibited OS growth and metastasis (data not shown).

MTA1 physically interacts with ATF4
As both MTA1 and ATF4 proteins are expressed in malignant OS [12], we investigated the possibility of crosstalk between MTA1 and ATF4. To this end, we first examined whether the two proteins could interact. Even though HEK293 cells have endogenous expression of MTA1 or ATF4 (Figure 2A), Myc-MTA1 coimmunoprecipitates Flag-ATF4 in HEK293 cells ( Figure 2B, left panel). We also observed this binding in a reverse experiment ( Figure 2B, right panel). We also found evidence of binding between the endogenous MTA1 and ATF4 in 143B and ZOS cells ( Figure 2C). To analyze the region of ATF4 required for binding to MTA1, we constructed several deletion mutants based on its reported domain structure ( Figure 2D), and applied them in an in vitro GST pull-down assay. The result demonstrated that both the p300 interaction site (1-90) and the central region including ODDD and β-TrCP recognition motifs (140-230) of ATF4, corresponded to the MTA1 binding domain. However, the C-terminal domain (270-351) was not required for the interaction with MTA1 ( Figure  2D). Thus, recombinant MTA1 directly binds with the N terminus of ATF4 (amino acids 1-224).

MTA1 stabilizes ATF4 protein by inhibiting its ubiquitination
Next, using MTA1+/+ or MTA1-/-MEF cell lines [13], we examined the role of MTA1 depletion on the levels of endogenous ATF4. Compared with ATF4 levels in wildtype MEFs, a marked reduction in the level of ATF4 protein in the MTA1-/-MEFs was observed ( Figure 3A). However, this ATF4 reduction could be reverted by reintroduction of MTA1 in the MTA1-/-MEFs ( Figure 3B). We found no discernible differences in ATF4 mRNA levels among the MTA1+/+, MTA1-/-, and MTA1-/-/MTA1 MEFs ( Figure  3C), suggesting a post-translational role of MTA1 in increasing the level of ATF4 protein. Furthermore, MTA1 knockdown in the 143B cells induced a reduction in ATF4 protein ( Figure 3D), whereas overexpression of MTA1 in the 143B cells increased the steady-state levels of ATF4 ( Figure 3E). Treatment with cycloheximide to inhibit protein synthesis further indicated that ATF4 stability increased in cells cotransfected with MTA1, but decreased in MTA1-depleted cells (data not shown).
ATF4 stability is determined through a series of post-translational modifications, including ubiquitination, which primarily relies on a phosphorylation-dependent interaction with the SCF (β-TrCP) ubiquitin ligase [14]. To test whether MTA1 stabilizes ATF4 by inhibiting its ubiquitination, we examined the role of MTA1 on endogenous ATF4 ubiquitination in 143B-MTA1 stable cells. As shown in Figure 3F, overexpression of MTA1 decreased endogenous ATF4 ubiquitination and increased ATF4 protein. Similarly, Myc-MTA1 expression in HEK293 cells also suppressed ATF4 ubiquitination ( Figure  3G). These findings suggest that MTA1 stabilizes ATF4 protein at least in part by inhibiting its ubiquitination.
We also measured the transcriptional activity of ATF4 using the luciferase reporter vector containing six tandem copies of ATF4 binding site (6 x ARE) [15]. Cotransfection of the MTA1 vector into U2OS cells activated the reporter activity, even in the absence of stress ( Figure 3H).

MTA1 promotes binding of ATF4 with HDAC1, which leads to ATF4 deacetylation
It was shown that MTA1 increased the stabilization of HIF-1α and p53 proteins by deacetylation, which is critically determined by HDAC1 recruitment [16]. Therefore, we examined whether HDAC1 was involved in the MTA1induced stabilization of ATF4. As shown in Figure 4A, the www.impactjournals.com/oncotarget    Protein extracts from indicated cells were subjected to IP with an anti-ATF4 antibody, followed by western blot analysis with the indicated antibodies F. HEK293 cells were transfected with the indicated plasmids and subjected to IP and western blot analysis as described above G. The ARE-tk-Luc (0.1 μg) reporter was cotransfected with the indicated amount of eukaryotic expression vector for MTA1 into U2OS cells H. Transfected cells were incubated for 24 hours in the presence or absence of 100 μM CoCl 2 , and luciferase activity was measured. Values are expressed as relative luciferase activity normalized to cotransfected renilla luciferase activity. expression of HDAC1 in 143B cells was largely induced under Thapsigargin (Tg)-treated ER stress, or via the overexpression of MTA1. Since there is a possibility that HDAC1 counteracts the acetylation of ATF4 and inhibits its degradation, we examined whether trichostatin A (TSA), a potent specific inhibitor of HDAC, decreased the stability of ATF4. As expected, TSA decreased the amount of ATF4 that was induced by either Thapsigargin or MTA1 expression, indicating that HDAC1 activity is closely correlated with the stability of ATF4.
Next, we examined the relative amounts of acetylated ATF4 in the presence of MTA1. When cells were treated with MG132, a proteosome inhibitor, ATF4 acetylation increased significantly, but acetylated ATF4 was not detected in the presence of MTA1 or Thapsigargin. Similarly, compared with MG132 alone, treatment with MG132 and Thapsigargin or MTA1 significantly decreased the amount of acetylated ATF4. However, TSA treatment significantly increased ATF4 acetylation in the presence of MTA1 and MG132, suggesting that HDAC1 Transfected cells were treated with or without thapsigargin (Tg) as indicated, then treated with or without 300 ng/ml TSA for 3 hours. Cells were collected and lysed with RIPA buffer. The expression of ATF4, HDAC1, MTA1, Myc-MTA1, and r-tubulin in whole lysates was analyzed by western blot (WB) analysis. 143B cells were treated with 10 μM MG132 or 300 ng/ml TSA for 1 or 3 hours, respectively B. Cell lysates (700 μg) were immunoprecipitated (IP) with anti-pan-Ac antibody, and then probed using anti-ATF4 antibody. The expression of indicated proteins was analyzed by western blot analysis. 143B cells were treated with Tg for 24 hours C. Cell lysates were immunoprecipitated with normal anti-IgG, anti-MTA1, or anti-ATF4 antibodies, and then probed using anti-HDAC1 antibody. The expression of ATF4, HDAC1, MTA1, and r-tubulin in the input was analyzed by western blot analysis. www.impactjournals.com/oncotarget deacetylates ATF4 and thereby blocks ATF4 degradation ( Figure 4B).
Given that MTA1 protein binds physically to HDAC1 [16], we examined the association between ATF4, MTA1, and HDAC1 in the presence of Thapsigargin. As shown in Figure 4C, MTA1, HDAC1 and ATF4 were efficiently co-precipitated, indicating that these proteins are capable of forming a complex in vivo in the presence of Thapsigargin. Together, our data indicate that MTA1 inhibits the acetylation of ATF4 by recruiting HDAC1, and increases the stabilization of ATF4 in OS, suggesting that HDAC inhibitors could potentially be used to treat OS.

MTA1 dictates the binding of ATF4 to PHD1, PHD3 and β-TrCP in OS cells
ATF4 protein abundance is dictated by two prolyl hydroxylase domain-containing proteins, PHD1 and PHD3, in mammalian cells, but not by PHD2 [17]. These PHDs belong to the superfamily of oxygenases that require O 2 , 2-oxoglutarate (2-OG), and Fe 2+ for their enzymatic activity [17]. The central region of ATF4 including the ODD domain and β-TrCP recognition motif is involved in binding to PHD1. The F-box protein β-TrCP, the receptor component of the SCF E3 ubiquitin ligase, is colocalized in the nucleus with ATF4 and decreases its stability [14]. Because MTA1, HDAC1, PHD1, PHD3 and β-TrCP bind to ATF4 protein and control its stability [14], we investigated whether MTA1 decreases the binding of ATF4 to each of these. In OS cells, both PHD3 and β-TrCP, but not PHD1, strongly bound to ATF4 in the presence of MG132, suggesting that binding of PHD3 and β-TrCP to ATF4 precedes ATF4 degradation. However, in the presence of MTA1, binding of ATF4 to PHD3 or β-TrCP was largely decreased, whereas binding between ATF4 and PHD1 increased ( Figure 5A). Similar results were observed when the ODD domain of ATF4 was used ( Figure 5B). Moreover, MTA1 knockdown in OS cells significantly reversed acetylation and binding of ATF4 ( Figure 5C), indicating that MTA1 predominantly blocks ATF4 ubiquitination via ATF4 deacetylation.

ATF4 transcriptionally activates the expression of MTA1 and HDAC1 in OS
Given that ATF4 is a transcriptional factor, we studied whether ATF4 enhances expression of MTA1 and HDAC1 in OS cell lines in which expression of the ATF4 gene is under control of an inducible doxycycline promoter. Expression levels of MTA1 and HDAC1 proteins were dramatically increased with doxycycline treatment ( Figure 6A). Moreover, the mRNA levels of MTA1 and HDAC1 were increased, indicating that ATF4 induces the expression of MTA1 and HDAC1 genes at the transcription level ( Figure 6B).
Analysis of the human MTA1 and HDAC1 promoters revealed potential binding sites for ATF4, and ChIP assays detected stress-inducible binding of ATF4 to these sites ( Figure 6C). siRNAs against ATF4 significantly reduced MTA1 and HDAC1 transcription after ER stress ( Figure 6D).
We also found that ATF4 expression dramatically decreased the amount of ubiquitinated MTA1 protein ( Figure 6F). Together, these data reveal an unexpected feedback loop among ATF4, MTA1 and HDAC1, which is likely to have significant consequences in environmentally compromised OS tissues and cells.

Correlation between ATF4 and MTA1 abundance
We found a significant positive correlation between the abundance of ATF4 and MTA1 in nine OS cell lines and in a cohort of OS patients ( Figure 7A and 7B). Together, these results strongly support the hypothesis that the observed crosstalk between ATF4 and MTA1 has physiological importance in OS.

DISCUSSION
In this study, we show that ATF4 promotes OS growth and metastasis. MTA1 stabilizes ATF4 by recruiting HDAC1, and ATF4 transcriptionally activates the expression of MTA1 in OS. Two mechanisms that could account for ATF4 protein stability have been described: (a) SCFβ-TrCP degrades ATF4 and (b) histone acetyltransferase p300 increases ATF4 stabilization by inhibiting ATF4/β-TrCP binding and subsequent degradation [14,18,19]. Although p300 can acetylate ATF4, its role in ATF4 stabilization is not due to p300 catalytic activity. p300 binds to the N-terminal region of ATF4, leading to sequestration of ATF4 into the nuclear speckles, where it is unavailable for β-TrCP binding [18]. This provides insight into a new mechanism for stabilization of ATF4 by MTA1, in which enzymatic activity of HDAC1 and deacetylation of ATF4 are required, and MTA1 inhibits binding of ATF4 with β-TrCP. Phosphorylation by an unknown kinase is probably required for subsequent recognition by SCFβ-TrCP [14]. Our data suggest that HDAC1 binds to ATF4 and promotes ATF4 stability via deacetyltransferase activity. Hence, we propose a molecular basis for the cell specificity of ATF4 accumulation in OS. www.impactjournals.com/oncotarget ATF4 was shown to induce expression of several target genes, including asparagine synthesase, Gadd153, VEGF, TRB3, osteocalcin, TFIIB and RAP30 [20,21,22]. Additionally, ATF4 transactivated gene expression by binding to various partners, including Nrf2, to induce heme oxygenase-1 (HO-1), or through its deacetylation by HDAC1 [23,24]. We demonstrated that MTA1 and HDAC1 were induced by ATF4 in OS, suggesting a positive feedback loop between ATF4 and MTA1/HDAC1.
Given that ATF4 is upregulated by signals in the tumor microenvironment, such as hypoxia [12], and is closely correlated to angiogenesis and adaptation of cancer cells to hypoxia [25], inhibiting ATF4 might be especially attractive for treating OS patients. Kai et al reported a novel pathway of chromatin remodeling by blocking the MTA1/HDAC1 complex, and suggested a combined treatment of resveratrol, which decreased MTA1 expression, and the HDAC inhibitor, SAHA [26]. Although this treatment cooperatively increased p53 levels and apoptosis in prostate cancer, it was unclear whether ATF4 expression was changed. In sum, our finding indicate that the ATF4/MTA1/HDAC1 axis promotes the development of OS. Therefore, inhibitors of HDAC1, ATF4 and MTA1 may prevent OS carcinogenesis.

Cell culture, antibodies and reagents
U2OS and 143B human osteosarcoma cell lines were originally purchased from ATCC, and cultured in DMEM and RPMI 1640 medium supplemented with 10% fetal bovine serum, respectively. ZOS cells were established from the primary tumor of an osteosarcoma patient and described previously [13,27]. Human osteosarcoma 143B cells (ATCC ® CRL-8303™), are suitable transfection host cells, thymidine kinase deficient (TK-) and resistant to BUdR [28]. Transfections were carried out using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instruction. All reagents in this study were of analytical grade and are commercially available. All mouse embryonic fibroblasts (MEFs) cell lines used in this study are from Dr. Rakesh Kumar's laboratory and have been described [13]. To establish cell lines stably expressing MTA1, MTA1-/-MEFs were transfected with expression vector encoding pcDNA6/V5-MTA1 or empty vector using FuGENEHD Transfection Reagent (Roche Applied Science, Indianapolis, IN). After 24 hours of transfection, cells were subjected to selection with 10 μg/ml blasticidin (Invitrogen) for 2 weeks. The resulting colonies were isolated and analyzed for V5-MTA1 expression by immunoblotting.

RNA extraction, semiquantitative RT-PCR and real-time qPCR
Total RNA extraction was performed using an RNeasy kit (Qiagen). RNA concentration and purity were measured using a NanoDrop spectrophotometer (Thermo Scientific). RNA was retro-transcribed using SuperScript III reverse transcriptase (Life Technologies) for use in subsequent PCR reactions. Platinum TAQ (Life Technologies) was employed for semi-quantitative RT-PCR. Real time qPCR was performed using TaqMan universal master mix II (Life Technologies) and different

Cell viability assay
OS cells were cultured overnight in a 96-well plate at a density of 5 × 10 3 cells/well. Control cells or cells stably overexpressing ATF4 were cultured in the presence of 5% FBS for 48 hours. Cell viability was assessed by MTT assay (Thermo Scientific, USA). The number of surviving cells was assessed by measuring absorbance at 490 nm for the dissolved formazan product after addition of 20 μL MTT solution (0.5 mg/ml) for 4 hours. Cell counting kit-8 (CCK-8; Dojindo, Japan) was used to determine OS cell proliferation based on the manufacturer's protocol. Briefly, cells were seeded in 96-well plates at a density of 5 × 10 3 cells/well and were treated as MTT assay. At different time points, 10 μL of CCK-8 solution was added to each well and incubated for 1 hour at 37°C. Absorbance (optical density, OD) was measured at 450 nm.

Wound healing and transwell invasion assay
For migration assay, cells were seeded in 6-well plates and grown to 80% confluency. The cell monolayers were scraped off with a sterile plastic tip and were cultured in DMEM with 1% FBS under normoxia (20% O 2 ) or hypoxia (1% O 2 ). Pictures were taken under a microscope at 0 and 24 hours points, and the distance between the two boundaries of an acellular area was measured. The results of the different treatment groups were expressed as a ratio to the distance of 0 hour point. In vitro invasive activity through a gel matrix was examined in 24-well plates. Cells were resuspended at a density of 1 × 10 5 cells/ ml. BD Falcon Cell Culture Inserts (San Jose, CA, USA) were coated with 50 μL of Matrigel and placed in each well. Wells were filled with 600 μL of complete medium, and the upper inserts were filled with 100 μL of the cell suspension. After 24 hours, viable invasive cells that adhered to the lower surface of the filter were fixed using 70% methanol and stained with 0.5% crystal violet. Cells on the upper surface of the filter were removed with cotton swabs. The cells in the lower level were counted.

Establishment of an in vivo lung metastasis model
Male BALB/c nude mice aged 3 to 4 weeks were purchased from Beijing Medical Experimental Animal Center (Beijing, China). All animal protocols were approved by the Institutional Animal Research Committee of Tongji Medical College. Control cells or cells stably overexpressing ATF4 or shATF4 were suspended in PBS and then subcutaneous and intramuscular injected into one nude mouse. Mice were sacrificed and the tumors were isolated. Tumor volume was calculated as (length × width 2 )/2. Volumes were calculated twice a week. Control tumors were harvested on days 34 and 40, and tumor weights were measured. Lung samples were harvested for further examination, fixed with formalin and embedded in paraffin. The number of lung metastases was determined by counting the number of metastatic nodules on the lung surface. H&E staining was performed to detect the lung metastases.

Immunohistochemistry
Dewaxed and rehydrated tissue sections were incubated overnight at 4°C with indicated antibodies and then washed with PBS. Biotinylated goat anti-rabbit IgG was then added to the sections for 30 minutes at RT. After the sections were washed with PBS, peroxidaseconjugated avidin was applied. Peroxidase activity was detected by exposing the sections to a solution of 0.05% 3, 3-diaminobenzidine and 0.01% H 2 O 2 in Tris-HCl buffer (3, 3-diaminobenzidine solution) for 3 to 6 minutes at RT. The sections were counterstained with hematoxylin. Images were scanned and analyzed using an Aperio Imagescope (Leica Biosystems). Immunohistochemical staining for MTA1/ATF4 in tissue was scored by two of the authors as positive or negative. An estimation of the immunohistochemical expression of the markers was determined by counting ≥100 cells in random highpower fields. The frequency of positive cells for each antibody was reported semi-quantitatively as follows: (−), No reaction; (+), mild with <30% of positive cells; (++), moderate with 30-60% of positive cells; and (+++), marked with >60% of positive cells. A positive expression was defined as the staining of >30% of the cancer cells (++ or +++).

Immunoprecipitation and western blot analysis
Protein extracts were obtained from tissues or cells lysed in buffer containing Complete Protease Inhibitor Cocktail (Roche Applied Science, Germany). For immunoprecipitation assays, cells were lysed using the Pierce Co-IP kit (Thermo Scientific, USA) according to the manufacturer's instructions. Proteins were resolved by 10% SDS-PAGE, transferred to PVDF membranes (Millipore, USA), incubated with the appropriate primary and secondary antibodies and detected using the ImmunoCruz Western Blotting Luminol Reagent (Santa Cruz, USA).

Luciferase assay
Reporter vectors containing six tandem copies of ATF4 binding site (6 x ARE) were co-transfected with a SV40-Renilla luciferase construct into U2OS cells using Oligofectamine™ reagent (Invitrogen). Two days after transfection, cells were lysed, and luciferase activity was determined using specific substrates in a luminometer according to the manufacturer's protocol. Transfection efficiency was normalized by determining the activity of Renilla luciferase.

Clinical tissue samples
One hundred thirteen OS and paired adjacent noncancerous tissue specimens were collected between 2011 and 2014 from the Tongji Hospital, Huazhong University of Science & Technology. The study was approved by the Committee for Ethical Review of Research in China, and the patients signed informed consent forms. All tissue samples were obtained from patients with OS. Samples were collected during surgery and were immediately snap frozen in liquid nitrogen until RNA and protein extraction.

Statistics
All graphs were generated and all statistical analyses (Student's two-tailed t test and log-rank test) were performed with GraphPad Prism software (GraphPad Software Inc.). Data shown are the mean ± SD of three independent experiments. *p < 0.05; **p < 0.001.