ATF3 Repression of BCL-X_L Determines Apoptotic Sensitivity to HDAC Inhibitors across Tumor Types

Histone deacetylase inhibitors (HDACi) are epigenome-targeting anticancer therapeutics with established clinical activity in several hematological malignancies (1). A number of distinct chemical classes of HDACi have been identified or developed including short-chain fatty acids (butyrate, valproic acid), hydroxamic acids (trichostatin A, vorinostat, belinostat, panobinostat, and pracinostat), tetrapeptides (romidepsin), and benzamidines (entinostat; ref. 2). Vorinostat and romidepsin are approved for the treatment of cutaneous T-cell lymphoma (CTCL; ref. 3), and belinostat is approved for the treatment of peripheral T-cell lymphoma (PTCL). In addition, combinatorial use of panobinostat with the proteasome inhibitor bortezomib is approved for refractory multiple myeloma (4), and pracinostat was recently granted breakthrough therapy designation with azacytidine in acute myelogenous leukemia (AML; ref. 5). Although responses to single-agent HDACi are limited in solid tumors (6), studies in non–small cell lung cancer and estrogen receptor–positive advanced breast cancer suggest they may have efficacy in combination therapy regimens (7, 8).

HDACis inhibit class I (HDACs 1, 2, 3, and 8) and class II HDACs (HDACs 4, 5, 6, 7, 9, and 10), which deacetylate lysine residues on target proteins (2). HDACis activate gene expression by inducing hyperacetylation of DNA-bound core histones, thereby increasing accessibility of the core transcriptional apparatus to DNA (1), or by hyperacetylating transcription factors, which can either increase or decrease their transcriptional activity (1). In addition, HDACi can elicit cellular effects independent of transcription by acetylating cytoplasmic proteins such as Hsp90 and tubulin (9, 10).

Although HDACis induce multiple effects on tumor cells, including inhibiting proliferation and inducing differentiation (1), their primary mechanism of antitumor activity is through the induction of apoptosis (2). In this regard, HDACis induce apoptosis primarily through the intrinsic/mitochondrial pathway (11), although in some tumor cell lines, the extrinsic/death receptor pathway is also activated (12, 13). HDACi-induced apoptosis has been linked with altered expression of key apoptotic regulators including upregulation of the proapoptotic molecules BAX (14), BAK (15), APAF1 (16), BMF (17), BIM (18), and DR5 (19), and downregulation of the antiapoptotic proteins SURVIVIN (20), BCL-X_L (21), and c-FLIP (22). However, HDACi regulation of these factors varies between cell type, and has not been systematically linked to apoptotic response (23). Furthermore, the mechanisms by which HDACis regulate the expression of pro- and antiapoptotic genes are only partially understood.

We previously identified a robust transcriptional response specifically associated with HDACi-induced apoptosis in colorectal cancer cell lines. This response involved the coordinate induction of multiple immediate-early (IE) response genes (FOS, JUN, EGR1, EGR3, ATF3, ARC, and NR4A1) and stress response genes (NDRG4, MT1E, MT1F, and GADD45B; ref. 24). The goals of this study were to determine whether this represents a generic transcriptional response that defines HDACi-induced apoptosis across tumor types, including CTCL and multiple myeloma where these agents currently have the greatest clinical activity. Second, we sought to determine whether this transcriptional response underpins HDACi-induced apoptosis by regulating expression of key apoptotic regulators.

Herein, we demonstrate that HDACis robustly induce expression of the IE genes FOS, JUN, and ATF3 in multiple tumor cell types, which correlated significantly with the magnitude of HDACi-induced apoptosis. We also demonstrate induction of these genes in 2 patients with CTCL treated with panobinostat. Functional studies revealed that ATF3 but not FOS or JUN was required for HDACi-induced apoptosis across tumor cell lines, and that the effects of ATF3 were mediated through repression of the prosurvival gene BCL-X_L (BCL2L1). These data provided a rationale for combining HDAC and BCL-X_L inhibitors, which successfully overcame inherent resistance to HDACi in a range of tumor types. Our findings establish the induction of ATF3 and subsequent repression of BCL-X_L as a consistent and key determinant of HDACi-induced apoptosis independent of tumor type. They also define the molecular basis for differential sensitivity to HDACi and identify avenues for predicting response and overcoming inherent resistance to HDACi through rational combination therapy.

All cell lines used for this study were obtained from the American Type Tissue Culture Collection or as gifts from collaborators listed in the Acknowledgments section. A total of 50 human cancer cell lines derived from multiple tumor types were used: Solid tumor cell lines used were PC-3, DU-145, LNCAP (prostate); HT-1197, HT-1376, 5637 (bladder); SK-MEL-3, SK-MEL-5, SK-MEL-28 (melanoma); MDA-MB-231, MDA-MB-468, MCF-7 (breast); A549, NCI-H292, NCI-H460, NCI-H358, NCI-H1650, NCI-H1975 (lung); RKO, LIM1215, Colo320, SW48, HCT116, SW948 (colon), IGROV1, SK-OV-3, JAM, OVCAR-8, OVCAR-5 (ovarian), OU-87 (glioblastoma); PANC-1 (pancreatic); ACHN (renal); 293T (embryonic kidney); A431 (epidermis); AGS (gastric), and Hep3B (hepatoma). Hematologic cancer cell lines used were HH, HuT-78, HuT-102, MJ (cutaneous T-cell lymphoma); Jurkat, Raji, U937 (lymphoma); LP-1, OPM-2, RPMI-8226, U266 (multiple myeloma); and K-562, KG-1, and KG-1A (leukemia). Cells were maintained at 37°C and 5% CO₂ in base medium DMEM for solid tumor cell lines or RPMI for hematogic cancer cell lines. Base medium was supplemented with 10% FCS, 2 mmol/L l-glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin. Wild-type and Atf3^−/− mouse embryonic fibroblasts were maintained in low-glucose DMEM supplemented with 10% FCS, 2 mmol/L l-glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin at 37°C in 10% CO₂. Methods for cell maintenance have been previously described (25). WT and FLAG-tagged hBCL-X_L–transduced mouse embryonic fibroblasts were maintained in DMEM, high-glucose media supplemented with 10% (v/v) FBS, 250 μmol/L l-asparagine, 50 μmol/L 2-mercaptoethanol, and 1 μmol/L HEPES. Cell lines were assessed for mycoplasma status using the MycoAlert assay (Lonza) and mycoplasma-negative frozen stocks used for a maximum of 2 months. Authenticity of frozen stocks of the A549, AGS, HCT116, PC3, U87, RPMI-8226, SKMEL28, MCF7, PANC1, HH, RKO, LIM1215, Colo320, SW48, and SW948 cell lines was determined by short-tandem repeat profiling using the GenePrint 10 system (Promega), and all found to be exact matches with published profiles.

Sodium-butyrate (NaBu) and valproate were obtained from Sigma. Vorinostat, belinostat, depsipeptide, entinostat, ABT-737, and ABT-199 were obtained from Selleck Chemicals. ABT-263 was obtained from ApexBio. Synthesis of A-1331852 was as described previously (26).

Apoptosis assays were performed as previously described by propidium iodide (PI) staining and FACS analysis (25). Cells were seeded in triplicate in 24-well plates. Seeding densities varied between 30,000 and 90,000 cells per well and were calculated such that control cell density approximated 80% confluence at the completion of the experimental period. Drug treatment was performed for 24 to 72 hours. Both attached and floating cells were harvested by scraping, washed in cold PBS, and resuspended in 50 μg/mL PI, 0.1% sodium citrate, and 0.1% Triton X-100. Cells were stained overnight at 4°C, and 10,000 cells were analyzed for DNA content using a BD FACS Canto II (BD Biosciences). The percentage of cells with a subdiploid DNA content was quantified using ModFit LT (Verity Software House).

Whole blood was collected in sodium-heparin tubes from 2 patients diagnosed with cutaneous T-cell lymphoma who participated in a single-arm, open-label, institutional phase II panobinostat trial (Clinicaltrials.gov identifier: NCT01658241). Patients received 30 mg panobinostat orally, 3 times weekly for up to 4 weeks. Both patients had >70% tumor involvement in peripheral blood mononuclear cells (PBMC). PBMCs were isolated by density centrifugation (Lymphoprep), according to the manufacturer's instructions. RNA from PBMC was purified and subjected to gene expression analysis using qRT-PCR. The clinical protocol, informed consent form, and other relevant study documentation were approved by the Institutional Review Board of the Peter MacCallum Cancer Centre. All patients gave written informed consent prior to study entry.

Total RNA was extracted using the RNeasy Mini Kit (Qiagen) and reverse-transcribed using random hexamers and the Transcriptor High Fidelity cDNA Synthesis Kit (Roche), according to the manufacturer's instructions. Quantitative RT-PCR was performed using Power SYBR Green PCR Master Mix (Applied Biosystems) on a 7500 Fast Real-Time PCR System (Applied Biosystems) according to the manufacturer's instructions. cDNA (10 ng) was amplified with 75 nmol/L forward and reverse primers in a 15 μL reaction. Primers used are listed in Supplementary Table S1.

Western blot analysis was performed as previously described (27). The source and dilutions of antibodies used are as follows: Rabbit anti-ATF3 (sc-188, Santa Cruz Biotechnology, 1:1,000), rabbit anti-FOS (cst-4384, Cell Signaling Technology, 1:1,000), mouse anti–c-JUN (cst-2315, Cell Signaling Technology, 1:1,000), rabbit anti–Ac Histone H3 (06-599, Merck Millipore, 1:10,000), goat anti–Histone H3 (sc8654, Santa Cruz Biotechnology, 1:5,000), rabbit anti–beta Tubulin (ab6046, Abcam, 1:20,000), mouse anti-actin (A5316, SIGMA, 1:10,000), and rabbit–anti-BCL-X_L (54H6, Cell Signaling Technology, 1:1,000 rat anti-FLAG antibody [WEHI, clone 9H1, 1:2500]).

The ATF3 overexpression vector was provided by Dr. Dakang Xu at Monash University (28). The AP-1 reporter construct was obtained from Clontech, Sp1/Sp3 reporter constructs were provided by Dr. Yoshihiro Sowa (Kyoto Prefectural University of Medicine), and pGL3-BCL-X_L reporter constructs were kindly provided by Dr. Ni Chen, Sichuan University, Chengdu, China (29).

Cell lines were transiently transfected with reporter constructs using the Lipofectamine 2000 transfection reagent (Invitrogen). Transfected cells were treated with HDACi for 24 to 48 hours and luciferase reporter activity determined using the dual-luciferase reporter assay Kit from Promega. Due to the strong effects of HDCAi treatment on TK-Renilla luciferase activity, reporter activity was normalized to total protein.

siRNAs targeting FOS, JUN, ATF3, and BCL-X_L were obtained from Dharmacon. siRNA transfection was performed using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer's instruction. Cells were harvested 24, 48, or 72 hours after transfection for subsequent analysis.

Animal studies were performed with the approval of the Austin Health Animal Ethics Committee. Eight-week-old female BALB/c nu/nu mice weighing approximately 16 g were obtained from the Australian Resources Centre (ARC). U87 cells (3 × 10⁶ cells) were injected subcutaneously into the right and left flank of each animal in a 150 μL suspension consisting of a 1:1 mixture of DMEM (Invitrogen) and BD Matrigel Basement Matrix (BD Biosciences). Once palpable tumors developed, mice were randomized into four groups to receive either vehicle [DMSO by intraperitoneal injection, and Phosal50 (60% Phosal PG, 30% PEG400, and 10% EtOH) by oral gavage], 50 mg/kg vorinostat via intraperitoneal injection, 25 mg/kg ABT-263 via oral gavage, or the combination. Mice were treated daily for 19 days. Tumor growth was monitored every second day by calliper measurement until the end of the experimental period or when tumors reached 1 cm³ in size. At this point, animals were euthanized, and tumors were excised and weighed.

In all cases, groups were compared using the Student's t test, with P < 0.05 considered to be statistically significant. Correlation analyses were performed using Pearson's correlation with P < 0.05 considered statistically significant.

To identify the molecular mechanisms underlying HDACi-induced apoptosis, we first stratified vorinostat-induced apoptotic responses in 50 human cancer cell lines representing common tumor types, including those displaying significant clinical response to HDACi (CTCL, multiple myeloma, leukemia, breast, and lung cancers; refs. 3, 7, 8). Sensitivity of the cell lines to vorinostat was highly variable (ranging from 2.5% apoptosis in U87 cells to 97.4% in RPMI-8226 cells), enabling separation into strong or weak responders (Fig. 1A). As observed clinically (3), vorinostat more potently induced apoptosis in hematologic cell lines (Fig. 1B). Among the solid tumor models, ovarian cancer lines were most sensitive, whereas prostate lines were most resistant (Fig. 1B). This spectrum of antitumor responses was replicated using NaBu, a member of the short-chain fatty acid subclass of HDACi (Fig. 1C and D). Differential sensitivity to HDACi was not due to differences in the extent of HDAC inhibition, as histone H3 acetylation was similarly increased by vorinostat in representative sensitive and resistant lines (Supplementary Fig. S1).

Using our comprehensive profile of HDACi-induced apoptosis, we next investigated the mechanisms likely to underpin HDACi response across multiple cancers. Based on our prior findings in colon cancer cells (24), we investigated whether the induction of the IE genes FOS, JUN, ATF3, EGR1, EGR3, and GADD45B is a general consequence of HDACi treatment, independent of tumor type. Using eight HDACi-sensitive cell lines representing solid and hematologic cancers, we found that vorinostat robustly induced these genes by 2- to 10-fold within 2 hours, and sustained their expression over 48 hours (Fig. 2A). Dose-dependent induction of these genes was confirmed in SK-MEL-28 (melanoma) and MCF7 (breast) cells (Fig. 2B), and corresponding increase in protein expression of c-FOS, c-JUN, and ATF3 was confirmed in 5 sensitive cell lines (Fig. 2C). Induction of this transcriptional response was independent of HDACi chemical subclass as FOS, JUN, ATF3, EGR1, EGR3, and GADD45B were also induced in the HH CTCL cell line treated with panobinostat, belinostat, depsipeptide, entinostat, and valproic acid (Supplementary Fig. S2).

To determine if HDACis induce IE genes in a clinical context, we assessed their expression before and after 4-hour panobinostat treatment in 2 patients with CTCL enrolled in an institutional phase II panobinostat trial (Clinicaltrials.gov identifier: NCT01658241). Both patients had >70% tumor load in their PBMCs and received 30 mg panobinostat orally. As in cell lines, panobinostat robustly induced IE genes in PBMCs in both patients establishing induction of this transcriptional response as a clinically detectable consequence of HDACi therapy (Fig. 2D).

We next examined if HDACi-induced apoptosis was coupled to the magnitude of IE gene induction by assessing HDACi induction of this transcriptional response in all 50 cell lines. Vorinostat-induced apoptosis across the 50 cell lines correlated significantly with the corresponding magnitude of FOS, JUN, and ATF3 induction, but not EGR1, EGR3, or GADD45B (Fig. 2E). Similar results were observed following treatment of the 50 cell lines with NaBu (Supplementary Fig. S3A). The preferential induction of FOS, JUN, and ATF3 in HDACi-sensitive cell lines was confirmed at the protein level in 2 representative sensitive and resistant cell lines, derived from different tumor types (Supplementary Fig. S3B).

FOS, JUN, and ATF3 encode members of the AP-1 family of transcription factors, which control transcription when bound to specific DNA sequences as homo- or heterodimers (30). To determine if induction of these genes by HDACi causes AP-1 activation and if the magnitude of AP-1 activation is associated with apoptotic sensitivity, AP-1 reporter gene assays were performed on 5 representative HDACi-sensitive and -resistant cell lines derived from multiple tumor types. Consistent with the preferential induction of FOS, JUN, and ATF3 in sensitive lines, HDACi induction of AP-1 reporter activity was significantly higher in the sensitive cell lines (Supplementary Fig. S3C).

Finally, we have previously demonstrated that the Sp1 and Sp3 transcription factors are required for HDACi induction of IE gene expression, and that HDACis preferentially induce Sp1/Sp3 reporter activity in HDACi-sensitive colon cancer cell lines (24). To determine if the differential induction of IE genes is linked to differential activation of Sp1/Sp3 transcription factors independent of tumor type, the 5 HDACi-sensitive and -resistant cell lines derived from different tumor types were transfected with an Sp1/Sp3 reporter construct and treated with vorinostat for 24 hours. Consistent with the findings in colon cancer cells, HDACi induction of Sp1/Sp3 reporter activity was significantly higher in sensitive cell lines, suggesting preferential activation of these transcription factors mediates IE gene induction independent of tumor type (Supplementary Fig. S3D).

To define the contributions of c-FOS, c-JUN, and ATF3 to HDACi-induced apoptosis, expression of each of these AP-1 proteins was knocked down in three HDACi-sensitive cell lines. We found that only ATF3 depletion was sufficient to attenuate HDACi-induced apoptosis (Fig. 3A and B). These effects were confirmed using multiple ATF3-targeting siRNAs (Supplementary Fig. S4). To test the role of ATF3 in HDACi-induced apoptosis in a different model, mouse embryonic fibroblasts (MEF) derived from wild-type and Atf3 knockout mice were treated with HDACi. As expected, vorinostat induced ATF3 mRNA only in wild-type MEFs (Fig. 3C). Atf3^−/− MEFs were significantly less responsive to vorinostat and NaBu-induced apoptosis than wild-type cells (Fig. 3D and E), collectively implicating ATF3 as a key mediator of HDACi-induced apoptosis in multiple cell types.

HDACi-induced apoptosis has been linked with altered expression of regulators of both the intrinsic and extrinsic apoptotic pathways; however, the majority of studies indicate a dominant role for the intrinsic (mitochondrial) pathway (2, 31, 32). To determine if ATF3 induction plays a role in altering expression of the key regulators of this pathway, we first determined the effect of HDACi treatment on expression of all known components of the intrinsic apoptotic pathway in 15 cell lines spanning a range of tumor types and HDACi sensitivities. HDACi significantly induced expression of BIM, BIK, BMF, and NOXA (PMAIP1) and downregulated expression of BCL-w (BCL2L2) in all cell lines, independent of apoptotic response (Supplementary Fig. S5). We next investigated if altered expression of any components of the intrinsic apoptotic pathway correlated with the magnitude of HDACi-induced apoptosis and the magnitude of HDACi induction of ATF3. This analysis identified BCL-X_L (BCL2L1) as a candidate ATF3-repressed gene, whose expression was inversely correlated with both the magnitude of HDACi-induced apoptosis and HDACi induction of ATF3 (Fig. 4A and B). Consistent with changes in its transcript levels, BCL-X_L protein was also preferentially repressed by HDACi in sensitive cell lines (Supplementary Fig. S6).

To directly determine if ATF3 is required for HDACi-mediated repression of BCL-X_L, we examined the effect of HDACi on BCL-X_L promoter activity using a series of BCL-X_L promoter reporter constructs (Fig. 4C; ref. 29). Vorinostat and panobinostat maximally repressed activity of the P1281 reporter (–664 downstream to +617 of the transcription start site) and also repressed the P828 and P1692 reporters (Fig. 4D). Conversely, minimal effect was observed on the P621 reporter, implicating key cis-acting sequences located between −4 and −664 bp upstream of the transcription start site that are required for HDACi-mediated repression of BCL-X_L promoter activity (Fig. 4D). Vorinostat also significantly repressed activity of the BCL-X_L P1281 reporter in two additional HDACi-sensitive cell lines, A549 and AGS (Supplementary Fig. S7). To determine if these effects were mediated through ATF3, experiments were repeated following ATF3 knockdown, which resulted in significant attenuation of HDACi-induced BCL-X_L promoter repression (Fig. 4E).

To determine if ATF3 can directly repress BCL-X_L promoter activity, we first assessed the effect of ATF3 overexpression alone on BCL-X_L promoter activity. Similar to the effects of HDACi, ATF3 overexpression repressed activity of the P828, P1281, and P1692 reporters but not the P621 reporter (Fig. 4F). Analysis of the promoter sequence −4 to −664 bp downstream of the transcription start site identified the presence of an AP-1 site and three CREB sites, which are putative ATF3-binding motifs (Fig. 4C). To directly establish ATF3 binding to this region in response to HDACi treatment, we performed ATF3 chromatin immunoprecipitation (ChIP) experiments which sequentially interrogated ATF3 binding along the BCL-X_L promoter. The most robust enrichment of ATF3 binding following vorinostat treatment was observed at regions R2 and R3 (Fig. 4G), overlapping the key regulatory region (−4 to −664) identified in the promoter reporter assays. Notably, HDACi and ATF3 overexpressions were able to repress the P828 promoter despite the lack of ATF3 binding to this region, suggesting that ATF3 may also indirectly repress BCL-X_L promoter activity.

Finally, to establish the requirement of ATF3 induction for HDACi-mediated repression of BCL-X_L at the endogenous level, ATF3 knockdown was performed in three sensitive cell lines prior to HDACi treatment. In each case, ATF3 knockdown markedly attenuated BCL-X_L repression in response to HDACi treatment (Fig. 4H), establishing ATF3 induction as a critical requirement for HDACi-mediated BCL-X_L repression.

We next examined the importance of BCL-X_L repression in HDACi-induced apoptosis. Knockdown of BCL-X_L in HDACi-refractory PANC1, U87, and PC3 cells significantly enhanced HDACi-induced apoptosis, implicating BCL-X_L repression as a key determinant of HDACi response (Fig. 5A). Conversely, BCL-X_L overexpression in FLAG-tagged hBCL-X_L MEFs conferred resistance to vorinostat-induced apoptosis compared with WT MEFs (Fig. 5B), collectively establishing BCL-X_L repression as a key determinant of HDACi-induced apoptosis.

These findings suggested that therapeutic targeting of BCL-X_L may have similar effects. To test this, the HDACi-resistant cell lines PANC1, PC3, and U87 were treated with vorinostat alone and in combination with the BH3 mimetics ABT-263 (navitoclax), which inhibits BCL-2, BCL-X_L, and BCL-w. Combination treatment significantly enhanced apoptosis compared with either agent alone in each cell line (Fig. 5C). Similar effects were obtained using its precursor compound, ABT-737 (Supplementary Fig. S8A). To directly determine the role of BCL-X_L, we next examined the effects of combining HDACi with the novel BCL-X_L–specific inhibitor, A-1331852 (33). Combination treatment significantly enhanced apoptosis in all 3 cell lines compared with either agent alone (Fig. 5D). In contrast, combination treatment with the BCL-2–specific inhibitor ABT-199 (venetoclax) resulted in modest to no enhancement of HDACi-induced apoptosis (Supplementary Fig. S8B). We next determined whether this combination could also be utilized in HDACi-sensitive cell lines, by enabling each drug to be used at significantly lower concentrations. Treatment of the HDACi-sensitive cell line HCT116 with a 2-fold lower concentration of vorinostat (2.5 μmol/L) and a 100-fold lower concentration of ABT-263 (0.1 μmol/L), or a 10-fold lower concentration of A-1331852 (1 μmol/L) to that used in resistant cells was still sufficient to induce >60% apoptosis (Fig. 5E and F).

As A-1331852 is not suitable for use in vivo, we next tested the effect of combination treatment with vorinostat and ABT-263 on growth of HDACi-refractory U87 xenografts in vivo. Daily treatment with the combination significantly inhibited tumor growth compared with control or either agent alone (Fig. 6A–C). Importantly, no differences in body weight were observed in either the single agent or combination treatment arms compared with control (Fig. 6D).

HDACis are an established treatment for hematologic malignancies (CTCL, multiple myeloma) and continue to be tested, mostly in combination, for activity in other tumor types (1). Comparatively, the activity of these agents in solid tumors is more limited. The goal of this study was to define the mechanisms of HDACi action in tumor cells in order to provide a framework for the rational design of drug combinations involving their use, and the identification of molecular determinants of sensitivity.

We previously demonstrated that HDACi-induced apoptosis in colon cancer cells is associated with a specific transcriptional response involving the induction of multiple IE response genes, including 3 members of the AP-1 transcription factor family, FOS, JUN and ATF3. We now demonstrate that this transcriptional response provides a robust and early readout of HDACi-induced apoptosis which transcends tumor cell type.

Although HDACi treatment preferentially induces expression of three AP-1 family members in sensitive cell lines, we found that only ATF3 is required for HDACi-driven apoptosis. The proapoptotic role for ATF3 identified herein is consistent with ATF3 overexpression alone being sufficient to induce apoptosis in prostate (34) and ovarian cancer cells (35), and the resistance of Atf3 knockout MEFs to UV-induced apoptosis (36). Furthermore, ATF3 is required for apoptosis induced by ER stress (37), anoxia (38), and the chemotherapeutic agents 5FU, etoposide, and cisplatin (39–41). Finally, ATF3 is required for apoptotic sensitization to HDACi combination therapy with cisplatin and agonistic anti-DR5 antibodies (41, 42) and for HDACi-induced apoptosis in bladder cancer cells (43). However, the subsequent mechanisms of apoptosis induction have not been investigated.

Prior studies have linked HDACi-induced apoptosis with altered expression of a number of pro- and antiapoptotic genes, particularly components of the intrinsic apoptotic pathway (43). However, these effects have not been investigated in the context of sensitivity across tumor cell type, and the mechanisms which underpin altered expression of these genes have not been systematically addressed. The current study identifies a uniform mechanism that determines HDACi-induced apoptosis, involving ATF3-mediated repression of BCL-X_L, which transcends tumor cell type. The role of ATF3 as a transcriptional repressor is consistent with prior reports (44), and our ChIP and reporter gene analysis indicate that repression of BCL-X_L in HDACi-treated tumor cells involves direct binding of ATF3 to the BCL-X_L promoter. Furthermore, we demonstrate that repression of BCL-X_L is central in HDACi-induced apoptosis, as both molecular and pharmacologic inhibition of this prosurvival factor markedly enhanced HDACi-induced apoptosis in vitro and in vivo, and BCL-X_L overexpression protects cells from HDAC-induced apoptosis, consistent with previous studies (18, 45).

However, we note that the molecular or pharmacologic inhibition of BCL-X_L alone did not induce apoptosis to the same extent as when BCL-X_L was inhibited in the presence of HDACi, suggesting the requirement for additional HDACi-induced molecular changes to drive apoptosis. In this regard, we did identify consistent induction of the proapoptotic BH3-only genes BIM, BIK, BMF, and NOXA in response to HDACi treatment, several of which has been shown to be required for HDACi-induced apoptosis (32, 46). Notably however, induction of these genes occurred uniformly across the cell lines, independent of apoptotic sensitivity, implying their altered expression is not the basis for differential HDACi response. We therefore propose a model whereby HDACi-induced apoptosis involves both the induction of proapoptotic factors such as BIM, BIK, BMF, and NOXA and the ATF3-dependent repression of the prosurvival factor BCL-X_L, of which the magnitude of induction of ATF3 and subsequent repression of BCL-X_L determines apoptotic response.

Delineating BCL-X_L repression as a key determinant for HDACi-induced apoptosis has significant implications for the rational design of strategies to enhance HDACi antitumor activity. We exploited this using navitoclax (ABT-263), a BH3-mimetic drug (that inhibits BCL-2, BCL-w, and BCL-X_L), and the BCL-X_L–specific inhibitor A-1331852, which significantly enhanced HDACi-induced apoptosis in tumor cells inherently refractory to HDACi. These findings have the potential to enhance the range of tumors amenable to HDACi treatment by overcoming inherent resistance and to potentially reduce toxicities in sensitive cells by enabling HDACi to be used at lower concentrations.

A further application of these findings could be in the selection of patients likely to respond to HDACi by assessment of the magnitude of FOS, JUN, and ATF3 induction following short-term HDACi treatment. The feasibility of this approach is supported by our demonstration of induction of these genes following 4-hour panobinostat treatment in 2 patients with CTCL with high circulating tumor load. This approach could potentially be extended to solid tumors where freshly isolated tumor cells in the form of biopsy material, organoids, patient-derived xenografts, or circulating tumor cells are assessed for FOS, JUN, and ATF3 induction following short-term HDACi treatment as a predictor of the likelihood of response.

Our findings also suggest a framework for identifying molecular biomarkers of HDACi response prior to drug treatment through detailed investigation of the molecular determinants of differential ATF3 induction among tumors. In this regard, our previous studies in colon cancer cells demonstrated that HDACi induction of IE genes, including ATF3, is dependent on the Sp1 and Sp3 transcription factors, and that HDACis preferentially induce Sp1/Sp3 reporter activity in HDACi-sensitive colon cancer lines (24). We now extend these findings by demonstrating that HDACis preferentially induce Sp1/Sp3 reporter activity in sensitive cell lines, independent of tumor type. A central role for Sp1 and Sp3 in regulating HDACi-induced apoptosis independent of tumor type is also plausible given their ubiquitous expression (47). However, SP1 and SP3 are not mutated in human cancers, and analyses in colon cancer cells suggest basal differences in expression are unlikely to be determinants of HDACi response (24). Notably, both SP1 and SP3 are posttranslationally modified by a number of mechanisms including acetylation, ubiquitination, and phosphorylation which can alter their activity (48). Exploration of whether such posttranslational modifications occur in response to HDACi treatment and identification of the factors regulating these changes, which may vary between sensitive and resistant cells, may provide novel insight into the basis for differential HDACi response. Notably, several other mechanisms of ATF3 induction have also been described, including induction by p53 (49), activation of JNK, ERK, and p38 signaling (50), and by ATF4 subsequent to activation of the ER stress/unfolded protein response pathway (37). In addition to modulating SP1 and SP3, HDACi can also affect these pathways which may contribute to the differential induction of ATF3 among tumors.

In summary, we have identified a specific transcriptional response associated with HDACi-induced apoptosis that transcends tumor type, involving the coordinate induction of FOS, JUN, and ATF3. We identify the induction of ATF3 and subsequent repression of BCL-X_L as a central mechanism of HDACi-induced apoptosis and applied these findings to develop rational drug combinations which overcome inherent resistance and enhance the activity of HDACi in a range of tumor types.

M. Dickinson reports receiving other commercial research support from, reports receiving speakers bureau honoraria from, and is a consultant/advisory board member for Novartis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.C. Chüeh, J.W.T. Tse, P. Ioannidis, M.R. Thompson, W.D. Fairlie, A.S. Dhillon, J.M. Mariadason

Development of methodology: A.C. Chüeh, J.W.T. Tse, P. Ioannidis, L. Togel, E. Lee, J.M. Mariadason

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.C. Chüeh, J.W.T. Tse, M. Dickinson, P. Ioannidis, L. Jenkins, B. Tan, I. Luk, R. Nightingale, J.M. Mariadason

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.C. Chüeh, J.W.T. Tse, M. Dickinson, P. Ioannidis, A.S. Dhillon, J.M. Mariadason

Writing, review, and/or revision of the manuscript: A.C. Chüeh, J.W.T. Tse, M. Dickinson, L. Togel, B.R.G. Williams, G. Lessene, E.F. Lee, W.D. Fairlie, A.S. Dhillon, J.M. Mariadason

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.C. Chüeh, J.W.T. Tse, L. Togel, M. Davalos-Salas, J.M. Mariadason

Study supervision: A.C. Chüeh, A.S. Dhillon, J.M. Mariadason

We thank Paul G. Ekert (Murdoch Children's Research Institute), Kaye Wycherley (Walter Eliza Hall Institute for Medical Research), Andrew Wei (Alfred Hospital), and Michael H. Kershaw (Peter Mac Cancer Centre) for providing us with the leukemia and multiple myeloma cell lines used in this study.

This study was funded by the National Health and Medical Research Council (NHMRC) of Australia (1008833 and 1066665), The National Institutes of Health (NIH1RO1 CA123316), an Australian Research Council Future Fellowship (FT0992234), an NHMRC Senior Research Fellowship (1046092) to J.M. Mariadason, Ludwig Cancer Research, and the Operational Infrastructure Support Program, Victorian Government, Australia. J.W.T. Tse was supported by an Australian Postgraduate Award.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.