An ARC-Regulated IL1β/Cox-2/PGE2/β-Catenin/ARC Circuit Controls Leukemia–Microenvironment Interactions and Confers Drug Resistance in AML

Acute myeloid leukemia (AML) is a highly heterogeneous hematologic malignancy with poor overall survival. Intrinsic characteristics of leukemia cells such as overexpression of antiapoptotic proteins and activation of survival signaling pathways, extrinsic microenvironmental factors such as alteration of growth factors/cytokines, and leukemia-niche interactions that further augment the changes in leukemia cells and microenvironments all support leukemia cell and leukemia stem/progenitor cell growth, survival, and resistance to therapy. Thus, targeting the survival and resistance mechanisms that are common in AML should benefit many patients with this disease. We previously reported that the apoptosis repressor with caspase recruitment domain (ARC) protein is a strong independent, poor prognosis factor in AML. ARC protects leukemia cells from apoptosis induced by multiple agents (1, 2). We subsequently found that ARC induces IL1β expression in AML cells, increases chemokine CCL2, CCL4, and CXCL12 expression in mesenchymal stromal cells (MSC) in an ARC/IL1β–dependent manner, and facilitates leukemia–microenvironment interactions through a NFκB/IL1β signaling network (3).

Bone marrow–derived MSCs are an essential structural and functional component of the bone marrow microenvironment, and they are critical for hematopoiesis (4). Within the context of leukemia, MSCs play an essential role in protecting leukemia cells from chemotherapeutic agents (5). IL1β was previously shown to promote apoptosis resistance in AML blasts (6), and the IL1 receptor accessory protein (IL1RAP) is reportedly overexpressed in AML stem/progenitor cells (7). Furthermore, targeting IL1RAP with a neutralizing antibody selectively killed AML stem cells (8). These data support the critical role of bone marrow IL1β signaling in AML cell survival. Furthermore, a recent study demonstrated that cancer cells release IL1β that strongly induces Cox-2/PGE2 in MSCs, which in turn activates β-catenin signaling and induces a stem cell phenotype in epithelial cancer cells (9).

The roles of Cox-2 and its enzymatic product PGE2 in inflammation and cancer are well documented. Cox-2–catalyzed prostaglandin synthesis is regulated by NFκB/IL1β signaling (10–12), which is frequently upregulated in cancer cells. PGE2 signals through EP1-EP4 receptors that regulate multiple signaling cascades including β-catenin (13). Cox-2 and PGE2 have been reported to enhance hematopoietic stem cell survival, self-renewal, engraftment, and homing (14). Treatment with the PGE2 analog dimethyl-PGE2 (dmPGE2) was found to upregulate CXCR4 through increasing HIF1α and enhance hematopoietic stem/progenitor cell homing and engraftment (15, 16).

Although PGE2 is known to possess potent tumor-stimulating activity (17), its role in leukemia is largely unknown. AML cell line HL-60 was reported to express high levels of microsomal PGE synthase 1 (mPGES-1), which is frequently induced concomitantly with Cox-2 by various proinflammatory stimuli and responsible for increased PGE2 production (18). mPGES-1 was undetectable in normal blood mononuclear cells and its inhibition reduced PGE2 production and viability of HL-60 cells (19). PGE2 from MSCs was shown to activate PKA signaling and antagonize p53-induced cell death (20). Furthermore, Wnt/β-catenin signaling is constitutively active in AML (21) and is required for the development of AML stem cells (22). Expression of β-catenin by AML cells portends enhanced clongenic capacities and poor disease prognosis (23).

We therefore hypothesized that ARC exerts its role in leukemia–stromal interactions by modulating PGE2 levels. In this study, we investigated Cox-2 expression in MSCs cocultured with AML cells in which ARC expression was genetically modified. We also investigated PGE2 levels in vitro in an AML–MSC coculture system, in fresh bone marrow samples from patients with AML and normal controls, and in vivo in immunodeficient mice xenografted with ARC knockdown (ARC KD) AML cells. We demonstrate that both Cox-2 expression and PGE2 generation are ARC/IL1β dependent and that ARC, regulated by β-catenin, is an integral component of an IL1β/PGE2/β-catenin circuit. Cox-2/PGE2, regulated by ARC and induced by AML–MSC coculture contributes to MSC-mediated chemoprotection in AML.

OCI-AML3 cells, provided by Dr. M. Minden (Ontario Cancer Institute, Toronto, Ontario, Canada) were validated by STR DNA fingerprinting using the AmpF_STR Identifier Kit according to the manufacturer's instructions (catalog number #4322288, Applied Biosystems). The STR profiles were compared with known ATCC fingerprints, and to the Cell Line Integrated Molecular Authentication database (CLIMA) version 0.1.200808 (http://bioinformatics.hsanmartino.it/clima/; ref. 24). The STR profile was identified as unique. Mycoplasma testing was performed using the PCR Mycoplasma Detection Kit from Applied Biological Materials (catalog number #G238) per the manufacturer's instructions. Authenticated and Mycoplasma-free cells are stored under liquid nitrogen and are never kept in culture for >4 months. Primary samples were acquired from patients with AML or normal controls after informed written consent following the institution approved protocol in accordance with Declaration of Helsinki. Patient characteristics are shown in Table 1. Mononuclear cells were isolated from primary samples by density–gradient centrifugation using Lymphocyte Separation Medium (catalog number #25-072-CV, Corning). Human MSCs were isolated from bone marrow samples obtained from healthy subjects as described previously (25). Cell lines were cultured in RPMI1640 medium and cells from primary samples and MSCs in α-MEM medium, both supplemented with 10% heat-inactivated FCS, 2 mmol/L l-glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin. Cells were kept at 37°C in a humidified atmosphere of 5% CO₂. ARC KD AML cells (2) and MSCs (26) were generated as described previously. For coculture experiments, leukemia cells were added to MSCs (4:1) that were plated the night before and cultured in α-MEM medium with supplements. AML cells, MSCs, or the cocultured cells were treated with IL1β (catalog number #200-01B) with or without IL1RA (catalog number #200-01RA) (PeproTech), dmPGE2 (16,16-dimethyl-PGE2, a PGE2 analogue; catalog number #14750, Cayman Chemical), Ara-C with or without Cox-2 inhibitor celecoxib (catalog number #C-1502, LC Laboratories), or β-catenin inhibitor C-82 (provided by PRISM Pharma; refs. 27, 28) with or without Ara-C.

PGE2 in the supernatant of cultured cells, mouse serum, and human bone marrow samples was determined by PGE2 ELISA (catalog number #KGE004B), and human IL1β in mouse serum and human bone marrow samples was determined by ELISA specific for human IL1β (catalog number #DLB50; both from R&D Systems) following the manufacturer's instructions.

OCI-AML3 cells were electroporated with a scramble control or the SMARTpool ON-TARGET plus CTNNB1 siRNAs (750 nmol/L; catalog number #L-003482-00-0005, Dharmacon) using an Amaxa apparatus (Solution T, catalog number #VCA-1002; program X-001; Lonza) following the manufacturer's instructions as described previously (28).

RNA isolation and TaqMan RT-PCR were carried out as described previously (29). Primers sets used are: β-catenin (Hs00355049_m1), ARC (Hs00358724_g1), and ABL (Hs01104728_m1; all from Thermo Fisher Scientific). The abundance of each transcript relative to that of ABL was calculated using the 2^−ΔC_tmethod, where ΔC_t is the mean C_t of the transcript of interest minus the mean C_t of the transcript for ABL housekeeping gene.

The promoter region of the ARC coding gene NOL3, identified using the Eukaryotic Primer Database (https://epd.vital-it.ch/index.php; residues 67173446-67174045, RefSeq NC_000016.10, human chromosome 16, GRCh38.p7) has two LEF1/TCF4E motifs (CTTTGTGC and GGCCAAAG) by searching PROMO 3.0 (http://alggen.lsi.upc.es/cgi-bin/promo_v3/promo/promoinit.cgi?dirDB=TF_8.3). We amplified by PCR and inserted this region between the Cla I and Nhe I sites in the lentivector pCDH-CMV-MSC-EF1a-Puro (System Biosciences), to replace the CMV promoter. We then inserted the open reading for copGFP between the Nhe I and BamH1 sites such that the start codon for the GFP moiety was in frame with the NOL3 start codon. We mutated the LEF1/TCF4E motifs using a QuikChange II XL Site-Directed Mutagenesis Kit as directed by the manufacturer (Agilent), except that we used NEBStable cells (New England Biolabs) in lieu of XL10-Gold cells. Mutants were identified by the presence of both Pml I and Sal I sites. Primers used for these constructions are listed in Supplementary Table S1. Lentivirus was prepared by transfecting HEK293T cells (ATCC) with an equimolar mix of reporter vector and packaging plasmids psPAX2 and pMD2.G (gifts of Didier Trono, Addgene) using JetPrime transfection reagent as directed by the manufacturer (Polyplus). OCI-AML3 cells were transduced with the lentivirus as described previously (2).

Protein levels were determined by Western blot analysis as described previously (3) using the Odyssey Infrared Imaging System for signal detection and Odyssey software version 3.0 for quantification (LI-COR Biosciences). Cytoplasmic and nuclear fractions were prepared as described previously (30). Antibodies against β-catenin (Cat#8480) and ARC (Cat#NBP2-41753) were purchased from Cell Signaling Technology and Novus, respectively. Histone H3 was used as loading control for nuclear fraction, α-tubulin for cytoplasm, and β-actin for total lysate.

After staining with Ghost Dye Violet 510 (catalog number #13-0870-T500, Tonbo Biosciences), cells were washed and fixed with 4% paraformaldehyde and permeabilized with 100% methanol, and then stained with Fc-blocker (catalog number #130-059-901, Miltenyi Biotec), followed with Cox-2-PE (catalog number #12282, 1:50, Cell Signaling Technology), CD90-PerCP/CyC5.5 (catalog number #328118), and CD45-Pacific Blue (catalog number #304029; BioLegend) in 5% BSA/PBS. The stained cells were analyzed using a Gallios flow cytometer (Beckman Coulter Life Sciences) and quantified using FlowJo analytic platform (BD Biosciences). Cox-2 levels in MSCs or AML cells were expressed as the geometric mean fluorescent intensity (MFI) difference of cells stained with Cox-2 antibody or with IgG in CD90⁺ or CD45⁺ cells, respectively.

AML patient samples were stained with a panel of metal-tagged antibodies for cell surface markers and intracellular proteins (Supplementary Table S2) and subjected to CyTOF analysis as described previously (3, 28, 31). Viable (cisplatin low) single cells were gated with FlowJo software (v10, FlowJo LLC) and exported as flow cytometry standard (FCS) data for subsequent analysis in Cytofkit (32). RPhenoGraph was used for unsupervised subset detection based on cell surface markers. t-SNE embedded FCS files were further analyzed in FlowJo and cell populations identified by RPhenoGraph were mimicked and gated on the t-SNE map for quantitation of intracellular marker expression. The expression level of each protein in the desired cell compartments is expressed as ArcSinh-transformed data.

Apoptosis was estimated in CD45⁺ cells by flow cytometry after Annexin V–Cy5.5 staining in the presence of 7-amino-actinomycin D (7-AAD) using LSRII flow cytometry (BD Biosciences). Apoptosis in primary samples was assessed in CD45⁺ and CD34⁺CD38⁻ cells after the cells were incubated with CD34-PE, CD38-PECy7, and CD45-APCH7 antibodies (BD Biosciences) and expressed as specific apoptosis:

NOD/SCIDIL2RγNull-3/GM/SF (NSGS) mice (male, 3 months old) were injected with either vector control or ARC KD OCI-AML3 cells (1 × 10⁶/mouse). A week later, control or ARC KD cell-injected mice (10/group) were either untreated or treated with Ara-C (100 mg/kg) via intraperitoneal injection, 3 times per week for approximately 5 and half weeks. Leukemia burden was assessed by flow cytometric measurement of human CD45⁺ (huCD45⁺) cells in mouse periphery blood, bone marrow, and spleen. Serum IL1β and PGE2 levels were determined. Tissue sizes were measured and protein expression in mouse tissues was determined by Opal/TSA (tyramide signal amplification) multispectral IHC and Vectra imaging (see below). Mouse survival was followed. All the mouse experiments were carried out following protocols approved by the Institution Animal Care and Use Committee at MD Anderson Cancer Center.

Paraffin-embedded mouse tissue slides were deparaffinized and rehydrated followed by antigen retrieval with Biogenex Laboratories Antigen Citra RTUSE (catalog number #NC0359148, Thermo Fisher Scientific) at 95°C for 15 minutes and blocked with 3% BSA in TBS-T for additional 15 minutes. Tissues were incubated with a primary antibody followed by HRP-conjugated IgG (broad spectrum) provided in SuperPicture Polymer Detection Kit (catalog number #878963, Thermo Fisher Scientific) for 15 minutes at room temperature and then labeled with TSA plus reagents following the manufacturer's protocols (PerkinElmer). The above steps were repeated sequentially for staining each of the following proteins: ARC, CD45, and β-catenin all specific for human. For ARC detection, tissues were stained overnight at 4°C with an anti-ARC antibody (1:4,000, catalog number #38916S, Cell Signaling Technology) and for β-catenin detection, they were incubated 1 hour at room temperature with an anti–β-catenin antibody (1:1,000; catalog number #LS-B4123, LifeSpan BioSciences Inc.). For huCD45 detection, tissues were stained 2 hours at room temperature with an anti-huCD45 antibody (1:2,000; catalog number #M0701, Agilent Dako). Opal fluorescein reagents used for TSA were Opal 540 (catalog number #FP1494001KT), Opal 570 (catalog number #FP1488001KT), Opal 650 (catalog number #FP1296001KT; PerkinElmer) for CD45, ARC, and β-catenin; respectively. Opal fluorescent reagent was diluted with 1× plus amplification diluent (1:150; catalog number #FP1498, PerkinElmer). Each issue was incubated with 200 μL diluted Opal reagent at room temperature for 5 minutes and washed with TBS-T twice, 3 minutes each. Finally, the tissue slides were counterstained with 2.5 μg/mL DAPI (catalog number #D21490, Thermo Fisher Scientific) at room temperature for 5 minutes, sealed by anti-fade fluorescent mounting medium (catalog number #S3023, Dako), and stored at 4°C.

Multispectral images were acquired using Vectra 3.0 automated quantitative pathology imaging system (CLS142338, PerkinElmer) with ×20 objective and software version 3.012 (PerkinElmer). Images were analyzed by InForm version 2.3 software (PerkinElmer). Briefly, a fluorophore library was created by monospectral labelled tissue images to define spectral curves for each fluorophore and counterstain to adjust background noise and positive staining of biomarkers. InForm software provides objective counting of cell population and biomarkers, allows tissue separation by machine training, and increases the accuracy of the statistical analysis. Algorithms were created for separating tumor tissues and normal tissues and for evaluating various protein levels via specific antibodies. For protein quantification, huCD45⁺ cells were selected from multiple fields and multiple cells in each field. The scan fields for spleens were selected from the central region of the tissue section to maximally cover the whole spleen avoiding the margins, and the fields for bone marrow were selected from the whole marrow compartments away from bone fragments. The expression level is expressed as mean optical density (OD)/huCD45⁺ cell in each field.

Experiments were conducted in triplicate unless otherwise indicated. One-way ANOVA or Student t test was performed to compare the differences between groups. Results were expressed as the mean ± SEM. Correlation coefficient was determined by Pearson correlation analysis. The combination index (CI), expressed as the mean of CI values at the 50%, 75%, and 90% effective doses was determined by the Chou–Talalay method (33). CI < 1.0 was considered synergistic; = 1.0 additive; and >1.0 antagonistic. Mouse survival was estimated by the Kaplan–Meier method and analyzed using log-rank statistics. Statistical significance was set at P < 0.05.

MSCs were treated with IL1β or cocultured with OCI-AML3 cells in the absence or presence of IL1R antagonist IL1RA (48 hours). Cox-2 protein levels were determined by flow cytometry in both MSCs (CD90⁺) and OCI-AML3 cells (CD45⁺). As expected, IL1β greatly induced the expression of Cox-2 in MSCs, which was suppressed by IL1RA. Cocultures of MSCs with OCI-AML3 cells also significantly increased Cox-2 expression in MSCs, and similarly this induction was inhibited by cotreatment of cells with IL1RA (Fig. 1A), suggesting that increased Cox-2 expression in MSCs cocultured with AML cells is mediated through IL1β. By comparison, OCI-AML3 cells expressed lower levels of Cox-2 than MSCs, and coculture of AML cells with MSCs also increased Cox-2 expression in AML cells in an IL1β-dependent manner, but to a lesser degree (Fig. 1A). Importantly, similar results were obtained when primary AML patient samples (Table 1, samples 1–3) were used (Fig. 1B) and IL1β, even at 25 pg/mL level, was sufficient to significantly increase Cox-2 in MSCs (P = 0.013).

We next cocultured MSCs with vector control or ARC KD OCI-AML3 cells, vector control or ARC KD MSCs with OCI-AML3 cells, or vector control or ARC KD MSCs with vector control or ARC KD OCI-AML3 cells (48 hours). We found that significantly less Cox-2 protein was induced in MSCs when ARC was knocked down either in AML cells (P = 0.00025) or MSCs (P = 0.0083) compared with the respective controls and that ARC KD in both MSCs and AML cells resulted in the lowest level of Cox-2 expression (P = 0.007, 0.021, or 0.001, versus ARC KD in MSCs, ARC KD in AML cells, or vector control; respectively; Fig. 1C), indicating that Cox-2 expression depends on ARC levels, which is in agreement with our previous finding that ARC regulates NFκB/IL1β signaling (3).

As shown in Fig. 1, although Cox-2 is expressed in OCI-AML3 cells and primary patient samples, AML cells express relatively lower Cox-2 than MSCs. We next determined PGE2 levels in the culture media of OCI-AML3 and MSCs in the absence or presence of IL1β (48 hours) and found that compared with OCI-AML3 cells, MSCs secreted >100-fold PGE2/cell. IL1β, at 1 to 100 ng/mL, significantly induced PGE2 production in MSCs (6.5- to 9.2-fold; P < 0.0001), whereas at only 100 ng/mL, IL1β significantly induced PGE2 production in OCI-AML3 cells (2.5-fold; P = 0.002; Fig. 2A), indicating that consistent with Cox-2 expression, MSCs are the major source of PGE2.

We then cocultured MSCs and OCI-AML3 cells for 48 hours and found markedly increased PGE2 levels in the medium of cocultured cells compared with MSCs or OCI-AML3 cultured alone. This increase was suppressed by IL1 receptor antagonist IL1RA or Cox-2 inhibitor celecoxib (Fig. 2B). We next cocultured MSCs with vector control or ARC KD OCI-AML3 cells for 48 hours and found that knocking down ARC in OCI-AML3 cells significantly decreased the secreted PGE2 in the coculture system (P < 0.0001; Fig. 2C, left). In addition, we found that in cocultured vector control or ARC KD MSCs with OCI-AML3 cells, knockdown of ARC in MSCs significantly reduced PGE2 production (P = 0.00081; Fig. 2C, right). Our data suggest that, as with Cox-2 expression, AML-MSC coculture induces the production of PGE2 in an ARC- and IL1β-dependent manner. Finally, coculture cells from peripheral blood samples of primary AML patients (n = 5, Table 1, samples 4–8) with MSCs (48 hours) markedly increased secreted PGE2, whereas PGE2 levels were significantly lower when patient cells were cocultured with ARC KD MSCs (P = 0.0027; Fig. 2D).

We previously showed that coculture of AML cells with MSCs increases β-catenin (27) and ARC expression in AML (2). To demonstrate that this effect is partially mediated through PGE2 signaling, OCI-AML3 cells were treated with a PGE2 analogue, dmPGE2 or cocultured with MSCs without or with Cox-2 inhibitor celecoxib. dmPGE2 induced protein levels of β-catenin and ARC (Fig. 3A). As expected, coculture with MSCs increased β-catenin and ARC in FACS-sorted OCI-AML3 (CD45⁺CD90⁻) cells after coculture (Fig 3A), and this increase was completely suppressed by celecoxib (Fig. 3A). Importantly, both PGE2 and cocultures primarily increased nuclear β-catenin in AML cells (Fig. 3B). To examine the biological relevance of increased PGE2 production in the leukemia–MSC system, we treated OCI-AML3 cells with Ara-C, the commonly used chemotherapeutic agent for AML therapy with or without dmPGE2 or MSCs. As shown in Fig. 3C, Ara-C induced time and dose-dependent apoptosis in OCI-AML3 cells, which was significantly suppressed by MSC coculture and by dmPGE2 cotreatment in a dose-dependent manner. Blocking PGE2 production with celecoxib (200 nmol/L) significantly reduced the protective effect of MSCs. Our data indicate that increased PGE2 production in leukemia–MSC coculture system activates β-catenin, increases ARC expression, and contributes to MSC-mediated chemoprotection.

We have demonstrated that in AML-MSC cocultures, ARC regulates IL1β in AML cells, induces Cox-2 in MSCs, and increases secreted PGE2, which in turn induces the expression of β-catenin in AML cells. Interestingly, in addition to β-catenin, PGE2 also increased ARC level. Once activated, β-catenin is translocated to the nucleus where it interacts with DNA-binding proteins T-cell factors (TCF)/lymphoid enhancer factors (LEF) that recognize and bind to specific sequence motifs in promoters and enhancers of target genes (34). Promoter analysis demonstrated that the ARC promoter region contains two binding sites for TCF4 (CTTTGTG and GCCAAAG)/LEF1 (CTTTGTGC and GGCCAAAG), suggesting that ARC may be regulated by β-catenin. To test that, we silenced β-catenin in AML cells by siRNAs. As shown in Fig. 4A, inhibition of β-catenin suppressed ARC expression both at the RNA and protein levels, suggesting that β-catenin regulates ARC expression transcriptionally, and that the PGE2-induced ARC increase is mediated through PGE2/β-catenin signaling.

We next generated an ARC promoter–driven GFP reporter and transduced the construct into OCI-AML3 cells. Silencing β-catenin decreased the level of GFP reporter in OCI-AML3 cells (Fig. 4B). We also mutated LEF1/TCF4E motifs in the ARC promoter region to generate an ARC mutant promoter–driven GFP reporter. Figure 4C shows that dmPGE2 markedly increased, and inhibition of β-catenin with a specific inhibitor C-82 decreased GFP levels in OCI-AML3 cells transduced with ARC promoter–driven GFP. dmPGE2 or C-82 had only minimal effects on GFP levels in cells transduced with ARC-mutant promoter-driven GFP, further supporting that PGE2/β-catenin signaling transcriptionally regulates ARC expression.

To assess the biological relevance of ARC/IL1β/PGE2/β-catenin signaling in AML in vivo, we injected NSGS mice with vector control or ARC KD OCI-AML3 cells. Mice were left untreated or treated with Ara-C and peripheral blood and tissues were collected 24 or 25 days after leukemia cell injection (Fig. 5A). We first determined the levels of IL1β and PGE2 in mouse serum by ELISA and found a statistically significant decrease in serum IL1β and PGE2 in mice transplanted with ARC KD compared with control AML cells (Fig. 5B). In addition, mice with ARC KD AML cells had lower leukemia burden in all tissues examined. Specifically, significantly less huCD45⁺ cells were found in mouse bone marrows and spleens (Fig. 5C), along with markedly reduced liver and spleen size (Fig. 5D) compared with mice injected with control cells.

We next determined the expression levels of ARC and β-catenin in huCD45⁺ cells in mouse spleen and bone marrow using Opal/TSA IHC and Vectra imaging that allows simultaneous measurement of multiple markers within a single tissue section. Figure 5E shows the individual or mixed fluorescent images of mouse spleen stained with huCD45, ARC, β-catenin, and DAPI (left) and quantification of ARC and β-catenin protein levels (right) in the spleens of all four experimental groups. We analyzed 35 fields in control cell-injected mice (7,164–10,873 cells/field, total 319,431 cells), 26 fields in control cell-injected mice with Ara-C treatment (2,635–4,777 cells/field, total 99,727 cells), 17 fields in ARC KD cell-injected mice (610–5,021 cells/fields, total 37,206 cells), and 22 fields in ARC KD cell-injected mice with Ara-C treatment (248–3,927 cells/field, total 45,431 cells); respectively. As expected, mice injected with ARC KD OCI-AML3 cells with or without Ara-C treatment expressed drastically lower levels of ARC protein in human spleen CD45⁺ cells (P < 0.0001) compared with those injected with control cells with or without treatment. Interestingly, their spleen huCD45⁺ cells also had significantly lower levels of β-catenin (P < 0.0001 with Ara-C treatment, P < 0.01 without Ara-C treatment), supporting the existence of an ARC/IL1β/PGE/β-catenin/ARC regulatory loop. Similarly, mice injected with ARC KD OCI-AML3 cells with or without Ara-C treatment expressed significantly lower levels of ARC and β-catenin in bone marrow huCD45⁺ cells compared with those injected with control cells without treatment. However, in bone marrow, Ara-C–treated groups had lower ARC and β-catenin levels compared with their respective untreated controls, especially for bone marrow cells obtained from control cell–injected mice (Supplementary Fig. S1). The reason for this difference is unclear. We did observe that in bone marrow samples, cells from Ara-C–treated mice did not attach to slides well compared with cells from untreated mice, which was not observed in spleen samples. Lower leukemia burden in mice with ARC KD AML cells was further demonstrated by Opal/TSA staining of huCD45 in spleen and bone marrow and Vectra imaging (Fig. 5E; Supplementary Fig. S1).

Furthermore, both ARC inhibition, and the chemotherapeutic drug Ara-C, significantly lowered leukemia burden in peripheral blood assessed by flow cytometry of huCD45⁺ cells (Fig. 5F) and prolonged survival (Fig. 5G). Mice with ARC KD AML cells lived longer (median survival, 33 days; P = 0.0006), similar to mice with control cells treated with Ara-C (34 days, P = 0.0003) compared with those with control cells not receiving treatment (27 days; Fig. 5G). Importantly, inhibition of ARC sensitized AML to Ara-C: mice injected with ARC KD cells treated with Ara-C showed the longest median survival time (40.5 days; P = 0.0006, 0.0004, and 0.0012 vs. control, control + Ara-C, and ARC KD; respectively; Fig. 5G).

To assess potential roles of IL1β and PGE2 in AML and the AML bone marrow microenvironment, we collected bone marrow from patients with AML (n = 8, Table 1, samples 9-16) and normal controls (n = 4) and determined IL1β and PGE2 levels: both IL1β and PGE2 levels were significantly higher in bone marrow from patients with AML compared with normal controls (Fig. 6A). We then determined β-catenin and ARC protein levels in AML patient bone marrow cells (Table 1, samples 17–20) by CyTOF mass cytometry in various cell compartments and observed strong correlations of β-catenin and ARC levels, in bulk as well as in CD34⁺CD38⁺ and CD34⁺CD38⁻ stem/progenitor cells (Fig. 6B). Furthermore, inhibition of β-catenin by C-82 decreased ARC and MSC coculture-induced ARC expression in CD45⁺ blasts and CD34⁺CD38⁺ and CD34⁺CD38⁻ stem/progenitor cells from primary samples (Fig. 6C; n = 2, Table 1, samples 21, 22). To demonstrate whether targeting the IL1β/PGE2/β-catenin/ARC cascade sensitizes to chemotherapy, we treated primary AML cells with C-82 and Ara-C (Table 1, samples 23-27): C-82 and Ara-C synergistically induced apoptosis in bulk (n = 5) and CD34⁺CD38⁻ cells (n = 4) even when cells were cocultured with MSCs (CI < 1; Fig. 6D).

ARC was first identified as an antiapoptotic protein. We reported that ARC is a strong adverse prognostic marker in AML and protects leukemia cells from therapeutic agent-induced cell death (1, 2) consistent with its antiapoptotic property. Interestingly, we also observed that ARC plays a role in leukemia–MSC interactions. On the basis of our previous finding that ARC modulates leukemia–microenvironment interactions by regulating NFκB/ILβ signaling (3), we here demonstrate that ARC regulates IL1β in AML, which then induces Cox-2 expression in MSCs and PGE2 production, which in turn activates β-catenin and increases ARC in AML, while ARC itself is regulated by β-catenin. The data support the concept that in addition to directly suppressing apoptosis, ARC, regulated by β-catenin, mediates leukemia–stromal interactions through the ARC-IL1β/Cox-2/PGE2/β-catenin circuit. Furthermore, ARC in MSCs also affects Cox-2 levels and PGE2 production in AML-MSC coculture. We previously reported that ARC in MSCs induces nuclear and phospho-NFκB (3). NFκB is known to regulate Cox-2–catalyzed PGE2 synthesis (10–12). Therefore, ARC is an integral component of and a regulator of leukemia–microenvironment interactions (Fig. 6E).

PGE2 was reported to suppress apoptosis in multiple cell types through various mechanisms such as PI3K/AKT signaling or regulating survivin expression (35–37). In this study, we found that PGE2 protects AML cells from chemotherapy and strongly induces the expression of antiapoptotic ARC protein in AML cells, a new mechanism of PGE2-mediated antiapoptotic effect. Furthermore, we demonstrated that MSCs confer AML drug resistance in a PGE2-dependent manner. Our data reveal that AML cells, microenvironmental factors, and leukemia–stromal interactions cooperatively support leukemia cells through the novel ARC/IL1β/Cox-2/PGE2/β-catenin circuit reported here.

We discovered that ARC is transcriptionally regulated by β-catenin. PGE2 treatment or coculture with MSCs induces β-catenin in AML in agreement with our previous finding that leukemia cells in the marrow had higher β-catenin levels than paired leukemia cells in circulation (27), suggesting that the bone marrow microenvironment induces β-catenin levels in AML cells at least in part through stromal cell secreted PGE2. Furthermore, using a xenograft model of human AML in NSGS mice, we demonstrated that inhibition of ARC by shRNA significantly decreased leukemia burden and serum IL1β/PGE2 levels, prolonged survival, and sensitized to chemotherapy. Interestingly, in addition to decreased ARC levels, low β-catenin levels in huCD45⁺ cells were also detected in mouse tissues injected with ARC KD AML cells compared with controls, supporting the ARC/IL1β/Cox/PGE2/β-catenin regulatory circuit in vivo.

We demonstrate that both IL1β and PGE2 levels are significantly higher in bone marrow from AML patients compared with normal controls and that the levels of bone marrow β-catenin and ARC strongly correlate in AML bulk and stem/progenitor cells, supporting the regulatory cascade in primary AML. It is technically challenging to establish ARC knockdown or overexpression in primary AML cells and currently no specific ARC inhibitor is available. However, targeting the components of the regulatory circuit can be explored. We previously reported that inhibition of β-catenin by its selective inhibitor PRI-724 (C-82 prodrug) has antileukemia activity in vivo in immunodeficient mice engrafted either with an AML cell line or AML PDX cells and sensitized to tyrosine kinase inhibitor in FLT3 mutated AML (27). We here show that C-82 decreases ARC and sensitizes to Ara-C in bulk and CD34⁺CD38⁻ AML cells even when they were cocultured with MSCs.

Evasion of immunity in hematologic malignancies has been increasingly recognized in recent years (38). Wnt/β-catenin and Cox-2/PGE2 signaling are both implicated in immunosuppression (39). As ARC is a component of the IL1β/Cox-2/PGE2/β-catenin cascade that connects β-catenin and IL1β and given our finding that high ARC expression is a strong adverse predictor of survival in AML, we envision that higher ARC levels in AML may also contribute to a more inflammatory and immunosuppressive microenvironment, a hypothesis that is currently under investigation.

In conclusion, we demonstrate that ARC is an integral component of a novel IL1β/Cox-2/PGE2/β-catenin circuit, plays a critical role in leukemia growth and the leukemia microenvironment, and may serve as a potential novel therapeutic target in AML. We are currently trying to develop inhibitors targeting ARC. The finding that ARC functions as a survival modulator through multiple mechanisms explains its strong adverse effect in AML. This sets ARC apart from other antiapoptotic proteins.

J.K. Burks is a consultant/advisory board member for Fluidigm. No potential conflicts of interest were disclosed by the other authors.

Conception and design: B.Z. Carter, V. Ruvolo, M. Andreeff

Development of methodology: P.Y. Mak, W. Tao, V. Ruvolo, J.K. Burks

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.Y. Mak, X. Wang, W. Tao, V. Ruvolo, D. Mak, H. Mu, J.K. Burks

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.Z. Carter, P.Y. Mak, X. Wang, W. Tao, H. Mu, J.K. Burks, M. Andreeff

Writing, review, and/or revision of the manuscript: B.Z. Carter, V. Ruvolo, M. Andreeff

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Mak, H. Mu, M. Andreeff

Study supervision: B.Z. Carter, M. Andreeff

We thank Drs. Numsen Hail for editorial support and Ivo Veletic for technical assistance. This work was supported in part by grants from the University Cancer Foundation via the Institutional Research Grant program at MD Anderson to B.Z. Carter and from the NIH (P01CA055164), Cancer Prevention Research Institute of Texas (CPRIT, RP121010), and by the Paul and Mary Haas Chair in Genetics to M. Andreeff and MD Anderson's Cancer Center Support Grant CA016672 (Flow Cytometry and Cellular Image Facility and Characterized Cell Line core).

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.