CXCR4 Chemokine Receptor Signaling Induces Apoptosis in Acute Myeloid Leukemia Cells via Regulation of the Bcl-2 Family Members Bcl-XL, Noxa, and Bak*

Background: The chemokine receptor CXCR4 plays a role in AML. Results: SDF-1, the ligand of CXCR4, induces apoptosis in AML cell lines and patient samples via modulation of Bcl-2 family members. Conclusion: SDF-1 induces apoptosis of AML cells via up-regulation of Bak and Noxa and down-regulation of Bcl-XL. Significance: SDF-1/CXCR4 signaling could induce AML cell apoptosis if bone marrow survival cues can be disrupted. The CXCR4 chemokine receptor promotes survival of many different cell types. Here, we describe a previously unsuspected role for CXCR4 as a potent inducer of apoptosis in acute myeloid leukemia (AML) cell lines and a subset of clinical AML samples. We show that SDF-1, the sole ligand for CXCR4, induces the expected migration and ERK activation in the KG1a AML cell line transiently overexpressing CXCR4, but ERK activation did not lead to survival. Instead, SDF-1 treatment led via a CXCR4-dependent mechanism to apoptosis, as evidenced by increased annexin V staining, condensation of chromatin, and cleavage of both procaspase-3 and PARP. This SDF-1-induced death pathway was partially inhibited by hypoxia, which is often found in the bone marrow of AML patients. SDF-1-induced apoptosis was inhibited by dominant negative procaspase-9 but not by inhibition of caspase-8 activation, implicating the intrinsic apoptotic pathway. Further analysis showed that this pathway was activated by multiple mechanisms, including up-regulation of Bak at the level of mRNA and protein, stabilization of the Bak activator Noxa, and down-regulation of antiapoptotic Bcl-XL. Furthermore, adjusting expression levels of Bak, Bcl-XL, or Noxa individually altered the level of apoptosis in AML cells, suggesting that the combined modulation of these family members by SDF-1 coordinates their interplay to produce apoptosis. Thus, rather than mediating survival, SDF-1 may be a means to induce apoptosis of CXCR4-expressing AML cells directly in the SDF-1-rich bone marrow microenvironment if the survival cues of the bone marrow are disrupted.

The CXCR4 chemokine receptor promotes survival of many different cell types. Here, we describe a previously unsuspected role for CXCR4 as a potent inducer of apoptosis in acute myeloid leukemia (AML) cell lines and a subset of clinical AML samples. We show that SDF-1, the sole ligand for CXCR4, induces the expected migration and ERK activation in the KG1a AML cell line transiently overexpressing CXCR4, but ERK activation did not lead to survival. Instead, SDF-1 treatment led via a CXCR4dependent mechanism to apoptosis, as evidenced by increased annexin V staining, condensation of chromatin, and cleavage of both procaspase-3 and PARP. This SDF-1-induced death pathway was partially inhibited by hypoxia, which is often found in the bone marrow of AML patients. SDF-1-induced apoptosis was inhibited by dominant negative procaspase-9 but not by inhibition of caspase-8 activation, implicating the intrinsic apoptotic pathway. Further analysis showed that this pathway was activated by multiple mechanisms, including up-regulation of Bak at the level of mRNA and protein, stabilization of the Bak activator Noxa, and down-regulation of antiapoptotic Bcl-X L . Furthermore, adjusting expression levels of Bak, Bcl-X L , or Noxa individually altered the level of apoptosis in AML cells, suggesting that the combined modulation of these family members by SDF-1 coordinates their interplay to produce apoptosis. Thus, rather than mediating survival, SDF-1 may be a means to induce apoptosis of CXCR4-expressing AML cells directly in the SDF-1-rich bone marrow microenvironment if the survival cues of the bone marrow are disrupted.
The bone marrow microenvironment provides a variety of signals that promote the survival of acute myeloid leukemia (AML) 2 cells. Both the endosteal and vascular niches provide an environment that supports normal hematopoietic stem cells as well as leukemic stem cells. A variety of cytokines, chemokines, and integrins present in these niches mediate both the homing and survival signals necessary to propagate leukemia. In addition, the expansion of hypoxic areas during the development of leukemia leads to additional signals that support the proliferation and survival of AML cells (1,2). Determining which of these signaling events are critical for the survival of AML cells will help to identify a strategy to disrupt the protection that the bone marrow microenvironment provides for AML cells.
CXCR4 is a G-protein-coupled chemokine receptor that signals in response to its sole physiological ligand, SDF-1 (stromal cell-derived factor-1; also known as CXCL12). CXCR4 has been characterized as a key player in the development of AML due to the abundant expression of SDF-1 within the bone marrow microenvironment. CXCR4 expression is essential for enabling AML cells to home to the bone marrow (1)(2)(3). Conversely, the CXCR4 antagonist AMD3100 causes mobilization of cells from the protective bone marrow microenvironment. This agent has been approved by the Food and Drug Administration for treatment of lymphoma and multiple myeloma in combination with chemotherapeutics to target the displaced cancer cells. In contrast, AMD3100 has not thus far been approved for treatment of AML (4). Although AMD3100 does mobilize a majority of AML cells from the bone marrow, the leukemic stem cells appear to resist mobilization, remain in the bone marrow, and contribute to the relapse seen in the murine models (5)(6)(7)(8)(9). These unexpected effects indicate the need to better understand the role of CXCR4 in AML.
In addition to initiating migration and ERK activation, CXCR4 has also been shown to contribute to apoptosis in a variety of cell types. In response to prolonged exposure to SDF-1, CXCR4 has been shown to induce apoptosis in T cells and colorectal carcinoma cells (10 -12). Moreover, lymphocytes, neurons, breast cancer cells, endothelial cells, and cardiomyocytes all undergo apoptosis in response to CXCR4 binding to gp120, the envelope glycoprotein of HIV-1 (13)(14)(15)(16).
Previous studies have suggested that CXCR4 can activate death signaling via either the extrinsic or intrinsic pathway. The extrinsic pathway is triggered when ligands on initiator cells bind to death receptors on target cells, leading to caspase-8 activation and downstream apoptotic events in those target cells. The intrinsic apoptotic pathway is activated by changes that tip the balance between anti-and proapoptotic members of the Bcl-2 family. Upon activation, the effector members of this family, Bak and Bax, oligomerize and permeabilize the outer mitochondrial membrane, resulting in the release of cytochrome c, activation of procaspase-9, and subsequent caspasemediated cleavages. The effectors are in turn regulated by the two other classes of Bcl-2 family members. The BH3-only-containing proapoptotic members, which include Bim, Puma, Bid, and Noxa, promote apoptosis by directly binding and inducing the oligomerizaton of Bak and/or Bax or by neutralizing the antiapoptotic members. The antiapoptotic family members directly bind and antagonize Bak and Bax or sequester the BH3only family members to prevent interactions with the effectors (17)(18)(19). In AML cells, various reagents have been described to induce apoptosis via altering levels of different Bcl-2 family members, suggesting that AML cells are sensitive to regulation of Bcl-2 family proteins (20 -23).
Many studies analyzing the role of CXCR4 in AML have focused on inhibiting CXCR4 signaling by utilizing AMD3100 or other CXCR4 antagonists. Based on these prior studies, we set out to determine the signaling mechanisms involved in SDF-1-mediated survival of AML cells. However, we found that SDF-1 did not promote AML cell survival but instead induced apoptosis of AML cells expressing CXCR4. Here, we characterize a novel SDF-1-mediated apoptosis pathway in AML cells and provide evidence suggesting that this intrinsic apoptotic mechanism is normally suppressed within the bone marrow microenvironment. These results describe a new role for SDF-1/CXCR4 signaling in AML cells and indicate that SDF-1/ CXCR4 functions in AML are more complex than previously appreciated.
Cells-After informed consent was obtained, samples of bone marrow were harvested from AML patients prior to chemotherapy according to an institutional review board-approved protocol. Following sedimentation on a Ficoll-Paque (1.077 g/cm 3 ) step gradient (24), mononuclear cells were recovered and cultured in Medium A (RPMI 1640 supplemented with 10% FCS, 2 mM L-glutamine, 1 mM sodium pyruvate, nonessential amino acids, and 5.5 M ␤-mercaptoethanol) at a density of 1 million cells/ml. The AML cell lines, KG1a and U937, and the E6 Jurkat acute T cell leukemia line (ATCC, Manassas, VA) were cultured in RPMI supplemented with 10% FCS at a cell density of 0.6 -1.0 million cells/ml. HeLa cells (ATCC) were cultured to confluence in RPMI supplemented with 10% FCS.
To generate CXCR4-expressing KG1a, U937, and E6 Jurkat cells, we transiently transfected cells with a plasmid encoding a CXCR4-YFP fluorescent fusion protein (25) using a BTX 830 square wave electroporator at 240 V for 10 ms, incubated for 16 -18 h to allow expression of the transgene, and treated as indicated in the various figures. The transfection efficiencies of KG1a and U937 with control YFP empty vector were typically 10 -20% and 10 -15%, respectively. The E6 Jurkat T cell line was transfected as described previously (25). Flow cytometric data were gated to compare cells expressing similar levels of YFP.
To stably suppress either Bak or Bim protein expression, KG1a cells were transduced with lentiviral particles that target the sequence CCCATTCACTACAGGTGAA for Bak or GAC-CGAGAAGGTAGACAATTGC for Bim in pSIH-H1, respectively (26,27). Following transfection of plasmids encoding shRNA or empty vector into HEK293 cells along with the packaging plasmids psPAX2 and pMD2.G, supernatants were collected and filtered through 0.45-m filters. After viral infection on two successive days, KG1a cells were cultured in medium containing 2 g/ml puromycin for a week. Bak-and Bim-specific antibodies were used to assay Bak and Bim expression via Western blot. To transiently suppress Noxa, Bak, or Bim protein expression, KG1a or Jurkat cells were transiently transfected as indicated above with the following siRNAs from Ambion (Austin, TX): Control siRNA (Silencer Negative Control 1); Noxa siRNA (sense, 5Ј-GGAGAUUUGGAGACAAAC-UTT-3Ј; antisense, 5Ј-AGUUUGUCUCCAAAUCUCCTG-3Ј); Bak siRNA (sense, GUACGAAGAUUCUUCAAAUTT; antisense, AUUUGAAGAAUCUUCGUACCA); and Bim siRNA (sense, 5Ј-GACCGAGAAGGUAGACAAUTT-3Ј; antisense, 5Ј-AUUGUCUACCUUCUCGGUCTT-3Ј). The cells were cultured for 6 -7 h prior to the addition of SDF-1. Depletion of Noxa, Bak, and Bim was confirmed by immunoblotting with specific antibodies using lysates prepared 24 h after transfection.
Tert-immortalized bone marrow-derived stromal cells (BMSC) were a kind gift of Dario Campana (St. Jude Children's Research Hospital, Memphis, TN) (28,29). BMSC were cultured in RPMI supplemented with 10% FCS, 2 mM L-glutamine, and 1 M hydrocortisone prior to the addition of AML isolates, at which time the coculture was maintained in Medium A.
Assaying CXCR4 Levels-Clinical AML isolates were cultured for 16 -18 h in Medium A and then assayed for CXCR4 cell surface expression by flow cytometry (30) after staining with a CXCR4 monoclonal antibody. For assaying total cellular CXCR4 levels, cells were incubated at 37°C for 10 min in BD Phosflow Lyse/Fix buffer; sedimented; permeabilized on ice for 30 min with BD Perm Buffer III; washed with BD stain buffer (BD Biosciences); stained with either CXCR4 mAb or a control, anti-EGFR mAb, for 30 min at room temperature; and assayed by flow cytometry. Where indicated, the KG1a cells were treated with 5 ϫ 10 Ϫ8 M SDF-1 for 20 min prior to staining for CXCR4.
Assays of Migration and Active ERK-KG1a cells were transfected with either CXCR4-YFP or YFP empty vector and cultured for 18 h prior to the migration assay or ERK activation assay. For the migration assay, the cells were stained with calcein-AM and assayed for migration as described (31). Where indicated, either 10 M PD325901, 10 M CI-1040, or DMSO (vehicle) was added 16 -18 h prior to the addition of SDF-1, or either 100 ng/ml pertussis toxin or its control toxin was added 4 h prior to the addition of SDF-1. For analysis of active phospho-ERK, cells were stimulated Ϯ 5 ϫ 10 Ϫ8 M SDF-1 and assayed by flow cytometry (31). Flow cytometric data were gated to compare ERK phosphorylation in cells expressing similar levels of YFP.
Apoptosis Assays-For assays measuring apoptosis, the indicated cells were cultured at 0.25 ϫ 10 6 cells/ml in the presence or absence of 1.3 ϫ 10 Ϫ8 M SDF-1 or 6 ng/ml CH.11 agonistic anti-Fas antibody (Millipore) for 16 -18 h and then assayed for apoptosis using one of the techniques listed below. Where indicated, 30 M AMD3100, 1 M CCX704, 1 M CCX771, 10 M PD325901, 10 M CI-1040 was added 1 h prior to the addition of SDF-1; 100 ng/ml pertussis toxin or its control toxin was added 4 h prior to the addition of SDF-1; cells were placed in a hypoxia chamber at 37°C, 5% CO 2 , and 1% O 2 for 16 -18 h prior to the annexin V staining; cells were incubated for 2 h under hypoxic or normoxic conditions and then treated with 1 M cytarabine for 16 -18 h in the indicated oxygen conditions; cells were cotransfected with CXCR4-YFP and a plasmid encoding the viral c-FLIP homolog MC159 (a kind gift from Greg Gores, Mayo Clinic, Rochester, MN), a dominant-negative caspase-9/ EGFP fusion protein (obtained from Emad Alnemri, Thomas Jefferson University, Philadelphia, PA), or human Noxa with L29A and D34A mutations (Noxa-2A) (32) fused to EGFP or S-peptide-tagged-Bcl-X L ; or cells were transfected with GFP-Bak or GFP-Noxa. MC159 expression was confirmed via semiquantitative RT-PCR with the following primers: MC159 Forward, ATGTCCGACTCCAAGGAGGTCCCTAG; MC159 Reverse, TCAAGTCGTTTGCTCGGGGCTGTCGCTG; GAPDH For-ward, GGCAAATTCCATGGCACCGTCAGG; GAPDH Reverse, GGAGGCATTGCTGATGATCCTGAGG.
For detecting cell surface phosphatidylserine, cells were stained with APC-conjugated annexin V as described (33). To detect fragmented nuclei, cells were washed with PBS, resuspended in 3:1 methanol/acetic acid, incubated at room temperature for 1 h, centrifuged, and mounted in 50% glycerol, 50% 0.1 M Tris-HCl (pH 7.4) containing 1 g/ml Hoechst 33258 (34). The cells were then visualized 16 h later with Hoechst/DAPI filters. Cleaved PARP and active caspase-3 were assayed by flow microfluorimetry in YFP-positive cells using an Alexafluor-647-conjugated antibody to cleaved PARP and an active caspase-3 apoptosis kit, respectively, according to the suppliers' instructions.
Apoptosis Assay with Clinical AML Isolates-AML isolates were cultured in Medium A for 1-2 h prior to plating on BMSC. The isolates were plated at a density of 0.25 ϫ 10 6 cells/ml onto a confluent layer of BMSC for 1 h, and then 5 ϫ 10 Ϫ8 M SDF-1 was added for 16 -18 h in Medium A. Annexin V staining was performed as described above.
Analyzing Noxa Stability-KG1a cells were cotransfected with CXCR4-YFP and Noxa2A-GFP; cultured for 16 h with the caspase inhibitor Q-VD-OPh in the presence or absence of SDF-1; and then treated with 25 g/ml cycloheximide for the indicated time, fixed with paraformaldehyde, and analyzed via flow microfluorimetry for Noxa2A-GFP expression in gated CXCR4-YFP-positive cells. The amount of Noxa2A-GFP remaining after the indicated cycloheximide treatment was determined as a percentage of the Noxa2A-GFP present at the 0 h time point.

CXCR4 Is Expressed at Variable Levels on AML Cells-In
initial experiments, we observed that CXCR4 is expressed at varying levels on the cell surface of primary AML cells from patient bone marrow (Fig. 1A). We also confirmed that CXCR4 is expressed intracellularly in patient samples, as indicated by the increased signal observed upon permeabilizing the plasma membrane prior to CXCR4 staining (Fig. 1B). Patient data shown in Fig. 1, A and B, are representative of all patients analyzed during the study period (supplemental Table 1). These results confirm previous findings that CXCR4 is expressed variably on the cell surface of AML isolates and that CXCR4 is maintained intracellularly at high levels in most AML isolates (9,37,38).
To begin to characterize the signaling pathways initiated by SDF-1/CXCR4 signaling in AML cells, we utilized the human M0 AML cell line KG1a, which has been extensively studied as a model of human AML. KG1a cells lack CXCR4 expression on the cell surface ( Fig. 1C) (9,37,39); however, KG1a cells do express CXCR4 that is maintained inside the cell, as shown by CXCR4 staining following permeabilization ( Fig. 1C) (9). To analyze the effects of SDF-1 treatment on KG1a, we transiently overexpressed CXCR4 at sufficiently high levels to achieve cell surface expression. A CXCR4-YFP fluorescent fusion protein was used for this purpose in order to permit the unambiguous identification of individual, transfected cells expressing similar levels of CXCR4 via flow cytometry. Here and below, we refer to these CXCR4-expressing KG1a cells as "KG1a-CXCR4." Control cells consisted of KG1a cells transiently transfected with a YFP-encoding vector in the same experiment. Compared with cells transiently transfected with a control vector, the KG1a-CXCR4 displayed elevated cell surface levels of CXCR4 (Fig.  1C). As expected, the level of CXCR4 on the cell surface of KG1a-CXCR4 cells was down-regulated in response to SDF-1 (Fig. 1C).
SDF-1 Signaling Induces Migration and ERK Activation in the KG1a-CXCR4-SDF-1/CXCR4 signaling is critical for homing of AML cells to the SDF-1-rich bone marrow microenvironment, where they are protected from chemotherapeutic agents (1,2,40). Consistent with these results, transient CXCR4 expression enabled SDF-1-induced migration of KG1a cells, in contrast to KG1a cells transfected with a vector control plasmid (p Ͻ 0.05; Fig. 2A).

FIGURE 1. CXCR4 is expressed at variable levels on AML cells.
A, bone marrow isolates were harvested from AML patients prior to chemotherapy. Patient samples were cultured for 16 -18 h before levels of CXCR4 protein on the cell surface were analyzed via incubation of intact cells with anti-CXCR4 mAb followed by flow microfluorimetry. Results from nine different patient isolates are shown. B, to assay intracellular CXCR4 protein levels, cells from AML isolates were fixed, permeabilized, and then incubated with either a CXCR4 antibody or a control antibody prior to flow microfluorimetry. Results of a representative isolate are shown; n ϭ 8. C, either endogenous, cell surface CXCR4 (left) or endogenous intracellular CXCR4 (middle) was assayed for the human AML cell line, KG1a; representative results are shown, n ϭ 3. Right, cell surface CXCR4 levels of KG1a cells transfected with either a control plasmid (filled histogram) or a plasmid encoding the CXCR4-YFP fluorescent fusion protein (open histograms) are shown, either before (Unstim.) or after 20 min stimulation with SDF-1 (SDF-1) that induced CXCR4 internalization; representative results from one experiment are shown; n ϭ 3.
Next we assessed activation of the ERK mitogen-activated protein kinase, a key initiator of SDF-1-induced survival pathways in a variety of cell types (41,42). Transfected cells were treated with SDF-1 for the indicated times, and levels of active, phosphorylated ERK in individual cells were assayed by flow microfluorimetry. KG1a-CXCR4 cells responded to SDF-1 treatment by significantly increasing levels of active, phosphorylated ERK at 2 and 5 min, with this response declining at 8 min (p Ͻ 0.05; Fig. 2, B and C). In contrast, there was no change in ERK phosphorylation in cells expressing empty vector. In summary, KG1a cells transiently expressing CXCR4 are able to migrate and increase mitogen-activated protein kinase signaling in response to SDF-1. These results are consistent with a previous report that similarly analyzed KG1a cells overexpressing CXCR4 (40).

SDF-1 Induces KG1a Cell Apoptosis via CXCR4-Because
KG1a-CXCR4 cells undergo SDF-1-induced migration and ERK activation, we anticipated that prolonged SDF-1 treatment would protect KG1a-CXCR4 cells from death. To assess survival, we assayed annexin V staining on KG1a-CXCR4 cells following 16 -18 h of SDF-1 treatment. Surprisingly, SDF-1 treatment resulted in a 3-fold increase in the percentage of annexin V-positive KG1a-CXCR4 cells compared with unstimulated cells (p Ͻ 0.05; Fig. 3, A and B). In contrast, cells transfected with the YFP vector control plasmid did not show a significant increase in annexin V-positive cells in response to SDF-1.
To rule out the possibility that increased annexin V staining was a result of plasma membrane reorganization without apoptosis, as has been observed in mitogen-stimulated lymphocytes (43,44), we stained SDF-1-treated cells with Hoechst 33258, a dye that allows visualization of nuclear morphological changes. As shown in Fig. 3C, KG1a-CXCR4 cells treated with SDF-1 displayed fragmented nuclei, a hallmark of apoptosis. Utilizing fluorochrome-conjugated antibodies specific for cleaved PARP and cleaved caspase-3, we also observed that SDF-1 treatment induced a significant increase in cleaved PARP and cleaved procaspase-3 in KG1a-CXCR4 cells (p Ͻ 0.05; Fig. 3

, D-F).
To confirm that SDF-1 was causing apoptosis via CXCR4, we examined the relationship between levels of cell surface CXCR4 and annexin V binding. Our results demonstrate that increased CXCR4 on the cell surface, as indicated by YFP fluorescence, led to a significantly increased percentage of annexin V-positive cells following SDF-1 treatment (p Ͻ 0.05; Fig. 3G). Similarly, the percentage of apoptotic cells also increased in a dose-dependent manner with increasing amounts of SDF-1 (Fig. 3H). The concentration of SDF-1 that achieves half-maximal increase in apoptosis is ϳ4.1 ϫ 10 Ϫ9 M, a result fully consistent with SDF-1 inducing apoptosis by binding to CXCR4 (25,45). In addition, use of the CXCR4-specific antagonist, AMD3100, prevented SDF-1-induced apoptosis in KG1a-CXCR4 cells (p Ͻ 0.05; Fig. 3I). Collectively, these results show that SDF-1 induces apoptosis in KG1a-CXCR4 cells via a mechanism requiring CXCR4.
SDF-1/CXCR4 Induces Apoptosis in the U937 AML Cell Line and in a Subset of Clinical AML Isolates-To ensure that the apoptosis induced by SDF-1 was not unique to KG1a cells, we utilized U937, an M2 AML cell line. U937 cells express low levels of CXCR4 on the cell surface (9), lower even than many clinical AML isolates. Accordingly, we transiently transfected U937 with CXCR4-YFP and gated on YFP-positive cells, which we designated U937-CXCR4 cells (Fig. 4A). In contrast to U937 cells transfected with the vector control plasmid, SDF-1 significantly increased the percentage of annexin V-positive U937-CXCR4 cells (p Ͻ 0.05; Fig. 4B).
In addition, we obtained a number of clinical AML isolates and assayed for the ability of SDF-1 to induce apoptosis via  AUGUST 9, 2013 • VOLUME 288 • NUMBER 32 endogenously expressed CXCR4. Interestingly, SDF-1 treatment increased the percentage of annexin V-positive cells in three of 10 isolates (Fig. 4C). Samples from patients 4 and 10, which had the highest CXCR4 cell surface expression, also displayed a substantial increase in SDF-1-induced apoptosis, suggesting that higher CXCR4 levels may correlate with increased apoptosis in response to SDF-1 (Figs. 1A and 4C). Unfortunately, we were unable to assess CXCR4 cell surface levels in the sample from patient 3 due to the limited number of cells obtained. Together, the results in Figs. 1-4 indicate that SDF-1/CXCR4 signaling can induce apoptosis in CXCR4-expressing AML cell lines and also in a subset of clinical AML isolates expressing high levels of endogenous cell surface CXCR4.

SDF-1 Modulates Bcl-X L , Noxa, and Bak to Induce Apoptosis
Hypoxia Protects CXCR4-expressing AML Cell Lines from Apoptosis-Because SDF-1 is abundant in the bone marrow microenvironment, it was unclear how CXCR4-expressing AML cells could thrive in this environment if they are sensitive to SDF-1-induced cell death. We hypothesized that an environmental signal must override or inhibit SDF-1-induced apoptosis. One possible inhibitory signal may derive from areas of hypoxia in the bone marrow microenvironment. Expansion of hypoxic regions has been noted to occur during the progression of leukemia, with leukemic cells adapting and then proliferating in these hypoxic regions (2). When KG1a-CXCR4 cells were exposed to SDF-1 in an environment of 1% oxygen, significantly fewer cells underwent apoptosis as compared with KG1a-CXCR4 cells treated with SDF-1 in ambient oxygen as measured by staining for either annexin V or cleaved PARP (p Ͻ 0.05; Fig. 5, A and B). Although SDF-1 was still able to induce an increase in apoptosis under hypoxic conditions, the overall level of apoptosis in SDF-1-stimulated cells under hypoxic conditions was low compared with normoxic conditions. These results suggest that hypoxia provides signals to protect AML cells from apoptosis. This protective effect of hypoxia on KG1a-CXCR4 cells treated with SDF-1 was not due to reduced CXCR4 cell surface levels (Fig. 5C). Similarly, both the unstimulated and SDF-1-treated U937-CXCR4 cells were significantly protected from apoptosis by the hypoxic environment (p Ͻ 0.05; Fig. 5, D and E), again without any reduction in cell surface CXCR4. Indeed, the U937-CXCR4 cells significantly increased levels of cell surface CXCR4 upon exposure to hypoxia (p Ͻ 0.05; Fig. 5F). To determine if hypoxia protects AML cells from apoptosis induced by a different treatment, we exposed KG1a cells to cytarabine, currently the most effective single agent for treatment of AML (46), and measured apoptosis under normoxic and hypoxic conditions. Compared with normoxic conditions, cytarabine-treated cells maintained in a hypoxic environment exhibited significantly less apoptosis (p Ͻ 0.05; Fig. 5G). These results suggest that hypoxia could provide a survival signal that allows AML cells to thrive in the bone marrow microenvironment, despite the presence of apoptosisinducing agents, including SDF-1.
CXCR7, ERK Activation, and G i Proteins Do Not Mediate SDF-1-induced Apoptosis in AML Cells-To further confirm that SDF-1 was inducing apoptosis via CXCR4, we assessed the role of CXCR7 (47)(48)(49)(50), the only other chemokine receptor known to bind SDF-1. CXCR7 was not detectably expressed on the cell surface of KG1a or U937 but was expressed on HeLa cells (Fig. 6A). Furthermore, treatment of the KG1a-CXCR4 cells with the CXCR7 antagonist, CCX771, did not inhibit the SDF-1-induced apoptosis compared with the KG1a-CXCR4 cells treated with the control compound, CCX704 (p Ͼ 0.05; Fig. 6B). Thus, CXCR7 does not mediate SDF-1-induced apoptosis in AML cells.
ERK activation, which often mediates survival signals, has also been reported to induce apoptosis in some cell types (51)(52)(53). To determine whether the SDF-1-induced ERK activation demonstrated in Fig. 2, B and C, contributes to SDF-1-induced apoptosis, we exposed cells to the MEK inhibitor PD325901 or CI-1040 before SDF-1 treatment. As shown in Fig. 6C, pretreating KG1a-CXCR4 cells with these MEK inhibitors did not substantially inhibit SDF-1-induced apoptosis, although the MEK inhibitors did effectively abolish ERK activation in response to SDF-1 (p Ͻ 0.5; Fig. 6, D and E).
Signaling by G i -type G proteins has previously been shown to induce apoptosis in hepatocytes and neutrophils (54, 55), and  CXCR4 is able to signal through G i -type G proteins in response to SDF-1 (25,48,56). To address whether CXCR4 utilizes G itype G proteins to induce apoptosis, we treated KG1a-CXCR4 cells with pertussis toxin (PTX), an exotoxin that inhibits G itype G proteins. PTX did not substantially inhibit SDF-1-induced apoptosis in KG1a-CXCR4 cells as compared with cells treated with the control toxin PTX-B (p Ͻ 0.5; Fig. 6F). However, PTX did significantly inhibit SDF-1-induced ERK activa-tion in KG1a-CXCR4 cells (p Ͻ 0.05; Fig. 6, G and H). Together, the results in Fig. 6 indicate that SDF-1/CXCR4-induced apoptosis is not mediated by pathways that require signaling by CXCR7, ERK, or G i -type G proteins.
Requirement for the Caspase-9/Intrinsic Death Pathway, but Not the Caspase-8/Extrinsic Pathway, for SDF-1/CXCR4-mediated Apoptosis-To further characterize the apoptosis induced by SDF-1/CXCR4 signaling in AML cells, we first determined
Role of SDF-1-mediated Bak Up-regulation in SDF-1/CXCR4mediated Apoptosis-To evaluate potential mechanisms for SDF-1-induced activation of the intrinsic pathway, we assessed the effect of SDF-1 on levels of mRNA encoding the antiapoptotic Bcl-2 family members Bcl-2, Bcl-2A1, Bcl-X L , and Mcl-1  Fig. 3A; n ϭ 3. D and E, KG1a-CXCR4 cells were treated with PD325901, CI-1040, or vehicle for 16 -18 h, stimulated with SDF-1 for 2 min, and then assayed for ERK activation as in Fig. 2, B and C. *, significantly different from control-treated cells, p Ͻ 0.05; n ϭ 3. G and H, KG1a-CXCR4 cells were cultured for 16 h, treated with PTX-B or PTX for 4 h, stimulated with SDF-1 for 2 min, and assayed for ERK activation as in Fig. 2, B and C. *, significantly different from SDF-1-stimulated control-treated cells, p Ͻ 0.05, n ϭ 3. Error bars, S.E. as well as the proapoptotic Bcl-2 family members Bim, Noxa, Bmf, BNIP3, Bak, and Bax in KG1a-CXCR4 cells. As shown in Fig. 8A, only Bak mRNA levels in KG1a-CXCR4 cells were substantially altered by SDF-1 treatment (p Ͻ 0.05). Consistent with these results, Western blotting showed that SDF-1 also up-regulates Bak at the protein level in KG1a-CXCR4 cells (Fig. 8B).
We next determined whether increased expression of Bak alone was sufficient to induce apoptosis. Transiently expressing a plasmid encoding GFP-Bak in KG1a cells resulted in a dramatic decrease in the number of GFP-positive cells compared with the control GFP vector (Fig. 8C). Treating the GFP-Bak expressing cells with the caspase inhibitor Q-VD-OPh resulted in a greater number of GFP-positive cells (Fig. 8C), suggesting that GFP-Bak is inducing apoptosis before expression levels get very high. Indeed, ϳ93% of GFP-Bak expressing cells were annexin V-positive in the absence of Q-VD-OPh (Fig. 8, D and E).
To further confirm the role of Bak in SDF-1-induced apoptosis, we created KG1a cells that had Bak stably knocked down by shRNA to ϳ30% of control levels (Fig. 8F). Compared with the KG1a-vector shRNA control cells expressing CXCR4-YFP, KG1a-shBak cells expressing CXCR4 exhibited significantly less apoptosis after treatment with SDF-1 (p Ͻ 0.05; Fig. 8G). As a control, despite an 85% reduction in Bim, a KG1a-shBim cell line expressing CXCR4 had similar levels of SDF-1-induced apoptosis as the KG1a-vector shRNA control cells (Fig. 8, H  and I). These results indicate that Bak is required to mediate SDF-1-induced apoptosis and that SDF-1-mediated up-regulation of Bak is sufficient to induce apoptosis in CXCR4-expressing AML cells.

SDF-1-induced Down-regulation of Bcl-X L and Up-regulation of Noxa
Contribute to Apoptosis of AML Cells-To assess the potential role of SDF-1-induced changes in other Bcl-2 family members, we assayed the protein levels of Bcl-X L , Bcl-2, Mcl-1, Noxa, and Bax in KG1a-CXCR4 cells treated with SDF-1. As shown in Fig. 9A, Noxa expression increased and Bcl-X L expression decreased in response to SDF-1 treatment, whereas Bcl-2, Mcl-1, and Bax remained unchanged. Noxa is a proapoptotic Bcl-2 family member that, upon binding to Bak increases apoptosis, whereas Bcl-X L is an antiapoptotic family member that inhibits apoptosis upon binding Bak (32,59,60).
To analyze the potential contribution of Bcl-X L down-regulation to SDF-1-induced death, we evaluated whether forced overexpresson of Bcl-X L could protect the cells. As anticipated, fewer KG1a-CXCR4 cells expressing S-peptide-tagged Bcl-X L bound annexin V compared with KG1a-CXCR4 cells expressing the vector control (p Ͻ 0.05; Fig. 9, B and C). Thus, in combination with Bak up-regulation, SDF-1 utilizes Bcl-XL down-regulation as an additional mechanism to mediate apoptosis in AML cells.
To determine whether increased expression of proapoptotic Noxa is sufficient to increase apoptosis in AML cells, we transiently expressed a plasmid encoding GFP-Noxa in KG1a cells (Fig. 9D). Similar to transient expression of GFP-Bak (Fig. 8, C  and D), treatment with Q-VD-OPh increased the number of cells expressing GFP-Noxa compared with the untreated cells (Fig. 9E), indicating that induction of apoptosis limits how much Noxa can be expressed in these cells. Likewise, ϳ92% of GFP-Noxa-expressing cells were annexin V-positive in the absence of Q-VD-OPh (p Ͻ 0.05; Fig. 9, F and G), indicating FIGURE 7. Requirement for the caspase-9/intrinsic death pathway but not the caspase-8/extrinsic pathway, for SDF-1/CXCR4-mediated apoptosis. A and C, KG1a or U937 cells were transiently transfected with CXCR4-YFP together with either a control plasmid (Vector) or a plasmid encoding either MC159 or DN-Casp-9-GFP. Cells were then assayed for SDF-1-dependent apoptosis via annexin V binding as in Fig. 3A. *, significantly different from SDF-1-treated control vector; p Ͻ 0.05; n ϭ 3. B, Jurkat T cells were transiently transfected with a plasmid encoding either a control vector or the caspase-8 inhibitor, MC159, and then stimulated with agonistic anti-Fas antibody for 16 -18 h and assayed for apoptosis via annexin V binding as in Fig.  3A. *, significantly different from the CH.11-treated control vector results; p Ͻ 0.05; n ϭ 3. D, immunoblots of whole cell lysates from A and C, showing overexpression of DN-Casp-9-GFP as compared with endogenous Procasp-9. The same membrane was stripped and reblotted for actin as a control. E, mRNA from cells transfected as in A-C was assayed via RT-PCR for expression of the mRNA encoding the viral protein, MC159. GAPDH mRNA was assayed as a control; n ϭ 3. Error bars, S.E. that increased expression of Noxa can mediate apoptosis in KG1a cells.
To address the mechanism by which SDF-1 increases Noxa expression, we analyzed the stability of Noxa protein in the cells stimulated with SDF-1. Because the elevated levels of endogenous Noxa that arise in response to SDF-1 rapidly cause the apoptosis of these cells (Fig. 9, F and G), we measured turnover of Noxa2A-GFP, a fluorescent fusion protein of Noxa that has amino acids Leu-29 and Asp-34 mutated to inhibit its interaction with Bak and ability to cause apoptosis (32). SDF-1 treatment significantly increased the stability of Noxa2A-GFP in KG1a-CXCR4 cells as compared with unstimulated cells (p Ͻ 0.05; Fig. 9H).
To assess whether the increase in Noxa protein expression contributes to apoptosis in response to SDF-1, we decreased Noxa expression via siRNA (Fig. 9I). Noxa depletion resulted in a significant decrease in SDF-1-induced apoptosis as well as in the basal level of apoptosis (p Ͻ 0.05; Fig. 9J). Thus, SDF-1 treatment increases Noxa expression and thereby apoptosis in AML cells via a mechanism that involves increased stability of the Noxa protein.
To determine whether these results were unique to AML cells, we examined the effect of SDF-1 treatment after CXCR4  Fig. S1B). Thus, increasing CXCR4 levels above endogenous levels permits SDF-1 treatment to drive apoptosis in Jurkat cells. To assess whether SDF-1 utilizes similar mechanisms to induce apoptosis in Jurkat cells as in AML cells, we used siRNA to knock down Noxa, Bak, and Bim (supplemental Fig. S1C). In contrast to cells depleted of Bim, depletion of either Noxa or Bak prevented SDF-1-induced apoptosis in Jurkat-CXCR4 cells (supplemental Fig. S1D) as it did in AML cells (Figs. 8 and 9). Thus, high levels of CXCR4 may be capable of mediating apoptosis in other cell types in addition to AML cells.
Collectively, the results in Figs. 8 and 9 indicate that SDF-1 potently induces the apoptosis of KG1a AML cells by a multifactorial intrinsic mechanism involving up-regulation of the apoptosis effector Bak and BH3-only protein Noxa with the simultaneous down-regulation of the antiapoptotic family member Bcl-X L .

DISCUSSION
Although the initial response rate to current AML therapies is a promising 60 -70%, relapse and death within a few years of initial treatment continues to be a frequent outcome. Relapse is most often mediated by resistance to therapy that is probably driven by the difficulty of eradicating the leukemia stem cells residing within the protected bone marrow microenvironment. Chemotherapeutics are not as effective at targeting the AML cells present in the bone marrow (1,2,61). AMD3100, the CXCR4 antagonist, showed initial promise in removing a large number of AML cells from the bone marrow to the bloodstream, where they could then be targeted by the chemotherapeutics. More recent evidence, however, indicates that AMD3100 may have a more limited benefit in AML because CXCR4 appears to only be required for initial homing of AML cells to the bone marrow, not for their retention within the bone marrow (1,9,62,63). Thus, targeting leukemic stem cells within the bone marrow microenvironment may be a more effective strategy for eliminating these cells.
We show here for the first time that SDF-1 treatment initiates an intrinsic apoptotic pathway in AML cells expressing CXCR4. Essential features of this pathway (Fig. 10) include upregulation of Bak expression at the mRNA and protein level as well as increasing the stability and protein expression of the Bak activator Noxa (32) and down-regulation of the Bak inhibitor Bcl-X L (59, 60). As summarized below, the effects of a variety of pathway perturbations, including treatment with AMD3100; forced alterations in expression of Bak, Noxa, and Bcl-X L ; and treatment with dominant negative procaspase 9 or caspase inhibitors, are consistent with this model.
We characterized the SDF-1-mediated apoptotic process utilizing the AML cell lines KG1a and U937 as well as AML isolates from patients. Transient expression of CXCR4 on KG1a resulted in the anticipated migration and ERK activation in response to short treatments with SDF-1 (Fig. 2), as has been previously characterized for SDF-1/CXCR4 signaling (40). Surprisingly, ERK activation in response to SDF-1 did not promote the survival of KG1a cells expressing CXCR4. Instead, an 18-h SDF-1 treatment triggered apoptosis, as indicated by phosphatidylserine exposure on the cell surface, nuclear fragmentation, activation of caspase-3, and cleavage of PARP (Fig. 3).
We also show here that SDF-1 induced apoptosis in 3 of 10 clinical AML isolates, whereas the level of apoptosis remained largely unchanged in the other patient isolates (Fig. 4C). Our additional studies confirmed that cell surface expression of CXCR4 varied widely among clinical AML isolates. Importantly, both of the AML isolates with high levels of CXCR4 had an increase in apoptosis in response to SDF-1 (cf. Figs. 1A and   4C). Unfortunately, the third clinical sample that exhibited SDF-1-induced apoptosis had too few cells to assay CXCR4 levels. Nonetheless, the correlation between high cell surface CXCR4 levels and SDF-1-induced apoptosis in the samples where both assays could be performed probably indicates that a subset of AMLs (i.e. the set with high cell surface CXCR4 expression) is sensitive to SDF-1-induced apoptosis. Further analysis of clinical samples will be required to determine whether SDF-1-sensitive phenotype tracks with any other clinicopathological parameter. Meanwhile, we also showed that CXCR4 was retained inside the cell at relatively high levels in all of the AML isolates assayed, in agreement with previous studies (9,37). Retention of pools of intracellular CXCR4 has been described in a variety of cancers; however, the mechanisms utilized to prevent transport to the cell surface have not been described (64 -66). In view of the relationship between cell surface CXCR4 expression and SDF-1-induced apoptosis, strategies to force trafficking of CXCR4 to the cell surface could alter the survival of these cancer cells.
Our studies showed that SDF-1 mediates apoptosis via a pathway that involves modulation of Bak, Noxa, and Bcl-X L as well as the initiator caspase of the intrinsic pathway, procaspase-9 (Fig. 10). We further showed that this pathway does not require signaling by CXCR7, ERK activation, or G i -type G proteins. To characterize the mechanism of this SDF-1-mediated apoptosis, we analyzed the SDF-1-induced changes in protein expression levels of Bcl-2 family proteins and we used the FIGURE 10. Proposed model of SDF-1-induced apoptosis. SDF-1-induced CXCR4 activation induces Noxa stabilization, Bak up-regulation, and Bcl-X L down-regulation. Collectively, these lead to Bax/Bak activation followed by activation of caspase-9 and caspase-3. Consistent with this model, increased expression of Noxa or Bak is shown to induce apoptosis in these cells, and Bcl-X L expression, Bak shRNA, or Noxa siRNA is shown to inhibit SDF-1-induced apoptosis. Likewise, AMD3100 and DN-Casp-9 inhibit apoptosis, providing further support for the involvement of CXCR4 and the caspase 9-mediated intrinsic pathway in this process.
procaspase-8 and procaspase-9 inhibitors. Further support for our model comes from the ability of Bak down-regulation, Noxa down-regulation, or Bcl-X L overexpression to protect the cells from SDF-1-induced apoptosis. We previously demonstrated that Noxa can directly bind and oligomerize Bak (32). Utilizing Noxa2A-GFP, a form of Noxa that cannot bind Bak and kill the cells, we show here that SDF-1 increases the half-life of Noxa as a means to mediate apoptosis. Conversely, we and others have shown that Bcl-X L has the ability to bind and neutralize active Bak (59,60). Thus, our results together indicate that SDF-1 treatment affects three related proteins in the mitochondrial apoptotic pathway in a concerted fashion to mediate apoptosis.
If SDF-1 can induce apoptosis in CXCR4-expressing AML cells, how can AML cells survive in the SDF-1-rich environment of the bone marrow? A variety of cancers have coopted hypoxia-driven signaling pathways in order to resist the death initiated by apoptosis-inducing agents, via a large variety of mechanisms including hypoxia-inducible factors, p53, and regulation of the Bcl-2 family members (67,68). We show here that hypoxic conditions significantly decrease the level of apoptosis seen in AML cells in either the absence or presence of apoptosis-inducing agents, such as SDF-1 or cytarbine (Fig. 5). Thus, hypoxia provides a survival signal that can override apoptosisinducing signals in AML cells. These observations highlight the importance of our ongoing studies of SDF-1-induced apoptosis in AML cells in hopes of understanding how its inhibition in the SDF-1-rich bone marrow microenvironment can be overcome.
In summary, the present results indicate that SDF-1 is not a factor utilized by the bone marrow microenvironment to protect AML cells. Instead, SDF-1 may be a means to induce robust apoptosis in CXCR4-expressing AML cells directly in the bone marrow microenvironment, particularly if the survival cues of the SDF-1-rich bone marrow can be disrupted. Further characterization of this novel pathway may provide valuable insights into targets that may be modulated to enhance the killing of AML cells.