HoxA10 Activates Transcription of the Gene Encoding Mitogen-activated Protein Kinase Phosphatase 2 (Mkp2) in Myeloid Cells*

HoxA10 is a homeodomain transcription factor that is frequently overexpressed in human acute myeloid leukemia. In murine bone marrow transplantation studies, HoxA10 overexpression induces a myeloproliferative disorder with accumulation of mature phagocytes in the peripheral blood and tissues. Over time, differentiation block develops in these animals, resulting in acute myeloid leukemia. In immature myeloid cells, HoxA10 represses transcription of some genes that confer the mature phagocyte phenotype. Therefore, overexpressed HoxA10 blocks differentiation by repressing myeloid-specific gene transcription in differentiating myeloid cells. In contrast, target genes involved in myeloproliferation due to HoxA10 overexpression have not been identified. To identify such genes, we screened a CpG island microarray with HoxA10 co-immunoprecipitating chromatin. We identified the DUSP4 gene, which encodes mitogen-activated protein kinase phosphatase 2 (Mkp2), as a HoxA10 target gene. We analyzed the DUSP4 5′-flank and identified two proximal-promoter cis elements that are activated by HoxA10. We find that DUSP4 transcription and Mkp2 expression decrease during normal myelopoiesis. However, this down-regulation is impaired in myeloid cells overexpressing HoxA10. In hematopoietic cells, c-Jun N-terminal kinases (Jnk) are the preferred substrates for Mkp2. Therefore, Mkp2 inhibits apoptosis by dephosphorylating (inactivating) Jnk. Consistent with this, HoxA10 overexpression decreases apoptosis in differentiating myeloid cells. Therefore, our studies identify a mechanism by which overexpressed HoxA10 contributes to inappropriate cell survival during myelopoiesis.

HoxA10 is a homeodomain transcription factor that is frequently overexpressed in human acute myeloid leukemia. In murine bone marrow transplantation studies, HoxA10 overexpression induces a myeloproliferative disorder with accumulation of mature phagocytes in the peripheral blood and tissues. Over time, differentiation block develops in these animals, resulting in acute myeloid leukemia. In immature myeloid cells, HoxA10 represses transcription of some genes that confer the mature phagocyte phenotype. Therefore, overexpressed HoxA10 blocks differentiation by repressing myeloid-specific gene transcription in differentiating myeloid cells. In contrast, target genes involved in myeloproliferation due to HoxA10 overexpression have not been identified. To identify such genes, we screened a CpG island microarray with HoxA10 co-immunoprecipitating chromatin. We identified the DUSP4 gene, which encodes mitogen-activated protein kinase phosphatase 2 (Mkp2), as a HoxA10 target gene. We analyzed the DUSP4 5-flank and identified two proximal-promoter cis elements that are activated by HoxA10. We find that DUSP4 transcription and Mkp2 expression decrease during normal myelopoiesis. However, this down-regulation is impaired in myeloid cells overexpressing HoxA10. In hematopoietic cells, c-Jun N-terminal kinases (Jnk) are the preferred substrates for Mkp2. Therefore, Mkp2 inhibits apoptosis by dephosphorylating (inactivating) Jnk. Consistent with this, HoxA10 overexpression decreases apoptosis in differentiating myeloid cells. Therefore, our studies identify a mechanism by which overexpressed HoxA10 contributes to inappropriate cell survival during myelopoiesis.
The 39 human and murine HOX genes are arranged in four paralog groups on four different chromosomes These genes encode homeodomain transcription factors that are highly conserved from Drosophila to humans. During embryogenesis, HOX gene transcription is activated 3Ј to 5Ј with the 3Ј-most genes regulating cephlad development and the 5Ј-most genes regulating caudal development (1). Transcription of the HOX genes is also tightly regulated during definitive hematopoiesis.
Genes 3Ј in the locus (HOX1 to -4) are actively transcribed in hematopoietic stem cells, and more 5Ј genes (HOX5 to -11) are actively transcribed in lineage-committed progenitors (2). Therefore, HOX1 to -4 genes are transcribed in CD34 ϩ CD38 Ϫ cells. In contrast, HOX5 to -11 genes (also referred to as ABD HOX genes) are transcribed throughout the CD34 ϩ compartment and down-regulated in CD34 Ϫ cells. In poor prognosis AML, the normal decrease in HoxA7, A9, and A10 expression in CD34 Ϫ cells does not occur (3,4). Consistent with this, mice transplanted with bone marrow overexpressing either HoxA9 or HoxA10 rapidly develop a myeloproliferative disorder, which evolves to acute myeloid leukemia over time. This myeloproliferative disorder is characterized by an increase in mature phagocytes in the peripheral blood and tissues. Murine studies also indicate that forced overexpression of HoxA9 or 10 has a common effect of expanding the bone marrow myeloid progenitor pool (5)(6)(7).
In contrast, the effect of HoxA9 and HoxA10 on myeloid differentiation is nonredundant. HoxA9 expression appears to be necessary for acquisition of the myeloid phenotype in normal and leukemic cells (5). In contrast, HoxA10 overexpression is associated with differentiation block in animal studies and in vitro (7). Despite the many elegant studies of the role of Abd HoxA proteins in myelopoiesis and leukemogenesis, relatively few genuine target genes have been identified for these homeodomain transcription factors. To address this issue, initial studies were performed to derive consensus sequences for HoxA protein-DNA binding. These studies identified consensus sequences for DNA binding of Hox-Pbx heterodimers (8). Pbx proteins are homeodomain transcription factors that increase the affinity of Hox proteins for DNA-binding sites. Based on the results of these studies, HoxA9 and HoxA10 target genes were identified that encode proteins involved in phagocyte functions. For example, a HoxA9/Pbx1 heterodimer activates homologous cis elements in the genes encoding the respiratory burst oxidase proteins gp91 phox and p67 phox in differentiating myeloid cells (9). Additional studies determined that transcription of these two oxidase genes is repressed by interaction of a HoxA10/Pbx1 heterodimer with the same cis elements in undifferentiated myeloid cells (10 -12). Interestingly, a HoxA10/Pbx2 heterodimer activates transcription of the gene encoding ␤3 integrin in endometrial cells (13). Since ␤3 integrin is expressed in myeloid cells, these results suggest * 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. 1  the possibility that HoxA10 is a multifunction transcription factor during myelopoiesis. In contrast, relatively few target genes that encode proteins involved in regulating cell proliferation or survival have been identified for any Abd HoxA protein. In hematopoietic stem cells and myeloid progenitor cells, these functions are regulated by a variety of different mechanisms. To identify HoxA10 target genes involved in myeloproliferation in differentiating myeloid cells, we coupled chromatin co-immunoprecipitation with high throughput screening of a CpG island microarray (14). By this approach, we identified DUSP4 as a potential HoxA10 target gene in undifferentiated myeloid cells. This gene encodes Dusp4 (dual specific phosphatase 4), also known as Mkp2 (mitogen-activated protein kinase phosphatase 2).
Mkp1 and -2 are structurally related proteins that are expressed in hematopoietic cells but have different substrate specificities. In vivo, Mkp2 preferentially dephosphorylates (and therefore inactivates) c-Jun N-terminal kinases 1 and 2 (Jnk1 and -2) but not p38 or extracellular signal-regulated kinase mitogen-activated protein kinases (15). Jnk1 is activated (phosphorylated) in response to genotoxic stress or hematopoietic cytokines and mediates apoptosis in differentiating myeloid cells (16,17). Therefore, if HoxA10 activates DUSP4 transcription, HoxA10 overexpression in myeloid malignancy might increase cell survival during cytokine-induced myelopoiesis. In this study, we investigate the role of HoxA10 in DUSP4 transcription and in regulation of apoptosis.

Plasmids and PCR Genomic Cloning
DUSP4 5Ј-Flank Reporter Constructs-Genomic clones for the DUSP4 5Ј-flank were obtained by PCR using genomic DNA isolated from U937 cells. The 3Ј-primer for each of the reactions encompassed ϩ30 to 0 relative to the ATG start codon. A series of 5Ј-primers were generated to amplify 5Ј-flank sequences, including Ϫ2933 to ϩ30 bp, Ϫ1772 to ϩ30 bp, Ϫ1722 to ϩ30 bp, Ϫ1205 to ϩ30 bp, and Ϫ1142 to ϩ30 bp. These genomic sequences were subcloned into the pCATE reporter vector (Stratagene, Cedar Creek, TX). Clones were completely sequenced on both strands, and the sequence was compared with the NCBI Human Genome Data Base to ensure that no mutations had been introduced.
Artificial Promoter Constructs-Artificial promoter/reporter constructs were generated as previously described (10,12) in the minimal promoter/reporter vector, p-TATACAT (18) (obtained from Dr. A. Kraft (Hollings Cancer Center at the Medical University of South Carolina, Charleston, SC)). Constructs were generated with four copies (in the forward direction) of the Ϫ1755 to Ϫ1717 bp sequence (referred to as "A") or the Ϫ1174 to Ϫ1136 bp sequence (referred to as "B") from the DUSP4 promoter (p-ADUSP4-TATACAT and BDUSP4-TATACAT, respectively).
cDNA Sequences-The cDNA for human HoxA10 was obtained from C. Largman (University of California, San Francisco, CA) (19). This cDNA sequence represents the major transcript in mammalian hematopoietic cells, encoding a 393amino acid, 55-kDa protein (20). Wild type HoxA10 cDNA sequence was subcloned into the pSR␣ vector for expression in mammalian cells (per the manufacturer's instructions (Stratagene)). The human Mkp2 cDNA was obtained from Clontech (Mountain View, CA) and subcloned into the pSR␣ vector for expression in mammalian cells.
Plasmids for Short Hairpin RNA (shRNA) 2 Expression-Plasmids were generated to express shRNA to human Mkp2 using the pLKO.1 vector (kindly provided by Dr. K. Rundell, Northwestern University, Chicago, IL). Oligonucleotide sequences for Mkp2-specific shRNA or scrambled control shRNA were designed with the assistance of the software on the Promega Web site (Promega, Madison, WI).

Oligonucleotides
Oligonucleotides were synthesized by the Core Facility of the Robert H. Lurie Comprehensive Cancer Center at Northwestern University. Double-stranded, synthetic oligonucleotides were generated representing the Ϫ1755 to Ϫ1717 bp "A" sequence from the DUSP4 promoter (5Ј-TAACTCCCTTGC-TCTGTGATTAATTCTCACTAACAAGA-3Ј), the Ϫ1174 to Ϫ1136 bp "B" sequence from the DUSP4 promoter (5Ј-CTAT-AAAACTGATTTAATGGCTTTAGATGAAAATCGATC-3Ј), or the Ϫ94 to Ϫ134 bp sequence of the CYBB promoter (5Ј-ttcagttgaccaatgattattagccaattttctgataaaa-3Ј). In these oligonucleotides, the HoxA10 core is in boldface type, and the Pbx core is in italic type.

Myeloid Cell Line Culture
The human myelomonocytic cell line U937 (22) was obtained from Andrew Kraft. Cells were maintained and differentiated as described (10,12,23). U937 cells were treated with 500 units/ml human recombinant IFN␥ for 24 or 48 h, as indicated (Hoffman-LaRoche).

Chromatin Immunoprecipitation and CpG Island Screening
U937 cells were cultured with or without IFN␥ for 48 h, as described (10,12,23). Chromatin immunoprecipitation and CpG island microarray hybridization were performed as described (14). Briefly, cells were incubated with formaldehyde and lysed, and lysates were sonicated to generate chromatin fragments with an average size of 2.0 kb. Lysates were precipitated with HoxA10 antiserum (Covance) or control preimmune serum. Chromatin was recovered by precipitation, proteins were stripped from the chromatin, and the chromatin was PCR-amplified. Several batches of immunoprecipitated, amplified chromatin were combined for each experiment.
Aliquots of HoxA10-specific and preimmune serum controlprecipitated, amplified chromatin were labeled with Cy3 or Cy5 by the random primer method. Labeled DNA was used to probe a CpG island microarray, as described (14). CpG island microarrays were obtained from the Microarray Center, University Health Network (Ontario Cancer Institute, Ontario, Canada). "Spots" with 3-fold enhancement in HoxA10-specific versus control preimmune serum-precipitated chromatin in three independent hybridization experiments were further considered. Dye swapping experiments were performed as controls for differences in efficiency in incorporation of Cy3 versus Cy5 into DNA. Arrays were scanned using an Agilent microarray scanner (G2565BA, Wilmington, DE), and feature intensity statistics were extracted using GenePix (Molecular Devices, Union City, CA). The GenBank TM accession number from the array was used to search the NCBI human genome data base for adjacent genes.
Specificity of chromatin immunoprecipitation was confirmed by independent chromatin immunoprecipitation with HoxA10-specific antiserum or preimmune serum control. Precipitated chromatin was PCR-amplified with overlapping primer sets that were designed to encompass the proximal 4 kb of 5Ј-flank sequences, based on location of the CpG island (Ϫ1734 to Ϫ2025 bp relative to the ATG start codon). Primer sets were designed to amplify Ϫ2025 to ϩ30 bp or Ϫ2025 to Ϫ3613 bp.

Electrophoretic Mobility Shift Assays (EMSA)
Nuclear extract proteins were prepared by the method of Dignam (24) with protease inhibitors (as described) (10). Oligonucleotides probes were prepared, and EMSA and antibody supershift assays were performed, as described (10,12). For all experiments, at least three independent batches of nuclear proteins were tested in at least two independent experiments. Integrity of the nuclear proteins and equality of protein loading was determined in control EMSA with a classical CCAAT box from the ␣ globin gene promoter. Antiserum to HoxA10 (not cross-reactive with other Hox proteins) was obtained from Covance Research Products (Richmond, CA) and from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA). Antibodies to Pbx1 and Pbx2 (not cross-reactive with each other or other Pbx proteins) were obtained from Santa Cruz Biotechnology. All EMSA were performed several times with at least two different batches of nuclear proteins, and representative data are shown.
Stable Transfectants-Stable U937 transfectants were generated with pSR␣ vectors to overexpress HoxA10, Mkp2, or empty vector control. Stable transfectants were selected in G418, as previously described (10). At least three transfectant pools were tested for each construct. In some experiments, U937 cells were transfected with PLKO.1 vectors to express shRNA for Mkp2 or scrambled shRNA control and selected with puromycin.

Western Blotting
Total cell lysate proteins from U937 cells or U937 stable transfectants (50 g) were separated by SDS-PAGE and transferred to nitrocellulose. Western blots were serially probed with antibodies to various proteins. Antibodies to tubulin, HoxA10, Jnk1, and phospho-Jnk were obtained from Santa Cruz Biotechnology. Each experiment was performed several times, and representative blots are shown.

Quantitative Real Time PCR
In some experiments, RNA was isolated from U937 cells using the Triazol reagent, according to manufacturer's instructions (Invitrogen). RNA was tested by denaturing gel electrophoresis to determine the integrity of the 18 and 28 S ribosomal bands. Primers were designed with the software from Integrated DNA Technologies, and real time PCR was performed using SYBR green according to the "standard curve" method. Results were normalized to 18 S and actin to control for RNA abundance in various samples.
In other experiments, chromatin co-immunoprecipitating from U937 cells with antibody to HoxA10, anti-acetyl-histone 3, or control preimmune serum was analyzed by real time PCR. Primers were designed to amplify ϳ80 bp centered around the A and B HoxA10 DNA-binding consensus sequences using the Integrated DNA Technologies software, as above. Results were normalized to PCR product abundance in nonprecipitated, total chromatin samples. Experiments were performed in triplicate for each of three different immunoprecipitation experiments.

Apoptosis Assays
Annexin V/propidium iodide (Beckman) double staining was used according to the manufacturer's instructions. The cells were washed with ice-cold culture medium, adjusted to a concentration of 1 ϫ 10 6 cells/ml, and incubated with annexin V-fluorescein isothiocyanate solution (2.5 g/ml) and propidium iodide (12.5 g/ml) on ice for 15 min, and cell preparations were analyzed on a BD Biosciences FACScan flow cytometer.

Statistical Analysis
Statistical significance of comparisons between two groups was determined by Student's t test. Statistical significance of comparisons between larger groups was determined using analysis of variance. Statistical calculations were performed using the SigmaPlot and SigmaStat software.

HoxA10 Interacts with the DUSP4 Promoter in Vitro and in
Vivo-HoxA10 target genes were identified by chromatin coimmunoprecipitation from U937 myeloid leukemia cells. This myeloid cell line undergoes differentiation in response to various cytokines, including IFN␥ (22). U937 differentiation is characterized by acquisition of mature phagocyte characteristics, including respiratory burst activity and phagocytosis (10,12,22,25). Differentiation is also characterized by cell cycle arrest within 24 h and programmed cell death over 72 h. Therefore, this cell line provides a reasonable model of the events of myelopoiesis.
For these studies, U937 cells were treated with formaldehyde to generate in vivo DNA-protein cross-links, and cell lysates were immunoprecipitated with antiserum to HoxA10 or control preimmune serum. Prior to immunoprecipitation, the lysates were sonicated to generate 2.0-kb chromatin fragments (14). Precipitated chromatin was labeled with Cy3 or Cy5 and used to hybridize a human CpG island microarray, as described (14). Triplicate array experiments were performed (with independent co-precipitations), and only "spots" with at least 3-fold enhancement in all three were further considered. One of the identified CpG islands localized to chromosome 8 in the 5Ј-flank of the DUSP4 gene ( Fig. 1A; location of the CpG island is Ϫ2025 to Ϫ1734 bp relative to the ATG start codon). This gene encodes dual specific phosphatase 4, also known as Mkp2 (mitogen-activated protein kinase phosphatase 2). This result was of considerable interest, since Mkp2 has been implicated in regulating apoptosis in myeloid cells via inactivation (dephosphorylation) of proapoptotic Jnk proteins (14).
Since co-precipitating DNA was sheared to generate 2.0-kb fragments, the location of the CpG island (ϳϪ2.0 kb) indicated that the HoxA10 binding site(s) should be within the proximal 4.0 kb of the DUSP4 5Ј-flank. Therefore, we performed initial experiments to verify interaction of HoxA10 with the DUSP4 5Ј-flank and to locate the binding site within either the proximal Ϫ2.0 kb of the 5Ј-flank or between Ϫ2.0 and Ϫ4.0 kb. To determine the impact of differentiation stage on HoxA10-binding, chromatin was co-immunoprecipitated from U937 cells with and without 48 h of IFN␥ differentiation. Precipitating chromatin was amplified by PCR, separated by acrylamide gel electrophoresis, and visualized by ethidium bromide staining. A, identification of a CpG island 2 kb from the DUSP4 transcription start site by chromatin immunoprecipitation coupled with screening a CpG island microarray. Chromatin specifically immunoprecipitating from U937 cells with an antibody to HoxA10 was compared with chromatin co-precipitating with preimmune serum. Chromatin was labeled with a fluorescent tag and used to probe a CpG island microarray. One of the identified CpG islands was located Ϫ1734 and Ϫ2025 bp 5Ј to the ATG translation start site of the gene that encodes Mkp2 (the DUSP4-gene). This CpG island includes a consensus sequence for HoxA10/Pbx binding (underlined). This putative consensus sequence is referred to as DUSP4 sequence A. B, HoxA10 interacts with the proximal DUSP4 5Ј-flank by chromatin immunoprecipitation. To verify the CpG island microarray results, independent chromatin immunoprecipitation experiments were performed with U937 cells with and without 48 h of IFN␥ differentiation. Lysates were immunoprecipitated with an antibody to HoxA10 or control preimmune serum. Co-precipitating chromatin was PCR-amplified with primers encompassing the 5Ј-flank of the DUSP4 gene from ϩ30 bp to Ϫ2.025 kb relative to the ATG translation start codon. PCR products were sized on an acrylamide gel with one-tenth of input chromatin as a positive control. These studies demonstrate HoxA10 interaction with the proximal DUSP4 5Ј-flank in U937 cells. These results also indicate that this interaction decreases with IFN␥ differentiation of this cell line. In contrast, PCR with primers encompassing the sequence from Ϫ2.025 to Ϫ3.613 kb relative to the ATG start codon did not demonstrate HoxA10 interaction (not shown). C, schematic representation of the DUSP4 5Ј-flank. Schematic representation of the DUSP4 5Ј-flank indicating the location of the CpG island, the A and B sequences (which are homologous to the derived Hox/Pbx DNA-binding consensus), and the 5Ј-flank truncations used to generate reporter constructs. The A and B sequences are indicated below, lined up with the derived consensus sequence for HoxA10/Pbx binding. JUNE 1, 2007 • VOLUME 282 • NUMBER 22

JOURNAL OF BIOLOGICAL CHEMISTRY 16167
Two primer sets were used to amplify fragments from Ϫ2025 bp to ϩ30 bp (relative to the ATG) or from Ϫ2025 to Ϫ3613 bp. For these experiments, nonprecipitated chromatin was a positive control, and chromatin co-precipitating with preimmune serum was a negative control. We found specific amplification of HoxA10 co-precipitating chromatin with the proximal primer set (Fig. 1B). Although not quantitative, these studies also suggested that HoxA10 interaction with the DUSP4 5Ј-flank decreased during U937 differentiation. In contrast, HoxA10 antibody did not co-precipitate chromatin with the more distal 5Ј-flank sequence (not shown).
The results of these chromatin immunoprecipitation studies suggest that HoxA10 interacts with a binding site within the proximal 2.0 kb of DUSP4 5Ј-flank in vivo. Therefore, we analyzed this 2.0-kb sequence to determine if it contained any HoxA10 DNA-binding consensus sequences. This analysis resulted in identification of two sequences that conform to the derived consensus for DNA binding of HoxA10-Pbx heterodimers. These sequences were found at Ϫ1731 to Ϫ1738 bp (referred to as "A") and Ϫ1158 to Ϫ1166 bp (referred to as "B") relative to the ATG start codon in the DUSP4 5Ј-flank (Fig. 1C).
These results provided initial evidence in support of our hypothesis that DUSP4 is a HoxA10 target gene. Therefore, we performed additional experiments to determine the impact on Mkp2 expression of myeloid differentiation and HoxA10 overexpression.

Mkp2 Expression Is Decreased by Differentiation and Increased by HoxA10 Overexpression in Myeloid
Cells-Mkp2 is expressed in myeloid cells, but differentiation stagespecific regulation of Mkp2 expression has not been previously investigated (14). Differentiation of U937 cells is associated with increased Jnk activation (phosphorylation) and Jnk-induced apoptosis. Since Mkp2 dephosphorylates Jnk, one might anticipate a decrease in Mkp2 expression as differentiation proceeds. Therefore, we initially investigated the impact of IFN␥ differentiation on Mkp2 mRNA abundance in U937 cells. For these studies, expression was analyzed by real time PCR using RNA isolated from U937 cells over a 72-h period of differentiation. We found that Mkp2 mRNA abundance in U937 cells decreased significantly after 24 h of IFN␥ treatment but did not decrease further between 24 and 72 h post-IFN␥ (F ϭ 0.40, p ϭ 0.68, n ϭ 4) ( Fig. 2A).
Based on these results, we investigated the impact of U937 differentiation and HoxA10 overexpression on Mkp2 protein abundance. For these studies, stable U937 transfectant pools were generated with a vector to overexpress HoxA10 or empty control vector. Transfectants were analyzed with and without IFN␥ differentiation. Three independent transfectant pools were selected, and the results of representative experiments are presented. Western blots of total cell lysates were serially probed with antibodies to Mkp2, HoxA10 (to verify overexpression), Jnk and phospho-Jnk (as a downstream Mkp2 target), and tubulin (as a loading control).
In control transfectants, we found that IFN␥ treatment decreased Mkp2 protein abundance and increased Jnk phosphorylation, consistent with our hypothesis (Fig. 2B). In contrast, Mkp2 protein expression was relatively preserved during differentiation of HoxA10-overexpressing U937 transfectants, and Jnk phosphorylation did not increase. These results suggest the possibility that HoxA10 overexpression decreases apoptosis in differentiating myeloid cells by impairing Jnk activation. Therefore, we performed additional studies to investigate the functional connection between HoxA10 overexpression, Mkp2 expression, and apoptosis in these cells.
HoxA10 Overexpression Decreases Mkp2-dependent Apoptosis in Differentiating Myeloid Cells-Based on the studies above, we hypothesized that HoxA10 overexpression in U937 cells would decrease apoptosis during IFN␥ differentiation in an Mkp2-dependent manner. This would provide physiologic relevance to our identification of DUSP4 as a potential HoxA10 target gene. To investigate the impact of HoxA10-induced Cells were analyzed with or without 48 h of IFN␥ differentiation. Total cell lysates were separated by SDS-PAGE, and Western blots (WB) were serially probed with antibodies to Mkp2, HoxA10 (to demonstrate overexpression), Jnk1, p-Jnk (to indicate activation), and tubulin (as a loading control). Mkp2 protein abundance decreases during IFN␥ differentiation of control U937 cells. In comparison, Mkp2 protein abundance is relatively increased in HoxA10-overexpressing U937 transfectants after differentiation. Phospho-Jnk1, but not total Jnk1, increases during differentiation of control U937 transfectants. In contrast, Jnk1 phosphorylation does not similarly increase in HoxA10-overexpressing U937 stable transfectants in response to IFN␥ differentiation. HoxA10 is overexpressed in U937 stable transfectants with the HoxA10 expression vector, consistent with expectations.
Mkp2 expression on apoptosis, we first studied the effect of overexpressing these proteins in U937 cells. Stable transfectants pools of U937 cells were generated with vectors to overexpress HoxA10, Mkp2, or empty control vector. Apoptosis of stable transfectants, with and without 48 h of IFN␥ treatment, was determined by annexin V staining.
In undifferentiated U937 transfectants, we found that overexpression of HoxA10 or Mkp2 did not significantly alter the percentage of apoptotic cells in comparison with control vector transfectants (p ϭ 0.48, F ϭ 0.83, n ϭ 3) (Fig. 3A). Consistent with previous studies, the percentage of apoptotic cells in control U937 transfectants increased significantly during IFN␥ differentiation (p Ͻ 0.05, n ϭ 3). In contrast, IFN␥ treatment did not significantly increase the percentage of apoptotic cells in U937 transfectants overexpressing HoxA10 or Mkp2 (p Ͼ 0.7, n ϭ 3). Therefore, there were significantly more apoptotic cells in IFN␥-treated control transfectants in comparison with HoxA10-or Mkp2-overexpressing cells (p ϭ 0.02, F ϭ 7.00, n ϭ 3). Overexpression of HoxA10 or Mkp2 in these stable transfectant pools was verified by Western blots of total cell lysates (representative blots are shown in Figs. 2B and 3B).
These results suggest that overexpression of either HoxA10 or Mkp2 protects differentiating myeloid cells from apoptosis. Therefore, we were interested in determining if the antiapoptotic effect of HoxA10 overexpression was dependent upon differentiation stage-inappropriate Mkp2 expression. To investigate this, U937 stable transfectant pools were generated with vectors to co-overexpress HoxA10 and an Mkp2-specific shRNA. Apoptosis in these cells was compared with control stable transfectants with empty expression vector plus vector to Apoptosis was determined by annexin V staining with and without 48 h of IFN␥ differentiation. IFN␥ differentiation increases apoptosis of control U937 cells, consistent with previous studies. Neither HoxA10 nor Mkp2 overexpression increases apoptosis in undifferentiated transfectants. In contrast, overexpression of either HoxA10 or Mkp2 blocks IFN␥-induced apoptosis. *, apoptosis that is significantly different in comparison with the other conditions. B, overexpression of Mkp2 in U937 cell stable transfectants. U937 cells were stably transfected with a vector to overexpress Mkp2 or empty control vector. Total cell lysates were separated by SDS-PAGE, and Western blots were serially probed with antibodies to Mkp2 and tubulin (as a loading control). Mkp2 is overexpressed in U937 stable transfectants with the Mkp2 expression vector, consistent with expectations. C, protection from apoptosis by HoxA10 overexpression in differentiating U937 cells is partly inhibited by knockdown of Mkp2. U937 cells were stably transfected with various combinations of a vector to overexpress HoxA10 or empty vector control and a vector to express a Mkp2-specific shRNA or scrambled control shRNA. Apoptosis was determined with and without 48 h of IFN␥ differentiation. Apoptosis increased during IFN␥ differentiation in U937 stable transfectants with control expression vector plus scrambled shRNA vector, consistent with the results above. Apoptosis was not significantly different from control in transfectants with control expression vector plus Mkp2-specific shRNA vector. IFN␥-induced apoptosis is decreased in U937 stable transfectants with HoxA10 expression vector plus scrambled shRNA control vector, similar to transfectants overexpressing HoxA10 alone. In contrast, co-expression of Mkp2-specific shRNA increases differentiation-induced apoptosis in HoxA10overexpressing cells. *, **, and #, a significant difference in apoptosis in stable transfectants with and without IFN␥. D, Mkp2 knockdown in U937 cell stable transfectants. U937 stable transfectants with a vector to express Mkp2-specific shRNA or scrambled shRNA were analyzed for inhibition of Mkp2 expression. Total cell lysates were separated by SDS-PAGE, and Western blots (WB) were serially probed with antibodies to Mkp2 and tubulin (as a loading control). Mkp2 expression is decreased in stable transfectants with the Mkp2-specific shRNA expression vector, as expected. express scrambled shRNA, a HoxA10 expression vector plus a vector to express scrambled shRNA, and an empty expression vector plus a vector to express Mkp2-specific shRNA. Apoptosis was determined as above (Fig. 3C).
In these studies, we found that the percentage of apoptotic cells significantly increased during IFN␥-induced differentiation of all of these stable transfectants with the exception of transfectants with HoxA10 expression vector plus control shRNA vector (Fig. 3C). Therefore, we found that expressing Mkp2-shRNA decreased the antiapoptotic effect of overexpressing HoxA10 in differentiating U937 cells. These results suggest that the antiapoptotic effect of HoxA10 overexpression in differentiating myeloid cells is at least partly abrogated by blocking Mkp2 expression. Efficient decrease in Mkp2 expression in stable transfectants with an Mkp2-specific shRNA vector was demonstrated by Western blot (Fig. 3D). These results suggest the functional significance of the impact of HoxA10overexpression on Mkp2 expression. Therefore, we analyzed the DUSP4 5Ј-flank to identify cis element(s) activated by HoxA10.
HoxA10 Overexpression Activates the DUSP4 Promoter in Myeloid Cells-To investigate the functional significance of this interaction, we generated a series of reporter constructs with various sequences from the DUSP4 5Ј-flank. DUSP4 clones for these studies were obtained from U937 genomic DNA by PCR. To facilitate generating the potentially most informative set of constructs, the location of the A and B HoxA10/Pbx DNA-binding consensus sequences were considered in designing these experiments. Therefore, we generated CAT reporter constructs with 2.93, 1.77, 1.72, 1.20, and 1.14 kb of DUSP4 5Ј-flank (relative to the ATG start) (see Fig. 1C). Based on the location of identified consensus sequences, the 2.93-and 1.77-kb constructs include both potential HoxA10binding sites. The 1.72-and 1.20-kb constructs include only the proximal (or B) HoxA10 binding site, and the 1.14-kb construct does not contain a HoxA10-binding consensus sequence.
U937 cells were co-transfected with these reporter constructs and a vector to overexpress HoxA10 or empty expression vector control. Reporter gene assays were performed with and without IFN␥ differentiation (Fig. 4A). In these studies, we found that IFN␥ differentiation significantly decreased reporter expression from all of the constructs with DUSP4 5Ј-flank sequences except for the 1.14-kb construct. These results suggest that cis elements responsible for the differentiation stage-specific decrease in Mkp2 expression are found between 1.14 and 2.93 kb of the DUSP4 5Ј-flank. Additionally, we found that HoxA10 overexpression significantly increased reporter activity from all of these constructs, except for the 1.14-kb DUSP4 5Ј-flank construct, with and without IFN␥ treatment. These results suggest that HoxA10-binding cis elements are located between 1.14 and 2.93 kb of the DUSP4 5Ј-flank, consistent with the location of the two HoxA10/Pbx DNA-binding consensus sequences. We found that the amount of HoxA10-induced reporter activity was not significantly different in transfectants with the 2.93versus 1.77-kb construct, with and without IFN␥ differentiation (p Ͼ 0.8, n ϭ 4). Therefore, we compared HoxA10-inducible reporter activity from the 1.77-kb DUSP4 construct, which contains two potential Hox/Pbx binding sites (the A and B sequences), with the 1.72-kb DUSP4 construct, in which the distal consensus sequence has been eliminated (i.e. truncated to eliminate the A sequence). We found that HoxA10 overexpression induced significantly less reporter activity from the 1.72-kb construct in comparison with the 1.77-kb construct (p Ͻ 0.02, n ϭ 4). This difference persisted with differentiation of the transfectants.
We also compared HoxA10-induced reporter activity from the 1.72-and 1.20-kb constructs, with and without differentiation. Both of these constructs include the proximal B consensus sequence but not the distal A sequence. We found that the amount of HoxA10-induced reporter activity of these constructs was not significantly different (p Ͼ 0.7, n ϭ 4). Therefore, we compared HoxA10-induced reporter activity of the 1.20-kb DUSP4 5Ј-flank construct, which contains the proximal HoxA10-binding consensus (the B sequence), with the 1.14-kb construct, which does not include a HoxA10-binding consensus (i.e. truncated to eliminate the B sequence). We found that HoxA10 overexpression did not induce a significant increase in reporter expression of the 1.14-kb construct, with and without IFN␥ differentiation.
These results suggest that HoxA10-inducible reporter activity decreases with deletion of the A sequence and is abolished by deletion of the B sequence. If the A and B sequences represent cis elements activated by HoxA10, HoxA10 overexpression would be expected to induce more reporter activity from constructs with both cis elements in comparison with constructs with only the more proximal B sequence. To test this hypothesis, we compared HoxA10-induced activation of constructs with both the A and B sequences (2.93-and 1.77-kb constructs) with activation of constructs with only the proximal B sequence (1.72-and 1.20-kb constructs). We found that overexpression of HoxA10 increased reporter expression from the 2.93-and 1.77-kb constructs by 79.5 Ϯ 4.7%. In contrast, overexpressed HoxA10 increased reporter expression from the 1.72-and 1.20-kb constructs by 44.8% Ϯ 3.1%. Therefore, overexpressed HoxA10 induced significantly more expression from constructs with both the A and B cis elements in comparison with constructs with only the B cis element (about 2-fold more, p Ͻ 0.02, n ϭ 8).
In contrast to these studies, neither HoxA10 overexpression nor IFN␥ differentiation significantly altered expression from the empty, control CAT reporter vector.
Overexpressed HoxA10 Activates Two Cis Elements in the DUSP4 Promoter-These results suggest that HoxA10 activates two different cis elements in the DUSP4 5Ј-flank. Therefore, we investigated whether either the A or B DUSP4 sequence functions as a HoxA10-activated cis element. For these studies, we generated constructs with four copies of the DUSP4 A (Ϫ1755 to Ϫ1717 bp) or B (Ϫ1174 to Ϫ1136 bp) sequence linked to a minimal promoter and CAT reporter (referred to as ADUSP4-TATACAT and BDUSP4-TATACAT, respectively) (Fig. 1C). These constructs or empty control pTATACAT vector were co-transfected into U937 cells with a vector to overexpress HoxA10 or empty expression vector. Reporter gene studies were performed with and without IFN␥ differentiation of the transfectants (Fig. 4B).
We found that the ADUSP4-TATACAT construct exhibited ϳ3-fold more reporter activity in comparison with control FIGURE 4. HoxA10 overexpression and myeloid differentiation influence activity of the DUSP4 promoter. A, HoxA10 overexpression increases activity of the DUSP4 promoter in U937 transfectants, with and without IFN␥ differentiation, and this activation depends on two consensus sequences for HoxA10-DNA binding. U937 cells were co-transfected with a reporter construct with Ϫ2.93, Ϫ1.77, Ϫ1.72, Ϫ1.20, or Ϫ1.14 kb of the DUSP4 5Ј-flank relative to the ATG start codon or empty reporter vector control and a vector to express HoxA10 or empty expression vector control. Reporter gene assays were performed with and without IFN␥ differentiation. Reporter activity of the 2.93-and 1.77-kb DUSP4 promoter constructs were not significantly different, were equivalently decreased by IFN␥ differentiation, and equivalently increased by HoxA10 overexpression. Activity of the 2.93-and 1.77-kb constructs (which contain both the A and B HoxA10-binding consensus sequences) was significantly greater than activity of the 1.72-and 1.2-kb constructs (which contain only the B HoxA10-binding consensus sequence), with and without IFN␥ differentiation. However, reporter activities of the 1.72-and 1.2-kb constructs were not significantly different, with and without IFN␥, with and without HoxA10-overexpression. In contrast, HoxA10 overexpression does not influence reporter expression from the 1.14-kb DUSP4 construct (which does not contain a HoxA10 DNA-binding consensus sequence). Neither HoxA10 overexpression nor IFN␥ treatment significantly influences reporter expression from the empty control reporter vector. * and **, statistically significant differences in reporter expression of constructs with versus without truncation of the A HoxA10 binding consensus sequence in HoxA10-overexpressing transfectants. # and ##, statistically significant differences in reporter activity of constructs with versus without truncation of the B HoxA10-binding consensus sequence in HoxA10-overexpressing transfectants. B, HoxA10 activates an artificial promoter construct with multiple copies of the A or B sequence from the DUSP4 5Ј-flank linked to a minimal promoter in U937 transfectants with and without IFN␥ differentiation. U937 cells were cotransfected with a minimal promoter-reporter vector with four copies of the DUSP4 A or B sequence (ADUSP4-TATACAT or BDUSP4-TATACAT) or minimal promoter-reporter control vector (p-TATACAT) and a vector to express HoxA10 or empty expression vector control. Reporter activity was determined with and without 48 h of IFN␥ treatment of the transfectants. Both the DUSP4 A and B sequence-containing constructs have significantly more reporter expression in comparison with control p-TATACAT in U937 transfectants. Activity of both the ADUSP4-TATACAT and BDUSP4-TATACAT vectors was significantly decreased by IFN␥ treatment of the transfectants. HoxA10 overexpression significantly increased reporter activity of both the ADUSP4-TATACAT and BDUSP4-TATACAT constructs in U937 transfectants, with and without IFN␥ differentiation. In contrast, IFN␥ treatment and HoxA10 overexpression did not significantly influence reporter activity from the p-TATACAT control vector. * and **, statistically significant difference in ADUSP4-TATACAT reporter activity with versus without HoxA10 overexpression. # and ##, statistically significant difference in BDUSP4-TATACAT reporter activity with versus without HoxA10 overexpression.
pTATACAT vector in undifferentiated U937 transfectants, suggesting that this sequence functions as a cis element (p ϭ 0.002, n ϭ 4). However, we found that the A sequence was a significantly less efficient positive cis element in IFN␥-treated transfectants. We also found that HoxA10 overexpression significantly increased reporter expression from the A oligonucleotide-containing reporter construct in both untreated and IFN␥-differentiated U937 transfectants. Indeed, reporter expression from ADUSP4-TATACAT was not significantly different in IFN␥-treated transfectants overexpressing HoxA10 from that in undifferentiated transfectants without HoxA10 overexpression (p ϭ 0.63, n ϭ 4). In other words, HoxA10 overexpression prevented the normal IFN␥-induced decrease in activity of the DUSP4 A cis element.
We performed similar experiments to determine if the more proximal HoxA10-binding DUSP4 sequence was a functional cis element, activated by HoxA10. In these studies, we found that inclusion of the DUSP4 B sequence in the artificial promoter vector significantly also increased reporter activity, suggesting that this sequence also functions as a cis element (p Յ 0.01, n ϭ 4). Similar to the DUSP4 A cis element, activity of the B cis element was significantly less in IFN␥-treated U937 transfectants in comparison with undifferentiated transfectants. We also determined the impact of overexpressing HoxA10 on the DUSP4 B cis element. In these studies, we found that HoxA10 overexpression significantly increased reporter activity from the DUSP4 B cis element-containing construct, with and without IFN␥ differentiation. As with the DUSP4 A cis element, there was no significant difference in reporter activity from the BDUSP4-TATACAT construct in undifferentiated transfectants without HoxA10 overexpression in comparison with differentiated transfectants overexpressing HoxA10 (p ϭ 0.26, n ϭ 4). These results further suggest that HoxA10 overexpression can reverse the impact of myeloid differentiation on Mkp2 expression, consistent with the studies above.
We also determined whether there were differences between the DUSP4 A and B cis elements in terms of efficiency of activation by overexpressed HoxA10. In undifferentiated, HoxA10-overexpressing U937 transfectants, we found that reporter activity of the ADUSP4-TATACAT construct was significantly greater than the BDUSP4-TATACAT construct (p ϭ 0.02, n ϭ 4). However, we found that reporter activity of the ADUSP4-TATACAT and BDUSP4-TATACAT constructs was not significantly different in IFN␥-treated, HoxA10-overexpressing U937 transfectants (p ϭ 0.27, n ϭ 4). In control experiments, neither HoxA10 overexpression nor IFN␥ differentiation significantly altered expression of pTATACAT vector.
HoxA10 Interacts with Two Cis Elements in the DUSP4 Promoter, in Vitro-These studies suggest that HoxA10 overexpression increases DUSP4 promoter activity in undifferentiated and differentiated U937 cells, and this effect depends on the A and B cis elements. We were interested in determining if either of these cis elements interact with HoxA10 in vitro and the impact of differentiation on this interaction. Therefore, we performed EMSA to determine whether HoxA10 binds either of these DNA sequences in vitro. First, we investigated whether oligonucleotide probes representing either the A or B DUSP4 sequence interact with a specific protein complex in vitro. For these studies, EMSA were performed with radiolabeled, double-stranded oligonucleotide probes representing the A or B sequences and nuclear proteins isolated from U937 cells with or without IFN␥ differentiation. At least three different batches of nuclear proteins were used for each of these experiments, and representative experiments are shown. For these studies, the same amount of nuclear protein was used in all of these binding assays, and the specific activities of these two probes were the same. Equality of protein loading was determined in control EMSA with probes representing the CCAAT box from the ␣-globin promoter, which binds the classical CCAAT factor CP1 (not shown) (10).
We found that these DUSP4 promoter sequence probes interact with a protein complex of similar mobility (Fig. 5A). We also found that binding of both of these complexes was decreased in assays with nuclear proteins from IFN␥-treated cells in comparison with untreated cells. Our results also suggest that probe A may have greater affinity for protein complex binding in comparison with probe B.
We further investigated these protein complexes for crossreactive binding specificities. In these studies, EMSA was performed with the A or B probe; nuclear proteins from U937 cells; and excess double-stranded oligonucleotide competitors representing the A or B sequence, the CYBB HoxA10-binding cis element, or irrelevant control oligonucleotide (Fig. 5B). In these studies, we found that the low mobility complexes that bound in vitro to these two probes have cross-competitive binding specificities with each other and with the CYBB cis element. These studies also suggest that oligonucleotide B has a lower affinity for binding the protein complex in comparison with probe A.
As discussed above, the A and B cis elements contain sequences homologous to the derived consensus for HoxA10 binding to DNA as a heterodimer with Pbx, a frequent binding partner. Therefore, we next determined whether protein complexes interacting with the A and B DUSP4 oligonucleotide probes were cross-immunoreactive with HoxA10 or Pbx proteins. In these studies, EMSA was performed with the A or B oligonucleotide probe; U937 nuclear proteins; and antibodies to HoxA10, Pbx1, Pbx2, or irrelevant control antibody (Fig. 5C). We found that the low mobility protein complexes interacting with the A and B probes were cross-immunoreactive with HoxA10 and Pbx2 but not Pbx1 or irrelevant antibody. The B probe also bound a higher mobility complex, which was crossimmunoreactive with HoxA10 or Pbx2 antibody. This latter complex may represent HoxA10 or Pbx2 binding to the B sequence as a monomer.
HoxA10 Interacts with Two Cis Elements in the DUSP4 Promoter, in Vivo-Based on these results, we investigated in vivo HoxA10 binding to the A and B cis elements in the DUSP4 promoter and the effect of IFN␥ differentiation on this binding. For these studies, chromatin co-immunoprecipitation was performed with a HoxA10 antibody or preimmune serum control using lysates of U937 cells with and without IFN␥ differentiation (as in Fig. 1B). Chromatin was analyzed by quantitative real time PCR with primers flanking the A or B cis elements (Fig.  6A). Primer sets were designed to amplify an approximately 80-bp DUSP4 5Ј-flank sequence centered around either the A or B consensus sequence.
In these studies, we found that endogenous HoxA10 binds to both of these cis elements in vivo in undifferentiated U937 cells. However, there was significantly more HoxA10 bound to the region of the A cis element, with and without IFN␥ differentiation (p Ͻ 0.004, n ϭ 12). Consistent with our in vitro studies, we also found that in vivo HoxA10 binding to the A and B cis elements decreased significantly during IFN␥ differentiation (p Ͻ 0.0001, n ϭ 12).
For a some previously described Hox target genes, Hox-containing protein complexes recruit transcriptional co-activators to the cis element. Since such transcriptional co-activators have histone acetyltransferase activity, this results in a local increase in acetylation of DNA-bound histones. Histone acetylation increases accessibility of the transcription start site, thereby favoring gene transcription. To correlate HoxA10-DNA binding with transcriptional activation, we also determined the effect of differentiation on abundance of acetylated histones in the region of the A and B cis elements. For these studies, chromatin was co-immunoprecipitated with an antiacetyl-histone 3 (H3) antibody or control antibody and analyzed by real time PCR with the A or B primers (Fig. 6B).
We found that significantly more acetyl-H3 was associated with the regions of the A and B cis elements in undifferentiated U937 cells in comparison with after differentiation (p Ͻ 0.001, n ϭ 12). Additionally, we found significantly more acetyl-H3 in the region of the A cis element in comparison with the region of the B cis element (p Ͻ 0.04, n ϭ 12). These results are consistent with the results of in vivo HoxA10 binding to these two cis elements in differentiating U937 cells.

DISCUSSION
In these studies, we identify DUSP4, the gene encoding Mkp2, as a HoxA10 target gene. We find that Mkp2 expression decreases during myeloid differentiation and that this decrease is at least partly due to decreased DUSP4 transcription. Also, we find that HoxA10 overexpression prevents down-regulation of Mkp2 expression and DUSP4 transcription during myelopoiesis. The downstream consequence of sustained Mkp2 expression is impaired apoptosis during differentiation of HoxA10-overexpressing myeloid cells. Therefore, these studies identify a novel HoxA10 target gene. Additionally, these studies provide a mechanism by which overexpressed HoxA10 induces resistance to apoptosis during myelopoiesis. These results are of potential significance to the phenotype of HoxA10-overexpressing malignant myeloid cells.
We demonstrate that HoxA10 activates DUSP4 transcription by interacting with at least two cis elements in the 5Ј-flank. These cis elements were identified by homology to a derived FIGURE 5. HoxA10 binds to the A and B DUSP4 sequences in vitro. A, two oligonucleotides representing Hox DNA-binding consensus sequences in the proximal DUSP4 5Ј-flank (A and B) interact in vitro with protein complexes of the same mobility. EMSA was performed with nuclear proteins from U937 cells (1 g) and oligonucleotide probes representing either the A or B sequence from the DUSP4 promoter. An equivalent amount of probe of the same specific activity was used in each of these binding assays to permit comparison between the two different probes. Both the A and B probes bind a low mobility complex that decreases in intensity in EMSA with nuclear proteins from IFN␥-treated U937 cells. This complex appears to bind with greater affinity to the A probe than to the B probe. The low mobility complex is indicated by the upper arrow, and free probe is shown by the lower arrow. B, the A and B DUSP4 sequences bind protein complexes in vitro with cross-competitive binding specificities with each other and with other Hox/Pbx consensus oligonucleotides. EMSA were performed with U937 nuclear proteins (1 g) and the A or B DUSP4 oligonucleotide probes. Binding reactions were incubated with a 200-fold molar excess of unlabeled, double-stranded oligonucleotide competitor. Homologous oligonucleotides compete for binding of the low mobility complex, but an unrelated oligonucleotide competitor does not. Also, there is cross-competitive binding specificity between the A and B probes and between these probes and the HoxA10/Pbx binding site from the CYBB promoter. Binding affinity of the B oligonucleotide is less than the A probe in these experiments. Because of the difference in intensity of the shifted bands generated by the A and B probes, different exposure durations were used for the two probes. C, the protein complexes that bind to the A and B DUSP4 sequences in vitro are cross-immunoreactive with HoxA10 and Pbx2. EMSA were performed with U937 nuclear proteins (1 g) and the A or B DUSP4 oligonucleotide probes. Binding reactions were preincubated with antibodies (Ab) to HoxA10, Pbx1, or Pbx2 or irrelevant control antibody (2 g each), as indicated. Binding of the low mobility complex to either the A or B probe is specifically disrupted by antibody to HoxA10 or Pbx2. The upper arrow indicates the specific, low mobility complex, and the lower arrow indicates free probe.
consensus sequence for DNA binding of HoxA10/Pbx heterodimers. We found that the activity of these two cis elements is additive in the context of the intact promoter. We also found that these cis elements are more active in undifferentiated myeloid cells than after cytokine-induced differentiation. Consistent with this differentiation stage-specific activity, binding of HoxA10 and acetylated histones to these cis elements is greater in undifferentiated myeloid cells.
In contrast to the current studies, we previously found that HoxA10 represses transcription of the CYBB and NCF2 genes in immature myeloid cells (10,12). Therefore, HoxA10 can either activate or repress transcription of various target genes in undifferentiated myeloid cells. In the current studies, we found that HoxA10 overexpression prevents the normal down-regulation of DUSP4 transcription during myelopoiesis. In contrast, in our previous studies, we found that HoxA10 overexpression did not induce differentiation stage-inappropriate repression of CYBB or NCF2 transcription (10,12).
During cytokine-induced differentiation, HoxA10 binding affinity for the negative CYBB and NCF2 cis elements decreases. We found this decrease was due to phosphorylation of specific tyrosine residues in the HoxA10 homeodomain (11). In contrast, overexpressed-HoxA10 induces as much activity from the DUSP4 cis elements in differentiating myeloid cells as in undifferentiated cells. These results suggest that HoxA10 transcriptional activation and repression functions are not similarly regulated by cytokine-mediated post-translational modification of HoxA10.
However, although overexpressed HoxA10 induces the same percentage increase in activity of the two DUSP4 cis elements in undifferentiated transfectants as in differentiated transfectants, the total amount of cis element activity is greater in undifferentiated transfectants. One possible explanation for this might be that there is less endogenous HoxA10 in the IFN␥-treated transfectants in comparison with untreated transfectants. This could result in a lower base-line activity of the cis elements. However, our previous studies indicate that HoxA10 protein abundance is relatively constant during differentiation of U937 myeloid cells. This is in contrast to the decrease in HoxA10 expression during the CD34 ϩ to CD34 Ϫ transition in normal myelopoiesis and represents an aspect of the transformed phenotype of these cells.
An alternative explanation for the decreased activity of these DUSP4 cis elements after differentiation, even in HoxA10overexpressing cells, is a decrease in activity of a crucial partner protein or co-activator. We find that HoxA10 binds the two DUSP4 cis elements as a heterodimer with Pbx2. This is in contrast to the HoxA10/Pbx1 heterodimer, which binds the CYBB and NCF2 cis elements. The two proteins in Hox/Pbx heterodimers are hypothesized to have different functions. For some cis elements, the Hox protein is thought to provide binding site specificity, and the Pbx protein is thought to increase binding affinity of the complex. It is possible that altered Pbx2 activity during differentiation influences HoxA10/Pbx2 complex binding to the two DUSP4 cis elements. This is an active area of investigation in the laboratory. Chromatin immunoprecipitation experiments were performed with U937 cells, with and without 48 h of IFN␥ differentiation. Lysates were immunoprecipitated with an antibody to HoxA10 or control preimmune serum. Co-precipitating chromatin was analyzed by real time PCR using primers designed to amplify ϳ80-bp sequences centered around the A and B cis elements. Results were normalized to input (unprecipitated) chromatin to control for total DNA abundance in the samples. HoxA10 interaction with either the A or B sequence decreased significantly during IFN␥ differentiation of the cells. Also, significantly less HoxA10 interacted in vivo with the B cis element in comparison with the A cis element, with and without IFN␥. * and **, statistically significant difference in the amount of A versus B chromatin co-precipitating with HoxA10 antibody. Insignificant amounts of preimmune serum co-precipitating chromatin were amplified by either the A or B primer sets, as indicated. B, IFN␥ differentiation decreases interaction of acetylated histone 3 with the A and B regions of the DUSP4 5Ј-flank in vivo. Chromatin immunoprecipitation experiments were performed with U937 cells, with and without 48 h of IFN␥ differentiation. Lysates were immunoprecipitated with an antibody to acetyl-histone 3 (H3) or preimmune serum as a negative control. Co-precipitating chromatin was analyzed by real time PCR using primers designed to amplify ϳ80-bp sequences centered around the A and B cis elements. Results were normalized to input (unprecipitated) chromatin to control for total DNA abundance in the samples. IFN␥ differentiation significantly decreases interaction of acetyl-H3 with the A and B regions of the DUSP4 promoter. Also, significantly more acetyl-H3 interacts with the A cis element in comparison with the B cis element, with and without IFN␥. * and **, statistically significant difference in A versus B chromatin co-precipitating with acetyl-H3 antibody. Insignificant amounts of preimmune serum co-precipitating chromatin were amplified by either the A or B primer sets, as indicated.
The two identified HoxA10-binding cis elements in the DUSP4 promoter are about 600 bp apart (Ϫ1157 to Ϫ1165 and Ϫ1132 to Ϫ1139 bp from the ATG start codon). Similarly, the CYBB promoter also includes several HoxA10-binding cis elements in the proximal promoter, each of which are several hundred bp apart (21). As in the current studies, these CYBB cis elements exhibit variable binding affinity for HoxA10. In the current studies, we found that the more proximal cis element had lower HoxA10 binding affinity, as determined by in vitro and in vivo binding studies. Consistent with this, the proximal cis element showed less activity in U937 transfection experiments, with and without IFN␥ differentiation. However, in the context of the whole promoter, we found that the A and B cis elements exhibit approximately equivalent activity. This may be related to the relative proximity of the lower affinity, B cis element to the transcriptional start site in the DUSP4 gene.
It is possible that the variable affinity of HoxA10 for these two cis elements provides a mechanism for a graded response to HoxA10-induced activation at various points during the differentiation program. Specifically, a decrease in HoxA10 abundance, such as normally occurs during the transition from CD34 ϩ to CD34 Ϫ progenitors (3), would have a greater impact on the function of the proximal DUSP4 cis element. However, further events during terminal differentiation or phagocyte activation might influence HoxA10 binding to the distal cis element. This would decrease Mkp2 expression, increase Jnk1 activation, and lead to apoptosis. This process could be deranged at various points by the extent of differentiation stage-inappropriate HoxA10 expression or by other factors that influence binding affinity of the HoxA10/Pbx2 complex to the DUSP4 cis elements. Further clarification of the role of multiple HoxA10 binding cis elements in differentiation stage-specific regulation of various target genes will be of interest.
A number of pathways are involved in programmed cell death during myelopoiesis. One pathway involves Jnk, also referred to as Sapk (stress-activated kinases) (27). Jnk is activated in response to genotoxic stresses and also in response to the "stress" of cytokines, such as granulocyte-macrophage colony-stimulating factor, interleukin-3, and IFN␥ (17). In human chronic myeloid leukemia, activation of Jnk1 can lead to apoptosis of the transformed leukocytes (28), suggesting a physiologic relevance to this pathway in malignant myeloid disease. Activation of Jnk is partly dependent upon activity of protein phosphatases of the Mkp family. In vivo, phospho-Jnk is the preferred substrate for Mkp2, and p38 mitogen-activated protein kinase is the preferred substrate for Mkp1 (15). These results suggest that HoxA10 induction of Mkp2 expression influences apoptosis via the Jnk phosphatase role of Mkp2.
We find that inhibition of differentiation-induced apoptosis by HoxA10 overexpression is at least partly dependent on increased Mkp2 expression in U937 cells. These results suggest that HoxA10 activation of DUSP4 transcription is physiologically significant to the phenotype of malignant myeloid cells overexpressing HoxA10. However, our studies indicate that not all of the apoptosis resistance of HoxA10-overexpressing cells was reversed by Mkp2 knockdown. It is possible that this reflects the efficiency of Mkp2 suppression by the specific shRNA. However, this result could also indicate that HoxA10 influences expression of additional target genes involved in programmed cell death in differentiating myeloid cells. Consistent with this latter possibility, not only did HoxA10 overexpression abolish differentiation-induced Jnk1 activation; phospho-Jnk1 is actually decreased in HoxA10-overexpressing U937 cells during IFN␥ differentiation. By coupling chromatin co-immunoprecipitation with CpG island microarray screening, we identified a number of other potential HoxA10 target genes involved in apoptosis, proliferation, and cell cycle regulation. These additional genes will be the subject of future studies.
Regulation of apoptosis in differentiating myeloid cells is complex. The current studies represent the first identification of a role for HoxA10 in this process. HoxA10 abundance decreases during normal myelopoiesis. Our studies suggest this decreases HoxA10 activation of DUSP4 transcription, especially via the proximal, low affinity cis element. This has implications for the impact of sustained HoxA10 expression during differentiation in various forms of acute myeloid leukemia (3). Specifically, these results suggest that sustained HoxA10-expression in maturing progenitors would sustain Mkp2 expression, impairing Jnk activation and therefore apoptosis. This could be a mechanism that contributes to myeloid progenitor expansion and leukocytosis in HoxA10-overexpressing murine models. Therefore, these studies suggest a potential pathway for molecular therapeutic targeting in myeloid malignancies characterized by overexpression of Abd HoxA proteins.