PU.1, Interferon Regulatory Factor (IRF) 2, and the Interferon Consensus Sequence-binding Protein (ICSBP/IRF8) Cooperate to Activate NF1 Transcription in Differentiating Myeloid Cells*

Nf1 (neurofibromin 1) is a Ras-GAP protein that regulates cytokine-induced proliferation of myeloid cells. In previous studies, we found that the interferon consensus sequence-binding protein (ICSBP; also referred to as interferon regulatory factor 8) activates transcription of the gene encoding Nf1 (the NF1 gene) in differentiating myeloid cells. We also found that NF1 activation requires cytokine-stimulated phosphorylation of a conserved tyrosine residue in the interferon regulatory factor (IRF) domain of ICSBP/IRF8. In this study, we found that ICSBP/IRF8 cooperates with PU.1 and interferon regulatory factor 2 to activate a composite ets/IRF-cis element in the NF1 promoter. We found that PU.1 binds directly to the NF1-cis element, and DNA-bound PU.1 interacts with IRF2, recruiting IRF2 to the cis element. This interaction requires cytokine-induced phosphorylation of specific serine residues in the PU.1 PEST domain and of a conserved tyrosine residue in the IRF domain of IRF2. We found that ICSBP/IRF8 interaction with the NF1-cis element requires pre-binding of PU.1 and IRF2. The conserved IRF domain tyrosine in ICSBP/IRF8 is required for interaction with the DNA-bound PU.1-IRF2 heterodimer. NF1 deficiency in myeloid progenitor cells results in cytokine hypersensitivity and myeloproliferation. Therefore, these studies identify a target gene for the previously observed tumor-suppressor effect of PU.1. Additionally, these studies identify a tumor-suppressor function for the “oncogenic” transcription factor, IRF2.

Previously, we found that NF1 transcription and Nf1 expression increase during cytokine-induced differentiation of myeloid leukemia cell lines or murine myeloid progenitor cells (4). We also found that cytokine-induced NF1 transcription requires the interferon consensus sequence-binding protein (ICSBP or IRF8) (4). ICSBP/IRF8 is expressed exclusively in myeloid and B-cells (10), and acquired ICSBP/IRF8 deficiency is found in bone marrow cells from subjects with chronic myeloid leukemia, AML, and MDS (11,12). Interestingly, ICSBP/ IRF8 deficiency induces myeloproliferation in mice, and myeloid cells from these mice are hypersensitive to the same cytokines as Nf1-deficient cells (4,13,14). Consistent with a role in NF1 transcription, proliferative abnormalities in ICSBP/ IRF8-deficient myeloid progenitor cells can be rescued by expression of the Nf1GAP-related domain (4).
In previous studies, we also found that activation of NF1 transcription requires cytokine-induced phosphorylation of a specific ICSBP/IRF8 tyrosine residue (Tyr-95) (15). This residue is in the conserved IRF domain that is thought to be involved in DNA-binding or protein/protein interactions of IRF proteins. ICSBP/IRF8 is a substrate for SHP1 and SHP2 protein-tyrosine phosphatases in undifferentiated myeloid cells (15,16). However, a constitutively activated SHP2 mutant, described in human subjects with MDS, AML, and juvenile myelomonocytic leukemia, dephosphorylates ICSBP/IRF8 in differentiated and undifferentiated myeloid cells (15,17). Such activated SHP2 mutants also induce cytokine hypersensitivity in myeloid progenitors (15,18).
IRF proteins regulate target gene transcription by interacting with several different DNA-binding site consensus sequences. ICSBP/IRF8 represses cis elements with PRDI consensus sequences (5Ј-TCACTT-3Ј) by interacting directly with DNA.
The ICSBP/IRF8-binding cis element in the proximal NF1 promoter has homology to composite ets/IRF consensus sequences found in myeloid-specific genes. This suggests possible involvement of PU.1 in NF1 transcriptional regulation and perhaps another IRF protein. The goal of these investigations is to determine the mechanism of cytokineinduced NF1 transcription in differentiating myeloid cells. This will be approached by identifying the components of the NF1 transcriptional activation complex.
Although composite ets/IRF consensus sequences have been identified in a number of genes involved in the inflammatory response, no such cis elements have previously been identified in target genes regulating proliferation. Identification of homologous cis elements that interact with common trans-factors in genes that regulate both differentiation and proliferation would suggest a common mechanism of cytokine activation of different types of genes. This could have implications for understanding the inter-relationship between differentiationprogression and proliferation-arrest during myelopoiesis.

Plasmids and PCR Mutagenesis
Protein Expression Vectors-The ICSBP/IRF8 cDNA was obtained from Dr. Ben Zion-Levi (Technion, Haifa, Israel), and the full-length cDNA was generated by PCR and subcloned into the mammalian expression vector pcDNAamp, as described (21). The cDNA for IRF2 in the pcDNAamp vector was obtained from Dr. Gary S. Stein (University of Massachusetts Medical School, Worcester, MA). IRF2 with mutation of a conserved tyrosine residue in the IRF domain (Tyr-109) to phenylalanine was generated by PCR using the Clontech "QuikChange" protocol, as described (16). Mutant clones were sequenced on both strands to verify that only intended mutations had been introduced. Wild type PU.1 and PU.1 mutants with serine 41 and 45, 148, or 132 and 133 changed to alanine were obtained from Dr. Michael L. Atchison (University of Pennsylvania School of Veterinary Medicine, Philadelphia) and subcloned into the pSR␣ mammalian expression vector. These cDNAs were also subcloned into the pGEX1 vector (Amersham Biosciences) for expression in Escherichia coli as glutathione S-transferase (GST) fusion proteins.
Reporter Constructs-An artificial promoter construct with four copies of the ICSBP/IRF8-binding cis element from the NF1 promoter (bp Ϫ320 to Ϫ336) linked to a minimal promoter and a CAT reporter (the p-TATACAT vector) was previously described (15). This construct is referred to as nf1TATACAT.

Myeloid Cell Line Culture
The human myelomonocytic cell line U937 (24) was obtained from Andrew Kraft (Hollings Cancer Center, Medical University of South Carolina, Charleston, SC). Cells were maintained and differentiated as described (21). For differentiation experiments, U937 cells were treated for 24 or 48 h with 500 units/ml human recombinant IFN␥ (Roche Applied Science) (21).

Isolation of Nuclear Proteins and Electrophoretic Mobility Shift Assays
Nuclear extract proteins were isolated from U937 cells by the method of Dignam et al. (25) (with the addition of protease inhibitors but not phosphatase inhibitors, as described). In some experiments, U937 cells were differentiated with 500 units/ml of IFN␥ before nuclear protein isolation. Oligonucleotides probes were prepared, and EMSA and antibody supershift assays were performed, as described (21). Antibodies to phosphotyrosine, ICSBP/IRF8, IRF1, IRF2, PU.1, and irrelevant, control GST antibody were obtained from Santa Cruz Biotechnology (Santa Cruz, CA).

Immunoprecipitation and Western Blots
Western Blots of U937 Lysates Proteins-U937 cells were lysed by boiling in 2ϫ SDS sample buffer. Lysate proteins (30 g) were separated by SDS-PAGE (12% acrylamide) and transferred to nitrocellulose, according to standard techniques. Western blots were serially probed with antibodies to Nf1, ICSBP/IRF8, IRF2, PU.1, and GAPDH (to control for loading). In other studies, nuclear proteins were isolated from U937 cells (with or without IFN␥ treatment). 30 g of protein were separated by SDS-PAGE (20% acrylamide gel) and transferred to nitrocellulose. Western blots were serially probed with antibodies to PU.1, IRF2, pERK2 (as a control for IFN␥ treatment), total ERK2 (as a loading control), or GAPDH (as a loading control).
Immunoprecipitation and Western Blots-Nuclear proteins were isolated from U937 cells with or without IFN␥ treatment and immunoprecipitated under denaturing conditions with antibody to PU.1, IRF2, phosphotyrosine (Tyr(P)), or irrelevant control antibody (anti-GST), as described previously (14,15). Precipitated proteins were collected with staphylococcus protein A-Sepharose, separated by SDS-PAGE, and transferred to nitrocellulose, as above. Western blots were serially probed with an anti-phosphotyrosine antibody (clone 4G10, Upstate) and antibodies to IRF2 or PU.1.
DNA Affinity Co-immunoprecipitation-Nuclear proteins were isolated from U937 cells with or without IFN␥ treatment. Proteins (300 g) were incubated with a biotin-labeled doublestranded oligonucleotide probe representing the Ϫ320to Ϫ336-bp NF1 promoter sequence or a specific mutant, non-ICSBP/IRF8-binding sequence. Probes were immunoprecipitated with anti-biotin antibody or irrelevant control antibody under nondenaturing conditions and recovered with staphylococcus protein A-Sepharose beads. Immunoprecipitates were separated by SDS-PAGE (10% acrylamide) and proteins transferred to nitrocellulose. Western blots were serially probed with antibodies to ICSBP/IRF8, IRF2, and PU.1.
In other experiments, 35 S-labeled, in vitro translated proteins were generated using rabbit reticulocyte lysate from in vitro transcribed RNA, as described previously (16). These proteins were also used in DNA affinity co-immunoprecipitation experiments with biotin-labeled probes, as described above. For these experiments, DNA/protein interaction was identified by autoradiography of SDS-PAGE.
Metabolic Labeling and Immunoprecipitation-U937 cells treated with IFN-␥ for 0, 24, and 48 h were incubated for 4 h at 37°C with [ 32 P]orthophosphate. Cells were lysed under denaturing conditions, and lysate proteins (100 g) were immunoprecipitated with antibody to PU.1 or irrelevant control antibody. Immunoprecipitates were collected with staphylococcus protein A-Sepharose beads and separated by SDS-PAGE (20% acrylamide). Phosphorylated PU.1 protein was identified by autoradiography of the fixed and dried gel.

In Vitro Protein Translation and GST Co-affinity Purification Assay
In vitro transcribed IRF2, Y109F IRF2, ICSBP/IRF8, Y95F ICSBP/IRF8, PU.1, S148A PU.1, or S132A/S133A PU.1 mRNAs were generated from linearized template DNA using the riboprobe system, according to manufacturer's instructions (Promega). In vitro translated proteins were generated in rabbit reticulocyte lysate, according to the manufacturer's instructions (Promega). Control (unprogrammed) lysates were generated in similar reactions in the absence of input RNA, and proteins were radiolabeled by including [ 35 S]methionine in the translation reaction. JM109 E. coli transformed with PU.1, IRF2, or ICSBP/IRF8 in the pGEX vector (or empty control vector) were grown to log phase, supplemented to 0.1 mM isopropyl 1-thio-␤-D-galactopyranoside, and incubated for 3 h at 37°C with shaking. The cells were harvested and resuspended in HN buffer (20 mM HEPES (pH 7.4), 0.1 M NaCl, 2 mM MgCl 2 , 0.1 mM EDTA, 0.5% Nonidet P-40, 0.1% Triton X-100, 2 mM phenylmethylsulfonyl fluoride, 5 mM NaF) and sonicated on ice (22). Debris was removed by centrifugation, and the lysate was incubated with glutathione-agarose beads (Sigma) and washed extensively. The beads were preincubated with control rabbit reticulocyte lysate to induce serine/threonine phosphorylation of PU.1/GST and tyrosine phosphorylation of ICSBP/GST or IRF2/GST as described (16,26). GST proteins were then incubated with [ 35 S]methionine-labeled in vitro translated protein.
Proteins were eluted with SDS-PAGE sample buffer, separated on 15% SDS-PAGE, and identified by autoradiography of the fixed and dried gel.

RESULTS
The Positive NF1-cis Element Binds a Multiprotein Complex-IFN␥ treatment of U937 cells increases NF1 promoter activity and ICSBP/IRF8 binding to a positive cis element in the NF1 promoter (4,15). To identify other proteins that interact with this cis element, we used in vitro DNA binding assays. EMSAs were performed with a radiolabeled, double-stranded oligonucleotide representing the NF1-cis element and nuclear proteins from U937 cells. We found that IFN␥ differentiation of U937 cells increases in vitro protein binding to this probe, consistent with our previous results ( Fig. 1A) (4,15).
We noted that the NF1-cis element has homology with the ets/IRF consensus sequences from the CYBB or NCF2 promoters. Therefore, we used double-stranded oligonucleotide competitors representing these cis element to investigate binding specificity of the NF1-protein complex. These competitors were compared with homologous oligonucleotide (ds NF1) or an oligonucleotide with mutation which abolishes protein binding to the NF1-cis element (ds mutNF1) (described (4)). In EMSA with the NF1 probe and nuclear proteins from IFN␥differentiated U937 cells, we found that the CYBB and NCF2 oligonucleotides compete for protein complex binding (Fig.  1B). This result suggests that the NF1-cis element interacts with PU.1 and one or more IRF proteins. Therefore, we performed EMSA with the NF1 probe, nuclear proteins from IFN␥-treated U937 cells, and antibodies to various transcription factors (Fig. 1C). In these studies, we found that the shifted complex is cross-immunoreactive with PU.1, ICSBP/IRF8, and IRF2 but not IRF1. Because the NF1-cis element appears to bind a multiprotein complex, we tested the ability of various combinations of PU.1, IRF2, and ICSBP/IRF8 antibodies to disrupt the complex (Fig. 1C). We found that the complex is completely disrupted by combining all three antibodies.
We also used DAPA to determine the impact of differentiation on assembly of the multiprotein complex on the NF1-cis element. To determine specificity of binding, wild type NF1-cis element probe (ds NF1) was compared with the binding-mutant probe (ds mutNF1). These oligonucleotides were biotinlabeled and incubated with U937 nuclear proteins. Probes were precipitated with anti-biotin antibody, and co-precipitating proteins were separated by SDS-PAGE and identified by Western blot (Fig. 1D).
In this assay, we also find that IFN␥ treatment increases interaction of PU.1, IRF2, and ICSBP/IRF8 with the NF1-cis element probe. In contrast, we found previously that PU.1 and IRF1 bind the CYBB-and NCF2-cis elements in undifferentiated myeloid cells. For those cis elements, differentiation increases ICSBP/IRF8 interaction with these proteins but does not increase DNA binding of PU.1 and IRF1. Our current results suggest differences in the activation of various ets/IRFcis elements, even in the same lineage.
Identification of Protein/Protein/DNA Interactions Required for Assembly of the NF1-cis Element-binding Complex-Previous studies suggest that PU.1 binding to ets/IRF consensus sequences is required to provide an anchor for IRF proteins. Therefore, we investigated the role of PU.1 in mediating IRF2 and ICSBP/IRF8 binding to the NF1-cis element. For these studies, we used in vitro translated, 35 S-labeled PU.1, IRF2, and ICSBP/IRF8 ( Fig. 2A). Because of differences in size, the three proteins can be detected simultaneously by SDS-PAGE. These in vitro translated proteins were tested for binding to biotinlabeled ds NF1 or ds mutNF1 probes in DNA-affinity purification assays (DAPA), as above. In these experiments the total amount of reticulocyte lysate in the binding assays was kept constant by inclusion of control (no RNA) lysate.
In initial assays, we tested each of these proteins individually for binding to the ds NF1 probe (Fig. 2B). We found that only FIGURE 1. PU.1, IRF2, and ICSBP/IRF8 interact with the positive cis element in the proximal NF1 promoter (؊320-to ؊336-bp). A, IFN␥ differentiation of U937 cells increases interaction of a protein complex with the Ϫ320to Ϫ336-bp NF1 promoter sequence. EMSAs were performed with a double-stranded oligonucleotide representing the Ϫ320to Ϫ336-bp sequence from the NF1 promoter (ds NF1 probe), and nuclear proteins were isolated from U937 cells. Binding of a protein complex increases in EMSA with nuclear proteins from IFN␥-differentiated cells in comparison with untreated U937 cells (indicated by an arrow). B, binding specificity of the protein complex that interacts with the Ϫ320 to Ϫ336-bp NF1 promoter sequence is cross-reactive with composite ets/IRF-cis elements. EMSAs were performed with the ds NF1 probe and nuclear proteins isolated from IFN␥-treated U937 cells. Binding reactions were preincubated with unlabeled double-stranded oligonucleotides representing the CYBB ets/IRF-cis element (dsCYBB), the NCF2 ets/IRF-cis element (dsNCF2), homologous oligonucleotide (dsNF1), or a mutant homologous oligonucleotide (dsmutNF1). The NF1-protein complex has cross-binding specificity with other ets/IRF-cis elements. The complex of interest is indicated by an arrow. C, protein complex binding to the Ϫ320 to Ϫ336 bp NF1 promoter sequence is cross-immunoreactive with PU.1, IRF2, and ICSBP. EMSAs were performed with the ds NF1 probe and nuclear proteins isolated from IFN␥-treated U937 cells. Binding reactions were preincubated with various combinations of antibodies (Ab) to PU.1, IRF2, and ICSBP/ IRF8, as indicated. The NF1-binding complex is disrupted by antibodies to PU.1, IRF2, and ICSBP/IRF8 but not IRF1. Consistent with binding of a multiprotein complex, combinations of antibodies are more efficient than any of these antibodies individually. The complex of interest is indicated by an arrowhead. D, binding of PU.1, IRF2, and ICSBP/IRF8 to the Ϫ320to Ϫ336-bp NF1 promoter sequence is increased by IFN␥ differentiation of U937 cells. DAPAs were performed with biotin-labeled ds NF1 and ds mutNF1 probes. Probes were incubated with nuclear proteins isolated from U937 cells with or without IFN␥ treatment, immunoprecipitated with an anti-biotin antibody, and co-precipitating proteins were separated by SDS-PAGE. Western blots (WB) were serially probed with antibodies to PU.1, IRF2, and ICSBP/IRF8. Binding of all three component proteins to the wild type probe is increased in nuclear proteins from IFN␥-differentiated cells.
PU.1 is able to interact independently with this probe, and none of the proteins interact with the ds mutNF1 probe. Previous studies indicate that phosphorylation of serine residues in the PU.1 PEST domain is important for PU.1 protein/protein/DNA interactions. For example, PU.1 serine 148 is necessary for recruitment of ICSBP/IRF8 to the CYBB promoter (21). Similarly, serines 132 and 133 in PU.1 are essential for PU.1 activation of macrophage-specific genes (27). Because these residues are phosphorylated in reticulocyte lysate (28,29), we used DAPA to investigate the role of PU.1 phosphorylation in the assembly of the NF1-binding protein complex. We found that in vitro translated wild type PU.1 and PU.1 with mutation of serine 148 or 132/133 to alanine (S148A PU.1 and S132A/ S133A PU.1, respectively) have similar binding to the NF1-cis element probe (Fig. 2C). This result is important for interpretation of transfection experiments as shown in the sections below.
We next investigated whether PU.1 recruits either IRF protein to the NF1-cis element. Biotin-labeled probe was incubated with PU.1 and either ICSBP/IRF8 or IRF2 (Fig. 2D). We found that PU.1 recruits IRF2 to the NF1-cis element but not ICSBP/ IRF8. Therefore, we tested the effect of PU.1 PEST domain residues on interaction with IRF2 by DAPA (Fig. 2D). In these studies, DNA-bound wild type and S148A PU.1 interact with IRF2 equivalently. In contrast, IRF2 does not interact with S132A/S133A PU.1 bound to the ds NF1 probe.
We previously identified a conserved tyrosine residue in the ICSBP/IRF8 IRF domain that is necessary for interaction with a PU.1-IRF1 heterodimer bound to the CYBB-or NCF2-cis element. Because IRF domain tyrosine residues are phosphorylated in reticulocyte lysate (16), we used DAPA to determine the role of the IRF2 IRF domain tyrosine (Tyr-109) in interaction with PU.1. We incubated PU.1 and wild type or Y109F IRF2 (tyrosine 109 mutated to phenylalanine) with the biotin-labeled ds NF1 or ds mutNF1 probe (Fig. 2D). We found that Y109F IRF2 interacts less efficiently with DNA-bound PU.1 than wild type IRF2.
We next investigated whether ICSBP/IRF8 can interact with the DNA-bound PU.1-IRF2 heterodimer. We found that in vitro translated ICSBP/IRF8 interacts with the ds NF1 probe (but not the ds mutNF1 probe) in binding assays with PU.1 and IRF2 (Fig. 2E). Because the conserved IRF domain tyrosine in ICSBP/IRF8 (Tyr-95) is required for activation of the NF1-cis element, we performed binding assays to determine the role of , and ICSBP/IRF8 were generated in reticulocyte lysate. These proteins were separated by SDS-PAGE and detected by autoradiography. The various proteins migrate differently based on size, as indicated by the arrows. B, in vitro translated PU.1 binds to the Ϫ320to Ϫ336-bp NF1 promoter sequence. DAPA was performed with biotin-labeled ds NF1 or ds mutNF1 probes and in vitro translated, 35 S-labeled PU.1, IRF2, or ICSBP/IRF8. Only PU.1 is able to bind to the ds NF1 probe in this assay. C, PU.1 with mutation of various PEST domain serine residues binds the Ϫ320to Ϫ336-bp NF1 promoter sequence. DAPA was performed with biotin-labeled ds NF1 or ds mutNF1 probe and in vitro translated (IVT), 35 S-labeled PU.1, S148A PU.1, or S132A/S133A PU.1. All of these forms of PU.1 bind the ds NF1 probe equivalently. D, in vitro translated IRF2 binds to the Ϫ320to Ϫ336-bp NF1 promoter sequence in the presence of PU.1. DAPA was performed with biotin-labeled ds NF1 or ds mutNF1 probe and in vitro translated, 35 S-labeled PU.1, S148A PU.1 or S132A/S133A PU.1 with ICSBP/IRF8, IRF2 or Y109F IRF2. Either wild type or S148A PU.1 recruits IRF2 to the ds NF1 probe in this assay. In contrast, none of the forms of PU.1 recruit ICSBP/IRF8 or Y109F IRF2 to the ds NF1 probe. E, Ϫ320to Ϫ336-bp NF1 promoter sequence interacts with a heterotrimer of in vitro translated PU.1, IRF2, and ICSB/IRF8P. DAPA was performed with biotin-labeled ds NF1 or ds mutNF1 probe and various combinations of in vitro translated, 35 S-labeled wild type, S148A, or S132A/S133A PU.1; wild type or Y95F ICSBP/IRF8; and wild type or Y109F IRF2. The ds NF1 probe interacts simultaneously with PU.1, IRF2, and ICSBP/IRF8. However, S148A PU.1 interacts only with IRF2 but not ICSBP/IRF8. Additionally, the interaction is disrupted by mutation of the conserved Tyr residue in the IRF domain of either IRF2 (Tyr-109) or ICSBP/IRF8 (Tyr-95).
Based on these results, it is possible that ICSBP/IRF8 interacts directly with IRF2 and that IRF2 interacts with DNAbound PU.1. However, it is also possible that DNA-bound PU.1 and IRF2 form a binding site that involves interaction of ICSBP/ IRF8 with both proteins. Otherwise, ICSBP/IRF8 interaction may depend on conformational changes of either PU.1 or IRF2, which occur when these two proteins interact with the DNAbinding site. Therefore, we tested the impact of mutating PU.1 serine 148 on interaction of ICSBP/IRF8 with PU.1 ϩ IRF2 bound to the NF1-cis element. We found that ds NF1-bound S148A PU.1 ϩ IRF2 does not recruit ICSBP/IRF8 to the binding site as efficiently as wild type PU.1 ϩ IRF2 (Fig. 2E).
These studies identify the order of binding of the activation complex proteins to the NF1-cis element. We also investigated interaction between these proteins using an in vitro assay system that does not require DNA binding of component proteins. For these studies, we expressed IRF2 in E. coli as a fusion protein with glutathione S-transferase. The IRF2/GST fusion protein was phosphorylated by incubation with reticulocyte lysate, as described previously (16). We investigated interaction with in vitro translated wild type versus S132A/S133A PU.1 versus S148A PU.1 by "pulldown" assay. Previously, we found that mutation of Ser-148 impairs the ability of PU.1 to interact with ICSBP/IRF8 in such assays (16). In this study, less in vitro translated S132A/S133A PU.1 co-purifies with IRF2/GST in comparison with the wild type PU.1 (Fig. 3A). In contrast, S148A PU.1 and wild type PU.1 copurify equivalently with IRF2/GST (Fig. 3B).
We also determined the role of IRF2 IRF domain phosphorylation for interaction with non-DNAbound PU.1. For these studies, PU.1 was expressed in E. coli as a GST fusion protein and phosphorylated by incubation with reticulocyte lysate, as described (29). Wild type and Y109F IRF2 were translated in vitro in reticulocyte lysate. We found that wild type IRF2 co-affinity purifies with PU.1/GST more efficiently than Y109F IRF2 (Fig. 3C). We also performed similar experiments to pursue the hypothesis that ICSBP/IRF8 tyrosine phosphorylation is necessary for interaction with IRF2. For these studies, IRF2 was expressed as a GST fusion protein, as described above, and incubated with in vitro translated wild type or Y95F ICSBP/IRF8. We found that the wild type protein co-precipitates more efficiently with IRF2/ GST than does Y95F ICSBP/IRF8 (Fig. 3D).
None of these in vitro translated proteins co-precipitate with control GST. Phosphorylation of GST fusion proteins and in vitro translated proteins is discussed under "Materials and Methods." PU.1, IRF, and ICSBP/IRF8 Cooperate to Activate NF1 Transcription-We next investigated the functional significance of PU.1 and IRF2 for NF1 transcription. In previous studies, we determine that overexpressed ICSBP/IRF8 activates transcription from an artificial promoter construct with multiple copies of the NF1-cis element in U937 transfectants (15). For this study, we used the same reporter system to determine the impact of overexpressed PU.1 and IRF2 on this NF1-cis element. Expression of IRF2 is ubiquitous, and PU.1 and ICSBP/IRF8 are expressed in all myeloid cell lines. Therefore, overexpressed ICSBP/IRF8 may have activated the NF1-cis element in our previous studies by interacting with endogenous PU.1 and IRF2. In this case, overexpression of various combinations PU.1, IRF2, and ICSBP/IRF8 might be expected to result in more reporter gene expression than overexpression of any protein individually.
In initial experiments, U937 cells were transfected with a reporter construct containing a minimal promoter with or without four copies of the NF1-cis element (nf1TATACAT and pTATACAT control, respectively). The cells were co-transfected with mammalian expression vectors to overexpress A, S132A/S133A PU.1 has lower affinity for IRF2 than does wild type. IRF2 was expressed as a fusion protein with glutathione S-transferase in E. coli (IRF2/GST). This fusion protein and control GST were used in affinity purification assays with in vitro translated (IVT), 35 S-labeled S132A/S133A PU.1, or wild type PU.1 (GST pulldown assay). In vitro translated S132A/S133A PU.1 has less affinity for IRF2/GST than does in vitro translated wild type PU.1. Neither of these in vitro translated proteins co-purifies with control GST. B, S148A PU.1 has equivalent affinity for IRF2 than does wild type. IRF2 was expressed as a fusion protein with glutathione S-transferase in E. coli. IRF2/GST and control GST were used in affinity purification assays with in vitro translated, 35 S-labeled S148A PU.1 or wild type PU.1 (GST pulldown assay). The affinities of S148A PU.1 and wild type PU.1 for IRF2/GST are equivalent. Neither of these in vitro translated proteins co-purifies with control GST. C, Y109F IRF2 has lower affinity for PU.1 than does wild type. PU.1 was expressed as a fusion protein with glutathione S-transferase in E. coli (PU.1/GST). This fusion protein and control GST were used in affinity purification assays with in vitro translated, 35 S-labeled Y109F IRF2 or wild type IRF2 (GST pulldown assay). In vitro translated Y109F IRF2 has less affinity for PU.1/GST than does in vitro translated wild type IRF2. Neither of these in vitro translated proteins co-purifies with control GST. D, Y95F ICSBP/IRF8 has lower affinity for IRF2 than does wild type. IRF2 was expressed as a fusion protein with glutathione S-transferase in E. coli. IRF2/GST and control GST were used in affinity purification assays with in vitro translated, 35 S-labeled Y95F ICSBP/IRF8 or wild type ICSBP/IRF8 (GST pulldown assay). In vitro translated Y95F ICSBP/IRF8 has less affinity for IRF2/GST than in vitro translated wild type ICSBP/IRF8. Neither of these in vitro translated proteins co-purifies with control GST.
We found that ICSBP/IRF8 overexpression activates the NF1-cis element-containing reporter construct, consistent with our previous results (15). Similarly, we found that overexpression of IRF2 also significantly increases reporter activity of the NF1-cis element-containing construct in both untreated and IFN␥-differentiated transfectants (p Ͻ 0.0001, n ϭ 5). PU.1 overexpression induces a smaller but statistically significant increase in reporter expression from the NF1-cis element in undifferentiated transfectants (p Յ 0.04, n ϭ 12) and a larger increase in differentiated transfectants (p Ͻ 0.0001, n ϭ 12).
Based on these results, we tested the effect of overexpressing combinations of these proteins (Fig. 4A). We found that cooverexpressing PU.1 with either IRF2 or ICSBP/IRF8 results in a statistically significant increase in activity of the NF1-cis element in comparison with overexpressing either IRF protein alone (p Յ 0.002, n ϭ 4 for all combinations, with and without IFN␥). We also tested the effect of co-overexpressing all three proteins on NF1-cis element activity. We were interested in determining whether the effect of overexpressing PU.1 and both IRF proteins was different from overexpressing PU.1 and either IRF protein alone. To determine this, the total amount of IRF expression vector was kept constant when comparing PU.1 ϩ IRF2 versus PU.1 ϩ ICSBP versus PU.1 ϩ IRF2 ϩ ICSBP.
We found that reporter expression from the NF1-cis element-containing construct is significantly greater in transfectants with PU.1 ϩ ICSBP ϩ IRF2 in comparison with PU.1 ϩ ICSBP or PU.1 ϩ IRF2 (p Յ 0.002, n ϭ 7; with and without IFN␥) (Fig.  4A). These results suggest that IRF2 and ICSBP are not functionally redundant for their impact on NF1 transcription. In control experiments, none of these proteins significantly influence reporter expression from the empty pTATACAT vector, with or without IFN␥ treatment of the transfectants.
Because these studies involve expressing multiple proteins from multiple vectors, we performed a control experiment to verify that these proteins could be simultaneously overexpressed. U937 cells were co-transfected with vectors to overexpress PU.1, IRF2, and ICSBP/ IRF8 or with an equivalent amount of control vector. Transfectants were incubated with IFN␥, as in the reporter gene assays above. Lysate proteins were separated by SDS-PAGE and analyzed for protein expression by Western blot (Fig.  4B). We found that these proteins are indeed co-overexpressed in U937 cells under these conditions. We also find that this cooverexpression increases expression of endogenous Nf1, consistent with the reporter assays.
PU.1 Is Essential for Activation of the NF1-cis Element in U937 Myeloid Cells-We hypothesize that PU.1 binding to the NF1-cis element is essential for binding of IRF2, which is essential for binding of ICSBP/IRF8. If this hypothesis is correct, decreasing endogenous PU.1 in U937 cells should decrease activity of the NF1-cis element. To investigate this hypothesis, U937 cells were co-transfected with a vector to express a PU.1specific short hairpin RNA (shRNA) or scrambled control shRNA. We found that expression of a PU.1-specific shRNA induces a small but significant decrease in reporter expression from the NF1-cis element in U937 transfectants with and without IFN␥ differentiation (p Ͻ 0.002, n ϭ 3) (Fig. 5A).
Therefore, we next determined the impact of co-overexpressing PU.1-specific shRNA with IRF2, ICSBP/IRF8, or IRF2 ϩ ICSBP/IRF8. This experiment would test our hypothesis that activation of the NF1-cis element by overexpressed IRF2 or ICSBP/IRF8 requires the presence of either endogenous or overexpressed PU.1. We found that expression of PU.1-specific shRNA (but not control scrambled shRNA) blocks activation of the nf1TATACAT reporter construct by overexpressed IRF2, ICSBP/IRF8, or IRF2 ϩ ICSBP/IRF8 (Fig. 5A). Indeed, we found that activity of the NF1-cis element containing construct is not significantly influenced by overexpression of these IRF proteins in the presence of PU.1 shRNA expression (p ϭ 0.61, F ϭ 0.54, n ϭ 3). Expression of PU.1-specific shRNA does not impact reporter expression from pTATACAT control vector. , and ICSBP/IRF8 cooperate to activate the NF1-cis element. U937 cells were co-transfected with an artificial promoter construct with multiple copies of the NF1-cis element (nf1TATACAT) or empty vector control (pTATACAT) and various combinations of vectors to overexpress PU.1, IRF2, and ICSBP/IRF8. Reporter gene assays were performed on transfectants with or without IFN␥ treatment. Overexpression of each of these proteins individually increases nf1TATACAT reporter expression ϩ IFN␥. Reporter expression from nf1TATACAT in transfectants overexpressing the three transcription factors simultaneously was significantly greater than in transfectants with either PU.1 ϩ IRF2 or PU.1 ϩ ICSBP/ IRF8, although the total amount of IRF vector was held constant. In contrast, none of these proteins influenced reporter expression from control pTATACAT. B, overexpressed PU.1, IRF2, and ICSBP/IRF8 increase expression of endogenous Nf1. U937 cells were co-transfected with vectors to overexpress PU.1, IRF2, and ICSBP/IRF8 or with empty control vector. Lysate proteins were isolated after 48 h of IFN␥ treatment, separated by SDS-PAGE, and Western blots serially probed with antibodies to PU.1, IRF2, ICSBP/IRF8, Nf1, and GAPDH (to control for protein loading). Increased PU.1, IRF2, and ICSBP/IRF8 expression correlates with increase in endogenous Nf1 expression.
Based on these results, we also investigated the effect of PU.1 knockdown on expression of endogenous Nf1. For these experiments, U937 transfectants with PU.1-specific or scrambled control shRNA were IFN␥-treated, and cell lysates were analyzed by Western blot (Fig. 5B). We found that decreased PU.1 expression in U937 cells is associated with decreased Nf1 expression. This is consistent with the reporter gene assays.

Mutation of PU.1 PEST Domain Serine Residues Prevents Functional
Interaction with IRF2 and ICSBP-We found that mutation of serine residues 132 and 133 in the PU.1 PEST domain impairs the ability of PU.1 to recruit IRF2 to the NF1cis element in vitro. Based on these results, we tested the impact of S132A/S133A PU.1 on activation of NF1 transcription. The reporter gene assays in this section were performed simultaneously with the experiments in Fig. 4A, but they are presented in separate graphs for convenience of interpretation. U937 cells were co-transfected with the NF1-cis element-containing reporter construct (or control pTATACAT) and a vector to express wild type or S132A/S133A PU.1 (Fig. 6A). We found S132A/S133A PU.1 induces significantly less expression from the NF1-cis element in comparison with wild type PU.1, with and without IFN␥-differentiation (p Յ 0.002, n ϭ 6). Indeed, reporter activity from the nf1TATACAT construct is not significantly different in transfectants with S132A/S133A PU.1 than in transfectants with control expression vector (p ϭ 0.122, n ϭ 6 for undifferentiated transfectants, and p ϭ 0.54, n ϭ 6 for IFN␥-treated transfectants). We hypothesize S132A/S133A PU.1 can bind the NF1-cis element but not recruit the IRF proteins necessary for transcriptional activation. Therefore, we tested the effect of S132A/S133A PU.1 on cooperation with IRF2 for NF1-cis element activation (Fig. 6A).
We also tested the impact of mutating PU.1 serine residues 132 and 133 on activation of the NF1-cis element by overexpressed IRF2 ϩ ICSBP (Fig. 6A). We found significantly less reporter expression from nf1TATACAT in transfectants with S132A/S133A PU.1 ϩ IRF2 ϩ ICSBP than in transfectants with wild type PU.1 ϩ these two IRF proteins. This was true in transfectants both with and without IFN␥-differentiation (p Յ 0.0003, n ϭ 6). Indeed, activity of the NF1-cis element is not significantly different in transfectants overexpressing S132A/S133A PU.1 ϩ IRF2 ϩ ICSBP than in transfectants with empty control expression vector (p Ն 0.7, n ϭ 4).
Our in vitro DNA binding assays indicate that mutation of serine 148 in the PU.1 PEST domain impairs the ability of the DNA-bound PU.1-IRF2 heterodimer to recruit ICSBP/ IRF8. Therefore, we investigated the impact of overexpressing S148A PU.1 on activation of the NF1-cis element. In initial experiments, we found that S148A PU.1 is significantly less efficient at activating this construct than wild type PU.1 in transfectants, with and without IFN␥ (p Յ 0.01, n ϭ 8). These results suggest that activation of this cis element by overexpressed PU.1 involves interaction with both endogenous IRF2 and ICSBP.
We also tested the ability of S148A PU.1 to induce NF1-cis element activation in cooperation with both overexpressed IRF2 and ICSBP/IRF8. We found that nf1TATACAT expression in transfectants with S148A PU.1 ϩ the two IRF proteins is less than half that in FIGURE 6. Specific PU.1 serine residues are required for cooperation with IRF2 and ICSBP and activation of the NF1-cis element. A, overexpressed S132/33A PU.1 does not cooperate with IRF2, ICSBP/IRF8, or IRF2 ϩ ICSBP/IRF8 to activate the NF1-cis element. U937 cells were co-transfected with an artificial promoter construct with multiple copies of the NF1-cis element (nf1TATACAT) or empty vector control (pTATACAT) and various combinations of vectors to overexpress S132A/S133A PU.1 or wild type PU.1 and IRF2 and ICSBP/IRF8. Reporter gene assays were performed with or without IFN␥ treatment of the transfectants. S132A/S33A PU.1 is less efficient at activation of the nf1TATACAT construct than wild type PU.1 alone, or with IRF2, ICSBP/IRF8, and IRF2 ϩ ICSBP/IRF8, with and without IFN␥ treatment. None of these overexpressed proteins significantly influenced reporter expression from control pTATACAT. B, overexpressed S148A PU.1 does not cooperate with IRF2, ICSBP/IRF8, or IRF2 ϩ ICSBP/IRF8 to activate the NF1-cis element. U937 cells were co-transfected with an artificial promoter construct with multiple copies of the NF1-cis element (nf1TATACAT) or empty vector control (pTATACAT) and various combinations of vectors to overexpress S148A PU.1 or wild type PU.1 and IRF2 and ICSBP/IRF8. Reporter gene assays were performed with or without IFN␥ treatment. S148A PU.1 is less efficient at activation of the nf1TATACAT construct than wild type PU.1 alone or with IRF2, ICSBP/IRF8, and IRF2 ϩ ICSBP/IRF8, with and without IFN␥ treatment. None of these overexpressed proteins significantly influenced reporter expression from control pTATACAT. transfectants with wild type PU.1 ϩ IRF2 ϩ ICSBP/IRF8 (p ϭ 0.001, n ϭ 4). This effect is even greater in IFN␥-treated transfectants (p Ͻ 0.00002, n ϭ 6).
Expression from the empty reporter vector pTATACAT is not significantly altered by expression of any of these proteins. Previous studies and our own preliminary data indicate that expression of wild type PU.1 and various serine mutants is equivalent in such transfection experiments (not shown in this study) (26).
Mutation of a Conserved IRF Domain Tyrosine Prevents IRF2 from Activating NF1 Transcription-We found that the conserved IRF domain tyrosine in IRF2 increases interaction with DNA-bound PU.1. Therefore, we investigated the role of this residue (Tyr-109 IRF2) in activation of the NF1-cis element. We performed U937 transfection experiments with the NF1-cis element-containing minimal promoter/reporter construct or pTATACAT control vector and vectors to overexpress various combinations of wild type or Y109F IRF2, ICSBP, and PU.1 (Fig.  7A). The reporter gene assays in this section were performed simultaneously with the experiments in Fig. 4A and Fig. 6, A and B, but are presented in a separate graph for convenience of interpretation.
We initially tested the impact of Y109F IRF2 overexpression alone on the NF1-cis element in the nf1TATACAT reporter construct. In these studies, we found significantly less reporter expression in transfectants with Y109F IRF2 in comparison with wild type IRF2 with and without differentiation (p Յ 0.002, n ϭ 6). We hypothesize that this is because of the inability of overexpressed Y109F IRF2 to interact with endogenous, DNAbound PU.1. This results in an inability to recruit ICSBP/IRF8. Therefore, we next tested the impact of mutating the conserved IRF2 IRF domain tyrosine on cooperation with overexpressed PU.1. We found that Y109F IRF2 ϩ PU.1 induces significantly less reporter expression from the NF1-cis element-containing construct than wild type IRF2 ϩ PU.1 both in untreated and IFN␥-treated transfectants (p Ͻ 0.002, n ϭ 6). Consistent with the hypothesis that overexpressed Y109F IRF2 is unable to interact with overexpressed PU.1, there is no difference in reporter activity in transfectants with PU.1 ϩ Y109F IRF2 and with PU.1 alone (p ϭ 0.64, n ϭ 3 for undifferentiated transfectants and p ϭ 0.25, n ϭ 3 in IFN␥-treated transfectants).
If ICSBP/IRF8 interaction with the NF1-cis element requires pre-binding of PU.1 and IRF2, overexpressed Y109F IRF2 would not be anticipated to increase activation of the NF1-cis element by overexpressed ICSBP/IRF8. Consistent with this, we found that activation of the NF1-cis element in transfectants overexpressing Y109F IRF2 ϩ ICSBP/IRF8 is not significantly different from reporter activity in transfectants overexpressing ICSBP/IRF8 alone (p ϭ 0.83, n ϭ 3 for undifferentiated transfectants and p ϭ 0.69, n ϭ 3 for IFN␥-treated transfectants). This results suggests ICSBP/IRF8 is interacting with only endogenous IRF2 in transfectants overexpressing this IRF2 mutant.
We next investigated the effect on activity of the NF1-cis element of combining Y109F IRF2 with both PU.1 and ICSBP/ IRF8. We found significantly less reporter expression in transfections with Y109F IRF2 ϩ PU.1 ϩ ICSBP in comparison with wild type IRF2 and these other two proteins (p Յ 0.03, n ϭ 6, with and without IFN␥). These results suggest that Tyr-109 in IRF2 is necessary for the overexpressed protein to interact with PU.1. This heterodimer recruits ICSBP/IRF8 and activates transcription.
In control experiments, overexpression of these proteins did not significantly alter reporter expression from the empty minimal promoter control vector. In additional control experiments, we demonstrate that expression of wild type and Y109F IRF2 is equivalent in U937 transfections (Fig. 7B).
Mutation of a Conserved IRF Domain Tyrosine Prevents ICSBP from Activating NF1 Transcription-In our studies above, we found that Tyr-95 in ICSBP/IRF8 is necessary for interaction with the DNA-bound PU.1-IRF2 heterodimer. Also, we previously found that phosphorylation of ICSBP/IRF8 Tyr-95 is essential for activation of NF1-cis element in U937 transfections (15). Therefore, we investigated the role of ICSBP/IRF8-tyrosine phosphorylation in functional interaction with PU.1, IRF2, and the NF1-cis element. These reporter gene assays were performed simultaneously with the experiments in Fig. 4A, Fig. 6, A and B, and Fig. 7A but are presented in a separate graph for convenience of interpretation.
For these studies, U937 cells were co-transfected with the NF1-cis element-containing reporter construct (nf1TATACAT) or control pTATACAT and various combinations of vectors to overexpress wild type or Y95F ICSBP/IRF8, IRF2, and PU.1 (or empty expression vector control) (Fig. 7C). In initial studies, we confirm that Y95F ICSBP/IRF8 induces significantly less reporter activity from the NF1-cis elementcontaining construct than wild type ICSBP, which is consistent with our previous results (15) (p Յ 0.01, n ϭ 6 for transfectants with and without IFN␥).
Therefore, we co-overexpressed PU.1 and wild type versus Y95F ICSBP/IRF8. We found that overexpressed PU.1 cooperates significantly less efficiently with overexpressed Y95F ICSBP/IRF8 than with wild type ICSBP/IRF8 in either untreated or IFN␥-differentiated transfectants (p Յ 0.01, n ϭ 4). We hypothesize that the interaction of overexpressed PU.1 ϩ ICSBP/IRF8 with the NF1-cis element involves interaction with endogenous IRF2. Because Tyr-95 ICSBP/IRF8 is required for interaction with PU.1 ϩ IRF2, we anticipate that nf1TATACAT reporter activity would be the same with overexpressed PU.1 as with PU.1 ϩ Y95F ICSBP/IRF8. Indeed, there is no difference in reporter activity in these transfectants (p ϭ 0.16, n ϭ 3 for undifferentiated and p ϭ 0.46, n ϭ 3 for differentiated transfectants). . Conserved tyrosine residues in the IRF domains of IRF2 and ICSBP/IRF8 are necessary for activation of the NF1-cis element. A, overexpressed Y109F IRF2 does not cooperate with PU.1 and ICSBP/IRF8 to activate the NF1-cis element. U937 cells were co-transfected with an artificial promoter construct with multiple copies of the NF1-cis element (nf1TATACAT) or empty vector control (pTATACAT) and various combinations of vectors to overexpress Y109F or wild type IRF2 and PU.1 and ICSBP/IRF8. Reporter gene assays were performed with or without IFN␥ treatment of the transfectants. Y109F IRF2 induces less reporter expression from nf1TATACAT than wild type IRF2 alone or in combination with PU.1, ICSBP/IRF8, or PU.1 ϩ ICSBP/IRF8 and with and without IFN␥ differentiation. In contrast, none of these proteins significantly influenced reporter expression from control pTATACAT. B, wild type and Y109F IRF2 are equivalently overexpressed in U937 cells. U937 cells were transfected with a vector to overexpress wild type or Y109F IRF2 (or empty vector control). Lysate proteins were separated by SDS-PAGE and serially probed with antibodies to IRF2 and actin (to control for loading). WB, Western blot. C, overexpressed Y95F ICSBP/IRF8 does not cooperate with PU.1 and IRF2 to activate the NF1-cis element. U937 cells were co-transfected with an artificial promoter construct with multiple copies of the NF1-cis element (nf1TATACAT) or empty vector control (pTATACAT) and various combinations of vectors to overexpress Y95F or wild type ICSBP/IRF8 and PU.1 and IRF2. Reporter gene assays were performed on transfectants with or without IFN␥ treatment. Y95F ICSBP/IRF8 induces less reporter expression from nf1TATACAT than wild type ICSBP/IRF8 alone or with PU.1, IRF2, or PU.1 ϩ IRF2 and with and without IFN␥ differentiation. In contrast, none of these proteins significantly influenced reporter expression from control pTATACAT. Therefore, we next tested the impact of ICSBP/IRF8 Tyr-95 phosphorylation on interaction with IRF2. We found that reporter expression from the NF1-cis element-containing constructs is not significantly different in transfectants overexpressing IRF2 ϩ Y95F ICSBP/IRF8 in comparison with transfectants overexpressing IRF2 only (p Ն 0.1, n ϭ 3 for both undifferentiated and differentiated U937 transfectants). Based on these results, we tested the impact of overexpressed Y95F ICSBP/IRF8 on cooperation with overexpressed PU.1 ϩ IRF2. We found expression of Y95F ICSBP/IRF8 ϩ PU.1 ϩ IRF2 induces significantly less expression from the NF1-cis elementcontaining construct than ICSBP/IRF8 ϩ PU.1 ϩ IRF2 in both untreated and differentiated transfectants (p Յ 0.01, n ϭ 4). Indeed, activity of the NF1-cis element is not significantly different in transfectants with PU.1 ϩ IRF2 ϩ Y95F ICSBP/IRF8 than transfectants with PU.1 alone (untreated transfectants p ϭ 0.43, n ϭ 3; IFN␥-treated transfectants p ϭ 0.46, n ϭ 3).
In control experiments, none of these combinations of overexpressed proteins influenced expression of the empty minimal promoter/reporter vector (pTATACAT). In previous control experiments, we demonstrated that wild type and Y95F ICSBP is equivalently overexpressed in U937 cells using this vector (15,16).
PU.1 Is Serine-phosphorylated during IFN␥ Differentiation of U937 Cells-If PU.1 is serine-phosphorylated in response to cytokine stimulation, this would provide a mechanism for increased NF1 transcription during myeloid differentiation. Initially, we determined the impact of IFN␥ differentiation on PU.1 abundance in U937 cells (Fig. 8A). We found that PU.1 protein increases with IFN␥ treatment, and there is a relative increase in a higher molecular weight immunoreactive band. This higher molecular weight band, which is readily detected on 20% acrylamide gels (but not lower % gels), might represent phosphorylated PU.1. The blot was stripped and re-probed for total ERK protein as a loading control and phospho-ERK as a control for IFN␥ signaling.
Therefore, we determined the impact of IFN␥ treatment on PU.1 phosphorylation in U937 cells. U937 cells were incubated for 48 h with or without IFN␥, followed by metabolic labeling with [ 32 P]orthophosphate. Cell lysates were immunoprecipitated under denaturing conditions with an antibody to PU.1 (or control antibody), separated by SDS-PAGE, and visualized by autoradiography. We found that IFN␥ differentiation of U937 cells increases PU.1 phosphorylation (Fig. 8B). This study shows multiple phospho-bands, suggesting multiple phosphorylated forms of PU.1 in these cells.
To determine whether these phospho-residues are serine/ threonine versus tyrosine, nuclear proteins from U937 cells were immunoprecipitated under denaturing conditions with an antibody to PU.1 (or control irrelevant antibody). Immunoprecipitates were separated by SDS-PAGE (7.5% acrylamide), and Western blots were serially probed with antibody to phosphotyrosine and PU.1 (Fig. 8C). We were not able to detect tyrosine-phosphorylated PU.1 by using this approach, consistent with previous results (28). PU.1 appears as a single band in this study because of the percent of acrylamide in the SDS-PAGE.
IRF2 Is Tyrosine-phosphorylated during IFN␥ Differentiation of U937 Cells-In previous studies, we found that IRF1 and ICSBP/IRF8 become tyrosine-phosphorylated during myeloid differentiation. If IRF2 is similarly modified, this might provide an additional mechanism for cytokine-induced Nf1 expression in differentiating myeloid cells. Therefore, we determined the tyrosine phosphorylation state of IRF2 in IFN␥-treated U937 cells. We initially investigated nuclear abundance of IRF2 in differentiating U937 cells (Fig. 8A). We found that total IRF2 protein is not significantly altered by IFN␥ treatment.
To determine the tyrosine phosphorylation state of IRF2 in these cells, U937 nuclear proteins were immunoprecipitated under denaturing conditions with an antibody to IRF2 (or control antibody). Immunoprecipitates were separated by SDS- Nuclear proteins were isolated from U937 cells before and after differentiation with IFN␥ for 48 h. Proteins were separated on SDS-PAGE, and Western blots (WB) were serially probed with antibodies to PU.1 and IRF2. Blots were also probed with antibody to phospho-ERK as a control for IFN␥ treatment and with total ERK as a loading control. Abundance of PU.1 increases with IFN␥ treatment of the cells, associated with the appearance of a dominant higher molecular weight band, representing phosphorylated forms of PU.1. In contrast, nuclear IRF2 abundance is not altered by differentiation of the U937 cells. B, IFN␥ differentiation of U937 cells increases PU.1 phosphorylation. U937 cells were incubated for 0, 24, or 48 h with IFN␥. Cells were metabolically labeled with [ 32 P]orthophosphate and cell lysates immunoprecipitated (IP) under denaturing conditions with antibody to PU.1 or irrelevant control antibody. Immunoprecipitates were separated by SDS-PAGE and phosphoproteins identified by autoradiography. Multiple phospho-forms of PU.1 are identified in IFN␥-treated U937 cells. C, IFN␥ differentiation of U937 cells does not induce tyrosine phosphorylation of PU.1. Nuclear proteins were isolated from U937 cells with or without 48 h of IFN␥ treatment. Proteins were immunoprecipitated under denaturing conditions with antibody to PU.1 (or control antibody). Immunoprecipitates were separated by SDS-PAGE and Western blots serially probed with antibody to phosphotyrosine and PU.1. Although PU.1 abundance increases slightly with in IFN␥-treated cells, no tyrosine-phosphorylated PU.1 is detected by this technique. D, IFN␥ differentiation of U937 cells increases IRF2-tyrosine phosphorylation. U937 nuclear proteins were isolated from cells with or without 48 h of IFN␥ treatment. Proteins were immunoprecipitated under denaturing conditions with an antibody to IRF2 (or control antibody). Immunoprecipitates were separated by SDS-PAGE and Western blots serially probed with antibodies to phosphotyrosine and IRF2. Although total IRF2 protein does not increase, the abundance of phosphorylated IRF2 is increased by IFN␥ treatment of U937 cells.
PAGE and Western blots serially probed with antibodies to phosphotyrosine and IRF2 (Fig. 8D). We found that IRF2-tyrosine phosphorylation increases during IFN␥ differentiation.

DISCUSSION
In previous studies, we determined that Nf1 activity during myelopoiesis is at least partly regulated by NF1 transcription (4,15). We also found that activation of the NF1 gene by hematopoietic cytokines is due in part to phosphorylation of a conserved tyrosine residue in the ICSBP/IRF8 IRF domain (15). In these studies, we determine that PU.1 binds to the NF1-cis element and recruits IRF2. We found that binding of this heterodimer is necessary for interaction of ICSBP/IRF8 with the NF1-cis element. Assembly of this activation complex also requires post-translational modification of all three proteins. Because PU.1 and IRF2 undergo post-translational modification during myelopoiesis, our studies identify an additional mechanism that regulates NF1 transcription in differentiating myeloid cells.
In this study, we note that the NF1-cis element is homologous to composite ets/IRF elements in the CYBB and NCF2 genes. However, we found that there are differences in protein binding to the NF1-cis element in comparison with the homologous cis elements in these oxidase genes. The CYBB-and NCF2-cis elements bind PU.1 and IRF1 in undifferentiated myeloid cells (16,21,22). This interaction provides basal transcription at early stages of differentiation (21). During terminal differentiation, the PU.1-IRF1 heterodimer recruits ICSBP/ IRF8 and the CREB-binding protein to the CYBB and NCF2 promoters (22). Assembly of this complex requires phosphorylation of PU.1 Ser-148 and of the conserved tyrosine residue (Tyr-95) in the ICSBP/IRF8 IRF domain (16).
In contrast, we found no equivalent activation complex binding the NF1-cis element in undifferentiated myeloid cells. In response to differentiating cytokines, an activation complex assembles on the NF1-cis element. Similar to the CYBB-and NCF2-cis elements, assembly of this complex requires initial binding of PU.1. We found that PU.1 recruits IRF2 to the NF1 promoter in a manner that requires phosphorylation of PU.1 serine residues 132 and 133. This interaction also requires the conserved IRF domain tyrosine in IRF2 (Tyr-109). Recruitment of ICSBP/IRF8 to the NF1-cis element by the PU.1/IRF2 heterodimer requires phosphorylation of the conserved tyrosine residue in the ICSBP IRF domain. This provides a specific molecular mechanism for our previous observations of the functional significance of this Tyr residue (15). We also find that recruitment of ICSBP to the NF1-cis element by the PU.1/ IRF2 heterodimer requires phosphorylation of serine 148 in PU.1.
These results suggest different residues in PU.1 are responsible for interaction with different IRF proteins. If these residues were substrates for different kinases (or phosphatases), differential signal-dependent events might regulate the interaction of PU.1 with various protein partners. This could specify PU.1 target gene activation patterns at various stages of differentiation or during the inflammatory response. Additionally, these results suggest there are differences in protein/protein/ DNA interactions between different ets/IRF-cis elements.
Previous studies found that PU.1 and IRF4 bind to a composite ets/IRF sequence in the immunoglobulin ⌲ gene (referred to as the ⌲3ЈE) (29). This interaction requires PU.1 Ser-148 and activates transcription in differentiating B-cells. Similar to the CYBB-, NCF2-, and NF1-cis elements, recruitment of the IRF protein to the ⌲3ЈE requires PU.1 binding. For this cis element, the minimal PU.1-consensus is 3Ј-GGAA-5Ј, consistent with our present results (30). Crystallography studies have been performed to obtain additional information regarding protein interaction with composite ets/IRF sequences. In these studies, the PU.1 ets domain and the IRF4 IRF domain were co-crystallized with the ⌲3ЈE. These studies indicate that the PU.1 PEST domain is sterically available for protein/protein interactions (31). These studies also indicate that the conserved IRF domain Tyr residue in IRF4 is positioned to interact with the PU.1 ets domain. Therefore, these crystallography results are not inconsistent with our model for PU.1/IRF protein interactions.
Based on our studies, we anticipate that interaction of DNAbound PU.1 with one IRF protein creates a binding site for a second IRF protein. We also anticipate that binding of this second IRF protein involves contact with PU.1 and the first IRF (Fig. 9). This conclusion is based on DNA affinity purification studies and on assays in which PU.1 interaction with ICSBP/ IRF8 or IRF2 was studied in the absence of DNA binding. In these studies, PU.1 interaction with ICSBP/IRF8 requires PU.1 Ser-148 and PU.1 interaction with IRF2 requires PU.1 Ser-132/133. We confirmed the model generated from in vitro binding assays with functional assays using transfection of U937 cells. We found that overexpression of these two IRF proteins has a more than additive impact on the NF1-cis element, indicating nonredundant roles for IRF2 and ICSBP/IRF8. These studies also indicate the functional importance of phosphorylation of Ser-132, -133, and -148 of PU.1, Tyr-109 of IRF2, and Tyr-95 of ICSBP/IRF8 for NF1 transcription. These results are consistent with the role of post-translational modification of these pro- Our previous studies suggest that PU.1 bound to the CYBB or NCF2 composite ets/IRF-cis elements interacts with IRF1 and that this heterodimer recruits ICSBP/IRF8 in differentiated myeloid cells. In contrast, our current studies indicate that PU.1 binding to the NF1 composite ets/IRFcis element recruits IRF2 and that this heterodimer recruits ICSBP/IRF8. MARCH 2, 2007 • VOLUME 282 • NUMBER 9 teins in regulating Nf1 expression in differentiating myeloid cells.

ICSBP/IRF8, IRF2, and PU.1 Activate Nf1 Expression
PU.1 is a key transcription factor in myelopoiesis. A number of previously identified PU.1 target genes confer mature phagocyte functional competence (11,21,22,(32)(33)(34)(35)(36). This is consistent with immunodeficiency found in PU.1-deficient murine models (37,38). However, mutations in the gene encoding PU.1 have been found in human AML (39,40). Additionally, studies of leukemia in ␥-irradiated mice documented frequent deletion in the PU.1 gene (41). These results suggest a possible role for PU.1 in leukemogenesis. To investigate this, other investigators generated a conditional PU.1-knock-out mouse (42). These animals develop a myeloproliferative disorder that evolves to AML over several months. These results suggest that PU.1 function is partly regulated by protein abundance. Consistent with this, PU.1 is expressed at low levels in hematopoietic stem cells and is relatively more abundant during monocyte versus neutrophil differentiation (43). However, target genes responsible for dysregulated myeloproliferation in PU.1-deficient cells had not been identified prior to this study.
The role of IRF2 in myelopoiesis and leukemogenesis is less clear. Relatively few IRF2 target genes have been identified in myeloid cells. Previous studies found IRF2 activates transcription of some genes involved in the innate immune response (23, 44 -46). IRF2 also activates the gene encoding histone H4 during G 1 /S transition (5). However, no IRF2 target genes that regulate cellular proliferation or cell cycle progression have been identified previously. Identification of NF1 as IRF2 target genes suggests that IRF2 deficiency might be expected to result in cytokine hypersensitivity. However, IRF2-deficient mice do not exhibit a myeloproliferative disorder (47). It is possible that IRF2 is functionally redundant with another IRF protein for regulation of genes involved in myeloproliferation. Alternatively, it is possible that IRF2 impacts target genes that both upand down-regulate proliferation. If so, the net impact of IRF2 might reflect the protein/protein or protein/DNA interactions in which it participates under a given set of conditions. Therefore, these studies determine that cytokine-induced NF1 transcription involves phosphorylation-dependent assembly of an activation complex that includes PU.1, IRF2, and ICSBP/IRF8. These studies provide a mechanism by which both PU.1 and IRF2 influence proliferation in differentiating myeloid cells. Clarification of such molecular mechanisms may suggest rational therapeutic targets for malignant myeloid disorders.