An ordered array of cold shock domain repressor elements across tumor necrosis factor-responsive elements of the granulocyte-macrophage colony-stimulating factor promoter.

The tumor necrosis factor-alpha-responsive region of the human granulocyte-macrophage colony-stimulating factor (GM-CSF) promoter (-114 to -31) encompasses binding sites for NF-kappaB, CBF, AP-1, ETS, and NFAT families of transcription factors. We show both here and previously that mutation of any one of these binding sites greatly reduces tumor necrosis factor-alpha induction of the GM-CSF promoter. Interspersed between these elements are sequences that when mutated lead to an increase in GM-CSF promoter activity. We have previously shown that two of these repressor elements bind proteins known as cold shock domain (CSD) factors and that overexpression of CSD proteins leads to repression of GM-CSF promoter activity in fibroblasts. CSD proteins are single strand DNA- and RNA-binding proteins that contact 5'-CCTG-3' sequences in the GM-CSF repressor elements. We show here that two newly identified repressor sequences in the proximal promoter can also bind CSD proteins. We have characterized the CSD-containing protein complexes that bind to the GM-CSF promoter and identified a novel protein related to mitochondrial single strand binding protein that forms part of one of these complexes. The four CSD-binding sites on the promoter occur in pairs on opposite strands of the DNA and appear to form an ordered array of binding elements. A similar ordered array of CSD sites are present in the promoters of the granulocyte colony-stimulating factor and interleukin-3 genes, implying a common mechanism for negative regulation of these myeloid growth factors.

Granulocyte-macrophage colony-stimulating factor (GM-CSF) 1 is one of a family of hematopoietic growth factors that control the survival, proliferation, and differentiation of hemopoietic progenitor cells as well as the functional activation of mature cells. GM-CSF functions in particular to regulate he-matopoietic cells of the myeloid lineage. GM-CSF expression is normally tightly regulated, and it is produced by a number of cell types following appropriate stimulation. These include myeloid, mesenchymal (fibroblast and endothelial cells), and lymphoid cells (1)(2)(3)(4). Inappropriate or constitutive expression of GM-CSF is implicated in a number of disease states, including myeloid leukemia, prostate and colorectal cancers, arthritis, and asthma (1,2,(5)(6)(7)(8)(9). Because the GM-CSF gene is primarily regulated at the level of transcription, it is important to investigate the mechanisms of repression as well as activation of this gene. Such studies are necessary to define the means by which the GM-CSF gene is maintained in a strictly silent state in the absence of stimulation and also to determine the mechanisms of rapid derepression of the gene and subsequent activation upon stimulation. This will then allow identification of defective regulatory pathways in diseases where GM-CSF disregulation is important.
Numerous transcriptional activators bind to and regulate the proximal GM-CSF promoter, which can be divided into two functional domains (see Fig. 1). Domain 1 (Ϫ114 to Ϫ71) contains the CK-1 and CK-2 elements conserved in a number of cytokine genes and binds a number of transcription factors including NF-kB, Sp1, and the CD28-responsive complex (1)(2)(3)(4). This region is responsive to T cell receptor activators and costimulators (10 -12), is involved in response to TNF-␣ in fibroblasts (13,14), and is required for constitutive expression in juvenile myelomonocytic leukemia cells (15). The NF-kB site is critical for expression in these cell types. Domain 2 (Ϫ70 to Ϫ31) binds CBF, AP1, ETS, and NFAT transcription factors (10, 16 -18). This region responds to TNF-␣ and interleukin (IL)-1 stimulation of fibroblast (13,14) and endothelial (19) cells and T cell receptor activation (10, 16 -18) and is required for constitutive expression in certain acute myeloid leukemia cell lines (20). The CBF, AP1, and ETS/NFAT transcription factor-binding sites have been shown to be essential for T cell receptor signaling in conjunction with the domain 1 NF-kB site (10,16,18), but it is not known which of these sites in domain 2 are required for activation in fibroblasts and endothelial cells.
In addition to the activators described above, we have identified nuclear complexes called NF-GMa, NF-GMb, and NF-GMc that bind to domain 1 of the GM-CSF promoter (13,14,21,22). The NF-GMa complex was found to also bind to the CK-1/ CK-2 equivalent regions of the genes for two other myeloid growth factors, granulocyte colony-stimulating factor (G-CSF) and IL-3 (23,24). The GM-CSF gene is coordinately regulated with the G-CSF gene in fibroblasts (25) and with the IL-3 gene in T cells (2), respectively, in response to certain stimuli. This complex was found to be TNF-␣-inducible in fibroblasts and was implicated in G-CSF promoter activation (23,24,26). The protein composition of this complex was not determined. In contrast, the NF-GMb/c complexes are implicated in repression. These complexes bound to two repressor sites in domain 1 that were functional in fibroblasts (3,4,13,14). NF-GMb contains two separate complexes, one containing a 42-kDa protein and the other containing a dimer of a 22-kDa protein, whereas NF-GMc represents the binding of a single 22-kDa protein. We identified these proteins as cold shock domain (CSD) proteins (4,(27)(28)(29)(30)(31)(32) by cloning of factors contacting the repressor elements and by subsequent analysis of the NF-GMb/c complexes (4,14). An interesting property of these proteins is that they bind to single-stranded DNA and in the case of the GM-CSF sites, to two repeated 5Ј-CCTG-3Ј elements on the noncoding (Ϫ) strand of domain 1 (13,14). CSD factors are expressed in all cell types, and consistently we have observed NF-GMb/c complexes in all cell types examined, including fibroblasts, endothelial cells, T cells, and myeloid cells (13,14). 2 CSD factors in addition to binding to single strand DNA can bind to double strand DNA and RNA. By virtue of their varied binding activities, these proteins are observed to be involved in transcriptional repression and activation and also in translational regulation (27)(28)(29)(30)(31)(32). In particular CSD factors appear to play a role in the strict regulation of expression of genes involved in growth regulation and stress responses (13,14,29,(33)(34)(35)(36). Analysis of CSD protein function on hematopoietic growth factor genes is at present restricted to the GM-CSF gene, where we found that overexpression of recombinant CSD proteins led to repression of domain 1 activity (14). Surprisingly, overexpression of CSD proteins was also shown to directly repress domain 2 in the absence of a similar arrangement of CSD-binding sites (14). The reason for this repression was unknown.
We now report the identification of two new CSD sites across domain 2 of the GM-CSF promoter that function as repressor elements in fibroblasts. We also define the TNF-␣-responsive sequences in domain 2 and find that they flank the CSD sites. The CSD-binding sites across domains 1 and 2 form an ordered regularly spaced array of repressor elements across the entire TNF-␣-responsive proximal GM-CSF promoter. We also find a similar array of CSD sites across the promoter regions of the G-CSF and IL-3 genes. We performed a detailed analysis of the protein composition of NF-GMb/c and NF-GMa complexes and find that distinct nuclear complexes bind to the different CSD sites across the GM-CSF promoter and determine that NF-GMa is also composed of CSD proteins. We propose mechanisms by which the different CSD complexes regulate growth factor promoter expression.

EXPERIMENTAL PROCEDURES
Plasmid Constructs -The human GM-CSF promoter constructs pGM41 and pGM43 have previously been described and were constructed by cloning the oligonucleotides GM41 (Ϫ65 to Ϫ31) and GM43 (Ϫ114 to Ϫ31), respectively, into the pBLCAT2 reporter vector (13). The construct pGM93 (Ϫ70 to Ϫ31) and the mutant constructs pGMm89, pGMm87, pGMm81, pGMm85, and pGMm95 were constructed by cloning respective oligonucleotides (with HindIII 5Ј and BamHI 3Ј ends) into pBLCAT2. Oligonucleotide sequences are shown in Fig. 1c. The bacterial expression vector pGEXBT contains the large EcoRI fragment from the B5 gt11 DbpB CSD cDNA expression clone (14) inserted into pGEX-4T-1 (Promega). This construct expresses a protein lacking the last 10 amino acids of DbpB CSD protein. It has previously been demonstrated that these last 10 amino acids do not affect recombinant DbpB binding to single strand DNA (37,38).
Oligonucleotides and Probe Preparation-All oligonucleotides were synthesized on an Applied Biosystems model 381A DNA synthesizer. Full-length oligonucleotides for retardations or cloning into reporter vectors (see Fig. 1, b and c) were purified from nondenaturing polyacryl-amide gels (39). Single strand DNA probes for gel retardation assays were prepared by end-labeling coding (ϩ) or noncoding (Ϫ) strand oligonucleotides with [␥-32 P]ATP and T4 polynucleotide kinase followed by gel purification.
Preparation of Recombinant Protein-The Escherichia coli strain JM109 transformed with pGEXBT was induced with isopropyl-1-thio-␤-D-galactopyranoside to produce recombinant GST-DbpB fusion protein, which was purified on glutathione-Sepharose beads as described by the manufacturer (Promega).
Preparation of Nuclear Protein, Affinity Purification, and Protein Sequencing-Crude nuclear extracts were prepared from HUT78 T cells as previously reported by us for extraction of NF-GMb/CSD complexes (13,21,22). Extracts contain NF-GMb/c and NF-GMa binding activity (see Fig. 5a). Crude HUT 78 nuclear extracts were heparin-Sepharose (HS) enriched for either NF-GMb/c or NF-GMa complexes as described previously (22). HS fractions enriched for NF-GMb/c (HSGMb; see Figs. 3 and 5) contained no detectable NF-GMa, and conversely fractions enriched for NF-GMa (HSGMa; see Fig. 5) were free of NF-GMb/c (13,21,22). For affinity purification concentrated HSGMa protein in 0.5ϫ TM buffer (22) containing a final concentration of 200 mM KCl and 10 g/ml poly(dI-dC) was applied to a 1-ml DNA affinity column. DNA affinity chromatography was carried out as described (40) except that the ligated oligonucleotides were coupled to Affi-Gel 15 matrix (Bio-Rad) in 0.1 M Hepes, pH 7.5. The oligonucleotide contains the IL-3 CK-1/CK-2 region, which is homologous to the GM-CSF CK-1/CK-2 domain 1 region (24). Of the three myeloid growth factor genes, GM-CSF, G-CSF, and IL-3, we previously found that the IL-3 region has the best affinity for NF-GMa (24). Specifically bound protein was eluted from the column with TM buffer containing 1 M KCl and rerun on a second affinity column. Protein eluates were monitored by both gel retardation assays and SDS-polyacrylamide gel electrophoresis. The 16-kDa affinity purified protein was transferred to polyvinylidene fluoride membrane, eluted, and analyzed by microsequencing (R. Simpson, Walter and Eliza Hall Institute, Melbourne, Australia).
Gel Retardation Analysis and UV Cross-linking-Gel retardation assays were performed using 0.25 ng of single strand 32 P-labeled oligonucleotide probe in a 10-l reaction mix of 0.5ϫ TM buffer (13,14,22) containing 200 mM KCl, 0.4 g of poly(dI-dC) and either 0.2 g of HS-enriched extract (HSGMa or HSGMb), 1.0 g of crude nuclear extract, 25 ng of recombinant CSD fusion protein (GST-DbpB), or 1 ng of affinity purified material. Retardation assays using recombinant protein also contained 2 g of bovine serum albumin. Reactions were incubated at room temperature for 20 min and analyzed on 12% nondenaturing polyacrylamide gels in 0.5ϫ TBE (21). Competition with unlabeled single strand oligonucleotides was performed by addition of protein and unlabeled probe, followed by immediate addition of the 32 P-labeled probe (14). For UV cross-linking, crude nuclear extracts were bound to 32 Plabeled single strand DNA probes in a 25-l retardation reaction and fractionated on a polyacrylamide gel as described above. The gel was exposed to UV light (340 nm) for 15 min, and retarded complexes were excised after exposure to x-ray film. Protein in excised bands was analyzed on 12% SDS-polyacrylamide gels (13,39).
Cell Culture and Transfections-Human embryo lung fibroblasts (Commonwealth Serum Laboratories) were grown in Dulbecco's modified Eagle's medium and 10% fetal calf serum. These cells were used for passages 14 -20 in all experiments. Human embryo lung fibroblasts were cotransfected with 15 g of reporter constructs using DEAEdextran as described (13,14). 24 h following transfection, cells were stimulated with TNF-␣ (100 units/ml) or left untreated for an additional 24 h. Cells were then harvested and CAT assays were performed (13,14). The percentage of [ 14 C]chloramphenicol conversion to acetylated forms via CAT activity in extracts was determined using PhosphorImager analysis (Molecular Dynamics).

Identification of Overlapping TNF-responsive Elements and
Repressor Elements in the GM-CSF Domain 2 Region-We previously reported that a domain 2 reporter construct (pGM41, Ϫ65 to Ϫ31; Ref. 14) was responsive to TNF-␣ in fibroblasts and that this activity was repressed by overexpression of the CSD proteins, DbpB and DbpA, that were cloned as GM-CSF domain 1 repressor site-binding proteins (14). To define the sequences responsible for activation and repression, mutations were made in the pGM41 construct (Fig. 1c) and transfected into human embryo lung fibroblasts, and cells were treated with TNF-␣ or left untreated (Fig. 2). Mutations included specific base changes previously shown to disrupt CBF, AP1 and ETS/NFAT binding. Mutation of the CBF, AP-1, or ETS/NFAT elements reduced both basal and TNF-␣-induced activity (Fig. 2). A repressor element was also identified. Mutation of a 5Ј-CC-3Ј (pGMm87) within a 5Ј-ACCA-3Ј sequence located between the CBF and AP1 sites resulted in a 50% increase in basal and TNF-␣-induced expression. An extended construct (pGM93) containing an extra five bases with a 5Ј-CCTG-3Ј sequence identical to the domain 1 CSD-binding sites was also analyzed (Fig. 2). The extended sequences caused a decrease in TNF-␣-inducible and basal expression. Mutation of the 5Ј-CCTG-3Ј element in this extended construct (pGM95) restored promoter expression, identifying this site as a second repressor element. Hence domain 2 contains at least three sites required for TNF-␣ response, and these are overlapped/flanked by two repressor elements.
Nuclear and Recombinant CSD Proteins Bind to Repressor Elements across Domain 2 on the Opposing Strand to CSDbinding Sites Identified on Domain 1-To determine whether the repressor elements identified were CSD-binding sites, domain 2 single strand wild type and mutant oligonucleotides ( Fig. 1c and Refs. 13 and 14) were analyzed in gel retardation assays for binding with HUT78 T cell extracts enriched for NF-GMb and NF-GMc CSD-containing nuclear complexes (HS-GMb) (Fig. 3a). Binding was compared with the GM-oligonucleotide (Fig. 3a, lane 1), which contains the noncoding (Ϫ) strand of the GM-CSF domain 1 CK-1/CK-2 region (Ϫ114 to Ϫ79) (Fig. 1b) and supports NF-GMb and NF-GMc complex formation. NF-GMb complex formation represents protein binding to both the 5Ј-CCTG-3Ј CSD repressor sites, whereas NF-GMc represents binding to only one site on the GM-oligonucleotide (Fig. 3a, lane 1) (13,14). The domain 2 GM93 coding (ϩ) strand oligonucleotide containing the two newly identified repressor elements supports both NF-GMb and NF-GMc-like complex formation while the GM41(-65 to -31) and GMm95 (-70 to -31; 5Ј repressor site mutated) coding (ϩ) strand oligonucleotides, containing only one repressor element, form only the NF-GMc-like complex (Fig. 3a, lanes 11, 7, and 15, respectively). No complex formation was observed on noncoding (Ϫ) strand domain 2 sequences (data not shown). Competition assays verified that the complexes forming on domain 2 were authentic NF-GMb/c complexes. As shown in Fig. 3a, the NF-GMb/c complexes formed on domain 2 coding (ϩ) strand oligonucleotides (GM41ϩ, GM93ϩ, and GMm95ϩ) were competed to a much greater extent by the wild type GM-CSF domain 1 noncoding (Ϫ) strand oligonucleotide (GMϪ) (lanes 8, 12, and 16) than by the GMm23Ϫ oligonucleotide (lanes 9, 13, and 17) containing mutations in both the CK-1/CK-2 region NF-GMb/ CSD sites (13,14). Consistent with these results the NF-GMb/c complexes on GMϪ were readily competed with the domain 2 oligonucleotides (lanes 4 -6). These data mapped one NF-GMb/c site to the 5Ј repressor site in domain 2 and the other to the Ϫ65 to Ϫ31 region containing the 3Ј repressor site.
Subsequent analysis of mutants in the GM41ϩ sequence (Ϫ65 to Ϫ31) mapped the second NF-GMb/c site to the 3Ј FIG. 1. Human GM-CSF proximal promoter oligonucleotide sequences. a, the GM-CSF proximal promoter is shown diagrammatically. The relative locations of conserved CK-1 and CK-2 elements and transcription factor-binding sites are indicated. The domain 1 (Ϫ114 to Ϫ71) and domain 2 (Ϫ70 to Ϫ31) regions are marked. The NF-kB site and the domain 2 region are responsive to TNF-␣ in fibroblasts (13,14). The NF-kB, CBF, AP1, and ETS/NFAT sites are required for T cell receptor signaling, and the CD28-responsive complex site is required for T cell costimulatory signaling (3,4). b, the sequence of coding (ϩ) and noncoding (Ϫ) strand wild type domain 1 CK-1/CK-2 region (Ϫ114 to Ϫ79) oligonucleotides (GMϩ and GMϪ, respectively) are shown (13,14). Sequences required for nuclear (NF-GMb/c) and recombinant CSD factor binding to the noncoding (Ϫ) strand are marked with asterisks. These sequences are repressor elements in fibroblasts (13,14). Base changes in the mutant GMm23 oligonucleotides are shown (13,14). c, the sequence of the wild type coding (ϩ) strand of domain 2 (Ϫ70 to Ϫ31) is given with CSD (data presented here), CBF, AP1, and ETS/NFAT sites (3, 4) marked. Nuclear and recombinant CSD binding is exclusive to the coding (ϩ) strand. Coding (ϩ) strand oligonucleotide sequences are listed under the domain 2 sequence. Only those bases that vary from the wild type sequence are indicated.
repressor site, identified above, as mutation of the 5Ј-CC-3Ј within the 5Ј-ACCA-3Ј repressor element (GMm87ϩ) abolished complex formation (Fig. 3b, lane 4). This type of binding site for CSD factors has only been reported for a viral gene (41,42) and has not been reported in a genomal gene. No other mutations affected NF-GMc complex formation, but there were shifts in mobility on the different mutant sequences. Competition with the wild type GM41ϩ oligonucleotide indicated that the complexes forming on these mutant sequences were authentic NF-GMc complexes (data not shown). This suggests therefore that NF-GMc may have a different conformation on the different mutant sequences and that the way NF-GMc complex formation occurs depends not only on the complexes binding site but also on the nature of surrounding sequences. To further confirm the requirement for both repressor elements in the formation of NF-GMb/c, mutations were made in one or the other or both sites across the extended GM93ϩ sequence (Ϫ70 to Ϫ31) (Fig. 3c). Mutation of any of the sites (GMm95ϩ, GMm103ϩ, or GMm107ϩ) resulted in loss of the NF-GMb complex but did not result in loss of the NF-GMc complex on GMm95ϩ (5Ј-CCTG-3Ј site mutated), and caused some reduction in NF-GMc on GMm103ϩ and GMm107ϩ (5Ј-ACCA-3Ј mutated) (lanes 2-4). Mutation of both sites (GMm105ϩ and GMm109ϩ) resulted in loss of all complex formation (lanes 6 and 7). These data are consistent with the way we observed NF-GMb/c complex formation on the domain 1 region where NF-GMb complex formation requires both sites for binding, whereas NF-GMc complexes can form on either CSD site (13,14).
We have previously demonstrated the binding of recombinant CSD protein to domain 1 NF-GMb/c sites (14). Domain 2 oligonucleotides were also tested for binding of recombinant CSD (Fig. 3d). As for nuclear NF-GMb/c CSD-containing complexes, recombinant GST-DbpB CSD protein binds exclusively to the coding (ϩ) strand of domain 2 (GM41ϩ and GM93ϩ; lanes 3 and 6) and requires the NF-GMb/c sites for binding. As shown, mutation of the single 5Ј-ACCA-3Ј binding element in the GM41ϩ oligonucleotide essentially abolished CSD binding (GMm87ϩ; Fig. 3d, lane 5). When the longer fragment containing two NF-GMb/c sites (GM93ϩ) was used in binding, mutation of the individual repressor elements (GMm95ϩ and GMm103ϩ) reduced binding, whereas mutation of both sites (GMm105ϩ) completely abolished binding (Fig. 3d, lanes 8 -10). Consistent with our observations for nuclear NF-GMb/c complexes, the binding of recombinant CSD protein to the mutant sequences results in altered mobility. Binding data for nuclear and recombinant CSD protein are summarized (Fig.  3d). Therefore, as observed for domain 1, we have shown that domain 2 contains a pair of repressor elements that bind both nuclear and recombinant NF-GMb/CSD factors, and we have identified a novel 5Ј-ACCA-3Ј NF-GMb/CSD-binding site.  (Fig. 4a). To determine which of the NF-GMb/CSD sites binds this new protein, UV cross-linking was performed using mutants in domain 2. This revealed that the 22-kDa protein bound to the 5Ј NF-GMb/CSD site in domain 2 (GMm103ϩ), whereas the 25-kDa protein bound to the 3Ј site (GMm95ϩ) (Fig. 4b, lane 3 and 4, respectively). Hence the distal three NF-GMb/CSD sites across the GM-CSF promoter with a 5Ј-CCTG-3Ј consensus bind the 22-kDa protein, whereas the newly identified 5Ј-ACCA-3Ј site binds the novel 25-kDa protein. In contrast, the 42-kDa protein, as we have previously shown on domain 1 (13), requires both domain 2 sites for binding (Fig. 4b, compare lanes 1, 3, and 4). Data are summarized diagrammatically in Fig. 4b. The nature of the 25-kDa protein is not known, but we have confirmed that this protein is a CSD factor by competition of the GMm95ϩ complex with a CSD polyclonal antibody. 2 Given the differences in CSD protein composition of complexes binding to 5Ј-CCTG-3Ј and 5Ј-ACCA-3Ј CSD sites, it is of interest that the GM-oligonucleotide (binding 42-and 22-kDa proteins) can compete for the GMm95ϩ complex (25 kDa). This is most probably due to the ability of all nuclear CSD-binding oligonucleotides to bind all CSD subtypes at the amounts of competitor required to observe a competition effect. That this is possible is suggested by experiments competing the GMm95ϩ complex (25 kDa) with an oligonucleotide that only binds the 22-kDa protein (GMm103ϩ). We found that over a wide range of competitor amounts GMm103ϩ could compete for GMm95ϩ complex for- FIG. 2. Identification of TNF-responsive elements and repressor elements in the GM-CSF domain 1. Wild type (pGM41 and pGM93) and mutant (pGMm89, pGMm87, pGMm81, pGMm85, and pGMm95) GM-CSF promoter reporter constructs were transfected into human embryo lung fibroblasts, followed by treatment with (ϩ) or without (Ϫ) TNF-␣ and CAT activity determined. CAT activity levels (average of at least three experiments) relative to unstimulated pBLCAT2, given as 1.0, are shown. Promoter constructs are shown diagrammatically. Repressor and activator elements are marked with boxes and circles, respectively. Sequences contained within promoter constructs are given in Fig. 1c. mation but that this competition was less efficient than competing with the self GMm95ϩ sequence. The other explanation for cross-competition between sequences binding the different 25-and 22-kDa CSD proteins is that the 25-kDa protein represents the binding of the 22-kDa protein to the 5Ј-ACCA-3Ј sequence in altered conformation relative to the way it binds to the 5Ј-CCTG-3Ј sequence. This is feasible given the mobility shifts seen for nuclear NF-GMb/c and recombinant CSD complexes observed on different domain 2 mutant sequences as discussed above (Fig. 3).

Analysis of Nuclear NF-GMb/c CSD-containing Nuclear Complexes Reveals a Novel 25-kDa Protein
NF-GMa Represents a Higher Order Complex of CSD Proteins Binding to Single Strand DNA-Studies performed above used extracts that were heparin-Sepharose enriched for NF-GMb/c binding activity. We have found that binding of crude extract to the domain 1 noncoding (Ϫ) strand oligonucleotide (GMϪ) reveals, in addition to NF-GMb/c, the binding of a more slowly migrating complex that we previously called NF-GMa (Fig. 5a, lane 1) (21,22). This complex also binds to the CK-1 regions of two other myeloid growth factor genes, G-CSF and IL-3, and it was found to be TNF-␣-inducible in fibroblasts (23,24,26). As for NF-GMb/c, NF-GMa binding activity could be enriched from crude extracts and separated from NF-GMb/c activity by heparin-Sepharose chromatography (21,22). Given the overlapping binding sites of NF-GMb/c and NF-GMa complexes, it was possible that NF-GMa was a higher order complex of CSD proteins or that it could act as a competitor for NF-GMb/c binding. To determine the relationship of NF-GMa and NF-GMb/c, NF-GMa was further investigated.
Competition experiments were carried out to determine the requirements for NF-GMa binding to the domain 1 CK-1/CK-2 region GMϪ oligonucleotide. As shown in Fig. 5a, the NF-GMa complexes from crude nuclear extracts (lanes 1-5) or extracts enriched for NF-GMa (HSGMa; lanes 6 -10) were competed by a control CSD-binding site oligonucleotide from the coding (ϩ) strand of the HPV18 enhancer (HPVϩ; lanes 4 and 9). This oligonucleotide competes for NF-GMb/c complexes and has been shown to be a good binding site for recombinant CSD proteins (14,43). The NF-GMa complex is not competed by a sequence (DR␣ϩ; lanes 5 and 10) that we and others have found is unable to bind nuclear or recombinant CSD factors (14,37,38). As for NF-GMa, the NF-GMb/c complexes are not competed by this sequence (Fig. 5a, lane 5) (14). Even though NF-GMa and NF-GMb/c complexes show the same binding characteristics to the control CSD-binding sequence (HPVϩ), the NF-GMa complex does not appear to require the NF-GMb/ CSD-binding sites defined in the GMϪ sequence. This is demonstrated by the ability of the GMm23Ϫ mutant (both 5Ј-CCTG-3Ј sites mutated) to compete for NF-GMa complex formation (Fig. 5a, lanes 3 and 8) but not for NF-GMb/c complexes as described previously (Fig. 5a, lane 3, and Refs. 13 and 14); hence NF-GMa readily binds this sequence.
UV cross-linking revealed that the NF-GMa complex formed from both crude nuclear extract (crude) and extract heparin-Sepharose enriched for NF-GMa (HSGMa) on the domain 1 GM-oligonucleotide, contained 42-and 22-kDa proteins, simi-lar in size to that observed in the NF-GMb complex, as well as a new 16-kDa protein (Fig. 5b, lanes 1-3). The NF-GMa complex formed on GMm23Ϫ was also analyzed by UV crosslinking (Fig. 5b, lane 4). NF-GMa on the mutant sequence contains the 42-and 16-kDa proteins but not the 22-kDa protein. This implies that the 42-kDa protein forms part of the NF-GMa complex without the need for binding to the CSD repressor sites but that the 22-kDa protein is dependent on these sites. These results imply that at least three CSD-containing complexes, NF-GMa, NF-GMb, and NF-GMc, can form on the GM-CSF domain 1 each with distinct sequence requirements for binding. A similarly migrating complex to NF-GMa was observed on domain 2, but cross-linking revealed the presence of a single 36-kDa protein (data not shown). The NF-GMa complex is therefore specific to the domain 1 CK-1/CK-2 region and does not form on domain 2.
As well as changing nuclear CSD factor DNA binding characteristics, it was possible that NF-GMa could compete with NF-GMb/c repressor complexes for binding to domain 1. This was examined in a retardation assay where addition of increasing amounts of heparin-Sepharose enriched NF-GMa (HSGMa) to enriched NF-GMb/c (HSGMb) resulted in inhibition of NF-GMb/c complex formation (Fig. 5c). This effect was prevented by inclusion of a competing GMm23Ϫ oligonucleotide that will titrate out NF-GMa but not NF-GMb/c complexes (data not shown). Hence the effect on the NF-GMb/c complexes was due specifically to the addition of NF-GMa. It appears therefore that NF-GMa may play a dual role in modulating CSD repressor function by competing with repressive NF-GMb/c complexes for binding to repressor elements and by altering the DNA binding characteristics of the 42-kDa protein so that it no longer contacts the domain 1 repressor elements.
Identification plex, DNA affinity chromatography was performed. To do this HUT78 T cell nuclear extract enriched for NF-GMa complex formation (HSGMa) was subjected to two rounds of affinity purification using the IL-3 CK-1/CK-2 region as a target for NF-GMa complex formation. The IL-3 CK-1/CK-2 region shares homology with the GM-CSF domain 1 CK-1/CK-2 region (see Fig. 7, a and b) and was shown to have a higher affinity for NF-GMa (24). The specifically bound protein was eluted in 1.0 M KCl. NF-GMa activity was assayed by gel retardation using the GM-CSF domain 1 noncoding (Ϫ) strand oligonucleotide. As shown in Fig. 6a the affinity purified material forms a complex (lane 2) that migrates at the same position as the NF-GMa complex formed from binding of the heparin-Sepharose purified material (HSGMa) (lane 1). Consistent with this complex being authentic NF-GMa, the affinity purified complex was competed by the GM-, GMm23Ϫ, and HPVϩ sequences but not by the DR␣ sequence (compare Fig. 6a, lanes 3-6, with Fig.  5a). The complex also contained the appropriate 42-, 22-, and 16-kDa proteins expected for NF-GMa as determined by UV cross-linking (Fig. 6b). Analysis of second round affinity purified NF-GMa on an SDS-polyacrylamide gel revealed the presence of a 16kDa protein band after silver staining (Fig. 6c, lane  4). The 42-and 22-kDa proteins were, however, not visible. This protein was isolated from the gel and prepared for microsequencing. A sequence of 10 amino acids was obtained from the N terminus of the protein (Fig. 6d). Data base searches revealed that the sequence matched the N-terminal sequence of mature human mitochondrial single strand binding protein (mtSSB) (44). The mature human mtSSB binds to single strand DNA and is similar in size (15.2 kDa) to the 16-kDa component of the purified NF-GMa complex (44,45). SSB proteins have also been shown to be able to both homodimerize and heterodimerize (44, 46 -48). Gel filtration chromatography of heparin-Sepharose-enriched NF-GMa (HSGMa) under native conditions showed that the protein in the NF-GMa complex had an apparent molecular mass of 62 kDa (data not shown). We have previously determined that the NF-GMb complex represents a mixture of two different types of complex, one containing a single 42-kDa protein and the other containing a pair of 22-kDa proteins (13,14). A molecular mass of 62 kDa is therefore consistent with the NF-GMa complex being composed of a single 42-kDa protein with a single 16-kDa protein (58 kDa) or a pair of 22-kDa proteins with a single 16-kDa protein (60 kDa). These results now implicate a second single-stranded binding protein, the mtSSB or a related protein, together with the CSD proteins, in the complexes that can bind to the domain 1 CK-1/CK-2 region of the GM-CSF gene.

Recombinant and Nuclear CSD Proteins Bind to the G-CSF and IL-3 Myeloid Growth Factor
Genes-Two other myeloid growth factor genes, G-CSF and IL-3, share conserved regulatory elements with GM-CSF, such as the CK-1 region, and have overlapping patterns of regulation (2,(23)(24)(25). These genes also bind the NF-GMa complex (23,24). Given this we examined the proximal promoter sequences of these genes for CSD-binding sites (Fig. 7a). Alignment of the domain 1 CK-1-containing regions did not show the repeated 5Ј-CCTG-3Ј elements, found on the noncoding (Ϫ) strand of the GM-CSF sequence, although there was a pair of CSD-like sites (Fig. 7, a and b). However, closer to the transcription start site a 5Ј-CCTG-3Ј/5Ј-ACCA-3Ј pair of potential CSD-binding sites identical to those on the domain 2 coding (ϩ) strand of the GM-CSF promoter were observed in both G-CSF and IL-3 promoters (Fig. 7, a and b). Single-stranded oligonucleotides spanning these sequences were tested for their ability to form NF-GMb/c-like complexes in gel retardation assays. As expected coding (ϩ) strand oligonucleotides spanning the conserved 5Ј-CCTG-3Ј/5Ј-ACCA-3Ј domain 2-like sequences from both G-CSF (D2) and IL-3 (D2) formed complexes that comigrated with NF-GMb and NF-GMc when incubated with HUT78 nuclear extracts enriched for NF-GMb/c (HSGMb) (Fig. 7b, lanes 7 and 13). The complexes formed with similar intensity to those formed on the GMϪ oligonucleotide (lane 1). Consistently these complexes were competed to a much greater extent by the GMϪ wild type oligonucleotide (lanes 8 and 14) than by the NF-GMb/CSDbinding site mutant oligonucleotide, GMm23Ϫ (lanes 9 and 15), suggesting that these complexes are authentic NF-GMb/c. The G-CSF domain 1 noncoding (Ϫ) strand oligonucleotide (D1) formed an apparent NF-GMc-like complex and a weak NF-GMb-like complex, whereas the IL-3 domain 1 oligonucleotide (D1) formed both complexes weakly (Fig. 7b, lanes 4 and 10). These complexes were also shown to be authentic by competition assays (lanes 5, 6, 11, and 12). The decreased complex formation on the domain 1 sequences is consistent with the reduced conservation of potential NF-GMb/CSD-binding sites in the G-CSF and IL-3 genes. The presence of both NF-GMb a, HUT78 T cell nuclear extracts enriched for NF-GMa (HSGMa) and affinity purified NF-GMa (affiGMa) were bound to labeled GM-CSF domain 1 noncoding (Ϫ) strand oligonucleotide (GMϪ) and assayed by gel retardation. The apparent NF-GMa complex formed using affinity purified NF-GMa material was competed with itself (GMϪ), the CSD site mutant (GMm23Ϫ), the control CSD-binding site (HPVϩ; Refs. 14 and 43), and the nonspecific DRa coding (ϩ) strand oligonucleotides (Refs. 14 and 79 and Fig. 5). A minus sign indicates no competitor. b, the NF-GMa complexes from crude and affinity purified (affiGMa) HUT78 extracts were analyzed by UV cross-linking. The size of cross-linked proteins is indicated. c, SDS-polyacrylamide gel electrophoresis of protein fractions from different steps of the NF-GMa purification. Tracks were loaded with crude, heparin-Sepharose (HSGMa), first round affinity (affi1 GMa), or second round affinity (affi2 GMa) material. Protein was visualized by silver staining. Positions of molecular mass markers are shown. The 16-kDa protein is indicated. d, a comparison of the N-terminal amino acid sequences of the mature human mtSSB and purified 16-kDa protein are shown. Because the nature of the first amino acid from the N terminus of the 16-kDa protein could not be determined, the presented sequence commences at the second amino acid of both the 16-kDa and mtSSB proteins. and NF-GMc complexes forming on the G-CSF and IL-3 oligonucleotides, however, suggests the presence of a pair of CSD sites in each domain as predicted. Consistent with these results, recombinant CSD fusion protein (GST-DbpB) also binds to the G-CSF and IL-3 domain 1 and domain 2 oligonucleotides with binding being strongest to the domain 2 regions (Fig. 7c). Given the conservation of a specific arrangement of CSD sites across the promoters of three myeloid growth factor genes, it is apparent that these genes may be subject to a common mechanism of repression and that the spatial arrangement of CSD sites is important to bring about this repression.

Common GM-CSF Proximal Promoter Elements Respond to Appropriate Signals in T Cells and
Fibroblasts-In analysis of the human GM-CSF promoter in fibroblasts, we previously demonstrated the involvement of the domain 1 NF-kB site and domain 2 sequences in response to TNF-␣ (13,14). We now show that the CBF, AP1, and ETS/NFAT sites (Fig. 1) are absolutely required for TNF-␣ response of domain 2. The NF-kB, CBF, AP1, and ETS/NFAT sites across domains 1 and 2 have been shown to be required for maximal activation in T cells in response to T cell receptor signals (3, 4, 10 -12, 16 -18). Extensive studies have not been performed regarding functional GM-CSF promoter elements in other cell types. Deletion studies in endothelial cells suggest a role for all three binding sites in domain 2 for IL-1 response (19), whereas mutation studies suggest that sequences across the AP1 and ETS/NFATbinding sites may be required for constitutive expression in some acute myeloid leukemia cell lines (20). A basic promoter unit may therefore be required for GM-CSF promoter function in a number of different cell types. The fact that mutation of any one transcription factor-binding site abolishes promoter activity suggests that all the sites act as a functional unit. A similar cooperative complex of factors has been described for the interferon-␤ promoter and termed an enhanceosome (49). The IL-2 promoter appears to operate in a similar manner (50). CBF and AP1 could clearly be involved in expression in fibroblasts because AP1 is widely expressed and TNF-␣-inducible, and CBF is constitutively expressed (4,16,18). The relevance of ETS/NFAT in fibroblasts is less clear because they have primarily been investigated in lymphoid gene expression (4, 51). We have not been able to detect NFAT protein in fibroblasts 2 ; hence, the most proximal site may bind an ETS family member (51). The way in which the promoter responds in different cell types and to different stimuli will depend on variations in levels of constitutive factors between cell types and the degree to which inducible factors respond to stimuli.
An Ordered Arrangement of Repressor Elements Binding CSD Proteins across the Proximal Promoters of the GM-CSF, G-CSF, and IL-3 Genes-We previously observed the binding of nuclear NF-GMb and NF-GMc complexes to two repeated 5Ј-CCTG-3Ј sequences on the noncoding (Ϫ) strand of domain 1 of the GM-CSF promoter (Figs. 1 and 7a). We subsequently found that mutation of these sites resulted in an increase in TNF-␣inducible expression directed by domain 1 and 2 TNF-␣-responsive regions (13). This identified the 5Ј-CCTG-3Ј sites as repressor elements. By screening a cDNA library with a single strand domain 1 probe, we determined that the 5Ј-CCTG-3Ј elements bound CSD proteins. Given we observed that recombinant and nuclear NF-GMb/c complexes bound common single strand DNA sequences and that NF-GMb/c complexes were competed by CSD consensus sequences and by CSD antibodies, we determined that NF-GMb/c represented nuclear CSD complexes (14). We now show that nuclear NF-GMb/c-like CSD complexes and recombinant CSD protein (DbpB) bind across the TNF-␣-responsive elements in domain 2. As for domain 1, two NF-GMb/CSD sites were identified, but in contrast to domain 1 the 5Ј and 3Ј sites were, respectively, a 5Ј-CCTG-3Ј and a novel 5Ј-ACCA-3Ј sequence and were on the coding (ϩ) strand of domain 2 (Figs. 1 and 7a). Mutation of either one of these sites, as for domain 1, resulted in an increase in TNF-␣-inducible expression, identifying these elements as repressors. The spacing between the four CSD sites in domains 1 and 2 was conserved bringing about an ordered regularly spaced arrangement of CSD repressor sites across the GM-CSF promoter. Overexpression of DbpB and DbpA CSD proteins confirmed that CSD proteins were the mediators of repression via the NF-GMb/CSD sites (14).
We previously proposed that the binding of NF-GMb/CSD proteins to single strand domain 1 DNA resulted in a local single strand structure blocking the binding of transcriptional activators that are dependent on double strand DNA for binding and activity (4,13,14). This proposed single strand structure can now be extended to contain the entire TNF-␣-responsive GM-CSF promoter, covering all activator binding sites and hence providing an efficient means of completely silencing the promoter. This model is shown in Fig. 8a. Even though individual NF-GMb/CSD elements can act as repressor elements, our previous and present transfection data reveal that all four CSD sites are required for maximal GM-CSF promoter repres-sion. For example, the pGM93 construct containing two sites is repressed to a greater extent than pGM41 containing one site, whereas a construct containing all four sites is maximally repressed (13,14). This is consistent with the idea of an extensive single structure across the GM-CSF promoter requiring binding to all four sites. Consistent with this model, single strand regions within double strand DNA, coinciding with CSD-binding sites have been detected in vitro (37,38,52). In vivo studies will confirm such a model. The binding of CSD proteins to opposite strands of the promoter may allow stabilization of such a single strand structure by interaction of CSD proteins (27,28) bound to either strand.
CSD proteins have also recently been shown to repress a number of genes including those for thyrotropin receptor (36), nicotinic acetylcholine receptor ␦ (53), major histocompatibility complex class I and II genes (34,54,55), and the grp78 gene (35). CSD proteins also bind to repressor sequences in the ␥ globin genes (38). Extensive characterization of CSD-binding sites across promoter elements has only been performed for the major histocompatibility complex DR␣ (52) and thyrotropin receptor (36) genes, but neither study reveals the ordered arrangement of CSD sites reported here. In the thyrotropin receptor gene, three CSD sites have been detected, one on the noncoding and two on the coding strand (36). These sites have a common sequence but are separated by large distances. The major histocompatibility complex DR␣ gene has two CSD-binding sites on opposing strands, but the exact location of the sites has not been determined (52). The study of the GMϪCSF promoter therefore reveals a unique arrangement of CSD sites that can efficiently function to repress an entire proximal promoter.
The location of the CSD sites in the GM-CSF promoter suggests that CSD proteins may be involved in repression of the GM-CSF promoter not only in fibroblasts but also in T cells and potentially in endothelial and myeloid cells. Consistently, sequences containing the 5Ј repressor site in domain 2 have been shown to have repressor activity in myeloid leukemic cell lines and in Jurkat, MLA144, and primary human T cells (4,20). In addition, sequences containing the 3Ј domain 2 repressor element have repressor activity in an acute myeloid leukemia cell line (20), and domain 1 has been reported to have repressor activity in endothelial cells (19). The sequences identified in the acute myeloid leukemia cell line bind a 45-kDa protein (20), consistent with the size of a CSD protein, and we have also identified CSD binding to the GM-CSF promoter in extracts from a number of GM-CSF expressing cells (data not shown). CSD proteins may be involved in maintaining tight regulation of the GM-CSF promoter in all expressing cell types. The signaling pathways and transcription factors activated in different cell types will dictate the ability of different signals to overcome the repressive effects of CSD binding.
In addition to GM-CSF, we analyzed the promoter sequences of two other myeloid growth factor genes, the human G-CSF and IL-3 genes (56,57). These genes have overlapping patterns of expression with GM-CSF (2,4,25). We find here that the unique arrangement of CSD sites in the GM-CSF gene is also apparent across the G-CSF and IL-3 proximal promoters (56,57). In each of the three genes the domain 1 NF-GMb/CSD sites are on the noncoding (Ϫ) strand, whereas the domain 2 sites are on the coding (ϩ) strand. CSD sites in the G-CSF and IL-3 genes overlap or are adjacent to activator sites in these genes (2, 58 -64). Some of these sites are in common with GM-CSF gene activator sites including SP1, CBF, NF-kB, CK-1, and CD28-responsive complex sites (3,4). The finding of a conserved arrangement of CSD sites suggests a common means of repression of the growth factor genes. Consistent with this, both domain 1 and domain 2 regions in the IL-3 gene have been shown to have repressor activity in T cells (62), and domain 2 in the G-CSF gene has repressor activity in CHU-2 cells (65). In addition we have confirmed by mutation analysis that the most 5Ј G-CSF domain 1 CSD site is required for nuclear CSD binding and that overexpression of CSD protein represses the TNF-␣-inducible expression of the G-CSF promoter in fibroblasts. 2 The sequence, spacing, and strand conservation of CSD sites in the three genes suggests an important role not only for binding of CSD proteins to DNA but also for CSD interactions with other regulatory proteins (35,51), and ultimately, the structure of the complex formed on the DNA resulting from binding these factors. The idea of a common single strand structure across all three genes is supported by the presence of CT-rich regions flanking the CSD sites. Such sites are susceptible to single strand DNA formation and could act as entry sites for CSD proteins (37,38) to enable them to bind and form a repressive single strand structure common to the three myeloid growth factor genes.
Distinct CSD-containing Nuclear Complexes Can Bind to the Domain 1 and Domain 2 Sites in the GM-CSF Promoter-From our previous analysis of the binding of nuclear NF-GMb/c complexes to mutant domain 1 oligonucleotides and from analysis of the protein composition of complexes by UV cross-linking, we interpreted our data (13,14) as follows: The NF-GMb complex represents two separate complexes, one containing a single 42-kDa protein requiring both CSD sites for maximal binding and the other containing two 22-kDa proteins, with one 22-kDa protein bound to each of the CSD sites. The NF-GMc complex represents the binding of the 22-kDa protein to one or other CSD site. This is summarized in Fig. 8b. The 42-kDa protein is the correct size for a full-length CSD protein as determined by Western analysis of nuclear extracts from a number of cell lines (43,52). Consistently full-length recombinant CSD proteins (DbpA and DbpB) also require both CSD sites for full binding to domain 1 (14). The 22-kDa protein probably represents a truncated CSD protein (discussed below). We now analyze the pro-tein composition of nuclear NF-GMb/c across domain 2 and interpret our data as summarized in Fig. 8b. As for domain 1, NF-GMb contains two complexes, one containing one single 42-kDa protein requiring both the 5Ј-CCTG-3Ј and 5Ј-ACCA-3Ј sites for full complex formation (Figs. 3c and 4b). Consistently recombinant DbpB CSD protein requires both sites for full binding (Fig. 3d). In contrast to domain 1, the second complex in domain 2 NF-GMb represents binding of both a 22-kDa protein to the 5Ј-CCTG-3Ј site and a novel 25-kDa protein binding to the 5Ј-ACCA-3Ј site (Fig. 4b). We have confirmed that the 25-kDa protein, like the 42-and 22-kDa proteins, is a cold shock protein by use of a CSD antibody. 2 Our data suggest that the 25-kDa protein is a separate subtype or an altered conformation of the 22-kDa protein contacting 5Ј-ACCA-3Ј. The NF-GMc complex on domain 2 most likely represents the binding of either a single 22-or 25-kDa protein. Taken all together, for the whole GM-CSF promoter (Fig. 8), a single 42-kDa CSD protein can contact the pair of sites in domain 1 or domain 2, a single 22-kDa protein can contact each of the first three 5Ј-CCTG-3Ј sites, and the 25-kDa protein contacts the 5Ј-ACCA-3Ј site. Hence the four repressor sites across the GM-CSF promoter that are required for full promoter repression bind a series of different CSD subtypes.
We are at present screening for cDNAs encoding the 22/25-kDa CSD subtypes. Proteins in the 22/25-kDa size range could be produced from reported alternatively spliced DbpA CSD cDNA sequences (42,52,66,67) or from potentially functional DbpB pseudogenes (67)(68)(69). Interestingly a chicken CSD factor binding to an 5Ј-ACCA-3Ј sequence in the Rous sarcoma virus long terminal repeat promoter represents a truncated form of DbpA, called YB-2 (41,42). Human homologues of YB-2 have also been identified (66,67); hence the 25-kDa protein that binds to a 5Ј-ACCA-3Ј sequence may represent a YB-2-type protein. The function of human YB-2 is unknown, but it is known that it lacks sequences in the C-terminal region relative to full-length CSD proteins. CSD proteins have three functional domains, an N-terminal, a central highly conserved cold shock, and a C-terminal domain (28,31,32). The central CSD domain is involved in single strand DNA binding and repression mechanisms (4, 14, 34 -36, 52-55), and the C-terminal region is involved in interaction with other transcription factors (28, 35, 51, 70 -72). Such protein interaction appears to be involved in both the mechanisms of repression and subsequent derepression upon stimulation (35,51,71). It can be seen therefore that full-length and truncated proteins may have different abilities to repress and also vary in the degree to which their repressive action can be reversed.
We cannot yet determine the precise role that each CSD subtype is playing at each repressor site until the subtypes are cloned, but our data do demonstrate that the full-length 42-kDa protein binds to sequences in common with those binding either the truncated 22-or 25-kDa protein. As we have observed that the relative levels of the 42-kDa versus 22/25-kDa factors vary between cell types, 2 it is probable that full-length and truncated proteins compete for binding sites in vivo. The relative amounts of the different CSD types may determine the ability of a gene to be repressed and derepressed in different cell types. It will also be of interest to determine the differences between 22-and 25-kDa proteins that dictate their binding to different sequences. At present the functional consequences of binding to different sequences is not apparent, but from mutation studies, however, the 22-kDa site (5Ј-CCTG-3Ј) in domain 2 does appear to be a stronger repressor element than the 25-kDa (5Ј-ACCA-3Ј) site (Fig. 2). The presence of alternative CSD subtypes or alternative conformational forms binding to both different and common elements enables a set up that can be manipulated in vivo to bring about appropriate gene regulation in different cell types.
CSD proteins have been shown to interact with transcription factors in solution (35,71,72) or to complex with a transcription factor on its double strand DNA-binding site (51). A complex of CSD proteins with heterologous proteins on single strand DNA has not, however, previously been reported. We show here that CSD proteins can interact with the GM-CSF promoter in association with a heterologous single strand DNA-binding protein to form the NF-GMa complex. The NF-GMa complex was originally identified as binding to the CK-1 regions of the GM-CSF, G-CSF, and IL-3 genes (23,24). We now show that NF-GMa is a complex of 42-and 22-kDa nuclear CSD proteins with a novel 16-kDa protein that forms on the noncoding (Ϫ) strand of the GM-CSF CK-1/CK-2 region in domain 1. We have determined the N-terminal sequence of the 16-kDa protein and found it to be identical to the N-terminal sequence of mature human mtSSB (44). Consistent with the properties of the 16-kDa protein, mtSSB has a molecular mass of 15.2 kDa and binds to single strand mitochondrial DNA (44,45). Human mtSSB belongs to a family of SSB proteins conserved from E. coli to mammals (44 -46, 73, 74). In E. coli and in the mitochondria of higher organisms these proteins have been implicated in the processes of DNA replication, recombination, and repair (74 -46) and in transcriptional derepression (75). In higher organisms mtSSB is primarily detected in mitochondria. Trace amounts have, however, been detected in the nucleus (77,78), and it has been demonstrated that overexpression of mtSSB can result in the activation of a nuclear gene, A␣ fibrinogen (78). Interestingly, a nuclear protein with N-terminal sequence conserved with mature mtSSB was isolated as binding to an IL-6 response element in the A␣ fibrinogen gene (78). This element, 5Ј-GAATTTCTGGGA-3Ј, has a similar sequence to that observed across the CK-1 region CSD site 1 in the growth factor genes we have investigated. The homology is particularly striking with the GM-CSF and G-CSF genes ( Figs.  1b and 7, a and b). mtSSB or related proteins may therefore have a broader role in the regulation of genes involved in growth and stress responses as is also the case for CSD proteins. Our identification of the 16-kDa component of NF-GMa as an mtSSB-like protein strengthens the idea that mtSSB type proteins can have a function in both mitochondria and the nucleus.
We have also shown that the association of the 16-kDa mtSSB related protein with the 42-and 22-kDa nuclear CSD proteins changes the specificity of the CSD factors for binding to domain 1. The 42-kDa protein no longer requires the NF-GMb/CSD sites to bind to domain 1 DNA when it is part of the NF-GMa complex. The 22-kDa protein may still need to interact with these elements because it is not present in the NF-GMa complex formed on the GMm23 mutant sequence that lacks functional CSD-binding sites. The NF-GMa complex can probably also form in solution in the absence of DNA because it can be separated by heparin-Sepharose chromatography from NF-GMb/c nuclear complexes containing only the 42-and 22-kDa CSD proteins. These findings are consistent with the reported abilities of both CSD and SSB proteins to complex with other proteins (44, 46 -48). We have also shown that the NF-GMa complexes can compete with NF-GMb/c complexes for binding to domain 1. Given the effects of the 16-kDa protein on CSD factor binding to repressor elements, formation of the NF-GMa complex may prevent the repressive function of the CSD proteins. This is supported by our observation of an increase in NF-GMa complex formation upon TNF stimulation of fibroblasts (23,24). Such a function for NF-GMa is consistent with the observed involvement of E. coli SSB in transcriptional derepression (75) and for mammalian mtSSB in the activation of the A␣ fibrinogen gene (78). Hence NF-GMa complex formation may be part of a mechanism to ensure rapid derepression of very tightly regulated genes such as GM-CSF and the other growth factor genes, G-CSF and IL-3, that also bind both the repressive NF-GMb/c complexes and the potentially antagonistic NF-GMa complex.
We have now characterized the entire TNF-␣-responsive proximal promoter of the human GM-CSF gene. We have identified the sequences required for both activation and repression of this promoter and have characterized a family of nuclear complexes containing single strand DNA-binding proteins that bind across these sequences. In doing so we determined a role for these single strand proteins in both the mechanisms of repression of the GM-CSF gene and potentially other growth factor genes and possibly in their subsequent activation.