Cold Shock Domain Factors Activate the Granulocyte-Macrophage Colony-stimulating Factor Promoter in Stimulated Jurkat T Cells*

Cold shock domain (CSD) family members have been shown to play roles in either transcriptional activation or repression of many genes in various cell types. We have previously shown that CSD proteins dbpAv and dbpB (also known as YB-1) act to repress granulocyte-macrophage colony-stimulating factor transcription in human embryonic lung (HEL) fibroblasts via binding to single-stranded DNA regions across the promoter. Here we show that the same CSD factors are involved in granulocyte-macrophage colony-stimulating factor transcriptional activation in Jurkat T cells. Unlike the mechanisms of CSD repression in HEL fibroblasts, CSD-mediated activation in Jurkat T cells is not mediated through DNA binding but presumably through protein-protein interactions via the C terminus of the CSD protein with transcription factors such as RelA/NF-κB p65. We demonstrate that Jurkat T cells lack truncated CSD factor subtypes present in HEL fibroblasts, which raises the possibility that the cellular content of CSD proteins may determine their final role as activators or repressors of transcription.

Cold shock domain (CSD) family members have been shown to play roles in either transcriptional activation or repression of many genes in various cell types. We have previously shown that CSD proteins dbpAv and dbpB (also known as YB-1) act to repress granulocytemacrophage colony-stimulating factor transcription in human embryonic lung (HEL) fibroblasts via binding to single-stranded DNA regions across the promoter. Here we show that the same CSD factors are involved in gran-

ulocyte-macrophage colony-stimulating factor transcriptional activation in Jurkat T cells. Unlike the mechanisms of CSD repression in HEL fibroblasts, CSDmediated activation in Jurkat T cells is not mediated through DNA binding but presumably through proteinprotein interactions via the C terminus of the CSD protein with transcription factors such as RelA/NF-B p65.
We demonstrate that Jurkat T cells lack truncated CSD factor subtypes present in HEL fibroblasts, which raises the possibility that the cellular content of CSD proteins may determine their final role as activators or repressors of transcription.
Cold shock domain (CSD) 1 proteins were originally identified in bacteria and have been shown to be highly conserved throughout evolution from bacteria to humans (1)(2)(3). CSD proteins have three functional domains: an N terminus, the central CSD, and a C-terminal domain. The N terminus region of the protein has not been well characterized but has been shown to contribute to single-stranded DNA binding (4,5). The highly conserved central cold shock domain (from which this family of proteins derives its name) contains an RNP1 motif that is essential for sequence-specific DNA and RNA binding (5-9) (see Fig. 1C). The C terminus of the protein has alternating basic and acidic domains and has been implicated in both nonsequence specific RNA binding and protein-protein interactions with transcriptional regulators like RelA, ZO-1, TATA binding protein, NF-Y, YY-1, and AP-2 (10 -15) (Fig. 1C). As a whole, the family of CSD proteins has been reported to bind to double-and single-stranded DNA and RNA and is involved in transcriptional repression and activation and in mRNA pack-aging, transport, localization, masking, stability, and translation (1-3, 16 -20).
GM-CSF belongs to a family of growth factors that control survival, proliferation, and differentiation of cells from the hematopoietic lineage of granulocytes and macrophages (20, 38 -41). GM-CSF is expressed by a wide variety of cells including myeloid, mesenchymal, and lymphoid cells in response to stress signals such as those derived from infection, inflammation, and blood loss (20,28,38,39). GM-CSF expression is triggered in T cells primarily via T cell receptor and coreceptor activation and is tightly regulated at the level of transcription (20,38). The human GM-CSF proximal promoter can be conveniently divided into two domains (domain 1, Ϫ114 to Ϫ71 and domain 2, Ϫ70 to Ϫ31), which are both important for GM-CSF regulation (21)(22)(23) (Fig. 1A). The two domains have many binding sites for transcription factors including: NF-B/ Rel, CBF, AP-1, ETS/NFAT, and CD28-responsive complex (CD28RC), which all act to mediate GM-CSF expression (reviewed in Refs. 20 and 38). In addition to the double-stranded DNA binding transcription factors mentioned above, we have shown previously that there are four CSD binding sites along the GM-CSF proximal promoter. There are two single-stranded DNA CSD binding sites along the noncoding (Ϫ) strand of domain 1 and two along the coding (ϩ) strand of domain 2 (Fig.  1, A and B) (21)(22)(23). We demonstrated that these CSD sites acted as repressor sites in human embryonic lung (HEL) fibroblasts, that CSD proteins could bind to these sites, and that overexpression lead to repression of TNF␣-mediated activation of GM-CSF (21)(22)(23).
T cells are a major source of GM-CSF in vivo, and we wished to determine whether CSD proteins played a role in GM-CSF transcription in these cells. We now report that CSD proteins are present in Jurkat T cells and bind to the defined CSD binding sites along the GM-CSF promoter. The Jurkat T cell nuclear CSD complexes that form on the GM-CSF promoter are distinct from those found in HEL fibroblasts. Using transient transfection assays, we previously showed that the CSD binding sites acted as repressor elements in HEL fibroblasts. However, when Jurkat T cells are stimulated with PMA/Ca 2ϩ ionophore to mimic T cell receptor activation, the CSD factors behave as coactivators. The coactivation seen does not function through the CSD binding sites but through potential protein-protein interactions mediated via the C-terminal domain of CSD proteins.

EXPERIMENTAL PROCEDURES
Plasmid Constructs-Luciferase reporter plasmids were constructed by inserting oligonucleotides encoding human GM-CSF promoter fragments into the pXP1 luciferase vector (Promega). The number after the pGM in each luciferase construct name indicates how many CSD binding sites are present. The plasmids pGM1-Luc, pGM2-Luc, and pGM4-Luc were constructed by cloning oligonucleotides (with HindIII 5Ј and BamHI 3Ј ends) spanning regions of the GM-CSF promoter (Ϫ65 to ϩ28), (Ϫ70 to ϩ28), and (Ϫ114 to ϩ28) into the HindIII/BamHI sites of pXP1 (Fig. 1A). The QuikChange TM site-directed mutagenesis kit (Stratagene) was used to mutate the CSD binding site in pGM1-Luc from 5Ј-ACCA-3Ј to 5Ј-AGGA-3Ј to create the pGM1-mut1-Luc reporter plasmid (Fig. 1A). The backbone expression vector pRcCMV was obtained from Invitrogen. Construction of the RelA expression plasmid has been described previously (42). Construction of the expression plasmids pSGdbpAv and pSGdbpB has been described previously (22). The dbpB deletion plasmid pSGdbpB⌬1 was created by digesting pSGdbpB with NarI, which removes the CSD and C-terminal domain of dbpB, followed by religation (pSGdbpB⌬1 contains the first 47 amino acids of FIG. 1. The GM-CSF proximal promoter and CSD factor dbpB truncation constructs. A, the sequence of the human GM-CSF proximal promoter is shown. Domain 1 (Ϫ114 to Ϫ71) and domain 2 (Ϫ70 to Ϫ31) regions are indicated. The binding sites for many double-stranded DNA binding transcription factors that mediate GM-CSF expression, including NF-kB, CBF, AP1, ETS/NFAT, and CD28RC, are underlined (20,38). Nuclear NF-GMb/c and recombinant CSD factor binding sites are indicated on the noncoding (Ϫ) strand in domain 1 and the coding (ϩ) strand in domain 2 with shaded boxes. GM-CSF promoter, luciferase reporter constructs used, pGMCK-1(2)-TK/Luc, pGM1-Luc, pGM1(mut1)-Luc, pGM2-Luc, and pGM4-Luc are shown diagrammatically underneath. Numbers indicate distance from the transcriptional start site, and boxes represent CSD binding sites. Mutation of the CSD binding site in pGM1(mut1)-Luc is represented by a box with an X in it, and the altered sequence is given below it. B, the sequence of the coding (ϩ) and noncoding (Ϫ) wild-type domain 1 (Ϫ114 to Ϫ79) oligonucleotides, GM and GMϪ, are shown (21,22). Sequences needed for nuclear NF-GMb/c and recombinant CSD factor binding to the noncoding strand are indicated by a box. Base changes in CSD binding sites of the noncoding (Ϫ) strand for each mutant oligonucleotide GMm19, GMm22, and GMm23 are shown (22). The sequence of the wild-type-coding strand (ϩ) domain 2 GM93ϩ oligonucleotide (Ϫ70 to Ϫ31) is given with CSD, CBF, AP1, and ETS/NFAT sites shown (21,22). Base changes in the CSD binding sites of the coding (ϩ) strand for each mutant oligonucleotide GMm95ϩ, GMm103ϩ, and GMm105ϩ are shown (23). C, the coding regions of dbpB contained in the expression plasmids pSGdbpB, pSGdbpB⌬1, and pSGdbpbB⌬2 are shown schematically. The N-terminal, CSD, and C-terminal regions of the protein are represented, and the numbers represent the amino acid number. The construct pSGdbpB encodes for full-length dbpB, pSGdbpB⌬2 (the first 173 amino acids), and pSGdbpbB⌬1 (the first 47 amino acids). dbpB) (Fig. 1C). The dbpB deletion plasmid pSGdbpB⌬2 was created by digesting pSGdbpB with EcoRI and Sau3AI, which removes most of the C-terminal domain of dbpB, and ligating this fragment into an EcoRI/ BglII-digested pSG5, pSGdbpB⌬2, containing the first 173 amino acids of dbpB (Fig. 1C).
Oligonucleotides and Probe Preparation-All oligonucleotides were purchased from GeneWorks (Adelaide, Australia), and full-length product was purified from nondenaturing polyacrylamide gels (43). Singlestranded DNA probes for gel retardation assays were prepared by end-labeling oligonucleotides with T4 polynucleotide kinase and [␥-32 P]ATP followed by gel purification.
Preparation of Recombinant Proteins-Construction of the bacterial expression construct for CSD protein dbpB (pGEXBT) has been previously described (23). The Escherichia coli strain MC1061 transformed with pGEXBT was induced with isopropyl-1-thio-␤-D-galactopyranoside to produce recombinant GST-dbpB fusion protein. Recombinant GST-dbpB was purified on glutathione-Sepharose beads as described by the manufacturer (Amersham Pharmacia Biotech). The bacterial expression construct for recombinant RelA was a gift of Dr. Steven Gerondakis, and protein was prepared according to the procedure described by Dunn et al. (42).
Gel Retardation Analysis, Competitions, and Antibody Blocking Experiments-Nuclear extracts were prepared from HUT78 T cells and Jurkat T cells as previously reported (44 45). Gel retardations were carried out using 0.25 ng of single-stranded 32 P-labeled oligonucleotide probes (Fig. 1A) in a 10-l reaction mix of 0.5ϫ TM buffer (21,22,45) containing 200 mM KCl, 0.4 g of poly(dI-dC), and either 1 g of nuclear extract or 25 ng of recombinant CSD fusion protein (GST-dbpB) or recombinant RelA. 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% (for nuclear extracts) and 6% (for recombinant proteins) nondenaturing polyacrylamide gels run in 0.5ϫ TBE (46). Competitions with unlabeled single strand oligonucleotides were performed by mixing protein and unlabeled probe followed by the immediate addition of the 32 P-labeled probe (22). Antibody blocking experiments were performed by adding protein and antibody and incubating for 5 min at room temperature before adding the 32 P-labeled probe. The reaction was then incubated for an additional 20 min at room temperature before being analyzed on polyacrylamide gels.
UV Cross-linking-For UV cross-linking, nuclear extracts were bound to 32 P-labeled single-stranded DNA probes (Fig. 1B), and the complexes were separated on a 12% polyacrylamide gel as described above. The gel was exposed to UV light (340 nm) for 15 min to cross-link bound protein to the DNA. The gel was exposed to x-ray film for 12 h at 4°C, and the retarded complexes were excised. Protein in the excised bands was analyzed on a 12% SDS-polyacrylamide gel (21,43).
Antibodies-The anti-CSD peptide antibody was raised by immunizing rabbits with the peptide (IKKNNPRKYLRSVGD) (dbpB amino acids 89 -103) conjugated to keyhole limpet hemocyanin (Imject conjugation kit, Pierce). Immunoglobulins were partially purified via ammonium sulfate precipitation. The specificity of the anti-CSD peptide antibody was verified by dot blot immunoblots against the CSD peptide, peptides to different regions of the dbpB protein, and recombinant glutathione S-transferase (data not shown).
Cell Culture, Transfections, and Luciferase Assays-The Jurkat T cell line was cultured in RPMI medium containing 10% fetal calf serum supplemented with L-glutamine, penicillin, and gentamycin antibiotics. Electroporation with a Bio-Rad Gene Pulser was used for transfection of Jurkat T cells at 270 V and a capacitance of 960 microfarad. 5 ϫ 10 6 cells were electroporated in 500 l of RPMI with 20% fetal calf serum per transfection with 5 g of reporter plasmid and 10 g of expression plasmid. Cells were stimulated 24 h post-transfection at a final concentration of 20 ng/ml PMA and 1 M calcium ionophore (A23187) and, 8 h post-stimulation, were assayed for luciferase activity as described by Osborne et al. (47).
Western Blot and Southwestern Analysis-Nuclear extracts from Jurkat and HUT78 T cells were isolated as described previously (44 45), and 10 g of protein was separated by 12% SDS-polyacrylamide gel electrophoresis before transfer to a nitrocellulose membrane via a Bio-Rad protein transfer apparatus. The filter was probed with CSD peptide antibody (1:1000 dilution) and developed with an ECL detection kit according to the recommendations of the manufacturer (Amersham Pharmacia Biotech). For Southwestern analysis, nuclear extracts were prepared as described above, and the filter was probed with 32 P-labeled oligonucleotides as described by Silva et al. (48).

Detection of CSD Complexes in Jurkat T Cells-
We have previously shown that both HEL fibroblasts and HUT78 T cells contain nuclear CSD proteins and that they bind to the human GM-CSF proximal promoter (21)(22)(23). We have also shown previously that the nuclear CSD complexes in HEL fibroblasts and HUT78 T cells are the same (21)(22)(23). To determine whether Jurkat T cells contained CSD proteins, Jurkat T cell nuclear extract was bound in a gel shift assay to single-stranded oligonucleotides spanning domain 1 and domain 2 of the human proximal GM-CSF promoter (Fig. 1B) and compared with HUT78 T cell CSD factor binding on the same oligonucleotides. When HUT78 T cell nuclear extract was used in a gel shift assay with the noncoding (Ϫ) strand of domain 1 (GMϪ), which contains two CSD binding sites, we observed the previously reported NF-GMb and NF-GMc nuclear complexes ( Fig. 2A, lane 2). Specific competitions, cross-linking, and antibody analysis have previously demonstrated that these HUT78 T cell complexes contain CSD proteins (21)(22)(23). In HEL fibroblast and HUT78 T cells, the nuclear NF-GMb complex migrates in multiple apparent conformational forms NF-GMb (1) and NF-GMb (2), and we have shown the different conformational forms have identical sequence specificity and protein content (21-23). 2 When Jurkat T cell nuclear extract was used in a gel shift assay we saw complexes that comigrated with the HUT78 T cell NF-GMb complexes but could not detect a NF-GMc-like complex ( Fig. 2A, lane 1). Using HUT78 T cell nuclear extract in a gel shift with the coding (ϩ) strand of domain 2 (GM93ϩ), which also contains two CSD binding sites, we observed three bands making up the NF-GMb complex and a single NF-GMc band ( Fig. 2A, lane 4) (21-23). 2 Jurkat T cell nuclear extract gave two bands that comigrated with the two slowest-migrating HUT78 T cell NF-GMb bands, and again, no NF-GMc-like complex was seen ( Fig. 2A, lane 3).
To verify that the Jurkat T cell NF-GMb complexes seen on domains 1 and 2 were authentic CSD-containing NF-GMb complexes, competition assays were performed. As shown in Fig.  2B, the unlabeled, wild-type, noncoding (Ϫ) oligonucleotide GMϪ (contains two CSD sites), when used as a competitor, inhibited formation of the Jurkat T cell NF-GMb complex on both domain 1 (GMϪ) (Fig. 2B, lane 7) and domain 2 (GM93ϩ) probes (Fig. 3A, lane 7). Competing with GMm23, domain 1 noncoding strand oligonucleotide, which has both CSD sites mutated, had no effect on NF-GMb complex formation (Fig. 2B,  lanes 3 and 8). As a positive control, we used an oligonucleotide from the coding (ϩ) strand of the human papillomavirus 18 enhancer (HPVϩ), which has been shown to bind recombinant and nuclear CSD proteins (22,49). HPVϩ was able to compete the Jurkat T cell NF-GMb complex binding almost as well as the wild-type GMϪ oligonucleotide (Fig. 2B, lanes 4 and 9). The NF-GMb complex was not competed by an oligonucleotide (nonspecific) that we and others have shown is unable to bind nuclear or recombinant CSD proteins (Fig. 2B, lanes 5 and 10) (5,22,50).
As further evidence that the Jurkat T cell NF-GMb complex contained CSD factors, we made polyclonal antibodies to the peptide (IKKNNPRKYLRSVGD, dbpB amino acids 89 -103) that represents a small region of the highly conserved CSD domain which is required for single-stranded DNA binding (7). This region of the CSD domain has identical sequence in both CSD factors, dbpB and dbpAv. The antibody raised to the CSD peptide was used in a gel shift with recombinant CSD protein dbpB (23). Increasing amounts of antibody in a gel shift assay with recombinant dbpB demonstrated that the CSD peptide antibody could block recombinant CSD protein dbpB binding to the GMϪ domain1 oligonucleotide (  lanes 2-4). The addition of pre-immune antibody (P.I) to the binding reaction mixture at the maximal concentration of CSD peptide antibody used showed no effect on mobility or complex formation in either the recombinant dbpB protein or the Jurkat nuclear extract gel shift (Fig. 2C, compare lanes 1 and 5 in both panels). The CSD peptide antibody was also able to block binding of the recombinant dbpB and Jurkat T cell nuclear extract to the coding (ϩ) strand of domain 2 (GM93ϩ) (data not shown).
To determine whether the Jurkat T cell NF-GMb CSD-containing complex bound to the GM-CSF promoter CSD binding sites, nuclear extract was used in a gel shift analysis with wild-type domain 1 (GMϪ) and domain 2 (GM93ϩ) oligonucleotides (each contain two CSD binding sites) and their respective CSD binding site mutants. These CSD binding site mutations (see Fig. 1B) have been previously described and characterized for HUT78 T cell and HEL fibroblast CSD binding (21, 23). 2 Gel shift analysis using the CSD mutant oligonucleotides on the noncoding (Ϫ) strand of domain 1 were compared with the  1 and 2), and domain 2 coding (ϩ) strand GM93ϩ (lanes 3 and 4) single-stranded oligonucleotides in a gel shift assay. Different conformational forms of NF-GMb are represented by (1), (2), and (3). ϫ represents nonspecific binding, and ss represents free single stranded labeled oligonucleotide. B, Jurkat T cell nuclear extract was bound in a gel shift assay to labeled GM-CSF domain 1 noncoding (Ϫ) strand oligonucleotide GM-(lanes 1-5) and to domain 2 coding (ϩ) strand oligonucleotide GM93ϩ (lanes 6 -10). The NF-GMb complex was competed with (GMϪ) (lanes 2 and 7), the CSD site mutant (GMm23Ϫ) (lanes 3 and 8), the control CSD binding site (HPVϩ) (lanes 4 and 9), and a nonspecific oligonucleotide (N.S.) (lanes 5 and 10). C, Jurkat T cell nuclear extract and recombinant dbpB were bound to labeled domain 1 noncoding (Ϫ) strand GMϪ oligonucleotide in a gel shift assay. Increasing amounts of anti-CSD peptide antibody were added to the reaction (lanes 2-4). 2-Fold more antibody than in lane 2 was added in lane 3, and 5-fold more was added in lane 4. As controls, no antibody was added to the reaction (denoted by a minus (Ϫ) sign, lane 1), and rabbit pre-immune (P.I) was added in the reaction as a negative control (lane 5). D, Jurkat T cell nuclear extract was bound to wild-type domain 1 noncoding (GMϪ) oligonucleotide (lane 1), domain 1 CSD mutants (GMm19Ϫ, GMm21Ϫ, GMm23Ϫ (lanes 2-4)), wild-type domain 2 noncoding (GM93ϩ) oligonucleotide (lane 5), and domain 2 CSD mutants (GMm95ϩ, GMm103ϩ, GMm105ϩ ((lanes 6 -8)). binding on the wild type oligonucleotide GMϪ (Fig. 2D, lane 1). Mutating one CSD site in GMm19 decreased binding significantly (Fig. 2D, lane 2), whereas mutating the other CSD site in GMm21 actually increased CSD binding but also resulted in an altered mobility of the complex (Fig. 2D, lane 3). Only when both CSD sites were mutated, in the GMm23 oligonucleotide, was all binding abolished (Fig. 2D, lane 4). Similar results were seen when we compared binding of Jurkat T cell nuclear extract on the wild-type coding (ϩ) strand of domain 2 (GM93ϩ) (Fig. 2D, lane 5) to the CSD mutant oligonucleotides. Mutating one CSD binding site in GMm95 almost abolished binding (Fig.  2D, lane 6) and mutating the other CSD site in GMm103 also significantly reduced binding (Fig. 2D, lane 7). Again, it was not until we mutated both CSD binding sites, in the oligonucleotide GMm105, that we lost all binding (Fig. 2D, lane 8), indicating that both CSD binding sites were needed for full binding. These results agree with previous observations for NF-GMb CSD complex formation in HUT78 T cells and HEL fibroblasts (21)(22)(23).
In summary, Jurkat T cells contain CSD-like proteins that can form a NF-GMb complex on domains 1 and 2 of the GM-CSF proximal promoter. This binding is dependant on the presence of the previously defined CSD binding sites. However, unlike other cells previously examined, Jurkat T cell nuclear extracts do not form a NF-GMc complex on the GM-CSF promoter.
To determine whether the 22-and 25-kDa proteins were completely lacking from Jurkat T cells or simply unable to bind DNA, Western and Southwestern experiments were performed. Jurkat and HUT78 T cell nuclear extracts and recombinant dbpB-GST fusion protein were fractionated on a Laemmli SDSpolyacrylamide gel electrophoresis gel and transferred to nitrocellulose, resulting in three replicate filter panels. One panel was probed with the CSD peptide antibody in a Western blot. The CSD peptide antibody recognized a single protein of ϳ75 kDa for recombinant dbpB-GST, which is the predicted size for the dbpB-GST fusion protein (Fig. 3B, lane 3). In the Jurkat and HUT78 T cell nuclear extract lanes the CSD antibody cross-reacted with many proteins (Fig. 3B, lanes 1 and 2). Proteins of the approximate size for the CSD factors identified in UV-cross-linking experiments (42,25, and 22 kDa) were detected. A strongly reacting band of ϳ42 kDa was seen in both Jurkat and HUT78 T cell extracts, whereas the smaller CSD factors 25 and 22 kDa were seen only in the HUT78 T cell extract (Fig. 3B, lanes 1 and 2).
To confirm that these were the CSD factors identified by UV-cross-linking analysis above, Southwestern experiments were performed. In the Southwestern experiments, one panel was probed with the wild-type GM-CSF domain 1 GM-oligonucleotide (contains two CSD binding sites; Fig. 3C, lanes 1-3), and the other panel was probed with the double CSD binding site mutant of the GMϪ oligonucleotide, GMm23- (Fig. 3C,  lanes 4 -6). Recombinant dbpB-GST can be seen to bind to the wild-type GMϪ probe (Fig. 3C, lane 3) but not to the CSD mutant GMm23Ϫ probe (Fig. 3C, lane 6). Proteins of the same size, as detected by UV cross-linking the HUT78 T cell NF-GMb/c complexes (42, 25, and 22 kDa), were detected binding to the GMϪ probe in this assay (Fig. 3C, lane 2). Several other bands were also observed that may represent CSD family members. In Jurkat T cell extracts, a band of 42 kDa (and a smaller protein of 39 kDa) was the strongest band observed when the GMϪ oligonucleotide was used as a probe; of most significance, the 22-and 25-kDa bands were not observed (Fig. 3B, lane 1).
In summary, UV cross-linking the Jurkat T cell NF-GMb complex identified only a 42-kDa protein, which is also detected here by the CSD peptide antibody, and as expected, the 22-and 25-kDa proteins seen in the HUT78 T cell nuclear extract were not detected in Jurkat T cells (Fig. 3B, lane 1). As expected, the 42-, 25-, and 22-kDa proteins either did not bind or the binding was significantly less to the CSD mutant probe GM23Ϫ (Fig.  3C, lanes 4 -6). Other proteins are seen binding to the GMϪ probe and not to the CSD mutant GMm23Ϫ. CSD proteins of greater than 42 kDa in size have been detected in nuclear extracts by other groups, and these proteins, seen binding here to the GMϪ probe, may represent other CSD family members that do not form part of the NF-GMb/c complexes.
CSD Factors Activate the GM-CSF Proximal Promoter in Jurkat T Cells-We have previously reported that overexpression of CSD proteins repress tumor necrosis factor ␣-mediated activation of the human GM-CSF proximal promoter in HEL fibroblasts (22). We wished to determine whether CSD proteins also acted as repressors of GM-CSF in Jurkat T cells. Surprisingly, we saw that when the GM-CSF reporter plasmid (pGM4-Luc), which contained all four CSD binding sites, was cotransfected into Jurkat T cells with CSD overexpression constructs for dbpAv (pSGdbpAv) and dbpB (pSGdbpB), that the CSD proteins behaved as activators in the presence of PMA/Ca 2ϩ ionophore (Fig. 4A). That they were only involved in activation of the GM-CSF promoter construct when the cells were stimulated with PMA/Ca 2ϩ ionophore indicated the CSD proteins either needed to be activated to function or that they cooperate with other proteins to function.
We wished to determine which region of the GM-CSF proximal promoter was required for CSD-mediated activation and the role, if any, that the four CSD sites across the promoter played in this activation. To do this, CSD overexpression constructs were cotransfected with reporter constructs containing variable numbers of CSD sites (Fig. 4B). Initially the pGMCK-1(2)-TK/Luc construct (Ϫ114 to Ϫ70) containing only the two domain 1 CSD sites of the GM-CSF promoter was used in experiments. Cotransfections showed that both CSD factors (pSGdbpAv and pSGdbpB) could coactivate this construct in the presence of PMA/Ca 2ϩ ionophore (Fig. 4B). Since the do-main 1 region was cloned upstream of the basal thymidine kinase (TK) promoter in pGMCK-1(2)-TK/Luc, CSD factors were tested for activator function on the TK promoter. We observed that CSD factors could not activate the TK promoter in the presence of PMA/Ca 2ϩ ionophore (data not shown). We next tested a construct containing only the domain 2 CSD sites to determine whether they too were targets for CSD-mediated activation. When pGM2-Luc (Ϫ71 to ϩ28, containing only the domain 2 CSD sites) was used in cotransfection experiments, CSD-mediated activation was still observed, but both basal (pSG5)-and CSD (pSGdbpAv, pSGdbpB)-coactivated expression was dramatically reduced relative to pGM4-Luc (Fig. 4B). When however, the pGM2-Luc construct was truncated by five base pairs to create pGM1-Luc, higher levels of activity and CSD factor-mediated activation were restored (Fig. 4B). The five-base pair truncation removed the 5Ј CSD site in domain 2, leaving only the single 3Ј domain 2 CSD site. These results therefore indicated that the 5Ј CSD binding sites had repressor activity.
To determine whether CSD-mediated coactivation of pGM1-Luc was acting through the single CSD site in this construct, coactivation levels were compared between pGM1-Luc (which contains one CSD site) and pGM1(mut1)-Luc (where the one CSD site has been mutated) (Fig. 4B). Both pGM1-Luc and pGM1(mut1)-Luc were cotransfected into Jurkat T cells with constructs encoding for CSD factors dbpAv and dbpB. The PMA/Ca 2ϩ ionophore stimulated levels of both basal-and CSDmediated coactivation in pGM1(mut1)-Luc were greater than those seen on pGM1-Luc (up to 2-fold greater) (Fig. 4B). These data taken altogether therefore demonstrate that both the domain 2 CSD sites have repressor activity. Our results also show that the coactivation effect of CSD proteins on the GM-CSF promoter did not require contact with these CSD sites. We were unable to repeat similar experiments on domain 1 CSD sites due to the overlap of the NF-B sites with the CSD sites. Mutation of the domain 1 NF-B sites results in loss of GM-CSF promoter activity (54).
The C-terminal Domain of dbpB Is Required for CSD-mediated Activation of the GM-CSF Proximal Promoter-Because interactions between CSD factors and other transcriptional regulators have been implicated in mechanisms of activation of a number of viral and cellular genes, a potential mechanism for the coactivation effect seen here was via interaction of CSD factors with other proteins. The C-terminal domain of dbpB has been demonstrated to be required for protein-protein interactions (10 -15). To explore the possibility that the coactivation effect was mediated via CSD protein-protein interactions, expression constructs were made encoding truncations of the dbpB CSD protein. In the dbpB deletion construct pSGdbpB⌬2, most of the C-terminal region of dbpB was deleted, and in pSGdbpB⌬1, both the C-terminal and CSD domain were deleted (Fig. 1B). The dbpB and dbpB deletion constructs were cotransfected into Jurkat T cells with either the pGM1-Luc or pGM1(mut1)-Luc domain 2 constructs or the pGMCK-1(2)-TK/ Luc domain 1 construct, and the cells were stimulated with PMA/Ca 2ϩ ionophore. Removal of most of the C-terminal region of the dbpB in pSGdbpB⌬2 resulted in a significant decrease in the level of PMA/Ca 2ϩ ionophore coactivation on all constructs (Fig. 5). Removal of both the C-terminal and CSD domain of dbpB in pSGdbpb⌬1 resulted in a return to near basal levels of expression on both domain 1 and 2 constructs (Fig. 5).
Taken altogether the results presented in Figs. 4 and 5 indicate, first, that at least the domain 2 CSD binding sites in the GM-CSF proximal promoter act as repressor elements, second, that the PMA/Ca 2ϩ ionophore coactivation by CSD proteins is not associated with the CSD binding sites (at least in domain 2), and third, that the coactivation seen is mediated primarily via the C-terminal protein interaction domain of dbpB.  Fig. 1A). Boxes represent CSD binding sites, the box with an X in it represents where the CSD binding site has been mutated, and ovals represent the binding sites of the corresponding transcriptional factor in the top diagram. In all transfection experiments cells were treated with or without PMA/Ca 2ϩ ionophore for 8 h, and luciferase activities are given relative to untreated pXP1 background plasmid cotransfected with pSG5, which is given a value of 1. All untreated values were below a relative luciferase activity of 10. A, the reporter plasmid pGM4-Luc was cotransfected into Jurkat T cells with the empty pSG5 expression vector and with expression plasmids containing full-length dbpAv (pS-GdbpAv) and full-length dbpB (pSGdbpB). B, each GM-CSF promoter luciferase reporter plasmid, pGM4-Luc, pGMCK-1(2)-TK/Luc, pGM2-Luc, pGM1-Luc, and pGM1(mut1)-Luc, was cotransfected into Jurkat T cells with pSG5, pSGdbpAv, and pSGdbpB. Only PMA/Ca 2ϩ ionophoretreated values are shown; all untreated values were below a relative luciferase activity of 10. CSD Factors and RelA Act to Synergistically Activate the GM-CSF Proximal Promoter-The C-terminal domain of CSD proteins has been shown to be involved in interactions with other transcription factors and, to date, of the transcription factors acting across the GM-CSF proximal promoter; only RelA/NF-B p65 has been reported to interact with CSD proteins (10 -15, 55). To investigate the potential role of a CSD-RelA interaction in the CSD-mediated activation of the GM-CSF promoter domain 1 sequences (which contains RelA/ NF-B p65 sites; Refs. 20 and 54), RelA and CSD expression constructs were cotransfected with the GM-CSF domain 1 reporter construct pGMCK-1(2)-TK-Luc (Fig. 6). In Fig. 6, we show as previously reported that RelA/NF-B p65 stimulated the pGMCK-1(2)-TK-Luc construct in unstimulated cells, whereas the CSD factors could not (56). When the pGMCK-1(2)-TK/Luc reporter construct was cotransfected with expression constructs for both RelA (pRcCMVRelA) and dbpAv (pSG-dbpAv) or dbpB (pSGdbpB), synergistic activation was observed (Fig. 6). Repeating the triple cotransfection with the dbpB C-terminal truncation (pSGdbpB⌬2) significantly reduced the synergistic activation (Fig. 6), suggesting that the C terminus of the CSD protein dbpB protein may be involved in interactions with RelA to bring about activation. Cotransfecting with the dbpB construct pSGdbpB⌬1 (missing both the C-terminal and CSD domains) further reduced the activation to basal pSG5 levels (Fig. 6). These experiments show that RelA/ NF-B p65 and the CSD proteins can cooperate to activate GM-CSF promoter function and that the C terminus of dbpB is essential for this cooperation. DISCUSSION Cold shock domain family members have been associated with diverse functions including the ability to both activate and repress transcription and translation. Here we investigated the role of the CSD proteins dbpAv and dbpB in regulation of GM-CSF promoter in Jurkat T cells. We have previously reported that HUT78 T cell and HEL fibroblast nuclear CSD proteins bind to defined single-stranded regions along domain 1 and domain 2 of the GM-CSF promoter as NF-GMb and NF-GMc complexes (21)(22)(23). We proposed that the binding of NF-GMb/c CSD proteins along the GM-CSF promoter resulted in or stabilized a single-stranded DNA structure, thereby preventing the binding of transcriptional activators that are dependent on double-stranded DNA for binding and activity. This proposed structure would be an efficient means of silencing the GM-CSF promoter and many other promoters where NF-GMb/c CSD proteins have been shown to act as repressors of transcription (13, 23, 25-27, 33, 57). The specific arrangement of NF-GMb/c CSD binding sites observed on the GM-CSF promoter, a pair of distal sites binding on the noncoding strand and a pair of proximal sites binding the coding strand, can also be found in the promoters of the granulocyte-colony stimulating factor and interleukin-3 cytokine genes with overlapping GM-CSF expression patterns (58,59).
Analysis of Jurkat T cell extracts identified nuclear CSD proteins binding to the CSD binding sites across the GM-CSF promoter as a NF-GMb-like complex. Competitions with singlestranded DNA CSD site sequences and CSD antibody experiments in addition to UV cross-linking indicated that the Jurkat T cell NF-GMb complex was made up of only a 42-kDa CSD protein. With the lack of the NF-GMc complex in Jurkat T cells, we also observed a lack of the 22-and 25-kDa CSD proteins previously identified in HEL fibroblasts and HUT78 T cells (21,23). These 22-and 25-kDa CSD proteins probably represent CSD splice variants or proteolytic cleavage products, as reported by others, that lack the C-terminal protein-protein interaction domain (22, 31-33, 35, 53).
Transient transfection in Jurkat T cells showed that in contrast to the repression observed in HEL fibroblasts, the CSD proteins, dbpAv and dbpB, functioned as coactivators on a 140-base pair region of the human GM-CSF proximal promoter. This activation was only observed when T cells were activated with PMA/Ca 2ϩ ionophore. We showed that both the GM-CSF domain 1 region (containing NF-kB/RelA sites) and the domain 2 region (containing the CBF, AP1, and ETS/NFAT sites) are able to independently respond to CSD-mediated activation. We also showed that this response to CSD-mediated activation did not require the CSD binding sites, at least in the Ϫ60 to ϩ28 region of domain 2. The CSD binding sites, however, still retain repressor function as defined by deletion and mutation.
CSD factors have been reported to activate a few cellular genes including c-Myc, ␣1(I) procollagen, and metalloproteinase (MMP-2) and also regulate many viral promoters such as human immunodeficiency virus, human T-cell lymphotrophic virus, and Rous sarcoma virus long terminal repeats and human polyomavirus JC late viral promoter (5,31,32,35,36,60), but the precise mechanisms of CSD-mediated activation has not been well defined. To determine the mechanism of CSDmediated activation in Jurkat T cells, we examined the possible role of CSD protein interactions with other transcription factors. Our data indicate that CSD factors dbpB and dbpAv are able to act with RelA in transient transfections to synergistically activate the GM-CSF promoter in the absence of PMA/ Ca 2ϩ ionophore stimulation. Our data suggest that this coactivation of GM-CSF by CSD and RelA proteins is via proteinprotein interactions. We demonstrated this by truncating dbpB, removing the C terminus of the protein, which has been shown to be essential for CSD protein-protein interactions (11,13,30,(61)(62)(63). When the dbpB deletion construct removing the C terminus was used, PMA/Ca 2ϩ ionophore-mediated coactivation was significantly reduced as was the ability of dbpB and RelA to synergize, indicating that the coactivation was mediated primarily through the C-terminal region of the CSD pro-FIG. 6. CSD factors and RelA act to synergistically activate the GM-CSF proximal promoter. The GM-CSF promoter domain 1 luciferase reporter plasmid pGMCK-1(2)-TK/Luc was cotransfected into Jurkat T cells with pRcCMV, pRcCMVRelA, pSG5, pSGdbpAv, or pS-GdbpB. pGMCK-1(2)-TK/Luc and pRcCMVRelA were cotransfected together into Jurkat T cells with pSG5, pSGdbpAv, pSGdbpB, pSGd-bpB⌬2, and pSGdbpB⌬1. Values are shown relative to pGMCK-1(2)-TK/Luc cotransfected with the pRcCMV backbone plasmid, which is given a value of 1. tein and, therefore, most likely via protein-protein interactions. Raj et al. (15,32) find that CSD and RelA can interact in solution and that this interaction leads to increased RelA binding to the human polyomavirus JC viral promoter. In these experiments RelA conversely decreased CSD binding to its single-stranded DNA binding site. This CSD-RelA interaction was implicated in activation of the human polyomavirus JC early promoter (32). Similarly, we have found that RelA can decrease CSD binding to its GM-CSF single-stranded DNA binding site (data not shown).
Taken together our data suggest a model whereby, first, in unstimulated Jurkat T cells, CSD proteins bind to singlestranded DNA in a structure as described for the repression of the GM-CSF gene by CSD proteins in HEL fibroblasts (21)(22)(23). This model has been suggested for a number of other genes repressed by CSD proteins (26,27,50). Upon T cell stimulation (mimicked by PMA/Ca 2ϩ ionophore) it has been well documented that many transcriptional activators (like NF-B p50, and RelA) are up-regulated, whereas the level of CSD proteins in the nucleus remains constant (reviewed in Refs. 38 and 20). 2 We and others (15) 2 have shown that RelA is able to dissociate dbpB from its single-stranded target, and we propose that this results in destabilization of the single-stranded DNA structure, facilitating a return of the DNA to its double-stranded conformation . This would allow the transcriptional activators (like NF-B, AP-1, CBF, ETS/NFAT) to bind initiating transcriptional activation.
We have presented data here that indicate a functional CSD-RelA interaction that is involved in activation of the GM-CSF promoter, domain 1 region. We have also shown CSD mediated activation via the domain 2 region, which contains no RelA sites, and hence, CSD proteins may also be able to interact with other transcription factors binding to this region. There is a growing list of proteins where CSD protein interactions have been implicated to affect promoter function (NF-B, mitrochondrial SSB, ZO-1, heterogeneous nuclear ribonucleoprotein K, TBP, NF-Y, YY-1, AP-2, SP-1, Tat, PuR␣) (5, 10 -13, 15, 23, 29, 30, 55, 64). Potential targets on the GM-CSF promoter domain 2 for CSD interactions could be SP1 and YY1, both of which have been shown to regulate GM-CSF function in T cells (20,65,66).
The complement of CSD protein subtypes in the cell may also affect CSD function. We have previously observed in HEL fibroblasts (with 42-, 22-, and 25-kDa CSD proteins) that CSD proteins acted to repress activation, whereas in T cells (with only the 42-kDa protein) CSD factors function to activate when the cells were PMA/Ca 2ϩ ionophore-stimulated. The reason for this difference between the cell types is as yet unclear but may be related to the proposed lack of C-terminal-protein interaction sequences in the 22-and 25-kDa CSD subtypes observed in HEL fibroblasts (22, 31-33, 35, 53). Taking into account our previous results and the reported function associated with each of the CSD protein domains, the 42-kDa CSD factor is implicated in repression (via the CSD domain) and activation (via the C-terminal domain), whereas the 22-kDa CSD factors may be involved only in repression (21)(22)(23). In fibroblasts, where CSD factors act to repress GM-CSF promoter function, UVcross-linking experiments indicated that the 22-kDa CSD protein sub-population makes up a greater proportion than the 42-kDa sub-population, suggesting that the CSD intracellular environment is predisposed toward truncated forms of the CSD proteins and, hence, repression (21)(22)(23). In Jurkat T cells, where CSD factors act to mediate activation of the GM-CSF promoter, only the full-length 42-kDa CSD protein is present. The 42-kDa CSD protein, although still able to bind DNA and potentially repress in the unstimulated state, is also able to interact with proteins in a stimulated cell via the C-terminal protein-protein interaction domain to bring about activation. Therefore the differences in effect when full-length CSD factors were overexpressed in Jurkat T cells and HEL fibroblasts may be due to the intrinsic difference in the ratio of CSD factor subtypes and resulting intracellular environment already present in the two cell types.
We have shown, therefore, that CSD proteins can act to repress or activate the GM-CSF gene in different cell types. Repression is associated with the CSD binding sites located across the promoter, whereas activation may relate to CSD cooperation with other transcription factors. We also raise the possibility that the cellular content of CSD proteins predetermines the functional outcome in the transcription response.