A High Affinity HSF-1 Binding Site in the 5′-Untranslated Region of the Murine Tumor Necrosis Factor-α Gene Is a Transcriptional Repressor*

Tumor necrosis factor-α (TNFα) is a pivotal early mediator of host defenses that is essential for survival in infections. We previously reported that exposing macrophages to febrile range temperatures (FRT) (38.5–40 °C) markedly attenuates TNFα expression by causing abrupt and premature cessation of transcription. We showed that this inhibitory effect of FRT is mediated by an alternatively activated repressor form of heat shock factor 1 (HSF-1) and that a fragment of the TNFα gene comprising a minimal 85-nucleotide (nt) proximal promoter and the 138-nt 5′-untranslated region (UTR) was sufficient for mediating this effect. In the present study we have used an electrophoretic mobility shift assay (EMSA) to identify a high affinity binding site for HSF-1 in the 5′-UTR of the TNFα gene and have used a chromosome immunoprecipitation assay to show that HSF-1 binds to this region of the endogenous TNFα gene. Mutational inactivation of this site blocks the inhibitory effect of overexpressed HSF-1 on activity of the minimal TNFα promoter (−85/+138) in Raw 264.7 murine macrophages, identifying this site as an HSF-1-dependent repressor. However, the same mutation fails to block repression of a full-length (−1080/+138) TNFα promoter construct by HSF-1 overexpression, and HSF-1 binds to upstream sequences in the regions −1080/−845, −533/−196, and −326/−39 nt in EMSA, suggesting that additional HSF-1-dependent repressor elements are present upstream of the minimal −85-nt promoter. Furthermore, although mutation of the HSF-1 binding site in the minimalTNFα promoter construct abrogates HSF-1-mediated repression, the same mutation fails to abrogate repression of this construct by high levels of HSF-1 overexpression or exposure to 39.5 °C. This suggests that HSF-1 might repress TNFα transcription through redundant mechanisms, some of which might not require high affinity binding of HSF-1.

Tumor necrosis factor-␣ (TNF␣) is a pivotal early mediator of host defenses that is essential for survival in infections. We previously reported that exposing macrophages to febrile range temperatures (FRT) (38.5-40°C) markedly attenuates TNF␣ expression by causing abrupt and premature cessation of transcription. We showed that this inhibitory effect of FRT is mediated by an alternatively activated repressor form of heat shock factor 1 (HSF-1) and that a fragment of the TNF␣ gene comprising a minimal 85-nucleotide (nt) proximal promoter and the 138-nt 5-untranslated region (UTR) was sufficient for mediating this effect. In the present study we have used an electrophoretic mobility shift assay (EMSA) to identify a high affinity binding site for HSF-1 in the 5-UTR of the TNF␣ gene and have used a chromosome immunoprecipitation assay to show that HSF-1 binds to this region of the endogenous TNF␣ gene. Mutational inactivation of this site blocks the inhibitory effect of overexpressed HSF-1 on activity of the minimal TNF␣ promoter (؊85/؉138) in Raw 264.7 murine macrophages, identifying this site as an HSF-1-dependent repressor. However, the same mutation fails to block repression of a full-length (؊1080/؉138) TNF␣ promoter construct by HSF-1 overexpression, and HSF-1 binds to upstream sequences in the regions ؊1080/ ؊845, ؊533/؊196, and ؊326/؊39 nt in EMSA, suggesting that additional HSF-1-dependent repressor elements are present upstream of the minimal ؊85-nt promoter. Furthermore, although mutation of the HSF-1 binding site in the minimal TNF␣ promoter construct abrogates HSF-1mediated repression, the same mutation fails to abrogate repression of this construct by high levels of HSF-1 overexpression or exposure to 39.5°C. This suggests that HSF-1 might repress TNF␣ transcription through redundant mechanisms, some of which might not require high affinity binding of HSF-1.
Tumor necrosis factor-␣ (TNF␣) 1 is an early pivotal mediator expressed in response to infection and injury (1). Although TNF␣ is essential for optimal host defense, persistent or inappropriately high TNF␣ expression has grave consequences, including multiorgan failure and death (2)(3)(4)(5). The pleiotropic nature of TNF␣ has lead to the evolution of stringent and redundant regulatory mechanisms imposed at transcriptional, translational, and posttranslational levels (6 -11).
We reported that exposure to febrile range hyperthermia suppresses TNF␣ expression in murine peritoneal macrophages, Kupffer cells, precision-cut liver slices, the murine Raw 264.7 macrophage cell line, human monocyte-derived macrophages, and the THP1 monocyte cell line (10,(12)(13)(14)(15)(16). We showed that the predominant mechanism of suppression of TNF␣ expression is by an abrupt and early cessation of TNF␣ transcription, and that the TNF␣ gene sequence between Ϫ85 and ϩ138 is sufficient to confer temperature responsiveness in murine macrophages (15). We also showed that the heat stressactivated transcription factor, heat shock transcription factor 1 (HSF-1) is activated at febrile range temperatures (FRT) to an alternate DNA binding form that acts as a repressor of gene expression (15) and binds to the minimal temperature-responsive TNF␣ gene sequence (nt Ϫ85/ϩ138). Furthermore, overexpression of HSF-1 represses the activity of a luciferase reporter construct-driven minimal TNF␣ promoter sequence (15).
HSF-1 is activated in response to various chemical and thermal stresses through a cascade of posttranslational modifications, including trimerization, nuclear translocation, DNA binding, and phosphorylation of its transactivation domain, the result being the activation of heat shock protein gene transcription (17)(18)(19). HSF-1 binds to conserved regulatory sequences known as heat shock response elements (HRE), a pentanucleotide nGAAn element that forms stable binding sites for HSF-1 when oriented in inverted dyad repeats (20). Recently, several studies have shown that HSF-1 can also act as a negative regulator of certain non-heat shock genes, including interleukin (IL)-1␤, c-fos, urokinase, and TNF␣ (15,21,22). In the case of IL-1␤, this effect requires stable binding of HSF-1 to the IL-1␤ promoter adjacent to an essential NF-IL-6 binding element (21), whereas, in the case of urokinase and c-fos the effect apparently does not require stable DNA binding (22).
The minimal temperature-responsive TNF␣ gene sequence (Ϫ85/ϩ138) was also inhibited by HSF-1 overexpression, but this sequence did not contain a complete cognate HSF-1 bind-ing sequence. It does, however, contain multiple nGAAn elements positioned in critical locations, including adjacent to an essential Sp-1 binding site, at the transcription start site, and 35 nucleotides downstream of the transcription start site. In the present study we have identified the high affinity binding site for HSF-1 in the minimal temperature-responsive TNF␣ gene sequence and showed that inactivating the site by mutation reverses the repression of this promoter fragment by HSF-1 overexpression.
Plasmids-TNF␣ promoter-luciferase reporter constructs used in the study have been described earlier (15). Three constructs in the pGL3 vector (Promega, Madison, WI), namely pTNF Ϫ1080/ϩ138 (Ϫ1080 to ϩ138 nt), pTNF Ϫ244ϩ138 (Ϫ244/ϩ138 nt), and pTNF Ϫ85/ϩ138 (Ϫ85/ϩ138 nt), respectively, were used in this study. For mutated constructs, mutations were introduced into pTNF Ϫ1080/ϩ138 plasmid (Mut-pTNF Ϫ1080/ϩ138 ) using a modification of the overlap extension PCR method. In the first step, the 5Ј and 3Ј partially overlapping fragments containing the desired mutant were amplified using Pfu DNA polymerase. Outside primers Luc_5 and Luc_3 that correspond to flanking pGL3 backbone sequences and complementary internal primers (49_Mut forward and reverse primers), which generated a 36-nt overlap in the gene fragment and introduced a 3-base (GAA to CCC at ϩ50 to ϩ52 nt) substitution (see Fig. 4), were used for the amplification. Following first step PCR, the products were gel-purified and the second amplification was performed with equal concentrations of the two first step products as template and the outside primers Luc5 and Luc3 using Taq polymerase. The final product was digested with KpnI and HindIII, gel-purified, and cloned into the KpnI/HindIII site of pGL3. Mutated constructs corresponding to pTNF Ϫ244/ϩ138 and pTNF Ϫ85/ϩ138 (Mut-pTNF Ϫ244/ϩ138 and Mut-pTNF Ϫ85/ϩ138 , respectively) were prepared by amplifying the respective fragments from Mut-pTNF Ϫ1080/ϩ138 using Taq polymerase and cloning the product into pGL3 vector as described earlier (15). The sequences of all constructs were confirmed by automated dideoxy sequencing (University of Maryland Biopolymer Core Laboratories).
Recombinant Human HSF-1-A glutathione S-transferase-human HSF-1 fusion construct in pGEX-2T (Amersham Biosciences, Inc., Piscataway, NJ) was expressed and purified over a glutathione-Sepharose 4B column (Amersham Biosciences, Inc.) according to the manufacturer's instructions. The fusion protein was cleaved using thrombin, and the purified HSF-1 protein was used in the study.
Cell Culture-The Raw 264.7 mouse macrophage cell line was purchased from the American Type Cell Collection (ATCC, Rockville, MD) and maintained in complete RPMI 1640 medium supplemented with 50 units/ml penicillin, 50 g/ml streptomycin, 2 mM L-glutamine, 1 mM sodium pyruvate, 10 mM HEPES buffer (Invitrogen, Gaithersburg, MD), pH 7.3, and containing 10% defined fetal bovine serum (fetal bovine serum, HyClone, Logan, UT) at 37°C in 5% CO 2 -enriched air. Cells were routinely tested for Mycoplasma infection using a commercial assay system (MycoTest, Invitrogen), and new cultures were established monthly from frozen stocks. All media and reagents contained less than 0.1 ng/ml endotoxin as determined by Limulus amebocyte lysate assay (Associates of Cape Cod, Falmouth, MA). Cell viability was determined by trypan blue dye exclusion. Cells were stimulated with LPS that was prepared by trichloroacetic acid precipitation from Escherichia coli 0111:B4 (Difco, Detroit, MI).
EMSA-Nuclear extract from Raw cells were prepared according to the method of Schreiber et al. (23) as we previously described (15), and total protein was measured using a commercial reagent (Bio-Rad, Hercules, CA) with bovine serum albumin as standard. Double-stranded oligonucleotides were radiolabeled using T4 polynucleotide kinase (Promega) and [␥-32 P]ATP according to the manufacturer's protocol. EMSA reactions containing 5 g of nuclear extract or the indicated amount of recombinant HSF-1, 0.035 pmol of radiolabeled oligonucleotide, 1 g of poly(dI/dC), 10 mM Tris-HCl, pH 7.8, 10% glycerol, 60 mM NaCl, 1 mM EDTA, and 1 mM dithiothreitol in volume of 20 l were incubated at room temperature for 30 min. Where indicated, excess unlabeled competitor double-stranded oligonucleotide or 1 l of anti-HSF-1 antibody (Santa Cruz Biotechnology, Santa Cruz, CA) were incubated with the nuclear extracts for 30 min at room temperature before the addition of the radiolabeled probe. The DNA⅐protein complexes were then electrophoretically resolved on 4% nondenaturing polyacrylamide gels. The dried gels were analyzed by phosphorimaging (PhosphorImager, Molecular Dynamics) and subsequently exposed to x-ray film.
Chromosomal Immunoprecipitation Assay-ChIP assay was performed using a kit from Upstate Biotechnology Inc. (Lake Placid, NY). Unless otherwise stated, all reagents were provided in the kit. In brief, Raw 264.7 cells were incubated at 39.5°C for 1 h and fixed by adding formaldehyde (Sigma Chemical Co., St. Louis, MO) to the medium to a final concentration of 1%. After 15 min the cells were washed with phosphate-buffered saline containing 1 M phenylmethylsulfonyl fluoride and 2 g/ml aprotinin and were collected by centrifugation. Cell pellets were resuspended in SDS lysis buffer and sonicated for three 10-s bursts using a Branson Sonifier 450 (duty cycle and output settings were 30 and 3, respectively). Sonicated cell lysates were diluted 10-fold using ChIP dilution buffer and precleared for 1 h at 4°C using 80 l of a 50% salmon sperm DNA saturated protein A-agarose beads. Immunoprecipitation was carried out at 4°C overnight, and immune complexes were collected with salmon sperm DNA saturated protein Aagarose beads. Antibodies used included two rabbit anti-HSF-1 antibodies (from Santa Cruz Biotechnologies and Stressgen) or, to control for nonspecific interaction, rabbit anti-IL-13 (R&D) or no antibody. After washing three times with immune complex wash buffer and twice with TE buffer, the complexes were eluted with 0.1 M NaHCO 3 and 1% SDS. Protein-DNA cross-links were reverted by incubating at 65°C for 4 h, and after proteinase K digestion, DNA was extracted with phenolchloroform and precipitated using ethanol. PCR was performed (30 cycles, denaturing at 94°C for 45 s, annealing at 63°C for 30 s, and extension at 72°C for 45 s) using primers specific for the murine TNF␣ sequence between Ϫ85 and ϩ138: 5Ј-ggatcctgtgctagcttccgagggttgaatgaga (forward) and 5Ј-ttcgaagcttggagatgtgcgccttg (reverse). As a positive control, immunoprecipitated DNA was also amplified using PCR primers specific for an HRE-containing 180-nt fragment of the murine HSP70 promoter: 5Ј-aactccgattactcaagggaggc (forward) and 5Јgattctgagtagctgtcagcg (reverse) using the same PCR conditions as for TNF␣ except for a 60°C annealing temperature.
Transfection and Reporter Gene Analysis-Cells were transfected using FuGENE 6 (Roche Molecular Biochemicals). 4 g of each test plasmid and 0.5 g of control (pRL-SV40, Promega) plasmid DNA were mixed with 15 l of FuGENE 6 in 100 l of medium. The mixture was incubated at room temperature for 15 min and then added to cells in 60-mm dishes. After 24 h, the cells were split into 24-well plates (1:12 per 60-mm plate). After an additional 24 h, the cells were stimulated with LPS at 37°or 39.5°C for 6 h. Cells were lysed, and reporter gene expression was analyzed using the Dual Luciferase Reporter assay kit (Promega) according to the manufacturer's protocol.
Statistical Analysis-Data are presented as mean Ϯ S.E. Differences between two groups of data were analyzed using the unpaired Student t test. Differences among more than two groups were tested by applying the Fisher protected least significant differences test applied to a oneway analysis of variance.

RESULTS
Organization of the Murine Minimal Promoter and 5Ј-Untranslated Region-We previously reported that the murine TNF␣ gene sequence spanning Ϫ85 to ϩ138 nt bound HSF-1 in EMSA competition assays and, when transfected into Raw 264.7 macrophages, conferred transcriptional repression by febrile range temperature (FRT; 39.5°C) or HSF-1 overexpression (15). The high affinity binding sequence for HSF-1 com-prises a minimum of two nGAAn elements arranged as an inverted dyad repeat (20) (Fig. 1A). Fig. 1 shows the location of nGAAn elements in the Ϫ85 to ϩ138 nt region of the murine TNF␣ promoter (Fig. 1C), the sequence of the heat shock response element (HRE) from the human HSP70 promoter (24) (Fig. 1B) and the sequences of each of the TNF␣ oligonucleotides used for the EMSA analysis of HSF-1 binding (Fig. 1C, underlined sequences). This fragment of the TNF␣ gene contains 10 nGAAn elements, but none arranged with perfect inverted dyad symmetry. Single nGAAn elements are positioned next to the Sp-1 binding site (Ϫ65 nt), at the transcription start site (Ϫ8 nt), and at nt ϩ72 and ϩ108 in the 5Ј-UTR. The 5Ј-UTR sequence spanning ϩ30 to ϩ68 contains an array of four nGAAn elements, three of which would form a perfect HRE if the "CT" sequence at nt ϩ57 and ϩ58 were inverted to "TC." In our EMSA analysis, we probed with oligonucleotides spanning Ϫ83/Ϫ43 and containing three nGAAn elements and the Sp-1 binding sequence, Ϫ15/ϩ5 containing one nGAAn and the transcription start site, ϩ30/ϩ68 and containing the imperfect HRE, and ϩ65/ϩ85 and ϩ101/ϩ121, each containing one nGAAn element.
Identification of the High Affinity HSF-1 Binding Sites in the Murine TNF␣ Gene-The capacity of each nGAAn-containing sequence in the minimal TNF␣ promoter/5Ј-UTR to bind HSF-1 was analyzed by EMSA using the oligonucleotides listed in Fig.  1 using the following three-step strategy. First, the capacity of a 100-fold excess of each oligonucleotide to compete and block HSF binding to a consensus HSF binding sequence was analyzed using the sequence from the human HSP70 promoter as radiolabeled probe and nuclear extracts from Raw 264.7 macrophages exposed to 39.5°C for 1 h as a source of HSF-1 ( Fig. 2A). Of the murine TNF␣ oligonucleotides analyzed, only ϩ30/ϩ68 blocked HSF-1 binding to the labeled HSP70 HRE sequence probe (Fig. 2A, lane 5). Competition for binding by this TNF␣ oligonucleotide was as complete as that of a comparable concentration of the HSP70 sequence itself (lane 2). By comparison, each of the other TNF␣ oligonucleotides studied (lanes 3, 4, 6, 7) failed to compete for HSF-1 binding. To confirm that HSF-1 binds with high affinity to the ϩ30/ϩ68 TNF␣ sequence, we repeated the EMSA analysis using each TNF␣ oligonucleotide as a radiolabeled EMSA probe (Fig. 2B). Of the TNF␣ sequences studied, only ϩ30/ϩ68 bound HSF-1 (lane 4), forming a complex that comigrated with the complex that formed on the HRE sequence from the HSP70 promoter (lane 1). Supershifting with anti-HSF-1 antibody (lane 7) confirmed that the observed complex contained HSF-1. To further confirm that HSF-1 could directly bind to TNF␣ ϩ30/ϩ68, we repeated the EMSA analysis with purified recombinant human HSF-1 (Fig.  3). TNF␣ ϩ30/ϩ68 bound rHSF-1 (lanes 7-10), but to a lesser degree than the HSP70 HRE (lanes 2-5). By comparison, the other TNF␣ sequence oligonucleotides failed to detectably bind rHSF-1 (data not shown).
Mutational Inactivation of the HSF-1 Binding Site in TNF␣ 30/68 -The ϩ30/ϩ68 TNF␣ sequence contains four nGAAn sites that form two possible partially overlapping HREs centered on nt ϩ49 and ϩ59, respectively (Fig. 1C). To further define the binding site for HSF-1 in this sequence we replaced the GAA sequences at either 50 -52 nt (49_Mut) or 60 -62 nt (59_Mut) with CCC (Fig. 4A) and analyzed the ability of the mutated ϩ30/ϩ68 oligonucleotides to bind HSF-1 in an EMSA competition assay (Fig. 4B). Using the wild-type ϩ30/ϩ68 oligonucleotide as the radiolabeled probe and nuclear extracts from Raw 264.7 cells exposed to 39.5°C for 1 h as a source of HSF-1, we found that unlabeled wild-type ϩ30/ϩ68 added at a 10-fold molar excess was sufficient to abrogate binding to radiolabeled ϩ30/ϩ68 (lane 2). Mutating the nGAAn sequence at ϩ59 (59_Mut, lanes 9 -12) did not change the capacity of ϩ30/ ϩ68 to compete for HSF-1 binding. In striking contrast, 49_Mut failed to compete for HSF-1 binding when added at 10to 100-fold excess (lanes 5-7) and only partially competed when added at 1000-fold excess (lane 8). Direct binding of HSF-1 to radiolabeled probe containing the wild-type and mutated ϩ30/ ϩ68 sequences (Fig. 3C) 4) similar to that formed on wild-type ϩ30/ϩ68 (lane 1), 49_Mut failed to bind HSF-1 under these conditions (lane 3). We extended these observations by introducing the GAA to CCC substitution at ϩ50/Ϫ52 into the full-length TNF␣ promoter/5Ј-UTR (Ϫ1080/ϩ138) sequence and analyzing HSF-1 binding to a PCR-amplified Ϫ85/ϩ138 fragment of the wildtype and mutated gene (Fig. 5). EMSA was performed by using each PCR product as radiolabeled probe, and nuclear extracts from Raw 264.7 cells exposed to 39.5°C for 1 h as a source of HSF-1. The wild-type TNF␣ PCR product formed a complex (lane 1) that was competed for by unlabeled wild-type ϩ30/ϩ68 oligonucleotide (lane 3) but not by 49_Mut oligonucleotide (lane 4) and was supershifted by anti-HSF-1 antibody (lane 5). In contrast, the comparable TNF␣ sequence containing the GAA to CCC mutation at ϩ50/Ϫ52 failed to form a detectable complex (lane 2). Thus, it appears that the HRE-like sequence centered on nt ϩ49 in the 5Ј-UTR is the only site in the minimal promoter/ 5Ј-UTR sequence (Ϫ85/ϩ138) that forms high affinity complexes with HSF-1 under these cell-free assay conditions.
Binding of HSF-1 to the Endogenous TNF␣ Gene-Although we have demonstrated that HSF-1 binds to naked DNA under artificial cell-free conditions, its ability to repress TNF␣ transcription in the cell requires it to bind to the endogenous TNF␣ gene in vivo. We used the ChIP assay to determine if HSF-1 binds to the TNF␣ gene in vivo in Raw 264.7 cells incubated at 39.5°C for 60 min. DNA was sonicated to yield fragments of ϳ500-nt length. PCR with primers specific for the HRE-containing murine HSP70 promoter sequence amplified detectable product of the predicted 180-nt length in samples immunoprecipitated with each of the anti-HSF-1 antibodies (Fig. 6, lanes  3, 4), whereas no PCR product was detectable in samples immunoprecipitated without antibody (lane 5) or with an irrelevant rabbit anti-IL-13 antibody (lane 2), thereby validating the ChIP assay in this model. PCR amplification using primers specific for the TNF␣ sequence spanning Ϫ85 to ϩ138, which includes the putative HSF-1 binding site, generated a detectable product of the predicted 223-nt length in samples immunoprecipitated with the anti-HSF-1 antibodies (Fig. 6, lanes 8, 9) but not in samples immunoprecipitated without antibody (lane 10) or with anti-IL-13 antibody (lane 7).
Effect of Mutational Inactivation of HSF-1 Binding on TNF␣ Transcriptional Activity-The functional significance of abrogating high affinity HSF-1 binding to the TNF␣ 5Ј-UTR in comparison with transcriptional repression was analyzed by introducing the GAA to CCC mutation at nt 50 -52 into each of three TNF␣ promoter-driven luciferase reporter constructs: 1) the minimal promoter/5Ј-UTR (pTNF Ϫ85/ϩ138 ) (Fig. 7A); 2) a 382-nt promoter/5Ј-UTR fragment (pTNF Ϫ244/ϩ138 ) containing the most proximal NF-B response element (Fig. 7B); and 3) a full-length 1.2-kb promoter/5Ј-UTR fragment (pTNF Ϫ1080/ϩ138 ) (Fig. 7C). Each of these promoter fragments was cloned into the NheI/HindIII site of pGL3, transiently transfected into Raw 264.7 cells, and the promoter activity was compared in 37°and 39.5°C cell cultures and in the presence of increasing concentrations of an HSF-1 expression plasmid at 37°C (15). The analysis was performed in both the presence and absence of LPS. Although TNF␣ promoter activity was higher in the presence of LPS, the pattern of FRT-and HSF-1-induced effects on wild-type and mutated TNF␣ reporter constructs was comparable in LPS-stimulated and unstimulated cells. The data from the LPS-stimulated cells are shown. HSF-1 overexpression reduced the activity of both wild-type pTNF Ϫ1080/ϩ138 (Fig. 7C) and pTNF Ϫ85/ϩ138 (Fig. 7A), as we have previously reported (15), and similarly reduced the activity of pTNF Ϫ244/ϩ138 (Fig.  7B). Mutation of the high affinity HSF-1 binding site (at nt ϩ50/Ϫ52) in pTNF Ϫ85/ϩ138 seemed to interfere with the inhibitory effect of HSF-1 overexpression (Fig. 7A). At 1 and 2 g of HSF-1 expression plasmid, the activity of the wild-type pTNF Ϫ85/ϩ138 construct was decreased by 28 and 40%, respectively; whereas, activity of the mutant construct was unchanged. However, cotransfection with 3 g of HSF-1 plasmid or incubation at 39.5°C inhibited the activity of the wild-type and mutated pTNF Ϫ85/ϩ138 construct by a similar extent.
HSF-1 overexpression inhibited both wild-type and mutated pTNF Ϫ244/ϩ138 (Fig. 7B), but the extent of inhibition of the mutated construct was significantly less than that of the wildtype construct at 2 g (22 versus 42%; p Ͻ 0.05) and 3 g of HSF-1 plasmid (36 versus 59%; p Ͻ 0.05). Both constructs were inhibited to a similar extent by exposing cells to 39.5°C. In contrast, in the full-length promoter (pTNF Ϫ1080/ϩ138 , Fig. 7C), mutation of the high affinity HSF-1 binding site in the 5Ј-UTR had no detectable effect on the inhibitory effects of HSF-1 overexpression or exposure to 39.5°C.
Interaction of HSF-1 with TNF␣ Promoter Sequences Upstream of Ϫ85-The transfection results suggested that additional HSF-1-responsive repressor elements might be present in the TNF␣ sequence upstream of the minimal promoter. To determine if HSF-1 binds to sequences upstream of Ϫ85, we prepared five partially overlapping fragments by PCR amplification spanning the region Ϫ39 to Ϫ1080. The amplified fragments were radiolabeled and used as a probe in EMSA using FIG. 2. EMSA analysis of HSF binding to TNF␣ oligonucleotides. Nuclear extracts from Raw 264.7 cells exposed to 39.5°C for 1 h were used as a source of HSF-1. A, competition for binding to HSF binding sequence from the human HSP70 promoter (HRE). Indicated unlabeled oligonucleotides were added at a 100-fold molar excess. The HSF doublet bands are indicted by the arrows. B, direct binding of HSF to radiolabeled HRE or individual TNF␣ oligonucleotides. In lane 7, the HRE-containing complex was supershifted with anti-HSF-1 (supershifted complex is indicated by the arrowhead).

DISCUSSION
In our earlier studies, we showed that TNF␣ expression is reduced in human and murine macrophages upon exposure to febrile temperature (10,12,15,16) and that the effect is largely caused by a reduction in TNF␣ transcription in the warmer FIG. 3. EMSA analysis of recombinant HSF-1 binding to the HSF binding sequence from the human HSP70 promoter and TNF␣ oligonucleotide ؉30/؉68. Radiolabeled oligonucleotide comprising the HSF binding sequence from the human HSP70 gene (HRE; lanes [1][2][3][4][5] or the TNF␣ ϩ30/ϩ68 sequence (lanes 6 -10) was incubated with the indicated concentration of recombinant human HSF-1. The doublet bands representing the HSF⅐DNA complex are indicted by the arrows .   FIG. 4. EMSA analysis of HSF binding to wild-type and mutated TNF␣ oligonucleotide ؉30/؉68. A, the sequences of the wild-type and two mutated ϩ30/ϩ68 oligonucleotides. B, competition for binding to wild-type ϩ30/ϩ68. Nuclear extracts from Raw 264.7 cells exposed to 39.5°C for 1 h were used as a source of HSF-1 and probed with radiolabeled wild-type ϩ30/ϩ68. Each unlabeled oligonucleotide was added at the indicated molar excess. The HSF⅐DNA doublet band is indicted by the arrow. C, direct binding of HSF to radiolabeled wildtype and mutated TNF␣ oligonucleotides. The complex on ϩ30/ϩ68 was supershifted with anti-HSF-1 antibody (lane 2), and the supershifted complex is represented by the arrowhead.
cells. We showed that exposing Raw 264.7 cells to febrile range temperature (39.5°C) activates the heat stress-activated transcription factor HSF-1, but although HSF-1 in the 39.5°C cell culture binds to its cognate DNA binding site, it fails to activate transcription from the HSP70 promoter. We showed that HSF-1 activation correlates with cessation of TNF␣ transcription in 39.5°C Raw 264.7 cell culture and HSF-1 overexpression reduces TNF␣ promoter activity in 37°C Raw 264.7 cell culture. Together, these data offered a persuasive argument that HSF-1 might act as a repressor of macrophage TNF␣ expression during exposure to febrile temperatures. This hypothesis was further supported by the studies of Xiao et al. (25) who showed that HSF-1 knockout mice demonstrated exaggerated TNF␣ expression following challenge with LPS compared with HSF-1-sufficient littermates. We previously showed that HSF-1 binds to the murine TNF␣ proximal promoter/5Ј-UTR sequence spanning nt Ϫ85 to ϩ138 under EMSA conditions. We have extended these findings in our present study by localizing the HSF-1 binding site between nt ϩ30 and 68 in the 5Ј-UTR of the murine TNF␣ gene. The sequence between nt ϩ49 and ϩ58, AGAACATCTT, with the exception of a simple inversion of the 3Ј-terminal CT dinucleotide (underlined), is the canonical HSF-1 binding sequence. We used EMSA analysis to show that this sequence specifically binds HSF-1 and that mutation of the nGAAn pentanucleotide at nt ϩ49/ϩ53 prevents binding of HSF-1, whereas the same substitution in a downstream GAAn sequence (nt ϩ60/Ϫ62) had no effect on HSF-1 binding. However, the inability of DNA⅐protein com- Immunoprecipitates were analyzed by PCR using primers specific for the HRE-containing HSP70 promoter sequence (lanes [1][2][3][4][5] or the TNF␣ sequence between Ϫ85 and ϩ138 containing the putative HSF-1 binding site (lanes 6 -10). Samples were immunoprecipitated either without antibody (lane 5, 10), with anti-HSF-1 antibody from two different sources, Stressgen (H1, lanes 3,8) or Santa Cruz Biotechnologies (H2, lanes 4,9), or with an irrelevant anti-human IL-13 antibody (lanes 2, 7). Lane 11 contains a 100-bp ladder. Lanes 1 and 6 contain the product of the PCR reaction with plasmid containing either the HSP70 promoter or the TNF␣ gene, respectively, added as template. Ϫ244/ϩ138 (B), or Ϫ1080/ϩ138 (C) and comparable fragments with a GAA to CCC mutation at nt ϩ50/Ϫ52 (MUT) were introduced into the PGL3 luciferase reporter construct and cotransfected into Raw 264.7 cells with the control plasmid pRL-SV40 alone or along with the indicated amount of HSF-1 expression plasmid. Cells transfected without the HSF-1 transfection plasmid were incubated at either 37°or 39.5°C. All cells were stimulated with 100 ng/ml LPS and harvested 6 h later. The ratio of experimental to control luciferase activity was calculated, and data were calculated as percentage of control values in cells cultured at 37°C without HSF-1. Mean Ϯ S.E.; n ϭ 6. *, p Ͻ 0.05 versus 0 HSF-1; †, p Ͻ 0.05 versus comparably treated wild-type-transfected cells.
plexes to form under EMSA conditions does not necessarily exclude an interaction in vivo. To provide further evidence that HSF-1 might bind to and repress the TNF␣ gene in vivo, we using the ChIP assay to determine if HSF-1 interacts with the TNF␣ gene in the living cell. We incubated Raw 264.7 cells for 1 h at 39.5°C, conditions that we have shown activate HSF-1 to a form that binds HSP70 promoter and TNF␣ gene sequences under cell-free EMSA conditions. We showed that two different anti-HSF-1 antibodies coimmunoprecipitated both HRE-containing HSP70 promoter sequence and TNF␣ sequence containing the putative HSF-1 binding site. Negative controls, immunoprecipitated without antibody or with an irrelevant antibody (anti-IL-13) from the same species (rabbit), did not contain detectable HSP70 or TNF␣ gene fragments demonstrating the specificity of the technique.
In transient transfection studies, the TNFϪ85/ϩ138 reporter construct containing a GAA to CCC substitution at nt ϩ50/Ϫ52 was significantly less sensitive to the inhibitory effect of overexpressed HSF-1 than was the wild-type construct. In fact, cotransfection of this mutated construct with 1 or 2 g of HSF-1 expression plasmid had no inhibitory effect on luciferase activity whereas the wild-type construct was inhibited by 30 -40% when cotransfected with the same amount of HSF-1 expression plasmid. These data suggest that HSF-1 represses transcription from the minimal TNF␣ promoter/5Ј-UTR fragment, in part, by binding to the HRE-like sequence at ϩ49/Ϫ58 nt within the 5Ј-UTR. The capacity of elements within the 5Ј-UTR region of genes to repress transcription has been reported to occur in other genes, including the collagen ␣1 gene and the potassium channel Kv3.1 gene (26,27). Based on the location of the putative repressor site in the murine TNF␣ gene, only 30 -60 nt downstream of the transcription start site, HSF-1 might repress TNF␣ transcription by blocking RNA polymerase processivity, as has been shown in the phage T7 model system in which binding of the lac repressor 13-15 nucleotides downstream of the initiation site blocks T7 RNA polymerase processivity (28).
Interestingly, cotransfection with 3 g of HSF-1 or incubation at 39.5°C caused comparable reductions in activity of the mutated and wild-type TNFϪ85/ϩ138 reporter constructs. The capacity of higher HSF-1 levels to repress the TNFϪ85/ ϩ138 construct containing the mutation at ϩ50/ϩ52 suggests that there might be additional pathways through which HSF-1 represses expression of the minimal TNF␣ promoter. These include lower affinity binding of HSF-1 to non-canonical HSF-1 binding sites elsewhere in this gene fragment that is not detected by EMSA yet still might block transcription complex formation. Alternatively, HSF-1 might repress TNF␣ transcription without directly binding to DNA. Chen et al. (22) have suggested that HSF-1 could interact either with an upstream signal transduction component or with a coactivating factor to specifically repress genes such as c-fos and urokinase-type plasminogen activator.
In contrast to the minimal TNF␣ promoter, the full-length 1.2-kb TNF␣ promoter construct was comparably inhibited by 1-3 g of HSF-1, and this inhibition was unaffected by the presence or absence of the GAA to CCC mutation at ϩ50/ϩ52. The intermediate length construct, TNFϪ244/ϩ138, exhibited a third pattern of responsiveness to the GAA to CCC mutation at ϩ50/Ϫ52 and to overexpression of HSF-1. The activity of the wild-type and mutated TNFϪ244/ϩ138 was comparably inhibited by the lowest HSF-1 expression but, when cotransfected with 2 and 3 g of HSF-1 expression plasmid, the wildtype construct was inhibited to a greater extent than the mutated construct. On the other hand, exposure to 39.5°C reduced the activity of all three constructs, and the repression was unaffected by mutational inactivation of the HSF-1 binding site at nt ϩ49/ϩ58.
The different patterns of response of each of the three constructs to HSF-1 overexpression and the mutational inactivation of the HSF-1 binding site at nt ϩ49/Ϫ58 suggest that additional sequences upstream of nt Ϫ85 might mediate transcriptional repression by HSF-1. Using overlapping ϳ200-nt PCR-generated fragments of TNF␣ upstream sequence, the sequence upstream of nt Ϫ85 was scanned for HSF-1 binding. Complex formation with rHSF-1 was detectable to the regions Ϫ1080/Ϫ845, Ϫ533/Ϫ196, and Ϫ326/Ϫ39 suggesting that redundant HSF-1 binding sites might be active in the TNF␣ gene. Alternatively, or in addition to direct binding to DNA, HSF-1 might interfere with transcription by binding to transcriptional activators as has been shown for STAT (29,30), TFIID, and related factors (31).
In summary, we identified a unique HRE-like sequence in the murine TNF␣ 5Ј-UTR that binds HSF-1 and is required for HSF-1-mediated transcriptional repression in the minimal mouse TNF␣ promoter. This offers proof of the concept that HSF-1 can repress gene transcription by binding to the 5Ј-UTR. Although the 5Ј-UTR HRE-like sequence of the mouse TNF␣ gene is not present in the human TNF␣ 5Ј-UTR, the results of this study suggest a number of additional redundant pathways through which exposure to febrile range temperature might inhibit transcription of both murine and human TNF␣.