A 3 (cid:42) cis -Acting Element Is Involved in Tumor Necrosis Factor- (cid:97) Gene Expression in Astrocytes*

Tumor necrosis factor- (cid:97) (TNF- (cid:97) ) contributes to demyelinating diseases in the central nervous system. Astro- cytes, the major glial cells in the CNS, do not constitutively express TNF- (cid:97) , but the TNF- (cid:97) gene is transcriptionally activated in response to a variety of stimuli, including TNF- (cid:97) itself. Because of the importance of TNF- (cid:97) in the CNS, we examined the mechanisms underlying transcriptional regulation of the TNF- (cid:97) gene in astrocytes. In transient transfection assays, a plasmid construct containing 1.3 kilobase pairs (kb) of 5 (cid:42) flanking sequence of the rat TNF- (cid:97) gene showed high basal activity that could not be further enhanced by TNF- (cid:97) stimulation. A “marked” 10-kb TNF- (cid:97) gene construct, which contains the whole TNF- (cid:98) gene with 1.2 kb of 5 (cid:42) flanking sequence, 1.1 kb of intergenic sequence, and the whole TNF- (cid:97) gene with 3 kb of 3 (cid:42) flanking sequence, was able to respond to TNF- (cid:97) stimulation. Analysis of a series of 5 (cid:42) and 3 (cid:42) deletion constructs of the marked TNF- (cid:97) genes demonstrated that upstream sequence elements such as NF- (cid:107) B are not required for TNF- (cid:97) induction and that TNF- (cid:97) responsive elements are located in the 3 (cid:42) flanking region of the TNF- (cid:97) gene. We also found that a TNF- (cid:97) -inducible DNase I-hypersensitive (DH) site is present in this 3 (cid:42) region whose deletion abolishes TNF- (cid:97) inducibility of the marked TNF- (cid:97) gene. Electrophoresis mobility shift assays showed that TNF- (cid:97) -inducible nuclear proteins, consisting of p50 and p65 NF- (cid:107) B proteins, specifically bind to two consecutive NF- (cid:107) B binding sites within the 3 (cid:42) DH site. These results indicate that TNF- (cid:97) -induced TNF- (cid:97) gene expression in astrocytes involves p50 and p65 NF- (cid:107) B proteins binding to downstream NF- (cid:107) B sites and concomitant modulation of the chromatin structure.

Tumor necrosis factor-alpha (TNF-␣) 1 is a proinflammatory cytokine recognized to be an important mediator of immunological and inflammatory responses in a variety of tissues, including the central nervous system (CNS). TNF-␣ contributes to the pathogenesis of inflammatory demyelinating dis-eases such as multiple sclerosis by promoting infiltration of inflammatory cells into the CNS, intracerebral immune responses, cytokine production, astrogliosis, and demyelination (for review see Ref. 1). Astrocytes, the major glial cells in the CNS, are capable of producing TNF-␣ upon exposure to multiple stimuli (2)(3)(4) and thus can serve as an endogenous source of TNF-␣ within the CNS. We have previously shown that rat astrocytes are capable of producing TNF-␣ at both the mRNA and protein levels in response to lipopolysaccharide (LPS) and TNF-␣ (2,5). TNF-␣ is not constitutively expressed in astrocytes; however, the TNF-␣ gene is transcriptionally activated upon exposure to these stimuli (5,6).
Most of the previous studies on transcriptional regulation of the TNF-␣ gene have focused on the promoter regions of the mouse and human TNF-␣ genes. Thus far, however, results have not been conclusive regarding the cis-acting elements and transcription factors involved in TNF-␣ gene expression. The role of the transcription factor NF-B in the regulation of TNF-␣ gene expression is controversial. NF-B is essential for LPS induction of the mouse TNF-␣ promoter in mouse macrophages (7,8), whereas the three B sites in the human TNF-␣ promoter are neither required nor sufficient for virus or LPS induction of the TNF-␣ gene in mouse monocytic cell lines (9). Phorbol ester (PMA) activation of the human TNF-␣ promoter in the U937 macrophage cell line appears to be mediated by the transcription factor AP-1 (10,11). Analysis of the upstream region of the human TNF-␣ gene in U937 cells showed that the sequence of Ϫ95 from the transcription start site (TSS) is sufficient for PMA induction and that AP-1 plays an important role in both basal and PMA-induced activity of the human TNF-␣ promoter (10,11). In contrast, Leitman et al. (12) demonstrated that the proximal human TNF-␣ promoter containing only 28 nt upstream and 10 nt downstream of the TSS is sufficient for PMA induction in U937 cells; no other upstream sequences were required, and the TATA box structure was important. The minimum region of the human TNF-␣ promoter required for PMA activation in T and B cells has been localized between Ϫ52 and ϩ89 with respect to the TSS, which is different from that in macrophages (13). The TNF-␣ promoter also can be activated by TNF-␣ itself, which appears to be mediated by a palindrome sequence present between Ϫ125 and Ϫ82 that resembles the consensus binding sequences for AP-1, CREB, and ATF (14). In addition to NF-B and AP-1, other transcription factors such as NFAT (15), Egr-1 (16), C/EBP␤ (17), and Ets (18) also appear to be involved in TNF-␣ gene expression. Furthermore, an involvement of downstream sequences in the TNF-␣ gene has been suggested (19,20). Taken together, it appears that the cis-acting elements and trans-acting factors involved in TNF-␣ gene expression are diverse and that the molecular mechanisms underlying transcriptional activation of the TNF-␣ gene are species-, tissue-, and stimuli-specific.
Because of the biological importance of TNF-␣ in the CNS, we have been studying how this gene is regulated in astrocytes.
Using a series of marked rat TNF-␣ constructs, we have determined that upstream sequence elements such as the NF-B sites are not required for TNF-␣ induction of the TNF-␣ gene and that the TNF-␣ responsive elements are present in the 3Ј flanking region of the TNF-␣ gene. We have also found that a TNF-␣-inducible DH site is present in the 3Ј flanking region whose deletion abolishes TNF-␣ inducibility of the marked TNF-␣ gene. We then identified TNF-␣-inducible nuclear proteins consisting of the p50 and p65 members of the NF-B family that specifically bind to the NF-B sites within the 3Ј DH site. These data suggest an involvement of 3Ј B sites and modulation of the chromatin structure in TNF-␣ gene expression in astrocytes. These findings collectively indicate that regulation of TNF-␣ gene expression in astrocytes differs from that previously described for monocytes, T-cells, and B-cells, suggesting cell type-specific mechanisms for control of this important cytokine.

MATERIALS AND METHODS
Primary Glial Cell Cultures-Primary glial cell cultures were established from neonatal rat cerebra as described previously (2). Meninges were removed from rat brains prior to glial cell dissociation and culture. Culture medium was Dulbecco's modified Eagle's medium, high glucose formula supplemented with glucose to a final concentration of 6 g/liter, 2 mM glutamine, 0.1 mM nonessential amino acid mixture, 0.1% gentamycin, and 10% fetal bovine serum (HyClone Laboratories, Logan, UT). After 14 days in primary culture, oligodendrocytes were separated from the glial cultures by mechanical dislodging, and astrocytes were obtained by trypsinization (0.25% trypsin, 0.02% EDTA). Astrocytes were monitored for purity by immunofluorescence and by nonspecific esterase staining for contaminating microglia. Astrocyte cultures were routinely Ͼ97% positive for glial fibrillary acidic protein, an intracellular antigen unique to astrocytes (21), and less than 2% of the cells were microglia based on their positive staining for nonspecific esterase.
Plasmid Constructions-All the TNF constructs originated from the plasmids pRTX, pRTSI, or pRTSII (22). The restriction enzyme sites used for plasmid constructions are indicated on the map of the TNF locus ( Fig. 2A). A 1359-bp AvaI fragment of intergenic sequence was blunt-end ligated with the pBLCAT3 vector (23) that was linearized with BglII to generate either pTNF(ϩ1229)CAT or pTNF(Ϫ1229)CAT. 5Ј deletion constructs were generated by exonuclease III digestion of pTNF(Ϫ1229)CAT digested with SphI and BamHI. Deletion points were determined by dideoxy chain-termination sequencing (USB).
pTNF6.5 was constructed by sequential subcloning of the 2.3-kb EcoRI-SacI and the 4.2-kb SacI fragments of the TNF-␣ gene into EcoRI and SacI sites of pGEM3z. pTNF10 was constructed as follows; two SacI sites of pTNF6.5 were destroyed by two steps of cutting, removing protruding ends, and religating. A 4.9-kb HindIII-SalI (SalI is in the vector) fragment (3Ј-half of the TNF locus) was ligated with a 2.8-kb HindIII-SalI fragment of pGL2-Basic vector (Promega). The resulting plasmid was cut with SacI and XhoI and ligated with a 5.3-kb SacI-XhoI fragment (5Ј-half of the TNF locus) to generate pTNF10. A series of marked TNF-␣ gene deletion constructs were made as follows; a 2.1-kb SalI-XhoI fragment was cloned into the same sites of pGL2-Basic vector to make the BspEI site in the 5Ј-untranslated region of the TNF-␣ gene unique. An 8-bp synthetic NotI linker (GCGGCCGC, New England Biolabs) was inserted to the BspEI site that had been cut and bluntended, which resulted in an extra 12-bp (CCGGGCGGCCGC) introduced to the wild type sequence. The 1.1-kb SacII-XhoI fragment containing the NotI linker sequence was substituted for the same site fragment of pTNF6.5 to generate pTNF6.5*. pTNF10* was generated by substitution of a 5.6-kb SacII-BspEI (BspEI is in the 3Ј flanking region of the TNF-␣ gene) fragment of pTNF6.5* for the same sequence of pTNF10. pTNF5.9* was generated by cloning the SacII (blunt-ended)-SalI (in the vector) fragment into the SmaI-SalI sites of pGEM3z. The 3Ј deletion constructs of the marked TNF-␣ gene were generated by utilizing the unique restriction enzyme sites in the 3Ј flanking region of the TNF-␣ gene that were also present in the vector. pTNF4.4*, -3.6*, -3.4*, and -2.7* were generated by deletions from pTNF5.9* of XbaI, SphI, SmaI, and PstI fragments, respectively. The SacII (blunt-ended)-EcoRI fragment was cloned into SmaI-EcoRI sites of pGEM3z to generate pTNF2.9*. pGL2-Control vector used as the internal control was purchased from Promega.
The template plasmid for the TNF-␣ antisense RNA probe was prepared as follows. The sequence from Ϫ70 to ϩ400 of gene pTNF6.5* (see Fig. 2C) was amplified by PCR. The primers were designed such that EcoRI and BamHI sites were generated at the 5Ј and 3Ј ends of the PCR product, respectively. The upstream primer was AGTTTTCCGCGAAT-TCGGGTTGAATGAGAGCTTTTC and the downstream primer was CCTTTCTTGCGGATCCCCTCCCACCCTACTTTGCTT. After digestion with EcoRI and BamHI, the 482-bp PCR product was cloned into pGEM4z to generate pTNF5Ј*. The plasmid for luciferase antisense RNA probe, pLUC5Ј*, was prepared in the same way. The sequence from 121 to 439 (sequence number according to manufacturer) of pGL2-Control was amplified by PCR and cloned into pGEM4z (see Fig. 2C). The upstream primer was CCATTCTCCGAATTCGCCCCATGGCT-GACTAAT and the downstream primer was CGGACATTTGGATC-CCGAAGTATTCCGCGTACG.
Transfection and Stimulation of Astrocytes-Transient transfection of primary rat astrocytes was performed by electroporation using a Bio-Rad gene pulser as described previously (24). Astrocyte cultures were trypsinized, 7-10 ϫ 10 6 cells resuspended in 250 -400 l of complete culture medium containing 20% fetal bovine serum, mixed with 1-20 g of plasmid DNA (CsCl, purified two times) with the same molar ratios, incubated for at least 10 min on ice, and then pulsed at 250 V and 960 microfarads. Cells transfected with the same plasmids were pooled, replated in 100-mm dishes with 10 ml of culture medium containing 10% fetal bovine serum, and allowed to recover for 24 -48 h. Transfected astrocytes were stimulated with either medium alone or rat recombinant TNF-␣ (Biosource Int., Camarillo, CA) (500 units/ml) for 18 h for the CAT assay or 2 h for RNA analysis. The CAT assay was done with the xylene extraction method as described previously (24).
RNA Analysis-Total cellular RNA was isolated using the guanidinium method as described previously (6). The antisense RNA probes were generated with an in vitro transcription kit (Ambion, Austin, TX) as described previously (6). pTNF5Ј* or pLUC5Ј* linearized with EcoRI was used as template DNA for TNF or luciferase antisense RNA probes, respectively. RNase protection assay (RPA) was performed with the RPA kit according to the manufacturer's instructions (Ambion) as described previously (6). 90 g of total cellular RNA were hybridized overnight at 43-45°C with TNF-␣ (2-5 ϫ 10 5 cpm) and luciferase (2-5 ϫ 10 5 cpm) probes in 20 l of hybridization buffer containing 40 mM PIPES, 80% deionized formamide, and 400 mM sodium acetate. The hybridization reaction was then incubated at room temperature for 1 h in the presence of 5 g/ml RNase A and 50 units/ml RNase T1. RNA protected from RNase digestion was precipitated and subjected to electrophoresis (5% polyacrylamide, 8 M urea gel) after heating at 90 -95°C for 3-4 min. Quantitation of the protected RNA fragments was performed by scanning the gel with the PhosphorImager (Molecular Dynamics, Mountain View, CA).
DNase I-hypersensitive Site Assay-Astrocytes were stimulated with medium alone or TNF-␣ (500 units/ml) for 20 min, harvested on ice after washing with cold PBS, and pelleted by centrifugation. The cell pellet was resuspended in 1 ml of hypotonic buffer containing 10 mM Tris-HCl, pH 8.0, 10 mM NaCl, and 5 mM MgCl 2 , incubated on ice for 30 min, and then centrifuged at 1400 ϫ g at 4°C for 4 min. The pellet was resuspended in 1 ml of hypotonic buffer containing 0.1% Nonidet P-40, incubated on ice for at least 10 min, and homogenized by 20 strokes with a type B pestle and 10 strokes with a type A pestle. Nuclei were purified through hypotonic buffer containing 8.5% (w/v) sucrose by centrifugation at 1500 ϫ g at 4°C for 10 min. The nuclei pellet was resuspended in 1 ml of digestion buffer containing 100 mM Tris-HCl, pH 8.0, 50 mM NaCl, 4.2 mM MgCl 2 , and 1 mM CaCl 2 and stored on ice until DNase I digestion.
The concentration of the nuclei was adjusted to 4 -5 ϫ 10 7 nuclei/ml. 100-l aliquots of the nuclei were incubated either in the absence or presence of varying amounts of DNase I (0.05-2.5 g/100 l, Sigma, DN-EP) at 37°C for 5 min. 100 l of stop solution containing 200 mM Tris-HCl, pH 8.0, 200 mM NaCl, 20 mM EDTA, 2% SDS, and 200 g/ml proteinase K (Boehringer Mannheim) was added and then incubated overnight at 37°C. 0.2 volume of 10 M ammonium acetate was added after phenol/chloroform extraction, and DNA was precipitated by adding 2 volumes of ethanol, followed by centrifugation at 6000 ϫ g (ϳ8500 cpm for Eppendorf microcentrifuge), and washed two times with 70% ethanol. DNA was dissolved in TE buffer and subjected to restriction enzyme digestion. RNase A (25 g/ml) was added and incubated at 37°C for 1 h, and DNA was purified by a standard procedure of phenol/ chloroform extraction and ethanol precipitation. DNA was run on a 1% agarose gel (Ͻ1 V/cm, for 16 -20 h), transferred to nitrocellulose membrane, and hybridized with labeled probes by standard methods.
EMSA was performed in a volume of 20 l containing 150 mM KCl, 2.5 mM MgCl 2 , 0.25 mM EDTA, 20% glycerol, 0.25 mM DTT, 10 mM Tris-Cl, pH 7.5, 1 g of polydeoxyinosinic-deoxycytidyl acid, 1 g of sonicated salmon sperm DNA, 8 -10 g of nuclear extract, and 10,000 -20,000 cpm 32 P-labeled oligonucleotide probe and incubated on ice for 30 min. Bound and free DNA were then resolved by electrophoresis through a 6% high ionic strength polyacrylamide gel (1 ϫ TGE (Tris/ glycine/EDTA) buffer) containing 5% glycerol at 250 V for 1 h. For competition assays, unlabeled probes were incubated with nuclear extracts on ice for 10 min followed by addition of labeled probes. For supershift assays, antibodies were incubated with the nuclear extracts for 15 min, followed by the addition of labeled probe. All the antisera against NF-B subfamilies (p65, p50, p52, RelB, and c-Rel) were the generous gift of Dr. Nancy Rice (National Cancer Institute, Frederick, MD) (25,26).

Expression of the TNF-␣ Gene Can Be Induced by TNF-␣ in
Primary Astrocytes-We have previously shown that rat astrocytes are capable of producing TNF-␣ at both the mRNA and protein levels in response to LPS (2). The TNF-␣ gene is not constitutively expressed in astrocytes; however, upon stimulation with LPS, TNF-␣ mRNA and protein are transiently expressed. Steady state levels of TNF-␣ mRNA are rapidly (15 min) detected after stimulation, are maximal at 2 h, and then decline (6). It has been shown that TNF-␣ induces its own expression at both the mRNA and protein levels in HL-60 cells (27). Here, we show that the TNF-␣ gene can be induced by TNF-␣ in astrocytes. As shown in Fig. 1, TNF-␣ mRNA is not expressed in unstimulated cells (lane 1) and can be induced by TNF-␣ in a dose-dependent manner (lanes 2-4). Further TNF-␣ dose-response experiments revealed that 500 units/ml TNF-␣ induced optimal TNF-␣ mRNA expression (data not shown); thus, this concentration was used for the remainder of the studies.
Functional Analysis of the Upstream Sequence of the Rat TNF-␣ Gene-To identify cis-acting regulatory element(s) required for TNF-␣ induction of the TNF-␣ gene, we made a series of constructs in which serially deleted upstream regions of the rat TNF-␣ gene were fused to the chloramphenicol acetyltransferase (CAT) reporter gene. Astrocytes were transiently transfected with those constructs and stimulated with medium alone or TNF-␣ for 18 h, and then CAT activity was measured. Table I summarizes the results of the transfections. The full-length promoter construct, pTNF(-1229)CAT, which contains the sequence from Ϫ1229 to ϩ130 with respect to the transcription start site (TSS) of the rat TNF-␣ gene (the AvaI fragment, see Fig. 2A), was not responsive to TNF-␣ stimulation, and a high basal level of CAT activity was detected in the absence of stimulation. The rat TNF-␣ gene has five B-related sites in its upstream region. The high basal activity was not affected by deletion to Ϫ235 bp, which eliminated four NF-B binding sites. Further deletion to Ϫ177, which removed the most proximal NF-B binding site, resulted in approximately a 3-fold decrease in basal activity compared with the full-length promoter. However, none of the deletion constructs responded to TNF-␣ stimulation. These data indicate that the 1359-bp upstream region of the rat TNF-␣ gene is not sufficient for TNF-␣-mediated activation of the TNF-␣ promoter.
Localization of TNF-␣ Responsive Sequence(s) in the 3Ј Flanking Region of the TNF-␣ Gene-The results described above led us to examine the 10-kb rat TNF locus with respect to potential TNF-␣ responsive sequence(s). We prepared a series of "marked" TNF-␣ genes in which a mutation was introduced to allow discrimination of those mutated genes from the endogenous wild type TNF-␣ gene (Fig. 2, A and B). The marked gene has an 8-bp synthetic NotI linker that is inserted to the BspEI site in the 5Ј-untranslated region of the TNF-␣ gene (Fig. 2C). An antisense RNA probe was generated such that it is complementary to the sequence of Ϫ70 to ϩ400 of the marked TNF-␣ gene containing the inserted NotI linker. In the RNase protection assay (RPA), this probe protects 359 nt of mRNA that are accurately initiated from the marked TNF-␣ gene promoter, whereas endogenous TNF-␣ mRNA is protected as two smaller fragments, 150 and 196 nt (Fig. 2C). This system also allows differentiation of authentic marked TNF-␣ mRNA from non-  3Ј cis-Acting Element in TNF-␣ Gene Expression specific transcripts such as readthrough transcripts. pGL2-Control, in which the SV40 promoter drives luciferase gene expression, was used as an internal control for transfection efficiency and RNA integrity. Similarly, the antisense RNA probe for the luciferase gene was generated such that the luciferase mRNA accurately initiated from the SV40 promoter was protected as ϳ250-nt fragments (Fig. 2C). Thus, it is possible to detect message from three genes in the same RPA gel; the transfected marked TNF-␣ gene, the transfected luciferase gene, and the endogenous TNF-␣ gene (see Fig. 2D). A number of preliminary control experiments of transient transfections with pTNF10* and/or vector plasmid showed that the 359-nt message is specifically transcribed from the promoter of the transfected TNF-␣ gene, that neither the vector nor the cotransfected pGL2-Control affect the regulation of the transfected TNF-␣ gene, and that the endogenous TNF-␣ gene was accurately regulated (data not shown).
To localize the TNF-␣ responsive sequence(s), we analyzed a series of 5Ј and 3Ј deletion constructs of the marked TNF-␣ genes (Fig. 2B) in transient transfection assays. The results are shown in Fig. 2D. The full-length construct, pTNF10*, which contains the entire TNF-␤ gene with 1.2 kb of 5Ј flanking sequence, the 1.1-kb intergenic space, and the entire TNF-␣ gene with 3 kb of 3Ј flanking sequence, showed basal activity and was able to respond to TNF-␣ with ϳ6-fold induction. Two 5Ј deletion constructs, pTNF6.5* and pTNF5.9*, were also capable of responding to TNF-␣ with ϳ5-fold induction, similar to what was observed with pTNF10*. pTNF6.5* (5Ј deletion at Ϫ634), lacking the TNF-␤ gene, retains three NF-B sites in the TNF-␣ promoter region. pTNF5.9* (5Ј deletion at Ϫ70) contains in the promoter region an Sp-1 binding site, an AP-2 binding site, and the TATA box (Fig. 2, A and B). These results indicate that, for TNF-␣ induction of the TNF-␣ gene, the B sites of the upstream region are not necessary, and the sequence of Ϫ70 is sufficient in the presence of downstream sequences.
Deletion points of the 3Ј deletion constructs, pTNF4.4*, pTNF3.6*, pTNF3.4*, pTNF2.9*, and pTNF2.7* are ϳ1.9 and ϳ1.1 kb, 746, 240, and 56 bp downstream of the polyadenylation signal sequence of the TNF-␣ gene, respectively (Fig. 2, A  and B). As shown in Fig. 2D, the constructs with deletions up to 746 bp (pTNF4.4*, pTNF3.6*, and pTNF3.4*) are capable of responding to TNF-␣. The more extensive deletion constructs, pTNF2.9* or pTNF2.7*, do not respond to TNF-␣ stimulation. As in the 5Ј deletion constructs, basal levels of marked TNF-␣ mRNA were detected in all the 3Ј deletion constructs. None of the deletions up to pTNF3.4* significantly affected the magnitude of TNF-␣ induction (ϳ4 -6-fold) of marked TNF-␣ gene expression, indicating that no other important regulatory sequences are present in regions outside of TNF3.4*. Endogenous trol. The regions pointed by EcoRI and BamHI indicate primer binding sites for PCR for the template DNA of the antisense RNA probes. The predicted, protected sizes of mRNA are indicated under the corresponding genes; probe, unprotected free probe (thick line indicates vector sequence), RT, readthrough transcript, 5ЈTNF(T), mRNA accurately initiated from transfected marked TNF-␣ gene promoter, TNF(E), endogenous TNF-␣ mRNA, 5ЈLUC, mRNA accurately initiated from the multiple transcription start sites of SV40 promoter of pGL2-Control. In this report, we describe the protected sizes of the TNF-␣ antisense RNA probe according to the previously described putative transcription start site (22). D, astrocytes were transfected with pGL2-Control (6 g/ transfection) and the marked TNF-␣ gene constructs at the same molar ratios (g of DNA/transfection: pTNF10*, 2.76 g; pTNF6.5*, 2.00 g; pTNF5.9*, 1.87 g; pTNF4.4*, 1.52 g; pTNF3.6*, 1.37 g; pTNF3.4*, 1.33 g; pTNF2.9*, 1.22 g; and pTNF2.7*, 1.17 g). Cells transfected with the same constructs were combined, replated in two 100-mm dishes, and allowed to recover for 44 h. Transfected astrocytes were then stimulated with medium alone (Ϫ) or 500 units/ml rat TNF-␣ (ϩ). 90 g of total cellular RNA were used for RPA. Representative of three experiments.  asterisk (B*). B, series of marked TNF-␣ gene deletion constructs (see "Materials and Methods" for details). The solid lines indicate inserts of the corresponding plasmids. The numbers used for plasmid names are identical to the insert sizes of the corresponding plasmids. The NotI linkers inserted into BspEI sites are indicated by asterisks. C, protected sizes of antisense RNA probes (see text for details). Top is for the marked TNF-␣ genes (pTNF10* through pTNF2.7*), and bottom is for the pGL2-Con-TNF-␣ gene expression was accurately regulated in all the transfections (ϳ15-fold induction), indicating that abolishment of TNF-␣ inducibility of pTNF2.9* or pTNF2.7* is specifically due to deletions of the sequences rather than possible impairment in the TNF-␣-mediated signaling pathway leading to activation of the TNF-␣ gene. These data demonstrate that the 500-bp downstream region of the TNF-␣ gene, i.e. the sequence between the EcoRI and SmaI sites ( Fig. 2A), is critical for TNF-␣ induction of the TNF-␣ gene in astrocytes.
Identification of TNF-␣-inducible DNase I-hypersensitive Sites in the 3Ј Flanking Region of the TNF-␣ Gene-We examined the chromatin structure of the endogenous TNF-␣ gene by looking at DNase I-hypersensitive (DH) sites, which are known, in essentially every case, to be associated with cisacting regulatory sequence elements such as enhancers (28,29). The 4.2-kb SacI region, containing the 3Ј region of the TNF-␣ gene, was examined for DH sites with a labeled 420-bp SacI-BamHI fragment abutting the 5Ј end of the SacI fragment (see Figs. 2A and 3C for map). As shown in Fig. 3A, there are no detectable DH sites in this SacI region under basal conditions; only the 4.2-kb parent band is detected. Two DH sites are rapidly (within 20 min) induced upon TNF-␣ stimulation. One (3Ј DH1) is mapped around the 3Ј deletion point of pTNF2.9* (the EcoRI site); the subfragment migrates along with the 1.2-kb band of the TNF-␣ plasmid marker. The fact that TNF-␣-inducible DH site overlaps with the region of the deletion, which abolishes TNF-␣ responsiveness of pTNF2.9*, strongly suggests that the TNF-␣ responsive sequence(s) is located in this region (around the EcoRI site). These results also suggest that the chromatin of that region undergoes structural changes from a closed to open configuration. The other DH site (3Ј DH2), which is much weaker than the 3Ј DH1, is ϳ800 bp downstream of the 3Ј DH1 (Fig. 3C). This 3Ј DH2 is not likely involved in TNF-␣ gene regulation, since deletion of this site (pTNF 3.4*) had no effect on TNF-␣ inducibility of the TNF-␣ gene (Fig. 2D).
TNF-␣ Induced NF-B (p50 and p65) Proteins Specifically Bind to the Two Consecutive B Sites in the 3Ј DH1 Region-Sequence analysis showed that the region of the EcoRI site (3Ј deletion point of pTNF2.9* or the 3Ј DH1) contains consensus sequences of binding sites for NF-B or AP-1, which are known to be inducible by TNF-␣ (30 -32). The canonical NF-B binding site (NF-B-IC-li in Fig. 4) is present 8 bp upstream of the EcoRI site. This B site has been shown to be bound by TNF-␣-induced NF-B-like protein and mediates TNF-␣ induction of the class II-associated invariant chain (li) gene in rat fibroblast cells (33). The PRDII sequence in the reverse orientation overlaps with the NF-B-IC-li, which is known to mediate viral induction of the human interferon-␤ gene via binding of NF-B (34). A sequence homologous to the NF-B-IC-li (one T deletion, indicated by B-like in Fig. 4) overlaps with the EcoRI site. The consensus sequence for the AP-1 binding site is found 84 bp downstream of the EcoRI site (35).
We wanted to examine the potential involvement of the two consecutive B-(like) sequences in TNF-␣-induced TNF-␣ gene expression in astrocytes by investigating nuclear proteins binding to these sites. Electrophoretic mobility shift assays were performed using the synthetic oligonucleotides described in Fig. 5A. For convenience, we designate the NF-B-IC-li and B-like as B and BЈ, respectively. We found that nuclear extracts from unstimulated astrocytes showed two weak com-  , and mB/IL-6 (lane 9) oligonucleotides were also used. The dried gel was exposed to film for 24 h at Ϫ70°C. D, nuclear extracts from astrocytes treated with medium alone (lanes 1 and 8) or TNF-␣ for 15 min (lanes 2-7 and 9-14) were incubated on ice for 10 min with unlabeled 250-fold excess amounts of B (lanes 3 and 12), mB (lanes 4 and 13), BЈ (lanes 5 and 10), mBЈ (lanes 6 and 11), and B/IL-6 (lanes 7 and 14) and then incubated on ice for 30 min with labeled B (lanes 1-7) or BЈ (lanes 8 -14) probes. The dried gel was exposed to film for 48 h at Ϫ70°C. E, nuclear extracts from astrocytes treated with medium alone (lane 1) or TNF-␣ for 15 min (lanes 2-14) were incubated at room plexes on the B/BЈ probe containing two B-(like) sequences and that the slower migrating complex (I) was enhanced by TNF-␣ stimulation, whereas the faster migrating complex (II) remained unchanged (compare lane 2 and lane 3 in Fig. 5B). Kinetic analysis with nuclear extracts from astrocytes that had been treated with TNF-␣ for different periods showed that complex I was immediately induced (ϳ12-fold induction) after 15 min of TNF-␣ treatment (lane 3 in Fig. 5B) and gradually decreased after 4 h of stimulation (lane 7 in Fig. 5B). Complex II was not changed at any point after TNF-␣ stimulation. The kinetics of TNF-␣ induction of complex I is functionally relevant to TNF-␣-mediated activation of the TNF-␣ gene, since the steady state level of TNF-␣ mRNA accumulation is maximally induced at 1-2 h after TNF-␣ treatment and subsequently diminishes.
To examine the specificity of these complexes, we performed competition assays with excess amounts of unlabeled probes. Complex I formed on the B/BЈ probe could be specifically competed away by a 250-fold excess amount of unlabeled B/ BЈ probe but not with the same amount of an unrelated sequence such as AP-1/TNF-␣ (this is the AP-1 sequence located in the 3Ј DH1) (Fig. 5C, lanes 3 and 10). Inhibition by the competitors was dose-dependent; we show only the 250-fold excess amount for all the competition data. To determine which B site of the B/BЈ probe is responsible for the formation of complex I, we carried out competition experiments with unlabeled B or BЈ probes. As shown in Fig. 5C, the B competitor can efficiently compete away complex I formed on the B/BЈ probe (lane 4), whereas the BЈ competitor is able to only partially compete away complex I (lane 6). The binding of complex I is B site-specific since mutation of the guanine nucleotides GGGA, which are critical for NF-B binding, of both B and BЈ probes abolishes the inhibition of complex I (lanes 5 and 7). To further ascertain the B site specificity of complex I, we utilized a B site from the IL-6 gene promoter (B/IL-6) as a competitor, which is similar to the B and BЈ sites but with different flanking sequences (36) (see Fig. 5A). Wild type B/IL-6 competitor could completely inhibit complex I binding to the B/BЈ probe, but mutant B/IL-6 did not (Fig.  5C, lanes 8 and 9). Complex II, which is not modulated by TNF-␣ treatment, was modestly inhibited by B/BЈ, B, and BЈ oligonucleotides (lanes 3, 4, and 6) and completely inhibited by the B/IL-6 oligonucleotide (lane 8).
We next performed similar competition experiments using B or BЈ as probes. Either B or BЈ alone is sufficient for formation of complex I, although the affinity of complex I for the B probe is ϳ6-fold higher than for BЈ (Fig. 5D, lanes 2 and  9). Complex I formation on the B or BЈ probes is B sitespecific since unlabeled B and BЈ are strong competitors (lanes 3 and 10), whereas mutants B and BЈ are not (lanes 4 and 11). The B/IL-6 oligonucleotide completely inhibits complex I formation on the B or BЈ probes (lanes 7 and 14), indicating that for both probes the flanking sequence is not important for complex I formation. The B competitor completely inhibited complex I formation on the BЈ probe (lane 12); however, the BЈ competitor moderately inhibited complex I on the B probe (lane 5). These results again indicate that the B site is more efficient than the BЈ site for formation of complex I. Thus, the proteins of complex I can bind to both B and BЈ sites independently but with different affinities. The competition data shown in Fig. 5, C and D, strongly suggested that complex I is an NF-B-like protein. To directly determine the identity of complex I, supershift assays were conducted with antibodies against various members of the Rel protein family (p65, p50, p52, c-Rel, and RelB) (25,26). As shown in Fig. 5E, antibodies against p65 or p50 supershifted complex I on the B/BЈ probe (lanes 3 and 4), while antibodies against p52, RelB, and c-Rel were without effect (lanes 5-7). As a control, normal rabbit serum did not affect the migration of complex I (lane 8). Similarly, migration of complex I on the B or BЈ probes was shifted by both anti-p65 and anti-p50 antibodies (lanes 10, 11, 13, and 14) but not by antibodies against p52, RelB, or c-Rel (data not shown). Complex II migration was not modulated by any of the antibodies tested, indicating that this complex does not consist of Rel protein family members. These results directly demonstrate that complex I consists of the p50 and p65 members of the NF-B family. It should be mentioned that although antisera against p52, RelB, and c-Rel did not affect the migration of complex I, we are not certain of their ability to recognize rat proteins. Thus, we cannot completely rule out the potential involvement of these proteins at this time. DISCUSSION We have investigated the mechanism(s) by which transcription of the TNF-␣ gene is induced by TNF-␣ in primary astrocytes. Using a series of marked TNF-␣ gene constructs, we demonstrate that TNF-␣ responsive sequences are located in the 3Ј flanking region of the TNF-␣ gene. We also find that a TNF-␣-inducible DNase I-hypersensitive site (3Ј DH1) is associated with the 3Ј flanking region that is critical for TNF-␣ inducibility of the marked TNF-␣ gene. Gel mobility shift assays identified nuclear proteins consisting of NF-B members p50 and p65 that specifically bind to NF-B sites within the 3Ј DH1; these proteins are immediately induced (ϳ12-fold induction) after TNF-␣ stimulation and gradually decrease over time.
Since the role of NF-B in the regulation of the TNF-␣ gene is controversial, we wanted to determine whether the upstream B sequences of the rat TNF-␣ gene played any role in TNF-␣ induction of the rat TNF-␣ gene in astrocytes. We observed that the rat TNF-␣ 5Ј promoter constructs showed high basal activity that could not be further enhanced by TNF-␣ stimulation ( Table I). The marked TNF gene system allowed us to search the 10-kb TNF locus for regulatory sequence elements by analyzing expression of deletion mutants of the TNF-␣ marked gene that could be differentiated from expression of the endogenous TNF-␣ gene. We demonstrated that the sequence of Ϫ70 from the TSS is sufficient, and a 500-bp region located 260 bp downstream of the polyadenylation sequence of the TNF-␣ gene is critical for TNF-␣ induction of the TNF-␣ gene (Fig. 2D). The five B-like sequences in the upstream region of the rat TNF-␣ gene were not required for TNF-␣ induction of the TNF-␣ gene.
As mentioned previously, some studies have suggested that the downstream sequence is involved in TNF-␣ gene expression. Using transgenic mice, Keffer et al. (19) showed that a 3Ј-modified TNF-␣ transgene could not be activated by LPS, whereas the wild type TNF-␣ transgene was LPS-inducible. In this study, however, whether the defect in the transgenic mice carrying the 3Ј-modified TNF-␣ gene was due to interference at the transcriptional level, in mRNA stability, or in translation of mRNA was not determined. Kuprash et al. (20) have recently reported that a B element located downstream of the TNF-␣ gene partially mediates LPS-induced TNF-␣ gene expression in murine macrophages. Although this downstream B element was shown to act as an LPS-responsive enhancer, it appears to temperature for 15 min without (lanes 1, 2, 9, 12) or with antisera against p65 (lanes 3, 10, 13), p50 (lanes 4, 11, 14), p52 (lane 5), RelB (lane 6), c-Rel (lane 7), or normal rabbit serum (NRS) (lane 8) and then were incubated on ice for 30 min with labeled B/BЈ probe (lanes 1-8), B probe (lanes 9 -11), or BЈ probe (lanes 12-14). The dried gel was exposed to film for 24 h at Ϫ70°C. be auxiliary in function since its activity is not significantly pronounced in the presence of three upstream B elements. However, the rat TNF-␣ gene in astrocytes seems to be mainly regulated by downstream sequence elements (possibly two consecutive B sites, see below), since deletion of all the upstream B sites did not affect the magnitude of TNF-␣-induced TNF-␣ gene expression (Fig. 2D). Similar observations were made for LPS stimulation, i.e. no upstream sequences are required, and LPS-responsive sequences are located in the same region as that for TNF-␣. 2 Two consecutive B-(like) sequences (B and BЈ) are present within the TNF-␣-inducible 3Ј DH1 site that is associated with the functionally critical region for TNF-␣-induced TNF-␣ gene expression (Fig. 4). We have identified a protein complex consisting of p50 and p65 components of NF-B that specifically binds to the B site (Fig. 5). This complex (complex I) is constitutively expressed at very low levels, immediately enhanced upon TNF-␣ stimulation, and then gradually decreases (Fig. 5B). The kinetics of TNF-␣ induction of complex I implies its involvement in TNF-␣-induced TNF-␣ gene expression. Interestingly, complex I also binds to the kBЈ site. The 3Ј end of the BЈ sequence is destroyed in the pTNF2.9* construct, which is not activated in response to TNF-␣, suggesting that the BЈ sequence could be important for TNF-␣ inducibility. In this regard, it is possible that B and BЈ cooperatively act to respond to TNF-␣ stimulation. This may be the reason that pTNF2.9* loses TNF-␣ inducibility even though it retains the B site that alone is sufficient for formation of complex I (Figs. 2D and 5D).
Our results suggest that expression of the TNF-␣ gene is under stringent regulatory control, which involves a transition of gene activity from repression to derepression to activation. This study provides some evidence supporting this idea. First, the transfected TNF-␣ promoter (both the promoter-CAT and marked gene constructs) shows high basal activity (Table I and Fig. 2D), whereas the endogenous TNF-␣ gene is transcriptionally silent under basal conditions (6). The chromatin structure of transiently transfected nonreplicating plasmids could be different from that of replicating DNA, which would more closely mimic the endogenous gene (37,38). The endogenous TNF-␣ gene thus could be under repression associated with the chromatin structure, which is not reflected in the transiently transfected TNF-␣ gene constructs. Second, the marked TNF-␣ genes with TNF-␣ inducibility showed basal activity but could be further activated upon TNF-␣ stimulation (Fig. 2D). These results suggest that the TNF-␣ gene is positively activated following derepression upon TNF-␣ stimulation. In fact, the lower degree of induction by TNF-␣ of the marked TNF-␣ genes (4 -6-fold, Fig. 2D) compared with the endogenous TNF-␣ gene (at least 15-fold, Fig. 1 and Fig. 2D) is due to the high basal activity of the transfected marked TNF-␣ genes. Third, TNF-␣ induces the 3Ј DH1 which is associated with B sites to which TNF-␣-inducible p50/p65 NF-B proteins bind (Figs. 3 and 5). The absence of a constitutive 3Ј DH1 suggests that B sites are restricted in accessibility to constitutive transcription factors (complex I in Fig. 5), which maintains the TNF-␣ gene in a repressed state under basal conditions. Upon TNF-␣ stimulation, the B sites become immediately accessible to increased amounts of TNF-␣ enhanced complex I following chromatin opening. Interestingly, a recent study has provided evidence that NF-B plays a direct role in modulation of chromatin structure (39).
These regulatory mechanisms would be beneficial for expression of a transient gene such as TNF-␣, which should be under tight repression followed by rapid induction upon stimulation. Recently, Probert et al. (40) showed that transgenic mice that overexpressed the TNF-␣ gene in the CNS spontaneously developed a chronic inflammatory demyelinating disease and that administration of a neutralizing TNF-␣ antibody prevented development of disease, indicating a direct role for TNF-␣ in the pathogenesis of this disease. Because of the detrimental effects of TNF-␣ within the CNS, it would be advantageous to have this gene under strict regulatory control.