Identification of Direct Genomic Targets Downstream of the Nuclear Factor-κB Transcription Factor Mediating Tumor Necrosis Factor Signaling*

Tumor necrosis factor (TNF) is a pro-inflammatory cytokine that controls expression of inflammatory genetic networks. Although the nuclear factor-κB (NF-κB) pathway is crucial for mediating cellular TNF responses, the complete spectrum of NF-κB-dependent genes is unknown. In this study, we used a tetracycline-regulated cell line expressing an NF-κB inhibitor to systematically identify NF-κB-dependent genes. A microarray data set generated from a time course of TNF stimulation in the presence or absence of NF-κB signaling was analyzed. We identified 50 unique genes that were regulated by TNF (Pr(F) <0.001) and demonstrated a change in signal intensity of ± 3-fold relative to control. Of these, 28 were NF-κB-dependent, encoding proteins involved in diverse cellular activities. Quantitative real-time PCR assays of eight characterized NF-κB-dependent genes and five genes not previously known to be NF-κB-dependent (Gro-β and-γ, IκBϵ, interleukin (IL)-7R, and Naf-1) were used to determine whether they were directly or indirectly NF-κB regulated. Expression of constitutively active enhanced green fluorescent·NF-κB/Rel A fusion protein transactivated all but IL-6 and IL-7R in the absence of TNF stimulation. Moreover, TNF strongly induced all 12 genes in the absence of new protein synthesis. High probability NF-κB sites in novel genes were predicted by binding site analysis and confirmed by electrophoretic mobility shift assay. Chromatin immunoprecipitation assays show the endogenous IκBα/ϵ, Gro-β/γ, and Naf-1 promoters directly bound NF-κB/Rel A in TNF-stimulated cells. Together, these studies systematically identify the direct NF-κB-dependent gene network downstream of TNF signaling, extending our knowledge of biological processes regulated by this pathway.

Tumor necrosis factor (TNF) 1 is a pleiotropic cytokine that mediates inflammatory signaling in infectious diseases (1). In the respiratory tract, alveolar macrophages encountering invading microorganisms inducibly secrete TNF. TNF, in turn, signals in a paracrine manner to induce expression of epithelial chemokine networks (2) and in an endocrine manner to induce the hepatic acute phase response (3,4). In this way, TNF plays a central role in amplifying the host inflammatory response to tissue infection.
TNF binds to ubiquitously expressed single pass type I transmembrane receptors of 55 and 75 kDa (TNFR-I and -II, respectively) that lack intrinsic kinase activity (5). TNF ligand binding initiates intracellular signaling by inducing the formation of receptor trimers that assemble submembranous protein complexes assembling on the aggregated cytoplasmic TNFR "death domain" (6). TNFR death domain-binding proteins include the TNFR-associated death domain TRADD (6) and the receptor interacting protein RIP (7). Subsequent recruitment of the TNF receptor-associated factor-2 (TRAF2) to the submembranous complex activates two important downstream signaling pathways, the c-Jun N-terminal kinase/p38 mitogen-activated protein kinase activating protein-1 (AP-1) pathway and the inhibitor of B kinase (IKK)-nuclear factor-B (NF-B) pathway (8,9). Experiments where specific mutations of TRAF2 and RIP are expressed have shown that the c-Jun N-terminal kinase and IKK pathways are separable intracellular signaling cascades (10).
The IKK complex is an upstream controller of the NF-B transcription factor (11). NF-B is a family of ubiquitously expressed, highly inducible cytoplasmic transcription factors (12)(13)(14). The prototypical NF-B complex is maintained in a cytoplasmic localization by association with the IB inhibitors, whose members bind and specifically inactivate the NF-B/Rel A subunit, preventing its nuclear entry (Refs. 15 and 16 and reviewed in Ref. 17). IKK induces IB proteolysis by phosphorylating IB specifically on serine residues 32 and 36 in its NH 2 regulatory domain, making it a substrate for ubiquitination and proteolysis through the 26 S proteasome (18) and calpain pathways (19). Upon IB degradation, the Rel A/NF-B1 heterodimer is released to enter the nucleus and stimulate gene expression by binding to regulatory regions of inducible promoters. Previous studies have shown the IKK-NF-B pathway is critical for mediating the TNF-induced cytokine cascade, because specific inhibitors of IKK-NF-B pathway block cytokine expression and inflammation in vivo (20,21). In addition, the IKK-NF-B pathway protects against apoptosis by inducing the expression of a variety of anti-apoptotic genes (10).
Although previous microarray studies have characterized the genomic responses to TNF (22), there has been no systematic evaluation or validation of the spectrum of genes directly under NF-B control. Earlier we exploited a robust microarray analysis of HeLa cells expressing a tightly tetracycline-regulated NF-B dominant negative inhibitor to identify NF-B-dependent genes downstream of respiratory virus replication (23). In this study, we sought to extend this analysis to the TNF pathway. HeLa cells were stimulated for 0, 1, 3, and 6 h with recombinant human (rh) TNF␣ in the presence or absence of NF-B signaling to generate a time series data set of genomic responses. Using two-way analysis of variance where TNF and NF-B inhibition were treated as experimental manipulations, we identified 28 NF-B-dependent genes encoding diverse biological pathways. Fourteen genes, including six not previously shown to be NF-B-dependent, were analyzed in detail to determine whether they satisfied independent experimental criteria for being bona fide genomic targets for inducible NF-B binding and transactivation. This experimentally defined NF-B-dependent data set furthers our understanding of its genomic and biological actions in inflammatory TNF signaling.

EXPERIMENTAL PROCEDURES
Cell Culture, Treatment, and Transfection-The human cervical epithelioid carcinoma cell line HeLa expressing tTA (HeLa Tet-Off) pBI-EGFP⅐IB␣ Mut was constructed earlier (23). HeLa Tet-Off were grown in medium containing 90% Dulbecco's modified Eagle's medium, 10% heat-inactivated fetal bovine serum, 4 mM L-glutamine, 0.1 mM nonessential amino acids, 1 mM sodium pyruvate, 100 mg/ml G418, 100 units/ml penicillin G sodium, and 100 mg/ml streptomycin sulfate in a humidified atmosphere of 5% CO 2 . For TNF stimulation, freshly isolated cells were split into two cultures, one group maintained in doxycycline (Dox), the other without for 7 days, a time at which IB␣ Mut expression was maximal. Thereafter, rhTNF␣ (25 ng/ml, final concentration) was added directly to the culture medium for the indicated times prior to harvest. Freshly plated cells in triplicate 60-mm plates were transiently transfected using Lipofectamine (Invitrogen), 8 g of indicated luciferase reporter plasmid, and 3 g of the constitutive alkaline phosphatase internal control plasmid, pSV2PAP (25). Cells were cultured for an additional 40 h and, where indicated, stimulated with rhTNF␣ for 6 h prior to harvest and assay for reporter activity.
-Fold induction of reporter activity was calculated by dividing the mean normalized luciferase activity (relative to alkaline phosphatase) from three treated cultures by the mean normalized luciferase activity from three untreated cultures.
Oligonucleotide Probe-based Microarray-Hu95Av2 GeneChip (Affymetrix Inc, Santa Clara, CA) containing 12,626 sequenced human genes was hybridized according to the manufacturer's recommendations and washed using both non-stringent (1 M NaCl, 25°C) and stringent (1 M NaCl, 50°C) conditions prior to staining with phycoerythrin streptavidin (10 g/ml final). Arrays were scanned using a Gene Array scanner (Hewlett Packard) and analyzed using GeneChip Analysis Suite 5 software (Affymetrix Inc). For each gene, 16 -20 probe pairs were immobilized as ϳ25-mer oligonucleotides that hybridize throughout the mRNA; each probe pair is represented as a perfect match oligonucleotide and a mismatch oligonucleotide as hybridization control. The absolute call (e.g. the gene is detected ("present") or not ("absent")) and the signal intensity (measure of mRNA abundance) were determined (26).
Data Analysis-Four independent experiments were performed identically including control (0 h) and 1, 3, and 6 h of TNF stimulation (25 ng/ml) in the presence or absence of doxycycline (2 g/ml) in growth medium. For comparison of the fluorescent intensity (signal intensity) values among multiple experiments, the signal intensity values for each "experimental" GeneChip were scaled to that of the "base" GeneChip. This was done first by calculating the "2% trimmed mean" (a measurement of global signal intensity) for each GeneChip. The trimmed mean is obtained by calculating the mean signal intensity of the chip after discarding the top and bottom 2% average difference values. Next, scaling was performed by multiplying all the signal intensity measurements in the experimental array by a scaling factor defined as the ratio of the base trimmed mean to that of the experimental trimmed mean (the base array was defined to be the 0 h control ϩ Dox GeneChip). An average of 4929 Ϯ 572 probe sets was detected on each GeneChip. The normalization factor ranged from 1.011 to 1.307 (1.18 Ϯ 0.10), and a high degree of reproducibility was observed in analysis of pairwise analysis of signal intensity measurements across the GeneChips. Because both the rhTNF␣ and Dox can be considered experimental treatments, the scaled average difference values were then subjected to a two-way analysis of variance with replications (ANOVA, Splus 6; Insightful Inc.) to determine which genes were significantly influenced by either the rhTNF␣ or Dox treatment. -Fold change was calculated as the signed ratio between treatment and control signal intensity. Agglomerative hierarchical clustering was performed using the weighted pair group method with arithmetic mean (Spotfire Array Explorer, version 8; Spotfire Inc., Cambridge MA) using Euclidian distance. To compare genes with different levels of expression, the data were Z-score normalized prior to cluster analysis (27). 2 RNA Quantitation-For Northern hybridization, acid guanidiumphenol-extracted total cellular RNA was fractionated by electrophoresis on a 1.2% agarose-formaldehyde gel, capillary transferred to a nitrocellulose membrane (Zeta-Probe GT; Bio-Rad), and hybridized with 1-2 ϫ 10 6 cpm/ml 32 P-labeled cDNA as previously described (23). The membrane was exposed to XAR film (Kodak) for 24 -48 h at 70°C and quantified by exposure to PhosphorImager cassette.
Quantitative real-time reverse transcriptase-polymerase chain reaction (Q-RT-PCR) assays were developed using TaqMan technology on an Applied Biosystems 7000 sequence detection system. Applied Biosystems assays-by-design or assays-on-demand 20Ј assay mix of primers and TaqMan MGB probes (FAM TM dye-labeled) for all our target genes and predeveloped 18 S rRNA (VIC TM -dye labeled probe) Taq-Man TM assay reagent (P/N 4319413E) for internal control were used for real-time PCR measurements. These assays were designed to span exon-exon junctions so as not to detect genomic DNA. All primers and probe sequences were searched against the Celera data base to confirm specificity. Validation experiments were performed to test the efficiency of the target and reference amplifications. All absolute values of the slope of log input amount versus CT are less than 0.1. Separate tubes (singleplex) for one-step RT-PCR were used with 50 ng of RNA for both target genes and endogenous controls using the TaqMan one-step RT-PCR master mix reagent kit (Applied Biosystems). The cycling parameters for one-step RT-PCR were: reverse transcription 48°C for 30 min, AmpliTaq activation 95°C for 10 min, denaturation 95°C for 15 s, and annealing/extension 60°C for 1 min (repeat 40 times) on ABI7000. Duplicate CT values were analyzed with Microsoft Excel using the comparative CT(⌬⌬CT) method as described by the manufacturer (Applied Biosystems). The amount of target (2 ϪDDCT ) was obtained by normalizing to an endogenous reference (18 S RNA) relative to a calibrator (one experimental sample). The primer and probe sequences for IL-6, IL-8, NF-B1, NF-B2, IB␣, IB⑀, Naf-1, TRAF1, and PTGES were obtained commercially (P/N 4331182; Applied Biosystems). For TNFAIP/A20, the probe sequence was 5Ј-CAATTGCCGTCACCGTTC-3Ј; the forward primer was 5Ј-AGCTTGTGGCGCTGAAAAC-3Ј, and the reverse primer was 5Ј-ACTGAGAAGTGGCATGCATGAG-3Ј.
Western Immunoblots-For Western immunoblot, a constant amount of cytoplasmic or nuclear extracts (200 -300 g as indicated) was boiled in Laemmli buffer, fractionated on 10% SDS-PAGE, and transferred to polyvinylidene difluoride membranes (Millipore, Bedford, MA) in CAPS/methanol (25). Membranes were blocked in 5% milk/Trisbuffered saline 0.1% Tween for 1 h and immunoblotted with primary antibody for 1 h at 4°C. Primary antibodies used were anti-FLAG M2-peroxidase conjugate (A8592; Sigma), affinity-purified anti-IB␣ or anti-Rel A (the latter two from Santa Cruz Biotechnology, Santa Cruz CA). Membranes were washed four times in Tris-buffered saline 0.1% Tween 20. Immune complexes were detected by reaction in the enhanced chemiluminescence assay (ECL; Amersham Biosciences) according to the manufacturer's recommendations. Equivalent protein loading was confirmed by probing the blot with anti-␤-actin (25).
Electrophoretic Mobility Shift Assays (EMSAs)-EMSAs were performed as described previously (23,28). In brief, 15 g of nuclear extracts (NE) were incubated with 40,000 cpm of 32 P-labeled duplex oligonucleotide probe and 2 g of poly(dA-dT) in binding buffer. The complexes were fractionated on 6% native polyacrylamide gels, dried, and exposed to Kodak X-AR film at 70°C. Competition was performed by the addition of 100-fold molar excess nonradioactive double-stranded oligonucleotide competitor at the time of addition of radioactive probe. The double-stranded oligonucleotides contained 5Ј-GATC overhangs. Plasmids-The human Naf-1 promoter, spanning Ϫ626/ϩ111 nucleotides relative to the cap site, was cloned by PCR using the sense primer, 5Ј-TTTAAAGGATCCAATGCTTA CGTGCCTTTTGG-3Ј, and the downstream primer, 5Ј-ATTTAAAAAGCTTGGGAGCTTGGGGAC-ACAG-3Ј using HeLa genomic DNA as a template (Bam H1 and Hind III sites are underlined). The PCR fragment was restricted with Bam H1/Hind III, gel purified, and ligated into the pOLUC firefly luciferase reporter (24). Site mutation of the Rel A binding site was introduced by PCR gene amplification by overlap extension using the sense mutagenic primer, 5Ј-AGTCTCCCTCCGTTTACTTGAAAAGGGGTGGGGCGGC-C-3Ј and the antisense mutagenic primer, 5Ј-CCACCCC TTTTCAAGT-AAACGGAGACTGGCAGGCCAC-3Ј, and the wild type Naf-1/LUC as template (mutated sites are underlined). The luciferase reporter genes driven by the three copies of the rat acute phase response element, (APRE) 3 -p59rAT/LUC, has been described (24). Plasmids were purified by ion exchange chromatography (Qiagen, Chatsworth, CA) and inserts sequenced to verify authenticity prior to transfection.
Chromatin Immunoprecipitation (ChIP) Assays-The ChIP assay was as described with modifications (29). On the day prior to the experiment, 2-4 ϫ 10 6 cells were plated in 0.5% bovine serum albumin containing growth medium. Cells were stimulated for the indicated times and sequentially cross-linked with disuccinimidyl glutarate and 1% formaldehyde in serum-free medium for 15 min at 37°C. The cells were washed, transferred to Eppendorf tubes, and solubilized in 400 l of SDS lysis buffer (1% SDS, 10 mM Tris, pH 8.0, 1 mM EDTA) with protease inhibitor mixture (Sigma). The samples were sonicated three times, 15 s at setting 2 until DNA fragments were 300 -400 bp or less. Equal amounts of DNA were immunoprecipitated overnight at 4°C in ChIP dilution buffer (50 mM NaCl, 1 mM HEPES, pH 7.4, 1% IGEPAL-630, 10% glycerol, 1 mM dithiothreitol) with 20 g of either anti-Rel A (sc-372; Santa Cruz Biotechnology) or IgG as indicated. Immunoprecipitates were collected with protein-A magnetic beads (Dynal, Inc.) and washed sequentially with ChIP dilution buffer, high salt buffer, LiCl buffer, and TE buffer (10 mM Tris, pH 8.0, 1 mM EDTA). DNA was eluted in 1 ml of elution buffer 1% SDS in 0.1 M NaHCO 3 . Samples were de-cross-linked in 200 mM NaCl at 65°C, 1 h. DNA was phenol extracted, ethanol precipitated, and used for PCR. PCR primers and conditions for semiquantitative PCR are shown in Table I. PCR products were fractionated by agarose gel chromatography and stained with ethidium bromide.

Dox-regulated Inhibition of NF-B Signaling-Because
NF-B is required for normal cell growth and survival (10, 30), we developed a tetracycline (Tet)-regulated cell system (Tet-Off) (31) by which an NF-B inhibitor could be tightly controlled and transiently induced. For this purpose, an expression plasmid encoding the non-degradable epitope-tagged IB␣ (IB␣ Ser32Ala/Ser36Ala, termed FLAG⅐IB␣ Mut) downstream of the Tet operator was stably introduced into cells expressing the Tet transactivator (tTA). FLAG⅐IB␣ Mut contains site mutations in the serine phosphoacceptor sites of IKK␤ that block its inducible degradation from a variety of stimuli and function as a potent dominant negative inhibitor of NF-B activation (23,32). The HeLa tTA/FLAG⅐IB␣ Mut transfectants cultured in the continuous presence of Dox did not express detectable amounts of FLAG⅐IB␣ Mut by Western immunoblot. In contrast, upon Dox withdrawal, FLAG⅐IB␣ Mut was strongly expressed (Fig. 1A). In earlier analysis of FLAG⅐IB␣ Mut expression, we demonstrated that FLAG⅐IB␣ Mut was expressed at similar levels to the endogenous IB␣ (23). Importantly, in the short durations used for expression and stimulation, these cells did not have detectable differences in spontaneous or viral-induced apoptosis (23). To demonstrate that these levels of FLAG⅐IB␣ Mut expression could inhibit NF-B activation, we performed EMSA on nuclear extracts of rhTNF␣-stimulated cells. In HeLa tTA/FLAG⅐IB␣ Mut cells cultured in the presence of Dox, binding of the Rel A⅐p50 heterodimer was rapidly induced within 15 min of rhTNF␣ treatment (Fig. 1B). By contrast, in cells cultured in the absence of Dox, no rhTNF␣ detectable changes in NF-B binding were observed in EMSA (Fig. 1B, right panel). We next determined whether the magnitude of FLAG⅐IB␣ Mut expression was sufficient to interfere with NF-B-dependent transcription. A previously characterized high affinity Rel A/NF-B1 binding site from the angiotensinogen gene (33) cloned as a multimeric element upstream of an inert promoter driving firefly luciferase reporter ((APREWT) 3 -p59rAT/LUC) was transfected into HeLa tTA/FLAG⅐IB␣ Mut cells maintained in the absence or presence of Dox and stimulated in the absence or presence of rhTNF␣. As seen in Fig. 1C, in the presence of Dox rhTNF␣ induced a 22-fold increase in normalized luciferase reporter activity (p Ͻ0.0001, n ϭ 3 independent experiments). However, in the absence of Dox, no statistically significant induction could be recorded (a 1.1-fold increase, p ϭ not significant). Together, these data confirmed that the Dox-regulated FLAG⅐IB␣ Mut expression was sufficient to block TNF-inducible NF-B-dependent transcription.
Oligonucleotide Microarray Identification of TNF-regulated NF-B-dependent Genes-To comprehensively identify the spectrum of genes activated by TNF and those controlled by NF-B, the tetracycline-regulated FLAG⅐IB␣ Mut-expressing cells were plated in parallel cultures in the absence or presence of Dox (2 g/ml). Each group of cells was then stimulated for 0, 1, 3, or 6 h with rhTNF␣, and RNA was harvested from all plates simultaneously ( Fig. 2A). The experiment was replicated in four independent experiments to obtain sufficient replicates for statistical analysis. The RNA was then quantitated and subjected to high density oligonucleotide microarray analysis using the Affymetrix platform (Hu95Av2 GeneChips).
To identify the genes regulated by TNF and those regulated by NF-B, the scaled signal intensity was subjected to two-way analysis of variance with replications considering rhTNF␣ stimulation and NF-B inhibition as independent experimental manipulations and filtered by -fold change criteria (Fig. 2B). Genes whose expression was under tonic NF-B control were first identified by those whose p value was Ͻ0.01 as a result of Dox treatment and whose steady state mRNA abundance TABLE I Primers and conditions used in ChIP assay Semiquantitative PCR assays were established for novel NF-B-dependent genes. For each promoter the common name is shown and the RefGene accession number, sequences of the sense (Sense) and antisense (Antisense) in 5Ј to 3Ј orientation, and PCR amplification conditions. For each, the size of the specific amplified fragment (in base pairs) is shown. was Ϯ 3-fold changed in the absence of TNF stimulation (0 h ϩ Dox versus 0 h Ϫ Dox) and are shown in Table II. Of the four genes identified, only two (prostaglandin synthase and Rel B) were also TNF regulated. Of these, Rel B had higher expression in cells in the presence of NF-B than its absence, indicating that Rel B is under tonic NF-B stimulatory control (Table II). Because our intent was to comprehensively identify NF-Bdependent genes downstream of TNF signaling, we next evaluated the data for those genes regulated by TNF. We found that 767 probe sets were changed by rhTNF␣ treatment, a p value (Pr(F)) of Ͻ0.05; this number was reduced to 343 probe sets at a p value of Ͻ0.01 and 159 probe sets at a p value of Ͻ0.001. The 159 probe sets were then filtered to include only those that differed by Ϯ 3-fold (relative to control) at any time during the TNF stimulation (Fig. 2B), yielding 61 probe sets. After elimination of duplicate probe sets, 50 genes met these criteria. The TNF-regulated genes were then grouped into those that were "NF-B-independent" and those that were "NF-B-dependent." Table III presents the individual data for the  22 TNF-regulated NF-B-independent genes, and Table IV presents the 28 TNF-regulated NF-B-dependent genes grouped by these criteria.
As a qualitative check for this classification scheme, the complete RNA expression profile as a function of rhTNF␣ and Dox treatment of the original 61 TNF-regulated probe sets was subjected to independent hierarchical clustering analysis. In this technique, the gene expression profile of a randomly selected probe set is grouped with its mathematically nearest neighbor. These two profiles are then combined into a single node, and the process is repeated iteratively until all of the probe sets are contained within the dendogram. The clustering algorithm generally grouped duplicate probe sets with one another (for example, JunB, NR4A1, NF-B1, NF-B1, and IL-8), indicating that the microarray is reproducibly measuring consistent changes in gene expression (Fig. 3). The clustering analysis identifies two major clusters; with the exception of CTGF (connective tissue growth factor), the top cluster contains those we identified as NF-B-independent (compare with Table III) and the bottom cluster NF-B-dependent (compare with Table IV). The genes encoding EXT1 and FST do not cluster with either the NF-Bdependent or -independent group, indicating that their grouping is indeterminant. Overall, these statistical filtering (Fig. 2B) and hierarchical clustering (Fig. 3) analyses largely converge on the same gene groupings.  (24) and affinity pull-down experiments (28), are indicated. C, FLAG⅐IB␣ Mutexpressing cells were transfected with luciferase reporter genes driven by multimeric NF-B binding site, termed (APREWT) 3 -p59rAT/LUC. Cells were stimulated with rhTNF␣ for 6 h prior to harvest and assay of luciferase reporter activity. Shown is XϮ S.D. of normalized reporter activity from triplicate plates taken from a representative transfection.

FIG. 2. Microarray identification of NF-B-dependent gene targets.
A, schematic view of experimental design. HeLa FLAG⅐IB␣ Mut cells were plated in parallel into cultures in the absence or presence of Dox (2 g/ml). After 4 days, cells were stimulated without (0 h) or with (25 ng/ml) rhTNF␣ in staggered doses. Cells were harvested simultaneously for RNA extraction. Experiments were conducted four independent times. B, schematic diagram of microarray data analysis pathway. A combined filter using statistical significance determined by two-way analysis of variance and -fold change cutoff was used to group the genes into three functionally distinct groups. See Tables I-III for individual probe sets and expression data.
The TNF-regulated NF-B-independent genes were analyzed by grouping into primary biochemical function, using pathway analysis tools from the Gene Ontology and Ingenuity databases (Table III). These genes encode proteins important in metabolism (particularly prostaglandin synthesis), signaling (dual specificity phosphatases and cAMP phosphodiesterases), transcription/nuclear factors (Jun/Fos, nuclear orphan receptor Max, GATA and zinc finger proteins), and immediate early genes with largely unknown function (immediate early response (IER3), TGF␤-inducible early response (TIEGR), early growth response-1 (EGR-1), and insulin-induced protein). The large majority are genes whose expression were induced by TNF; only Frizzled, poly(ADP-ribose) polymerase, GATA-2, and the insulin-induced protein were genes whose expression were inhibited by TNF.
Similarly, the TNF-regulated NF-B-dependent genes were classified by primary biochemical function (see Table IV). We noted that the biochemical functions of the NF-B-dependent genes were quite distinct from those that were NF-B inde-pendent, the former encoding cytokines/growth factors (IL-8, IL-6, CXCL-2/3, CTGF), cell surface receptors (CD83, IL-7, and TNFR-9), signaling proteins in the TNF pathway (TNFAIP/A20 and TRAF-1), and transcription factors (largely of the NF-B pathway itself, including NF-B1/2, Rel B, IB␣/⑀, and c-Rel). Only 2 of these 28 genes were inhibited by TNF (CTGF, follistatin), with the other 26 being up-regulated. Of these upregulated genes, the expression of all was significantly reduced when the cells were TNF stimulated in the absence of NF-B signaling. Interestingly, both CTGF and follistatin were also the only genes whose expression was increased during the TNF time course as a result of NF-B inhibition. These findings suggest that the majority of NF-B-dependent genes are positively activated by NF-B and that CTGF and follistatin may be down-regulated by NF-B.
Validation of TNF-regulated NF-B-dependent Gene Expression Profiles-To visualize and validate our identification of TNF-regulated NF-B-dependent genes, the normalized signal intensity measurements were retrieved for the IL-8, IL-6, and TABLE II Genes whose basal expression is controlled by NF-B Genes were identified by p value [Pr(F) Ͻ 0.01] in two-way analysis of variance and filtered by 3-fold changes in mean signal intensity in unstimulated samples (0 h ϪDox vs. 0 h ϩ Dox). For each of the four genes identified, the primary cellular function is shown and the common name, the GenBank™ accession number (GenBank™), the chromosomal locus (Locus), and the signed fold change (Fold Ch). Negative value means expression was inhibited in the absence of NF-B, positive value means expression was greater in the absence of NF-B and the Pr(F) as a result of TNF treatment or NF-B inhibition, respectively. TABLE III TNF-regulated genes whose expression are NF-B-independent Genes were identified by p value [Pr(F) Ͻ 0.001] as a result of TNF treatment, whose Pr(F) Ͼ 0.001 as a result of NF-B inhibition and whose mean signal intensity level was 3-fold changed at any time throughout the TNF stimulation (1, 3, or 6 h vs. 0 h ϩ Dox). Columns are as described in Table I. Nur4a, nuclear orphan receptor 4a; ZF9, zinc finger 9; IER 3, immediate early response gene 3; RGS14, regulator of G protein signaling; TIEGR, TGF-inducible early growth response; Hsp, heat shock protein; EXT, (multiple) exostoses; EGR, early growth response. Naf-1 genes and plotted as a function of TNF treatment (Fig.  4A). Expression of IL-6 and IL-8 was significantly inhibited at all TNF treatment times in the absence of NF-B. By contrast, Naf-1 expression was undetectable at the 0-and 1-h time points; 3 and 6 h after rhTNF␣ treatment, the inducible expression was significantly reduced in cells expressing FLAG⅐IB␣ Mut. Northern blot analysis was used to confirm and validate the mRNA profile changes inferred by microarray. The Northern blot hybridization signals closely follow the microarray-derived signal intensity profiles (Fig. 4B). Importantly, expression of these genes was significantly inhibited in the HeLa tTA/FLAG⅐IB␣ Mut cells stimulated in the absence of Dox, validating our microarray data analysis scheme. The peer-reviewed literature was next surveyed for studies characterizing NF-B binding sites in the group of 28 NF-Bdependent genes (Table V). Of these, reports describing 23 binding sites experimentally shown to bind NF-B were described for 16 of the genes. In fact, multiple NF-B binding sites were found in the CD83 (34), TNFAIP/A20 (35), TRAF-1 (36), NF-B1 (37), and NF-B2 (38) gene promoters. In this analysis, we identified five TNF-inducible genes (Gro-␤ and -␥, IL-7R, IB⑀, Naf-1) not previously demonstrated to be NF-B regulated. Q-RT-PCR assays were established for these mRNAs, and their expression in response to TNF stimulation in the HeLa FLAG⅐IB␣ Mut cells was measured (Fig. 4C). In the presence of Dox, genes exhibited a wide range of induction, with Naf-1 being induced 115-fold, Gro-␤ 62-fold, Gro-␥ 36-fold, B94 17-fold, and IL-7R or Rel B being induced ϳ7-fold over unstimulated values. Moreover, we noted that the expression profiles were quite distinct, with rapid induction of Gro-␤ and -␥ observed within 30 min, whereas the time for peak expression of IL-7R, B94, and Rel B was delayed (maxima occurred at 3 h), and Naf-1 peaked afterward (maxima at 9 h or later). For all these genes, expression levels in response to TNF stimulation in the absence of Dox was significantly reduced, indicating a requirement for NF-B in their expression. In addition, unstimulated Rel B expression was reduced in the absence of Dox (0 h, ϩ Dox versus 0 h, Ϫ Dox, Fig. 4C), confirming that its basal, as well as its TNF-induced, expression is NF-B-dependent (see Table II). Together, these findings validate the NF-Bdependent data set identified by microarray analysis and extend our understanding of NF-B-dependent gene control to additional members of the chemokine family, control of interleukin signaling, and NF-B autoregulation (see "Discussion").
Ectopic NF-B/Rel A Is Sufficient to Activate NF-B-dependent Gene Expression-TNF activates a variety of intracellular signaling cascades that may work in concert with NF-B to activate target gene expression (10). To determine whether NF-B binding alone was sufficient to transactivate the TNFregulated NF-B-dependent genes, the effect of expressing constitutively active Rel A was explored. In preliminary studies, transient expression of an EGFP⅐NF-B/Rel A fusion protein (39) potently activated NF-B-dependent reporter gene activity, inducing APRE-LUC activity by ϳ1,115-fold; this effect TABLE IV TNF-regulated NF-B-dependent genes Genes were identified by p value [Pr(F) Ͻ 0.001] as a result of TNF treatment, whose Pr(F) Ͻ 0.001 as a result of NF-B inhibition and whose mean signal intensity level was 3-fold changed at any time throughout the TNF stimulation (1, 3, or 6 h vs. 0 h ϩ Dox). Columns are as described in Table I. Signal ratio, signal intensity on the microarray in the absence of NF-B signaling expressed as a percentage of that produced in the presence of NF-B. CXCL, CXC motif ligand; TNFAIP, TNF ␣-induced protein; CCL, CC motif ligand; CTGF, connective tissue growth factor; GFPT2, glutamine-fructose-6-phosphate transaminase 2; IRF, interferon response factor; TTP, tristetraprolin; Naf-1, Nef-associated factor-1. was sequence-specific because the reporter activity of the NF-B binding site mutation (APRE Mut2) was relatively unaffected (Fig. 5A). We next transfected pEGFP⅐NF-B/Rel A into HeLa cells and determined changes in gene expression patterns by Q-RT-PCR. Compared with mRNA levels seen in empty vector-transfected controls, pEGFP⅐NF-B/Rel A strongly induced the expression of TRAF-1 by 70-fold, Naf-1 by 38-fold, Gro-␤ by 25-fold, and IB␣ by 15-fold, whereas it modestly induced NF-B-1 and -2 and IL-8 (Fig. 5B). By contrast, IL-6 and IL-7R expression was weakly induced by ϳ2-fold or less. The magnitude of induction of these genes probably differs from that seen during TNF induction because of a requirement for other signaling cascades in promoter activation (for example, IL-6 and -8 both have AP-1 sites that would be transactivated by the TNF-induced c-Jun NH 2 -terminal kinase-AP-1 signaling pathway (40)). Nevertheless, these observations suggested that activated NF-B, in the absence of any other TNFinduced signaling cascade, was sufficient for transactivation of 10 of 12 of these representatives of the NF-B-dependent gene data set.
NF-B-dependent Genes Are Induced in the Absence of New Protein Synthesis-We noted that the temporal induction of Naf-1 had markedly different kinetics than that of the cytokines, IL-8 and IL-6 (see Fig. 4, B and C). This finding along with the demonstration that TNF induces a variety of transcription factors raised the possibility that some members of our identified NF-B-dependent gene network may be "indirectly" NF-B regulated. Indirectly regulated genes would require the synthesis of an intermediate transacting factor, whose expression itself was NF-B-dependent. To test whether this was possible, we exploited the fact that TNF-induced NF-B activation is protein synthesis-independent because the signaling molecules required for the initial release of cytoplasmic NF-B are preformed (10,41).
HeLa cells were TNF-stimulated in the absence or presence of cycloheximide (CHX) at concentrations sufficient to inhibit Ͼ98% of amino acid incorporation into protein (33). As seen in Fig. 6A, in the absence of CHX rhTNF␣ rapidly induces IB␣ proteolysis (compare 15 min to control), followed by its resynthesis after 1-3 h of treatment. In the presence of CHX, IB␣ proteolysis was not affected; however, its resynthesis was significantly reduced (compare 3 h ϪCHX to 3 h ϩCHX, Fig. 6A). Consistent with the inhibition of IB␣ resynthesis, we found that TNF-induced activation of NF-B binding is much greater in the presence of CHX 1-3 h after stimulation ( Fig. 6B and Ref. 41). Enhanced NF-B binding is found at this time because IB␣ and TNFAIP/A20 resynthesis is inhibited; these two negative autoregulators of the NF-B pathway are involved in termination of NF-B signaling (see "Discussion" and Ref. 42). To further show that enhanced Rel A accumulation was seen in CHX-treated samples, Western immunoblot was performed on purified nuclear extracts (Fig. 6C) where Rel A signal normalized to ␤-actin was increased 8-fold at 1-and 2-fold at 3 h relative to the TNF-only-treated samples. These data indicate that exaggerated NF-B activation was produced in the absence of new protein synthesis. We next measured changes in steady state mRNA abundance by Q-RT-PCR (Fig. 6D). As seen in Fig. 6D, TNF-induced expression of the NF-B-dependent genes was strongly maintained in the presence of CHX. In fact, with the exception of NF-kB-1 and -2 and IL-7R, expression in the presence of CHX was 5-fold or greater than that produced by TNF in the absence of CHX. Importantly, the NF-B2 and Naf-1 genes, both of which show later expression profiles (Table IV and Fig. 4), were still induced in the presence of CHX, strongly suggesting that transactivation of these genes is likely to be directly mediated by preformed transcription factors (NF-B).
Identification of Uncharacterized NF-B Binding Sites in NF-B-dependent Genes-The promoters of genes not previously reported to have NF-B binding sites were searched for the presence of high affinity NF-B binding sites using a positional weight matrix (PWM) algorithm. PWMs represent a scoring matrix for all possible bases at every position within a binding motif that allows for weighting of invariant nucleotides in a binding site to contribute to a greater degree than would degenerate positions in the binding motif. Moreover, the PWM score is related to the Gibbs free energy of binding (43,44), and statistical probabilities can be assigned to predicted sites based on the probability of that sequence randomly occurring in nonregulatory DNA regions (45). With this, high affinity NF-B binding sites (matrix scores of Ͼ0.9) were predicted for nine additional genes (Table V). To determine whether these represented bona fide NF-B binding sites, these putative sites from the Gro, TNFIAP-1/B94, IL-7R, Rel B, IB⑀, and Naf-1 promoters were chemically synthesized as duplex oligonucleotides and subjected to EMSA using either control or TNF-stimulated NE FIG. 3. Hierarchical clustering of TNF-regulated genes. 61 TNF-regulated genes (Pr(F) Ͻ0.001 by TNF treatment) whose expression changed 3-fold at any time throughout the stimulation were Zscore normalized and subjected to hierarchical clustering. Two major clusters are identified. Note that duplicate probe sets, representing the same gene, co-cluster. (Fig. 7A). All the binding sites tested were bound by apparently identical complexes as the constitutive NF-B1⅐p50 homodimer and the TNF-inducible Rel A/NF-B1 complex that recognized the APRE (Fig. 7A and Ref. 24). Sequence specificity for these complexes on Naf-1 binding was determined (Fig. 7B) where WT and site mutation of the Naf-1 binding site was used in competition assay. Both the constitutive NF-B1⅐p50 and the inducible Rel A/NF-B1 complex were specifically competed by the wild type, but not mutant, oligonucleotide. To further show that Rel A recognized the DNA binding sequences, recombinant GST⅐Rel A (12-308) was used for binding in EMSA (Fig. 7C). GST⅐Rel A (12-308) bound Rel B, Naf-1, and Gro to a similar degree as the APRE and bound B94, IL-7, and IB⑀ slightly less strongly. To further underscore that GST⅐Rel A (12-308) bound in a sequence-specific manner, competition in EMSA was performed. As compared with the APRE standard, GST⅐Rel A   FIG. 4. Validation of selected NF-B-dependent genes. A, the mean and S.D. of the normalized signal intensities were retrieved from the microarray data set and plotted as a function of TNF stimulation. Shown are the individual plots for IL-8, NF-B1, and Naf-1 in the presence or absence of Dox. B, total RNA (from cells treated as in Fig. 2A) was analyzed by Northern blot probed with indicated radiolabeled cDNA. Shown is an autoradiographic exposure. Bottom panel, ␤ thymosin-10 (␤-Thymo) was used as an internal control for RNA recovery (74). C, validation of six novel NF-B-dependent genes. Q-RT-PCR analysis of RNA extracted from cells treated as in Fig. 2A for B94, Gro-␤,␥, IL7R, Rel B, and Naf-1. Shown is -fold change (mean Ϯ S.D.) of Q-RT-PCR signal relative to unstimulated cells (0 h, ϩ Dox).
The Naf-1 Gene Promoter Is Directly Transactivated by NF-B/Rel A-Our microarray analysis identifies Naf-1 as a previously uncharacterized member of the NF-B-dependent gene network. We noted that Naf 1 expression induced by TNF is markedly delayed from that of the rapidly responding cytokine group (Fig. 4), even though Naf-1 expression is induced by ectopic expression of EGFP⅐Rel A (Fig. 5), is TNF-inducible in the absence of new protein synthesis (Fig. 6), and contains a high affinity NF-B binding site (Fig. 7). These characteristics strongly suggested to us that the promoter is directly bound and transactivated by NF-B/Rel A. To further determine whether this was the case, the Ϫ626/ϩ111 human Naf-1 promoter luciferase reporter vector was cotransfected with increasing concentrations of EGFP⅐NF-B/Rel A into HeLa cells. EGFP⅐NF-B/Rel A strongly activated Naf-driven luciferase reporter activity (Fig. 7E). Introduction of a site mutation in the NF-B element in the Naf-1 promoter that did not bind Rel A/NF-B1 in TNF-stimulated NE (Fig. 7B) or recombinant Rel A (Fig. 7D) in EMSA rendered the promoter inert to Rel A-induced transactivation (Fig. 7E). To demonstrate that Rel A directly bound to the endogenous Naf-1 promoter, ChIP assays were performed on a time series of TNF-stimulated cells. Immunoprecipitation of Rel A-associated chromatin showed rapid inducible Rel A binding to the endogenous Naf-1 promoter (Fig.  7F). Interestingly, the kinetics of Rel A binding are indistinguishable from that of binding to the endogenous IB␣ promoter; these data further indicate that Naf-1 is directly transactivated by NF-B and is a bona fide member of the TNFregulated NF-B-dependent gene network.
Inducible and Dox-dependent NF-B/Rel A Binding to Endogenous IB⑀, Gro-␤/␥, and Naf-1 Promoters-To further validate the direct NF-B-dependent gene data set, ChIP assays were established for the novel NF-B-dependent genes. ChIP was used to determine binding of Rel A to the endogenous IB⑀, Gro-␤/␥, and Naf-1 genes in the HeLa tTA/FLAG⅐IB␣ Mut cells in comparison to the IB␣ promoter used as a positive control. In the presence of Dox, TNF rapidly induces the binding of Rel A to these novel endogenous genes (Fig. 8). Conversely, in cells stimulated in the absence of Dox, the binding of Rel A is significantly reduced (Fig. 8, right lanes). These findings validate our identification of an NF-B-dependent gene data set.

TABLE V Promoter analysis of NF-B-dependent genes
Published work on the NF-B-dependent genes was surveyed for description of functional NF-B binding sites and tabulated. When no relevant publications were found, 500 bp of the promoter was retrieved and scanned for high affinity binding sites by weighted matrix algorithm (46,63). Score, matrix similarity score (scale 0 -1). Induc?, whether the gene is inducible by an NF-B activator (64); Y, yes; NF, not found.

DISCUSSION
TNF is a pleiotropic cytokine that acts as an essential mediator of pulmonary cytokine cascade, hepatic acute phase response, and regulator of leukocyte activation and apoptosis (1). Upon its ligation to cell surface receptors, TNF induces protein recruitment to cytoplasmic death domains, assembling a signaling complex composed of TRADD, FADD, TRAF2 (and others) that activate divergent intracellular signals, including the c-Jun NH 2 -terminal kinase-AP-1 and the IKK-NF-B pathways to induce genomic responses in the target cell (8,10). Although the IKK-NF-B pathway is critical for inducing tissue inflammation and preventing TNF-induced programmed cell death, surprisingly little is known about its downstream gene targets. In this study, we have exploited high density oligonucleotide arrays in conjunction with a tightly tetracycline-regulated expression of an NF-B inhibitor to dissect TNF-regulated genes under NF-B control. Using a widely accepted two-tiered filtering strategy that employs both statistical significance and a minimal -fold change "cut-off," we identified 28 unique genes that are NF-B-dependent during the first 6 h of TNF stimulation. Our findings indicate that over half (28 of 50) of the TNF-regulated genes are under NF-B control and that 26 of these 28 are up-regulated by TNF. We have validated our findings by demonstrating that 10 of 12 tested NF-B network members can be up-regulated by ectopic Rel A expression in the absence of TNF signaling and that their induction is independent of new protein synthesis. Moreover, we have identified bona fide NF-B binding sites in five promoters not previously known to be NF-B regulated and have shown that the endogenous genes inducibly bind NF-B/RelA. Together these observations strongly suggest that these genes are direct targets of NF-B. Our data further indicate that NF-B controls genes with different functional activities that govern a multitude of biological processes, including NF-B autoregulation, NF-B subunit expression, and chemokine expression.
A previous cDNA microarray study identified the spectrum of TNF-induced genes in HeLa cells within the first 4 h of stimulation (22). In this study, the expression of 16 TNF-induced genes was validated and examined for NF-B dependence using short interfering RNA-mediated Rel A knockdown. These results showed that the expression levels of 12 genes were partially inhibited by Rel A knockdown, 7 of which were identified in our analysis, including Gro-␣, IL-6, Cox-2, B94, A20, IB␣, and Syndecan 4 (compare Ref. 22 and Table III). Because NF-B dependence was determined only for these 16 selected genes, our study design extends this work by comprehensively identifying genes requiring NF-B for basal expression (Table II), experimentally identifying the NF-B-dependent data set (Fig. 2), and demonstrating direct NF-B binding by EMSA and ChIP assays. Further comparison between our studies is complicated by the differences in microarray platform technology (in terms of robustness of hybridization signal and probe set mapping) and differences in post-hybridization data analysis. FIG. 7. Binding site analysis of NF-B-dependent promoters. A, control (Ϫ) or TNF-stimulated (25 ng/ml, 15 min) NE were extracted and used for EMSA with indicated radiolabeled duplexes (bottom). The bound complexes are shown. B, competition for Naf-1 binding in EMSA was performed. Control (Ϫ) or TNF-stimulated (25 ng/ml, 15 min) NE were incubated with labeled Naf-1 WT probe in the presence or absence of 100-fold excess of unlabeled WT or Mut duplex prior to nondenaturing electrophoresis. Bound complexes are shown. C, recombinant GST⅐Rel A (12-308) was purified (24). 100 ng was incubated with indicated radiolabled duplexes prior to EMSA. An autoradiogram is shown. D, recombinant GST⅐Rel A (12-308) was incubated with labeled APRE WT or Naf-1 WT in the presence or absence of 100-fold excess of unlabeled duplex prior to EMSA. E, Ϫ626/ϩ111 Naf-1/LUC and the NF-B site mutation, Naf-1⌬B/LUC, were transfected into HeLa epithelial cells in the presence of increasing amounts of pEGFP⅐NF-B/Rel A (indicated on X axis). Empty pEGFP was added to the transfections to make an equal amount of expression vector in all samples. Shown is normalized luciferase reporter activity from a representative transfection in triplicate (X Ϯ S.D.). F, ChIP assay on control or TNF-stimulated HeLa cells. Antibodies used were anti-Rel A (C-20; Santa Cruz) or IgG (PI) as indicated. Immunoprecipitated DNA was PCR amplified with primers specific for the IB␣ or Naf-1 promoters using a cycle number that yielded linear signals in control reactions. Shown is the ethidium bromide-stained agarose gel.
Autoregulation of the TNFR-IKK-NF-B Pathway-Our results indicate that inhibitory members of the TNF-IKK-NF-B signaling pathway are themselves NF-B-dependent, suggesting that the NF-B pathway is under negative autoregulatory control at multiple levels. A major functional group of NF-Bdependent genes is the ankyrin repeat (AR)-containing proteins, a secondary structural domain common to proteins involved in the inhibition of NF-B translocation and binding (17). AR repeat domains are found in NF-B1, NF-B2, BCL-3, IB␣, and IB⑀. Several of these AR-containing proteins participate in the "NF-B-IB" loop, an autoregulatory feedback loop that begins when cytoplasmically liberated NF-B enters the nucleus to stimulate transcription (15,25,47). Activated NF-B induces the resynthesis of the AR repeat-containing proteins that serve to recapture and inactivate NF-B, restoring the cell to homeostasis. The previously characterized ARcontaining NF-B-dependent targets include IB␣ and BCL-3, which display distinct specificities for specific NF-B isoforms and distinct kinetics of induction. IB␣ binds relatively selectively to Rel A and is rapidly resynthesized (Ͻ1 h after TNF stimulation), whereas BCL-3 binds selectively 50-kDa NF-B1 and is more slowly resynthesized (3-9 h) (17,48). This finding suggests that inactivation of NF-B isoforms by the NF-B-IB autoregulatory loop is highly temporally orchestrated. Our findings here indicate that both NF-B-1 and -2 are slowly induced by TNF relative to IB␣. NF-B-1 and -2 also bind a distinct spectrum of NF-B transactivators, with NF-B1 binding Rel A (49) and NF-B2 binding Rel B (33). In this sense, induction of NF-B1/2 by NF-B/Rel A may serve as a redundant autoregulatory mechanism to the actions of IB␣. In addition, our study indicates that IB⑀, a recently described member of the IB family that binds to c-Rel with high affinity (50), is also a member of the NF-B-IB autoregulatory loop. Our findings indicate that IB⑀ is a direct NF-B target, having identified an atypical NF-B binding site in the IB⑀ promoter that binds RelA, and show the endogenous gene binds NF-B in response to TNF. Further studies will be needed to determine the exact contribution of IB⑀ to cytoplasmic sequestration of c-Rel or Rel A after TNF stimulation.
A second level of negative autoregulation of the TNF-IKK-NF-B signaling pathway is the TNAIP/A20 protein, whose expression is strongly and rapidly induced by TNF in an NF-B-dependent fashion (Table IV). A20 is now known to be a ubiquitin ligase that associates with RIP and mediates its degradation through the ubiquitin-proteasome pathway (51), inhibiting TNF-induced NF-B signaling and apoptosis (52). Without A20 expression, TNF induces chronic NF-B activation and inflammation in a mouse model (53). Recently we have developed a mathematical model to describe the time-dependent evolution of the NF-B pathway after TNF stimulation (42). Consistent with the findings for the A20-deficient mouse, our modeling studies indicate that without the second level of autoregulation at the level of the IKK mediated by A20, the IB autoregulatory pathway alone is unable to terminate NF-B binding activity. Even in the presence of robust IB␣ resynthesis, the persistently activated IKK will continuously degrade IB␣. Finally, we also note that the receptor adapter protein, TRAF-1, is also TNF-induced by activated NF-B. TRAF-1 is a unique member of the TRAF family that lacks the NH 2 -terminal RING finger motif (54). The current role of TRAF-1 in signaling has not yet been resolved, but it is known to associate with TNFRI, II, CD30, and TNFS9 where it may, like A20, influence caspase activation and affect cellular apoptosis. Therefore, A20 (and perhaps TRAF-1) are involved in a second level of autoregulation of the TNF-IKK-NF-B signaling pathway by disrupting TNFR coupling with IKK.
A third level of negative autoregulation is mediated by the zinc finger protein, tristetraprolein (TTP)/zinc finger protein 36 (Table IV). TTP is a prototypic member of the CCCH motifcontaining zinc finger family that promotes the degradation of mRNA transcripts containing AU-rich elements, including TNF␣, in their 3Ј-untranslated regions (55). Interestingly, mice deficient in TTP develop chronic inflammatory disease (arthritis and dermatitis) and have exaggerated TNF secretion from activated macrophages, indicating the essential role for TTP in terminating TNF signaling by degrading TNF mRNA (Ref. 56 and references therein). Together, our findings strongly suggest that NF-B pathway autoregulates target proteins to actively terminate TNF signaling at multiple levels of the signaling pathway.

NF-B Initiates a Program of NF-B Expression Controlling Subunit
Exchange-This study indicates that expression of the transactivating Rel B and c-Rel isoforms is under positive NF-B control and is the first to indicate Rel B is under tonic NF-B regulation. An earlier report examining the effect of ectopic c-Rel expression reported that it inhibited endogenous c-Rel expression (57). Our findings that c-Rel is induced by TNF treatment in an IB-sensitive manner (Table IV) can be reconciled in that other combinations of NF-B heterodimers have different transcriptional consequences than c-Rel homodimers. The enhanced expression of Rel B and c-Rel may mediate the phenomenon of exchange of NF-B subunits on NF-B-responsive promoters (29). Here, the rapidly translocated Rel A/NF-B1 heterodimers initially bound to their target promoters are replaced by Rel B/NF-B2 and c-Rel/NF-B2 heterodimers. This dimer exchange is thought to produce, in some cases, transcriptional shut-off of activated genes (58). Our array findings suggest that NF-B initiates a coordinated cascade of NF-B-subunit expression that sequentially binds target genes during the evolution of the TNF response and could be thought of as an additional level of NF-B autoregulation.
Role of NF-B in TNF-induced Chemokine Expression-The actions of TNF initiate the multistep process by which circulating leukocytes are recruited into inflamed tissues. A critical step in this process occurs when activated leukocytes are stimulated to migrate through the vascular endothelium toward chemical gradients of chemoattractant cytokines ("chemo- FIG. 8. Dox dependence of NF-B/Rel A binding to newly identified NF-B-dependent genes. HeLa FLAG⅐IB␣ Mut cells were plated in parallel into cultures in the absence or presence of Dox (2 g/ml). Cells were either stimulated with nothing (Ϫ) or 20 ng/ml TNF (ϩ) for 1 h, and cross-linked chromatin was subjected to immunoprecipitation with anti-RelA. Immunoprecipitated DNA was PCR amplified using primers specific for the indicated genes (shown at left). To ensure that equal amounts of chromatin were subjected to immunoprecipitation, input DNA was subjected to PCR (Input, bottom). nc, negative control; genomic, genomic DNA (25 ng).
kines") (reviewed in Ref. 59). Chemokines are a superfamily of small basic (heparin-binding) proteins divided into three major functionally distinct groups based on the number and spacing of highly conserved NH 2 -terminal cysteine residues, termed the C, CC, and CXC chemokines. A striking finding of our functional classification is that the rapid expression of TNFinduced cytokines is completely NF-B-dependent. Of these, IL-8 and CXCL-1, -2 and -3 (Gro-␣/␤/␥) are CXC motif-containing chemokines, responsible for inducing neutrophil and monocyte chemotaxis (59,60). Exodus-1 is a CC motif-containing chemokine that stimulates recruitment of monocytes and immature dendritic cells (61). Although IL-8 and Gro-␣ have been extensively demonstrated to be NF-B-dependent, our study demonstrates Gro-␤/␥ are also direct targets of NF-B action. Together these NF-B-dependent chemokines play important roles in coordinating the recruitment of distinct subsets of activated leukocytes, including neutrophils and monocytes, to the site of inflammation, initiating the process of TNF-induced tissue inflammation. Our findings suggest a mechanism by which inhibition of NF-B signaling could reduce leukocyte chemotaxis in inflammatory diseases.
Insights into Activation of NF-B-dependent Genes-For us, a surprising finding from this study is that the expression of NF-B-dependent genes is not tightly temporally linked with the appearance of NF-B binding in the nucleus. Although many genes, such as the chemokines, are rapidly induced by TNF in an NF-B-dependent manner, other genes, such as Naf-1, show delayed kinetics of expression. Our findings that Naf-1 expression is induced by ectopic Rel A expression and is TNF-inducible independently of new protein synthesis, that the promoter contains a high affinity NF-B binding site that is transactivated by Rel A, and that the endogenous promoter is rapidly bound by Rel A after TNF stimulation suggests to us that Naf-1 expression is under direct NF-B control. These findings also suggest that additional regulatory steps in NF-B activation of Naf-1 are operative. Although previous ChIP experiments have suggested that NF-B binding occurs in two distinct "waves" in lipopolysaccharide-stimulated macrophages (29), our findings do not support this as a universal mechanism controlling late gene activation by NF-B. As seen in Fig. 7D, similar rates of NF-B binding to the early (IB␣) and late (Naf-1) promoters are seen, even though the kinetics of mRNA induction are markedly different, with IB␣ being maximally induced 9-fold at 1 h and Naf-1 peaking at 115-fold at 9 h (Fig.  4, Table IV). Rather, our findings are consistent with a mechanism by which a rate-limiting step in promoter activation occurs after NF-B associates with its targets in chromatin. These mechanisms will require further investigation but could involve promoter-specific chromatin remodeling events or differences in the rate of coactivator or polymerase recruitment.
NF-B Controls Genes in the Absence of Exogenous Activation-Real-time measurements of NF-B/Rel A fluorescent protein fusions indicate that this protein dynamically shuttles through the nucleus of unstimulated cells (62), a finding consistent with the low levels of Rel A binding in unstimulated nuclei (Fig. 7D). If so, it has not been completely resolved whether the constitutive shuttling of NF-B is sufficient for target gene activation. A novel outcome of our study design is the identification of four genes whose basal expression are under this "constitutive" NF-B control (Table II). Of these, Rel B expression is apparently tonically NF-B activated because its basal expression is reduced when the FLAG⅐IB␣Mut protein is induced (Fig. 4C). Careful scrutiny of the data in Table  IV indicates that the basal expression of several other genes is also significantly inhibited by NF-B inhibitor, but less than 3-fold, indicating that NF-B may control the basal activity of more than just these four genes shown in Table II. Independent information that basal levels of NF-B are present in our system include the findings that Rel A immunoactivity can be demonstrated in unstimulated nuclei (Fig. 6C) and that NF-B is associated with target genes in the absence of stimulation (Fig. 7F). The mechanism by which basal levels of stem cell growth factor, prostaglandin synthase, and cholinergic nicotinic receptor are induced by NF-B inhibition may be indirect and will require further study.
In summary, we have used high density oligonucleotide microarrays to dissect the contribution of NF-B-dependent gene expression to the TNF response and experimentally validate that these genes are directly under NF-B control. Our studies have extended the identities of the autoregulated members of the IKK-NF-B pathway and suggest autoregulation occurs at multiple levels in the signaling pathway from the level of the receptor (TNFAIP/A20, TRAF-1) to ankyrin repeat molecules (NF-B1/2, IB␣/⑀) and mRNA stability (TTP) and subunit exchange. NF-B mediates the majority of TNF-induced chemokine expression and is responsible for significant alterations in the ability of cells to metabolize arachidonic acid. We have identified high affinity Rel A binding sites in the Gro ␤/␥, B94, Rel B, IL-7R, IB⑀, and Naf-1 genes and experimentally demonstrated that the endogenous Gro ␤/␥, IB⑀, and Naf-1 promoters inducibly bind Rel A in an IB-dependent manner. Finally, our findings suggest that NF-B-dependent genes are expressed at markedly different times, indicating that ratelimiting steps in gene activation occur after NF-B binds its genomic targets. These intriguing observations will be the focus of further work in understanding the molecular actions of NF-B.