Characterization of the transforming growth factor-beta 1-induced apoptotic transcriptome in FaO hepatoma cells.

We have previously shown that transforming growth factor-beta(1) (TGF-beta(1))-induced apoptosis in FaO hepatoma cells is mediated by cytochrome c release, apoptosome formation, and caspase activation. Although TGF-beta(1) acts via the SMAD signaling pathway to initiate de novo gene transcription, little is known about the downstream gene targets that are involved in the regulation of apoptosis. Therefore, in this study, we used in-house microarrays (approximately 5500 genes) to identify pathway-specific gene clustering in TGF-beta(1)-treated cells. A total of 142 genes showed time-dependent changes in expression during TGF-beta(1)-induced apoptosis. The polycaspase inhibitor benzyloxycarbonyl-VAD-fluoromethyl ketone, which, on its own, had no effect on gene transcription, blocked TGF-beta(1)-induced cell death and significantly altered the expression of 261 genes, including 185 down-regulated genes. Cluster analysis identified up-regulation of early response genes (0-4 h) encoding for the extracellular matrix and cytoskeleton, including the pro-apoptotic CTGF gene, and delayed response genes (8-16 h), including pro-apoptotic genes. A second delayed response cluster (44 genes) was also observed when TGF-beta(1)-induced caspase activation was blocked by benzyloxycarbonyl-VAD-fluoromethyl ketone. This cluster included genes encoding stress-related proteins (e.g. Jun, ATF3, TAB1, and TANK), suggesting that their up-regulation may be in response to secondary necrosis. Finally, we identified an early response set of nine down-regulated genes that are involved in antioxidant defense. We propose that the regulation of these genes by TGF-beta(1) could provide a molecular mechanism for the observed elevation in reactive oxygen species after TGF-beta(1) treatment and may represent the primary mechanism through which TGF-beta(1) initiates apoptosis.

TGF-␤ 1 induces its many biological effects by binding to specific receptors on the plasma membrane, initiating a serine/ threonine kinase-catalyzed signaling pathway. Although all the subsequent downstream effects in the apoptotic pathway have not been delineated, it is clear that caspase activation is an essential component, as benzyloxycarbonyl-Val-Ala-Asp-fluoromethyl ketone (Z-VAD-fmk), a polycaspase inhibitor, not only blocks activation of caspase-3 and caspase-7, but also abrogates apoptotic cell death (6,7). In addition, studies in hepatoma cell lines have shown that caspase-2, -8, and -9 are also processed and activated during TGF-␤ 1 -induced cell death (14,15). Activation of effector caspases (caspase-3 and caspase- 7), which kill the cell by cleaving and inactivating/activating key proteins (for review, see Refs. 16 and 17), can occur by one of two major pathways involving either stimulation of cellsurface death receptors or perturbation of mitochondria (18). In both pathways, a caspase cascade is activated by a two-step mechanism in which initiator (apical) procaspases (procaspase-8 and procaspase-9) are recruited and activated within large multiprotein complexes known as the death-inducing signaling complex and the apoptosome, respectively (for review, see Ref. 19). In the case of mitochondrion-mediated cell death, the release of cytochrome c is a common response to many apoptotic stimuli (20,21), initiating the ATP/dATP-dependent oligomerization of Apaf-1 to form the apoptosome. This large multiprotein complex recruits and facilitates autoprocessing of caspase-9 to form an Apaf-1⅐caspase-9 holoenzyme complex, which then recruits and processes the effector caspases (22)(23)(24)(25)(26)(27).
The formation of the Apaf-1-containing apoptosome complex is central to the activation of the caspase cascade in mitochondrion-mediated cell death, and we have recently shown that TGF-␤ 1 -induced apoptosis in FaO hepatoma cells induces cytochrome c release and assembly of the ϳ700-kDa apoptosome complex, which then activates the effector caspases (caspase-3 and caspase-7) (28). Cytochrome c release during TGF-␤ 1 -induced apoptosis has subsequently been confirmed in fetal hepatocytes (29). Furthermore, overexpression of Bcl-x L (30), an anti-apoptotic Bcl-2 family member, protects against TGF-␤ 1induced cytochrome c release and apoptosis in prostate epithelial cells. Thus, although TGF-␤ 1 -induced apoptosis is a receptor-mediated phenomenon, it does not involve death-inducing signaling complex formation and direct primary activation of caspase-8 (28), but acts by triggering the mitochondrial caspase activation pathway.
The mechanism by which TGF-␤ 1 initiates cytochrome c release is as yet unknown. TGF-␤ 1 induces its other varied biological effects by acting through specific transmembrane type I and II serine/threonine kinase receptors (TGFBR-1 and TGFBR-2). The cytokine binds to TGFBR-2, which then phosphorylates and activates the TGFBR-1 kinase (1,32), which, in turn, phosphorylates the receptor-associated Smad2 and Smad3 proteins. These are then released from the receptor complex and bind to Smad4 to form a heterotrimeric complex, which translocates to the nucleus (33,34). SMAD complexes interact directly or indirectly with TGF-␤ 1 -responsive promoter sequences and, in combination with other transcription factors, regulate the transcription of specific genes (1,35). This mechanism has been well established in a variety of TGF-␤ 1 signaling paradigms, and recent studies have shown that overexpression of dominant-negative Smad2 and Smad3 and also the inhibitor Smad7 not only prevents SMAD-mediated signal transduction, but also abrogates the apoptotic effects of TGF-␤ 1 (36,37). Thus, TGF-␤ 1 -induced apoptosis appears to require changes in the expression levels of key proteins. In support of this, earlier studies have shown that TGF-␤ 1 -induced apoptosis in adult and fetal rat hepatocytes is blocked by cycloheximide (7,38). Furthermore, we have shown that TGF-␤ 1 -induced apoptosis in FaO hepatoma cells requires both de novo transcription and translation, as it is blocked by actinomycin D and cycloheximide (see Fig. 1) (39). As TGF-␤ 1 -induced apoptosis occurs via transcriptional activation, some of the resultant and downstream changes in gene expression must be involved in the release of cytochrome c and induction of apoptosis. Therefore, we have used DNA microarrays to characterize time-dependent changes in gene expression during TGF-␤ 1induced apoptosis. Our results have identified distinct clusters of both up-and down-regulated genes that may act in a coordinated manner to initiate and amplify the apoptotic program. At early times, a cluster of genes involved in protection against reactive oxygen species (ROS) was down-regulated, allowing an increase in ROS, which would be predicted to induce cytochrome c release and subsequently apoptosome formation. A second cluster of pro-apoptotic genes was up-regulated at later times, thereby amplifying the apoptotic response. Rather surprisingly, in the presence of TGF-␤ 1 , the polycaspase inhibitor Z-VAD-fmk up-regulated a number of genes involved in stress responses, suggesting that the expression of some TGF-␤ 1regulated genes is caspase-dependent. In conclusion, our studies show that TGF-␤ 1 induces apoptosis by time-dependent changes in the expression of a number of critical genes, which then act in a concerted manner to initiate and propagate apoptotic cell death.

EXPERIMENTAL PROCEDURES
Cell Culture-FaO cells were cultured in Ham's nutrient mixture F-12 with Glutamax supplemented with 10% fetal calf serum, 100 units/ml penicillin, and 100 g/ml streptomycin and maintained in a humidified atmosphere of 5% CO 2 at 37°C. Cells were seeded at a density of 2 ϫ 10 4 /cm 2 on day 0. After 24 h, the medium was changed to low serum Ham's nutrient mixture F-12 containing 1% fetal calf serum. After another 24 h, the medium was changed again, and the cells were treated with 0.5 ng/ml TGF-␤ 1 (2 g/ml stock in 4 mM HCl containing 1 mg/ml bovine serum albumin) or an HCl control. In some experiments, cells were pretreated with 50 M Z-VAD-fmk (200 mM stock in Me 2 SO), 0.1 g/ml cycloheximide (0.1 mg/ml stock in phosphate-buffered saline), or 250 nM actinomycin D 30 min before TGF-␤ 1 administration.
Annexin V Staining-Apoptosis was assessed using the annexin V staining method adapted for adherent cells as previously described (28). Attached cells were trypsinized (0.5ϫ trypsin/EDTA in phosphate-buffered saline) and combined with the detached cells, therefore composing the total cell population, which was then pelleted (200 ϫ g, 5 min, 4°C), resuspended in 10 ml of fresh medium containing 10% fetal calf serum, and incubated for 20 min at 37°C. Cells (0.5 ϫ 10 6 ) were pelleted and resuspended in 1 ml of annexin V buffer (10 mM HEPES/NaOH (pH 7.4), 150 mM NaCl, 5 mM KCl, 1 mM MgCl 2 , and 1.8 mM CaCl 2 ). Annexin V (1.5 l) was added, and the cells were incubated for a further 8 min at room temperature before labeling with 30 l of propidium iodide (50 g/ml) for 1 min and subsequent analysis by flow cytometry.
RNA Extraction and Labeling-Attached cells were removed from the flasks with a scraper and pelleted (200 ϫ g, 5 min, 4°C) with the detached cells. Cell pellets were then washed with ice-cold phosphatebuffered saline, snap-frozen, and stored at Ϫ80°C. Array construction, RNA extraction and labeling, hybridization, and analysis of fluorescence were all carried out as previously described (40), except that the optimal hybridization time was found to be 48 h.
Data processing was carried out using ConvertData Version 3.3.3 2 before importing files into GeneSpring Version 4.0.4 (Silicon Genetics, Redwood City, CA) for overall analysis. K-means clustering analysis using smooth correlation was performed according to the advanced analysis techniques manual supplied with GeneSpring. Genes were selected as significantly altered if they showed a consistent change in three different experiments and the mean value represented a -fold change of at least Ϯ1.5 (where ϩ1 represents a 100% increase, i.e. double the control value, and Ϫ1 represents a 50% decrease, i.e. half the control value). This method of -fold change analysis gives equal emphasis to underexpressed and overexpressed genes while avoiding the loss of detail that occurs when logarithms are used. Using this formula [ϩ/Ϫ(ext(abs(log(ratio))))Ϫ1] to calculate the change in gene expression means that the y axis can be numbered from positive to negative infinity. The gene designations used in the figures and tables are those approved by the HUGO Gene Nomenclature Committee, although in some cases, the alternative, more usual designation (published) is also shown.
Sequencing-For practical purposes, only the sequences of the genes discussed under "Results" were confirmed using an ABI PRISM ® Big-Dye TM Terminator Version 3.0 cycle sequencing kit.
Reagents-TGF-␤ 1 was purchased from R&D Systems (Oxford, UK). All cell culture reagents were obtained from Invitrogen (Paisley, Scotland). The caspase inhibitor Z-VAD-fmk was purchased from Enzyme Systems Products (Dublin, CA). Annexin V was purchased from Bender Medsystems Diagnostics GmbH (Vienna, Austria). All other chemicals were purchased from Sigma (Poole, Dorset, UK).

Global Changes in Gene Expression during TGF-␤ 1 -induced
Apoptosis-Previous studies have shown that TGF-␤ 1 -induced apoptosis requires de novo protein synthesis (7,38) and involves receptor-associated SMAD proteins (36), indicating that TGF-␤ 1 -induced apoptosis requires transcriptional and translational alterations in the expression of key proteins. Therefore, we reasoned that TGF-␤ 1 should induce a set of early response genes that would initiate apoptotic cell death and that other genes might be activated later in response to the cellular changes brought about by the cell death program. To characterize these changes, we investigated gene expression throughout the time course of TGF-␤ 1 -induced apoptotic cell death and also examined the effect of transcriptional (actinomycin D) and translational (cycloheximide) inhibitors on both apoptosis and gene expression. In addition, we also used the polycaspase inhibitor Z-VAD-fmk to block caspase activation and thereby elucidate any potential gene changes caused by activation of the execution phase of the cell death program. Both actinomycin D and cycloheximide significantly inhibited TGF-␤ 1 -induced apoptosis in FaO cells at 4 and 8 h after treatment ( Fig.  1), although by 16 h, the degree of inhibition was less marked.
To identify potential mediators of TGF-␤ 1 -induced apoptosis, gene expression profiles were analyzed from the various treatments using in-house human DNA microarrays containing 5548 -5784 cDNA clones and included as many IMAGE clones for apoptotic genes as were available at the time of the experiments. These arrays have been successfully used to detect the changes in gene expression in human cell lines after exposure to various apoptotic stimuli. 5 Homologene software 6 was used to determine, for several key genes, the degree of identity between curated/calculated human and rat orthologs. In the main, the observed homology between rat and human clones was Ͼ80%. Using cross-species arrays, it is possible to get false negatives, but not false positives; and in our experiments, only a small number of the genes on the arrays exhibited no hybridization (see below), thereby validating our approach.
Experiments with the actinomycin D-and cycloheximideinhibited FaO cells showed that both cycloheximide and actinomycin D affected a very large number of genes that are constitutively expressed (data not shown). Consequently, it was difficult to identify those genes that were specifically involved in TGF-␤ 1 -induced apoptosis. We therefore decided to focus on the time-dependent changes in gene expression that occur during TGF-␤ 1 -induced apoptosis, with a view to correlating these changes with the corresponding apoptosis-related biochemical changes that we have detailed in a previous study (28). In this study, we showed that, in the first 4 h after TGF-␤ 1 treatment, there was little or no caspase processing and only a small increase in annexin V-positive cells ( Fig. 1 in this study and Figs. 1-3 in Ref. 28). After 8 h, caspase processing and activity were initiated and continued to increase to a maximum between 16 and 24 h. We therefore examined the gene expression changes at 4, 8, and 16 h after exposure to TGF-␤ 1 , when ϳ12, ϳ25, and ϳ40% of the cells were apoptotic, respectively ( Fig. 1). A second set of samples that had been co-treated with TGF-␤ 1 and Z-VAD-fmk was also analyzed. Apoptosis was completely inhibited by Z-VAD-fmk at 4 and 8 h and was still extensively inhibited (70 -80%) at 16 h.
Gene expression profiles of the normalized data were collated from three independent experiments and revealed that, in all treatments, ϳ5000 genes (ϳ90% of the total gene array) hybridized successfully to FaO cDNA. The majority of the genes that were detected showed little or no alteration in their expression levels. However, after TGF-␤ 1 treatment, 142 genes were significantly changed (Ͼ1.5-fold), with 32 genes up-regulated and 110 genes down-regulated. Unexpectedly, pretreatment with Z-VAD-fmk, which abrogates apoptosis by blocking downstream caspases, resulted in an additional 44 up-regulated and 75 down-regulated genes, giving a total of 261 genes with significantly altered gene expression. These changes were not caused by Z-VAD-fmk itself, as cells treated with Z-VADfmk alone showed no significant alterations in gene expression at any of the time points.
Genes Up-regulated by TGF-␤ 1 Cluster into Discrete Functional Groups-To further investigate the alterations in gene expression induced by TGF-␤ 1 , the 76 up-regulated genes were analyzed using a clustering algorithm that groups genes according to their time-dependent expression profiles. This analysis assumes that co-regulated genes will cluster together and therefore either have similar functions or are involved in the same biochemical pathway/phenomenon. A smooth correlation analysis technique using K-means clustering, as described under "Experimental Procedures," was used to identify co-regulated genes. Three clusters were identified ( Fig. 2 and Table I).
The TGF-␤ 1 -induced changes in gene expression in clusters A (extracellular matrix and cytoskeleton) and B (apoptosis) were largely unaffected by co-treatment with TGF-␤ 1 and Z-VADfmk, whereas cluster C (stress response) was detected (except for FAT10) only in the presence of both Z-VAD-fmk and TGF-␤ 1 .
Cluster A: Extracellular Matrix and Cytoskeleton-This gene cluster contains 25 genes, nine of which were significantly induced by 4 h and remained up-regulated throughout the time course ( Fig. 2A). This cluster includes a number of genes coding for structural proteins such as TNNT2 (troponin T), DSP (desmoplakin), ARHB (ras homolog gene family, member B), SDC4 (proteoglycan), and SMTN (smoothelin), consistent with the fundamental role played by TGF-␤ 1 in regulating cytoskeletal proteins and in extracellular matrix remodeling (41). Significantly, this cluster also contains CTGF (connective tissue growth factor), which has been shown to induce apoptosis and to activate caspase-3 in human aortic smooth muscle cells (42). 5 B. Coyle, manuscript in preparation. 6 Available at www.ncbi.nlm.nih.gov/HomoloGene/. CTGF belongs to a family of immediate-early growth-response genes (43), including NOV, ELM1, COP1, WISP3, and CYR61, which were also found in cluster A.
Cluster B: Amplification of TGF-␤ 1 -induced Apoptosis by Upregulation of Apoptosis-related Genes-This apoptotic cluster contains 26 genes, none of which reached a significant level of expression until 8 h, but thereafter gradually increased with time ( Fig. 2B and Table I). This cluster contains anti-apoptotic genes XIAP and cIAP2 and pro-apoptotic genes CASP8, TP53, BAK1, BAD, and NOTCH4. Up-regulation of these genes was delayed, and they are therefore unlikely to play a role in the initial events that initiate the apoptotic cascade. However, up-regulation of these genes was generally unaffected by Z-VAD-fmk, suggesting that the TGF-␤ 1 -induced increase in these pro-apoptotic genes serves to maintain and amplify the apoptotic process.
Cluster C: TGF-␤ 1 Induces a Stress Response in the Absence of Caspase Activation-This cluster contains 25 genes that, in the main, were up-regulated only in the presence of Z-VADfmk. It contains genes that have been implicated in the activation of NF-B, viz. TWEAK, TAB1 (TGF-␤ 1 -activated kinasebinding protein-1), and TANK (TRAF-associated NF-B activator) (Fig. 2C). Several other stress-related transcription factors were also up-regulated, including ATF3, a member of the CCAAT/enhancer-binding protein family (44), and Jun, the major form of the AP-1 transcription factor. Together, these gene changes suggest that the cells may undergo a stress response that may be important in the caspase-independent cell death induced by TGF-␤ 1 at later time points (28). Interestingly, this cluster also contains FAT10, which, although it was up-regulated in the TGF-␤ 1 -alone treatment, exhibited an even greater increase in the presence of Z-VAD-fmk. FAT10 is a ubiquitin-like protein that forms covalent conjugates and induces apoptosis (45).
During TGF-␤ 1 -induced Apoptosis, Genes Encoding Proteins Involved in Antioxidant Defense Are Down-regulated-Of the 185 significantly down-regulated genes, including a number of oncogenes and cell division genes, a substantial proportion (44 genes) were metabolic enzymes, suggesting that the cells may undergo an adaptive response and switch off nonessential functions. However, the most interesting genes that were observed to be down-regulated are those known to be involved in antioxidant pathways (Fig. 3 and Table II). Importantly, these nine genes include both the first, GLCLC (glutamate-cysteine ligase catalytic subunit (␥-glutamylcysteine synthetase)), and second, glutathione synthetase, enzymes of the glutathione synthesis pathway, and all were down-regulated independently of Z-VAD-fmk. Notably, GLCLC was already down-regulated by 4 h, indicating that this may be a primary transcriptional response. The down-regulation of antioxidant defense mechanisms may facilitate TGF-␤ 1 -induced apoptosis, which is known to involve the generation of ROS (38).
Gene Expression Is Significantly Altered in the Presence of Z-VAD-fmk and TGF-␤ 1 -Although the initial aim of these experiments was to identify upstream mediators of TGF-␤ 1induced apoptosis, the significant changes in gene expression observed in the presence of Z-VAD-fmk pretreatment were very intriguing. Therefore, we analyzed the data further to identify genes whose expression during TGF-␤ 1 -induced apoptosis was significantly changed (p Ͻ 0.05) by Z-VAD-fmk treatment. This analysis showed that 24 genes were up-regulated in the presence of Z-VAD-fmk ( Fig. 4 and Table III); and interestingly, 17 of these genes are members of the stress-response cluster shown in Fig. 2C. These results imply that cellular stress responses are enhanced when the caspase-dependent apoptotic pathway is blocked. The time course of these changes in stressrelated genes shows that they occurred late, and this correlates with the delayed necrosis that we have previously observed in FaO hepatoma cells treated with TGF-␤ 1 (28). These results indicate that TGF-␤ 1 can potentially alter the expression of an additional set of genes, except that these changes in gene Changes in gene expression were quantified by comparing mRNA intensity in control and TGF-␤ 1 -treated FaO hepatoma cells and analyzed by GeneSpring software. Genes were selected as significantly altered if they showed a consistent (i.e. always up or down) change in three different experiments and the mean value showed a -fold change of at least Ϯ1.5 (where ϩ1 represents a 100% increase, i.e. double the control value, and Ϫ1 represents a 50% decrease, i.e. half the control value). The 76 significantly up-regulated genes were analyzed by GeneSpring and sorted into three groups using K-means clustering by smooth correlation as described under "Experimental Procedures." The clusters were then classified on the basis of the function of the majority of genes. A, extracellular matrix (ECM) and cytoskeleton. In addition to the five labeled genes, this cluster contains ERF, ANXA13, KIAA0824, SDC4, CITED1, CYR61, C4A, HREV107, ODC1, BTG2, DSP, CES1, GPX3, ABCB1, CD5, NSF (N-ethylmaleimide-sensitive factor), S100A6, IREB2, fos, and DUSP6. B, apoptosis. Functional details and changes in expression of the proteins encoded by the genes in this cluster are listed in Table I and Supplemental Table I, respectively. C, stress response. In addition to the seven labeled genes, this cluster contains TARBP1, TSC22, CGR19, CSNK1G2, BRF1, DCTN4, COL14A1, TWEAK, SAT, IGFBP5, AES, NR5A2, CAMP, ID3, APPBP1, SFRS8, DDIT3/GADD153, and CASP8AP2/FLASH. expression are suppressed by the action of active caspases. Consequently, inhibition of the caspase cascade by Z-VAD-fmk then allows the appropriate genes to be up-or down-regulated.
Confirmation of Selected Gene Changes by RT-PCR-Significant changes in selected gene expression as detected by the microarrays were validated by RT-PCR. Due to the large number of gene changes observed, we selected relevant genes from each of the functionally significant clusters or groups we had identified by GeneSpring analysis. Thus, from cluster A (extracellular matrix and cytoskeleton), we selected CTGF and ARHB for RT-PCR analysis. At 4 h, both these genes were significantly up-regulated as shown by microarray analysis (Fig. 2A) and clearly confirmed by RT-PCR (Fig. 5A). Similarly, GLCLC and SEPP1 from the antioxidant group of genes were shown to be down-regulated by the microarray analysis (Fig. 3) and confirmed by RT-PCR (Fig. 5A), particularly with GLCLC, which was markedly down-regulated. The pro-apoptotic gene CASP8 and TANK were shown by the microarray analysis to be up-regulated at later times (16 h), particularly in the presence of Z-VAD-fmk; and this was confirmed by RT-PCR (Fig. 5C). However, the RT-PCR results also showed that TGF-␤ 1 alone produced marked up-regulation of these two genes. Up-regulation of FAT10, which was clustered in the stress group (Fig.  2C), was confirmed by RT-PCR and was clearly a response to TGF-␤ 1 treatment, which was markedly enhanced in the presence of Z-VAD-fmk. jun expression as measured by RT-PCR was up-regulated in all the treatments (Fig. 5B), whereas in the microarray results, it was only significantly up-regulated in the TGF-␤ 1 /Z-VAD-fmk-treated cells (Figs. 2C and 4). However, the changes in ATF3 expression as measured by RT-PCR (Fig. 5B) correlated exactly with the microarray results (Figs. 2C and 4). DISCUSSION TGF-␤ 1 initiates many of its biological effects, including apoptosis, by signaling through SMAD proteins and TGF-␤ 1responsive promoter sequences that regulate the expression of a variety of proteins (36,46). In this study, we used DNA microarrays to study the changes in gene expression during TGF-␤ 1 -induced apoptosis. We have shown that there was a time-dependent and coordinated expression of discrete clusters of genes. Interestingly, many genes were down-regulated by TGF-␤ 1 , demonstrating that transcriptional repression of specific genes is involved in the initiation of TGF-␤ 1 -induced apoptotic cell death. The 32 transcripts specifically up-regulated by TGF-␤ 1 clustered into immediate-early responses (upregulated by 4 h) ( Fig. 2A) and delayed responses (up-regulated by 8 h or later) (Fig. 2B). The immediate-early response genes encode many structural proteins, consistent with the fundamental role of TGF-␤ 1 in regulating cytoskeletal proteins and in extracellular matrix remodeling. Only a few specific genes (ϳ12) are known to have SMAD-responsive elements. However, a recent microarray study in dermal fibroblasts has iden-  tified more SMAD-responsive genes (47). These included five related genes encoding for collagen, TIMP-1 (tissue inhibitor of metalloproteinase-1), an irreversible inhibitor of collagenases, ARHB (rhoB), and DSP (desmoplakin I). The latter two genes were also up-regulated in the extracellular matrix and cytoskeleton cluster in FaO cells ( Fig. 2A). This cluster also includes CTGF, which promotes fibroblast proliferation and extracellular matrix formation and has been implicated in liver fibrosis (48) and TGF-␤ 1 -induced apoptosis in MCF-7 and human aortic smooth muscle cells (49,50). The CTGF gene contains a TGF-␤ 1 -responsive element and is strongly up-regulated in MCF-7 cells (which do not normally express CTGF mRNA) during TGF-␤ 1 -induced apoptosis (49). Overexpression of CTGF induces apoptosis, which is abrogated by CTGF antisense oligonucleotides (49). Thus, the up-regulation of CTGF we observed in FaO cells (Fig. 5) could amplify the apoptotic effects of TGF-␤ 1 and/or alternatively induce apoptosis in surrounding cells.
The apoptotic related gene cluster (cluster B) encodes a number of both pro-and anti-apoptotic proteins, including XIAP, cIAP2, CASP8, TP53, BAK1, BAD, and NOTCH4. The upregulation of CASP8 is in agreement with previous studies showing that caspase-8 is processed/activated during TGF-␤ 1induced apoptosis (15,28) and would augment apoptotic cell death. However, the significance of other gene changes is more difficult to assess, as some of the proteins (e.g. XIAP) could act as inhibitors of apoptosis, whereas BAK1 and BAD would be expected to amplify the apoptotic response.
Although the main action of IAP proteins is to prevent apoptosis by inhibiting active caspases (51), XIAP has also been shown to associate with TGFBR-1 and to potentiate TGF-␤ 1induced signaling (52). XIAP binds to TAB1, an activator of TAK1 (TGF-␤ 1 -associated kinase-1), a mitogen-activated protein kinase kinase kinase (53). Interestingly, TAB1 was upregulated in the presence of Z-VAD-fmk ( Fig. 2C and Table III) and is a member of the stress-response gene cluster, which also contains jun, ATF3, and GADD153. The microarray analysis showed that Jun expression was markedly up-regulated at later time points only in TGF-␤ 1 /Z-VAD-fmk-treated cells. In contrast, the RT-PCR data showed that TGF-␤ 1 alone also induced the up-regulation of jun, which agrees with previous studies showing that TGF-␤ 1 induces an immediate up-regulation of AP  Gene expression was analyzed as described in the legend to in Fig. 2, and listed below are genes that are involved in antioxidant defense and that were down-regulated by TGF-␤ 1 (see also Fig. 3) 4. Statistically significant alterations in gene expression in the presence of Z-VAD-fmk. The data were analyzed by K-means clustering for any genes whose expression was statistically different using a non-parametric test (p Ͻ 0.05) when co-treated with Z-VAD-fmk (ZVAD-FMK). This analysis produced a group of 25 genes, 24 of which were upregulated only in the presence of Z-VADfmk. The genes and functions of the proteins they encode are listed in Table III  and Supplemental Table II. including DNA-damaging agents (57), oxidant injury (58), and natural products such as flavonoids and isothiocyanates (59).
ATF3 homodimers bind to the ATF/cAMP-responsive element consensus sites and repress transcription of downstream genes. However, ATF3 also forms a nonfunctional heterodimer with GADD153, thereby up-regulating gene transcription (60). GADD153, a small nuclear protein, dimerizes with other CCAAT/enhancer-binding proteins and is normally expressed at very low levels, but is markedly up-regulated during endoplasmic reticulum stress (61) and apoptosis (62). Parallel upregulation of ATF3 and GADD153 transcripts has also been observed during MG132-induced apoptosis in MCF-7 cells (63) and Jurkat cells. 7 These results suggest that GADD153/ATF3 heterodimers may be formed and that genes normally repressed by ATF3 would be up-regulated, perhaps facilitating the necrotic cell death induced by TGF-␤ 1 after prolonged exposure (28).
The stress-response cluster also contains TAB1 and TANK, which have been implicated in the activation of NF-B. TAK1 and its activator TAB1 have been shown to activate IB kinase, thus stimulating NF-B activation (64). Similarly, TANK can stimulate activation by forming a signaling complex containing TANK, TRAF2, and TBK1, a novel IB kinase-related kinase (65). Together, these data suggest that the cells may undergo an NF-B-related stress response in the presence of Z-VADfmk. Interestingly, 9 of the 185 down-regulated genes encode proteins involved in antioxidant defense, particularly the glutathione redox cycle, which is the major defense system against ROS by normal aerobic mitochondrial metabolism. In the absence of antioxidants, ROS would increase to toxic levels; and significantly, ROS have previously been implicated in TGF-␤ 1induced apoptosis in fetal hepatocytes (29,38,66). After TGF-␤ 1 treatment, there is an early increase in ROS and a decrease in glutathione levels (66), which precede a decrease in 7 B. Coyle, unpublished data.

TABLE III
Genes whose expression is altered by Z-VAD-fmk treatment Gene expression was analyzed as described in the legend to Table I. Of the total number of genes whose expression (except for IGFBP1) was significantly altered, we identified a subset of 24 genes whose expression was significantly altered (p Ͻ 0.05) by Z-VAD-fmk treatment. Gene names IMAGE numbers, level of significance, a summary of functions, and homology of rat to human clones are shown (see Footnote 6). CNP, c-type natriuretic peptide; LPS, lipopolysaccharides; C/EBP, CCAAT/enhancer-binding protein. a Clones whose identity upon sequencing proved to be different from that assigned to their respective (as shown) IMAGE number. The relevant expression ratio for each of the genes shown is listed for each time point in Supplemental Table II. FIG. 5. Verification of selected gene expression changes by RT-PCR. RT-PCR was carried out on treated and control samples as described under "Experimental Procedures." Glyceraldehyde-3phosphate dehydrogenase (GAPDH) expression was used as a loading control. In A, the expression of selected genes is shown at each time point, with and without Z-VAD-fmk. In those experiments in which Z-VAD-fmk was shown to affect the TGF-␤ 1 response (B), additional Z-VAD-fmk-alone samples were also run for comparison. In C, the expression of two apoptotic genes is shown for the 16-h time point. Casp-8, caspase-8.
⌬ and the release of cytochrome c (29). Significantly, the TGF-␤ 1 -inducible transcription factor TIEG1 (TGF-␤ 1 -inducible early response gene-1) also induces apoptosis via an increase in ROS and loss of mitochondrial ⌬ (67). TGF-␤ 1induced oxidative stress and apoptosis can be blocked in part by antioxidant treatment (66,67) and accentuated by inhibitors of glutathione synthesis (29). Glutathione is synthesized in a two-step reaction from glutamate and cysteine to form L-␥glutamylcysteine, which is then conjugated with glycine to produce reduced glutathione (for review, see Ref. 68). Our data show that two key enzymes (GLCLC and glutathione synthetase) involved in glutathione synthesis and CTH (cystathionine ␥-lyase), the terminal enzyme involved in synthesizing cysteine from methionine, are down-regulated at a very early stage in TGF-␤ 1 -mediated cell death ( Fig. 3 and Table II). This would lead to a decrease in the levels of glutathione and an increase in ROS. In support of this conclusion, a recent study has shown that TGF-␤ 1 -induced glutathione depletion in alveolar epithelial cells is due to down-regulation of GLCLC (␥-glutamylcysteine synthetase) (69). This would lead to a decrease in glutathione levels and a subsequent increase in ROS, resulting in cytochrome c release and apoptosis. Consistent with this hypothesis, a recent study has shown that superoxide can directly trigger cytochrome c release and apoptosis in HepG2 cells via interaction with voltage-dependent anion channel (VDAC) independently of pro-apoptotic Bcl-2 proteins (70).
The down-regulation of a battery of enzymes that are involved in protecting the cell against oxidative stress suggests that these genes are coordinately regulated. Cells treated with agents such as tert-butylhydroquinone are protected against oxidative stress via the antioxidant-responsive element. Antioxidant-responsive element-like elements have been found in the promoter regions of the rat NQO1 and human NQO2 (NAD-PH:quinone oxidoreductase) genes, rat glutathione synthetase, and other antioxidant-related enzymes (for review, see Ref. 71;Ref. 72). Interestingly, a recent microarray analysis of tertbutylhydroquinone-treated human neuroblastoma cells has identified 63 genes that are significantly increased, many of which are involved (including GLCLC, GSR, and GSTM3 (brain glutathione S-transferase)) in protecting the cell against oxidative stress (73). In contrast TGF-␤ 1 appears to act in the opposite manner to tert-butylhydroquinone and leads to downregulation of many of these genes, perhaps by the induction of an as yet unidentified antioxidant-responsive element suppressor protein. Significantly, the time-dependent suppression of other antioxidant enzymes (viz. manganese-superoxide dismutase, copper/zinc-superoxide dismutase, and catalase) has also been demonstrated in primary hepatocytes following exposure to TGF-␤ 1 (74).
Another gene that was markedly up-regulated during TGF-␤ 1 and TGF-␤ 1 /Z-VAD-fmk treatment was FAT10, which encodes a ubiquitin-like protein that is synergistically induced by interferon-␥ and tumor necrosis factor-␣ and which has been shown to induce apoptosis in a caspase-dependent manner (45). There is increasing evidence that TGF-␤ 1 signaling is regulated by proteasomal degradation of component members of the pathway. For example, TIEG1 is rapidly induced by TGF-␤ 1 and serves to down-regulate the negative feedback inhibition of the inhibitory protein Smad7, which is also induced by TGF-␤ 1 (75). TIEG1 interacts with SIAH1 (seven in absentia homlogue-1), a ubiquitin-protein isopeptide ligase, and is targeted for destruction by the proteasome (76). Other studies have shown that activated Smad2 is polyubiquitinated and degraded by the proteasome, thereby terminating the signaling pathway (31). Inhibition of proteasomal degradation would be predicted to prolong TGF-␤ 1 signaling, whereas in contrast, we have found that proteasome inhibitors (MG132 and lactacystin) abrogated TGF-␤ 1 -induced apoptosis. 8 This suggests that the involvement of the proteasome in TGF-␤ 1 -induced apoptosis implicates some other as yet identified target and perhaps that FAT10 targets an anti-apoptotic molecule for proteasomal degradation.
The microarray data indicate that TGF-␤ 1 induces apoptosis by specifically down-regulating a set of genes that encode enzymes involved in protecting the cell against ROS. These anti-8 C. Freathy and K. Cain, unpublished data.
FIG. 6. Schematic representation of the possible mechanism of TGF-␤ 1 -induced apoptosis. The scheme shows some of the possible pathways involved in TGF-␤ 1 -induced apoptosis based on the known biochemical events and the possible effects of up-regulation (1) or downregulation (2) of target genes that have been identified in this study. Thus, in this scheme, the activation of the TGF-␤ 1 receptor leads to SMAD-mediated transcriptional changes in key metabolic proteins. The down-regulation of GSH and GLRX and its subsequent effects on ROS, coupled with up-regulation of BAK1 and BAD, induce cytochrome c release, apoptosome formation, and activation of the caspase cascade. The up-regulation of CTGF could also lead to amplification of the apoptotic response by activating the CTGF receptor and inducing apoptosis by this pathway. oxidant enzymes are rapidly turned over, and down-regulation would lead to an increase in ROS, cytochrome c release, and activation of caspases via the apoptosome complex. The later up-regulation of apoptogenic proteins such as caspase-8, BAD, and BAK1 could serve to amplify the apoptotic response. In addition, the early up-regulation of CTGF may trigger apoptosis in neighboring cells, thus providing an additional pathway of amplification (Fig. 6). In conclusion, our study has shown that TGF-␤ 1 -induced apoptosis involves the coordinated induction and repression of specific gene clusters that initiate and propagate the process of cell death.