A Large Scale Genetic Analysis of c-Myc-regulated Gene Expression Patterns* 210

The myc proto-oncogenes encode transcriptional regulators whose inappropriate expression is correlated with a wide array of human malignancies. Up-regulation of Myc enforces growth, antagonizes cell cycle withdrawal and differentiation, and in some situations promotes apoptosis. How these phenotypes are elicited is not well understood, largely because we lack a clear picture of the biologically relevant downstream effectors. We created a new biological system for the optimal profiling of Myc target genes based on a set of isogenic c-myc knockout and conditional cell lines. The ability to modulate Myc activity from essentially null to supraphysiological resulted in a significantly increased and reproducible yield of targets and revealed a large subset of genes that respond optimally to Myc in its physiological range of expression. The total extent of transcriptional changes that can be triggered by Myc is remarkable and involves thousands of genes. Although the majority of these effects are not direct, many of the indirect targets are likely to have important roles in mediating the elicited cellular phenotypes. Myc-activated functions are indicative of a physiological state geared toward the rapid utilization of carbon sources, the biosynthesis of precursors for macromolecular synthesis, and the accumulation of cellular mass. In contrast, the majority of Myc-repressed genes are involved in the interaction and communication of cells with their external environment, and several are known to possess antiproliferative or antimetastatic properties.

The myc proto-oncogenes encode transcriptional regulators whose inappropriate expression is correlated with a wide array of human malignancies. Up-regulation of Myc enforces growth, antagonizes cell cycle withdrawal and differentiation, and in some situations promotes apoptosis. How these phenotypes are elicited is not well understood, largely because we lack a clear picture of the biologically relevant downstream effectors. We created a new biological system for the optimal profiling of Myc target genes based on a set of isogenic c-myc knockout and conditional cell lines. The ability to modulate Myc activity from essentially null to supraphysiological resulted in a significantly increased and reproducible yield of targets and revealed a large subset of genes that respond optimally to Myc in its physiological range of expression. The total extent of transcriptional changes that can be triggered by Myc is remarkable and involves thousands of genes. Although the majority of these effects are not direct, many of the indirect targets are likely to have important roles in mediating the elicited cellular phenotypes. Myc-activated functions are indicative of a physiological state geared toward the rapid utilization of carbon sources, the biosynthesis of precursors for macromolecular synthesis, and the accumulation of cellular mass. In contrast, the majority of Myc-repressed genes are involved in the interaction and communication of cells with their external environment, and several are known to possess antiproliferative or antimetastatic properties.
The Myc protein is a member of the basic region/helix-loophelix/leucine zipper (b/HLH/Zip) family of transcriptional regulators and is capable of exerting both transactivation and transrepression activities (1,2). Transactivation is mediated by binding as an obligate heterodimer with the b/HLH/Zip factor Max to the consensus sequence CA(C/T)GTG (the E box) (3). Transrepression is less well understood (4,5). In either mode Myc is a weak transcriptional regulator, exerting most of its effects within the 2-5-fold range. In a general sense, the upregulation of Myc strongly enforces proliferation and growth, antagonizes cell cycle withdrawal and differentiation, and in some situations promotes apoptosis (6 -8). In agreement, the down-regulation of Myc results in the attenuation of both cell division and cell growth as well as protection against some apoptotic processes (9 -13). Despite extensive research, the specific mechanisms by which these highly evident biological end points are achieved are not well understood. This is largely because a comprehensive list of biologically relevant Myc target genes has not yet been defined.
A wide variety of techniques have been employed in the hunt for Myc targets, ranging from differential expression screens, promoter analysis, and informed guesswork (14 -16) to the modern methods of microarray profiling, serial analysis of gene expression, and chromatin immunoprecipitation (17)(18)(19)(20)(21)(22)(23). This search has been complicated by several factors. First, the weak transcriptional effects of Myc present significant experimental challenges. Second, by all recent indications the total set of Myc targets may be very large. Third, not all E boxes are bound by Myc, and transient transfection studies do not adequately reflect regulation in a chromosomal context. Fourth, comparing tumor cells expressing amplified Myc with nonderegulated counterparts is complicated by the nonisogenic nature of the cells. A widely used approach has been to compare cell lines engineered to overexpress ectopic Myc with parental cells (17,19,21,23). However, it is questionable to what extent this approach can detect genes that respond optimally to physiological changes in Myc expression.
In an attempt to circumvent the latter problem, some time ago we generated c-myc null cells that were derived by gene targeting from an immortalized but otherwise nontransformed rat fibroblast cell line (9). To date, the c-myc Ϫ/Ϫ cells have been used in two limited profiling experiments that examined the expression of 4,400 rat (18) and 6,355 mouse (24) cDNAs and expressed sequence tags in spotted glass slide microarray formats. To create a new biological system for the optimal profiling of Myc target genes, we have reconstituted c-myc Ϫ/Ϫ cells with the conditionally active, tamoxifen-specific c-Myc-estrogen receptor fusion protein (MycER) 1 (25). These new cell lines allow the modulation of Myc activity from essentially null to supraphysiological.
To achieve maximum consistency in expression profiling we sought a simple experimental regimen in which the only changing parameter was the expression of c-Myc and in which a change in c-Myc status elicited and clear and significant change in phenotype. We chose to use randomly cycling, exponential phase cultures, and we developed conditions such that cells experienced a constant environment and were in a balanced, steady state of growth for significant periods of time. Under these conditions c-myc null cells displayed a pronounced phenotype, a 2-3-fold reduction in macromolecular synthesis accompanied by a commensurate slowing of the cell cycle (9). Most importantly, we showed that under these conditions both c-myc ϩ/ϩ and c-myc Ϫ/Ϫ cultures cycled uniformly, namely, that there were no cohorts of differentially cycling or noncycling cells within a given culture (26).
Expression profiling using a total of 81 Affymetrix GeneChip arrays was performed in three experiments (Fig. 1). First, we compared c-myc ϩ/ϩ (TGR), c-myc Ϫ/Ϫ (HO), and c-myc Ϫ/Ϫ cells reconstituted with a constitutive c-myc transgene (HOmyc3). This revealed the total number of genes that respond to a sustained loss of c-Myc under exponential growth conditions. Second, c-myc Ϫ/Ϫ cells reconstituted with the conditional c-mycER transgene (HOmycER) were stimulated with 4-hydroxytamoxifen (OHT), and data were collected during a 16-h time course. This revealed the kinetics of the responses to Myc activation. Finally, the time course of induction with OHT was performed in the presence of cycloheximide, revealing a subset of direct transcriptional targets of c-Myc. All experiments, including the growth of cells and preparation of RNA, were performed on three separate occasions (independent biological replicates), and all data were subjected to a statistical analysis of significance.

Cell Lines and Culture Conditions
TGR-1 is a hprtsubclone of the Rat-1 cell line (27). HO15.19 (referred to as HO) is a homozygous c-myc null derivative of TGR-1 generated by gene targeting (9). HOmyc3 was derived from HO15.19 by constitutively expressing murine c-myc cDNA using a retroviral vector (10). HOmyc3 cells express c-Myc protein at three to four times the level seen in TGR cells. HOMycER12 and HOMycER104 were derived in the same fashion to express MycER (25). MycER is a hybrid protein consisting of the entire c-Myc polypeptide at its N terminus and the ligand (estrogen) binding domain of the human estrogen receptor at the C terminus. In the MycER construct used here the estrogen binding domain has been mutated to be specific for the agonist OHT. Retroviral vectors were packaged in BOSC cells (28), and supernatants were used to infect HO15.19 cells. Colonies were selected with 120 g/ml hygromycin (Calbiochem), ring cloned, and expanded into clonal cell lines. The mRNA encoding the MycER protein is thus expressed constitutively from the viral long terminal repeat promoter, and the activity of this promoter is not affected by OHT. OHT is instead believed to elicit a conformational change in MycER which allows the protein to become biologically active. All cultures were grown in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 10% calf serum (Hyclone) at 37°C in an atmosphere of 5% CO 2 , except BOSC cells, which were supplemented with 10% fetal bovine serum (Hyclone). Great care was taken that cultures were cycling asynchronously and were in rapid and exponential phase of growth (26). Briefly, cells were always split at subconfluent densities (Ͻ50%) and at relatively low dilution (1:10 for c-myc ϩ/ϩ and 1:4 for c-myc Ϫ/Ϫ cells). Cultures can thus be maintained continuously at densities of 10 -50% confluence (to avoid any contact inhibition), and the relatively frequent passaging (every 3-4 days) and medium changes maintain a rapid growth rate. This regimen was followed for a minimum of two passages before cells were harvested for other experiments. MycER was activated with 200 nM OHT (Sigma). Dose-response studies showed that 200 nM OHT is saturating for the activation of MycER. OHT was dissolved in absolute ethanol at 1 mM and stored at Ϫ80°C. Mock-treated cultures received vehicle (ethanol) at a final concentration of 0.02%. Protein synthesis was inhibited with 20 g/ml cycloheximide (Sigma) which was added 30 min before the addition of OHT. BrdUrd labeling and flow cytometry were performed as described previously (9), except that the Vectastain Elite ABC and Novared kits (Vector Laboratories) were used for histochemical staining.

Molecular Biology Procedures
The c-myc, full-length MycER (25), and a deletion mutant of MycER missing amino acids 106 -143 (29) cDNAs were cloned into the HpaI site of the pLXSH retroviral vector (30) using standard procedures (31). Total RNA for Northern hybridization and microarray analysis was isolated using TriZol reagent (Invitrogen). Total RNA for quantitative real time PCR (qPCR) was isolated using the RNaqueous-4PCR kit (Ambion). Northern hybridization was performed using the formaldehyde gel method, and 32 P-labeled probes were synthesized using the random oligonucleotide labeling method from gel-purified restriction or PCR fragment templates as described previously (32). qPCR was performed using the Applied Biosystems Prism 7700 Sequence Detector and software. Primers were designed using Primer Express software (Applied Biosystems) for amplification of 100-bp fragments. cDNA was generated using the TaqMan reverse transcription kit and amplified using the SYBR green PCR and reverse transcription PCR kits (Applied Biosystems). Amplification efficiencies were determined by serial dilution of template cDNA for each gene. All samples were run in triplicate. Glyceraldehyde phosphate dehydrogenase (GAPDH) was used as the internal standard. GAPDH was used because microarray profiling showed that the signals for six distinct GAPDH probe sets were equivalent between TGR, HO, and HOmyc3 cell lines under our exponential phase culture conditions. Protein samples were prepared by lysing whole cells in radioimmune precipitation assay buffer (33) supplemented with protease inhibitors. Immunoblotting was performed as described previously (10,34). The following antibodies were used: c-Myc (Upstate Biotechnology, cat. 06-340), neomycin phosphotransferase II, (5 Prime -3 Prime, Inc., cat. 7-511721), and actin (Sigma, cat. A5316). Horseradish peroxidase-conjugated secondary antibodies were from Jackson Immunoresearch. Signals were visualized using the ECL reagent (Amersham Biosciences).

Microarray Analysis (Fig. 1 Experiments)
Target cRNA was prepared and hybridized (45°C, 16 h) to GeneChip rat U34 arrays according to the manufacturer's directions (Affymetrix). Hybridized arrays were washed and stained using the GeneChip fluidics station 400 and scanned using the Agilent GeneArray scanner. Signals were analyzed using Microarray Suite 5.0 software (Affymetrix). Data were normalized using a set target intensity of 1,500, published to a data base using MicroDB 3.0, and analyzed in Data Mining Tool 3.0 software (Affymetrix). Analysis of each cell line and/or condition was based on three biological replicates (RNAs prepared from independent experiments performed at different times). The replicates were used to calculate the means and standard deviations for the expression values of all probe sets for each cell line and/or condition. Probe sets were considered present if they received a present call in two of the three biological replicates. Pairwise comparisons between cell lines and/or conditions were made using Student's t test (p Ͻ 0.05).

Analysis of Experiment 1
Myc-influenced expression patterns were assigned to four categories: 1) activated by Myc; 2) repressed by Myc; 3) activated by overexpression of Myc; and 4) repressed by overexpression of Myc (for examples, see Fig. 4). Probe sets were categorized based on the following criteria.
Category 1: Activated by Myc-The ratio of the average signal intensity between HOmyc3 and HO and/or between TGR and HO was Ն2.0. Category 2: Repressed by Myc-The ratio of the average signal intensity between HOmyc3 and HO and/or between TGR and HO was Յ0.50. Category 3: Activated by Myc Overexpression-HO/Myc3 average intensity values were significantly greater than those of TGR and those of HO (p Ͻ 0.05), TGR and HO average intensity values were not statistically different (p Ͻ 0.05), and the ratio of average intensity values between TGR and HO was less than 1.4. 14 probe sets were moved from category 1 to category 3 based on visual inspection (in these cases, the ratio of the average intensity values between HO and HOmyc3 was greater than or equal to twice the ratio of average intensity values between HO and TGR. Category 4: Repressed by Myc Overexpression-The HOMyc3 average intensity values were significantly less than those of TGR and those of HO (p Ͻ 0.05), TGR and HO average intensity values were not statistically different (p Ͻ 0.05), and the ratio of average intensity values between TGR and HO was greater than 0.7. Nine probe sets were moved from category 2 to category 4 based on visual inspection (in these cases, the ratio of average intensity values between HO and HOmyc3 was less than or equal to half the ratio of average intensity values between HO and TGR).

Analysis of Experiment 2
Probe sets were considered responsive to OHT if, at any given time point, they displayed statistically significant (p Ͻ 0.05) differences between OHT and vehicle-treated replicates, and the fold change between the means was Ն1.5. Of the 535 probe sets on the U34A chip which were identified in the initial comparison of TGR, HO and HOmyc3 cell lines (experiment 1), 142 probe sets were OHT-responsive by the above criteria. An additional 460 probe sets satisfied the OHT inducibility test (experiment 2, Fig. 1) but failed the 2-fold induction limit set in the comparison of TGR, HO, and HOmyc3 cells (experiment 1). 76 probe sets were recovered from this list and designated as c-Myc targets if the inducibility in the TGR, HO, and HOmyc3 comparison (experiment 1) was Ն1.5. The resultant 218 (142 ϩ 76) MycER-responsive probe sets out of the total 611 (535 ϩ 76) probe sets represent 180 nonredundant genes. 75 of the 180 genes (41%) identified in HO/my-cER12 cells satisfied the same statistical criteria in HOmycER104 cells. Of the remaining 105 genes, 49 (27%) were already deregulated by the high basal Myc activity in HOmycER104 cells, 36 (20%) behaved qualitatively similarly in HOmycER104 cells but failed the t test, and 20 (11%) failed the fold change test or behaved anomalously. Because the MycER protein is capable of eliciting low Myc activity even in the absence of OHT, we also asked whether this "leakiness" could mask potential responses to Myc activation if, for example, probe sets were already maximally induced/repressed before the addition of OHT. Probe sets were considered leaky if the average intensity values in HOmy-cER12 cells were statistically different (p Ͻ 0.05) and greater (for Myc-activated genes) or smaller (for Myc-repressed genes) than the average intensity values in HO cells. Probe sets that were leaky, nonresponsive to OHT, unable to respond to elevated levels of Myc (nonresponsive to OHT in HOmycER104 cells and/or not overexpressed in HOmyc3 versus TGR cells), and expressed above a threshold intensity value of 500 in TGR cells comprised less than 10% of the probe sets identified in experiment 2.

Analysis of Experiment 3
Probe sets were considered to be direct Myc targets if differences in expression between samples treated with cycloheximide plus OHT and those treated with cycloheximide alone at any time point were statistically significant (p Ͻ 0.05) and had a magnitude of Ն1.5. Probe sets were classified as indirect targets of Myc if differences in expression between samples treated with cycloheximide alone were not statistically different (p Ͻ 0.05) from the untreated control and if differences in expression at any time point between samples treated with cycloheximide plus OHT and cycloheximide alone were not statistically significant (p Ͻ 0.05).
To assess the effect of OHT alone on RNA expression (in the absence of the MycER transgene) TGR and HO cells were treated with OHT for 16 h (or vehicle for the same time period), and RNA was extracted and subjected to microarray analysis. 288 of the total 8,799 probe sets on the U94A chip were affected by OHT by a factor of Ն2 in either TGR or HO cells. 2 of the OHT-affected probe sets are on the list of 180 genes reported in Table I. However, these probe sets were affected by OHT only in HO cells and not in TGR cells. The genes affected by OHT alone in HO cells are ␣-mannosidase II (M24353) and cytosolic Na/K-transporting ATPase, B subunit (AA859920). The expression of these genes was clearly affected in a comparison of TGR, HO, and HOmyc3 cells in the absence of OHT; however, because part of their response in HOmycER cells may be the result the effect of OHT alone, further examination may reveal them to be indirect Myc targets.

Complementation of c-myc Ϫ/Ϫ Cells with a Conditional c-mycER Transgene-
The c-mycER cDNA (25) was introduced into c-myc Ϫ/Ϫ cells using the retrovirus vector LXSH. Selection with hygromycin in the absence of OHT resulted in an approximately equal mixture of slowly and rapidly growing colonies. Cells in the slow growing colonies displayed a flattened and spread-out morphology typical of c-myc Ϫ/Ϫ cells, whereas cells in the rapidly growing colonies had the fibroblastic morphology characteristic of the parental c-myc ϩ/ϩ cells. Expression of a c-mycER transgene containing an internal deletion of Myc sequences (MycER⌬106 -143) did not generate any fast growing colonies. The fast growing colonies are thus likely to be the result of low levels of Myc activity elicited by the MycER transgene even in the absence of OHT. Colonies of both types were picked and expanded into cell lines. Two representative clones were chosen for further study: HOmycER12, derived from a slow growing colony, and HOmycER104, derived from a rapidly growing colony.
As expected, HOmycER104 cells expressed higher levels of MycER protein than HOmycER12 cells ( Fig. 2A). Several assays were performed to examine the functionality of the c-mycER transgene. Because c-Myc is known to repress its own promoter, we examined the expression of the Neo mRNA, which is encoded by one of the c-myc gene targeting vectors and was placed under control of the endogenous c-myc promoter by the homologous recombination event. Treatment of both HOmycER12 and HOmycER104 cells with OHT clearly reduced the expression of the Neo mRNA (Fig. 2B). In contrast, repression of Neo by OHT was not observed in the HO or HOmycER⌬106 -143 cell lines (Fig. 2B, legend). Because expression of Myc accelerates the cell cycle and thus increases the fraction of cells in S phase (9), we measured S phase content of HOmycER12 and HOmycER104 cultures using BrdUrd labeling (Fig. 3). The percentage of cells in S phase was increased from 32 to 48% in HOmycER12 cells and from 42 to 57% in HOmycER104 cells after a 24-h treatment with OHT.
We next examined the expression of several known c-Myc target genes in HOmycER12 cells in the absence and presence of OHT (Fig. 2C). Activation of MycER resulted in the upregulation of CAD mRNA and down-regulation of gadd45 mRNA, whereas the expression of ornithine decarboxylase mRNA was essentially unchanged or only weakly activated. The same behavior has been documented for these genes by comparing TGR and HO cells under similar exponential growth conditions (35). In control experiments, OHT had no effect on the expression of CAD, gadd45, or ornithine decarboxylase mRNAs in either TGR or HO cells (Fig. 2C, legend). We there-fore conclude that HOmycER12 cells in the absence of OHT resemble c-myc Ϫ/Ϫ (HO) cells and that addition of OHT elicits a c-myc ϩ/ϩ (TGR)-like phenotype.
In preparation for experiments using OHT induction in the presence of cycloheximide we examined the effect of cycloheximide on the expression of the MycER protein (Fig. 2D). Expression of the MycER protein declined rapidly and was barely detectable at 8 h after cycloheximide and OHT addition. Because c-Myc protein is known to be unstable, with a half-life estimated in the range of 30 -40 min (36), the turnover of the MycER protein after the addition of cycloheximide was not unexpected. This observation unfortunately limits the utility of cycloheximide as a tool to investigate the direct action of Myc to targets that respond within a relatively short time frame.
Genes Differentially Expressed in c-myc Ϫ/Ϫ Cells-Total RNA was extracted from exponentially cycling cultures of TGR, HO, and HOmyc3 cells, and expression profiling was performed using Affymetrix U34 rat GeneChips. Each cell line was grown on three separate occasions, and each of the corresponding RNAs (total of nine RNA samples comprising three biological replicates) was hybridized to the three available Affymetrix rat GeneChips (U34A, U34B, U34C; experiment 1 in Fig. 1). 5,732 probe sets displayed statistically significant differences (p Ͻ 0.05) between TGR and HO cells and/or between HO and HOmyc3 cells. Adopting an expression differential cutoff of Ն2-fold between the means of TGR and HO and/or HO and HOmyc3 reduces the number of probe sets to 1,527. These probe sets were then grouped into four categories according to Where indicated, OHT was included in the culture medium for 16 h before harvest. The MycER⌬ protein contains an internal deletion (amino acids 106 -143) and thus migrates faster than MycER. Actin was used as the loading control. X-band, a cross-reacting protein of unknown identity. B, repression of knock-in Neo mRNA by MycER. HOmycER12 and HOmycER104 cells were harvested in the exponential phase of growth and analyzed by qPCR. Where indicated, OHT was included in the culture medium for 16 h before harvest. GAPDH was used to normalize for equal input of RNA. Repression of Neo mRNA by the 16-h OHT treatment was 1.7 Ϯ 0.22-fold in the HOmycER12 cell line and 3.8 Ϯ 1.1-fold in the HOmycER104 cell line. To demonstrate that the effect of OHT on Neo mRNA expression is dependent on MycER, HO, and HOmycER⌬106 -143 cells were treated with OHT for 16 h (or propagated in parallel without OHT), and RNA was harvested and analyzed by qPCR. All data were normalized to GAPDH and are expressed relative to the no OHT condition, which was set to a value of 1.0 for each cell line. The values in the presence of OHT were: HO, 0.97 Ϯ 0.17; HOmycER⌬106 -143, 0.86 Ϯ 0.11. C, effect of MycER on expression of gadd45, CAD, and ornithine decarboxylase (ODC) mRNAs. HOMycER12 cells were harvested in the exponential phase of growth and analyzed by Northern hybridization. Where indicated, OHT was included in the culture medium for 48 h before harvest. GAPDH was used as the loading control. To demonstrate that the effect of OHT on gadd45, CAD, and ornithine decarboxylase mRNA expression is dependent on MycER, TGR and HO cells were treated with OHT for 48 h (or propagated in parallel without OHT), and RNA was harvested and analyzed by qPCR. All data were normalized to GAPDH and are expressed relative to the no OHT condition, which was set to a value of 1.0 for each gene and cell line. The values in the presence of OHT were as follows: CAD, TGR 0.95 Ϯ 0.20 and HO 1.14 Ϯ 0.15; gadd45, TGR 1.07 Ϯ 0.13 and HO 1.11 Ϯ 0.12; ornithine decarboxylase, TGR 0.93 Ϯ 0.13 and HO 1.45 Ϯ 0.11. D, expression of MycER protein in the presence of cycloheximide (CHX). HOMycER12 cells in exponential phase of growth were treated with OHT and cycloheximide. Note that cycloheximide was added 30 min before OHT. Samples were harvested at the indicated times and analyzed by immunoblotting. Actin was used as the loading control.  Fig. 4. The remaining 52 probe sets (3%) exhibited patterns of expression whose biological relevance to c-myc is not clear. To ascertain the accuracy of the microarray analysis, we examined the mRNA expression levels of 7 Myc-activated, 6 Myc-repressed, and 4 unaffected genes using qPCR. In 17 of 17 cases the qPCR data confirmed the microarray results. Because the U34A GeneChip contains most of the known rat genes it was used in subsequent experiments.
Kinetic Analysis of c-Myc Target Gene Regulation-Exponentially cycling cultures of HOmycER12 cells were treated with either OHT or vehicle (ethanol), and samples were collected 2, 4, 8, and 16 h after treatment (experiment 2 in Fig. 1). The zero time point reference sample was harvested at the time of drug addition, resulting in 9 total RNA samples. The time course experiment was performed on three separate occasions to obtain three biological replicates for a total of 27 RNA samples. Of the 611 probe sets differentially expressed on the U34A chip in the TGR, HO, and HOmyc3 comparison (experiment 1 in Fig.  1), 218 were responsive to OHT in experiment 2. Because of some redundancy present on the chips, the 218 probe sets responsive to MycER correspond to 180 unique genes or expressed sequence tag clusters. They were categorized further according to their kinetics of induction as early, mid, or late responding if the change in expression was first evident at 2-4 h, 8 h, or 16 h after the addition of OHT, respectively. Finally, within these categories genes were grouped according to general function (Table I). Representative induction profiles are shown in Fig. 5. The HOmycER104 cell line was also profiled in a time course of OHT induction with samples collected 0, 8, and 16 h after treatment. There was a high degree of concordance between the HOmycER12 and HOmycER104 data sets, thus providing additional verification (Table I).
Direct Targets of c-Myc-Next, we sought to determine which of the 180 unique genes and expressed sequence tags that we identified as MycER-responsive may be direct targets. Exponentially cycling cultures of HOmycER12 cells were treated with either OHT plus cycloheximide or cycloheximide alone, and samples were collected 4, 8, and 16 h after treatment (experiment 3 in Fig. 1). The zero time point reference sample did not contain either drug. As previously, the time course experiment was performed in three biological replicates. 21 of 180 OHT-responsive genes were designated as direct targets. Interestingly, all 21 were in the Myc-activated category. In addition, we identified 24 activated genes and 16 repressed genes that appeared to be bona fide indirect targets. It is well documented that cycloheximide alone can strongly influence gene expression. These effects have the potential of significantly masking the influence of Myc on the expression of bona fide target genes and underscore the importance of doing cycloheximide only controls at all time points. Indeed, all of the genes that failed the criteria of a direct or indirect target showed significant induction or repression resulting from cycloheximide alone. In these cases we do not believe that a clear distinction between a direct and an indirect target can be made. Representative plots of time courses in various categories are shown in Fig. 5, and the data for all 180 OHT-responsive genes are summarized in Table I. DISCUSSION The three Affymetrix U34 GeneChips provide the most extensive coverage of the rat genome available (26,261 probe sets, 20,691 unique genes and expressed sequence tag clusters). The data set presented here is thus the most comprehensive analysis of Myc-influenced gene expression profiles to date. In total, we identified 1,527 probe sets differentially expressed by 2-fold or more. Given that 43% of all U34 probe sets were expressed in Rat-1 fibroblasts, ϳ14% of the active transcriptome is responsive to Myc within the 2-fold differential expression cutoff. Because the U34 chips cover approximately half of the rat genome, ϳ3,000 probe sets (ϳ2,400 genes) can be estimated to be Myc-responsive in this cell type. However, if the differential expression criterion is relaxed to statistical significance only, the Myc-responsive transcriptome becomes greater than 50% of all active genes.
The significantly increased yield of Myc-responsive genes achieved in this study is the result primarily of our ability to modulate Myc expression from almost zero to supraphysiological. This is clearly evidenced by the fact that a previous report (17), which relied on ectopic MycER expression in a normal (c-myc ϩ/ϩ ) cell background, identified only 36 MycER-responsive targets from a total of 6,416 genes surveyed. Both studies used Affymetrix technology and very similar criteria for data analysis. It is thus clear that although a subset of Myc targets can respond to elevated Myc levels, the great majority of responses occur in the range of physiological expression.
Given the widespread effects of Myc on gene expression, it is noteworthy that only 36% of differentially expressed probe sets responded acutely to Myc activation in the HOMycER12 cell line. 22% of the MycER-responsive genes have been identified in previous studies. One limitation of the MycER activation regimen is the response time of repressed genes because any observable effects depend on the turnover of the preexisting mRNA. However, Myc-repressed genes comprised 45% of the OHT-responsive set and 51% of the total pool of differentially expressed probe sets, indicating that the extent to which Mycrepressed genes are being underestimated in the 16-h OHT time course is likely to be minor.
The most likely explanation for the large fraction of MycERnonresponsive genes is that they represent longer term adaptive responses to the loss of c-Myc function. Although the existence of direct targets with delayed responses cannot be ruled out, the fact that many genes respond rapidly argues that factors other than Myc are likely to account for the slow kinetics. Indeed, the list of OHT-responsive genes includes many     indirect targets, demonstrating that even indirect effects can be rapid enough to score in the 16-h time course. The morphological changes that accompany the addition of OHT to HOmycER12 cells are slow, taking effect between 24 and 48 h. It is not unreasonable that the extensive adjustments of cellular physiology which take place in response to loss and/or gain of Myc activity would extend over one or even several cell cycles. It is clear that the great majority of these changes are reversible because 95% of probe sets (on all three chips) that are differentially expressed in c-myc Ϫ/Ϫ (HO) cells relative to c-myc ϩ/ϩ (TGR) cells are reverted to a c-myc ϩ/ϩ pattern of expression in the HOmyc3 cell line.
Although the activation of MycER with OHT in the presence of cycloheximide has been used frequently to differentiate direct and indirect action of Myc, our data indicate that this method has limited resolution. The major problems are the short life span of the MycER protein and the presence of significant changes in gene expression caused by cycloheximide alone. Of the 180 OHT-responsive genes, 119 (66%) were significantly affected by cycloheximide alone. Of the remaining genes, 21 were activated and direct, 24 were activated and indirect, and 16 were repressed and indirect. No directly repressed genes were apparent. One possible explanation is that the resolution of the cycloheximide methodology becomes critically limiting when the short half-life of MycER and the effects of cycloheximide are superimposed on the slow response of many repressed genes. Another possibility is that repression by Myc depends on interaction with other proteins, such as Miz-1, and that this interaction or the activity of the interacting proteins is masked by cycloheximide artifacts.
Among Myc-activated genes, the relative proportion of direct and indirect targets was approximately equal (21 and 24%, respectively). Because there are no obvious reasons why this relationship should not also hold for genes subject to cycloheximide effects, by extrapolation we can expect ϳ47 directly activated targets among the 180 OHT-responsive genes. Because 38% of Myc-responsive probe sets were found on the U34A chip, and the set of 3 chips covers approximately half of the rat genome, we can expect ϳ247 directly activated Myc targets in a rat fibroblast under exponential growth conditions. Although our simple and easily controlled experimental design greatly facilitates expression profiling, there are several reasons why it may be underestimating the total spectrum of Myc-regulated genes. First, a gene may not be affected equally by Myc under all growth conditions. For example, the induction of Myc after serum stimulation of quiescent cells could contribute significantly to the regulation of genes that may respond only weakly under balanced, steady-state growth conditions. Second, some genes may only be able to respond to Myc during a specific segment of the cell cycle. Third, cell line-or cell type-specific effects are also likely to be encountered. Fourth, some genes are detected poorly or not at all by the current U34 probe sets (e.g. p15 (Ink4b), p27 (Kip1), and Cdk7).
MycER-responsive genes identified in our profiling screen have diverse functions (Table I). The largest single category on the Myc-activated list (22 of 101 genes) are enzymes involved mostly in carbon assimilation, anabolic pathways, and energy metabolism. Only 6 of these have been reported previously (17,21,37,38). What is striking is the preponderance of enzymes that catalyze the first committed and regulated steps of major pathways, such as glycolysis, biosynthesis of purines, pyrimidines, polyamines, fatty acids, phospholipids, S-adenosylmethionine (a key molecule in one carbon transfer reactions), creatine (important, as creatine phosphate, in short term energy storage), and NADPH (needed in most reductive anabolic reactions). In addition to being rate-limiting and regulated allosterically, many of the genes are regulated transcriptionally, and their activation is correlated with rapid growth and proliferation.
Also prominent on the Myc-activated list are functions that positively impact protein synthesis, including proteins involved in the synthesis and processing of rRNA, the biogenesis of ribosomes, and translation initiation and elongation factors. 17 genes fall in this category, including RNA polymerase I, 8 of which have been reported previously. The expression of 13 ribosomal proteins was decreased in c-myc Ϫ/Ϫ (HO) cells, but only one was OHT-responsive in MycER cells, indicating that they are likely to be indirect targets. However, because many were weakly inducible at late times in MycER104 cells, it is possible that ribosomal protein genes can also respond to very high Myc levels, such as those found in tumor cells (20).
The appearance of several protein folding functions on the Myc-activated list, both cytoplasmic and mitochondrial, is also consistent with an increased capacity for protein biosynthesis. Notable among these are the mitochondrial chaperones and chaperonins prohibitin, BAP-37, Hsp60, Hsp10, and GrpE, all of which have been identified previously as possible Myc targets. Interestingly, chaperones have been shown to have important roles in the control of apoptosis. The GroEL/GroES homologs Hsp60/Hsp10 have proapoptotic effects involving caspase-3 activation (39,40). In contrast, Hsp27 and ␣B crystallin, identified as Myc-repressed in our analysis, have been shown to function as negative effectors of apoptosis through their ability to sequester cytochrome c from Apaf-1 (41) and inhibit the maturation of caspase-3 (42), respectively. We also identified a component of the mitochondrial permeability transition pore complex as a Myc-repressed gene.
In a general sense, the majority of Myc-activated metabolic functions are indicative of a physiological state geared toward the rapid utilization of major carbon sources and the biosynthesis of precursors for the synthesis of DNA, RNA, proteins, and lipids. Combined with the up-regulation of the machinery for protein synthesis and folding these changes would promote the accumulation of cellular mass, which is required to support ongoing cellular proliferation. Myc also impacts the expression of key G 1 phase cell cycle regulators, raising the question as to which functions, metabolic or cell cycle, are the primary effectors. The preponderance of genes that promote metabolism and cell growth, as well as the documentation of increasing numbers of bona fide direct Myc targets in this category, makes it very unlikely that all these effects are secondary. Indeed, preliminary evidence indicates that both metabolic and cell cycle functions may be equally important: overexpression of an enzyme involved in one carbon metabolism (serine hydroxymethyl transferase) or a cell cycle regulator (Cdk4) both partially rescued the slow growth phenotype of c-myc Ϫ/Ϫ cells (43,44).
The Myc-repressed genes stand in stark contrast to the Myc-activated targets: metabolic and protein synthesis functions are absent, and the list is dominated by genes involved in cell adhesion, cell-cell contact, extracellular matrix synthesis and modification, and vesicular trafficking. In particular, the latter category has not been identified previously as Myc-responsive. This list includes genes involved in vesicular transport of secreted proteins, such as the calcium-binding protein P22, ␣-fodrin, the secretory carrier membrane protein-1, the small G protein ARF2, and phospholipase D. In addition, proteins involved in vesicle docking and fusion, including cellubrevin, Rab10, the Rab effector GM130, and the Rab GDP-dissociation inhibitor, were found to be repressed by Myc. In a general sense, a significant fraction of Myc-repressed genes are involved in the interaction and communication of cells with their external environment. It is especially interesting to note that several of these targets have been shown to possess tumor suppressor and antimetastatic properties. By repressing genes involved in vesicular trafficking and cellular adhesion inappropriate Myc expression may thus create a permissive environment for aggressive tumor cell invasion.
Although much work will be needed to sort out direct and indirect targets and to integrate fully the functions of the activated and repressed genes, this study has significantly expanded our appreciation of the impact of Myc on cellular physiology and has revealed a number of intriguing novel candidates for drug targets. The total extent of transcriptional changes that can be triggered in response to Myc activity is remarkable, and it should be noted that many of the indirect targets are likely to have important roles in mediating the elicited cellular phenotypes.