Transcriptional Regulation of Mitotic Genes by Camptothecin-induced DNA Damage: Microarray Analysis of Dose- and Time-dependent Effects

CPT³ is a DNA topoisomerase I inhibitor that blocks the DNA religation of topoisomerase I cleavage complexes. During DNA replication, the collision between topoisomerase I cleavage complexes and DNA replication forks generates double-strand breaks (1). Therefore, CPT is most potent during DNA replication (2). DNA damage caused by CPT arrests cells in G₂ phase and can trigger rapid apoptosis in certain cell types (3). CPT derivatives are used clinically to treat many types of solid tumors, including colorectal and ovarian carcinomas, but not all cancer cells are equally sensitive to CPT. Although decreased topoisomerase I expression levels and mutations within topoisomerase I can play important roles in CPT resistance (4), evidence indicates that molecular responses to the DNA damage can also contribute to differential CPT sensitivity (5). These responses, including ones related to cell cycle control, DNA repair, and the apoptotic process, can be both cell-type specific and damage specific. It has been shown that the expression levels of p21/WAF1, cyclin B1, bcl-2, and bax change after CPT treatment (6, 7). Transcriptional changes after DNA damage can be either primary responses to the damage or secondary to cellular processes responding to the damage. Identification of changes in gene expression levels can provide a useful link between DNA damage and drug cytotoxicity (8, 9). cDNA microarray technology (10, 11, 12) makes it possible to measure changes in the relative expression levels of thousands of genes and, therefore, to study multiple cellular processes simultaneously.

In this study, using pin-spotted cDNA microarrays, we measured relative changes in expression of 1694 genes after brief CPT treatment of HCT116 cells. HCT116 is a MSH-2-deficient, nonapoptotic colon cancer-derived cell line that expresses functional p53 protein (13). HCT116 cells arrest in G₂-M phase after DNA damage, but unlike leukemia cell lines, they do not undergo rapid apoptosis (14). This property was used to separate gene expression changes directly related to DNA damage from those associated with apoptosis. To obtain more homogeneous response to CPT treatment and distinguish changes in gene expression related to cell cycle from changes related to DNA damage, we carried out experiments using HCT116 cells synchronized in S-phase. The synchronization allowed us to discern cell cycle-dependent down-regulation and modest up-regulations that would have been buried in an unsynchronized (mixed) population of cells. By performing time- and dose-dependent experiments, we found that up-regulation of p53-related stress response genes was associated with high-dose CPT treatment or prolonged exposure to APH. In contrast, protracted expression of mitosis-related genes was correlated with cell cycle delay in G₂. These results support the twin propositions that DNA damage can disrupt normal cell cycle-related gene expression and that this disruption may be involved in G₂ cell cycle arrest. High and low concentrations of CPT produced qualitatively different changes in patterns of gene expression.

APH and CPT were purchased from Sigma Chemical Co.-Aldrich and dissolved in DMSO to concentrations of 3 and 10 mm, respectively. Cy3-dUTP, Cy5-dUTP, deoxynucleotide triphosphates, Oligo(dT)_12–18, and Oligo(dA)_40–60 were purchased from Amersham Pharmacia Biotech (Piscataway, NJ). Superscript II reverse transcriptase and hCOT-1 were purchased from Life Technologies, Inc. (Gaithersburg, MD). Seed cultures of human colon cancer cell line HCT116 were obtained from the Developmental Therapeutics Program, National Cancer Institute, and the cells were cultured in RPMI 1640 (Life Technologies, Inc.) containing 10% fetal bovine serum (BioAtlantic, GA) and 2 mm l-glutamine (Life Technologies, Inc.). The cells were grown to passage 19 and used in the experiments.

HCT116 cells, cultured to near confluence, were trypsinized and plated at 3.6 × 10⁶ cells/150-mm² dish (Becton Dickinson, Franklin Lakes, NJ). After a 24-h incubation, the cells were synchronized in medium containing 1 μm APH for 17 h or for the time indicated. CPT treatment was carried out as follows on synchronized cells after 17 h with APH. The APH-containing medium was removed, and the cells were washed twice with APH-free medium, and then cultured in drug-free medium for 2 h. Portions of the cells were then treated for 75 min with 0, 20, or 1000 nm CPT. In each case the medium contained 0.01% DMSO. The CPT/DMSO-containing medium was then removed, and the cells were washed twice with drug-free medium. Thereafter, cells were cultured in drug-free medium for the indicated time.

Cells were trypsinized, and the resulting cell suspensions were centrifuged at 1000 rpm for 10 min. The cells were fixed by resuspending them in 0.5 ml of 70% ethanol for 30 min, centrifuging them at 1500 rpm for 10 min, and washing the pellets twice with ice-cold PBS to remove residual ethanol. The cell pellets were resuspended in 0.5 ml of PBS containing 50 μg/ml propidium iodide (Sigma Chemical Co.-Aldrich, St. Louis, MO) and 100 μg/ml RNase (Sigma-Aldrich), incubated at 37°C for 30 min, and then analyzed using a FACScan flow cytometer (Becton Dickinson).

Total RNA was isolated from cells by the RNeasy Midi Kit (Qiagen, Valencia, CA) and concentrated to 5 μg/μl in diethyl pyrocarbonate-treated water (Research Genetics, Huntsville, AL). The RNA samples were aliquoted and stored at −70°C. cDNA microarray experiments were carried out using a protocol developed by the Laboratory of Cancer Genetics of the National Human Genome Research Institute,⁴ with minor modifications. Cy3 and Cy5 labeling was done with 54 and 72 μg of total RNA, respectively. Seventy-five μm Cy3-dUTP or Cy5-dUTP was used in each labeling reaction. Cy3- and Cy5-labeled cDNA samples, one experimental and the other a reference sample, were mixed in equal amounts. The mixture was then hybridized with a pin-spotted cDNA microarray. The microarrays were spotted robotically at the National Cancer Institute Microarray Facility, Advanced Technology Center. Each microarray contained 2076 cDNA spots corresponding to known cancer-related genes. Some of the genes were printed on the microarray more than once to provide replicates for reproducibility analysis.⁵ The hybridization data were acquired using the Avalanche Fluorescence Scanner (Molecular Dynamics, Sunnyvale, CA). Image analysis was carried out using a set of software tools developed by Y. Chen (15) at the Laboratory of Cancer Genetics, National Human Genome Research Institute.⁶

To remove the systematic bias caused by the chemical difference between Cy3 and Cy5, each microarray study was performed twice using “reciprocal labeling,” i.e., for the first microarray, the experimental sample was labeled with Cy3 and the reference sample was labeled with Cy5, whereas for the second microarray, the reference sample was labeled with Cy3 and the experimental sample was labeled with Cy5. Throughout the data analysis process, differential expression was represented by:

where GMR_i denotes the geometric mean of ratios, 1 and 2 denote the two arrays used, i denotes the array pair, and E and R denote the experimental and reference samples, respectively.

Before any of the analyses, we removed the data from empty spots and reagent-control spots [poly(dA), Cot I, Cy3, and Cy5]. On each array, 2076 of the original 2208 spots remained, and 242 of those spots were in duplicate. We also removed data from spots identified as visibly flawed (<20 in all for the 36 arrays). We then normalized the ratio data by Gaussian-kernel fitting⁷ and normalized the resulting values for each cell type so that the sum of the log-transformed values was the same for each cell sample. We next filtered the data to obtain genes whose expression differed by more than 1.7-fold in the reciprocal-averaged ratio (GMR_i). This threshold criterion was, to a degree, arbitrary. It was based, however, on quality control experiments in which we found that >99% of spots had reciprocal-averaged ratios between 1.7 and 1/1.7 when two identical samples were analyzed (Fig. 3). In addition, to further remove possible outliers (such as, e.g., the three points in the right panel of Fig. 3), we selected only spots that met the 1.35-fold change criterion for both of the duplicate arrays. We further required that the data point meet these criteria for more than 1 design point in the set of 18. This procedure selected a total of 33 different genes, including 3 that were represented twice in the microarray.

Most statistical analyses were carried out using Matlab 5.3 (Mathworks Inc., Natick, MA). We clustered the genes using agglomerative hierarchical average-linkage based on the Euclidian distance metric. For each gene, the ratio value was based on the reciprocal average, as shown in Eq. A. The data are presented in the form of color-coded CIMs generated in Matlab. We introduced CIMs previously for compact visualization of “high dimensional” biological and pharmacological data (16, 17, 18), and they have since been used by others (19). In the CIMs here, only the genes were clustered (to bring similar expression patterns together); the experiments were left in natural order.⁸

The protocol is summarized in Fig. 1. To obtain a relatively homogeneous cell population before CPT treatment, we synchronized HCT116 cells using APH. APH, an inhibitor of DNA polymerase α, blocks the cell cycle in S-phase. HCT116 cells have a doubling time of 16 h, and incubation of HCT116 cells with 1 μm APH for 17 h arrested most cells in S-phase. Cell cycling resumed immediately after removal of APH. After 17 h of exposure to APH and 2 h in APH-free medium, ∼80% of the cells were still in S-phase. The cells were then treated with or without CPT for 75 min and finally incubated in CPT-free medium for the times indicated in Fig. 1.

Cell cycle progression after APH synchronization and CPT treatment was examined using FACS after different times of culture in drug-free medium. Subpopulations of cells containing 2N (G₁ phase) and 4N (G₂ phase) DNA were separated into two distinct peaks. A FACS study of dose dependence (over a range from 1 nm to 10 μm CPT) identified CPT concentrations that caused reversible and irreversible G₂ arrest (results not shown). Gene expression changes were measured after treatment with 0, 20, or 1000 nm CPT. As shown in Fig. 2, the cells not treated with CPT resumed cell cycle progression after removal of APH, and the time taken to complete the first cycle was close to the normal doubling time of HCT116 cells. The cells treated with 20 nm CPT slowed down slightly in S-phase and were significantly delayed in G₂-M. The apparent G₂-M phase of cells treated with 20 nm CPT was approximately twice as long as that of the control cells (0.01% DMSO). The cells treated with 1000 nm CPT showed significant S-phase delay (8–10 h), and G₂-M arrest was not reversed even after a 60-h incubation in drug-free medium (data not shown).

The microarray studies included two series of experiments. The first series examined changes in gene expression caused by CPT treatment. In these experiments, total RNA was isolated from cells harvested at the time points indicated in Fig. 1, and the reference sample was total RNA isolated from cells harvested at the end of the 17-h APH synchronization (2 h before CPT treatment). This series generated 36 microarray data sets, i.e., 18 reciprocally labeled pairs of arrays. Included were duplicate sets for cells collected 5 h after treatment with 1000 nm CPT.

The second series of experiments examined the effects of APH treatment on gene expression. In these experiments, total RNA isolated from APH-treated cells was compared with total RNA isolated from nontreated cells. The APH treatment was carried out for 6, 12, 17, and 24 h. This series of experiments produced 10 microarray data sets, i.e., 5 reciprocally labeled pairs of arrays. Included were duplicate data sets for cells harvested after 17 h of treatment with APH. In addition, to evaluate random variation in the experimental procedure, we carried out 17-h APH synchronization three different times using the same conditions, and the three independent samples were compared with each other in pairwise fashion. After calibrating the signals from both the Cy3 and Cy5 channels and averaging the ratios of two reciprocally labeled duplicates, we found that >99% of the genes in the three control data sets showed ratio differences <1.7-fold (Fig. 3). Data reproducibility was also assured by the high correlation between duplicate gene pairs (represented by the same clone or different clones). As illustrated in Fig. 4, the duplicate clones clustered together.

To identify changes in gene expression induced by drug treatment and to remove random noise (isolated spikes) in the data sets, we filtered the data to obtain genes whose expression differed at more than one experimental time point by >1.7-fold in either direction from the reference for the reciprocally averaged pair and by >1.35-fold for both of the arrays. A total of 33 genes met those criteria, and 3 of them were represented twice on the microarray chip. Thus, we selected a total of 36 cDNA spots. Expression data for the 36 spots from 18 CPT experiments were clustered based on the expression patterns and represented in the form of a CIM (Fig. 4). The gene definitions were based on human Unigene build #95.

Clustering the genes brings out patterns in the data by grouping together genes with similar expression profiles. Of the 33 genes that exhibited changes according to the criteria established, 27 clustered tightly within three groups (Fig. 4,A). Group I genes were up-regulated in the control and low-dose CPT-treated cells, but the up-regulation took place at a later time in cells treated with low-dose CPT. The difference in timing of up-regulation closely matched the difference in the timing of mitosis. Those genes were not up-regulated after high-dose CPT treatment. Group II genes were down-regulated during the period of time when low-dose CPT-treated cells were recovering from G₂ delay. In the case of high-dose CPT treatment, some of these genes showed an abortive down-regulation shortly after drug removal. Group III genes were up-regulated in the cells treated with high-dose CPT during S-phase delay and/or G₂ arrest, but these genes did not change in the cells treated with low-dose CPT. Time-dependent expression changes for the 36 spots are shown in a line graph in Fig. 4 B.

The changes in expression identified by cDNA microarray were largely consistent with the results obtained using Northern blot analysis (Table 1). In summary, Northern blot analyses were run for five of the genes, each over eight experimental conditions. IMP dehydrogenase 1, which showed no significant difference in ratio from sample to sample in the microarray analysis, was used as a normalization reference. The overall Pearson correlation coefficients (r) relating microarray and Northern expression levels for the eight conditions were as follows: for p21, r = 0.88; for 14-3-3ς, r = 0.92; for cyclin B1, r = 0.79; for aurora/IPL1, r = 0.91; for p16-INK4a, r = 0.78. The overall average of the correlation coefficients for the five genes was, therefore, r = 0.86

To assess the effect of APH treatment itself on gene expression, we examined data for the 36 spots from the five APH time points. The 36 spots were ordered in Fig. 4 according to the cluster order generated from the CPT experimental data. The results are shown in the five columns at the right side of the CIM in Fig. 4. The group I genes were down-regulated in most of the APH-treated cells, whereas some of the group III genes were up-regulated in the cells treated with APH for 17 and 24 h. Group II genes did not show any consistent changes.

The basic mechanism of action for CPT has been well studied and characterized (1, 2). CPT generates replication-mediated double-strand DNA breaks, which in turn induce reversible or permanent cell cycle arrest in G₂-M. In this study, we used cDNA microarrays to investigate the molecular consequences of CPT’s activity at the transcriptional level. Among other observations, we found that reversible cell cycle arrest in G₂-M after low-dose CPT treatment was associated with delayed up-regulation of mitosis-related genes, which are normally up-regulated during G₂. In contrast, treatment with high-dose CPT elevated expression of some p53-responsive genes.

To study dose- and time-dependent effects of CPT treatment on cell cycle progress and gene expression, we carried out the experiments using APH-synchronized cells treated with 1000, 20, or 0 nm CPT (all in 0.01% DMSO). The time dependence of treatment effects was revealed by comparing each sample with an internal second-color control harvested after synchronization with APH but without any other treatment. This direct comparison of all CPT- and sham-treated samples with a single internal control permitted us to map out the time course of expression levels after treatment.

Because the cytotoxicity of CPT is cell-cycle specific and because CPT treatment directly affects cell cycle progression, we reasoned that we could optimize our chance of seeing down-regulations and modest up-regulations of genes if we synchronized the cells and then treated them at the appropriate point in the cell cycle. Among the possible methods (none of them perfect) for synchronizing cells, we selected the commonly used technique based on incubation with APH, an inhibitor of DNA polymerases. However, APH itself can induce gene expression changes. It can activate p53 and up-regulate p21/WAF1 expression (20). Mitosis may also be perturbed immediately after treatment with high-dose APH (21). To minimize (although not completely eliminate) such confounding influences, we used a minimum concentration of APH for synchronization. In this study, we did not observe perturbation of mitosis after removal of APH (Fig. 2), and a colony formation assay did not indicate significant toxicity (data not shown). To monitor gene expression changes caused by APH treatment, we compared gene expression in cells incubated with and without APH. Stress-related responses, including moderate elevation of p21/WAF1 transcript, were observed after prolonged APH treatment (Fig. 4 A), but they had little effect on cell cycle progression after removal of APH. Because the internal control for the APH experiments was an unsynchronized cell sample and that for the CPT-treatment experiments was an APH-treated sample, the appearance of up-regulated gene expression after CPT treatment indicated an additional up-regulation above the already increased expression caused by APH. For example, p53-responsive genes were up-regulated after APH treatment. Subsequent treatment with high-dose CPT induced additional up-regulation, whereas treatment with low-dose CPT or no CPT largely maintained the already elevated transcription levels.

On the basis of cluster analysis (Fig. 4 A), the 33 changing genes fell into three predominant groups based on their patterns of gene expression:

Group I genes tended to be maximally up-regulated during the time that cells were in transit through G₂-M. Their increased expression correlated well with the period of time during which most of the APH-treated control cells or the low-dose CPT-treated cells were beginning to exit from mitosis. Among six genes in this group, cyclin B1, aurora/STK15, Nek2, and hRad6 are directly involved in mitotic regulation. Through formation of a complex with CDK1 (p34^cdc2), cyclin B1 promotes entry into mitosis, and its expression is tightly regulated in the cell cycle. Its expression is low in G₁ but up-regulated throughout the S and G₂ phases (22). Cyclin B1 associates with centrosomes at about the time of transition of the centrosome’s microtubule organizing center from an interphase to a mitotic pattern (Ref. 23 and references cited therein). STK15, a member of the aurora/IPL1 kinase family, is a short-lived protein associated with the mitotic spindle and centrosome. It is required for cell cycle progression and genome stability (24, 25). The STK15 gene is amplified and overexpressed in multiple human tumor cell types and can induce excessive centrosome duplication. Nek2 is another protein kinase whose expression is up-regulated during the S and G₂ phases. Nek2 is a core component of the centrosome. It is probably required for separation of centrosomes and for the G₂-to-M transition (26, 27, 28). hRad6 codes a ubiquitin-conjugating enzyme critical for degradation of proteins that maintain the G₂ or M states of the cell (29). When analyzing a CIM of the entire 1694-gene data set (results not shown), we identified a tight cluster branch that contained all of the group I genes as well as other genes showing expression patterns similar to those of group I but not meeting the criteria used to select the set of 33 genes. That branch is shown in Fig. 5. Most of the additional genes are mitosis related. Among them, ckshs2 and CENP-F kinetochore protein have been shown to have regulated expression patterns similar to that of cyclin B1 (30).

The mitosis-related function and changes in expression level during cell cycle suggest that up-regulation of this group of genes is a part of the cell cycle-regulatory mechanism in G₂-S. It is well established that DNA damage inhibits cyclin B1 expression and relocalizes cyclin B1 protein to the cytoplasm (29, 31, 32). The inhibition of cyclin B1 expression may be attributable to an increase in active p53 protein (33, 34). As noted above, in addition to cyclin B1, we found that CPT treatment delayed or inhibited the up-regulation of several other mitosis-related genes. This inhibition occurred before completion of DNA replication when cells were still in S or S-like phase. The disruption of transcriptional regulation suggests that cells arrested in G₂-like phase after DNA damage are different from normal G₂ cells. Although DNA replication was apparently completed, those cells may have been missing necessary mitotic machinery and therefore not have been ready to proceed to mitosis. This lack of complete mitotic machinery may be involved in G₂ arrest and mitotic catastrophe. The latter occurs in some types of cells after G₂ arrest after DNA damage (14, 35, 36). At this time, we do not know the upstream event that caused disruption of transcriptional regulation, and we do not know how this event relates to the G₂ checkpoint. However, it seems unlikely that the disruption of transcriptional regulation in cell cycle is directly attributable to p53 activation because the inhibition of mitotic gene expression could occur without further activation of genes in the p53 pathway (see below).

The implication of TRAF5 and p16-INK4a in group I is not clear at this point. TRAF5 is an adapter protein that links signaling from tumor necrosis factor receptor to nuclear factor-κB and to the apoptosis-regulating kinase ASK1 (37, 38). Via both mechanisms, TRAF5 tends to protect cells from apoptosis. p16-INK4a tends to block entry into S-phase by binding to cdk4/6, thereby preventing the activation of cdk4/6 by cyclin D. In all of our experiments, p16-INK4a exhibited expression patterns very similar to those of cyclin B1. This behavior is consistent with reports of p16 accumulation during G₂ delay of UV-treated HeLa cells (Ref. 39 and references cited therein).

These genes were down-regulated during the period of G₂ delay in the cells treated with low-dose CPT. This down-regulation seemed not to be associated with G₂ per se because it was not seen in control cells as they passed through G₂, nor in the early period of G₂ delay in the cells treated with low-dose CPT. It was only during the extension of G₂ that the down-regulation occurred. This finding suggests that the down-regulation of group II genes is intimately related to a checkpoint that reversibly delays the transition from G₂ to mitosis. The absence of down-regulation in the high-dose CPT experiment suggests that most of the cells that were in S-phase during CPT treatment never reached a normal G₂ state. Some genes in this experiment showed an immediate transient down-regulation, perhaps reflecting the subpopulation of cells that were already at or near G₂ at the time of CPT treatment (see Fig. 2). The group II genes are heterogeneous in function, and they have not been associated previously with DNA damage response or normal cell cycle regulation. Some of them are associated with calcium-dependent pathways, cellular metabolism, and signal transduction pathways.

Group III genes are characterized by increased expression after high-dose, but not low-dose, CPT treatment. These genes are inducible by DNA damage and are associated with cell cycle arrest and apoptosis. Five of the eight genes in the group are known to block cell cycle progression and to be activated by p53 in response to DNA damage: (a) the cyclin-dependent kinase inhibitor p21 (35, 40); (b) the chaperone 14-3-3ς (14, 41); (c) the type IIC protein phosphatase Wip1 (42, 43); (d) the apoptosis-inducer CD95/Fas (44); and (e) the damage-specific DNA-binding protein subunit DDB2 gene (45). HCT116 is p53 wild type and exhibits p53 function (13). It is possible that the up-regulation of p53-responsive genes in HCT116 after high-dose CPT treatment was attributable to activation of p53. Although we do not have direct data on p53 protein in these experiments, this pattern of up-regulation of five p53-responsive genes in these p53 wild-type cells strongly suggests that p53 protein activity was up-regulated after high-dose CPT treatment. These observations are consistent with present understanding of the transcriptional role of p53 in cellular responses to DNA damage.

In contrast to the observation with high-dose CPT, group III genes showed no up-regulation after low-dose CPT treatment, suggesting that p53-dependent cell death- and cell cycle-related genes were not significantly up-regulated under conditions of reversible cell cycle arrest. Two of the p53-responsive genes (p21 and 14-3-3ς) were in fact down-regulated, perhaps reflecting recovery from the up-regulation induced by exposure to APH. This behavior suggests a remarkable distinction between the reversible G₂ delay seen in these experiments after low-dose CPT treatment and the more permanent G₂ arrest that followed extensive DNA damage after high-dose CPT.

In addition to p53-responsive genes, group III also included the genes for TGF-β superfamily protein and IFN-induced 17-kDa protein. The TGF-β superfamily protein is involved in signal transduction events that regulate cell growth and differentiation (46). It has been reported to be up-regulated in cells treated with the DNA topoisomerase II inhibitor VP-16 (47). The highly correlated expression pattern seen here indicates a possible relationship of TGF-β superfamily protein to DNA damage and p53 activity, although such associations have not previously been reported.

In conclusion, we find it striking that 27 of the 33 genes selected objectively from the overall set of 1694 represented on the microarrays fell coherently into three tightly clustered patterns of behavior. These patterns appear to define signatures for the initial molecular consequences of CPT treatment. Even more unexpected, changes in gene expression during the first cell cycle after treatment revealed patterns after a relatively nontoxic dose of CPT that were qualitatively different from those after a highly toxic dose, although there was no gross cell death during the course of the experiments in either case. We were not surprised to see that the high dose precipitated broad changes not seen at the low dose, but we were surprised to find that the reverse was also true, that the low dose led to changes not seen at the high dose. On the basis of these observations, we propose that there is a fundamental difference between the gene expression changes associated with reversible G₂ delay that follows mild DNA damage and permanent G₂ arrest that follows more extensive DNA damage. The fate of cells after DNA damage is probably determined by coordination and balancing of multiple parallel processes at both the transcription and protein levels. Directed experimental investigations will be required to test the hypotheses generated by these microarray experiments. In particular, it will be instructive to follow up the suggestion of these experiments that the differences involve patterns of regulation in the pathways of p53 and other stress-response genes, as well as in the mitotic machinery, especially with regard to centromere formation and function.