p53 accumulation and activation in response to various cellular stresses
To determine the overall and p53-dependent transcriptional responses to different types of inflammatory stress, we exposed the TP53wt and TP53−/− HCT116 colon cancer cell lines to NO•, H2O2, hydroxyurea, or hypoxia in time course experiments for up to 24 hours. The conditions for each exposure were chosen for comparable activation of the p53 network using p53 protein accumulation and posttranslational modification as surrogate indicators.
The Western blots shown in all display similar p53 accumulation, although with different kinetics, upon exposure to the various stress conditions. Maximal accumulation of p53 was observed after 4 hours of exposure to H2O2, 8 hours of exposure to NO•, or 12 hours of exposure to hydroxyurea or hypoxia. All stress conditions resulted in p53 phosphorylation with kinetics similar to those of p53 accumulation, indicating p53 activation. As expected, no p53 expression was observed in TP53−/− cells (Supplementary Fig. 1). NO• or H2O2 exposure resulted in a comparable degree of DNA damage by Comet Assay (Supplementary Fig. 2A and B). Based on those observations, we selected three time points, representing early, maximal intermediate, and late transcriptional response, for expression profiling. The first, 1-hour exposure, already showed p53 accumulation and/or phospho-Ser15 p53 for all stresses. The second represented the maximum of p53 protein accumulation, which occurred at 8 hours for NO•-treated cells, at 4 hours for H2O2, and at 12 hours for hydroxyurea- and hypoxia-treated cells. The third time point was 24 hours of exposure for all conditions. We did not observe any significant toxicity from any of the treatments over the 24-hour period of exposure.
Figure 1 Western blot of p53 accumulation and posttranscriptional modification. HCT116 cells were exposed to (A) 1 mmol/L Sper/NO, (B) hypoxia, (C) 0.1 mmol/L H2O2, or (D) 1.5 mmol/L hydroxyurea (HU) for the indicated times (hours). Numbers below the blots indicate (more ...)
Treatment-specific gene expression profiles
Genes significantly modulated by each exposure were identified by BRB Array Tools using class comparison analysis with univariate F test (P < 0.001) and a global permutation test (P < 0.004). For this analysis, each time variable was considered a class, and each exposure type was analyzed separately. There were 1,997 genes modulated by NO•, 838 genes by H2O2, 1,054 genes by hydroxyurea, and 1,652 genes by hypoxia. shows a dendrogram of hierarchical clustering of the samples based on the 4,047 genes identified by F tests as modulated by at least one of the exposure types. The hybridization replicates for each condition cluster together, indicating reproducibility of the data across arrays. The modulated genes clearly distinguish “early-response” and “late-response” samples (). The early-response branch includes the 4-hour H2O2 and all 1-hour samples, subdivided into the four treatments. The late-response branch includes all 8-, 12-, and 24-hour exposures and discriminates hydroxyurea- or H2O2-treated cells from NO•-or hypoxia-treated cells. As an indicator of the robustness of the clustering, we computed the “R” measure after the dendrogram was cut into three clusters (A, dashed horizontal line). The R measure of 1.0 indicates that the clustering is highly robust and emphasizes that the pattern of gene modulation is unique to each exposure type. Of note, unbiased selection of 3,367 genes filtered for log ratio variation >0.01 and no more than 5% missing values across all experiments, resulted in similar discrimination of early versus late responses and overall similarity of response between hypoxia and NO• as well as between hydroxyurea and H2O2 (Supplementary Fig. 3).
Figure 2 Time course of treatment-specific gene expression profiles as determined by univariate F test (P < 0.001) and controlled by a global permutation test. A, hierarchical clustering of samples based on the compiled list of 4,047 genes identified by (more ...)
The “uniqueness” of each stress condition is also highlighted by the Venn diagram in . The majority of the modulated genes changed their expression significantly in response to only one type of exposure. Only 40 genes of 4,047 total were significantly affected by all four treatments. Assuming total independence of the exposures, the expected value would be (1,396/4,047)4 × 4,047 = 57.3. The 40 genes are listed in along with a hierarchical gene cluster analysis using individual array replicates. Most of them were up-regulated at 1 hour but changed to different degrees over time. Six genes in a small cluster were up-regulated at the intermediate and late time points, preferentially in response to hypoxia and NO•.
Nonoverlapping genes from each treatment shown on the Venn diagram in are further categorized in as up-regulated or down-regulated. About two thirds or 64% (1,270 of 1,997) of the genes modulated by NO• were down-regulated. Similarly, 61% (648 of 1,054) of the genes were down-regulated by hydroxyurea and 54% (455 of 838) by H2O2. The gene expression profiles of cells exposed to hypoxia differed from those of the other three treatments; only 21% (343 of 1,652) were down-regulated.
Using PathwayAssist, we further analyzed the 40 genes whose change was common to all four exposures (Venn diagram, center, ). The analysis looked for the shortest paths connecting genes with each other or with different types of nodes (e.g., enzymes, cellular processes, or functional classes). The resulting Biological Association Network (BAN) is presented in . Early-response genes JUN and FOS, as well as p53, were incorporated into the BAN occupying central places. Cellular processes, such as apoptosis, death, and proliferation, were also identified and incorporated into the network (highlighted). So were the nodes for “hypoxic treatment” as well as the “HIF-1 complex.” The 40 genes regulated in common by all stress conditions are associated with many of the cellular networks (including the p53 network) known to be involved in responses to severe stress.
Figure 3 BAN using PathwayAssist for the 40 genes with gene expression changes common to all four exposures. The colors represent PathwayAssist categories: red, gene; yellow, cell process; green, small molecule; orange, protein functional class; gray, cell object; (more ...)
To facilitate interpretation of the significant time course–related genes from the four exposures, as determined by F
tests, we used High-Throughput GoMiner, a program that organizes genes of interest in the context of the Gene Ontology (19
). That program enabled us to focus on groups of genes with similar functions or genes involved in the same network (18
). A hierarchical clustering of the resulting Gene Ontology categories is depicted in . Several of the categories were affected only by NO•
and hypoxia (upper third of the category clusters). That observation supports the close relationship between NO•
and hypoxia detected in the previous analysis (). Two clusters consisting of 16 Gene Ontology categories were affected by all four exposures. Interestingly, one-half of those are cell cycle–related categories. Included were G1
-S and G2
-M cell cycle checkpoints and M phase–specific microtubule processes. The category “DNA replication,” identified in NO•
and hypoxia exposures, overlapped with the H2
-related gene expression profile. The lower half of the hierarchical cluster contains Gene Ontology categories, affected only by NO•
or hypoxia. However, a subset of those categories, mainly “DNA replication initiation,” “Protein metabolism,” and “Macromolecule metabolism,” was shared by H2
exposure and hypoxia.
Figure 4 Functional categorization by High-Throughput GoMiner based on the Gene Ontology of the sum of 4,047 differentially expressed genes from . These data were visualized by the publicly available software “Genesis” using the Pearson (more ...)
Cell cycle analysis by flow cytometry (fluorescence-activated cell sorting)
Gene Ontology analysis indicated that cell cycle regulation was a prominent feature for all exposures. Therefore, we analyzed cell cycle and DNA synthesis profiles in TP53wt
cells at the end of each exposure ( ). As expected, hydroxyurea resulted in S-phase arrest, with >65% BrdUrd incorporation in TP53wt
cells compared with ~35% in the respective untreated cells. In contrast, H2
led to G2
-M arrest, more pronounced in the TP53−/−
cells than in the TP53wt
ones. That pattern is typical for DNA-damaging agents. The preferential bypass of G1
arrest and increase in G2
-M arrest observed is also typical for DNA damage response in p53-deficient cells. Very significant changes in the cell cycle profile were observed after NO•
exposure as well. There was a pronounced decrease in BrdUrd incorporation in cells with S-phase DNA content, a condition previously described as quiescent S phase (4
), along with a significant arrest in G2
-M. The effect of hypoxia was similar to that of NO•
, with a complete loss of active (BrdUrd positive) S-phase and G2
Figure 5 Functional cell cycle analysis by fluorescence-activated cell sorting. A, HCT116 and HCT116 TP53−/− cells were labeled with BrdUrd during the last 30 minutes of a 24-hour exposure as indicated, followed by double staining with FITC-conjugated (more ...)
p53-dependent gene expression profiles
To examine further the contribution of p53 activation to gene expression following different types of cellular stress, we did pairwise comparisons between gene expression in HCT116 and HCT116 TP53−/− cells for each time point and each treatment, using the univariate two-sample t test (P < 0.001) with randomized variance. A total of 842 genes, of which 399 were induced, were differentially expressed in a p53-dependent way upon NO• exposure. The other three treatments resulted in fewer genes differentially expressed in a p53-dependent way. Exposure to H2O2 resulted in the differential expression of 272 genes, of which 187 were induced. Exposure to hydroxyurea resulted in the differential expression of 283 genes, of which 184 were induced, and exposure to hypoxia resulted in the differential expression of 315 genes, of which 177 were induced. Altogether, a total of 1,396 single genes were affected in a p53-dependent way by at least one of the four exposures, and 746 of them were induced. The 746 up-regulated genes contained 28 of a list of 53 previously reported p53 transcriptional target genes (Supplementary Table 1). That constituted a 15-fold enrichment over random chance. The highest frequency of known p53 target genes (19 of 187, 10.2%) was observed in response to H2O2. There was specificity within the p53 transcriptional response in that some of the known p53 target genes were induced by only one treatment. For example, GADD45A and TP53I11 (PIG11) were induced by NO•; GPX1 was induced by H2O2; APAF1 was induced by hydroxyurea; and IGFBP3 was induced by hypoxia.
To assess the overall contribution of p53 activation, we used the set of 1,396 genes found to be p53 dependent in at least one of the individual t tests to generate a sample cluster tree (Supplementary Fig. 4), based on the array replicates, time points, and stress conditions (as in ). The resulting cluster tree showed remarkable similarities to the dendrogram consisting of genes that changed significantly over the time course (F test, A). Hybridization replicates of each condition clustered together. The arrays that defined “early” and “late” responses in were similarly grouped, and there were consistent modulations of gene expression over time across the different types of exposure. The relationships between NO• and hypoxia and between hydroxyurea and H2O2 seen in were also preserved. Hierarchical clustering of samples from the intermediate time points based on the same 1,396 p53-dependent genes yielded the expected discrimination between TP53wt and TP53−/− genotypes (data not shown).
As indicated in , the temporal pattern of gene modulation was quite consistent, regardless of type of exposure. However, each exposure type was always clearly discriminated from the others, suggesting a degree of specificity in the p53-mediated transcriptional response. That specificity was further evidenced by the Venn diagram in , which was generated using all 1,396 p53-dependent genes. Only a small fraction of genes were modulated in two, three, or all four p53-mediated gene expression profiles. This observation is consistent with , in which genes modulated by all exposures, regardless of p53 dependence, were similarly grouped.
Figure 6 Analysis of p53-mediated stress-related genes as determined by the univariate t test with randomized variance model. A, Venn diagram based on the cumulative 1,396 genes that discriminate between p53wt and p53−/− cells at any time point (more ...)
The 14 genes in the center of the Venn diagram in are listed in together with a dendrogram generated using the array replicates corresponding to the maximum p53 accumulation by Western blot (). The majority of genes in the group are p53 target genes and showed varying degrees of p53-dependent up-regulation [BTG2, DDB2, FDXR, p53CSV (HSPC132), PLAB, SESN1, and SNK (PLK2)]. In addition, DKK1, another p53 target gene, was induced by NO•, H2O2, or hydroxyurea, but repressed by hypoxia at 24 hours (data not shown). The general consistency of those results suggests that the data set and t test analysis may be useful in the identification of new p53 target genes.
The total number of genes differentially expressed (induced or repressed) in a p53-dependent manner for each time point and exposure type is depicted in . As expected, there was a trend toward an increasing number of genes with time of exposure. It was also clear from these diagrams that the kinetics of p53-mediated gene expression is specific to each exposure type. More than 80% of the genes differentially expressed by 1 hour of exposure to NO• were repressed, whereas almost 90% of the genes differentially expressed by 1 hour of exposure to hypoxia were induced. Those proportions were very similar to changes in gene expression found by the F test (). They indicate that signal transduction pathways intersecting with p53 transcription are uniquely activated upon each type of stress exposure. The proportions of induced genes are maximal at the same time point that showed maximum accumulation of p53 protein by Western blot ().
The identity and fold expression change of the top 20 most up-regulated or down-regulated genes for each of the four stress conditions are shown in the Supplementary Table 2A–D. Greater expression changes were observed with NO• and hypoxia than with hydroxyurea and H2O2. The top 20 gene lists included 12 previously described p53 target genes for hydroxyurea, nine for NO•, six for H2O2, and one for hypoxia.
Novel candidate p53 target genes
Well-characterized p53-responsive genes usually contain p53-binding sites within either their promoter or their first intron sequences. The p53-responsive element (p53RE) has been defined as two decamers of the palindromic consensus sequence RRRCWWGYYY (where R is a purine, W is adenine or thymine, and Y is a pyrimidine) separated by a 0- to 13-bp spacer. Because of the degeneracy of this sequence, it has been difficult to reliably identify p53RE in the regulatory regions of genes solely based on adherence to the consensus sequence. A novel approach uses the so-called position weight matrix (PWM) method (20
). We have developed an algorithm based on a PWM generated by combining 34 experimentally validated p53RE (Supplementary Table 1) from genes known to be transcriptional targets of p53. That PWM was applied to all genes in the center of the Venn diagram in (genes modulated in a p53-dependent manner by all four stresses). All previously validated p53REs within that set of genes were found by the PWM algorithm. Most interestingly, novel putative p53REs were found in the predicted protein KIAA0247
, the putative exonuclease FLJ12484
, the recently identified p53CSV
), and the serine-threonine kinase CNK
). The sequences of the putative p53REs in those genes are provided as Supplementary Table 3. Validation of the putative p53 target genes is currently under way in our laboratory.
Validation of array data by quantitative reverse transcription-PCR
For validation of the microarray results, we used quantitative real-time RT-PCR. We selected eight genes [SNK (PLK2), PMAIP1, DDB2, CNK (PLK3), RAI3, POLE2, CPE, and CDKN1C] with well-characterized functions related to, among others, cell cycle, differentiation and division, DNA repair, apoptosis, and secretion. Overall, independent validation by RT-PCR showed a good correlation with the microarray analysis (Pearson r = 0.822, P < 0.0001). The data are presented as Supplementary Fig. 5 and Supplementary Tables 4 and 5. A more detailed description of the validation experiments is also available in the Supplement.