Portrait of transcriptional responses to ultraviolet and ionizing radiation in human cells

Abstract

To understand the human response to DNA damage, we used microarrays to measure transcriptional responses of 10 000 genes to ionizing radiation (IR) and ultraviolet radiation (UV). To identify bona fide responses, we used cell lines from 15 individuals and a rigorous statistical method, Significance Analysis of Microarrays (SAM). By exploring how sample number affects SAM, we rendered a portrait of the human damage response with a degree of accuracy unmatched by previous studies. By showing how SAM can be used to estimate the total number of responsive genes, we discovered that 24% of all genes respond to IR and 32% respond to UV, although most responses were less than 2-fold. Many genes were involved in known damage-response pathways for cell cycling and proliferation, apoptosis, DNA repair or the stress response. However, the majority of genes were involved in unexpected pathways, with functions in signal transduction, RNA binding and editing, protein synthesis and degradation, energy metabolism, metabolism of macromolecular precursors, cell structure and adhesion, vesicle transport, or lysosomal metabolism. Although these functions were not previously associated with the damage response in mammals, many were conserved in yeast. These insights reveal new directions for studying the human response to DNA damage.

Figure 1

Correlation between northern blots and microarray measurements of gene expression. The logarithms of fold-changes R(i) from northern blots for 20 genes were plotted against the logarithms of fold-changes from the microarray measurements in the current study. The northern-blot data were obtained from Tusher et al. (19). The logarithms of fold-changes from microarray data were obtained by averaging the logarithms of the pairwise fold-changes for all 15 samples. The error flags indicate the SD of the logarithms of the pairwise fold-changes for the 15 samples. Fifteen of the twenty-one (71%) genes plotted had SDs that crossed the line of identity x = y. Four of the genes had low ranks by SAM associated with FDR > 50% (open circles). The squared correlation coefficient R² = 0.823 was obtained using the remaining genes, which were contained in the set of genes with FDR < 10%. One gene (cyclin F) is represented by two probe sets on the microarray, and values for both probe sets are plotted (gray circles). One gene was analyzed by quantitative PCR (closed square).

Figure 2

Effect of number of samples on FDR. SAM was used to identify probe sets responsive to IR. The graph shows curves of FDR (expressed as a percentage) as a function of the number of probe sets called significantly changing. Each curve was generated for a given set of samples from 2, 4, 7, 10 or 15 individuals. The sets containing 2, 4 and 7 samples were non-overlapping. For example, set 7a included seven samples (1–7), and set 7b included seven different samples (8–14). Increasing the number of samples led to a dramatic decrease and stabilization in the FDR. Note that SAM sometimes generated anomalously high values for FDR when the number of probe sets called significant was small.

Figure 3

Estimate of total number of differentially expressed probe sets. The net number of significant probe sets was plotted as a function of the FDR from an analysis by SAM of all 15 samples. The net number of significant probe sets is the number called significant multiplied by (1−FDR). The net number reached an asymptotic value, providing an estimate of the total number of damage-responsive probe sets.

Figure 4

Distribution of fold-changes for damage-responsive probe sets. The histograms show the distribution of fold-changes for 1932 IR-responsive (upper panel) and 3143 UV-responsive (lower panel) probe sets, which were identified by SAM with an FDR of 10%. The bins between 2 and ∞ represent probe sets with more than 2-fold changes. The fold-change was not available (NA) for about 100 IR-responsive probe sets and 200 UV-responsive probe sets, because these probe sets had a negative value for expression either before or after exposure to DNA damage. Relatively few genes deemed significant by SAM had less than 1.1-fold responses even though 43 and 34% of the genes represented on the microarray had fold-changes in this range after IR and UV, respectively.

Figure 5

Functional categories of top-ranked damage-responsive genes. The 200 top-ranked IR-responsive probe sets (upper panel) and 200 top-ranked UV-responsive probe sets (lower panel) were categorized by function. The probe sets were identified by SAM with the additional criterion that the response to DNA damage was at least 1.3-fold in magnitude. Genes with more than one function were assigned to the more specific category. Thus, genes with anti-apoptotic functions were assigned to the ‘Apoptosis’ category, although they could have been assigned to the ‘Cell cycle/proliferation’ category. Genes in the ‘DNA repair’ category were not assigned to the ‘Stress response’ or ‘Cell cycle/proliferation’ categories. The highest specificity categories were ‘DNA repair’, ‘Metabolism of macromolecular precursors’ and ‘Apoptosis’. ‘Signal transduction’ was considered to be the least specific category. For example, MAPKAPK2 is involved in signal transduction, but was assigned to ‘Stress response.’

Figure 6

Hierarchical clustering of damage-responsive genes and samples. Data are shown for the top-ranked 200 UV-responsive probe sets and top-ranked 200 IR-responsive probe sets. The dendrogram to the left of the heat map shows clustering of the 350 probe sets (50 probe sets responded to both UV and IR). The dendrogram above the heat map shows clustering of the cell lines by treatment. The values used for clustering were the logarithm of the ratio of treatment (UV or IR) to mock treatment, and the scale to the lower right shows the fold-change indicated by each color. Yellow color represents induced expression following UV or IR, and blue color represents repressed expression. Gray spots on the heat map represent negative ratios of treatment to mock treatment for which logarithms values could not be computed. These cases occurred rarely and were generated because hybridization to the mismatched probes for that gene was greater than hybridization to the matched probes. There were 25 established p53-responsive genes, which are marked with a black bar to the right of the heat map. Four of these genes, TNFRSF6, cyclin A2, cyclin B1 and BAX, were represented by two probe sets each, for a total of 29 probe sets. Putative p53-responsive genes are marked with a gray bar. The green bars indicate clusters that are enriched in genes involved in apoptosis and DNA repair. The red bar highlights a cluster that was strongly repressed by IR and enriched for genes involved in regulating the cell cycle and proliferation.