Changes in genomewide occupancy of core transcriptional regulators during heat stress

Abstract

Organisms respond to heat stress by reprogramming gene expression. In Saccharomyces cerevisiae, heat-induced genes tend to be regulated by factors that belong to the Spt-Ada-Gcn5 acetyltransferase (SAGA) transcription regulatory pathway, whereas heat-repressed genes tend to be regulated by a parallel pathway involving transcription factor IID (TFIID). Here, we examine whether heat stress affects the occupancy of representative factors of each pathway at promoter regions throughout the yeast genome. Representatives of the SAGA pathway include the TATA binding protein, Spt3, and Mot1. Representatives of the TFIID pathway include the TATA binding protein, TAF1, and Bdf1. We find that heat stress causes disassembly of the TFIID pathway at genes that are inhibited by stress. In contrast, heat induces assembly of the SAGA pathway at stress-induced genes, although many also assemble along the TFIID pathway. Other genes were found to assemble almost exclusively along the TFIID pathway. Strikingly, these genes are lowly transcribed and are generally not induced. Thus, heat stress leads to factor assembly along each pathway but with distinct transcriptional outcomes. Further investigation of these pathways reveals that Bdf1 and Mot1 negatively regulate the SAGA pathway in different ways. The findings suggest that Bdf1 blocks assembly, whereas Mot1 promotes disassembly of the transcription machinery.

Fig. 1.

Changes in gene expression and factor occupancy during heat stress are linked. (A) Key to the Venn diagrams shown in B–H.(B–H) Inner circles correspond to the 10% of all tail-to-head genes that exhibited the greatest increase (Upper) or decrease (Lower) in gene expression upon heat shock. Tail-to-head refers to those genes that do not share their upstream region with another gene. Outer circles correspond to the 10% of all tail-to-head genes that have intergenic regions with the greatest increase (Right) or decrease (Left) in factor occupancy upon heat shock. We chose a cutoff of 10% because this fraction of the genome tends to be significantly reprogrammed during heat stress (28, 30). Genes narrowly missing this cutoff might nevertheless be regulated by stress. Other types of comparison or cutoffs yielded similar results (Table 3). If no relationship exists, then ≈10% of each data set is expected to overlap by chance. The number of genes meeting each criterion and for which we have data are indicated within each circle. Each factor assayed by chIP is presented separately as indicated.

Fig. 3.

Cluster analysis of chIP-chip data. (A) A total of 2,715 tail-to-head intergenic regions were filtered to include only those having an absolute value >0.5 in at least one chIP-chip experiment. A cutoff of 0.5 ensured that the value was >3 SD (3σ) from the median value for the untagged control. Although this cutoff minimized the number of false positives, it also excluded genes with significant but modest changes, resulting in an underestimate of group membership. Filtered intergenic regions (excluding “input” and “no tag”) were next clustered by K-means. Clusters 1–4 are indicated to the left and are represented by 98, 218, 172, and 59 intergenic regions, respectively (see also Table 4, which is published as supporting information on the PNAS web site). Within each cluster, the data were arranged hierarchically. Columns (arrays) were also clustered hierarchically. Similar results were obtained when intergenic regions of divergently transcribed (head-to-head) genes were also included (Fig. 9, which is published as supporting information on the PNAS web site), or when different filtering criteria were used. Increased factor occupancy (Left) or increased expression of the adjacent downstream gene (Right) during heat stress is shown by shades of red, and decreases are shown by shades of green, scaled by the left and right side of the key, respectively. Gray indicates missing data. The percentage of genes in each cluster that are SAGA-dominated is indicated (5). The genomewide distribution is 10%. (B) Genes identified in A were sorted by fold changes in gene expression. Sliding windows (size = 50 genes, step size = 1) through this data set were examined by plotting the average fold change in gene expression in each window against the percentage of genes that fell into clusters 1–4. The curves represent smooth fits. Similar results were obtained by using gene expression profiles published elsewhere (28), ruling out fortuitous cluster-specific expression artifacts. (C) The curve from cluster 1 in B is shown along with curves generated by dividing cluster 1 data in half, representing genes that lost Bdf1 and genes that were generally unchanged.

Fig. 2.

Correlation of genomewide changes in factor occupancy. (A–J) Median-centered fold changes in factor occupancy are plotted on a log₂ scale for all intergenic regions upstream of genes in tail-to-head or head-to-head configurations. Head-to-head refers to divergently transcribed genes, which share their upstream region with another gene. X axis labels are shown to the right, and y axis labels are shown on top. (K) A representative correlation for downstream promoterless intergenic regions (i.e., flanking genes are in a convergent tail-to-tail configuration). Correlation R² values are indicated.