Identifying transcription factor functions and targets by phenotypic activation

Abstract

Mapping transcriptional regulatory networks is difficult because many transcription factors (TFs) are activated only under specific conditions. We describe a generic strategy for identifying genes and pathways induced by individual TFs that does not require knowledge of their normal activation cues. Microarray analysis of 55 yeast TFs that caused a growth phenotype when overexpressed showed that the majority caused increased transcript levels of genes in specific physiological categories, suggesting a mechanism for growth inhibition. Induced genes typically included established targets and genes with consensus promoter motifs, if known, indicating that these data are useful for identifying potential new target genes and binding sites. We identified the sequence 5'-TCACGCAA as a binding sequence for Hms1p, a TF that positively regulates pseudohyphal growth and previously had no known motif. The general strategy outlined here presents a straightforward approach to discovery of TF activities and mapping targets that could be adapted to any organism with transgenic technology.

Fig. 1.

Microarray expression data resulting from overexpression and/or deletion of 57 TFs that cause growth inhibition when overexpressed. Only TFs that contain expression profiles significantly above microarray noise when overproduced are shown. The diagram shows all 5,222 genes represented on the array after removal of dubious ORFs, transposable elements, mitochondria-encoded genes, and bad spots on the array. (A) Overexpression experiments. z-score-transformed data are shown (see Methods). Genes are ordered such that those with the greatest level of induction when a given TF is overexpressed are grouped, and then TFs are ordered according to the number of genes meeting this criterion. The color scale reflects z-score, which reflects noise-corrected log(ratio) (see Methods) and extends from ≈3-fold induction (red) to ≈3-fold reduction (green). (B) Microarray expression data (z-score transformed) of the corresponding deletion mutants. Rows and columns are in the same order as A, except that four essential TFs are missing. (C) Induction of specific functional classes of genes in response to TF overexpression. The columns are in the same order as A. Induction was scored with the WMW P value (see Methods).

Fig. 2.

Behavior of known TF targets in response to overexpression or deletion of the TF and compared with a similar analysis of genomewide ChIP-chip data from Harbison et al. (4). Each point indicates, for the TF indicated, the WMW P value (see Methods) for the difference of medians between the ranked TRANSFAC targets and those of all other ORFs; i.e., a point with a higher −log(P) value indicates that the median of TRANSFAC targets is shifted higher toward the top of the ranked list of genes. For our data, the z-scores are ranked; for Harbison et al. data, the P values are ranked. Only TFs with P < 0.05 in either Harbison et al. or this study are shown.

Fig. 3.

Promoter analysis of differentially regulated genes in response to TF overexpression. (A) Motifs identified by RankMotif compared with known DNA-binding motifs for overexpressed TFs. Binding sites are displayed as logos in which the height of each letter is proportional to its weight in determining the motif. The purple underlined portion indicates bases consistent with the known binding site. The likelihood of the known motif matching the RankMotif consensus is given (formula and code are available on request). The orange underlined portion of the HMS1 motif shows the six bases that match the gel-shifted segment in B. (B) Gel mobility-shift assays. The purified DBDs of Gcn4p, Upc2p, and Hms1p TFs were incubated with oligonucleotides containing two tandem copies of the motif sequence predicted by RankMotif. The same amount of purified MBP-TF DBD was used for each oligonucleotide in the binding reaction.