Associations between TFs and target genes were extracted from Beyer et al.
]. We used the subset of TF-regulated gene associations labeled as highly confident
by the authors. We built a set of CTFPs based on the compilation of computationally-predicted CTFPs by four different methods [6
]. We selected those TF pairs predicted as cooperative at least by two methods. The resulting amount of CTFPs was 32, composed by the pairing of 26 distinct TFs (Additional file 3
Following the Breadth-First Search algorithm described by Yu & Gerstein [11
], we built a directed network of TFs as a multi-layered hierarchical structure (Figure ). We merged the two upper layers of the network (with 8 and 2 TFs, respectively) in order to avoid a low number of TFs that would hinder statistical calculations. The final network had four layers (the bottom layer termed layer-1
, the topmost layer termed layer-4
) composed by 148 TFs and 96 regulatory interactions (Additional file 4
). This network will be referred to as regulatory hierarchy
. A slightly different implementation of the algorithm places all targets of a TF in the same level, thus forcing all interactions in the hierarchy to point downward or horizontally, but never upwards [19
]. We also built and analyzed this hierarchy (Additional file 5
, Additional file 6
). The enrichment in CTFPs for level n
was calculated as the ratio of the probability of finding a CTFP in that level vs random expectation. The statistical significance was calculated using 10 [5
] random hierarchies where the target genes were randomly exchanged between TFs.
We obtained mRNA expression data for the following cellular conditions: diauxic shift [20
], cell cycle [21
], sporulation [22
], and six environmental stress conditions: heat, acid, alkali, peroxide, NaCl and sorbitol [23
]. Expression levels (copies/cell), apparent half-life of the transcripts (in minutes) and transcriptional frequency (mRNAs/hour) were obtained from Holstege et al
]. Correlation between expression levels was calculated using a Spearman's correlation test. We downloaded the protein activity profiles of the TFs in our sets from 17 experiments of the database RegulonProfiler, where TF activity profiles are inferred from genomewide changes in mRNA expression patterns of groups of genes with similar regulation (called ChIP-based regulons), which allowed the authors to quantify the post-translational activity of TFs [25
]. Only activity profiles with E-value
0.05 were considered. Correlations were calculated using a squared Spearman's correlation test [26
]. In all cases, the distribution of the correlation of the activity levels of CTFPs was compared against the distribution of the activity levels of 1000 non-CTFPs by means of a KS test.
Information on the essentiality of yeast proteins was downloaded from the Yeast Deletion Project [28
]. Information on synthetic lethals was obtained from the BioGrid database [29
]. Association between essentiality and transcriptional cooperativity was calculated by means of a Fisher's test.
Protein evolutionary rate for TFs was obtained from Xia et al
]. The correlation between evolutionary rates and expression levels for CTFPs was calculated as the dn/ds ratio for members of the same CTFP and the ratio of their expression levels. A Spearman's test was used to calculate the correlation.
Protein functions were extracted from the FunCat catalogue [31
]. Being FunCat a hierarchical classification, we used first-level functions with experimental evidence, which amounted to 16 different functions. Functional similarity between TFs was calculated as in Aguilar & Oliva [10
]. We defined two TFs as co-functional if their functional similarity was larger than the 90th
percentile of the distribution of the functional similarity values for all TF pairs.
We first measured the enrichment for each function in each level of the regulatory hierarchy by means of a z
-score, using a random model consisting in 10 [5
] regulatory hierarchies where the gene functions controlled by the TFs were randomly exchanged. We repeated this analysis using the Gene Ontology functional annotation at depth level 2 (which is roughly equivalent to the first level of FunCat) [32
]. TFs annotated at lower levels were re-annotated with the corresponding parent terms of level 2. Only experimental annotations were used. The R software was used for all statistical tests [33
Availability of supporting data
The data sets supporting the results of this article are included within the article and its additional files.