Functional enrichment of genes associated with motif families. Each point represents a motif family. For each family, the set of genes that contain a motif from the family in their promoter was identified. The P-value of the best match to a ChIP experimental profile (x-coordinate) and the highest coregulation achieved in a collection of gene expression experiments (y-coordinate) were computed. The vast majority of true motif clusters (green) are strongly supported by functional data, and are well separated from the respective P-values computed with the same motif clusters but randomly shuffled promoter-gene association (red).

The effect of binding site affinity on Leu3 multimodality. (A) Cluster structure in a fragment of the Leu3 selection network. Black nodes, high-affinity (k_d < 50 nM) motifs; white, low-affinity (k_d > 170 nM); small gray, unknown affinity. Arcs connect neighbors with high substitution rate. Arcs with low substitution rates are not shown. As only motifs with measured affinities and their neighbors are presented, some real families may appear fragmented. Note that no component contains both high- and low-affinity nodes. (B) Rate of substitution as a function of affinity change within and between the families. Substitutions with similar effect on log(k_d) are grouped together, and their joint rate is plotted. The rates of substitution between high-affinity sites and from high- to low-affinity sites differ significantly (P < 0.01). (C) Motif abundance box-plots for different affinity intervals. Both nonfunctional motifs (undetectable k_d-s) and medium-affinity motifs (50 nM < k_d < 170 nM) have very low abundances compared to motifs in the high- and low-affinity intervals, which were also identified as families in the selection network. This may indicate that medium-affinity motifs are selected against to avoid ambiguity of site modality and to increase transcriptional program robustness.

Partial view of the selection network. Motifs are represented as circles; motifs that have conservation rate >0.2 at 2 SD are drawn as large circles, motifs with conservation rate >0.2 at 3 SD are shown as large filled circles. Substitutions are shown as color-coded arcs. Black, nonnegative substitution rate (ρ > 0 and ρ > -0.5 at 2 SD); cyan, negative rates (ρ < -0.5 at 2 SD). A family in the selection network is a cluster of motifs that are interconnected via high or neutral substitution rate arcs. Low-rate substitution arcs separate families. Arc directions are not shown for readability, but 91% of the negative rate arcs point from a family motif to a motif outside the family. We annotated each family by a consensus motif, and by the names of the transcription factors that match that consensus (if such are known). Many known transcription factors are identified as families of motifs, some of which are shown in this figure. Families can be further analyzed for intricate intrafamily relations. Certain transcription factors (e.g., Reb1) correspond to more than one family due to multiple functionalities, as discussed later. Transcription factors with binding sites that are less frequent in the genomes (e.g., Gcn4) have larger variance on rate estimations, and thus are harder to cluster robustly. An interactive version of this figure is available on our Web site.

Multimodality of the Reb1 transcription factor. A part of the selection network containing motifs that are associated with Reb1 is shown in greater detail. All nodes represent variants of the Reb1 consensus. Note that in this figure, all selection network arcs, including those with low confidence, are plotted, and reverse complement motifs are not combined. (A) Reb1 motifs. (B) Reb1 reverse complementing motifs. Black arcs represent high substitution rate (low selective pressure). Blue arcs represent low substitution rate (high selective pressure). A clear two-family structure emerges, where low-rate arcs are separating a large Reb1 family from a smaller one. The structure is mirrored in the reverse complement motifs. The Reb1 promoter (C) contains two autoregulatory sites, each located in a different family, with distinct binding site affinities (Ideker et al. 2001). This raises the hypothesis of two distinct Reb1 modes of operation, each activating a different group of motifs in specific concentration. Note that nodes in the network are octamers and that the Reb1 consensus is a septamer.

Selection on the ESR1 consensus. The color-coded matrices specify substitution rates from the motif to its neighbors (upper MSSM) and vice versa (lower MSSM). Rows in the matrices represent nucleotides; columns stand for positions in the octamer. The number inside the cell indicates the substitution rate when the neighbor has the nucleotide associated with the row in the position associated with the column, and all other motif nucleotides are unchanged. Blue and cyan cells indicate low rates (ρ < 0 at 2 SD and 0.5 SD, resp.); red and pink cells indicate high rates (ρ > 0 at 2 SD and 0.5 SD, resp.). For example, the substitution of T with C at position 4 (GCGATGAG →GCGACGAG, upper matrix, cell C4) has the rate -3.4. The SDs of the rate estimations are given in parentheses. Most substitutions that perturb the ESR1 consensus have a very low rate, with few exceptions that appear not to disrupt functionality. On the other hand, several substitutions that lead to the consensus appear with high rates, suggesting the possibility of motif optimization toward the preferable binding site.