Quantitative description of gene response to TFs. Overlapping Hunchback and Knirps gradients combine two symmetrical zones, where the even-skipped expression stripes 4+6 and 3+7 are formed (A). The Hunchback cluster in the eve 3+7 enhancer has a lower PWM cutoff (site affinity) and a lower statistical significance (less sites) than that in the eve 4+6. In contrast, the Knirps cluster in eve 3+7 enhancer has more high-affinity sites than that in the eve 4+6 (B). The detected composition of clusters for Hunchback and Knirps in the even-skipped locus supports the existing biological model and might be used for interpretation of gene response to the upstream regulators.

Recognition of known regulatory interactions. The data from Table 2 are represented as a scatter plot. The correlation values (X-axis) for tested motif/gene locus combinations are plotted vs. the cluster significance (-Log[E], Y-axis). Motifs are shown on the right side; data points for each motif have the same color. Brackets (A–C) mark three distinct correlation groups. The cluster significance cutoff (the cluster E-value cutoff) was set for each motif on the basis of a global training procedure (see Table 1). Data points in the upper right corner correspond to the best confidence of recognition; poorly performing motifs have a consistently low E-value (Gt, Eve).

Search for the maximum of the correlation between the location of Bicoid clusters and the location of Bicoid-driven CRM sequences in eight different genes. Y-axis shows the size of the resolution window; X-axis shows the PWM cutoff values; the color intensity scale shows the correlation (the Pearson association coefficient) value. The correlation also depends on the cluster significance (the E-value cutoff), which is shown by three consequent projections with the increasing E-value cutoff (A–C). The global maximum of correlation for the Bicoid motif is found at the PWM cutoff values 4.1–4.3, for the windows ∼500–600 bp and at the E-value cutoff 0.0005 (B). The CC maximum has nonmonotonous distribution in the parametric space, caused, in part, by a specific ratio between low- and high-scoring PWM matches, i.e., by specific ‘cluster composition.’

Discovery of CRM sequences using clustering of single motifs. Results of mapping of cis-regulatory modules in the giant locus are shown in comparison with the distribution of known functional elements. (A) Relative distribution of Bicoid and Giant expression patterns along the embryo. (B) The distribution of Kruppel and (C) the distribution of Bicoid clusters. Besides the known posterior enhancer (Berman et al. 2002), the method indicates the presence of two Bicoid clusters upstream and downstream of the gene. Notice that the Bicoid cluster located downstream of the giant CDS (B, in position ∼10000) contains low-affinity Bicoid sites (PWM is below 4.2, the established optimal cutoff for Bicoid; see Table 1) and does not represent a candidate CRM. We also found that this sequence consists of several long direct repeats (∼100 bp), which is not typical for Drosophila CRMs. In contrast, the upstream candidate region contains a very strong cluster of closely spaced high-affinity Bicoid sites.

Representation of clustered recognition motifs. The distribution of clusters for the Bicoid motif in the locus of even-skipped is shown. The color intensity scale represents statistical significance of constitutive clusters. A map of known functional elements in the eve locus is given on the bottom of each panel (X-axis: relative position in the locus in base pairs). (A) Dependence of the cluster significance on the PWM site score cutoff (Y-axis) at the fixed window size (500 bp, see also the red mark on B). (B) How the size of the resolution window (Y-axis) affects the cluster significance at the fixed PWM cutoff (red mark on A). The representation given shows that Bicoid clusters in eve stripe 2 and eve stripe 1+5 regions have distinct ratios of high- and low-affinity binding sites (the cluster composition). Both of these regions in the eve locus have been shown to be responsible for formation of corresponding eve expression stripes in a fly embryo. The weak ‘Bicoid’ cluster inside the eve late1 element indicates the presence of the clustered motif related, but not identical to Bicoid. It is known, however, that the eve late2 region is under control of other homeodomain proteins (such as Eve itself) that also have the conserved core consensus sequence TAAT. The best correlation obtained for Bicoid clusters with their cognate CRMs is 0.76 for the eve locus (Table 2, eve locus vs. Bicoid motif).