Dynamic CRM occupancy reflects a temporal map of developmental progression

Abstract

Development is driven by tightly coordinated spatio-temporal patterns of gene expression, which are initiated through the action of transcription factors (TFs) binding to cis-regulatory modules (CRMs). Although many studies have investigated how spatial patterns arise, precise temporal control of gene expression is less well understood. Here, we show that dynamic changes in the timing of CRM occupancy is a prevalent feature common to all TFs examined in a developmental ChIP time course to date. CRMs exhibit complex binding patterns that cannot be explained by the sequence motifs or expression of the TFs themselves. The temporal changes in TF binding are highly correlated with dynamic patterns of target gene expression, which in turn reflect transitions in cellular function during different stages of development. Thus, it is not only the timing of a TF's expression, but also its temporal occupancy in refined time windows, which determines temporal gene expression. Systematic measurement of dynamic CRM occupancy may therefore serve as a powerful method to decode dynamic changes in gene expression driving developmental progression.

Figure 1

Transcription factors occupy CRMs in highly dynamic patterns. (A) k-Means clustering of binding profiles of four TFs (Twi, Tin, Mef2, and Bin) shows emerging temporal-binding patterns, which are similar in all four cases. Each row represents a TF-bound region, each column a time window of development. The intensity of blue represents the level of ChIP enrichment (log2 of the peak height). Colored vertical bars represent early (green), continuous (yellow), and late (red) bound regions. (B) Hierarchical clustering of total CRM occupancy, using the average TF-binding enrichment at a given time point, for all 2813 unique CRMs. (C) Representative examples of individual CRMs from (B), illustrating that different temporal-binding patterns can occur in the same locus either in a coordinated (CRMs #659, 153, 895, 38) or cascading (CRMs #81, 147, 392, 134) manner. Blue represents the level of ChIP enrichment, yellow represents no binding.

Figure 2

Relationship between TF binding, motif quality, and temporal occupancy. (A) The distribution of quantitative-binding signal (log2 ChIP peak height) for different temporal classes of CRMs bound by the TF Twist shows that there is a clear distinction between early, continuous, and late CRMs. All TFs are shown in Supplementary Figure 3. (B) Average enrichment of all motifs in CRM classes defined by binding of respective TFs. The appropriate motif is significantly enriched in bound CRMs compared with non-bound CRMs (NB). There is no difference in motif enrichment between the three temporal classes. (C) Quantitative TF binding (ChIP enrichment) is only weakly correlated (Pearson's r=0.192) with motif strength as measured by the PWM log-odds score. Exemplified using Twi binding at 4–6 h, where each dot represents a single CRM (all TFs shown in Supplementary Figures 5 and 6). A number of CRMs with no detectable Twist binding (blue dots) contain high-affinity motifs. (D) Motifs in bound CRMs are significantly more conserved than in non-bound CRMs, but there is no significant difference in motif conservation between temporal classes. NB, non-bound CRMs.

Figure 3

Temporal TF binding correlates with temporal gene expression and function. (A) Proportion of target genes with CRMs bound at different time points (green, early; yellow, mid-stages; red, late stages), divided into groups based on the first stage that they are expressed in the embryo. In all, 33% of genes with CRMs bound at 2–4 h initiate expression at 2–3.5 h, whereas an additional 33% initiate expression at 3.5–6 h. Significantly enriched classes peaking in different time points are marked with asterisks (Fisher's exact test ^*P<0.05, ^**P<0.001). (B) Representative examples of expression patterns of genes targeted by temporally bound CRMs. The only expression of the rosy gene in mesodermal and/or muscle occurs at early stages in development, the expression of fasIII initiates in visceral and somatic muscle at mid-embryogenesis, whereas CG5080 is strongly expressed in somatic muscle at late stages. (C) Genes with occupied enhancers early in development have different functions than those that are bound by the same TF at later stages. GO categories (with the number of genes indicated in brackets) showing differential enrichment between target genes grouped by temporal-binding classes. The blue shade corresponds to the –log of the P-value (Benjamini–Hochberg corrected). For the full set of GO terms see Supplementary Table 3. (D) The upper panel displays the number of CRMs bound by Mef2 and/or Biniou out of the 69 CRMs associated with 18 muscle contractile genes. Bottom panel shows the temporal expression pattern of the 18 genes, which initiate at 8–9 h of development, reaching high levels of expression by 11–12 h, reflecting the temporal occupancy of their CRMs. Solid line represents mean expression and dashed lines represent 1 s.d. above and below the mean.