The role of chromatin accessibility in directing the widespread, overlapping patterns of Drosophila transcription factor binding

Abstract

Background: In Drosophila embryos, many biochemically and functionally unrelated transcription factors bind quantitatively to highly overlapping sets of genomic regions, with much of the lowest levels of binding being incidental, non-functional interactions on DNA. The primary biochemical mechanisms that drive these genome-wide occupancy patterns have yet to be established.

Results: Here we use data resulting from the DNaseI digestion of isolated embryo nuclei to provide a biophysical measure of the degree to which proteins can access different regions of the genome. We show that the in vivo binding patterns of 21 developmental regulators are quantitatively correlated with DNA accessibility in chromatin. Furthermore, we find that levels of factor occupancy in vivo correlate much more with the degree of chromatin accessibility than with occupancy predicted from in vitro affinity measurements using purified protein and naked DNA. Within accessible regions, however, the intrinsic affinity of the factor for DNA does play a role in determining net occupancy, with even weak affinity recognition sites contributing. Finally, we show that programmed changes in chromatin accessibility between different developmental stages correlate with quantitative alterations in factor binding.

Conclusions: Based on these and other results, we propose a general mechanism to explain the widespread, overlapping DNA binding by animal transcription factors. In this view, transcription factors are expressed at sufficiently high concentrations in cells such that they can occupy their recognition sequences in highly accessible chromatin without the aid of physical cooperative interactions with other proteins, leading to highly overlapping, graded binding of unrelated factors.

Figure 1

DNaseI accessibility and in vivo DNA binding by transcription factors across the eve locus. DNA binding in stage 5 embryos is shown as ChIP-chip scores (blue) for 675-bp windows that fall above a 1% FDR threshold for 21 sequence-specific transcription factors, TFIIB and the transcriptionally active form of RNA polymerase II (POLII). The sequence-specific factors are grouped into three major regulatory classes that regulate patterning along the Dorsal-Ventral axis of the embryo (D-V), initiate patterning along the Anterior-Posterior axis (Early A-P), or establish later pair rule patterns along the Anterior-Posterior axis (Pair rule A-P). DNaseI accessibility at stage 5 is shown for 75-bp windows of sequence tag density (red) along with the locations of accessible regions above the 5% FDR threshold (black bars). At the bottom, the locations of major RNA transcripts are shown (grey) as well as the autoregulatory CRM (Auto) and the four stripe initiation CRMs (S3/7, S2, S4/6 and S1/5) (green). Nucleotide coordinates in the genome are given in base pairs.

Figure 2

DNaseI accessibility and in vivo DNA binding by transcription factors across a 200-kb genomic region. The figure is labeled using the same conventions in Figure 1 except that the RNA transcript locations are shown in light blue at the bottom.

Figure 3

Levels of factor occupancy and genome accessibility correlate. The median DNase-seq tag density in non-overlapping cohorts of 200 1-kb ChIP-chip peaks is shown down the ChIP-chip rank list (continuous lines). The ChIP-chip data are from stage 5 embryos and the DNaseI accessibility data are from stages 5 (green) and 14 (purple). The 95% confidence limit for median DNaseI accessibility of each cohort is indicated. Shown also is the percent of ChIP-chip peaks that are overlapped by 5% FDR accessible regions in stage 5 embryos (dashed green line). The regions most highly bound by transcription factors are to the left along the x-axis and results are plotted as far as the 25% FDR cutoff. The location of the ChIP-chip 1% FDR threshold is indicated by the vertical black dotted line. Results for the regulatory transcription factors (a) Dichaete (D) and (b) Twist (TWI) are shown. Additional file 7 shows plots for all 21 regulators.

Figure 4

Factor recognition sites in DNaseI accessible regions are more highly bound in vivo than sites in closed chromatin. Separately for each transcription factor, all significant recognition sequences in the euchromatic genome for four affinity cohorts were identified using PWMs derived from in vitro DNA binding data (Table 1) [17,70]. In addition, matches to ten PWMs derived by random permutation of nucleotide position order were derived for each factor. Sites in each affinity cohort for both the genuine and scrambled PWMs present were each classified as either accessible or inaccessible, using the 5% FDR DNaseI accessible regions to define accessible regions (Table 1). The median ChIP-chip scores (y-axis) for the 500-bp regions ±250 bp around recognition sites in each affinity cohort were plotted separately for accessible (red lines) and inaccessible (blue lines) genomic regions. Dark red and blue lines show results for the genuine factor PWMs, light red and blue lines the median result for the scrambled PWMs. The highest affinity cohort is to the left (x-axis). Web logo representations of the PWM representing the highest and lowest affinity cohorts of genuine recognition sites are shown at the bottom. The 95% confidence limits for the median ChIP-chip scores are indicated. Plots for (a) CAD, (b) GT, (c) KNI, and (d) HRY are shown. Additional file 8 provides similar plots for all 16 factors for which sufficiently accurate PWMs are available.

Figure 5

Accessibility better explains in vivo occupancy than does intrinsic affinity. We identified and grouped 150-bp local peaks of accessibility within DNaseI accessible regions into non-overlapping cohorts of 200 peaks down the DNase-seq rank list. (a) The median ChIP-chip score in each cohort for each factor. (b) The sum of occurrences of recognition sequences that match the factor's PWM (P < 0.003) in each cohort for each factor. The bottom row in each panel shows the relative DNase-seq scores for each cohort. Data for each factor were normalized by scaling the median value for each row and plotted as a heat map. The correlation coefficients of the data for each factor with the DNase-seq scores are shown on the right. The correlations are calculated using data for each accessible region, not the cohort average values.

Figure 6

Levels of HB factor occupancy and DNaseI accessibility change between developmental stages. The level of hunchback (HB) binding and DNaseI accessibility to the Caudal (cad; left) and hb; right) genes are shown at stages 5 and 9. The figure is labeled using the same conventions in Figure 1 except that the locations of the regions above the ChIP-chip 1% FDR threshold are indicated by black horizontal lines beneath the continuous traces of ChIP-chip scores. Additional file 9 shows similar results for Medea (MED).

Figure 7

Temporal changes in levels of HB occupancy correlate with changes in DNaseI accessibility. We identified the 1-kb regions ±500 bp of the peak nucleotide of binding for each of the 400 regions most highly bound by HB at (a) stage 5 and (b) stage 9. (a) Scatter plot of the ratio of ChIP-chip scores at stage 5 over those at stage 9 (x-axis) versus the ratio of DNase-seq scores at stage 5 over those at stage 9 (y-axis). (b) Scatter plot of the ratio of ChIP-chip scores at stage 9 over those at stage 5 (x-axis) versus the ratio of DNase-seq scores at stage 9 over those at stage 5 (y-axis). The Pearson correlation coefficients (r) for each comparison are shown in the top right of each panel. See Additional file 11 for similar plots for MED.