Distinct properties of cell-type-specific and shared transcription factor binding sites

Abstract

Most human transcription factors bind a small subset of potential genomic sites and often use different subsets in different cell types. To identify mechanisms that govern cell-type-specific transcription factor binding, we used an integrative approach to study estrogen receptor α (ER). We found that ER exhibits two distinct modes of binding. Shared sites, bound in multiple cell types, are characterized by high-affinity estrogen response elements (EREs), inaccessible chromatin, and a lack of DNA methylation, while cell-specific sites are characterized by a lack of EREs, co-occurrence with other transcription factors, and cell-type-specific chromatin accessibility and DNA methylation. These observations enabled accurate quantitative models of ER binding that suggest tethering of ER to one-third of cell-specific sites. The distinct properties of cell-specific binding were also observed with glucocorticoid receptor and for ER in primary mouse tissues, representing an elegant genomic encoding scheme for generating cell-type-specific gene regulation.

Figure 1. Shared and cell-specific sites have unique sequence properties

A) The locations of ECC-1 specific (blue), T-47D specific (yellow) and shared (red) ER binding sites are distributed across human chromosomes with an overall lack of shared sites. B) Less than 10% of the total set of identified ER binding sites is consistently bound in ECC-1 and T-47D (top panel). Less than 20% of cell-specific ER binding sites contain EREs, while 46% of shared sites harbor EREs (bottom panel). C) The distribution of the predicted relative affinity for the best match to an ERE is shown for each ECC-1 specific (solid blue), T-47D specific (solid yellow) and shared (solid red) ER binding sites as well as ECC-1 and T-47D ER binding sites in which the order of the nucleotides has been shuffled (dashed lines). All three sets of observed binding sites (solid lines) are enriched over background (dashed lines); however, shared sites represent the highest affinity matches to an ERE. See also Figure S1.

Figure 2. Relationship between chromatin accessibility and cell-specific binding

A) DNaseI hypersensitivity data is comprehensive, capturing >90% of p300 and CTCF binding sites in each cell type, while indicating that ~65% of ER binding sites are found in open chromatin prior to treatment with E2. B) Most ER binding sites that are cell-specific and lack strong matches to an ERE are found in open chromatin prior to E2 treatment, while ER binding sites that are shared between the two cell lines and have strong EREs are found in closed chromatin prior to treatment. C) Bound EREs found in closed chromatin prior to E2 treatment are stronger matches to an ERE than sites found in open chromatin for ECC-1 specific (blue), T-47D specific (yellow) and shared (red) ER binding sites. See also Figure S2.

Figure 3. Cell type-specific DNA methylation is found at cell-specific ER binding sites

A) Typical examples of ER binding signal (grayscale read density, black is highest read coverage), DNase I hypersensitive sites (black bars) and DNA methylation (green: unmethylated, red: methylated). B) ECC-1 specific (blue), T-47D specific (yellow) and shared (red) ER binding sites have cell type-specific DNA methylation with the unbound cell type exhibiting methylation. C) DNA methylation is rarely present in open chromatin, while CpGs in closed chromatin can be either methylated or unmethylated. See also Figure S3.

Figure 4. Transcription factors co-occur with ER at cell-specific binding sites

A) Motifs were identified in ECC-1 (left panel) and T-47D (right panel) ER binding sites after EREs were masked out. B) The expression levels from RNA-seq data of transcription factors that recognize each family of motifs were analyzed and identified ETV4, FOXA1 and GATA3 as potential ER interacting factors. Error bars represent s.e.m. C) Cell-specific ER binding sites overlap ETV4, FOXA1 and GATA3 binding sites measured by ChIP-seq at a higher rate than shared ER binding sites. In each case, the overlap is significantly higher than CTCF, which serves as a negative control. D) ER binding sites that co-occur with these transcription factors in ECC-1 cells (left panel) and T-47D cells (right panel) lack strong matches to an ERE. See also Table S1.

Figure 5. Model of cell type specificity in ER binding site selection

A) The model depicts the differences between shared and cell-specific binding sites with respect to the presence of EREs (red bars), chromatin accessibility, DNA methylation (“me” yellow circles) and co-occurring transcription factors (“TF” orange and purple ovals). B) Model predictions (gray) of the relative expression of each factor between cell lines are highly correlated with observed transcript levels from RNA-seq experiments (white). C) ROC curves show that the thermodynamic model performs significantly better when a tethering model of ER binding is implemented. D) The distributions of differences between the probability of ER binding in the tethering model and the full model are significantly different between the shared binding sites (red) and either the ECC-1 specific (blue) or T-47D specific (yellow) binding sites. See also Figure S4.

Figure 6. Cell-specific sites observed in primary mouse tissues as well as for the glucocorticoid receptor

A) ER binding shows a high degree of cell type specificity between mouse liver (red) and uterus (blue). B) The canonical ERE motif was identified using ER binding sites found in either liver or uterus. C) Similar to results in human cancer cell lines, tissue-specific ER binding sites are significantly less likely to harbor EREs than shared sites bound by ER in both liver and uterus. D) Very few GR binding sites are bound in both ECC-1 and A549. E) The distribution of the best sequence match to an GRBE is shown for each ECC-1 specific (blue), A549 specific (yellow) and shared (red) GR binding site. F) Sites without a strong match to a GRBE are more likely to be found in open chromatin prior to treatment with dexamethasone than sites without a strong match to a GRBE. See also Figure S5.

Figure 7. Evidence supporting the model of ER binding site selection

A) ER binding sites found within 50 kbp of the transcription start site of genes that show cell type-specific gene expression changes most often share the same cell type specificity in binding. B) Luciferase enhancer assays indicate that shared sites with an ERE show strong context-independent activity and cell-specific ER binding sites without EREs drive less activity; however, that activity is cell type-specific. C) High affinity bound EREs (red) are under more evolutionary constraint than low affinity bound EREs (orange) and bound EREs are under more evolutionary constraint than unbound EREs (blue and green). The levels of constraint are highly correlated with the information content of the ERE motif. See also Figure S6 and Table S2.