Genome-Wide Identification of Basic Helix-Loop-Helix and NF-1 Motifs Underlying GR Binding Sites in Male Rat Hippocampus

Abstract

Glucocorticoids regulate hippocampal function in part by modulating gene expression through the glucocorticoid receptor (GR). GR binding is highly cell type specific, directed to accessible chromatin regions established during tissue differentiation. Distinct classes of GR binding sites are dependent on the activity of additional signal-activated transcription factors that prime chromatin toward context-specific organization. We hypothesized a stress context dependency for GR binding in hippocampus as a consequence of rapidly induced stress mediators priming chromatin accessibility. Using chromatin immunoprecipitation sequencing to interrogate GR binding, we found no effect of restraint stress context on GR binding, although analysis of sequences underlying GR binding sites revealed mechanistic detail for hippocampal GR function. We note enrichment of GR binding sites proximal to genes linked to structural and organizational roles, an absence of major tethering partners for GRs, and little or no evidence for binding at negative glucocorticoid response elements. A basic helix-loop-helix motif closely resembling a NeuroD1 or Olig2 binding site was found underlying a subset of GR binding sites and is proposed as a candidate lineage-determining transcription factor directing hippocampal chromatin access for GRs. Of our GR binding sites, 54% additionally contained half-sites for nuclear factor (NF)-1 that we propose as a collaborative or general transcription factor involved in hippocampal GR function. Our findings imply a dose-dependent and context-independent action of GRs in the hippocampus. Alterations in the expression or activity of NF-1/basic helix-loop-helix factors may play an as yet undetermined role in glucocorticoid-related disease susceptibility and outcome by altering GR access to hippocampal binding sites.

Figure 1.

Validation of experimental design. (a) Plasma corticosterone measured prior to the start of the infusion (0 minutes) and at 10, 20, and 30 minutes thereafter. Blood samples were drawn by hand through indwelling cannulas. Corticosterone rises after the start of the infusion and does not differ between NSC (black circles, n = 30 to 31) and stressed (gray squares, n = 17 to 20) groups. (b) Plasma ACTH levels rose in stressed group rats, in line with activation of the central stress response. ^#P < 0.0001 (between groups); *P < 0.005 (different from 0 minutes). n = minimum of 30 (NSC) or 17 (stressed). (c) Plasma corticosterone in trunk blood for animals taken for ChIP-seq. All animals received similar corticosterone infusions regardless of group or replicate (one-way ANOVA, n = 8 per replicate). (d) PFC c-fos expression is significantly different between groups. The two stressed group replicates have higher c-fos levels compared with either NSC group. ^#P < 0.0001 (compared with NSC replicates), n = 8. NSC replicates were not different from each other, and stressed replicates were not different from each other. Means ± standard error of the mean are shown.

Figure 2.

GR ChIP-seq in rat hippocampus reveals known and unknown GR binding sites. (a–f) Browser tracks showing regions of interest in the genome. Tag density profiles are shown for each group indicating regions called as GR-bound peaks. Known genes (RefSeq) are indicated along with the direction of the coding strand (chevron). (a and b) Promoter region for the clock gene period 1 (Per1) known to contain binding sites for GRs in rat hippocampus (56), and the promoter region of metal ion chelator metallothionein-2a (108). (c) The upstream region of the Ddit4 gene also known to contain GR binding sites. The presence of these known sites within the peak list highlights the success of the experimental protocol. (d–f) Novel GR binding sites in the hippocampus. Several locations flanking the calcium-calmodulin–dependent kinase 2a (Camk2a) gene and the promoter of the thyroid hormone receptor α (Thra) gene bind GRs. A weaker intronic binding site is also observed in the Bdnf gene.

Figure 3.

(a) Scatter plot of the log₂ of peak intensity at the same chromosome position in NSC and stressed groups. (b) Log ratio (M) versus mean average (A) plot for NSC and stressed peaks. Peaks significantly different between groups were set to display as red dots. There are no peaks significantly different between groups by DESeq. (c) Heat map plotting tag density over the center of the peak (±3 kb) supports the view that GR binding is not significantly different between groups. Peaks are ranked according to their tag density. (d) Distribution of hippocampus GR binding sites across the genome annotated according to nearest transcription start site. Most binding events occurred in intergenic or intronic regions with comparatively little occurring at promoters or within coding regions. UTR, untranslated region.

Figure 4.

Motif analysis identifies candidate binding partners for GRs in rat hippocampus. (a) De novo motif analysis of DNA sequences underlying GR peaks in rat hippocampus returns motifs that strongly resemble a palindromic GRE, NF-1 half-site, and a basic helix–loop–helix transcription factor most likely NeuroD1 or Olig2. The P value shows significance level for the enrichment of the motif indicated. The score indicates the similarity of the identified motif to the known motifs for the factor indicated, with higher percentages equivalent to greater confidence in an accurate match. The percentage of peaks containing motifs that match this model is also indicated (percentage targets). s.p., specific probability. (b) GR binding sites in rat hippocampus identified with high confidence by Polman et al. (65) were additionally examined by de novo motif discovery using the same settings as above. A motif matching a GRE half-site (underline) was more degenerate and poorly represented within peaks overall. (c) Motifs for the factors identified are located toward the center of the GR binding site. (d) The frequency of discovered motifs was separately determined for GR peaks containing a fully palindromic GRE, and for the remaining GR peaks that did not contain a GRE palindrome. By default, all GR peaks containing a full GRE also contained a GRE half-site, and so this was not determined (nd) for the full GRE containing population. Most peaks absent a full GRE contained a GRE half-site instead.

Figure 5.

Overlap of GR binding sites (with dexamethasone or corticosterone) with preaccessible chromatin defined by DNase I hypersensitivity (without dexamethasone or corticosterone) in the (a) 3134 mammary carcinoma cell line, and (c) in adrenalectomized mouse liver. Redrawn from data in Stavreva et al. (67) and Grøntved et al. (37), respectively. De novo motif discovery for DHS peaks and GR binding sites in (b) 3134 cells, or (d) mouse liver, produces many of the same motifs. Motifs are arranged in order of significance with P values for DHS motifs between 1e-7713 and 1e-42, and P values for GR binding site motifs between 1e-2918 and 1e-53. All motifs indicated have a confidence score of >80%. (e) GR binding enrichment was determined by ordering peaks according to tag density and counting motifs in the top and bottom third, respectively. Weaker peaks have fewer occurrences of transcription factor motifs. GREs, NF-1 half-sites, NeuroD1, Olig2, Atoh1, Lhx3-like, AP-1, and Egr2 motifs particularly are less well represented in weaker GR binding sites. Conversely, GRE half-sites and myocyte enhancer factor 2 motifs are better represented in weaker binding sites. (f) Composite motifs containing GREs, NF-1, and NeuroD1/Olig2 are more common in stronger peaks whereas weaker binding sites are more likely to contain isolated GRE sequences. GRE/H, full GRE or GRE half-site; NF-1-H, NF-1 half-site; ND1, NeuroD1; O2, Olig2. (g) ChIP assay showing the presence of NF-1 at four out of five selected GR binding sites in rat hippocampus is not corticosterone/GR-dependent. Intergenic peak 1912 may be a de novo site requiring GR to recruit NF-1. The liver specific Tat gene promoter does not recruit GRs in this tissue and is shown as a negative control for NF-1 binding. Means ± standard error of the mean are shown; n = minimum of 7 for all sites except the Tat-negative control, where n = a minimum of 3.