Conservation analysis predicts in vivo occupancy of glucocorticoid receptor-binding sequences at glucocorticoid-induced genes

Abstract

The glucocorticoid receptor (GR) interacts with specific GR-binding sequences (GBSs) at glucocorticoid response elements (GREs) to orchestrate transcriptional networks. Although the sequences of the GBSs are highly variable among different GREs, the precise sequence within an individual GRE is highly conserved. In this study, we examined whether sequence conservation of sites resembling GBSs is sufficient to predict GR occupancy of GREs at genes responsive to glucocorticoids. Indeed, we found that the level of conservation of these sites at genes up-regulated by glucocorticoids in mouse C3H10T1/2 mesenchymal stem-like cells correlated directly with the extent of occupancy by GR. In striking contrast, we failed to observe GR occupancy of GBSs at genes repressed by glucocorticoids, despite the occurrence of these sites at a frequency similar to that of the induced genes. Thus, GR occupancy of the GBS motif correlates with induction but not repression, and GBS conservation alone is sufficient to predict GR occupancy and GRE function at induced genes.

Genomic response elements coordinate gene transcription networks through the binding of regulatory factors that serve as sensors of physiological and environmental cues. Typically, these elements are composites, comprised of binding sites for multiple distinct factors (1–3) clustered within conserved genomic segments (some extending to 2 kb), whose effects are integrated to elicit specific transcriptional regulatory outcomes.

In response to corticosteroid signaling, the glucocorticoid receptor (GR) occupies primary glucocorticoid response elements (GREs) and alters expression of genes that mediate a range of essential biological processes, such as development and immune responses (4, 5). Within a given cell type, binding of GR to GREs can either up- or down-regulate transcription in a gene-specific manner. For example, in human A549 alveolar epithelial cells, GR occupancy at the ETNK2 GRE triggers transcriptional induction (3), whereas binding of the receptor at the IL8 GRE drives repression (6). At a major class of GREs, GR binds directly to 15-bp GR-binding sequences (GBSs) composed of imperfect palindromic hexamers separated by 3 bp (3). At a distinct class of GREs, GR tethers indirectly to DNA through protein–protein interactions (7). The regulatory significance of the different classes of GREs is not well understood.

Previously, we found that GR-occupied GREs are evenly distributed upstream and downstream from the transcription start site (TSS) of their target genes, with the majority >10 kb from the corresponding promoter (3). Strikingly, although the 15-bp GR-occupied GBSs are highly variable among different GREs, the precise sequence of the GBS within an individual GRE is highly conserved across multiple species. Moreover, individual GREs at different genes appear to retain specific “architectural signatures,” including distinct GBSs and binding motifs for other factors.

Studies with GR and its close relative, androgen receptor, established that receptor occupancy at their corresponding response elements is commonly a determinant of transcriptional responsiveness (1, 3), and that the core binding sites at these genomic elements are evolutionarily conserved (3). Thus, we wished to examine whether the conservation, per se, of the receptor-binding sequences at their target genes was sufficient to predict receptor occupancy in vivo. In the present study, we identified glucocorticoid responsive targets, computationally inferred 15-bp GBS motifs, and tested whether species conservation of these sites alone was sufficient to predict GR-occupied genomic GREs.

Results

Identification of GR Target Genes and GBS Motifs.

We began our study by performing expression microarray profiling to identify genes responsive to glucocorticoids in mouse C3H10T1/2 mesenchymal stem-like cells. Single-stranded cDNA, synthesized from samples obtained from cells treated for 90 min with ethanol vehicle or 1 μM dexamethasone (dex), a synthetic glucocorticoid, was labeled and hybridized with the arrayed oligonucleotide probes. Approximately 100 genes were found to be responsive to dex; 69 up- and 17 down-regulated genes were validated in C3H10T1/2 cells by quantitative PCR (>1.5-fold difference compared with ethanol treated samples) [supporting information (SI) Table S1].

To identify GBS motifs at genes responsive to dex in C3H10T1/2 cells, we scanned computationally 32 kb upstream and 32 kb downstream from the TSSs of these genes. We generated a GBS positional weight matrix using 79 GBSs (Table S2) identified through chromatin immunoprecipitation-microarray (ChIP-chip) tiling experiments (3). Successive 15-bp windows progressing across the 64 kb surrounding each target gene TSS were scored against this matrix, and the sites falling within the 90th percentile were defined as GBSs. Importantly, the lowest-scoring site included in this set had a higher score than verified GBSs (data not shown) (3, 8), demonstrating that we likely included only sites that are recognized by GR in vitro. Scanning 69 dex-induced genes across 64-kb windows, we identified 325 GBSs, an average of 4.7 sites per gene (Table 1). This motif was detected at a similar or perhaps slightly lower frequency (3.5 sites per gene) at genes repressed by dex (Table 1). In addition, the GBS motif also occurred at a similar frequency near randomly chosen genes that were unresponsive to dex in C3H10T1/2 cells. Importantly, the GBS motif was similarly distributed across the 64-kb windows encompassing the TSSs at genes induced, repressed, and unresponsive to dex (data not shown), indicating that sequences potentially recognizable by GR are widely distributed in the genome.

Table 1.

Identification of GR targets and GR recognition motifs

Features	Responsive	Induced	Repressed	Unresponsive
No. of genes	86	69	17	40
Total no. of GBSs	384	325	59	181
Average no. of GBSs per gene	4.5	4.7	3.5	4.5
Frequency of GBS occurrence	14 kb	14 kb	18 kb	14 kb
Range of GBSs per gene	0–11	0–11	1–7	1–10
Total no. of conserved GBSs	107	93	14	37
Average no. of conserved GBSs per gene	1.2	1.4	0.8	0.9

Frequency of GBSs at genes activated, repressed, or unresponsive to dex. GBSs were identified by computational scanning of 64 kb surrounding the transcription start sites of the mouse genes shown in Table S1. Human aligned GBS sequences were obtained from Ensembl Genome Browser. GBSs that contain at least nine identical bases across the mouse and human genome are considered conserved, and any GBSs that contain internal gaps at positions 4–12 of the 15-bp motif are considered not conserved.

Sequence Conservation of GBSs Is Sufficient to Predict GR Occupancy at Dex-Induced Genes.

We showed previously that the precise sequences of GR-occupied GBSs within individual GREs at genes induced by dex are highly conserved across four mammalian species (human, mouse, rat, dog) (3). Based on this finding, we wished to assess whether GBS species conservation at dex-induced genes is sufficient to predict GR occupancy. We began with simple pairwise comparisons, examining mouse–human identity with no weighting bias introduced at any base position across the motif; the mouse–human evolutionary separation is ≈100 million years (9). We then assessed GR occupancy using chromatin immunoprecipitation (ChIP) at these sites in C3H10T1/2 cells. At the 69 genes induced by dex, 49 of them (Table S3) contained a conserved GBS motif (defined here as 9–15 bp identical between mouse and human). Whereas GR was bound at only 13 of the 39 (33%) weakly conserved sites (9–11 of 15 base pairs identical between mouse and human) (Table 2), we detected GR binding at 16 of the 34 (47%) moderately (12–13 identical bases) conserved sites and at 17 of the 20 (85%) highly conserved (14–15 identical bases) sites. In contrast, only four of the 76 (5%) tested nonconserved sites (<9 base identity) were bound by GR (Table 2). In summary, we found that mouse–human sequence identity of GBSs correlated directly with the likelihood of these sites being bound by GR (Fig. 1A). Thus, pairwise (mouse–human) sequence conservation of GBSs at GR-induced genes is sufficient to predict genomic sites occupied by GR in vivo.

Table 2.

GR occupancy at genes induced, unresponsive, or repressed by glucocorticoids

Number of mouse–human identical bases	Dex-induced genes, number of GR-occupied GBSs*	Dex-unresponsive genes, number of GR-occupied GBSs†	Dex-repressed genes, number of GR-occupied GBSs‡
15	7/7	0/1	0/0
14	10/13	0/3	0/4
13	9/14	1/4	0/1
12	7/20	0/5	0/1
11	6/15	1/10	0/1
10	5/16	0/10	0/2
9	2/8	0/4	0/5
<9	4/76	0/13	0/45

*GR occupancy of GBSs at dex-induced genes assessed by chromatin immunoprecipitation. DNA samples immunoprecipitated from C2H10T1/2 cells treated for 90 min with ethanol or dex were quantified with qPCR using specific primers spanning the corresponding GBS. Relative amplification was normalized to a genomic region near the Hsp70 gene, which does not bind GR.

^†GR binding at genes unrsponsive to dex. A set of genes unresponsive to dex in C3H10T1/2 cells (Table SI) was chosen at random.

^‡GR occupancy at genes repressed by dex.

Fig. 1.

Pair-wise sequence conservation of GBSs is sufficient to predict GR occupancy at dex-induced genes. (A) Direct correlation between sequence conservation of GBSs and GR occupancy at dex-induced genes. The number of identical bases of the GBSs was plotted against the percentage within that population that was occupied by GR in C3H10T1/2 cells. All of the conserved GBSs identified at genes induced by dex are represented in the graph. (B) GR occupancy of conserved and nonconserved GBSs at dex-induced genes. GR occupancy at all of the conserved GBSs was experimentally tested by chromatin immunoprecipitation. For the nonconserved GBSs, the number of GR-occupied sites was extrapolated from 76 tested sites (Table 2).

Of the total 325 GBSs identified at 69 genes up-regulated by dex, we found 232 sites (72%) that were not conserved (<9 bp identity between mouse and human) (Table 1, Fig. 1B). Extrapolation from the tested sites suggested that only 12 (5%) would be GR bound in C3H10T1/2. Assessing the 93 conserved sites (≥9-bp identity between mouse and human) identified near these genes, we correctly predicted GR occupancy at 46 (50%) (Table 2, Fig. 1B); those 46 occupied sites were associated with 31 (45%) of the 69 genes induced by dex. Notably, prediction rate is a direct function of sequence conservation stringency: we predicted GR occupancy at ≈80% of GBSs with >12-bp mouse–human identity and at 100% of GBSs with full identity between the two species. Thus, by applying computational and conservation analysis, we could predict precise 15-bp positions bound by GR within the 64,000-bp regions sampled. Of note, both GR-occupied and nonoccupied GBSs (Table S3) were distributed evenly upstream and downstream from the associated TSSs (data not shown), suggesting that GR occupancy cannot be predicted based on the location of the GBS motif and further highlights the significance of our analysis. Our analysis identified both known GBSs, such as those near the promoters of Sgk and Mt2 (3, 10), and previously uncharacterized sites, such as those near Peli1, Tob2, Bteb1, and Bcl2l1 (Fig. 2A,Fig. 2B).

Fig. 2.

Identification of known and previously uncharacterized GR-occupied GBSs. (A) Examples of conserved GBSs bound by GR in C3H10T1/2 cells. GBSs found near the indicated genes are shown, together with corresponding human sequences obtained from the Ensembl Genome Browser. Displayed in red letters are bases that are identical in the mouse and human genomes. (B) Examples of GR occupancy of GBSs in C3H10T1/2 cells. ChIP experiments were performed and analyzed as indicated in Table 2. The data represent an average of at least three independent experiments and are plotted with standard error of mean. The genes shown correspond to the dex-induced transcript nearest the corresponding GBS.

GR-Occupied GBSs Are Functional in Reporter Assays.

We subcloned 500-bp DNA fragments encompassing our GR-occupied GBSs into a luciferase reporter vector to assess their potential to function in cells as glucocorticoid response elements. In C3H10T1/2 cells transfected with these reporters, luciferase activity was stimulated >2-fold for 9 of the 11 reporters tested upon treatment with dex (Fig. 3). To confirm that GR was indeed recognizing the 15-bp GBSs predicted in our computational analysis, we mutated highly constrained positions within GBS motif at the nine dex-responsive reporters. In each case, the dex-stimulated luciferase response was compromised in the mutants (Fig. 3), indicating that our computational analysis was indeed identifying GR-bound GBSs that could confer glucocorticoid responsiveness.

Fig. 3.

GR-occupied GBSs are functional in C3H10T1/2 cells. Cells transfected with reporters each bearing a single copy of a 500-bp GBS-containing fragment were treated with ethanol or dex, harvested, and the activity of the firefly luciferase reporter was normalized to renilla luciferase activity. Displayed is the ratio of relative light units in log scale of the reporters treated with dex or ethanol, averaged over at least four independent experiments. Black bars represent the reporters with wild-type GBSs (showing standard error of mean), whereas the gray bars correspond to mutant reporters, in which the GBSs were mutated to sequences that abrogate GR binding (see Materials and Methods and Table S6).

GBS Conservation and Occupancy at Genes Unresponsive to Glucocorticoids.

We showed previously that GR occupancy in a given cell type appears to be strongly restricted to genes that are actually glucocorticoid responsive within those cells (3). Consistent with this finding, we detected bound GR at only 2 of 37 (5%) conserved GBSs in C3H10T1/2 at genes not regulated by dex in those cells and at none of the 13 nonconserved sites tested (of 144 total) (Table 2 and Table S4). Recall, in contrast, that GR occupied conserved sites at 45% of the dex-induced genes. The two occupied GBSs at the unresponsive genes are nevertheless notable, however, as they appear to represent the relatively rare cases in which events subsequent to receptor binding may serve as primary determinants of glucocorticoid responsiveness.

Distinct GBS Occupancy at GR-Induced vs. -Repressed Genes.

GR can either activate or repress gene transcription. We found that GBSs occurred at similar frequencies at these two classes of genes (Table 1), and that mouse GBSs were conserved in the human genome at only slightly lower rates at repressed relative to induced genes. To our surprise, we failed to detect GR binding at any GBSs associated with GR repressed genes (Table 2 and Table S5), whereas at least 45% of the induced genes contained GR-bound GBSs. Thus, there appear to be at least three constraints on GR occupancy of GBS motifs: first, GR generally fails to bind at GBSs associated with genes that are not GR regulated in a given cell context; second, GR typically does not occupy GBSs that are not well conserved at dex-induced genes; and third, GR binding is not detected even at conserved GBSs linked to GR repressed genes. We conclude that GBS conservation is a strong predictor of GR occupancy and function at GR induced genes, but that GBS alone is not sufficient to specify GR occupancy.

Discussion

In this study, we examined whether sequence conservation of GBSs across two mammalian species is sufficient to predict GR-occupied GREs. We first identified glucocorticoid responsive genes in mouse C3H10T1/2 cells (Table S1). We then inferred sites recognizable by GR at these genes using previously defined GBSs to derive a receptor-binding motif positional weight matrix. Next, we determined the level of sequence conservation of these sites between mouse and human and tested GR occupancy using ChIP. Strikingly, we found that the level of conservation of these sites correlated directly with the predictability of these sites being occupied by GR at dex-induced genes (Fig. 1A).

Using sequence conservation analysis, we predicted sites of GR occupancy at nearly half of the genes induced by dex. The remaining genes may be secondary targets, have GR-occupied GBSs outside of the 64 kb examined here, or may have GR-binding sequences that do not conform to the motif applied in this study. Alternatively, these genes may be bound by GR at poorly conserved GBSs, which comprise only ≈20% (12 of 58 GBSs) of GR-occupied sites (Fig. 1B). Similarly, it was observed that 10–14% sites bound by a regulatory factor within a particular yeast species were not conserved in binding or sequence in orthologous yeast genomes (11). However, it should be noted that we restricted our analysis to positionally conserved sites, and that these sites found in the mouse genome may actually be conserved in sequence but shifted location within the human genome. Indeed, occupancy of some regulatory factors has been shown to shift in location between mouse and human genomes (12). Moreover, GBSs that are not conserved between mouse and human may be conserved in genomes of other species, and extending our study beyond the current pairwise analysis may be informative. Finally, some nonconserved sites may reflect selected changes that enable species evolution.

Among the genes that are glucocorticoid-unresponsive in C3H10T1/2 cells, we expect a subset of GBSs found at these genes to be responsive in other cell contexts and the remainder to be unresponsive under any conditions. We predict that some of the 37 conserved GBSs at genes that are not bound and unresponsive in C3H10T1/2 cells may in fact be occupied and functional in other cell contexts. In contrast, GBS motifs that are not conserved at unresponsive genes are unlikely to be GR-bound and functional in any context. For the conserved GBSs not occupied by GR, it is possible that occupancy and functionality may be precluded by their native chromatin context or absence of other regulatory factors. If the former were the primary determinant, we would expect that these conserved GBSs would be GR-occupied and responsive to glucocorticoids in heterologous nonchromatinized reporters.

We found that GBSs occurred at similar frequencies at glucocorticoid-induced and repressed genes in C3H10T1/2 cells. However, none of the GBSs at dex-repressed genes were bound by the receptor, indicating that GR occupancy at these sites is specific to dex-induced genes and precluded from repressed genes. GR has been demonstrated to be an active repressor (6); thus, our study suggests that GR may occupy specific sites with distinct sequences to elicit transcription repression. Consistent with this notion, repression of POMC (13, 14) and osteocalcin (15) by glucocorticoids is directed by binding of GR to DNA sequences distinct from the GBS motif applied in this study. In addition, GR tethers through protein–protein interactions to DNA-bound nonreceptor factors, such as NFκB and AP-1, to repress transcription of genes such as interleukin-8 (6) and collagenase (16), respectively. It will be interesting to examine whether GR indirectly binds to DNA through interaction with these factors at the dex-repressed genes identified in this study. Thus, the GBS motif examined here appears to specify transcriptional activation by GR but not repression; moreover, GR fails to occupy the motif close to GR repressed genes.

In general, we found that GR occupancy is remarkably site-selective. This is despite the fact that GBSs are widespread across the mouse genome, as predicted by the rather modest conservation constraints across the 15-mer. Hence, GBSs occurred at comparable frequencies near genes activated, repressed, or unresponsive to dex, yet bound GR was detected at these sites only at activated genes. For example, precisely the same GBS (GGAACAGAATGTTCA) appears at the activated Dusp1 and unresponsive Adamts19 genes, but it is conserved and occupied by GR only at the former gene (Tables S3 and S4). Indeed, we observed selective GR occupancy even at GBSs associated with activated genes. The same GBS (AGAACAGTCTGTGCT) is present at Il6ra and Adm genes (data not shown), both of which are induced by dex in C3H10T1/2 cells, but is occupied only by the receptor at the Il6ra gene. These and other findings suggest strongly the presence of additional elements that either promote or inhibit GR binding at GBSs in particular contexts and in a manner that also relates to GBS conservation. In the simplest case, such elements might be proximal to the GBSs themselves. Indeed, for 50 GREs examined in another study (3), we found conserved regions of up to 2 kb that surround GR-occupied GBSs and identified enriched motifs resembling non-GR factor-binding sites within the GREs; it is tempting to speculate that those sites may be important for directing receptor occupancy at specific GBSs, and it will be interesting to determine whether the nonconserved GBSs lack such sequence signatures.

In this study, we were able to predict GR-occupied GREs by exploiting sequence conservation of the GBSs with the constrained and flexible base positions weighted equally, consistent with our previous finding that individual GBSs are strongly conserved across the full 15-bp motif (1). Structural studies have determined that GR makes specific contacts with as few as four bases of the 15-bp GBS (17). Thus, it is tempting to speculate that the receptor-binding sites are not merely docking sites for GR but rather serve additional functions. Because GR regulates transcription in a highly gene-specific manner (18), the precise sequence of the receptor-binding site may harbor a “regulatory code” that confer allosteric effects on GR (7) to direct such gene specificity.

Materials and Methods

Cell Culture, Plasmids, and Reporter Analysis.

C3H10T1/2 cells were grown in DMEM with 10% FBS, pyruvate, penicillin, and streptomycin in 5% carbon dioxide atmosphere. For all experiments, the cells were supplemented with 1 μg/ml of insulin upon treatment with ethanol or dex (1 μM). For reporter analysis, inserts were cloned into PGL4.10 E4TATA plasmid using KpnI and XhoI sites and GBS mutagenesis was carried out by using QuikChange (Stratagene) (see Table S6 for primer sequences). ≈200,000 C3H10T1/2 cells were transfected with 200 ng of the reporter constructs and 200 ng of pRL Luc (Promega) using Amaxa 96-well shuttle with SEC solution and 96-CM-137 program (Amaxa Biosystems). Three hours after transfection, cells were treated with EtOH vehicle (0.1%) or 100 nM dex for 6–8 h and harvested, and luciferase activity was measured as described for the dual luciferase reporter system (Promega) using a Tecan Ultra Evolution plate reader (Tecan).

RNA Isolation, Reverse Transcription, Quantitative PCR (qPCR), and Microarray Analysis.

The RNA isolation, reverse transcription, and qPCR steps were performed as described (18). For the microarray analysis, we exposed C3H10T1/2 cells with dex or DMSO for 90 min in triplicate and hybridized the transcribed cDNA to arrayed MEEBO (whole-genome mouse oligo set, Invitrogen) oligonucleotide probes. We analyzed the arrays using the limma package in BioConductor (19). Primers for cDNA amplification using qPCR are displayed in Table S1.

Computational and Conservation Analysis to Identify Putative GR-Binding Sites.

The GBS positional weight matrix was derived by using GBSs identified in GR ChIP-chip experiments (3). Briefly, the GBSs were extracted from the GREs by using BioProspector; BioProspector analysis was performed by using nucleotide widths (w) 14 and 15. The GBSs found in the top-scored motifs and the second top-scored motif generated using w14 (Table S2) were subsequently used for the positional weight matrix. BioPerl and the TFBS module (20) were used to scan for GBS motifs within 64-kb DNA segments surrounding the TSS of the closest glucocorticoid-responsive gene. All sites that scored within the top 90th percentile were considered GBSs. Ensembl compara API (Ensembl Release 45) (21) was used to extract the aligned human sequences for GBSs.

ChIP.

ChIP assays were performed as described (6). Briefly, cells were formaldehyde cross-linked, the reaction stopped with glycine, chromatin sheared with sonication, and chromatin samples immunoprecipitated with 8 μg of N499 anti-GR antibody per 15-cm plate of cells with Protein A/G beads (Santa Cruz Biotechnology). After washing the beads, chromatin samples were extracted once with phenol-chloroform and purified by using a Qiaquick column (Qiagen). The relative amplification of dex- vs. ethanol-treated samples at each genomic site was assessed by using qPCR (Table S3–S5). We considered genomic sites GR-occupied GBSs when we observed at least a 2-fold enrichment in dex-treated samples (averaged over at least three independent experiments). The primers used for amplifying the ChIP samples are displayed in Tables S3–S5.