Genomewide analysis of Drosophila GAGA factor target genes reveals context-dependent DNA binding
Abstract
The association of sequence-specific DNA-binding factors with their cognate target sequences in vivo depends on the local molecular context, yet this context is poorly understood. To address this issue, we have performed genomewide mapping of in vivo target genes of Drosophila GAGA factor (GAF). The resulting list of ≈250 target genes indicates that GAF regulates many cellular pathways. We applied unbiased motif-based regression analysis to identify the sequence context that determines GAF binding. Our results confirm that GAF selectively associates with (GA)n repeat elements in vivo. GAF binding occurs in upstream regulatory regions, but less in downstream regions. Surprisingly, GAF binds abundantly to introns but is virtually absent from exons, even though the density of (GA)n is roughly the same. Intron binding occurs equally frequently in last introns compared with first introns, suggesting that GAF may not only regulate transcription initiation, but possibly also elongation. We provide evidence for cooperative binding of GAF to closely spaced (GA)n elements and explain the lack of GAF binding to exons by the absence of such closely spaced GA repeats. Our approach for revealing determinants of context-dependent DNA binding will be applicable to many other transcription factors.
Sign up for PNAS alerts.
Get alerts for new articles, or get an alert when an article is cited.
Transcription factors control gene expression patterns by binding to specific sequence elements in regulatory regions of the genome. The sequence specificity of a transcription factor is often inferred from in vitro experiments, but in vitro specificity is an unreliable predictor of in vivo binding (reviewed in ref. 1). In many cases, in vivo association of a transcription factor with its consensus sequence is strongly influenced by the presence or absence of other factors and by the local chromatin structure. This interplay between local molecular context and the binding of a transcription factor is still poorly understood. Here, we describe how a combination of large-scale mapping of target genes and bioinformatics approaches can reveal several aspects of the molecular context that determine the target specificity of a DNA-binding protein.
Drosophila GAGA factor (GAF) is a sequence-specific DNA-binding factor with several functions. Mutations in the GAF-encoding Trl gene affect viability and display distinct developmental phenotypes (2). GAF is involved in both gene activation (3–6) and gene repression (7, 8) and plays a role in the modulation of chromatin structure (4, 9) and mitotic chromosome segregation (10).
In vitro, GAF binds to the sequence (GA)n, with optimal binding requiring at least 2.5 GA repeats (11, 12). In agreement with this, the solution structure of the DNA-binding domain of GAF complexed to a GAGAG-containing DNA element shows contact sites of the protein with all five base pairs (11). Mutational analysis has confirmed that (GA)n motifs are necessary for GAF antirepressor activity (13). Studies of GAF binding in the native chromatin context have identified a number of in vivo GAF target sequences that indeed contain (GA)n elements (14–16). Taken together, these data strongly argue that GAF binds to (GA)n sequences.
In the sequenced portion of the Drosophila genome, GAGAG elements occur on average once every 652 bp (data not shown), which would predict that virtually every gene has several molecules of GAF bound in its immediate vicinity. However, staining of larval salivary gland polytene chromosomes with GAF-specific antibodies shows a clear banded pattern (13, 17). Thus, GAF is unlikely to bind to every GAGAG element in the genome. The observation that GAF binds to the AAGAG satellite repeat only during mitosis (18) further suggests that GAF binding can be modulated by local molecular features, the nature of which is unknown.
Here, we report the large-scale identification of in vivo GAF target loci in the Drosophila genome. We measured binding of GAF to thousands of loci by using the recently described “chromatin profiling” approach (16). We expressed a fusion protein consisting of GAF linked to Dam methyltransferase in Drosophila Kc cells and subsequently used a DNA microarray-based method to detect the resulting GAF-directed adenine methylation pattern. When corrected for the methylation pattern obtained with untethered Dam, this GAF-directed methylation pattern reflects the in vivo binding pattern of GAF (16, 19).
We previously reported that methylation by tethered Dam spreads in cis over 2–5 kb from a protein binding sequence (19). On the one hand, this limits the mapping resolution of the chromatin profiling technique to a few kb. On the other hand, it allows for the use of conventional cDNA arrays to detect binding of proteins to upstream and downstream regulatory sequences, provided that the binding sites are located within the methylation spreading distance from transcribed regions. As we demonstrate below, unbiased bioinformatics analysis of such binding profiles can be used to uncover some of the rules that govern context-dependent binding of transcription factors.
Materials and Methods
Chromatin Profiling Experiments.
Chromatin profiling of GAF was performed as described (16) by using spotted microarrays containing the Drosophila Gene Collection (release 1) (20) and 430 additional cDNA and genomic fragments. All measured ratios were log2-transformed and normalized to the median value of the entire array. Data from three independent experiments (one with reversed dye orientation) were averaged. A total of 331 cDNA and genomic DNA fragments that were spotted in duplicate on the arrays showed a high correlation between the two spots (r = 0.97; mean difference between the two spots 0.06 ± 0.07), further confirming the accuracy of our measurements. To test whether log ratios were significantly different from 0, we used the cyber-t algorithm (21), followed by a correction for multiple testing (22), setting the estimated false discovery rate to 0.05.
Construction of Sequence Files.
EST and genomic sequences were obtained from the Berkeley Drosophila Genome Project (BDGP), release 2. For 5,459 ESTs we were able to identify unique matching genomic regions (megablast against the BDGP database). For each of these, the precise chromosomal coordinates of the 5′ and 3′ boundaries of the matching region were determined. reduce and GAGAG spacing analyses were restricted to microarray data obtained from 4,402 ESTs that matched to genomic regions <10 kb in size for which at least 10-kb upstream and downstream flanking sequence could be obtained. Coordinates of introns, exons, and nontranscribed sequences were obtained from BDGP genome annotation files. Perl scripts that were written for this purpose are available on request.
reduce Analysis and Analysis of GAGAG Spacing.
The sequences of the probed loci (optionally including flanking sequence on both sides, as well as the sequence of the introns, exons, or intergenic regions they contain were determined by using the Berkeley Drosophila Genome Project whole-genome sequence and annotation (GFF) files (release 2) and dedicated Perl scripts. reduce analysis was performed as described (23) by using software available at http://bussemaker.bio.columbia.edu/reduce (see also Supporting Text, which is published as supporting information on the PNAS web site, www.pnas.org). In the linear model underlying reduce, each occurrence of a given motif is assumed to contribute equally to the localization of GAF at a given locus; the coefficients in the model are determined by performing a least-squares fit to the chromatin profiling log ratios. All motifs from a large class are scored based on how much the fit to the data would improve by their inclusion in the model.
Results
Identification of ≈250 Target Loci of GAF.
Previously, we reported the use of the DamID chromatin profiling approach to screen ≈300 Drosophila genes for GAF binding (16). Here, we extended this approach to >6,000 genes. Briefly, a fusion protein consisting of Dam methyltransferase and the 519-aa isoform (17) of GAF was expressed in Drosophila Kc cells. This leads to preferential methylation of GAF binding sites in the genome. Methylated genomic DNA fragments were purified, fluorescently labeled, and used to probe microarrays containing 6,280 unique cDNA fragments. Methylated DNA purified from cells expressing unfused Dam was labeled with a different fluorochrome and used as a reference probe. Normalized fluorescence ratios represent the targeted/untargeted methylation ratios, and therefore the relative GAF binding to the probed loci (16).
Because we used cDNA microarrays in our assay, we were only able to directly measure methylation levels in exons in the probed loci. However, targeted methylation “spreads” in cis over ≈2–5 kb (19). Binding of GAF-Dam to upstream, downstream, and intronic sequences may therefore be detected as increased methylation of the nearby exon sequences, provided that the GAF binding sites are located within a few kilobases from a probed exon. We define a “target gene” as a gene for which the corresponding cDNA probe on the array detects a significantly elevated GAF-Dam/Dam methylation ratio (Fig. 1, Table 3, which is published as supporting information on the PNAS web site, and Materials and Methods).
Figure 1
At an estimated false discovery rate of 0.05, we identified 262 cDNA probes for which GAF-targeted methylation levels were significantly elevated (Table 4, which is published as supporting information on the PNAS web site). Of these, 219 probes corresponded to 208 unique previously annotated genes and three repetitive elements (some genes were represented by two cDNAs). The remaining 43 positive cDNAs matched genomic loci that had not been annotated. Importantly, the GAF binding patterns are strikingly different from the binding patterns of six other Drosophila proteins [HP1, HP1c, Su(var)3–9, dMyc, dMad, and an ortholog of mammalian Max; B.v.S., F. Greil, J.D., A. Orian, and R. Eisenman, unpublished work], supporting the specificity of the chromatin profiling technique.
The identified GAF target genes appear to cover a broad variety of functions and include genes that encode proteins involved in growth and development, signaling, heat shock response, and metabolic pathways (Table 1). Thus, GAF may regulate a wide range of cellular processes and pathways. Our data confirm the previously reported GAF binding to heat shock protein genes (14), but not the reported very weak binding to the 28S rDNA and histone gene loci (14). Although for other individual loci in our list of target genes the binding of GAF may need to be confirmed by independent methods such as chromatin immunoprecipitation (14), global analysis of our GAF binding patterns as presented below strongly supports the specificity of our mapping technique.
Table 1
| Gene Ontology category | GAF target genes in category |
|---|---|
| Biological processes | |
| Amino acid metabolism | SelD, Gad1, slgA |
| Catabolism | Gad1, eff, slgA |
| Cell communication | PGRP-LA, Dad, grk |
| Cell growth and/or maintenance | Fer1HCH, Cdk4, stg, PGRP-LA, LamC, SelD, Gad1, eff, ash2, Alhambra, slgA, chic |
| Defense response | PGRP-LA, CG3829, spz, Myd88 |
| Developmental processes | mfas, grk, fray |
| Heat shock response | Hsp22, Hsp26, Hsp27 |
| Imaginal disc growth factor | Idgf2, Idgf3, Idgf1 |
| Ligand binding or carrier | Fer1HCH, Nrv2, CG6783, Ras85D, dome, Cyt-c-d, chic |
| Mitosis | LamC, eff, eIF-4E, stg |
| Protein amino acid dephosphorylation | CG11597, stg, PP2A–B′, dome, puc |
| Protein amino acid phosphorylation | Cdk4, par-1, for, par-1, Pak3, gish, fray, CG10967, for, Lk6 |
| Signal transduction | Dad, grk, Myd88 |
| Molecular functions | |
| Actin binding | cpb, bif, chic |
| DNA binding | Eli, Tis11, E2f, D1, HmgD, Alhambra, Rbf |
| Establishment/maintenance of chromatin architecture | HmgD, ash2, eIF-4E |
| Glutathione transferase | CG17531, CG17533, CG5224, CG1681 |
| Heat shock protein | Hsp22, Hsp26, Hsp27 |
| Hydrolase | Nrv2, CG11597, Idgf3, Faa, PP2A–B′, CG8689, Ras85D, aay, puc, CG9026, ESTS:172F5T, CG10992 |
| Hydrolase, acting on glycosyl bonds | Idgf3, CG8689, ESTS:172F5T |
| Integral plasma membrane protein | Nrv2, PGRP-LA, CG3829 |
| Ion transporter | Nrv2, BcDNA:LD28120, Sip1 |
| Kinase | SelD, CG1216, CG4798 |
| Ligand | Idgf3, grk, spz |
| Oxidoreductase | Gs1, BEST:CK02318, Mdh, CG6199, slgA, CG15093, Sptr |
| Oxidoreductase CH-OH group of donors | Mdh, CG15093, Sptr |
| Phosphatase | CG11597, PP2A–B′, aay, puc, stg, dome |
| Protein serine/threonine kinase | Cdk4, for, par-1, fray, CG10967, for, Lk6 |
| Serpin | sp1, Spn43Aa, sp4 |
| Signal transducer | Idgf3, dome, grk, spz |
| Transcription factor | E2f, E11, CG11799, CG11867, Iola, NK7.1 |
| Transferase | CG1140, Ugt35a, CG8782, SelD, CG5431, CG1681, CG1216, CG4798, Ugt35b |
| Translation initiation factor | eIF-4a, eIF-4E, Syx1A |
| Transporter | Nrv2, BcDNA:LD28120, Sip1 |
| UDP-glucuronosyltransferase | Ugt35a, Ugt86Da, Ugt35b |
Standardized gene annotations were taken from the Drosophila Gene Ontology database (release February 2002). A complete list of target genes can be found in Table 4.
reduce Analysis Identifies in Vivo Binding Motifs of GAF.
To confirm the in vivo binding sequence of GAF, we used an unbiased bioinformatics method. reduce is a motif-based regression analysis method originally designed for the discovery of regulatory elements based on microarray expression data (23). Here, we applied the same algorithm to find sequence motifs whose occurrence correlates with the chromatin profiling data for GAF. A major advantage of reduce is that it analyzes the entire set of probed loci and does not rely on clustering or prior partitioning into “target” and “nontarget” loci. Instead, reduce uses the full quantitative dataset obtained from one or more chromatin profiling experiments. Moreover, the output of reduce includes statistical parameters that indicate the correlation strength (represented as a t value) and statistical significance (P value) for each sequence motif, taking into account corrections because of the parallel testing of many motifs.
To account for the cis-spreading of targeted methylation, we performed reduce by using the sequences of the genomic regions corresponding to the cDNAs on the microarray, including introns and 2 kb of flanking sequence added to both the 5′ and 3′ ends (Fig. 2A). Thus, we made no prior assumptions on the location of GAF binding sites relative to the transcribed regions, but instead analyzed the complete genomic regions where binding of GAF-Dam would in principle be detectable.
Figure 2
For all possible sequence motifs up to 7 nt we tested whether their occurrence in the probed genomic regions correlates with GAF binding. The results show that (GA)n repeats indeed are strongly correlated with GAF binding (Table 2 and Table 5, which is published as supporting information on the PNAS web site). When ranked by correlation, all of the top 20 motifs contain (GA)n repeats. This finding demonstrates the specificity of our chromatin profiling technique and confirms that GAF binds selectively to (GA)n motifs in the native chromatin context.
Table 2
| Rank | Motif | r 2 | t | −log10 (P value) | Matches | Loci with match |
|---|---|---|---|---|---|---|
| 1 | AGAGAG | 0.08411 | 19.646 | >16 | 6,958 | 2,677 |
| 2 | GAGAG | 0.08353 | 19.572 | >16 | 22,454 | 4,084 |
| 3 | AGAGA | 0.07613 | 18.61 | >16 | 25,640 | 4,120 |
| 4 | GAGAGA | 0.07043 | 17.846 | >16 | 6,675 | 2,720 |
| 8 | AAGAGAG | 0.05598 | 15.788 | >16 | 1,940 | 1,384 |
| 11 | GAGAGAG | 0.05345 | 15.406 | >16 | 2,762 | 1,217 |
| 13 | AGAGAGC | 0.04817 | 14.585 | >16 | 1,848 | 1,350 |
| 14 | AGAGAGA | 0.04673 | 14.353 | >16 | 2,876 | 1,309 |
| 16 | AAGAGA | 0.04087 | 13.382 | >16 | 7,993 | 3,373 |
| 17 | GAGAGAA | 0.03944 | 13.137 | >16 | 1,731 | 1,333 |
| 22 | AAAGAGA | 0.03592 | 12.514 | >16 | 3,054 | 2,023 |
| 26 | GAGAGCG | 0.03245 | 11.872 | >16 | 2,125 | 1,528 |
| 28 | AGAGAGT | 0.03229 | 11.842 | >16 | 1,178 | 981 |
| 35 | AGAGAA | 0.03012 | 11.424 | 15.5 | 8,221 | 3,418 |
| 36 | GAGAGC | 0.02949 | 11.302 | 15.2 | 7,275 | 3,269 |
| 42 | CGAGAGA | 0.02823 | 11.051 | 14.6 | 1,268 | 1,055 |
| 5 | CTCTC | 0.06802 | 17.514 | >16 | 21,401 | 4,062 |
| 6 | TCTCT | 0.0619 | 16.653 | >16 | 24,773 | 4,115 |
| 7 | CTCTCT | 0.06175 | 16.632 | >16 | 7,068 | 2,711 |
| 9 | CTCTCTT | 0.05526 | 15.68 | >16 | 1,861 | 1,339 |
| 10 | GCTCTCT | 0.05501 | 15.641 | >16 | 1,772 | 1,307 |
| 12 | TCTCTC | 0.04922 | 14.751 | >16 | 6,624 | 2,665 |
| 15 | TCTCTT | 0.04443 | 13.979 | >16 | 7,390 | 3,298 |
| 18 | CTCTCTC | 0.03852 | 12.976 | >16 | 2,742 | 1,150 |
| 19 | CTCT | 0.03762 | 12.818 | >16 | 88,116 | 4,205 |
| 20 | GCTCTC | 0.03709 | 12.723 | >16 | 6,627 | 3,103 |
| 23 | CGCTCTC | 0.03579 | 12.49 | >16 | 1,858 | 1,325 |
| 31 | TCTCTTT | 0.03069 | 11.537 | 15.7 | 2,923 | 1,970 |
| 33 | TCTCTCT | 0.03017 | 11.435 | 15.5 | 2,971 | 1,300 |
| 34 | CTCTTT | 0.03017 | 11.435 | 15.5 | 8,940 | 3,540 |
| 44 | TTCTCT | 0.02798 | 11 | 14.4 | 7,486 | 3,313 |
| 46 | CGCTCT | 0.0274 | 10.882 | 14.2 | 5,904 | 2,986 |
| 30 | AAACAA | 0.03103 | 11.601 | 15.9 | 25,931 | 4,137 |
| 45 | AACAAA | 0.02752 | 10.906 | 14.2 | 25,282 | 4,122 |
| 49 | ACAAA | 0.02648 | 10.693 | 13.7 | 61,740 | 4,202 |
| 57 | AAAACAA | 0.02563 | 10.515 | 13.3 | 10,584 | 3,556 |
| 24 | TTGTT | 0.03248 | 11.878 | >16 | 57,441 | 4,204 |
| 27 | TTGTTT | 0.03232 | 11.847 | >16 | 23,214 | 4,121 |
| 43 | TGTTT | 0.02816 | 11.036 | 14.5 | 58,404 | 4,202 |
| 50 | TTTGT | 0.02648 | 10.692 | 13.7 | 59,213 | 4,203 |
| 52 | TTTGTT | 0.02636 | 10.667 | 13.6 | 22,839 | 4,111 |
| 32 | TACATA | 0.03065 | 11.528 | 15.7 | 14,749 | 3,790 |
| 40 | ATACATA | 0.02898 | 11.201 | 14.9 | 6,679 | 2,766 |
| 54 | ATACA | 0.02587 | 10.565 | 13.4 | 37,679 | 4,191 |
| 58 | ACATAT | 0.02559 | 10.507 | 13.3 | 13,943 | 3,864 |
| 39 | TATGTA | 0.02907 | 11.217 | 14.9 | 14,268 | 3,822 |
| 53 | ATATGTA | 0.02605 | 10.603 | 13.5 | 6,207 | 2,891 |
A variety of (GA)n motifs displayed highly significant correlation (P < 10−12) with GAF binding, but only if n ≥ 2. No significant correlation was found for the trinucleotide motif GAG (P = 0.2), which was previously reported to bind GAF in vitro (24). Thus, in the native chromatin context, at least two GA repeats are necessary for GAF recruitment, and 2.5 or 3 repeats appear to be optimal. Interestingly, the reduce algorithm found roughly equal correlation values for (GA)n and (CT)n motifs. This finding demonstrates that GAF binds with approximately the same frequency in either orientation relative to the direction of transcription.
A few additional motifs display a weaker but highly significant correlation with GAF binding, in particular several variants of TnGTm or its reverse complement AnCAm, and (AT)nACA(TA)m or its reverse complement (TA)nTGT(AT)m (Table 2). Given the high specificity of GAF for (GA)n repeats in vitro, these unrelated motifs are unlikely to recruit GAF directly. Rather, we speculate that these motifs bind other sequence-specific factors that provide a favorable context for GAF binding to (GA)n motifs.
Although the overall correlation between GAF binding and the occurrence of (GA)n elements is highly significant, it is far from perfect. For example, the number of GAGAG matches in a probed locus (in both orientations) and the GAF-binding log ratio only correlate with r = 0.34. In part, this imperfect correlation may be attributed to some random noise in our GAF mapping data. In addition, because a variety of (GA)n motif variants and a few other motifs correlate with GAF binding, each individual motif contributes to GAF binding to a limited extent. Below we will demonstrate that another explanation lies in the fact that specific regions in the genome, such as exons and 3′ downstream regions, bind GAF poorly, even though (GA)n elements occur with approximately the same frequency in these regions.
GAF Binds to Nontranscribed and Transcribed Regions.
Because our previous estimate of cis-spreading of targeted methylation (19) was of limited accuracy, we tested how the correlation between GAF binding and the presence of (GA)n elements was affected by including more or less flanking sequence in the reduce analysis. Throughout the remainder of this article, we will focus on the interaction of GAF with the motif GAGAG, which is the minimal high-affinity binding motif in vitro (11). This motif was one of the highest ranking motifs in the reduce analysis (Table 2). We use the t statistic tGAF:GAGAG of the Pearson correlation between the number of occurrences of GAGAG in the sequence and the observed GAF binding.
The results (Fig. 2 B and C) show that inclusion of ≈2 kb of flanking sequence results into a maximum value for tGAF:GAGAG. This finding indicates that GAF associates with GAGAG elements that are located upstream or downstream of the probed exons. Addition of >2 kb of flanking sequence leads to a weaker correlation, presumably because binding of GAF to sites >2 kb away from the probed regions does not add significantly to the methylation levels of the probed exons. This finding is in agreement with our previous estimate of 2- to 5-kb cis-spreading of targeted methylation (19).
Strikingly, if flanking sequences are left out completely, tGAF:GAGAG is still highly significant. This finding demonstrates that there is considerable binding of GAF to GAGAG elements within transcribed regions.
GAF Binding to Nontranscribed Regions.
To study the interaction of GAF with nontranscribed regions in more detail, we first determined the correlation between GAF binding and the occurrence of GAGAG elements in predicted nontranscribed sequences located within 2 kb of the probed loci (as indicated in Fig.2A). As expected, we found a highly significant correlation tGAF:GAGAG in these nontranscribed regions (Fig. 3 Left).
Figure 3
We then investigated whether GAF binds preferentially to upstream or downstream nontranscribed regions by comparing the respective contributions of these regions to the observed correlation. We separated all intergenic regions within 2 kb from probed loci into three categories: between two divergent genes (exclusively upstream), between two convergent genes (exclusively downstream), and between two tandem genes (mixed upstream/downstream). For each category we determined the value of tGAF:GAGAG (Fig. 3). The results show that GAF preferentially associates with GAGAG elements in upstream intergenic regions, compared with downstream intergenic regions (Δt = 6.9; P = 5 × 10−12). As expected, the category of regions that can be regarded as both upstream and downstream (“tandem”) showed intermediate levels of correlation.
It is important to note that the observed correlations may be interpreted as an indication of the relative average binding of GAF per GAGAG element. Our observations therefore imply that GAGAG elements in downstream regions are occupied less frequently than in upstream regions. Because upstream and downstream noncoding regions harbor GAGAG elements at almost equal density (Fig. 3 Right), we conclude that GAF preferentially binds to upstream regions.
GAF Is Excluded from Exons.
Using the same approach, we tested whether GAF preferentially binds to GAGAG elements located in introns or exons (Fig. 3). Strikingly, we found a clear correlation between GAF binding and the occurrence of GAGAG in introns, yet no such correlation was detectable in exons (tGAF:GAGAG(exon only) = −0.3). Thus, GAF binds significantly to GAGAG elements in introns, yet fails to interact with GAGAG elements in exons. A more detailed multivariate analysis suggests that this exclusion from exons is particularly strong in relatively long exons (see Table 6, which is published as supporting information on the PNAS web site).
GAF Binds to Introns Throughout Transcribed Regions.
GAF is often bound near promoter regions, where it can facilitate initiation of transcription (25). Enhancer elements can be located within introns, and it is therefore possible that GAF associated with introns facilitates transcriptional initiation. However, one report has suggested that GAF binds in some genes throughout the transcribed region and perhaps may control transcript elongation (14). If intron-associated GAF plays a role in elongation, then it may be expected that GAF binds to introns irrespective of the distance to the promoter. We tested this by comparing the binding of GAF to all first and last introns of the probed loci (Fig. 3). Strikingly, the results show that the value of tGAF:GAGAG in last introns is at least as high as in first introns. This finding indicates that GAF binding to introns is not limited to promoter-proximal introns, which is in agreement with a role for GAGA factor in transcript elongation.
GAF Binds Preferentially to Closely Spaced Pairs of GAGAG Motifs.
The striking difference in GAF binding between exons and introns argued that the association of GAF with GAGAG is modulated by additional molecular cues. In theory, GAF binding could be either selectively inhibited in exons (for example, by a chromatin folding rendering GAGAG elements inaccessible) or selectively enhanced in introns and upstream intergenic regions (by cooperative interactions). In vitro, GAF is able to form oligomeric complexes and displays cooperative binding to closely spaced (GA)n elements (26, 27). We therefore investigated whether such cooperative binding could explain the observed regional differences in GAF binding.
To test whether GAF preferentially binds to clustered GAGAG elements in vivo, we ranked loci by their level of GAF binding and compared the spacing of GAGAG elements in the 500 loci with strongest GAF binding to the GAGAG spacing in the 500 loci with weakest GAF binding. The results (Fig. 4) reveal that GAF target loci are indeed enriched in GAGAG elements that are spaced by less than ≈20 bp. The degree of clustering of GAGAG elements in target loci is much higher than can be attributed to random spacing (Fig. 4A), and the 500 control loci with no GAF binding do not show clustered GAGAG elements (Fig. 4B). Taken together with previously reported in vitro binding studies (26, 27), this result strongly suggests that cooperative GAF binding occurs in the native chromatin context. Note that the clustering of GAGAG pairs is only significant at odd distances, suggesting that there is evolutionary pressure to preserve the even/odd character of (GA)n repeats even over distances up to at least 10 bp.
Figure 4
Comparative analysis shows that in the 500 probed loci with high GAF binding, intergenic regions and introns contain 40.3% and 43.3%, respectively, of all 641 pairs of GAGAG elements spaced <10 bp apart, whereas exons harbor only 16.4% (see Fig. 4A). Because 45% of the DNA in these loci consists of exon sequences, closely spaced GAGAG elements are significantly underrepresented in exons (P = 3 × 10−55; binomial distribution). It is possible that this lack of clustering of GAGAG motifs explains for a large part the absence of GAF binding to exons.
Discussion
The genomewide analysis presented here confirms that GAF is a pleiotropic regulatory protein that binds in vivo to (GA)n motifs. A few distinct other motifs correlate with GAF binding, suggesting that certain other sequence-specific factors may provide a favorable context for GAF binding. The identity of these factors is presently unknown.
Importantly, we find that not all (GA)n elements in the genome are occupied by GAF. GAGAG elements in exons, but not introns, appear to be devoid of GAF, and GAGAG elements in downstream nontranscribed regions show weaker binding of GAF than in upstream regions. Furthermore, we provide evidence for cooperative binding of GAF to (GA)n elements that are closely spaced. The underrepresentation of such closely spaced (GA)n elements in exons provides a possible explanation for the lack of binding of GAF to exons. Recently, a genomewide study in yeast showed that the Rap1p protein abundantly associates with its cognate binding sequence in promoter regions, but rarely binds to the same sequence in ORFs (28). Exclusion from exons may therefore be a more general phenomenon. Although a “genomewide mechanism that marks promoter regions in chromatin” was proposed to explain the Rap1p distribution (28), our results argue that the spacing of binding motifs can be a major determinant of protein targeting.
Our finding that GAF binds to introns irrespective of the intron-promoter distance is in agreement with a previously suggested role for GAF in transcriptional elongation (14). Interestingly, GAF is able to recruit a chromatin remodeling complex (13), which may facilitate the passage of the elongation complex through nucleosome-packaged genes.
We have demonstrated here that chromatin profiling combined with reduce analysis provides a powerful tool for studying context-dependent binding of transcription factors. Even though the chromatin profiling data were obtained with conventional cDNA arrays, the combination with reduce allowed us to reveal the preferred sequence motifs of GAF, as well as the distribution of GAF over functionally distinct subregions of genes, such as introns, exons, and nontranscribed regions. Thus, reduce increases the resolution and analytical power of chromatin profiling. This approach should be applicable to a variety of DNA-binding factors.
Abbreviation
- GAF
- GAGA factor
Acknowledgments
We thank Ineke van der Kraan for technical assistance, Steve Henikoff for critically reading the manuscript, and Roel van Driel for support. B.v.S. was in part supported by the Royal Netherlands Academy of Sciences. H.J.B. was partly supported by National Institutes of Health Grant 1P20LM007276-01.
Supporting Information
Supporting Information
Supporting Information
- Download
- 94.31 KB
Supporting Information
Supporting Information
- Download
- .47 KB
Supporting Information
Supporting Information
- Download
- .42 KB
Supporting Information
Supporting Information
- Download
- .48 KB
Supporting Information
Supporting Information
- Download
- 5.14 KB
8000Table3.xls
- Download
- 1.15 MB
8000Table4.xls
- Download
- 69.50 KB
8000Table5.xls
- Download
- 143.00 KB
References
1
M D Biggin Nat Genet 28, 303–304 (2001).
2
G Farkas, J Gausz, M Galloni, G Reuter, H Gyurkovics, F Karch Nature 371, 806–808 (1994).
3
G E Croston, L A Kerrigan, L M Lira, D R Marshak, J T Kadonaga Science 251, 643–649 (1991).
4
Q Lu, L L Wallrath, H Granok, S C Elgin Mol Cell Biol 13, 2802–2814 (1993).
5
J A Weber, D J Taxman, Q Lu, D S Gilmour Mol Cell Biol 17, 3799–3808 (1997).
6
M Okada, S Hirose Mol Cell Biol 18, 2455–2561 (1998).
7
K Hagstrom, M Muller, P Schedl Genetics 146, 1365–1380 (1997).
8
R K Mishra, J Mihaly, S Barges, A Spierer, F Karch, K Hagstrom, S E Schweinsberg, P Schedl Mol Cell Biol 21, 1311–1318 (2001).
9
G Wall, P D Varga-Weisz, R Sandaltzopoulos, P B Becker EMBO J 14, 1727–1736 (1995).
10
K M Bhat, G Farkas, F Karch, H Gyurkovics, J Gausz, P Schedl Development (Cambridge, UK) 122, 1113–1124 (1996).
11
J G Omichinski, P V Pedone, G Felsenfeld, A M Gronenborn, G M Clore Nat Struct Biol 4, 122–132 (1997).
12
R C Wilkins, J T Lis Nucleic Acids Res 25, 3963–3968 (1997).
13
T Tsukiyama, P B Becker, C Wu Nature 367, 525–532 (1994).
14
T O'Brien, R C Wilkins, C Giardina, J T Lis Genes Dev 9, 1098–1110 (1995).
15
H Strutt, G Cavalli, R Paro EMBO J 16, 3621–3632 (1997).
16
B van Steensel, J Delrow, S Henikoff Nat Genet 27, 304–308 (2001).
17
C Benyajati, L Mueller, N Xu, M Pappano, J Gao, M Mosammaparast, D Conklin, H Granok, C Craig, S Elgin Nucleic Acids Res 25, 3345–3353 (1997).
18
J S Platero, A K Csink, A Quintanilla, S Henikoff J Cell Biol 140, 1297–1306 (1998).
19
B van Steensel, S Henikoff Nat Biotechnol 18, 424–428 (2000).
20
G M Rubin, L Hong, P Brokstein, M Evans-Holm, E Frise, M Stapleton, D A Harvey Science 287, 2222–2224 (2000).
21
A D Long, H J Mangalam, B Y Chan, L Tolleri, G W Hatfield, P Baldi J Biol Chem 276, 19937–19944 (2001).
22
Y Benjamini, Y Hochberg J R Stat Soc B 57, 289–300 (1995).
23
H J Bussemaker, H Li, E D Siggia Nat Genet 27, 167–171 (2001).
24
R C Wilkins, J T Lis Nucleic Acids Res 26, 2672–2678 (1998).
25
H Granok, B A Leibovitch, C D Shaffer, S C Elgin Curr Biol 5, 238–241 (1995).
26
K R Katsani, M A Hajibagheri, C P Verrijzer EMBO J 18, 698–708 (1999).
27
M L Espinás, E Jimenez-Garcia, A Vaquero, S Canudas, J Bernues, F Azorin J Biol Chem 274, 16461–16469 (1999).
28
J D Lieb, X Liu, D Botstein, P O Brown Nat Genet 28, 327–334 (2001).
Information & Authors
Information
Published in
Classifications
Copyright
Copyright © 2003, The National Academy of Sciences.
Submission history
Received: July 22, 2002
Accepted: December 30, 2002
Published online: February 24, 2003
Published in issue: March 4, 2003
Acknowledgments
We thank Ineke van der Kraan for technical assistance, Steve Henikoff for critically reading the manuscript, and Roel van Driel for support. B.v.S. was in part supported by the Royal Netherlands Academy of Sciences. H.J.B. was partly supported by National Institutes of Health Grant 1P20LM007276-01.
Authors
Metrics & Citations
Metrics
Citation statements
Altmetrics
Citations
If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.
Cited by
Loading...
View Options
View options
PDF format
Download this article as a PDF file
DOWNLOAD PDFGet Access
Login options
Check if you have access through your login credentials or your institution to get full access on this article.
Personal login Institutional LoginRecommend to a librarian
Recommend PNAS to a LibrarianPurchase options
Purchase this article to get full access to it.