CRP is one of the seven “global” transcription factors in E. coli known to regulate >50% of the cell's transcription units (3). CRP's activity is triggered by binding of the second messenger cAMP in response to glucose starvation and other stresses (4). Many catalogues of CRP regulated promoters have been assembled, based on either transcriptome analysis (5–7) or bioinformatics (8, 9). These analyses conclude that CRP directly affects up to 200 promoters, although neither approach directly measures CRP binding at promoters. The computational analysis of Robison et al. (8) identified ≈200 high-affinity DNA targets for CRP, predominantly located in noncoding parts of the E. coli chromosome. That study also predicted >10,000 lower-affinity CRP targets scattered throughout the chromosome, with little bias to noncoding sequences. In this work, we used ChIP-chip analysis to take snapshots of the distribution of CRP in living E. coli cells. The advantage of this methodology (reviewed in ref. 10) is that it provides a direct measure of binding at different targets, enabling us to identify specific sites where CRP binds strongly and show that the weaker sites noted by Robison et al. are, indeed, occupied.

Chromatin Immunoprecipitation. Monoclonal mouse antibodies against CRP and the β or σ⁷⁰ subunits of RNA polymerase were obtained from Neoclone (Madison, WI), and rabbit polyclonal anti-MelR was provided by E. Tamai (Kagawa University, Takamatsu City, Japan). Immunoprecipitations using antibodies against MelR or subunits of RNA polymerase were done exactly as described in ref. 14. Immunoprecipitations using antibodies against CRP were done by using a modified version of this protocol, where the conditions for the initial antibody–nucleoprotein incubation were altered from 90 min at room temperature to 7 h at 4°C (all subsequent washing steps were also done at 4°C). DNA obtained from a single immunoprecipitation (typically between 0.2 and 0.4 μg) was resuspended in 20 μl of sterile water and labeled as described below. Note that the DNA was not amplified before labeling and hybridization. For each immunoprecipitation, we also prepared a control sample, and the two samples were differentially labeled with Cy5 or Cy3 (see below). We noticed that, for example, it was possible to “immunoprecipitate” DNA from MG1655 Δcrp cells with anti-CRP, and such DNA, isolated from appropriate strains, was used as a background control when needed. Our reasoning was that such nonspecific immunoprecipitated DNA may differ from total genomic DNA.

Microarray Design, Labeling, and Hybridizations. Microarrays were designed and produced by Oxford Gene Technology specifically to analyze DNA obtained from ChIP experiments. Arrays consisted of 21,321 60-mer oligonucleotide probes that match MG1655 sequences at intervals of ≈160 bp. DNA generated from immunoprecipitations was labeled by Klenow random priming, incorporating Cy3-dCTP or Cy5-dCTP (GE Healthcare), and spun through a Qiaquick column (Qiagen, Valencia, CA), following the manufacturer's protocol. DNA was hybridized to microarrays in an Agilent Technologies microarray chamber by using microarray buffer (1 M NaCl/50 mM Mes, pH 7/20% formamide/1% Triton X-100) and rotated at 55° over 60 h. The arrays were then washed with Wash 1 [6× standard saline phosphate/EDTA (0.18 M NaCl/10 mM phosphate, pH 7.4/1 mM EDTA] (SSPE)/0.005% N-lauryl sarcosine] and Wash 2 (0.06% SSPE/0.18% polyethylene glycol 200), both for 5 min at room temperature.

Fig. 1 shows the results from two experiments in which DNA from E. coli growing in minimal defined media was immunoprecipitated with antibodies directed against either MelR or CRP. For each experiment, DNA samples were prepared from E. coli MG1655 and a mutant carrying a melR or crp deletion. Because both MelR and CRP are known to be active in our growth conditions, we expected that immunoprecipitated DNA from MG1655 would be enriched with target sequences for MelR or CRP, whereas DNA from the melR or crp deletion strains would contain only a nonspecific background of fragments precipitated by each antibody. After immunoprecipitation, DNA samples isolated from MG1655 and melR or crp cells were labeled with the fluorescent dyes Cy5 and Cy3, respectively. Samples were combined and hybridized to the microarray. Arrays were scanned and, for each of the 21,321 probes on the array, the Cy5/Cy3 ratio was calculated. These ratios were then plotted as a scattergram against their respective locations within the MG1655 genome. As expected, the data show that MelR binds to a single location in the mel operon (Fig. 1A), whereas CRP binds to multiple sites (Fig. 1B). After scrutinizing scans from eight independent experiments, we were able to identify 68 unambiguous and reproducible peaks, which must correspond to locations of high occupation by CRP. These locations are listed in Table 1, which is published as supporting information on the PNAS web site, and some examples of the peaks are shown in Fig. 1C. Of the 68 targets, 29 correspond to sites at promoters that had been validated previously. The other 39 are targets where CRP had not previously been found to bind. Whereas 5 of these targets fall within ORFs, 34 are located at or close to promoters. Further analysis shows that the average “score” for these newly identified CRP targets, as defined by Robison et al. (8), is –9.7. The fact that the average score for the >10,000 suggested targets for CRP in the E. coli chromosome is only –11.5 supports the argument that the newly identified CRP targets are likely to be bona fide sites for transcriptional regulation.

A striking difference between the profiles of DNA fragments immunoprecipitated by anti-MelR and anti-CRP concerns the “background.” For MelR, the background is very even, reflecting the fact that MelR binds at a single location (Fig. 1A). In contrast, for CRP, the background is very noisy because of a large number of low-intensity signals (Fig. 1B). We interpret this finding as being due to CRP binding to the many thousands of weak sites scattered throughout the E. coli chromosome, predicted by Robison et al. (8). Interestingly, for one segment of the pattern, corresponding to ≈1 Mbp centered near coordinate 2,500,000, the average signal intensity is lower, more clearly seen in Fig. 1D, that shows a smoothed version of the data at 200,000-bp resolution. Our interpretation for this finding is that access by CRP to its sites is limited in this segment of the E. coli chromosome. As a control, the CRP immunoprecipitation was repeated three times with cells grown in the presence of 0.4% glucose. Recall that, in these conditions, the cell's cAMP level is lowered, so that CRP becomes inactive (4). The scattergrams of data from these experiments (Fig. 1D) show that binding to all targets is lost, and the observed background resembles that seen with MelR.

Measurements of the Genome-Wide Distribution of RNA Polymerase. ChIP-chip analysis was used to determine the genome-wide DNA-binding pattern of RNA polymerase in midlog E. coli cells growing in LB medium. Immunoprecipitated DNA samples, prepared from E. coli MG1655 by using an antibody against the RNA polymerase β subunit and, as a control, from MG1655 Δcrp by using anti-CRP, were labeled with Cy5 and Cy3, respectively. The samples were combined and hybridized to the microarray, which was then scanned, and Cy5/Cy3 ratios were calculated for each probe. Fig. 2A shows these ratios, plotted as a scattergram against probe location within the MG1655 genome. The experiment shows ≈90 very prominent signals due to RNA polymerase binding. Many of these signals locate to genes required for essential cellular processes, such as translation and ATP synthesis (a complete annotation of the scattergram and a list of the top targets is given in Fig. 4 and Table 2, respectively, which are published as supporting information on the PNAS web site). This result is consistent with previous findings that, in rapidly growing cells, most RNA polymerase is engaged in transcribing the subset of genes needed for protein synthesis (15). Comparatively low signals are observed for RNA polymerase binding at the majority of genes that are less-frequently transcribed (e.g., the galETKM operon).

To attempt to catalogue all of the potential promoters, the analysis was repeated after cells had been treated with rifampicin for 15 min. The resulting profile of immunoprecipitated DNA is shown in Fig. 2B. Recall that rifampicin inhibits transcript elongation (16), and, thus, we expected an RNA polymerase molecule to be trapped at any functional promoter. Analysis of the data (by using a Cy5/Cy3 cut-off of 20) shows 529 locations that are occupied by RNA polymerase (see Table 3, which is published as supporting information on the PNAS web site). Only 36 of these locations were within ORFs and >400 bp away from the 5′ end of the nearest gene. A total of 394 peaks are located in noncoding sequences >150 bp long and adjacent to the 5′ end of a gene. An additional 58 signals are located at the 5′ end of a gene that appears to be part of an operon. Intriguingly, there are 41 locations where a signal is observed between transcription units that converge. This distribution is similar to that reported by Herring et al. (12). Some examples of individual signals are shown in Fig. 2C.

RNA Polymeraseomics. The clarity and reproducibility of the ChIP-chip analysis of the distribution of RNA polymerase in growing cells (Figs. (Figs.2A2A and 4) suggested that it might be possible to differentially label immunoprecipitated DNA from cells grown in different conditions and then use the microarray to identify locations where the binding of RNA polymerase is increased or decreased. Because most of these changes will correspond to changes in transcription frequencies, this procedure would provide an alternative to transcriptomic analysis of gene-expression patterns. Hence, in a first attempt, we prepared two midlog-phase cultures of MG1655 and pulsed one with 5 mM IPTG 40 min before crosslinking. After ChIP with the anti-RNA polymerase β-subunit antibody, the plus and minus IPTG samples were labeled with Cy5 or Cy3, respectively, mixed, and hybridized to the microarray. Fig. 3A shows the Cy5/Cy3 ratio measured for each of the probes plotted against its coordinate on the MG1655 genome. The results show that IPTG primarily induces the binding of RNA polymerase to the lacZYA operon. Fig. 3A (Inset) shows the distribution of RNA polymerase binding across the lacZYA operon: More RNA polymerase is located toward the 5′ than the 3′ end of the operon.

To investigate a more complex transcriptional response, 5 mM salicylic acid was added to a midlog-phase culture of MG1655. The consequence of this addition was to reduce the growth rate sharply and to induce the multiple-antibiotic-resistance regulon and other stress responses (17). The experiment was done exactly as above, with the plus and minus salicylic acid samples labeled with Cy5 or Cy3, respectively. Fig. 3B shows the log₁₀ Cy5/Cy3 ratio measured for each of the probes plotted against its coordinate on the MG1655 genome (a full annotation is shown in Fig. 5, which is published as supporting information on the PNAS web site). The results show that salicylic acid provokes redistribution of RNA polymerase from genes encoding stable RNA, components of the translation machinery, and other proteins essential for rapid growth to genes encoding proteins required to survive the stress (the principal changes are listed in Table 4, which is published as supporting information on the PNAS web site). Intriguingly, around half of the sites at which RNA polymerase binding increases are immediately adjacent to down-regulated genes. For example, Fig. 3Ci shows the redistribution of RNA polymerase at the serT (serine tRNA1) locus. Similar local rearrangements in RNA polymerase binding can be seen throughout the chromosome and are highlighted in Fig. 5. In each case, the geometry of the redistribution is similar, with the sites of maximal and minimal RNA polymerase binding separated by ≈3,000 bp.

Most E. coli transcription factors interact directly at just one or a few loci, whereas some interact at many different targets (3). The global regulator CRP is estimated to interact at ≈200 regulatory regions in E. coli (5–7). Because at least 22 of these regions control the expression of other transcription factors (3), it has been difficult to unravel direct and indirect effects of CRP. Hence, in our ChIP-chip experiment, which was performed in conditions in which CRP is known to be active, we expected to find up to 200 targets. The fact that we were able clearly to define only 68 targets could be due to many factors. For example, it is possible that CRP is unable to occupy certain targets in noninducing conditions. More trivially, it is possible that CRP bound at certain loci fails to crosslink to target DNA. However, the main factor that prevented us from clearly defining more targets was the “noisy” background signal that we attribute to the binding of CRP at the many thousand weak DNA sites for CRP predicted from sequence analysis (8). The conventional view is that global regulators are transcription factors whose target sites are located at scores of promoters. Our analysis suggests that this view is incorrect for CRP and that CRP binds to a range of many thousand different sites, with only a small proportion of bound CRP directly affecting transcription initiation. Thus, CRP should be considered as a chromosome-shaping protein (see ref. 21), in addition to its function as a promoter-specific regulator (22). Note that DNA-bound CRP bends its target sharply (4), presumably contributing to the compaction of the E. coli chromosome. Unexpectedly, our experiment identified a ≈1-Mbp segment of the E. coli chromosome where the binding of CRP to weaker sites is reduced. The predicted density of weak DNA sites for CRP in this segment is not unusual: It covers ≈20% of the genome and ≈20% of the predicted DNA sites for CRP (8). Hence, we suggest that our observation may be due to some higher-order structure in the bacterial chromosome, as discussed by Boccard et al. (23). Note that, although background binding by CRP in this segment of the E. coli chromosome is reduced, it contains many CRP-dependent promoters (7) and other highly transcribed genes (Fig. 2). We suggest that the measurement of weak binding of proteins throughout bacterial chromosomes by ChIP-chip may have important applications in understanding higher-order folded chromosome structures.

We have also used ChIP-chip to monitor the distribution of RNA polymerase in growing E. coli cells. Because RNA polymerase levels in E. coli are limiting, we can assume that most DNA-bound RNA polymerase is associated with transcribed sections of the chromosome (15). We found that, as expected, most RNA polymerase is associated with the small proportion of genes encoding components of the translational machinery and other proteins essential for rapid growth. It is significant that these loci are scattered throughout the E. coli genome, and they may contribute to the organization of discrete supercoil domains (24). Hence, the small number of transcription foci reported by Cabrera and Jin (25) must be formed by higher-order structures that bring nonadjacent segments of the E. coli chromosome together. Note that most of the highly expressed transcription units listed in Table 2 are arranged such that the direction of transcription points away from the replication origin, and, as suggested by Dworkin and Losick (26), transcribing RNA polymerase at these loci may help to drive chromosome segregation.

The experiment illustrated in Fig. 2A permits us to detect RNA polymerase binding at only the most-strongly expressed genes. To attempt to identify all of the potential promoters, we repeated the ChIP-chip analysis after blocking transcription elongation with rifampicin. In these conditions, each active promoter is expected to become occupied, and, after performing such an experiment with E. coli grown in minimal media, Herring et al. (12) identified >1,000 loci at which RNA polymerase is bound. After our experiment, with E. coli grown in rich media, we identified >500 RNA polymerase-binding sites (Fig. 2B and Table 3). In agreement with Herring et al., most of these sites correspond to bona fide promoters, and there are very few targets for RNA polymerase within ORFs. Interestingly, however, some RNA polymerase molecules bind to targets between cistrons, where there is no obvious requirement for an initiating RNA polymerase. These locations may correspond to sites from which small RNA molecules are transcribed.

In the final part of our work, we showed that the ChIP-chip method could be used to measure changes in RNA polymerase distribution provoked by environmental change. These experiments are formally equivalent to many transcriptomic studies, and, yet, by measuring RNA polymerase directly, extra information is provided. Thus, whereas we showed that IPTG induced the lac operon, we could also measure the distribution of RNA polymerase molecules within the operon. The skewed distribution of RNA polymerase across this transcription unit may be evidence for barriers to elongation that disrupt the transit of some RNA polymerase molecules. In an experiment with salicylic acid, we found that genes encoding the translation apparatus were down-regulated, whereas genes that encode stress-response proteins are switched on. Unexpectedly, we identified many sites adjacent to strongly down-regulated genes where RNA polymerase appears to accumulate after salicylic-acid shock. In some instances, these sites of RNA polymerase accumulation correspond to the locations of stress-response genes whose transcription is induced by salicylic acid (17). We speculate that some of these sites are locations where excess RNA polymerase, generated by the down-regulation of rRNA and tRNA transcription, can be stored until environmental conditions change. Presumably, it is no coincidence that some stress-response genes are located close to sites where large pools of RNA polymerase become available when growth is stalled.