Before discussing whole-genome maps of histone modifications, it is useful to consider the genomic distribution of nucleosomes (that is, the histone octamers) themselves. The positioning of nucleosomes is believed to be crucial for gene regulation. Two recent papers 89 describe the whole-genome mapping of histone-genome interactions in the budding yeast Saccharomyces cerevisiae. ChIP-on-chip experiments were performed with epitope-tagged histone H2B or H4, and endogenous H3 (using an antibody that is presumably insensitive to the H3 modification state). The results indicate that there are locus-specific differences in nucleosome density. Most notably, the upstream regions of highly active genes display a reduced nucleosome density compared to the upstream regions of inactive genes. Depletion of histone H2B in promoter regions was also detected by a genome-wide mapping technique that combines ChIP with mass sequencing of short sequence tags 10. A similar depletion of nucleosomes, although of lower magnitude, was observed in coding regions of highly active genes, when compared to inactive genes 89. Although it cannot be completely ruled out that local differences in cross-linking efficiency or epitope accessibility account for the observed differences, the results are in agreement with detailed studies of selected regions, such as the Pho5 promoter 1112. Thus, these data suggest that, as a general rule, gene activation in yeast is accompanied by reduced nucleosome density.

In order to understand the interplay between different histone modifications during transcriptional regulation, it is necessary to construct systematic genome-wide location maps of all histone modifications and the proteins that control and interact with them. Two laboratories have taken important steps towards this goal by global mapping of a large number of histone modifications. Kurdistani et al. 13 performed ChIP-on-chip in S. cerevisiae using 11 antibodies, each of which specifically recognizes a different acetylated histone lysine residue. They used whole-genome arrays of coding regions (ORFs) and intergenic regions (IGRs), thereby creating complete genomic maps of these histone acetylations. Schübeler et al. 14 performed similar ChIP-on-chip experiments in a cultured cell line from Drosophila melanogaster, using antibodies against six distinct histone modifications. But because the complete Drosophila genome is not yet available on microarrays, these mapping experiments were done with cDNA arrays, thus providing a view of transcribed regions only. One modification (H3K4 dimethylation) was also mapped using a genomic tiling array covering the left arm of chromosome 2. Strikingly, both papers report that the genomic distributions of virtually all tested histone modifications are highly similar to one another. Schübeler et al. 14 found that H3 and H4 acetylation, H3K4 di- and tri-methylation, and H3K79 di-methylation, are all strongly correlated with each other, having pairwise Pearson correlation coefficients of at least +0.7. Along the same lines, Kurdistani et al. 13 report that all pairs of 11 tested acetylation sites show significant positive correlations, with the majority of correlation values (both in ORFs and in IGRs) being well above +0.5.

Before we reflect on the potential biological implications, the possibility should be considered that some kind of biological or experimental bias is responsible for the strong correlations. One possible source of bias could be local differences in nucleosome density. It is conceivable that in genomic regions with high nucleosome densities ChIP analysis of any histone modification might yield stronger signals than in regions with low nucleosome density 15 (Figure 2). Unfortunately, the papers by Schübeler et al. 14 and Kurdistani et al. 13 did not address this putative bias by including control experiments with antibodies against histone epitopes that lack potential modification sites. To illustrate the putative consequences of this bias, we compared the yeast histone acetylation data from Kurdistani et al. 13 with yeast nucleosome-density maps 89 (Table 1). Strikingly, this analysis reveals that especially within IGRs there is a considerable correlation between H2B, H3 or H4 density and ChIP of several of the lysine acetylation variants. These results argue that ChIP of histone modifications may be biased by local differences in nucleosome density. Given the strict 2:2:2:2 stochiometry of histones in the nucleosome, it is likely that the bias is similar for modifications of all four histones.

Kurdistani et al. 13 reported that the acetylation levels in IGRs of all tested lysines in histones H2A, H2B and H4 are negatively (although weakly) correlated with expression levels of the corresponding genes (Table 1). But when the results are corrected for the nucleosome-density bias, it may well be that these modifications are positively correlated with gene expression. Likewise, the reported weakly positive correlations between gene expression and acetylation of H3K9, H3K18 and H3K27 in IGRs may in fact be more positive when corrected for nucleosome density. To further investigate this bias, we compared the correlations of each lysine acetylation with nucleosome density and with gene expression levels (Figure 3). The striking linear relationship between points on the plot shown in Figure 3 indicates that these correlations are not independent, underscoring the notion that the links between histone modification and gene expression cannot be interpreted safely without a correction for nucleosome density. Obviously, the bias caused by nucleosome-density differences may have implications for further analyses and interpretations of all (previously published) histone-modification maps.

It is not known whether depletion of nucleosomes in transcriptionally active regions is a general phenomenon in Drosophila. Schübeler et al. 14 performed ChIP-on-chip with an antibody against phosphorylated H3S10 as a control, and reassuringly found that this epitope does not correlate with any of the other modifications, although one might argue that this is not an entirely satisfying control because H3S10 phosphorylation is primarily a mitosis-specific mark 16. Importantly, Schübeler et al. 14 report strong positive correlations between all mapped acetylation and methylation patterns and gene expression levels. Correction for nucleosome depletion in active regions would further increase these correlations, and therefore would only strengthen the conclusion that H3 and H4 acetylation, H3K4 di- and tri-methylation, and H3K79 di-methylation are all enriched in regions of high transcriptional activity. Although other experimental biases cannot be ruled out, these results argue that, at least in coding regions of the fly genome, these histone modifications are closely linked.

Interestingly, a recent mapping study of histone modifications in human cells suggests that linkage between active histone modifications might be a common theme in eukaryotic genomes. By combining ChIP with hybridization on tiling oligonucleotide arrays, Bernstein et al. 17 showed that more than 90% of H3K4 methylated sites on human chromosomes 21 and 22 are also H3K9K14-acetylated. Virtually no variation was found in H3 density, suggesting that in human cells differences in nucleosome occupancy are minor, and thus are unlikely to cause severe biases in the mapping of histone modifications.

Unraveling the role of histone modifications can be greatly facilitated by genome-wide analysis of histone-modifying enzymes. Such studies of histone deacetylases (HDACs) and histone acetyltransferases (HATs) in yeast argue that acetylation of most histone lysine residues is in general positively, not negatively, associated with gene expression. ChIP-on-chip mapping of histone acetylation patterns after deletion of either Rpd3 or Hda1 (both are HDACs) revealed increased acetylation of several histone lysine residues 18. Comparison of these 'deacetylation maps' with expression profiles obtained after deletion of either HDAC revealed that increased acetylation is generally accompanied by increased transcription 18. This was found for six different acetylation positions, including H4K12, H4K16, and H2BK16. Furthermore, genome-wide mapping of the binding sites of the HATs Gcn5 and Esa1 in yeast show that these enzymes are primarily recruited to active genes 19. These data further support the view that for many lysines on the nucleosome acetylation is generally positively linked to transcription. While there seems to be a general correlation between histone acetylation and gene expression, some histone deacetylases seem to positively regulate gene expression, possibly by maintaining the amount of histone acetylation that is optimal for efficient transcription 20.

Another study employed expression profiling of yeast strains in which lysines on histone H4 were systematically mutated to arginine (which mimics unacetylated lysine) 21. Mutations of lysines 5, 8, and 12 had identical effects on the expression of a broad set of genes. These effects were cumulative: changes in expression were proportional to the number of mutations. In contrast, mutation of H4K16 affected a distinct small group of genes. These results point to a nonspecific, cumulative regulatory role for H4K5, H4K8, and H4K12 acetylation.

Theoretically, the large number of possible histone modifications creates the potential for an extremely complex regulatory code. Nevertheless, careful comparison of the recent whole-genome studies of histone modifications, as discussed above, suggests a rather simple picture: most of the mapped modifications, including acetylation of all tested nucleosomal lysines and methylation of H3K4 and H3K79, are positively correlated with levels of gene expression. This raises the possibility that the different 'active' histone marks are linked. So far, no histone modifications have been identified that mark transcriptionally inactive regions in budding yeast. In other eukaryotes, however, a number of such repressive marks have been reported, such as methylation of H3K9, H3K27, and H4K20. The degree of global linkage of these repressive histone marks remains to be established.

How may the linkage of various active histone modifications be established? In part, it may be a consequence of the broad substrate-specificity of some histone-modifying enzymes. For example, the HDAC Rpd3 and the HAT Gcn5 both control the acetylation state of several lysines on H3 and H4 1822. Furthermore, several histone-modifying enzymes might act at the same nucleosome because they interact with each other (for example, the TAC1 chromatin-remodeling complex has both H3K4 methyltransferase activity and HAT activity 23), or because they are recruited by the same signal (for example, the H3K4, H3K36, and H3K79 methyltransferases in yeast all require components of the Paf1 complex for their activity 24). In addition, certain histone modifications may recruit complexes containing other modifying enzymes and thus trigger cascades of histone modifications (for example, methylation on H4R3 by PRMT1 facilitates acetylation by p300 on H4K8 and H4K12 25) It is conceivable that such cascades could be initiated by different initial modifications, yet ultimately result into the same set of modifications.

Why would multiple histone modifications be employed in parallel to control transcription? Partially redundant functions of histone marks in chromatin might confer robustness on epigenetic regulation. Thus, transcriptional responses might be largely independent of small changes in any single modification and the cell might have several options to induce transcriptional changes (changes in transcription can be elicited by inhibition or activation of one of several different histone-modification pathways) 26. Despite the evidence that many active histone modifications are globally linked, it is likely that there are also differences between combinations of histone modifications that mark specific genes (Figure 1). It should be emphasized that the current ChIP-on-chip maps do not contain information about the timing of (sequential) events, as the maps were generally obtained from asynchronous cell populations. Moreover, ChIP-on-chip does not have single-nucleosome resolution, which may complicate the interpretation: it is formally possible that certain nucleosomes in a particular IGR or coding region carry functionally important modifications that differ from those in the surrounding nucleosomes. Indeed, detailed ChIP analysis of individual genes suggests that trimethyl-H3K4 occurs predominantly in the 5' part of transcribed regions whereas dimethyl-H3K4 has a much broader distribution 17272829. Furthermore, the different phenotypes and expression profiles that result from inactivation of the enzymes that control the active marks argue that the respective histone modifications can have distinct effects on gene expression.

The high correlations between histone modifications as reported by Schübeler et al. 14 and Kurdistani et al. 13 may have been overestimated as a result of differences in nucleosome densities or other unknown biases. Interestingly, for a series of additional analyses of the yeast histone acetylation maps, Kurdistani et al. 13 employed a data-normalization procedure called variance normalization. The first step of this normalization involves the subtraction (for each probed locus) of the average of all 11 measured acetylation levels from each individual acetylation level. If the presumed nucleosome-density bias is indeed similar for all histone epitopes, then this normalization procedure would at least partially correct for the bias, although the residual values after such a subtraction may have a high noise content. Intriguingly, after variance normalization and the subsequent use of a clustering algorithm, Kurdistani et al. 13 identified clusters of functionally related genes that were enriched in certain histone-acetylation patterns. Further experiments are needed to reveal the functional relevance of the relationships between histone-modification patterns and gene functions.

In addition to revealing interesting biological information, recent whole-genome mapping studies have also revealed some technical hurdles that still need to be overcome if we are to construct highly reliable epigenetic maps. First, improved experimental methods are required. Inclusion of control ChIPs, of modification-free histone epitopes or tagged histones, may help to solve the potential bias problems associated with ChIP-on-chip. It will also be crucial to verify whole-genome ChIP results by an independent method because the exact choice of procedures used for ChIP (such as the type of cross-linker, the extent of cross-linking, fractionation of chromatin, and experimental conditions that may cause cellular stress responses) can affect the outcome of a ChIP-on-chip experiment 19303132. At present, no alternative genome-wide method is available for ChIP analysis of histone modifications. The DamID technology, which employs tethered Dam methyltransferase 33, provides an independent means for verifying the global binding data of histone-modifying enzymes and other chromatin-associated proteins, but it is not suitable for mapping histone modifications.

A second challenge lies in the integration of histone-modification maps with other whole-genome chromatin datasets, such as gene-expression profiles, global maps of DNA methylation, DNase-hypersensitive sites, and the binding of histone-modifying enzymes and other chromatin-interacting proteins. For example, a recent mapping study in Arabidopsis revealed tight links between methylation of H3K9, hypomethylation of H3K4, and methylation of DNA in a heterochromatic region of the genome 34. Whole-genome datasets are extremely complex, however, and each mapping approach yields different information. Moreover, most of the current data should be regarded as probabilistic because of the rather high noise levels. Advanced statistical techniques, such as Bayesian network models 35, may be essential for the successful integration of different types of data and the eventual construction of useful biological models. Despite these challenges, the genome-wide exploration of chromatin structure and function is well on its way.