We analyzed conformation of GC- and AT-rich isochores along yeast chromosome III using the 3C methodology. 3C is used to detect the relative frequencies of interaction for different pairs of genomic loci. 3C data can be used to determine the overall spatial conformation of chromosomes and chromosomal sub-domains 1417181920. This approach, as previously described in detail 212223, involves three steps. First, formaldehyde cross-linking is used to trap pairs of interacting chromatin segments (via protein/protein/DNA cross-links). Second, cross-linked chromatin is solubilized and then digested and ligated at low concentration so that cross-linked segments will be preferentially joined. Third, ligation products are detected and quantified by PCR using pairs of primers specific to each pair of interacting loci. Relative levels of different PCR products correspond to the relative interaction frequencies of the various locus pairs.

We chose to analyze isochore domains along chromosome III because of their relatively large size (up to 90 kb), which allows detailed 3C analysis. Our previous analysis of these isochores revealed structural differences but did not address whether these differences in interaction frequencies could reflect differences in chromatin compaction 14. Here we addressed this issue in detail. Nuclei were isolated from alpha-factor arrested (G1) haploid wild-type yeast cells and 3C was performed. Interaction frequencies for pairs of sites located within the GC- and AT-rich domains along the right arm of chromosome III (positions 100-190 kb and 190-280 kb, respectively) were measured. When these frequencies are plotted against the distance between the loci of each pair (the genomic site separation) an inverse relationship between interaction frequency and genomic distance is observed. Moreover, sites located in the GC-rich isochore domain interact less frequently than sites located in the AT-rich isochore domain (Figure 1a).

We next determined whether the difference in interaction frequencies was simply due to lower levels of formaldehyde cross-linking in the GC-rich isochore compared to the AT-rich isochore. We reasoned that formaldehyde cross-linking during the 3C procedure would reduce restriction enzyme digestion efficiency due to cross-linking of proteins to restriction sites and that any differences in cross-linking in GC- and AT-rich domains should be detectable as differences in their susceptibilities to restriction enzyme digestion. We first used a PCR based method that detects partially digested chromatin to confirm that digestion efficiency is inversely proportional to the level of cross-linking (Additional data file 2). We then assessed the digestion efficiencies for several sites located in the GC-rich and AT-rich regions. The fraction of protected restriction sites, and thus the level of cross-linking, in the GC-rich regions was slightly higher than, but not significantly different from, that observed in the AT-rich domain (Additional data file 1 and 2). Similar previous 3C analyses have also shown that digestion and cross-linking efficiency is relatively constant throughout large chromosomal regions 192425. These results imply that the two types of domains have undergone very similar levels of cross-linking and thus that the difference in interaction frequencies in GC- and AT-rich domains as detected by 3C reflects a difference in spatial conformation.

Interaction frequencies are proportional to the local concentration of the loci and, therefore, differences in interaction frequencies within the GC-rich and AT-rich domains are most straightforwardly attributable to a difference in effective volume between these domains, with the GC-rich isochore occupying a larger volume per kb of DNA (that is, being less compact).

Further details of differences in compaction between GC- and AT-rich domains are provided by analysis of 3C data using a suitable polymer model 14262728. The model used here (equation 1) is the same as that used previously 142629, but is slightly re-arranged in order to allow assessment of chromatin compaction by including a parameter L that reflects chromatin compaction:

X ( s ) = ( k × L − 3 ) × 0.53 × β − 3 / 2 × exp ⁡ ( − 2 β 2 ) × S − 3 nm 3 mol liter with β = s S × ( 1 − s c ) MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2Caerbhv2BYDwAHbqedmvETj2BSbqee0evGueE0jxyaibaiKI8=vI8tuQ8FMI8Gi=hEeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciGacaGaaeqabaqadeqadaaakeaafaqabeqadaaabaGaamiwaiaacIcacaWGZbGaaiykaiabg2da9iaacIcacaWGRbGaey41aqRaamitamaaCaaaleqabaGaeyOeI0IaaG4maaaakiaacMcacqGHxdaTcaaIWaGaaiOlaiaaiwdacaaIZaGaey41aqlcciGae8NSdi2aaWbaaSqabeaacqGHsislcaaIZaGaai4laiaaikdaaaGccqGHxdaTciGGLbGaaiiEaiaacchadaqadaqaaiabgkHiTmaalaaabaGaaGOmaaqaaiab=j7aInaaCaaaleqabaGaaGOmaaaaaaaakiaawIcacaGLPaaacqGHxdaTcaWGtbWaaWbaaSqabeaacqGHsislcaaIZaaaaOWaaSaaaeaacaqGUbGaaeyBamaaCaaaleqabaGaae4maaaakiaab2gacaqGVbGaaeiBaaqaaiaabYgacaqGPbGaaeiDaiaabwgacaqGYbaaaaqaaiaabEhacaqGPbGaaeiDaiaabIgaaeaacqWFYoGycqGH9aqpdaWcaaqaaiaadohaaeaacaWGtbaaaiabgEna0oaabmaabaGaaGymaiabgkHiTmaalaaabaGaam4CaaqaaiaadogaaaaacaGLOaGaayzkaaaaaaaa@7403@

This model describes chromatin in terms of three key features: flexibility, apparent circularity and level of compaction (expressed in nm/kb). The parameter s is the genomic site separation between two loci (in kb) and X(s) is the interaction frequency. The parameter S is the length of the Kuhn's statistical segment in kb, which corresponds to two times the persistence length and is a measure for the flexibility of the chromatin fiber. The parameter c is the apparent circle size of the fiber (in kb). In the case of a fiber engaged in an unconstrained random walk, c will be infinitely large, in which case β equals s/S; any other value of c implies the presence of constraints on the path of the chromatin fiber. The parameter k is the efficiency of cross-linking 14. Finally, L is the contour length (in nm) of 1 kb of chromatin, referred to as the mass density, and is a measure for the level of compaction of the chromatin fiber. Fitting interaction frequencies to equation 1 yields values for S, c and for [k × L^-3]. Values for the individual parameters k and L cannot be directly obtained from this analysis and the combined parameter [k × L^-3] will be referred to as the apparent compaction factor. However, if k is known to be constant, as appears to be the case in the present study (above), variations in this combined parameter can be interpreted as differences in the value of L.

When interaction frequencies for the GC- and AT-rich domains, from three independent cultures were fitted to equation 1 (Figure 1a; Table 1), significant differences between the two types of domains become apparent. First, chromatin in the GC-rich domain is significantly more flexible than chromatin in the AT-rich domain (that is, S is smaller, P < 0.05). Second, the GC-rich domain appears to be in a circular conformation, with an apparent circle size of around 200 kb reflecting the presence of constraints on the chromatin path, consistent with our previous findings 14. For the AT-rich domain, in contrast, we had to assume (for two out of three cultures) that c is infinitely large in order to obtain a good fit, implying the apparent lack of such constraints. Third, an approximately three-fold lower value for the apparent compaction factor [k × L^-3] was obtained for the GC-rich domain than for the AT-rich domain (P < 0.01).

The difference in the value of [k × L^-3] for the GC- and AT-rich domain (Table 1) could reflect differences in cross-linking efficiency (k) or compaction (L). Since no difference in cross-linking efficiency between GC- and AT-rich domains could be detected, this analysis indicates that there is a 2.5-fold difference in the value of L^-3 (average of three independent yeast cultures) and thus an approximately 1.4-fold difference in the value of L. In other words, the contour length of 1 kb of chromatin in the GC-rich isochore region is approximately 40% larger than the contour length of 1 kb of chromatin in the AT-rich isochore.

We next examined functional differences between GC- and AT-rich isochores by determining the relationship between base composition of genes and their transcriptional activity throughout the genome in wild-type yeast cells. First, genes were divided into categories based on the average base composition of the surrounding 4 kb region (that is, the average base composition of a gene was determined using a 4 kb window centered around the transcription start site). Genes were then divided into six groups, approximately equal in size, based on regional base composition (Figure 1b). Genes located within 30 kb of telomeres were omitted because these genes are under epigenetic control due to their close proximity to telomeric heterochromatin. Excluding such genes, the final dataset comprised 5,568 open reading frames.

Next we determined average steady-state mRNA levels of genes in each group. The transcriptional activity of each gene is known from data obtained by Bernstein et al. 30. Using their dataset, we find that expression levels of individual genes within each group vary widely, but that the most GC-rich genes as a group are, on average, significantly more transcriptionally active than the most AT-rich isochore group (Figure 1b). Previously, Marin et al. 31 reported a similar positive correlation between mRNA levels and GC content of genes in yeast.

Histone modifications can affect the conformation of chromatin fibers and are correlated with gene expression (for example, 32333435). Given the differences in chromatin conformation and transcriptional activity of GC- and AT-rich chromatin domains, we hypothesized that these domains may also display differences in histone modification status. We used a genome-wide dataset of histone modification levels in wild-type yeast cells obtained by Kurdistani et al. 36 to determine average histone modification levels of GC- and AT-rich regions.

We again divided all genes into six groups based on their base composition, exactly as described above. For each group we determined the average level of each of 11 histone modifications (Figure 2a-d; Additional data file 3). We found that 4 out of 11 modifications (histone H4 Lys8 (H4K8) and Lys12 (H4K12), and histone H3 Lys9 (H3K9) and Lys18 (H3K18)) are enriched in GC-rich chromatin and depleted in AT-rich chromatin. The levels of the remaining seven modifications were not clearly correlated with base composition (histone H4 Lys16 (H4K16), histone H3 Lys14 (H3K14), LysK23 (H3K23) and Lys27 (H3K27), histone H2A Lys7 (H2AK7), and histone H2B Lys11 (H2BK11) and Lys16 (H2BK16); Additional data file 3). Interestingly, modifications of both H3 and H4 are correlated with base-composition, whereas modifications of H2A and H2B are not. These results demonstrate that GC- and AT-rich chromatin domains display distinct levels of H3 and H4 acetylation (Figure 2e) and provide additional evidence for structural and functional differences of isochore domains in yeast.

The histone deacetylase Rpd3p acts as a repressor of a number of specific target genes throughout the genome 373839. In addition, Rpd3p has been shown to affect the global pattern of histone acetylation, over and above its specific effects at target promoters 40. This global activity is weak, affecting histone acetylation levels only up to two-fold. The significance of these more global weak effects on chromatin structure and gene expression is not well understood. We were interested in the possibility that the global effects of Rpd3p may modulate structural and functional differences between GC- and AT-rich chromatin. To test this, we used 3C to analyze changes in chromatin conformation of GC- and AT-rich domains along chromosome III in an rpd3Δ mutant.

Interaction frequencies between sites located in the GC- and AT-rich isochore domains of chromosome III were determined and plotted against genomic site separation, as described above for wild-type cells (Figure 3a-c). As in the wild type, the GC-rich domain exhibits lower interaction frequencies than the AT-rich domain. However, the magnitude of the difference in interaction frequencies between the two domains is greater in the rpd3Δ mutant than in the wild type (compare Figures 3a and 1a; Table 1). This effect can be seen most clearly by normalizing both datasets to the interaction frequencies observed in one of the two domains, for example, the AT-rich domain (see Materials and methods). Such a comparison reveals that all interaction frequencies in the GC-rich domain are approximately 25% lower in the rpd3Δ mutant than in the wild type (Figure 3b-d). This effect is statistically significant (P < 0.001; Figure 3d) and was observed in three independent rpd3Δ cultures (Table 1). We also analyzed a set of interactions along the right arm of chromosome VI, which is characterized by a high GC-content, and found a similar significant decrease in interaction frequencies (Figure 3d).

As discussed above, a difference in interaction frequency between GC- and AT-rich domains could result either from a difference in chromatin compaction or a difference in cross-linking efficiency, and the two possibilities can be distinguished by assessing the efficiency of restriction digestion. When such analysis was performed for rpd3Δ cells, we again found, as for wild-type cells, no significant difference in digestion efficiency between GC- and AT-rich isochore domains (Additional data file 2). We conclude that Rpd3p differentially affects the conformation of these GC-rich and AT-rich domains, which results in further exaggeration of their difference in conformation. These observations are important for two reasons. First, they reveal a previously unrecognized base-composition-sensitive effect of this histone deacetylase. Second, they suggest that Rpd3p normally acts to keep the two types of isochore domains from being even more different in conformation than they would otherwise tend to be.

To more fully characterize chromatin conformation in rpd3Δ cells, interaction frequencies were fitted to equation 1 (Figure 3a-c; Table 1). Flexibility and apparent circularity of chromatin did not significantly change in rpd3Δ cells compared to wild-type cells (Table 1). However, the statistically significant reduction in interaction frequencies in the GC-rich isochore compared to the AT-rich isochore resulted in a five-fold difference in apparent compaction factor [k × L^-3] compared to a 2.5-fold difference observed in wild-type cells. Analysis of data from three wild-type cultures and three rpd3Δ cultures shows that this effect on the fold difference in [k × L^-3] is reproducible and significant (P < 0.05). Application of the restriction digestion assay described above further reveals that, as for wild-type cells, this difference is not ascribable to a differential change in efficiency of cross-linking efficiency (k) (Additional data file 2).

We conclude that the difference in relative compaction L of the chromatin fiber in the GC- and AT-rich isochores has changed in rpd3Δ cells. Specifically, in rpd3Δ cells, the value of L is 1.7-fold higher, and compaction correspondingly lower, in the GC-rich isochore compared to the AT-rich isochore.

Our results suggest that Rpd3p activity differentially affects GC-rich and AT-rich chromatin. We next tested whether this effect was also reflected in differential modulation of expression of GC- and AT-rich genes throughout the yeast genome. This question was addressed using a genome-wide dataset generated in Tsukiyama's laboratory 37 that describes the effects of deletion of RPD3 on transcription throughout the yeast genome.

First, we examined whether the genes in the relatively large 90 kb GC-rich and AT-rich isochores along chromosome III are differentially affected by deletion of RPD3. We calculated the average change in transcription in rpd3Δ versus wild-type cells along chromosome III as a function of gene position along the chromosome. Comparison of average base composition with average global change in transcription shows that deletion of RPD3 had little effect on transcription of the central AT-rich isochore domain. In contrast, transcription in the GC-rich isochore domains was significantly more increased (Figure 4a; P < 0.001).

Next we analyzed whether deletion of RPD3 has a general differential effect on expression levels of GC- and AT-rich genes throughout the genome. We calculated the effect of deletion of RPD3 on expression of the same six groups of genes with different base compositions as described above (Figure 1b). We found that all six groups exhibit increased average levels of transcription in the rpd3Δ mutant (genome-wide average Log(rpd3Δ/WT) = 0.08) but that the magnitude of this effect varies in proportion to GC content. More GC-rich genes are significantly more up-regulated in rpd3Δ cells than more AT-rich genes (Figure 4b). These data confirm that elimination of Rpd3p affects most regions of the genome 1538 and, in addition, reveal a previously unappreciated fact that base composition is an important feature in determining the magnitude of this effect.

To characterize the base-composition sensitive effect of deletion of RPD3 in more detail, we analyzed whether it was related to the level of expression of genes in wild-type cells. First, we determined the general relationship between mRNA levels of genes in wild-type cells and the fold change in expression in rpd3Δ cells. We found that deletion of RPD3 most strongly activated genes that are expressed at relatively low levels in wild-type cells (Figure 4c), as expected for deletion of a transcriptional repressor. Next we analyzed whether this relationship is different for GC- and AT-rich genes. Interestingly, for both the most GC-rich and AT-rich groups of genes we found a similar negative correlation between transcript level in wild-type cells and increase in transcription in rpd3Δ cells. Importantly, however, for all levels of transcription, GC-rich genes are more up-regulated upon deletion of RPD3 than AT-rich genes that are expressed at similar levels in wild-type cells (Figure 4d). These observations reveal that Rpd3p mediates transcriptional control via two independent effects. At one level, Rpd3p-mediated inhibition is correlated with steady-state expression levels of genes. At the second level, Rpd3p inhibits transcription in a GC content-dependent manner. The GC content-dependent activity is not correlated with the steady-state expression level of genes.

These observations suggest that the base composition-dependent activity of Rpd3p is not dependent on local and gene-specific control of promoter activity, but instead may be related to more general features of chromatin conformation in GC-rich regions of the genome. In that case, we predict that the base composition-dependent activity of Rpd3p will be independent of local targeting to specific target genes. To test this we analyzed the effects of deletion of UME6. Ume6p recruits Rpd3p to many of its specific target promoters and the effects of deletion of UME6 display many similarities to those observed upon deletion of RPD3 41. We used a dataset obtained by Fazzio et al. 37 to determine whether deletion of UME6 differentially affects GC- and AT-rich genes. Interestingly, we did not find significant base composition-dependent changes in gene expression (Figure 5). Therefore, Ume6p-dependent recruitment does not appear to be involved in base composition-dependent activity of Rpd3p. We propose that the non-targeted global activity of Rpd3p affects transcription and chromatin conformation in a base composition-dependent manner.

To determine whether the base composition-dependent effects of Rpd3p are direct and not due to indirect effects of altered expression of a downstream target gene, we analyzed the patterns of Rpd3p binding and Rpd3p-mediated histone H4 deacetylation. Relative levels of Rpd3p-binding throughout the yeast genome have been determined by Humphrey et al. 42. Using these data, we determined the relative average levels of Rpd3p binding to genes in each of the six base-composition-based groups defined above (Figure 1b). We found that the level of bound Rpd3p is significantly higher for the most GC-rich genes than for the rest of the genome (P < 0.01; Figure 6a).

For analysis of Rpd3p-mediated histone H4 acetylation, we employed a dataset of Bernstein et al. 30, who analyzed H4 acetylation levels in intergenic regions throughout the genome in wild-type and rpd3Δ cells. We found that elimination of Rpd3p increases H4 acetylation of GC-rich genes more strongly than that of AT-rich genes (Figure 6b). These observations imply that Rpd3p binds more strongly to GC-rich genes, resulting in lower levels of histone acetylation and, thereby, directly affects chromatin conformation and expression level of GC-rich genes.

To determine whether the base composition-dependent effect of rpd3Δ is due to loss of histone deacetylase activity, we investigated the effect of treatment with the histone deacetylase inhibitor trichostatin A (TSA) on expression of GC- and AT-rich genes in wild-type cells. Bernstein and co-workers 41 have analyzed genome-wide changes in gene expression at various time points after addition of TSA. We have analyzed their data in the same way as described above to determine whether TSA treatment differentially affects expression of GC-rich and AT-rich genes. We found that after 30 and 60 minutes of exposure to TSA, GC-rich genes are more activated than AT-rich genes, whereas no such effect was observed after 15 minutes (Figure 7a-c). This result confirms that histone deacetylation plays an important role in differentially modulating GC- and AT-rich genes. We do note that the base composition-dependent effect of TSA treatment occurs more slowly than up-regulation of specific Rpd3p target genes, which is already observed after 15 minutes of TSA treatment 41.

TSA inhibits not only Rpd3p, but also another globally acting histone deacetylase, Hda1p. Therefore, to determine whether the base composition-dependent effect of TSA treatment is a result of inhibition of Rpd3p as well as Hda1p, or is specifically due to inhibition of Rpd3p only, we analyzed the effects of deletion of HDA1. Using an expression dataset obtained by Bernstein et al. 41, we found that deletion of HDA1 does not result in base composition-dependent changes in gene expression (Figure 7d). This result suggests that base-composition sensitive activity is specific for the histone deacetylase activity of Rpd3p.

Yeast strains were derived from SK1. The genotype of the wild-type strain (NKY2997) is: MATa, ho::LYS2, lys2, ura3, nuc1::LEU2. The genotype of the rpd3Δ strain (JDY172) is MATa, ho::LYS2, lys2, ura3, nuc1::LEU2, rpd3::KanMX4. The RPD3 gene was deleted using a standard PCR based gene disruption strategy 48.

Nuclei were isolated from G1-arrested cells and 3C analysis was performed as described previously 14. A control PCR template was generated by digestion and random ligation of purified yeast genomic DNA (NKY2997). All templates were titrated to determine the linear range of PCR. PCR products were quantified on 1.5% agarose gels in the presence of ethidium bromide. Interaction frequencies were calculated in triplicate by determination of the ratio of the amount of PCR product obtained with the 3C template divided by the amount of PCR product obtained with the control template 14. Reproducibility was confirmed by analysis of three independent wild-type cultures and three rpd3Δ cultures. Data from wild-type cells and rpd3Δ cells was normalized so that the average Log of the fold difference in interaction frequencies in rpd3Δ cells compared to wild-type cells for loci in the AT-rich domain of chromosome III was zero.

The chromosomal position of each open reading frame was defined as its 5' start site. Base composition was determined using EMBOSS Isochore 49 with a window size of 4 kb and a step size of 1 kb. As a result, base composition values are assigned every 1 kb. To determine the regional base composition of each gene, the base composition value determined for the position closest to its 5' start site was chosen.

For each open reading frame the fold change in transcription in an rpd3Δ mutant compared to wild type was obtained from a dataset generated by Fazzio and co-workers 37 and log-transformed. Sliding window analysis was used to plot regional effects of deletion of RPD3 along chromosomes III (Figure 4a). A window of 30 kb was selected. The step size was 1 open reading frame.

Steady-state transcription levels in exponentially growing wild-type cells were obtained by Bernstein and co-workers 30. Histone acetylation levels in wild-type cells were determined by Kurdistani and co-workers 36. Rpd3p binding to intergenic regions was determined using data obtained by Humphrey and co-workers 42. The change in the level of histone H4 acetylation of intergenic regions in rpd3Δ cells was determined using data obtained by Bernstein et al. 30. Rpd3p binding and change in acetylation levels were assigned to a downstream open reading frame as described in 3042. Datasets describing the effect of TSA treatment on gene expression were obtained by Bernstein et al. 41. Data were log-transformed and zero-centered.