Statistical assessment of the global regulatory role of histone acetylation in Saccharomyces cerevisiae

Abstract

Background: Histone acetylation plays important but incompletely understood roles in gene regulation. A comprehensive understanding of the regulatory role of histone acetylation is difficult because many different histone acetylation patterns exist and their effects are confounded by other factors, such as the transcription factor binding sequence motif information and nucleosome occupancy.

Results: We analyzed recent genomewide histone acetylation data using a few complementary statistical models and tested the validity of a cumulative model in approximating the global regulatory effect of histone acetylation. Confounding effects due to transcription factor binding sequence information were estimated by using two independent motif-based algorithms followed by a variable selection method. We found that the sequence information has a significant role in regulating transcription, and we also found a clear additional histone acetylation effect. Our model fits well with observed genome-wide data. Strikingly, including more complicated combinatorial effects does not improve the model's performance. Through a statistical analysis of conditional independence, we found that H4 acetylation may not have significant direct impact on global gene expression.

Conclusion: Decoding the combinatorial complexity of histone modification requires not only new data but also new methods to analyze the data. Our statistical analysis confirms that histone acetylation has a significant effect on gene transcription rates in addition to that attributable to upstream sequence motifs. Our analysis also suggests that a cumulative effect model for global histone acetylation is justified, although a more complex histone code may be important at specific gene loci. We also found that the regulatory roles among different histone acetylation sites have important differences.

Figure 1

Model validation by comparing the R² for the real versus randomly permutated datasets. The R² obtained by applying the motif selection and fitting equation 2 (with sequence motif information only) procedures to randomly permutated and real data. The histogram is obtained based on 50 randomly permutated samples. The arrow on the right marks the R² for the real data. Results for the coding regions are represented here. See the main text for details.

Figure 2

Dependency of transcription rates on histone acetylation levels (ac) after controlling for confounding effects. (a) Transcription rates versus intergenic H3K9 and K14 acetylation levels controlling for H4 acetylation levels. (b) Transcription rates versus intergenic H4 acetylation levels controlling for H3K9 and K14 acetylation levels. (c) Same as (a) except that coding region histone acetylation data are used. (d) Same as (b) except that coding region histone acetylation data are used. All data are log-transformed. Genes are sorted by transcription levels. A sliding smoothing window of 20 genes is applied to the transcription rates and histone acetylation data.