Quantitative Models Identify Histone Signatures of Poised Genes Prior to Cellular Differentiation

Background: Recent studies have shown that histone marks are involved in pre-programming gene fates during cellular differentiation. Bivalent domains (marked by both H3K4me3 and H3K27me3) have been proposed to act in the histone prepatterning of poised genes; however, bivalent genes could also resolve into monovalent silenced states during differentiation. Thus, the histone signatures of poised genes need to be more precisely characterized.

Results: Using a support vector machine (SVM), we show that the collective histone modification data from human blood hematopoietic cells (HSCs) could predict poised genes with fairly high predictive accuracy within the model of directed erythrocyte differentiation from HSCs. Surprisingly, models with single histone marks (e.g., H3K4me3 or H2A.Z) could reach comparable predictive powers to the full model built with all of the nine histone marks. We also derived an H2A.Z and H3K9me3-based Naive Bayesian model for inferring poised genes, and the validity of this model was supported by data from several other pluripotent/multipotent cells (including mouse ES cells).

Conclusion: Our work represents a systematic quantitative study that verified that histone marks play a role in pre-programming the activation or repression of specific genes during cellular differentiation. Our results suggest that the relative quantities of H2A.Z modification and H3K9me3 modification are important in determining a corresponding gene’s fate during cellular differentiation.