Histone modification levels are predictive for gene expression

Abstract

Histones are frequently decorated with covalent modifications. These histone modifications are thought to be involved in various chromatin-dependent processes including transcription. To elucidate the relationship between histone modifications and transcription, we derived quantitative models to predict the expression level of genes from histone modification levels. We found that histone modification levels and gene expression are very well correlated. Moreover, we show that only a small number of histone modifications are necessary to accurately predict gene expression. We show that different sets of histone modifications are necessary to predict gene expression driven by high CpG content promoters (HCPs) or low CpG content promoters (LCPs). Quantitative models involving H3K4me3 and H3K79me1 are the most predictive of the expression levels in LCPs, whereas HCPs require H3K27ac and H4K20me1. Finally, we show that the connections between histone modifications and gene expression seem to be general, as we were able to predict gene expression levels of one cell type using a model trained on another one.

Fig. 1.

Modeling framework. Models are equations that linearly relate the levels of histone modifications to the measured expression value. corresponds to a vector of length L (the number of promoters), where the components are the transformed levels of a histone modification i (, with N_i representing the number of tags in each promoter), a is the y intercept , and the b_i to the slope associated with . y denotes a vector of length L whose components are the expression values. In the one-modification models, i can be any of the 39 modifications or two control IgG antibodies. In the two-modifications models, i and j are chosen to cover all combinations of two modifications without repetition. In the three-modifications models, i, j, and k are chosen to cover all combinations of three modifications without repetition. The full model incorporates all 41 variables.

Fig. 2.

Quantitative relationship between histone modification levels and expression. (A) Scatterplot with the predicted expression value in CD4+ T-cells using the full linear model on the x axis and the measured expression value in CD4+ T-cells on the y axis. The shades of blue indicate the density of points; the darker color, the more points. The red line indicates the linear fit between predicted and measured expression (y = 0.99x + 0.02), which are highly correlated (r = 0.77), indicating a quantitative relationship between levels of histone modifications at the promoter and gene expression levels. (B) Comparison of prediction accuracy between all possible one-modification, two-modifications, three-modifications models, and the full model for CD4+ T-cells. Models are sorted by ascending prediction accuracy along the x axis. The best models using only a small subset of modifications almost reach the prediction accuracy of the full linear model. (C) Bar plot showing the frequency of appearance of different histone modifications in best scoring three-modifications models (142 models) for CD4+ T-cells. Best scoring models are defined as reaching at least 95% of prediction accuracy of the full linear model. (D, E) Expression values of genes, which were at least 5-fold up or down regulated in CD36+ and CD133+ cells with respect to CD4+ T-cells, predicted using model parameters trained on data from CD4+ T-cells. The predicted and measured expression values are highly correlated for both CD36+ (D) (r = 0.75; 1,412 genes) and CD133+ (E) (r = 0.63; 1,243 genes) cells. The equations of the regression line for both CD36+ and CD133+ cells (y = 0.43x + 6.04 and y = 0.53x + 6.17, respectively) show a high value of the intercept and a slope different from one due to the fact that the levels of the histone modifications were not normalized across cell types.

Fig. 3.

Differences in transcriptional regulation between HCPs and LCPs. (A, B) Bar plots showing the frequency of appearance of different histone modifications in best scoring three-modifications models for HCPs (B) (50 models) and LCPs (C) (40 models) in CD4+ T-cells. Best scoring models are defined as reaching at least 95% of prediction accuracy of the full model trained on HCPs and LCPs, respectively. Only the top ten modifications are depicted. (C) Normalized cumulative tag counts in the region of -500 base pairs to 3,000 base pairs surrounding the transcription start site of RefSeq genes in CD4+ T-cells for the five important modifications identified by our analysis.