Predicting enhancer transcription and activity from chromatin modifications - PubMed (original) (raw)

. 2013 Dec;41(22):10032-43.

doi: 10.1093/nar/gkt826. Epub 2013 Sep 12.

Affiliations

Predicting enhancer transcription and activity from chromatin modifications

Yun Zhu et al. Nucleic Acids Res. 2013 Dec.

Abstract

Enhancers play a pivotal role in regulating the transcription of distal genes. Although certain chromatin features, such as the histone acetyltransferase P300 and the histone modification H3K4me1, indicate the presence of enhancers, only a fraction of enhancers are functionally active. Individual chromatin marks, such as H3K27ac and H3K27me3, have been identified to distinguish active from inactive enhancers. However, the systematic identification of the most informative single modification, or combination thereof, is still lacking. Furthermore, the discovery of enhancer RNAs (eRNAs) provides an alternative approach to directly predicting enhancer activity. However, it remains challenging to link chromatin modifications to eRNA transcription. Herein, we develop a logistic regression model to unravel the relationship between chromatin modifications and eRNA synthesis. We perform a systematic assessment of 24 chromatin modifications in fetal lung fibroblast and demonstrate that a combination of four modifications is sufficient to accurately predict eRNA transcription. Furthermore, we compare the ability of eRNAs and H3K27ac to discriminate enhancer activity. We demonstrate that eRNA is more indicative of enhancer activity. Finally, we apply our fibroblast trained model to six other cell-types and successfully predict eRNA synthesis. Thus, we demonstrate the learned relationships are general and independent of cell-type. We provided a powerful tool to identify active enhancers and reveal the relationship between chromatin modifications, eRNA production and enhancer activity.

PubMed Disclaimer

Figures

Figure 1.

Figure 1.

Methodological overview. (a) The levels of 24 chromatin marks at each P300-bound enhancer were used as inputs to optimize a logistic regression model that best separates eRNA+ (positive) and eRNA− (negative) training samples. (b) Sample profiles of the 24 histone modifications for the eRNA+ (left) and eRNA− (right) enhancers. Both enhancers are enriched with H3K4me1 but depleted of H3K4me3. The left enhancer is actively producing bi-directional eRNA transcripts and is enriched with histone acetylations and H3K79me1 but depleted of H3K27me3 and H3K9me3. No eRNA transcripts were detected on the right enhancer locus, which is enriched in H3K27me3 and H3K9me3 but depleted of histone acetylations and H3K79me1.

Figure 2.

Figure 2.

Chromatin modifications predict eRNA synthesis. (a) AUCs obtained using the full and top four-modification models. (b) Distribution of AUCs and MCCs using all _m_-modification models with m taken from 1 to 6 and from 18 to 24. Because the number of combinations of 7–17 variables was too high, they were omitted. The red curves represent AUC and MCC from the top _m_-modification model, respectively. (c) Five of the top four-modification models with their AUCs and MCCs. (d) The number of acetylations in the best-scored four-modification models. The best-scored four-modification models were the ones with at least 95% of the AUCs and MCCs obtained with the full model. (e) The frequency of appearance of the 24 chromatin marks in the best-scored four-modification models.

Figure 3.

Figure 3.

eRNA transcription is a robust indicator of active enhancers. (a) Expression levels of genes associated with active and inactive enhancers identified by three different marks: measured eRNAs, predicted eRNAs and H3K27ac. (b) Heatmaps of GRO-seq and H3K27ac signals for four classes of enhancers: eRNA+K27ac+, eRNA+K27ac−, eRNA-K27ac+ and eRNA−K27ac−. The enhancer windows were centered at p300-binding peaks. (c) Average profiles of GRO-seq signals for the four classes of enhancers in (b). (d) Average profiles of H3K27ac signals for the four classes of enhancers in (b). (e) Expression levels of the genes associated with the four classes of enhancers in (b). (f) Luciferase assay testing the enhancer activity for the eRNA+, eRNA−, H3K27ac+ and H3K27ac− enhancers.

Figure 4.

Figure 4.

Application of the logistic regression model across six cell-types. (a) AUCs of the top six- and four-modification models tested on the IMR90 and mESC cells. All models were trained on IMR90 cells. (b) Performance comparison of the top four-modification models, which were trained and tested on the same (IMR90) and different (trained on IMR90 but tested on mESC) cell-types. (c) Of the 18 predicted eRNA+ (P-eRNA+) enhancers, 15 were positive in luciferase reporter assays. Of the 21 predicted eRNA− (P-eRNA−) enhancers, only three were positive in luciferase reporter assays. (d) Average profiles of GRO-seq signals for the four classes of enhancers in mESC cells: eRNA+K27ac+, eRNA+K27ac−, eRNA−K27ac+ and eRNA−K27ac−. The enhancer windows were centered at p300-binding peaks. (e) Average profiles of H3K27ac for the four classes of enhancers in mESC cells. (f) Heatmaps of GRO-seq and H3K27ac signals for the four classes of enhancers in mESC cells. (g) Expression levels of genes associated with the four classes of enhancers in mESC cells. (h) Expression levels of genes associated with the four classes of enhancers in H1, ME, TBL, MSC and NPC cells.

Figure 4.

Figure 4.

Application of the logistic regression model across six cell-types. (a) AUCs of the top six- and four-modification models tested on the IMR90 and mESC cells. All models were trained on IMR90 cells. (b) Performance comparison of the top four-modification models, which were trained and tested on the same (IMR90) and different (trained on IMR90 but tested on mESC) cell-types. (c) Of the 18 predicted eRNA+ (P-eRNA+) enhancers, 15 were positive in luciferase reporter assays. Of the 21 predicted eRNA− (P-eRNA−) enhancers, only three were positive in luciferase reporter assays. (d) Average profiles of GRO-seq signals for the four classes of enhancers in mESC cells: eRNA+K27ac+, eRNA+K27ac−, eRNA−K27ac+ and eRNA−K27ac−. The enhancer windows were centered at p300-binding peaks. (e) Average profiles of H3K27ac for the four classes of enhancers in mESC cells. (f) Heatmaps of GRO-seq and H3K27ac signals for the four classes of enhancers in mESC cells. (g) Expression levels of genes associated with the four classes of enhancers in mESC cells. (h) Expression levels of genes associated with the four classes of enhancers in H1, ME, TBL, MSC and NPC cells.

References

    1. Barski A, Cuddapah S, Cui K, Roh TY, Schones DE, Wang Z, Wei G, Chepelev I, Zhao K. High-resolution profiling of histone methylations in the human genome. Cell. 2007;129:823–837. - PubMed
    1. Wang Z, Zang C, Rosenfeld JA, Schones DE, Barski A, Cuddapah S, Cui K, Roh TY, Peng W, Zhang MQ, et al. Combinatorial patterns of histone acetylations and methylations in the human genome. Nat. Genet. 2008;40:897–903. - PMC - PubMed
    1. Birney E, Stamatoyannopoulos JA, Dutta A, Guigo R, Gingeras TR, Margulies EH, Weng Z, Snyder M, Dermitzakis ET, Thurman RE, et al. Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature. 2007;447:799–816. - PMC - PubMed
    1. Heintzman ND, Hon GC, Hawkins RD, Kheradpour P, Stark A, Harp LF, Ye Z, Lee LK, Stuart RK, Ching CW, et al. Histone modifications at human enhancers reflect global cell-type-specific gene expression. Nature. 2009;459:108–112. - PMC - PubMed
    1. Visel A, Blow MJ, Li Z, Zhang T, Akiyama JA, Holt A, Plajzer-Frick I, Shoukry M, Wright C, Chen F, et al. ChIP-seq accurately predicts tissue-specific activity of enhancers. Nature. 2009;457:854–858. - PMC - PubMed

MeSH terms

Substances

LinkOut - more resources