Modeling gene regulation from paired expression and PNAS PLUS chromatin accessibility data Zhana Durena,b,c, Xi Chenb, Rui Jiangd,1, Yong Wanga,c,1, and Wing Hung Wongb,1 aAcademy of Mathematics and Systems Science, National Center for Mathematics and Interdisciplinary Sciences, Chinese Academy of Sciences, Beijing 100080, China; bDepartment of Statistics, Department of Biomedical Data Science, Bio-X Program, Stanford University, Stanford, CA 94305; cSchool of Mathematical Sciences, University of Chinese Academy of Sciences, Beijing 100049, China; and dMinistry of Education Key Laboratory of Bioinformatics, Bioinformatics Division and Center for Synthetic & Systems Biology, Tsinghua National Laboratory for Information Science and Technology, Department of Automation, Tsinghua University, Beijing 100084, China Contributed by Wing Hung Wong, May 8, 2017 (sent for review March 20, 2017; reviewed by Christina Kendziorski and Sheng Zhong) The rapid increase of genome-wide datasets on gene expression, gene expression data, accessibility data are available for a diverse set chromatin states, and transcription factor (TF) binding locations offers of cellular contexts (Fig. 1, blue boxes). In fact, we expect the an exciting opportunity to interpret the information encoded in amount of matched expression and accessibility data (i.e., measured genomes and epigenomes. This task can be challenging as it requires on the same sample) will increase very rapidly in the near future. joint modeling of context-specific activation of cis-regulatory ele- The purpose of the present work is to show that, by using ments (REs) and the effects on transcription of associated regulatory matched expression and accessibility data across diverse cellular factors. To meet this challenge, we propose a statistical approach contexts, it is possible to recover a significant portion of the in- based on paired expression and chromatin accessibility (PECA) data formation in the missing data on binding location and chromatin across diverse cellular contexts. In our approach, we model (i)the state and to achieve accurate inference of the gene regulatory localization to REs of chromatin regulators (CRs) based on their in- relations. In our approach, key events in the regulatory process, teraction with sequence-specific TFs, (ii) the activation of REs due to such as recruitment of chromatin remodeling factors to a regu- CRs that are localized to them, and (iii) the effect of TFs bound to latory element, activation of regulatory elements, etc., are activated REs on the transcription of target genes (TGs). The transcrip- regarded as latent unobserved variables in a statistical model that tional regulatory network inferred by PECA provides a detailed view describes the relations among these variables and the gene ex- trans cis of how -and -regulatory elements work together to affect pression variables, conditional on accessibility data on the regula- BIOPHYSICS AND gene expression in a context-specific manner. We illustrate the fea- tory elements. By fitting this model to expression and accessibility COMPUTATIONAL BIOLOGY sibility of this approach by analyzing paired expression and accessi- data across a large number of cellular contexts, we can infer many bility data from the mouse Encyclopedia of DNA Elements (ENCODE) details of the gene regulatory system helpful in the interpretation of and explore various applications of the resulting model. new data or the generation of new hypotheses. We end this Introduction with comments on related works. gene regulation | transcription factor | regulatory element | Several methods have recently been proposed to detect tran- chromatin regulator | chromatin activity scription factor (TF) binding sites by “footprinting” in which the presence of a bound TF is reflected by the shape of the DNase- ver since the emergence of high-throughput gene expression seq (or ATAC-seq) profile around its binding site (6–8). These Eexperiments (1), computational biologists have been inter- works focus on the effect of TF binding on the frequency of ested in the inference of gene regulatory relationships from gene cleavage near the site and do not attempt to model gene regulatory expression data across diverse cellular contexts corresponding to diverse cell types and experimental conditions (Fig. 1, red boxes). Significance However, progress has been hindered by the fact that gene expres- sion measurements provide little information on underlying regula- tory mechanisms such as transcription factor binding and chromatin Chromatin plays a critical role in the regulation of gene ex- modification. To fill this gap, chromatin immunoprecipitation- pression. Interactions among chromatin regulators, sequence- based methods (2, 3) have been developed for the genome-wide specific transcription factors, and cis-regulatory sequence ele- mapping of transcriptional regulator binding locations and the ments are the main driving forces shaping context-specific detection of epigenetic marks characteristic of specific chromatin chromatin structure and gene expression. However, because states. For example, by performing thousands of ChIP-seq exper- of the large number of such interactions, direct data on them iments, the Encyclopedia of DNA Elements (ENCODE) consor- are often missing in most cellular contexts. The purpose of the tium has generated such data for many chromatin marks and present work is to show that, by modeling matched expression transcriptional regulators on a small number of cell lines (Fig. 1, and accessibility data across diverse cellular contexts, it is green boxes). However, because a large number of transcriptional possible to recover a significant portion of the information in regulators and chromatin marks have to be analyzed one by one, it the missing data on binding locations and chromatin states and is unlikely that such comprehensive data will become available for to achieve accurate inference of gene regulatory relations. many other cell lines. For most cellular contexts, the desired data will remain missing in the foreseeable future (Fig. 1, gray boxes). Author contributions: R.J., Y.W., and W.H.W. designed research; Z.D. and X.C. performed – research; Z.D., R.J., Y.W., and W.H.W. analyzed data; and Z.D., R.J., Y.W., and W.H.W. On the other hand, it is known that many of the protein DNA wrote the paper. interactions important for gene regulation occur in regulatory Reviewers: C.K., University of Wisconsin; and S.Z., University of California, San Diego. elements (REs) such as enhancers and insulators, which com- The authors declare no conflict of interest. pose only a small portion of the noncoding sequences in a ge- nome. The REs active in gene regulation in a given cellular state Freely available online through the PNAS open access option. tend to have an open chromatin structure so that they are ac- Data deposition: The sequence data reported in this paper have been deposited in the Gene Expression Omnibus (GEO) database, https://www.ncbi.nlm.nih.gov/geo (accession cessible for binding by relevant transcriptional regulators. This no. GSE98479). suggests that many of the relevant regulatory relations may be 1To whom correspondence may be addressed. Email: [email protected], ruijiang@ revealed by analyzing the accessible REs. Fortunately, genome-wide tsinghua.edu.cn, or [email protected]. measurement of chromatin accessibility is now straightforward by This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. recentmethodssuchasDNase-seq(4)orATAC-seq(5).Similarto 1073/pnas.1704553114/-/DCSupplemental. www.pnas.org/cgi/doi/10.1073/pnas.1704553114 PNAS Early Edition | 1of10 Downloaded by guest on September 25, 2021 ChIP-seq Biological RNA-seq DNase-seq Ctcf Zmiz1 Bhlhe40 Cebpb Chd1 Chd2 Ep300 Fli1 Gabpa Gata1 Gata2 Hcfc1 Jund Mafk Maz Myb Myod1 Myog Pax5 Rad21 Rdbp Sin3a Srf Tbp Tcf12 Ubtf Usf1 Usf2 Zc3h11a Zkscan1 Znf384 Fosl1 Kat2a Max Polr2a Smc3 system Cell types ENCODE sample ID E2f4 Ets1 Jun Mxi1 Myc Rcor1 Rest Tal1 Tcf3 Muscular SkMuscle SkmuscleC57bl6MAdult8wks Circulatory G1E-ER4 G1eer4S129ME0Diffd24h G1 E G1eS12 9ME0 Cerebrum CerebrumC57bl6MAdult8wks Nervous Cerebellum CerebellumC57bl6MAdult8wks WholeBrain WbrainC57bl6ME18half Respiratory Lung LungC57bl6MAdult8wks NIH-3T3 Nih3t3NihsMImmortal LgIntestine LgintC57bl6MAdult8wks Liver liver129dlcrME14half Digestive Liver LiverC57bl6MAdult8wks Liver LiverC57bl6ME14half Excretory Kidney KidneyC57bl6MAdult8wks Endocrine FatPad FatC57bl6MAdult8wks GenitalFatPad GfatC57bl6MAdult8wks 416B 416bC57bl6MAdult8wks A20 A20BalbcannMAdult8wks B-cell(CD19+) Bcellcd19pC57bl6MAdult8wks B-cell(CD43-) Bcellcd43nC57bl6MAdult8wks MEL MelC57bl6MAdult8wks Spleen SpleenC57bl6MAdult8wks Lymphatic Thymus ThymusC57bl6MAdult8wks T-Naïve TnaiveC57bl6MAdult8wks Fig. 1. Genome-wide data for gene-regulatory inference. Each row represents a particular cellular context under which multiple types of genome-wide data may be available. In this paper we illustrate our method by analyzing data from 25 contexts studied in the mouse ENCODE project covering a variety of mouse cell types and developmental stages. Expression data (from RNA-seq, red boxes) and chromatin accessibility data (from DNase-seq or ATAC-seq, blue boxes) are available for each context, but most of the location data (from ChIP-seq data, green boxes) for transcriptional regulators are missing. We expect that the number of contexts (i.e., number of rows) with expression and accessibility data will increase rapidly in