2017-胞嘧啶甲基化对人类转录因子DNA结合特异性的影响.pdf

RESEARCH ARTICLE SUMMARYDNA METHYLATIONImpact of cytosine methylation onDNA binding specificities of humantranscription factorsYimeng Yin,Ekaterina Morgunova,Arttu Jolma,Eevi Kaasinen,Biswajyoti Sahu,Syed Khund-Sayeed,Pratyush K.Das,Teemu Kivioja,Kashyap Dave,Fan Zhong,Kazuhiro R.Nitta,Minna Taipale,Alexander Popov,Paul A.Ginno,Silvia Domcke,Jian Yan,Dirk Schbeler,Charles Vinson,Jussi Taipale*INTRODUCTION:Nearly all cells in the humanbody share the same primary genome sequenceconsisting of four nucleotide bases.One of thebases,cytosine,is commonly modified by meth-ylation of its 5 position in CpG dinucleotides(mCpG).Most CpG dinucleotides in the humangenome are methylated,but the level of CpGmethylation varies with genetic location(pro-moter versus gene body),whether genes areactive versus silenced,and cell type.Researchhas shown that the maintenance of a particularcellular state after cell division is dependent onfaithful transmission of methylated CpGs,aswell as inheritance of the mother cells rep-ertoire of transcription factors by the daughtercells.Thesetwomechanisms ofepigenetic inher-itance are linked to each other;the bindingof transcription factors can be affected by cy-tosine methylation,and cytosine methylationcan,in turn,be added or removed by proteinsthat associate with transcription factors.RATIONALE:The genetic and epigenetic lan-guage,which imparts when and where genesare expressed,is understood at a conceptuallevel.However,a more detailed understandingisneeded of thegenomicregulatorymechanismby which methylated cytosines affect transcrip-tionfactorbinding.Becausecytosinemethylationchanges DNA structure,it has the potentialto affect binding of all transcription factors.However,a systematic analysis of binding ofa large collection of transcription factors toall possible DNA sequences has not previouslybeen conducted.RESULTS:To globally characterize the effectofcytosinemethylationontranscriptionfactorbinding,we systematically analyzed bindingspecificities of full-length transcription factorsand extended DNA binding domains to un-methylated and CpG-methylated DNA by usingmethylation-sensitive SELEX(systematic evo-lution of ligands by exponential enrichment).We evaluated binding of 542 transcription fac-torsandidentifiedalargenumberofpreviouslyuncharacterizedtranscriptionfactorrecognitionmotifs.Bindingofmostma-jorclassesoftranscriptionfactors,including bHLH,bZIP,and ETS,wasinhib-itedbymCpG.Incontrast,transcriptionfactorssuchas homeodomain,POU,and NFAT proteins preferred to bind methyl-ated DNA.This class of binding was enrichedinfactors withcentralrolesinembryonicandorganismal development.The observed binding preferences were val-idated using several orthogonal methods,in-cluding bisulfite-SELEX and protein-bindingmicroarrays.In addition,the preference of thepluripotency factor OCT4 to bind to a mCpG-containing motif was confirmed by chromatinimmunoprecipitation analysis in mouse embry-onic stem cells with low or high levels of CpGmethylation(due to deficiency in all enzymesthat methylate cytosines or contribute to theirremoval,respectively).Crystal structure analysisof the homeodomain proteins HOXB13,CDX1,CDX2,and LHX4 revealed three key residuesthat contribute to the preference of this devel-opmentally important family of transcriptionfactors for mCpG.The preference for bindingto mCpG was due to direct hydrophobic inter-actions with the 5-methyl group of methylcyto-sine.In contrast,inhibition of binding of othertranscription factors to methylated sequenceswas found to be caused by steric hindrance.CONCLUSION:Our work constitutes a globalanalysis of the effect of cytosine methylationon DNA binding specificities of human tran-scription factors.CpG methylation can influ-ence binding of most transcription factors toDNAin some cases negatively and in otherspositively.Ourfindingthatmanydevelopmen-tally important transcription factors prefer tobind to mCpG sites can inform future analysesof the role of DNA methylation on cell differ-entiation,chromatinreprogramming,andtran-scriptional regulation.RESEARCHYin et al.,Science 356,502(2017)5 May 20171 of 1The list of author affiliations is available in the full article online.*Corresponding author.Email:jussi.taipaleki.seCite this article as Y.Yin et al.,Science 356,eaaj2239(2017).DOI:10.1126/science.aaj2239Systematic analysis of the impact of CpG methylation on transcription factor binding.The bottom left panel shows the fraction of transcription factors that prefer methylated(orange)or unmethylated(teal)CpG sites,are affected in multiple ways(yellow),are not affected(green),or do not have a CpG in their motifs(gray),as determined by methylation-sensitive SELEX(topleft).The structure and logos on the right highlight how HOXB13 recognizes mCpG(blue shadingindicates a CpG affected by methylation).ON OUR WEBSITERead the full articleat http:/dx.doi.org/10.1126/science.aaj2239.on December 27,2018 http:/science.sciencemag.org/Downloaded from RESEARCH ARTICLEDNA METHYLATIONImpact of cytosine methylation onDNA binding specificities of humantranscription factorsYimeng Yin,1Ekaterina Morgunova,1Arttu Jolma,1Eevi Kaasinen,1Biswajyoti Sahu,2Syed Khund-Sayeed,3Pratyush K.Das,2Teemu Kivioja,2Kashyap Dave,1Fan Zhong,1Kazuhiro R.Nitta,1Minna Taipale,1Alexander Popov,4Paul A.Ginno,5Silvia Domcke,5,6Jian Yan,1Dirk Schbeler,5,6Charles Vinson,3Jussi Taipale1,2*The majority of CpG dinucleotides in the human genome are methylated at cytosine bases.However,active gene regulatory elements are generally hypomethylated relative to theirflanking regions,and the binding of some transcription factors(TFs)is diminished bymethylation of their target sequences.By analysis of 542 human TFs with methylation-sensitive SELEX(systematic evolution of ligands by exponential enrichment),we foundthat there are also many TFs that prefer CpG-methylated sequences.Most of these are inthe extended homeodomain family.Structural analysis showed that homeodomainspecificity for methylcytosine depends on direct hydrophobic interactions with themethylcytosine 5-methyl group.This study provides a systematic examination of the effectof an epigenetic DNA modification on human TF binding specificity and reveals that manydevelopmentally important proteins display preference for mCpG-containing sequences.The methylation of cytosine at CpG dinu-cleotides(mCpG)plays an important rolein the regulation of human genome archi-tecture and activity.Most CpG dinucleotidesin mammalian genomes are methylated,but the methylation pattern is not uniform.Nucleosome-associated DNA has a lower rate ofmethylation than the more accessible linker se-quences located between nucleosomes(1,2).Inaddition,methylation patterns vary between celltypes(3),and the changes correlate with geneexpression.Gene bodies of highly expressed genesare heavily methylated(4,5),whereas active generegulatory elements have a low degree of meth-ylation(68).Methylation of DNA is thought to regulatetranscription both directly and indirectly.CpGmethylation can directly repress transcriptionby preventing binding of some transcription fac-tors(TFs)to their recognition motifs for example,(912).In addition,mCpG dinucleotides can berecognized by a specific class of proteins,themethyl-CpG domainbinding proteins,some ofwhich can recruit histone deacetylases and arethoughtto promote local chromatin condensation(7,13).It is generally thought that methylationserves as a barrier to reprogramming(14,15)andthat binding of TFs to previously methylated sites(16)or removal of CpG methylation(7)is involvedin cellular differentiation or reprogramming.Methylation patterns are inherited across celldivisions.This is accomplished by the methyl-transferase DNMT1,which associates with theDNA replication fork,methylating the newlysynthesized DNA strand at positions where itstemplate strand is methylated(7).This process,together with the inheritance of cytoplasmic de-terminants such as TFs,forms the basis of theepigenetic memory that allows cellular inheri-tance of acquired characteristics,such as thestate of differentiation for example,(14).Therole of TFs and DNA methylation in epigeneticinheritance is thus well established.Severalstudies have also characterized the interplaybetween these key determinants by analyzingbinding of individual TFs to methylated sites(9,10,12,17,18)and/or analyzing binding ofmultiple TFs to a limited number of sequences(19,20).However,systematic analysis of bind-ing of a large collection of TFs to all possibleDNA sequences hasso far not been conducted.HT-SELEX in the presence andabsence of CpG methylation revealsTF binding specificitiesTo globally characterize the effect of cytosinemethylation on TF binding,we performed a HT-SELEX high-throughput systematic evolutionof ligands by exponential enrichment(21,22)analysis of 1000 human TF extended DNA bind-ing domains(eDBDs;details are given in table S1and the methods)and 550 full-length TFs.Thiscollection included 84%of the high-confidenceTFs described by Vaquerizas et al.(23)(classes aand b;table S1).The assay was performed withunmethylated DNA ligands and DNA ligands thatwere methylated(24)using the CpG-specific cy-tosine 5-methylase M.SssI before each selectioncycle(see the methods for details;Fig.1A).Theligands contained a 40base pair(bp)randomsequence and were sequenced before the assayand after each selection cycle.The resulting datawere analyzed using the previously describedAutoseed pipeline,a de novo binding motif dis-covery method based on identification of distinctseed sequences that are subsequently used togenerate position weight matrix(PWM)modelssee(25)and the methods.We(21,26,27)andothers have previously established that motifsgenerated using HT-SELEX are similar to thoseobtained using other state-of-the-art methods,such as protein-binding microarrays or bacterialone-hybrid assays(28,29),and are biologicallyrelevant on the basis of their ability to predictTF binding in vivo(3032).Comparedwiththeothermethods,HT-SELEXis able to detect longer binding motifs because ofthe high complexity of the input library(26).Although SELEX measures enrichment of se-quences and not affinity of binding per se,theorder in which sequences are enriched is thesame as the order of their affinities.In addition,we have previously shown that the motifs ob-tained from early HT-SELEX cycles are similarto those obtained by methods that more directlymeasure affinity,such as oligonucleotide compe-tition assays(33)or assays that compare enrich-ment across a single SELEX cycle(26).Thus,themotifs presented here can be directly used formotif matching using a threshold,where onlythe rank of the affinities affects the outcome.However,obtained scores should be consideredas estimates rather than true affinity measures.ThemediansuccessrateofmotifdiscoveryperTF was 47%,and in total,data were obtained for444 eDBDs and 227 full-length TFs(table S2).Asdescribed previously(26),a relatively low rate ofsuccess was observed for C2H2 zinc finger pro-teins,SMAD proteins,and SANT/Myb proteins,likely becauseofthe verylongrecognitionmotifsof some C2H2 proteins,the factthat SMAD pro-teins act as obligate heterotrimers,and the mis-classificationofmanySANT/MybproteinsasTFsdespitetheirlack ofkeyaminoacidsrequiredforDNA binding(25,34).The median coverage for the TF families was60%(Fig.1B).ThecoverageofindividualTFswasconsiderably higher than that reported in previ-ous systematic studies(26,3537).For example,thisstudy,using both unmethylated and methyl-ated ligands,and our previously published HT-SELEX data(26)respectively recovered modelsfor 542 and 411 TFs,representing 343 and 239distinctly different specificities(Fig.2 and tableS2).The motifs obtained inthisstudy were highlyRESEARCHYin et al.,Science 356,eaaj2239(2017)5 May 20171 of 151Division of Functional Genomics and Systems Biology,Department of Medical Biochemistry and Biophysics,Karolinska Institutet,SE 141 83 Stockholm,Sweden.2Genome-Scale Biology Program,Post Office Box 63,FI-00014 University of Helsinki,Helsinki,Finland.3Laboratory ofMetabolism,National Cancer Institute,National Institutes ofHealth,Room 3128,Building 37,Bethesda,MD 20892,USA.4European Synchrotron Radiation Facility,38043 Grenoble,France.5Friedrich-Miescher-Institute for Biomedical Research(FMI),Maulbeerstrasse 66,4058 Basel,Switzerland.6Facultyof Science,University of Basel,Petersplatz 1,4003 Basel,Switzerland.*Corresponding author.Email:jussi.taipaleki.seon December 27,2018 http:/science.sciencemag.org/Downloaded from consistentwithearlierdataforTFsthathavebeenstudied previously(figs.S1 and S2A).Most motifsthat were different from any of the previouslydeterminedHT-SELEXmotifswereforC2H2zincfinger proteins whose specificity had not beendetermined earlier or represented known TFswhose preference for mCpG was not previouslyrecognized,such as the motifs newly identifiedfor a collection of homeodomain proteins in thisstudy(fig.S2).In total,HT-SELEX models nowexist for 632 of the 1400(23)human TFs.Analysis of the new HT-SELEX data also re-vealed a mechanism of evolution of TF bindingspecificityutilized by the BARX homeodomain,ELF3 ETS-family TFs,and bHLH-family TFNEUROD1wherebytheadditionofanAT-hookdomain next to the main DBD results in pref-erence for a flanking AT-rich sequence(Fig.3).CpG methylation has a major effect onTF-DNA bindingUsing the methyl-SELEX process(see the meth-ods),it is possible to determine the effect ofCpG methylation on TF binding,if the TF en-riches both sequences that do and do not containCpG dinucleotide(s).It is,however,difficult todetermine the effect of methylation on TFs thathave a very strong or an absolute requirementfor a CpG in their motif,because they can yieldthe same CpG-containing motif or fail to yieldany motif when DNA is methylated.To allowmeasurement of the effect of CpG methylationin such cases and to validate the results ofmethyl-SELEX,we subjected most of the TFswhose recognition motifs contained CpG se-quences to one cycle of bisulfite-SELEX(Fig.4A;see the methods).Using this method,it is possibleto determine in one reaction the preference ofTFs for unmethylated or methylated forms oftheir CpG-containing recognition sequences.Theresults from bisulfite-SELEX confirmed most ofthe results of methyl-SELEX,and because of itshigher sensitivity,bisulfite-SELEX also revealedmany additional TFs with preferences for un-methylated or methylated CpG(for a compar-ison,see table S3).Combining the results of methyl-SELEX andbisulfite-SELEX showed that some TFs did notrecognize sites with CpG sequences,and theirbinding was thus not influenced by CpG methyl-ation(fig.S3A;“no CpG”class).A second class ofTFs recognized CpG-containing sequences,butmethylation of the CpG had little effect onbinding(fig.S3B;“little effect”class)(38,39).A third class ofTFs did not bind to,or boundmoreweakly to,methylated versions of their recog-nition sequences(Fig.4,B and C,and fig.S3C;“methyl-minus”class).In most cases(82%),t