ReLA, a local alignment search tool for the identification of distal and proximal gene regulatory regions and their conserved transcription factor binding sites.

0.61 MB·English

by  Gonzalez, Santi

MANUSCRIPTCATEGORY: ORIGINALPAPER
BIOINFORMATICS ORIGINAL PAPER Vol.28no.62012,pages763–770
doi:10.1093/bioinformatics/bts024
Genome analysis
AdvanceAccesspublicationJanuary16,2012
ReLA, a local alignment search tool for the identification of distal
and proximal gene regulatory regions and their conserved
transcription factor binding sites
Santi González1†, Bàrbara Montserrat-Sentís1†, Friman Sánchez2,
Montserrat Puiggròs1,3, Enrique Blanco4, Alex Ramirez2 and David Torrents1,5∗
1JointIRB-BSCprogramonComputationalBiology,BSCc/JordiGirona29,2DepartmentofComputerArchitecture,
Campus Nord UPC D6-117, Jordi Girona 1-3, 08034 Barcelona, 3Computational Bioinformatics, National Institute of
Bioinformatics, 4Departament de Genètica i Institut de Biomedicina (IBUB), Universitat de Barcelona, Diagonal 643,
08028 Barcelona, Catalonia and 5Institució Catalana de Recerca i Estudis Avançats (ICREA) Pg. Lluís Companys 23,
08010 Barcelona, Catalonia
AssociateEditor:JohnQuackenbush
ABSTRACT (Abeeletal.,2008;Goñietal.,2007).However,theincorporation
Motivation: Thepredictionandannotationofthegenomicregions ofnovelbiologicalknowledgeintotheseprogramsisnotnecessarily
involvedingeneexpressionhasbeenlargelyexplored.Mostofthe improving the quality of their predictions, which still contain a
energy has been devoted to the development of approaches that substantialfractionoffalsepositives.
detecttranscriptionstartsites,leavingtheidentificationofregulatory Currently,methodsthatdenovodetectandcharacterizeproximal
regionsandtheirfunctionaltranscriptionfactorbindingsites(TFBSs) regulatory regions show specificity levels <50% (Van Loo and
largely unexplored and with important quantitative and qualitative Marynen,2009).Eventhoughphylogeneticfootprintingusingpre-
methodologicalgaps. alignedhomologousregulatoryregionsofferspromisingresultsin
Results: We have developed ReLA (for REgulatory region Local the identification of Conserved Regulatory Modules (CRMs) of
Alignment tool), a unique tool optimized with the Smith–Waterman transcriptionfactorbindingsites(TFBSs)(BlanchetteandTompa,
algorithmthatallowslocalsearchesofconservedTFBSclustersand 2003; Blanco et al., 2007; Pavesi et al., 2007; Sebestyen et al.,
thedetectionofregulatoryregionsproximaltogenesandenhancer 2009; Tokovenko et al., 2009; Tonon et al., 2010), they cannot
regions. ReLA’s performance shows specificities of 81 and 50% define the regulatory region itself in most real scenarios because
whentestedonexperimentallyvalidatedproximalregulatoryregions they are based on global alignment strategies and require that all
andenhancers,respectively. matchingbindingsitesacrossallcomparedsequencesarelocatedin
Availability: ThesourcecodeofReLA’sisfreelyavailableandcan thesame(orsimilar)positionwithineachsequence,i.e.theyrequire
beremotelyusedthroughourwebserverunderhttp://www.bsc.es/ predefinedandpre-alignedregulatoryregions.Asaresult,inspiteof
cg/rela. theexistingmethodologies,stillthemostcommonandreliableway
Contact: [email protected] toidentifyproximalregulatoryregionsingenomesistheanalysis
Supplementary information: Supplementary data are available at ofafewnucleotides(typicallyupto1000)immediatelyupstreamof
Bioinformaticsonline. annotated transcription start sites (TSSs), which likely constitutes
the proximal promoter. But the annotation of gene starts is still
Received on September 28, 2011; revised on December 19, 2011;
unsolved,particularlyfornon-humanspecies.Forexample,asimple
acceptedonJanuary9,2012
searchintheENSEMBLdatabase(Hubbardetal.,2009)identified
(cid:3)
substantial fractions of vertebrate genes without annotated 5UTR
1 INTRODUCTION (from 17% in mouse to 91% in opossum, 42% for human). This
result is even more dramatic within invertebrates. Other problems
The identification of the genomic regions that control the
that constitute a barrier for the automatic inference of promoters
transcriptionofgenesstillremainsachallengedespitetherecentand
(eveninhumanormouse)aretheabundanceandoverpredictionof
continuous development of new experimental and computational
alternativetranscripts.Takentogether,mostcomputationalmethods
methodologies(Tompaetal.,2005).Multipleautomaticapproaches
that detect or align promoters strongly depend on or are coupled
havebeenproposed,rangingfromthosethatsearchforphylogenetic
withtheannotationofuntranslatedgeneregions,whichisgenerally
conservation of sequence or transcription factor binding motifs in
insufficientforthispurpose(Guigoetal.,2006).
non-transcribed DNA regions (Blanchette and Tompa, 2003; Van
Ontheotherhand,thecomputationalidentificationofenhancers
LooandMarynen,2009)totheanalysisofDNAphysicalproperties
is even more complex. These regulatory regions that work
characteristic of regions expected to bind transcription factors
in cooperation with promoters throughout multiple structural
∗Towhomcorrespondenceshouldbeaddressed. constraints are, apparently, delocalized relative to the genes that
†Theauthorswishittobeknownthat,intheiropinion,thefirsttwoauthors are controlling (Arnosti and Kulkarni, 2005). Therefore, their
shouldberegardedasjointFirstAuthors. identification through computational methods requires strategies
©TheAuthor(s)2012. PublishedbyOxfordUniversityPress.
ThisisanOpenAccessarticledistributedunderthetermsoftheCreativeCommonsAttributionNon-CommercialLicense(http://creativecommons.org/licenses/
by-nc/3.0),whichpermitsunrestrictednon-commercialuse,distribution,andreproductioninanymedium,providedtheoriginalworkisproperlycited.
[18:5625/2/2012Bioinformatics-bts024.tex] Page:763 763–770
MANUSCRIPTCATEGORY: ORIGINALPAPER
S.Gonzálezetal.
based on local alignments. Some existing methods, like rVISTA, ofconservedTFBSs)foreachofthecomparisonsperformedbetweenthe
look for conserved TFBS clusters between regions that have referencesequencewitheachoftheothers(seepseudocodeinSupplementary
been pre-aligned with local alignment tools, such as BLASTz in Material). The overall stringency of the searches, the reliability of the
rVISTA(Loots and Ovcharenko, 2004), while others use directly resultingpredictionsandtheconservationoftheTFBSsbetweentheinput
sequencescanproducedifferentpredictions.Insteadofselectingauniversal
localalignment-basedsearchstrategies,liketheEnhancerElement
andfixedsetofparametersforeachofthesearches,whichwouldyieldone
Locator(EEL)thatusestheSmith–Watermanalgorithm(Palinetal.,
unique prediction, we chose to run recursively each pairwise comparison
2006).Thesetoolshaveshowngoodpredictionratesonenhancers,
(reference sequence against each of the other input sequences) with a
andalsoshowimportantlimitationsregardingthenumberofspecies
differentsetofparametersgeneratingacollectionofpreliminarypredictions.
thattheycananalyse,therequiredparametrizationandtheaccuracy A set of posterior selection filters (see below) is then applied on these
oftheprediction. preliminarypredictionstocomeupwithafinalpredictionofthepromoter
Toovercometheselimitations,andtoprovidenovelandimproved region.
solutionstothepredictionofregulatoryregions,wehavedeveloped Eachofthesepair-wisecomparisonsiscarriedoutintwodifferentscoring
ReLA, a public local-based alignment tool that is capable of scenarios (10/−1 and 20/−1 match/mismatch scores, both with an open
detecting promoters and enhancers by identifying clusters of and extension gap penalties of −2 and with two overlapping thresholds
to remove redundant sites (using a window of three or five nucleotides,
regulatorymotifsconservedinanypositionwithinlargehomologous
seeabove);i.e.atotaloffourcomparisonsareperformedoneachpairof
DNAregions(i.e.independentlyofgeneannotation).Considering
sequencesandeachsetofmatricesdefined.Thesespecificcombinationsof
thewiderangeofpotentialusersofthistool,wehavealsodeveloped
parametersweredeterminedbymonitoringandmaximizingtherelationship
auser-friendlywebserverforremotepredictions.Thesourcecode
betweensensitivityandspecificityusingacollectionof10knownpromoters
ofReLAisdistributedalsoasastandaloneprogramthatcanbeused oftheABSdatabase(Blancoetal.,2006a)(seeSupplementaryMaterialand
locallyinUnix-basedcomputationalplatforms. www.bsc.es/cg/rela/downloads).These10regionswereexcludedfromthe
performanceevaluation.Wealsoobservedthatthebestpredictionsobtained
during the benchmarking were those with sizes between 200 and 600nt.
2 MATERIALS AND METHODS Preliminarypredictionscoveringshorterregionsusuallyinvolvedtoofew
conserved TFBSs, while larger predictions tend to connect distant and,
2.1 FromDNAtoTFBSsequences
apparently,unrelatedbindingsites.Forthisreason,preliminarypredictions
The first step of our method consists on the mapping of putative TFBSs outsidethisrangeofsizesarenotconsideredduringthegenerationofthe
sequences along the input sequences according to a certain catalogue of finalprediction.
positionfrequencymatrices(PFMs)(inthedocumentationincludedwiththe
programandinthewebserver,weprovideguidanceforselectingasetof 2.3 Outputgeneration
homologoussequences).UserslocallyrunningReLAshouldprovidetheir
As part of the results, ReLAgenerates a graph showing the distribution
own PFM files (information about accepted file formats is also provided
ofallacceptedpreliminarypredictionsonthereferencesequence.Fromthe
with the program and in the web server). It has been shown that the
analysisoftheoverlapamongthesepreliminarypredictions,wegeneratethe
selectionofparticularcollectionsofmatricesyieldsslightlydifferentresults
finalpredictionbyselectingtheregion,between200and1000ntlongthat
(Blancoetal.,2006b).Afterevaluatingdifferentoptions(datanotshown),
containsthehighestnumberofpreliminarycandidatepredictions.Together
weobtainedtheoptimalresultsbyusingPFMsfromTRANSFAC(Matys
with the final prediction on the reference sequence, additional consensus
etal.,2006).Weclassifiedthiscollectionofmodelsintothreesubsets:whole
regionsarealsodefinedineachoftheothersequences,whichcorrespondto
collectionofTRANSFACPFM,thefirst600andthefirst400PFMsranked
those(ifany)thatmatchthefinalpredictedpromoterregion.Wealsoprovide
by their information content.These three sets are included as the default
thelistofallconservedTFBSs.Fromthislist,asubsetofthemostconserved
optioninthewebserver.TheidentificationofpotentialTFBSsisperformed
TFBSs(specifically,thosewithinthetop10%ofconservation)isselected
using the MATSCAN software (Blanco et al., 2006b) at high levels of
andusedtogenerateamultiplealignmentingraphicalformat.
stringency: 80% of similarity threshold. For this study, we calculate the
similarityscoreusingSS=[(current–min)/(max–min)],where,‘current’
istheactualmatchingscore,‘min’istheminimumpossiblematchingscore 2.4 Webserver
and‘max’,themaximumpossiblematchingscoreofaparticularPFM,as We have implemented ReLA as a web server that can be accessed at
described elsewhere (Kel et al., 2003). Since the next Smith–Waterman http://www.bsc.es/cg/rela.Theunderlyingsearchengineisdistributedalso
steprequiresasinglesequenceofTFBSs,andbecauseMATSCANresults asastandaloneprogram.WehavedesignedtheReLAwebsitetomeetthe
usually contain PFM that overlap in all possible ways, we next simplify requirements of non-expert users. The web version provides a graphical
thisoutput.Weremovethisoverlapbyslidingawindowofnbp(n=3or representation of the putative promoter region predicted in all the input
5,bothconditionsareincludedintheglobalsearch,seebelow).Foreach sequences(Fig.5)andaplaintextdescriptionoftheresults.Asmanyas
overlappingPFMstartingatthefirstpositionofeachofthe3or5ntsliding twosuboptimalsolutionscanbeprovidedthroughtheweboneachsetof
window,wesystematicallykeptthemostinformativeone,maximizingthe inputsequencestopotentiallypredictalternativepromoters.
overallinformationcontentofthesequence.Topreservetherelativedistance
between motifs during the comparison, we insert a ‘spacer box’every n
2.5 Selectionofdatabasesforevaluation
consecutive nucleotides in regions free of predictions. In summary, we
convert each input DNAsequence into a sequence of highly informative To validate the results obtained and to compare our method with other
non-overlappingTFBS,whichisusednextinthecomparativesearches. existingprogramsinsimilarsearchingconditions,threedifferentworking
subsetsorreportedregulatoryregionsweregeneratedfromthreedifferent
databases:AnnotatedregulatoryBindingSitesdatabase[ABS;(Blancoetal.,
2.2 Comparativesearches
2006a)],EukaryoticPromoterDatabase[EPD;(Schmidetal.,2006)]and
For our searches, we have modified the classical local alignment Smith– VistaEnhancerDatabaseBrowser[VISTA;(Viseletal.,2007)]Tofollow
Watermanalgorithm(SmithandWaterman,1981)todealwithsequences common criteria and to be consistent with the annotation of ABS (as
ofTFBSs(associatingauniquethree-lettercombinationtoeachTFBS)and 500ntpromoters),wetransformedtheseTSSintoregionsbyconsidering
to provide the best scoring local alignment (i.e with the highest density aspromoterregion500bpupstreamfromtheEPDTSS.VISTAEnhancer
764
[18:5625/2/2012Bioinformatics-bts024.tex] Page:764 763–770
MANUSCRIPTCATEGORY: ORIGINALPAPER
ReLA
Browserisadatabaseofregionscontainingexperimentallyvalidatedhuman the same strategies that have been used for fast local sequence
andmouseenhancerstestedintransgenicmice. comparisons of protein sequences. In particular, we adapted the
Theworkingsubsetsweregeneratedaccordingtothreedifferentfiltersto Smith–Watermanalgorithm(SmithandWaterman,1981)tomakeit
facilitatetheautomationofthevalidationprocessandtoensurereliability capableofcomparinganddetectingthebestoptimallocalalignment
of the evaluation protocol: (i) genes associated to selected regulatory
of regions with similar sequences of TFBSs. In our procedure,
regions must have at least three orthologous one-to-one genes according
each TFBS is internally transformed into symbols of an arbitrary
toENSEMBLorthologydata;(ii)thepromoterfragmentsselectedshould
alphabet, as if they were amino acid or nucleotides in traditional
notoverlapwithcodingregions;and(iii)theyhavetobeinourscanned
protein–proteinandDNA–DNAcomparativesearches.Thissearch
region:asdescribedintheSection3,forABSandEPDitis500bpupstream
ofthefirstcodonofthegene,andforVista,50000bpupstreamoftheclosest algorithm is the core of a pipeline, referred from now on as
gene(seebelow). ReLA(forREgulatoryregionLocalAlignmenttool).Thecomplete
Applyingthesethreefiltersweobtained75humanandmousepromoters procedurecanbedividedintothreemajorsteps.First,inputDNA
fromABS,and740fromEPD.Inbothcases,5000bpupstreamfromthefirst sequences are transformed into sequences of TFBSs by mapping
methioninewereusedtocheckandcomparetheaccuracyofthemethod. withtheMATSCANsoftware(Blancoetal.,2006b)allthePFMs
FromVISTAEnhancerBrowser,weendedupwith44enhancerslayingin provided; secondly, the resulting TFBS sequences are compared
the5000bpupstreamofaknowngene.
with each other to identify conserved groups of TFBSs using
the modified Smith–Waterman algorithm under different scoring
2.6 Evaluationprotocol scenarios. Finally, all the resulting preliminary alignments are
For the evaluation of the promoter prediction programs, we followed a evaluatedtoproducethefinalpredictionofthepromoterregion.
modifiedversionoftheDistance-basedvalidationevaluationprotocolfrom
Abeeletal.(2009).Takingintoaccountthatwewereevaluatingpromoter 3.2 EvaluationofReLA’sperformance
genesandwewereconsideringdistances,wecalculatedrecallandprecision
We first applied ReLAto a collection of experimentally validated
valuesas
promoter and enhancer regions, both to define its internal search
Precision=correctpredictions/totalpredictions parametersandtoevaluateitsperformance.Despiteongoingefforts
Recall=discoveredgenes/totalgenes of acquiring experimental data, on functional TFBS, still the
vast majority of detailed and reliable data can only be retrieved
Forthoseprogramsthatprovidesinglepositionsasoutputs(TSSpredictors, from the literature. In this direction, the ABS database (Blanco
ARTS,EponineandPromoterExplorer),weconsideredthesequence±500 et al., 2006a) is the result of one of the few initiatives to gather,
fromTSSfortheevaluation.InthecaseofTFM,weobtainedconserved
from the literature, promoters with two or more of their TFBSs
binding sites as result and considered the fragment between the first and
experimentallyvalidated.Forthisreasonweusedthisdatabasefor
thelastoneforevaluation.Acorrectpredictionwasconsiderediftherewas
ReLA’sevaluation.Weselectedthesubsetof73(35humanand38
anoverlapbetweenthepredictionandthe500bpupstreamofthedefined
mouse) promoters from this database that showed, in ENSEMBL
TSS. For all programs, we considered the unique or the best prediction,
exceptforPromoterExplorerthatdoesnotranktheresults.Sincewewere (Hubbard et al., 2009), one-to-one orthologous relationship with
usingalreadyfilteredpromotergenesinsteadofbigDNAfragments,wedid at least three out of seven chosen vertebrate species (human or
not discarded any prediction further of 500nt from theTSS as it is done mouse,rat,horse,dog,cow,opossumandchicken).Byreproducinga
elsewhere(Abeeletal.,2009).Forallprogramsofourevaluation,weused commonandrealisticsearchscenario,wheretheTSSand5(cid:3)UTRsof
defaultsettingsdefinedbythecorrespondingdevelopers.ForEELruns,we queryhomologousregionsarenotknown,wecollectedtheputative
usedthemouseandhumansequencesforeachoftheorthologousgroups upstreamregionofthesegenesandtheircorrespondingorthologous
andappliedtheparametersdescribedforthisspeciespairelsewhere(Palin
regions.Theseupstreamregionscomprise5000bpupstreamDNA,
etal.,2006).Allthedatausedforthevalidationprocedureareavailableat
fromthefirstannotatedcodonaccordingtotheencodedENSEMBL
www.bsc.es/cg/rela/downloads. (cid:3)
protein.Fromthemeasurementofthelengthof5UTRsregionsof
‘known’ENSEMBLgenes, we previously had estimated that this
3 RESULTS selectionof5000bpissufficienttocapturetheproximalpromoters
for>85%ofknownvertebrategenes(datanotshown).Inaddition,
3.1 RationaleandunderlyingsearchstrategyofReLA
wehavealsocomparedtheresultingperformanceofReLAwiththe
Thegoalofthisstudyistodevelopanovelmethodologythatwould prediction ratio of other reported TFBS-based search tools: TFM
overcomethecurrentlimitationsmentionedabovebyfocusingon: (Tonon et al., 2010) and rVISTA(Loots and Ovcharenko, 2004),
(i) the detection of conserved TFBS; (ii) the use of local search as well as with that of TSS predictors:ARTS (Sonnenburg et al.,
strategies;(iii)simplicityofuse;and(iv)alowcomputationalcost 2006);Eponine(DownandHubbard,2002)andPromoterExplorer
to perform genome-wide searches. For this, we decided to use (Xieetal.,2006),allofthemrunonthesameregions(Table1).
Table1. PerformanceresultsonABSpromoters
ReLA TFM rVISTA(1) PromoterExplorer Eponine ARTS
Recall 0.81 0.6 0.37 0.51 0.2 0.14
Precision 0.81 0.61 0.46 0.69 0.21 0.14
Predictiontype DefinedregionswithconservedTFBS ConservedTFBS TSS
765
[18:5625/2/2012Bioinformatics-bts024.tex] Page:765 763–770
MANUSCRIPTCATEGORY: ORIGINALPAPER
S.Gonzálezetal.
Table2. PerformanceresultsonEPDTSSs
ReLA TFM PromoterExplorer Eponine ARTS
Recall 0.56 0.49 0.78 0.23 0.17
Precision 0.56 0.51 0.67 0.27 0.17
Following the same strategy, we alternatively evaluated ReLA
using740regionsderivedfromtheEukaryoticPromoterDatabase
[EPD; (Schmid et al., 2006)], which, despite not being ideal for
thispurpose,setsourtoolintothecontextofpreviousevaluations
of these other existing search strategies: ARTS, Eponine and
PromoterExplorer against which we have also compared ReLA’s
predictions(Table2).
From all resulting predictions, we calculated different
performance parameters, such as recall and precision by adapting
an evaluation protocol used for the comparison of a large number
ofTSSspredictingmethods(Abeeletal.,2009).Thisadaptationis
necessarybecausethedifferentmethodsweusedprovidedifferent Fig. 1. Prediction of the proximal regulatory region of the Cyclin E1
type of outputs, e.g. ARTS, Eponine, PromoterExplorer provide (CCNE1)geneinsevenvertebratespecies.Typicalsearchscenariowhere
single TSS positions, TFM and rVISTAlists of conserved TFBS, theTSSformostofthespeciescomparedisnotknownormissannotated:
andReLAdelimitedregionsofconservedTFBSs. no TSS information was available for cow, dog and opossum, whereas
From the results of this evaluation, we observe that the in chicken is wrongly placed. The predicted promoter for these species
appearsindifferentlocationwithintheinputsequence.Dashedlinesshow
overallperformanceisdifferentbetweenthedifferentmethodsand
thedistributionofallprimarypredictionsalongtheseregions.Consensus
databases:whileReLA’sperformancewasthebestonABSentries,
predictions are delimited by the first and the last coloured box (each
PromoterExplorer on EPD regions outperformed it. Interestingly,
box corresponds to a conservedTFBS). Red horizontal lines indicate the
ReLA’s precision values for EPD regions are lower than those
experimentally characterized regulatory region. Initial fragments of each
shown with ABS. To discard a possible bias in the performance transcriptareshownontheright:non-codingregionsinblue,andcoding
of ReLAtowards specific promoter types that are more abundant exonsingreen.Thenumbersindicatethepositionofthecodingexonsinthe
in the ABS database, we divided all EPD and ABS regions in humanmRNA.ENSEMBLTSSsareindicatedwitharrows.
different promoter classes (Section 2) and calculated the same
precisionvaluesforeachpromotertypeandeachpredictionmethod
(cid:3)
separately (Supplementary Tables S1 and S2). This analysis has observed when the annotation of orthologous gene 5UTRs and
shown that, despite ABS appears to be enriched in TATA-box firstexonsispracticallyabsent,asoccursformostoftheavailable
containing promoters (a 42% versus a 20% in EPD), ReLA’s genomes. These results highlight the potential of using ReLAfor
performanceisnotaffectedbythis,asprecisionvaluesaresimilar thesystematicidentificationandannotationofregulatoryregionsin
among most of the promoter types present inABS and EPD. It is non-modelorganisms,suchaschicken,cow,dog,opossumandany
also worth noticing that predictors based onTFBSs show a better otherthathasincompletegeneandcDNAdata.
performanceontheABS,whichisalsobasedonTFBS,thanwiththe
TSS-basedEPDentries,whereTSSpredictorstendtodobetter.We
3.3 Predictionofmultiplealternativepromoters
didnotfindanysignificantdifferencewhencomparingperformance
valuesforHumanorMouseentries(datanotshown). During the evaluation of ReLA, we also observed that, in
InordertohaveasenseoftheTFBSconservationlevels,upon some cases, the distribution of preliminary predictions along
which ReLA is able to build predictions, we have also analysed the reference sequence highlighted two regions with similar
thedistributionofthenumberofconservedboxesdetectedwithin scores, which could correspond to alternative promoters. From
allABS and EPD results. This analysis shows that, indeed, there these two options, ReLA selects the one that generated more
is a wide range of TFBS conservation, both in number and in preliminary predictions (Section 2). Suboptimal solutions, i.e.
composition(SupplementaryFig.S1).Similarly,fromtheanalysis potentialalternativepromoters,canbeobtainedbysimplymasking
of the contribution of each of the species in ReLA’s performance, the previously predicted regions in the reference sequence and
weobservethatalltheothervertebratesusedinthisstudycontribute running ReLA again. For a number of such cases, we confirmed
substantiallytothefinalpredictioninhuman:forexample,cowand the presence of twoTSSs through the analysis of ESTs or known
dog contribute to ∼60% of the predictions whereas opossum and alternatively transcribed full-length mRNAs. For this reason, we
chickento36and32%,respectively(SupplementaryTableS3). have implemented this option in the web server, where the user
AdetailedinspectionofReLA’sresultsonABSentriesuncovered can launch a second search run to find suboptimal solutions.
some interesting features. Often, the promoters that we identified Figure2showsthebesttwopredictionsofregulatoryregionslocated
on each of the species present different locations within the input upstream from SLC7A7, an amino acid transporter gene, which
5000bp region (Fig. 1). This typical scenario, which must be hasbeenexperimentallyproventohavetwoalternativepromoters
necessarily solved with local-based comparative approaches, is (Puomilaetal.,2007).
766
[18:5625/2/2012Bioinformatics-bts024.tex] Page:766 763–770
MANUSCRIPTCATEGORY: ORIGINALPAPER
ReLA
Fig.2. Predictionofalternativeknownpromoterregionsofthesolutecarrier Fig.3. PredictionoftheproximalregulatoryregionoftheE2Ftranscription
family 7 member 7 (SLC7A7). Searches were done on 15000bp region factor1(E2F1)usingthesequenceofthewholegene.Predictedregulatory
upstreamofthefirstaminoacidannotatedintheENSEMBLdatabasefor region along a highly conserved region of 15722 bp that includes the
thehumanSLC7A7gene.Dashedlinesshowthedistributionofpreliminary E2F1 complete gene transcript and 5000bp upstream of the first amino
predictionsinthefirstandsecondrun.Finalfirstandsecondpredictionsare acid annotated in ENSEMBL. Dashed line shows the distribution of all
enclosedinaboxanddelimitedbythefirstandthelastconservedTFBS(each preliminarypredictionsalongthisregion.Aschematicrepresentationofthe
designedbyacolouredbox).Initialfragmentsofeachtranscriptareshown structureofthisgeneisshown:non-codingregionsareinblue,andcoding
ontheright:non-codingregionsinblue,andcodingexonsingreen.The exonsingreen.Thenumbersdesignatethepositionofthecodingexonsin
numbersindicatethepositionofthecodingexonsinthemRNA.ENSEMBL the mRNA, according to human. Data related to DNAconservation from
TSSsareindicatedwitharrows.Distancesarenotdrawnatrealscale. UCSCarealsoincluded[http://genome.ucsc.edu;(Kentetal.,2002)].
ThefindingoftwohighscoringregionsinanyReLAprediction theidentificationofenhancers,whichareoftenlocateddistantfrom
couldsuggest,insteadofthepresenceofanalternativepromoter,the other functional elements. In order to test ReLA’s capabilities in
existenceofhighlyconservedcodingexons,whichwouldconstitute enhancerdetectionwehavegatheredacollectionofexperimentally
a false positive prediction. Thus, the identification of regulatory validatedhumanenhancersfromtheVISTAdatabasewithactivity
regions with ReLAwould be based only on the level of sequence assessed on transgenic mice (Visel et al., 2007). To test ReLA,
conservationandthepresenceofhighlyconservednon-regulatory weselected44enhancersthatarelocatedwithinthefirst50000bp
DNA, like coding exons, could negatively influence the results. upstream of a known gene. In order to search for each of these
To discard this, we studied how this scenario can affect ReLA distal regulatory regions we extracted up to 50000bp from the
predictions. The example in Figure 3 shows a positive promoter most upstream TSS annotated for the closest gene in human and
predictionwhenallsevenorthologousinputsequencesincludethe from each of the corresponding one-to-one orthologous genes in
complete region of the E2F1 gene and the additional upstream mouse,rat,horse,dog,cow,opossumandchicken.Thefirstrunof
untranslated regions (a total of 15000nt each). In this case, the ReLA on these 44 regions generated 40 predictions, from which
distribution of hits along the human sequence shows two high 11(28%) overlapped with the annotated enhancer. Considering
scoringregionsthatappeartosharesimilarconservationlevelsof that the regions selected for the search theoretically contain other
nucleotides. One of these fragments constitutes the third exon of unknown regulatory regions (promoters, for instance) that could
(cid:3)
this gene, while the other matches the 5UTR, the TSS and the match with the first prediction, we performed a second run on
corepromoter.ReLAisabletosuccessfullydiscriminatethecorrect the remaining 29 cases, which yielded nine other positive hits. In
promoter region, including sites that have been experimentally total, with two iterative runs, ReLAshowed a positive predictive
proven (Blanco et al., 2006a). In particular, the two TFBS that valueof50%ofthescreenedsubsetofannotatedhumanenhancers.
ReLA scores highest in conservation among input sequences are Asimilarpredictionrate(49%)isobtainedoverthesameenhancer
preciselydescribedintheABSdatabaseastwoE2F1factorbinding benchmark set when using a specific enhancer locator tool, EEL
sitesnecessaryforself-regulationduringthetransitionfromG1toS (Palinetal.,2006)thatalsoreliesonlocalsearchstrategies(EEL
phaseinthecellcycle(Johnsonetal.,1994).Interestingly,thethird searchesimpliedonlyhumanandmousesequences,asitdoesnot
TFBS following the conservation ranking corresponds to ADF1, acceptmorethantwosequencespersearch).Itisworthmentioning
(cid:3)
whichwaslocatedwithinthe5UTRandisknowntobindthesame that an important difference between both methods is that ReLA
motifsrecognizedbytheE2F1factorinmice(Hsiaoetal.,1994) provides more precise results, as the regions predicted are shorter
Despite these results, we cannot discard the possibility that exons (up to 750nt long, with an average of 485nt) than those coming
arewronglypredictedaspromotersincertainsituations.Therefore, fromEEL(upto11563nt,withanaverageof2644nt).
we recommend performing preliminary evaluations of the coding These results indicate that ReLA is capable of searching large
potential within the input sequences, for example, by comparing genomic DNA fragments and identifying multiple proximal and
them against protein sequence databases with BLASTX (Altschul distal regulatory regions, which makes this tool suitable for
et al., 1997). Putative coding regions should be preferentially genome-wide screenings and across several genomes (see an
maskedfromtheinputsequencestoensurethecorrectprediction. exampleinFig.4).
Tofurtherexemplifythisfeature,wealsoperformedagenome-
wide analysis on a 109kb long ENCODE region [ENm011;
3.4 Identificationofenhancers
chr11:1858751-1968592; (Birney et al., 2007)] that includes six
The local nature of the underlying search engine and the capacity genes coding for, at least, 11 transcripts, with their corresponding
to compare large DNAsequences makes ReLAa suitable tool for intergenic regions. By using SYT8 and MRPL23 flaking genes
767
[18:5625/2/2012Bioinformatics-bts024.tex] Page:767 763–770
MANUSCRIPTCATEGORY: ORIGINALPAPER
S.Gonzálezetal.
4 DISCUSSION
Taking into account the available methods to in silico recognize
generegulatoryregions,asubstantialimprovementisnecessaryto
accuratelyannotategenesandpromotersequencesinmostgenomes.
HerewereportthedevelopmentofReLA,acomputationaltoolto
identify such regulatory regions using genome-wide comparisons.
ReLAis distributed as a standalone program and as a web server.
OurapproachismostlybasedinanadaptationofthepopularSmith–
Watermanalgorithmthatisabletorapidlyidentifycoincidencesof
TFBSsbetweentwosequences(conceptuallysimilartotraditional
protein–protein comparisons). ReLAis able to efficiently process
longsequencesinstandardcomputationalplatforms(e.g.lessthan
aminutetoobtaintheresultsshowninFig.5).Wehaveevaluated
theaccuracyofReLA,firstinadatasetofexperimentallyvalidated
humanandmousepromoters,onanextensivecollectionofvalidated
TSS from the Eukaryotic Promoter Database, as well as on an
experimentallyvalidatedcollectionofrVISTAenhancers.Wehave
reached maximums of 0.81 of recall and precision levels onABS
sequencesOntheotherhand,andsurprisingly,ReLA’sperformance
resultslowerwhenusingEPDTSSentries.Apossibleexplanation
forthisobservationcouldbethatReLAperformsbetteroncertain
typesofpromoterregions.But,afterweclassifiedallABSandEPD
entriesintodifferentpromotertypesaccordingtotheircomposition
and evaluated their associated performance obtained with all the
methods used here for the validation, we observed that ReLA’s
Fig.4. PredictionoftheSALL1enhancerinsixvertebrategenomes.The accuracy is similar among most of the identified promoter types.
upperpanelshowstheSALL1geneanditscorresponding50kbupstream We cannot discard though that other uncontrolled biases present
regionforsixvertebratespecies.Coordinatesandstrandoftheseregionsare: either inside the underlying search methodology of each of the
Chicken(chr11:6148098–6213605,−),Mouse(chr8:91551143–91618061, protocolsusedhereorintheuseddatabasescouldactuallyexplain
+),Rat(chr19:19227337–19293298,−),Dog(chr2:67080056–67144522, the different behaviour observed. It is worth noticing that overall,
−), Cow (chr18: 18688671–1875278, +) and Human (chr16: 51169886– TFBS-basedpredictionmethodsperformbetterontheTFBS-based
51235181,+).Ineachline,wedisplaythestructureofthegene(greenboxes
ABSdatabasethanontheTSS-basedEPD,whereTSSpredictorsare
arecodingexons,whileblueareuntranslated).KnownandpredictedTSSs
doingbetter.Inanycase,thelevelsofprecisionandrecallobtained
arealsoshown.ReLA’spredictionsareshownforeachspeciesasgroupsof
withReLAaresufficienttoprovidereliablepredictionsthatguide
coloredboxes.Please,notethattheseregionsarenotdrowntoscale.ReLA’s
posteriorexperimentalvalidation.Thisstudyalsodemonstratesthe
predicted enhancer regions expanded from 223 and 233bp for dog and
mouse,301bpforcow,to359,360and366bpforrat,humanandchicken, benefitsofusingtheSmith–Watermanalgorithmtodirectlysearch
respectively.Thelocationsoftheexperimentallyprovenregions(asshown for conserved binding sites, as it outperforms other methods like
inrVISTAdb)aredisplayedasgreenboxes.Thebottomlineofthispanel rVISTAandTFM, that are based on pre-aligned DNAand global
showsthesequenceconservationprofile(accordingtohumancoordinates; search strategies. See the example in Figure 1, which cannot be
http://genome.ucsc.edu).Inthebottompanel,wedisplaythealignmentof solved using global alignment approaches. Please note that other
theconservedTFBSdetectedwithineachofthepredictedenhancerregions. methods based on similar strategies could not be included in the
TFBSarelabelled(inTRANSFACformat)anddifferentiatedusingarbitrary
comparison, as they did not provide results on our benchmark set
shapes and colours. Coordinates shown here indicate the position of the
becauseoflimitationsinthesize(MMETA)andonthenumberof
predictedenhancerwithinthe50kbinputsequence.
sequences[Conreal(Berezikovetal.,2005)].
Furthermore, we also show that ReLA is able of predicting
as anchors, we identified and characterized the corresponding alternative promoters and even enhancer regions, dealing with
orthologous regions for mouse, rat, dog, cow and chicken. The multiple suboptimal solutions in most cases. Our approach is
complete analysis consisted in 10 iterative ReLA searches and suitableforintegratingacomputationalannotationpipelineinwhich
implied the screening ofTFBS sequences in >600kb of genomic otherpredictivemethodssuchashomologysearches(e.g.BLAST
DNA.Inordertoobtainanestimationoftheperformanceonthese againstproteindatabases)canassistintheimprovementofthefinal
genome-wide conditions, we have taken as positive predictions predictions.
those that match ChIP-Seq transcription evidences (Birney et al., Insummary,webelievethatthedevelopmentofReLAconstitutes
2007),aswellasthosefallingimmediatelyupstreamofannotated asignificantstepforwardinthefieldofthepredictionofregulatory
gene starts. This count shows that 8 out of 10 predictions have regions, as it shows the highest predictive power reported so far.
evidence of expression or regulation (Supplementary Fig. S2). ReLA is able to locally compare multiple large genomic regions
Pleasenotethatwecannotdiscardthatadditionalrunswouldprovide and identify non-alignable conservation events across different
other overlooked regulatory regions and, at some point, also false genomes.Thisisrelevantifweconsiderthatthelimitedinformation
positives. regarding regulatory regions in eukaryotes is restricted to human
768
[18:5625/2/2012Bioinformatics-bts024.tex] Page:768 763–770
MANUSCRIPTCATEGORY: ORIGINALPAPER
ReLA
acrosssequencedvertebrategenomes,weareactivelysearchingfor
ways of improving further its predictive power by, for example,
applying more sophisticated scoring systems and accepting even
largerDNAregionswithlowadditionalcomputationalcosts.
ACKNOWLEDGEMENTS
We thank Mar Albà, Steven Laurie and our entire group for
constructive feedback during the development of this work and
during the writing of the manuscript.We also thank Jan andAina
SagristàfordesigningReLA’slogo.
Funding: Ministerio Español de Ciencia e Innovación (BIO2006-
15036).
ConflictofInterest:nonedeclared.
REFERENCES
Abeel,T.etal.(2008)Genericeukaryoticcorepromoterpredictionusingstructural
featuresofDNA.GenomeRes.,18,310–323.
Abeel,T. et al. (2009) Toward a gold standard for promoter prediction evaluation.
Bioinformatics,25,i313–i320.
Altschul,S.F.etal.(1997)GappedBLASTandPSI-BLAST:anewgenerationofprotein
databasesearchprograms.NucleicAcidsRes.,25,3389–3402.
Arnosti,D.N. and Kulkarni,M.M. (2005) Transcriptional enhancers: Intelligent
enhanceosomesorflexiblebillboards?J.Cell.Biochem.,94,890–898.
Berezikov,E.etal.(2005)CONREALwebserver:identificationandvisualizationof
conservedtranscriptionfactorbindingsites.NucleicAcidsRes.,33,W447–W450.
Birney,E.etal.(2007)Identificationandanalysisoffunctionalelementsin1%ofthe
humangenomebytheENCODEpilotproject.Nature,447,799–816.
Blanchette,M.andTompa,M.(2003)FootPrinter:Aprogramdesignedforphylogenetic
footprinting.NucleicAcidsRes.,31,3840–3842.
Blanco,E.etal.(2006a)ABS:adatabaseofAnnotatedregulatoryBindingSitesfrom
orthologouspromoters.NucleicAcidsRes.,34,D63–D67.
Blanco,E.etal.(2006b)Transcriptionfactormapalignmentofpromoterregions.PLoS
Comput.Biol.,2,e49.
Blanco,E.etal.(2007)Multiplenon-collinearTF-mapalignmentsofpromoterregions.
BMCBioinformatics,8,138.
Down,T.A. and Hubbard,T.J. (2002) Computational detection and location of
transcriptionstartsitesinmammaliangenomicDNA.GenomeRes.,12,458–461.
Goñi,J.R.etal.(2007)DeterminingpromoterlocationbasedonDNAstructurefirst-
principlescalculations.GenomeBiol.,8,R263.
Guigo,R.etal.(2006)EGASP:thehumanENCODEGenomeAnnotationAssessment
Project.GenomeBiol.,7(Suppl.1),S21–31.
Hsiao,K.M.etal.(1994)MultipleDNAelementsarerequiredforthegrowthregulation
ofthemouseE2F1promoter.GenesDev.,8,1526–1537.
Hubbard,T.J.etal.(2009)Ensembl2009.NucleicAcidsRes.,37,D690–D697.
Johnson,D.G.etal.(1994)AutoregulatorycontrolofE2F1expressioninresponseto
positiveandnegativeregulatorsofcellcycleprogression.GenesDev.,8,1514–1525.
Kel,A.E.etal.(2003)MATCH:Atoolforsearchingtranscriptionfactorbindingsites
Fig.5. SnapshotoftheReLAwebserveroutput.Graphicalrepresentation inDNAsequences.NucleicAcidsRes.,31,3576–3579.
oftheputativepromotersandalignmentsofTFBSsofthehumanE2F1gene, Kent,W.J. et al. (2002) The human genome browser at UCSC. Genome Res., 12,
aswellasthelistsofpredictedregionsandconservedbindingmotifs.See 996–1006.
Sections2and3foracompleteinterpretationofeachoftheresultsprovided. Loots,G.G.andOvcharenko,I.(2004)rVISTA2.0:evolutionaryanalysisoftranscription
factorbindingsites.NucleicAcidsRes.,32,W217–W221.
Matys,V.etal.(2006)TRANSFACanditsmoduleTRANSCompel:transcriptionalgene
and mouse, e.g. from 2540 vertebrate entries in the Eukaryotic regulationineukaryotes.NucleicAcidsRes.,34,D108–D110.
Promoter Database (Schmid et al., 2006), 2067 (81%) belong Palin,K.etal.(2006)Locatingpotentialenhancerelementsbycomparativegenomics
to these two species. Thus, with this tool in hand we can now, usingtheEELsoftware.Nat.Protocols,1,368–374.
Pavesi,G.etal.(2007)WeederH:analgorithmforfindingconservedregulatorymotifs
not only fill missing gaps in the annotation of the genomes of
andregionsinhomologoussequences.BMCBioinformatics,8,46.
model organisms, mostly with the identification of enhancers and Puomila,K.etal.(2007)Twoalternativepromotersregulatetheexpressionoflysinuric
alternative promoters, but also to start a reliable and consistent proteinintolerancegeneSLC7A7.Mol.Genet.Metab.,90,298–306.
annotationofconservedpromotersthroughouttherestofgenomes Schmid,C.D. et al. (2006) EPD in its twentieth year: towards complete promoter
thathavelittleornoinformationregarding5(cid:3)UTRsandoftenfirst coverageofselectedmodelorganisms.NucleicAcidsRes.,34,D82–D85.
Sebestyen,E.etal.(2009)DoOPSearch:aweb-basedtoolforfindingandanalysing
coding exons (Fig. 1). Beyond the current performance of ReLA
commonconservedmotifsinthepromoterregionsofdifferentchordateandplant
and,asweareplanningagenome-widesearchofregulatoryregions genes.BMCBioinformatics,10(Suppl.6),S6.
769
[18:5625/2/2012Bioinformatics-bts024.tex] Page:769 763–770
MANUSCRIPTCATEGORY: ORIGINALPAPER
S.Gonzálezetal.
Smith,T.F. and Waterman,M.S. (1981) Identification of common molecular Tonon,L. et al. (2010) TFM-Explorer: mining cis-regulatory regions in genomes.
subsequences.J.Mol.Biol.,147,195–197. NucleicAcidsRes.,38(WebServerIssue),W286–W292.
Sonnenburg,S.etal.(2006)ARTS:accuraterecognitionoftranscriptionstartsinhuman. VanLoo,P.andMarynen,P.(2009)Computationalmethodsforthedetectionofcis-
Bioinformatics,22,e472–e480. regulatorymodules.Brief.Bioinform.,10,509–524.
Tokovenko,B. et al. (2009) COTRASIF: conservation-aided transcription-factor- Visel,A.etal.(2007)VISTAEnhancerBrowser–adatabaseoftissue-specifichuman
bindingsitefinder.NucleicAcidsRes.,37,e49. enhancers.NucleicAcidsRes.,35,D88–D92.
Tompa,M.etal.(2005)Assessingcomputationaltoolsforthediscoveryoftranscription Xie,X.etal.(2006)PromoterExplorer:aneffectivepromoteridentificationmethod
factorbindingsites.Nat.Biotechnol.,23,137–144. basedontheAdaBoostalgorithm.Bioinformatics,22,2722–2728.
770
[18:5625/2/2012Bioinformatics-bts024.tex] Page:770 763–770