Constitutive patterns of gene expression regulated by RNA ...

RNA-binding proteins (RBPs) are key regulators of post-transcriptional events [3] and influence gene expression by acting at various steps ... Skiptomaincontent Advertisement SearchallBMCarticles Search ConstitutivepatternsofgeneexpressionregulatedbyRNA-bindingproteins DownloadPDF DownloadPDF Research OpenAccess Published:02January2014 ConstitutivepatternsofgeneexpressionregulatedbyRNA-bindingproteins DavideCirillo1,2,DomenicaMarchese1,2,FedericoAgostini1,2,CarmenMariaLivi1,2,TeresaBotta-Orfila1,2&GianGaetanoTartaglia1,2  GenomeBiology volume 15,Article number: R13(2014) Citethisarticle 11kAccesses 32Citations 6Altmetric Metricsdetails AbstractBackgroundRNA-bindingproteinsregulateanumberofcellularprocesses,includingsynthesis,folding,translocation,assemblyandclearanceofRNAs.RecentstudieshavereportedthatanunexpectedlylargenumberofproteinsareabletointeractwithRNA,butthepartnersofmanyRNA-bindingproteinsarestilluncharacterized.ResultsWecombinedpredictionofribonucleoproteininteractions,basedoncatRAPIDcalculations,withanalysisofproteinandRNAexpressionprofilesfromhumantissues.Wefoundstronginteractionpropensitiesforbothpositivelyandnegativelycorrelatedexpressionpatterns.Ourintegrationofinsilicoandexvivodataunraveledtwomajortypesofprotein–RNAinteractions,withpositivelycorrelatedpatternsrelatedtocellcyclecontrolandnegativelycorrelatedpatternsrelatedtosurvival,growthanddifferentiation.Tofacilitatetheinvestigationofprotein–RNAinteractionsandexpressionnetworks,wedevelopedthecatRAPIDexpresswebserver.ConclusionsOuranalysisshedslightontheroleofRNA-bindingproteinsinregulatingproliferationanddifferentiationprocesses,andweprovideadataexplorationtooltoaidfutureexperimentalstudies. BackgroundWiththeadventofhigh-throughputproteomicandtranscriptomicmethods,genome-widedataaregivingpreviouslyunprecedentedviewsofentirecollectionsofgeneproductsandtheirregulation.Recently,approachesbasedonnucleotide-enhancedUVcross-linkingandoligo(dT)purificationhaveshownthatanumberofproteinsareabletobindtoRNA[1,2].RNA-bindingproteins(RBPs)arekeyregulatorsofpost-transcriptionalevents[3]andinfluencegeneexpressionbyactingatvariousstepsinRNAmetabolism,includingstabilization,processing,storing,transportandtranslation.RBP-mediatedeventshavebeendescribedusingrecognitionandregulatoryelementsinRNAsequences[4,5]aswellasexpressionprofiles[6]thataretissuespecificandconservedacrossspecies[7–9].Althoughheterogeneityingeneregulationisresponsibleforphenotypicvariationandevolution[10],verylittleisknownaboutconstitutiveexpressionpatternscontrolledbyRBPs[11,12],whicharethesubjectofthiswork.Datafromrecenttranscriptomicandproteomicstudies[13,14]arebecomingattractiveforstudyingmechanismsofgeneregulation[15,16].Despitetheincreasingamountofgenomicdata,thedevelopmentofcomputationalmethodsforintegrating,interpretingandunderstandingmolecularnetworksremainschallenging[17,18].Herewecombineourpredictionsofprotein–RNAinteractions,basedoncatRAPIDcalculations[19,20],withtheinformationobtainedfromexpressiondatatoinvestigateconstitutiveregulatorymechanisms.ThecatRAPIDapproachhasbeenpreviouslyemployedtopredictproteinassociationswithnon-codingRNAs[21,22]aswellasribonucleoproteininteractionslinkedtoneurodegenerativediseases[23,24].Ourtheoreticalframeworkhasbeenusedtounravelself-regulatorypathwayscontrollinggeneexpression[25].ThecatRAPIDomicsalgorithm,validatedusingphotoactivatable-ribonucleoside-enhancedcross-linkingandimmunoprecipitation(PAR-CLIP)data,hasbeenrecentlydevelopedtopredictprotein–RNAassociationsatthetranscriptomicandproteomiclevels[26].Usingcomprehensiveandmanuallyannotateddatabasesofexpressionprofilesinhumantissues,atbothproteinandRNAlevels,weinvestigatedthecorrelationbetweenRBPactivityandregulation.Thelinkbetweeninteractionpropensityandexpressionlevelswasexploitedtorevealthefine-tunedfunctionalsub-networksresponsibleforregulatorycontrol.Toexploretheresultsfurther,wedevelopedthecatRAPIDexpresswebserver[27].ResultsInthisstudy,wefocusedonthemRNAinteractomesofRBPsdetectedthroughnucleotide-enhancedUVcross-linkingandoligo(dT)purificationapproaches[1,2].Exploitinggeneontology(GO)annotations[28]forprotein-codinggenes,wesystematicallyanalyzedprotein–RNAinteractionsandexpressiondataforhumantissues.Atpresent,fewstudieshaveinvestigatedhowalteringproteinexpressionaffectstheabundanceofRNAtargets.InterrogatingtheGeneExpressionOmnibus(GEO)[29]andArrayExpressdatabases[30],wefoundtwohumanproteins,ELAV-likeprotein1(orhumanantigenR,HuR)[31]andProteinlin-28homologB(LIN28B)[32,33],whoseknock-downhasbeenshowntoaltertheexpressionoftargetgenesidentifiedbyPAR-CLIP(seeMaterialsandmethods).Ourpredictions,madeusingthecatRAPIDalgorithm[26],identifiedexperimentallyvalidatedinteractionswithhighsignificance(HuR:P = 10-8;LIN28B:P = 10-3;Fisher’sexacttest;seeMaterialsandmethods).Theinteractionswereeffectivelydiscriminatedfromnon-interactingpairsusingscoredistributions(LIN28B:P = 10-4;HuR:P = 10-16;Student’st-test;seeMaterialsandmethods).Hence,catRAPIDisverygoodatpredictingphysicalinteractionsbetweenaproteinandRNApartners(otherstatisticaltestsaregiveninMaterialsandmethodsandAdditionalfile1).TounderstandtheregulationofHuRandLIN28Btargetsbetter,westudiedtherelationbetweeninteractionpropensitiesandexpressionlevels.WefoundthattheexpressionofpredictedHuRtargetsisaltered(log-foldchange,LFC)whenHuRisknockeddown(P  0.SimilarlyforLIN28B,57%ofexperimentalinteractionsand56%ofpredictedassociationshadLFC > 0.Figure1RelationbetweenproteinandRNAregulation.(A)HuRinteractome:ourpredictions,madeusingcatRAPID[26],indicatethatexpressionlevelsofRNAtargetschangeuponHuRknock-down(log-foldchanges,LFC),inagreementwithexperimentalevidence[31](P  0.4[1])andcalculatedtheinteractionpropensitieswithhumantranscripts.WeconsideredbothcanonicalRBPs(thatis,containingRNA-bindingdomains)andputativeRBPs(thatis,lackingRNA-bindingdomains)[1].WithrespecttotheRNA-bindingabilityoffull-lengthsequences,thecontributionofdisorderishigheratlowinteractionpropensityscoresandbecomesnegligibleathighinteractionpropensities(seeMaterialsandmethodsandFigure 3A).Nevertheless,theroleofstructuraldisorderismorepronouncedinproteinslackingcanonicalRNA-bindingdomains,indicatingthatunfoldedregionsmightbeabletopromoteinteractionswithRNA(Figure 3B).Figure3RNA-bindingabilityandstructuraldisorder.(A)Foreachprotein,wecalculatedRNAinteractionswithfull-lengthsequencesaswellasstructurallydisorderedregions[1,36].Whentheinteractionpropensityscoreofadisorderedregionexceedsthatofthefull-lengthprotein(pointsabovetheredline),disorderisconsideredtopromoteinteractionwithRNAmolecules.(B)For66%oftheproteins(137entries),disordercontributesatlowinteractionpropensities,whilefull-lengthproteinsequencesdominateathighinteractionpropensities(Mann–WhitneyUtest).Overall,fromlowtohighinteractionpropensities,thecontributionofdisorderdecreasesprogressivelywithrespecttothatofthefull-lengthprotein(redandgreylines),inagreementwithapreviousanalysis[25].TheroleofdisorderismorerelevantinproteinslackingcanonicalRNA-bindingdomains(greyline),indicatingthatunstructuredregionsmighthavedirectinvolvementincontactingRNA.Interactionpropensitiesareaveragedperprotein.RBD,RNA-bindingdomain.FullsizeimageInapreviousstudyweobservedthatcatRAPIDscorescorrelatewithchemicalaffinities[21],whichsuggeststhattheinteractionpropensitycanbeusedtoestimatethestrengthofassociation[21,26].Hence,ourresultsindicatethatstructuraldisordermightcontributetolow-affinityinteractionswithRNA(Figure 3A,B),whichisinagreementwithwhathasbeenobservedforprotein–proteinassociations[37,38].Asamatteroffact,ithasbeenreportedthatdisorderregionsareabletopromotepromiscuousandnon-specificinteractions[39].DiscussionBecausetheyareassociatedwithtranscriptionalcontrolofgeneexpression,RBPsplayfundamentalrolesinhealthanddisease.Indeed,bybindingtotheirtargetmRNAs,RBPscaninfluenceproteinproductionatdifferentlevels(transcription,translationandprotein/mRNAdegradation).Protein–RNAcomplexesareverydynamicandcanundergoextensiveremodeling.Thus,theycancontrolthespatiotemporalregulationoftargetgeneexpressionandtheoverallswitchingonandoffofthedistinctsetsofgenesinvolvedinbiologicalprocessessuchascellcycleprogression,celldifferentiation,cellresponsetometabolicstimuliandstressconditions,organmorphogenesisandembryonicdevelopment.Co-expressionandinteractionpropensityarefeaturesofcellcyclecontrolAthighinteractionpropensities(AUC > 95%;seeMaterialsandmethods),theICsubsethasmoreGOtermslinkedtocellcyclecontrolandhousekeepingfunctionssuchasnucleobasemetabolismandpurinebiosynthesis(Figure 4andAdditionalfile3:TableS1).Inparticular,mRNAsinteractingwithco-expressedRBPscodefornegativeregulatorsofcellproliferationandmigration(translation,signalingandmetaboliteutilization).WefoundanumberoftumorsuppressorsintheICsubset(AHRR,BAX,BRMS1,CDKN1A,CDKN2A,CTBP1,DAB2IP,DKK3,FLCN,FOXP1,GADD45G,GALR1,GTPBP4,HIC1,IGFBP3,IRF8,KLF4,MEN1,MLH1,NF2,NR0B2,PARK2,PAWR,PAX4,PAX5,PCGF2,PHB,PML,PPP1R1B,PPP2R4,PTPRJ,PYCARD,RHOA,SIRT2,TFAP2A,TNFAIP3,TRIM24,TSC2,TSG101,UCHL1).Interestingly,90%ofICgenesannotatedwithmorefunctionalcategories(381outof422)arelistedinthegeneindexoftheNationalInstitutesofHealth’sCancerGenomeAnatomyProject[40].Termsassociatedwithinhibitionofcellularpathways(especiallythenegativeregulationofphosphorylationandregulationofproteinserine/threoninekinaseactivity)arealsomoreprevalentintheICsubsetwhenimmunochemistrydataareused.Figure4GOenrichmentforinteractingmRNA–RBPpairscorrelatedinexpression(ICsubset).UsingthecatRAPIDscoredistribution,wecountedmRNAGOenrichmentassociatedwithdifferentareasunderthecurve(seeMaterialsandmethods).Thecolorgradient(yellowtored)indicatestheAUCvalues(numberofinteractions:20,702,804forAUC > 50%,10,351,402forAUC > 75%,2,070,280forAUC > 95%).WefoundthatcellcycleprocesseshavemorehighlyinteractingmRNA–RBPpairs(AUC > 95%)thatarecorrelatedinexpression.AUC,areaunderthecurve;GO,geneontology;IC,interactingandco-expressed;RBP,RNA-bindingprotein.FullsizeimageAsmutationsalteringtumorsuppressionleadtoaberrantproliferativeevents,wespeculatethatdownregulationofspecificgenesisamechanismforpreventingindiscriminatecellulargrowth.Inagreementwiththishypothesis,ithasbeenreportedthatsomaticlossoffunctionofthetumorsuppressortuberoussclerosis2(TSC-2)leadstothedevelopmentofbenignandmalignantlesionsinthemyometrium,kidneyandothertissuessharingcommonfeaturessuchasalowrateofrenewalanddefectsinthemitochondrialrespiratorychainassociatedwithoncogenesis[41,42].ThisgeneisannotatedinallthefunctionalcategoriesprevalentintheICsubset.Intriguingly,itispredictedthatTSC-2mRNAinteractsstronglywithNuclearProtein5A(NOP56).Theinteractionpropensityis175correspondingtoanAUCof99.5%.Thisproteinisanessentialcomponentofthesplicingmachinery[43]thatisdifferentiallyexpressedinleiomyomaanddownregulatedinresponsetohypoxia[44].Itispossiblethathypoxia-dependentrepressionofNOP56expression[45–47]isaprotectivemechanismagainstfastgrowthandpotentialtumorprogression.Indeed,ithasbeenreportedthatNOP56andTSC-2arenotdifferentiallyexpressedinrenalcarcinomasandoncocytomas[48,49](ArrayExpress:E-GEOD-12090;ArrayExpress:E-GEOD-19982),indicatinglossofregulationduringmalignantprogression.Basedontheseobservations,weproposethatdownregulationofRBPspromotingthetranslationofdysfunctionaltumorsuppressorscanpreventindiscriminatecellulargrowthandthatlossofcontrolcandestineacelltomalignancy(additionalexamplesarereportedinAdditionalfile1).Anti-expressionandinteractionpropensityarefeaturesofrepressingprocessesForAUC > 95%,theIAsubsethasmoretermsassociatedwithcelldifferentiationprocesses(forexample,proximal/distalpatternformation)aswellasinflammation(forexample,positiveregulationofisotypeswitching),whichareknowntobetightlylinked[50–52].Infact,anumberofdifferentiationcytokines(IL18,IL23andEBI3/IL27)andstimulatorsofcytokineproduction(CD28andCD80CCR2/CD192)areinthesubset.Moreover,alargefractionofentriesisalsolinkedtoprotein–DNAcomplexassemblyandregulationoftranscriptioninitiationfromRNApolymeraseIIpromoter(Figure 5andAdditionalfile3:TableS1).Ithasbeenshownthat94%ofgenesinIAenrichedfunctionalcategories(124outof132)arelistedintheannotatedgeneindexoftheNationalInstitutesofHealth’sCancerGenomeAnatomyProject[40].Remarkably,termsclearlyassociatedwithcelldifferentiationandinflammation(especiallyregulationofembryonicdevelopmentandBcellactivationinvolvedinimmuneresponse)aremoreprevalentintheIAsubsetwhenimmunochemistrydataareused.Figure5GOenrichmentforinteractingmRNA–RBPpairsanti-correlatedinexpression(IAsubset).UsingthecatRAPIDscoredistribution,weevaluatedmRNAGOenrichmentassociatedwithdifferentareasunderthecurve(seeMaterialsandmethods).Acolorgradient(cyantoblue)showstheAUCvalues(numberofinteractions:20,702,804forAUC > 50%,10,351,402forAUC > 75%,2,070,280forAUC > 95%).WefoundthatcelldifferentiationprocessesaremoreprevalentininteractingmRNA–RBPpairs(AUC > 95%)thatareanti-correlatedinexpression.AUC,areaunderthecurve;GO,geneontology;IA,interactingandanti-expressed;RBP,RNA-bindingprotein.FullsizeimageIAgenessharethecommonfunctionalpropertyofregulatingsurvival,growthanddifferentiationprocesses.AsRBPsplayacrucialroleinrepressinggeneexpression[53,54],IAassociationscouldbeinvolvedintheregulationofproliferativeevents.Indeed,adulttissuesareconstantlymaintainedatthesteadystate[13]butadramaticreawakeningofgrowth,survivalanddifferentiationgenesoccurineitherphysiologicalconditions(forexample,woundhealing[50])orpathologicalprogressiontocancer[55].IntheIAset,wefoundYTHDC1(YT521-B),whichisaubiquitouslyexpressedmemberofthenovelRNA-bindingYTH-domainfamily[56].YTHDC1repressesgeneexpressionbyeithersequesteringsplicingfactorsordirectlybindingtotranscripts[57–59](Additionalfile2:FigureS5A).AmongthetranscriptsthatwepredicttobepotentiallytargetedbyYTHDC1,wefoundseveralproto-oncogenesortumor-associatedgenessuchasRET,PRMT2,RARGandHOXA9(RET:interactionpropensity = 166;PRMT2:interactionpropensity = 209;RARG:interactionpropensity = 194;HOXA9:interactionpropensity = 165;allcorrespondingtoanAUCof99.5%).Inparticular,alternativelysplicedvariantsofPRMT2wererelatedtosurvivalandtheinvasivenessofbreastcancercells[60,61],whilehighexpressionofRARGandHOXA9hasbeenobservedinhumanhepatocellularcarcinomasandacuteleukemia[62,63].WehypothesizethatperturbationoftheregulationbyYTHDC1ofpotentiallyoncogenicgenessuchasRET,PRMT2,RARGandHOXA9couldbeinvolvedinthepathogenesisofrelatedtumors.Infact,experimentalstudiessupporttheimplicationsforYTHDC1incancerprogressionwithregardtoangiogenesis,growthfactorsignaling,immortalization,geneticinstability,tissueinvasionandapoptosis[59,64,65].Similarly,thetranslationalsilencerTIA-1,alsoreportedtoinducemRNAdecay[66–68],ispredictedtointeractwiththeubiquitouslyexpressedNAP1L1transcript(interactionpropensity = 113correspondingtoanAUCof95%),consistentwithiCLIPdataforHeLacells(ArrayExpress:E-MTAB-432)[69](Additionalfile4:TableS2).DeregulationofNAP1L1expressionhasbeendocumentedforseveraltumorssuchassmallintestinecarcinoidneoplasia[70],neuroendocrinetumors[71],ovariancancer[72]andhepatoblastomas[73].WehypothesizethatTIA-1playsafundamentalroleinthepost-transcriptionalregulationofNAP1L1andthatalterationofthisregulatoryprocesscontributestoNAP1L1-associatedtumordevelopment.Wenotethatrepressionofaberrantinteractionscanbeachievedbygenesilencing,whichpreventsthepotentialstabilizingactionofRBPsonspecifictranscripts(Additionalfile2:FigureS5B).Forinstance,theNodalgeneisnormallysilencedinadulttissuesanditsexpressionisassociatedwithtumorprogression[74].SinceNodalisamemberoftheTransformingGrowthFactorβ(TGFB)superfamilyandcontrolsmesodermformationandaxialpatterningduringembryonicdevelopment[74],itispossiblethatNodalinteractionswithspecificRBPsleadtopathogenesisinadulttissues.OurpredictionsindicatethatthetranscriptNodalinteractswithanumberofanti-expressedRBPs(ADD1,API5,ARCN1,CANX,CAPRIN1,CCT6A,DKFZP434I0812,GSPT1,HSP90AB1,PKM,PUF60,XRCC5,YTHDC1andYWHAZ).SincetheexactmechanismregulatingNodalisatpresentunknown,wegeneratedalistofproteinpartnersthatcouldbeexploitedforfutureexperimentalstudies(Additionalfile5:TableS3).ConclusionsComparativeexpressionstudiesprovideimportantinsightsintobiologicalprocessesandcanleadtothediscoveryofunknownregulationpatterns.Whileevolutionaryconstraintsontissue-specificgeneexpressionpatternshavebeenextensivelyinvestigated[7–9,75,76],theconstitutiveregulationofRBP-mediatedinteractionsisstillpoorlyunderstood[11,12].IthasbeenpreviouslyobservedthatcellularlocalizationandgeneexpressionlevelsimposestringentconditionsonthephysicochemicalpropertiesofbothproteinandRNAsequences[77,78],butlarge-scalecomputationalanalysesofconstitutiveRBP-mediatedregulatorynetworkshaveneverbeenattemptedbefore.Ourstudyshowsforthefirsttimethattheintegrationofinsilicopredictions[19]withexvivoexpressionprofiledata[6,34]canbeusedtodiscoverdistinctfeaturesofRBPbiologicalfunctions.WeobservedanenrichmentofuniqueandfunctionallyrelatedGOtermsforRBP–mRNApairsassociatedwithhighinteractionpropensitiesandspecificexpressionpatterns.Inouranalysis,co-expressionofinteractingmRNA–RBPpairs(ICset)islinkedtoregulationofproliferationandcellcyclecontrol,whileanti-expression(IAset)isacharacteristicfeatureofsurvival,growthanddifferentiation-specificprocesses.WedonotexcludethatRBP–mRNAassociationsdisplayingpoorinteractionpropensities(NICandNIAsets)mighthaveimportantevolutionaryimplicationsasspatiotemporalseparationandlimitedchemicalreactivitycouldbewaystoavoidaberrantassociations[55].WefoundthatRNA-bindingproteinsareenrichedinstructurallydisorderedregionsandthatunfoldedpolypeptidefragmentspromoteassociationwithRNAmoleculesatlowinteractionpropensities.Asdisorderedproteinsarehighlyreactive[37],itisreasonabletoassumethatinteractionwithRNAneedstobetightlyregulatedtoavoidcellulardamage[39].Inthisregard,ourresultsexpandatthenucleicacidlevelwhathasbeenpreviouslyobservedforthegeneralpromiscuityofnativelyunfoldedproteins[38,79].Inconclusion,wehopethatourstudyofprotein–RNAinteractionandexpressionwillbeusefulinthedesignofnewexperimentsandforfurthercharacterizingribonucleoproteinassociations.AlistofproposedinteractionsandaserverfornewinquiriesareavailableatthecatRAPIDexpresswebpage[27].MaterialsandmethodsPredictionforLIN28BandHuRinteractionsWeperformedanumberofteststoassessthequalityofourcalculations(seesectiononRNA-bindingprotein–mRNAinteractionpropensity)usingPAR-CLIPdata[31,33].Inthisanalysis,weusedalltheRNAinteractionspresentinourdataset(positiveset:285sequencesforLIN28Band579forHuR)and,duetotheunavailabilityofnon-boundRNAs,thefulllistofhumantranscripts(negativeset:105,000sequences).Forthesignificanceofinteractionpredictions,weperformedFisher’sexacttestcomparingthetop1%ofpredictedinteractionswiththeremainingprotein–RNAassociations(HuR:P = 10-8;LIN28B:P = 10-3).Fisher’sexacttestwascomputedusingequalamounts(thatis,1%ofthetotalinteractions)ofrandomlyextractednegativesubsets(HuR:P = 10-7;LIN28B:P = 0.0002;Additionalfile2:FigureS3).Forthesignificanceofscoredistributions,weusedStudent’st-testtocomparethescoredistributionofpositivesandnegatives(HuR:P = 10-16;LIN28B:P = 10-4).WealsoperformedStudent’st-testusingrandomextractionsofnegativesubsets,eachcontainingthesamenumberofRNAsaspositives(LIN28B:P = 0.03;HuR:P  r th − n no ‒ int r > r th n no ‒ int r > r th (1) InEquation(1),thecorrelationcoefficientrfollowsthedistributionofprotein–RNAexpressionandtheparameterr th  > 0correspondstoanAUCspanningtherange50%to99.5%(Figure 2B).Similarly,fornegativelycorrelatedexpressions(Figure 2D): enrichment anti ‒ expressed interactions = n int r < l th − n no ‒ int r < l th n no ‒ int r < l th (2) InEquation(2),theparameterl th   0.1)intheothers. 3. Theenrichmentmustbeconservedwithrespectto:(a)theentirehumantranscriptome(thatis,includingRNAslongerthan1,200nucleotidesandindependentlyofexpressiondata),(b)thecompletesetofanalyzedgenes(thatis,includingRNAsshorterthan1,200nucleotidesandwithavailableexpression)and(c)allgenesunderthesameAUC(thatis,consideringbothinteractingandnon-interactingpairsatthetwotailsofthedistribution). 4. ThePvalueoftheGOtermmustbenon-significant(P > 0.1)in:(a)thecompletesetofanalyzedgenescomparedtothehumantranscriptome(significancewouldindicateenrichmentirrespectiveofthesubsetassignment)and(b)thelistoftranscriptscompatiblewithcatRAPIDlengthrequirementscomparedtothehumantranscriptome(significancewouldindicatelengthbiasinthestatistics;seesectiononlengthbiasstatistics). 5. Theenrichmentmustbeconservedaftersequenceredundancyreductiontothe80%identitythreshold. LengthbiasstatisticsDuetotheconformationalspaceofnucleotidechains,predictionofRNAsecondarystructuresisdifficultwhenRNAsequencesare>1,200nucleotidesandsimulationscannotbecompletedonstandardprocessors(2.5 GHz;4to8 GBmemory).ToseewhetherGOenrichmentisbiasedbythecatRAPIDlengthrestriction,weusedahypergeometrictest(seesectionontheRNA-bindingprotein–mRNAinteractionpropensity).IfaGOtermisenrichedinthelength-restrictedset,itisexcludedapriorifromtheanalysisbecausegenesannotatedinthatGOtermwouldbeonlyselectedforthelengthrange.Thus,weimposedthatGOtermsmustbenon-significant(P > 0.1)inthelength-restrictedsetofgenes(seesectionongeneontologyanalysis).ThisconditionensuresthatthereisnobiasduetolengthrestrictionsforanyGOtermenrichedinaparticularsubset(Additionalfile3:TableS1).AnalysisofRNA-bindingproteinsequencedisorderThecontentofdisorderedregionsintheRBPsequenceswascomputedusingIUPred[36].Foreachprotein,weextractedstructurallydisorderedregions(IUPredscorehigherthan0.4)andcalculatedtheirinteractionsagainstthereferencetranscriptome.Wecomparedtheinteractionpropensitiesofeachdisorderedregionwiththatofthefull-lengthproteinandassessediftherewasanincreaseordecreaseoftheinteractionpropensityscore(Figure 3A).ThecontributionofthedisorderedregionwasevaluatedusingaMann–WhitneyUtest,whereasignificantincrease(P