integration of transcriptome, proteome and …ruedi aebersold# affiliation: department of biology,...

Integration of transcriptome, proteome and phosphoproteome data elucidates the genetic control of molecular networks Authors JanGroßbach*Affiliation:ClusterofExcellenceinCellularStressResponsesinAging‐associatedDiseases,Cologne,Germanymail:jan.grossbach@uni‐koeln.deLudovicGillet*Affiliation:DepartmentofBiology,InstituteofMolecularSystemsBiology,ETHZurich,Zurich,Switzerlandmail:[email protected]ément‐Ziza*Affiliation:CenterforMolecularMedicineCologne,Cologne,GermanyLesaffre,Marcq‐en‐Baroeul,Francemail:[email protected]:ClusterofExcellenceinCellularStressResponsesinAging‐associatedDiseases,Cologne,Germanymail:corinna.schmalohr@uni‐koeln.de

OlgaT.SchubertAffiliation:DepartmentofHumanGenetics,UniversityofCalifornia,LosAngeles,LosAngeles,CA,USAmail:[email protected]: NovoNordiskResearchCenterSeattle,Inc.,Seattle,WA,USAMail:[email protected]:DepartmentofBiology,InstituteofMolecularSystemsBiology,ETHZurich,Zurich,Switzerlandmail:[email protected]#Affiliation:DepartmentofBiology,InstituteofMolecularSystemsBiology,ETHZurich,Zurich,SwitzerlandandFacultyofScience,UniversityofZurich,Zurich,Switzerlandmail:[email protected]#Affiliation:ClusterofExcellenceinCellularStressResponsesinAging‐associatedDiseases,Cologne,GermanyCenterforMolecularMedicineCologne,Cologne,Germanymail:andreas.beyer@uni‐koeln.de*equalcontribution#correspondingauthor

.CC-BY-NC 4.0 International licenseacertified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under

The copyright holder for this preprint (which was notthis version posted July 16, 2019. ; https://doi.org/10.1101/703140doi: bioRxiv preprint

https://doi.org/10.1101/703140http://creativecommons.org/licenses/by-nc/4.0/

Großbach,Gillet,Clément‐Ziza,etal.

Page2

Abstract GenomicvariationaffectscellularnetworksbyalteringdiversemolecularlayerssuchasRNAlevels,proteinabundance,andpost‐translationalproteinmodifications.However,itremainsunclearhowthesedifferentlayersareaffectedbygeneticpolymorphismsandgiverisetocomplexphysiologicalphenotypes.Toaddressthesequestions,wegeneratedhigh‐quality transcriptome, proteome, and phosphoproteome data for a panel of 112genetically diverse yeast strains.While genetic effects on transcript abundancesweregenerally transmitted to the protein level, we found a significant uncoupling of thetranscript‐protein relationship for certain protein classes, such as subunits of proteincomplexes.Theadditionalphosphoproteomicsdatasuggeststhatthesamegeneticlocusoftenaffectsdistinctsetsofgeneswithineachoftheselayers.Inparticular,QTLstendedto affect upstream regulatory proteins at the phosphorylation layer, whereasdownstream pathway targets were typically affected at the transcript and proteinabundancelayers.Underscoringtheimportanceofregulatoryproteinphosphorylationinlinking genetic to phenotypic variation is the finding that the number of proteinphosphositesassociatedwithagivengeneticlocuswasmorepredictiveforitsinfluenceoncellulargrowthtraitsthanthenumberoftranscriptsorproteins.Thisstudyshowshowmulti‐layeredmolecularnetworksmediatetheeffectsofgenomicvariantstomorecomplexphysiologicaltraitsandhighlightstheimportantroleofproteinphosphorylationinmediatingtheseeffects.




Multiomicssystemsgenetics

Page3

Introduction Genetic polymorphisms are importantmodifiers ofmany physiological traits, such asbodyheightordiseasesusceptibility.Differencesinthesetraitsarecausedbyalterationsoftheunderlyingmolecularnetworks1.Therearevariouswayshowgeneticvariantscaninfluence these molecular networks. For instance, genetic variants might either actindirectly through multiple layers, e.g. by first changing mRNA levels, which theninfluenceproteinlevels,ordirectlyonindividual layers,e.g.bychangingproteinlevelsthrough altered protein stability2. Changes in the sequence of an mRNA can affecttranslation rates, hence altering protein abundance, while the concentration of itscorrespondingmRNAstaysconstant(reviewedin3).Thenetactivityofproteinscanbemodulatedbychangingtheproteinlevelsand/orbychangingtheirspecificactivities,e.g.by increasing or decreasing regulatory phosphorylation. Varying concentrations of akinasecanaffectphosphorylationlevelsofitstargets,which,inturn,mightleadtotheirdegradation4,activationorinhibition5,orchangesintheirsubcellularlocalization6.Thelocalconcentrationofaspecificphosphorylatedprotein(e.g.,atranscriptionfactor)mightinturnaffecttranscriptionofyetothergenes.Thus,thecrosstalkbetweenthedifferentmolecular layers is characterized by a great diversity of mechanisms, includingbiosynthetic‐andsignallingprocesses2.High‐throughput molecular profiling (‘omics’) technologies have enabled theinvestigation of the impact of genetic variation on individual molecular layers on agenomicscale.Suchdatacanbeusedtoidentifygeneticvariantsthatexplainsomeofthevariationinthemeasuredtraits,termedquantitativetraitloci(QTLs).Forexample,withcurrentRNAsequencing(RNA‐seq)technologiesitispossibletodetectexpressionQTLs(eQTLs)forvirtuallyalltranscriptspresentinacell7.Likewise,massspectrometryhasbeen used to identify QTLs for hundreds of proteins (pQTLs)8–11 andmetabolic traits(mQTLs;reviewedin12).Ashortcomingofmanyexistingstudiesisthateitheronlyonemolecularlayerwasmeasured,typicallytranscripts,or,ifmultiplelayersweremeasured,transcripts and proteins were not isolated from the same cultures, reducingcomparabilityduetopotentialdifferencesinenvironmentaleffects.Furthermore,post‐translationalproteinmodifications(PTMs),suchasphosphorylation,havebeenlargelyneglectedandnosystematicphosphorylationQTL(phQTL)studieshavebeenperformedtodate.A comprehensive viewof the effects of geneticperturbationson integratedmolecularnetworks requires accurate quantitative measurements at different molecular layersfrom the same samples14. Therefore, we designed a multi‐omics QTL study in





Page4

recombinant offspring of a cross of two budding yeast strains, with four keydistinguishing features: (i) we quantified transcripts, proteins, and proteinphosphorylation levelsatagenomicscale;(ii)byusingtheSWATH‐MStechnologywecouldreproduciblyquantifyalargenumberofproteinsacrossvirtuallyallsamples;(iii)RNA,protein,andphosphoproteinsampleswereobtainedfromthesameyeastcultures,whichgreatlyfacilitatedtheintegrationofthedata;and(iv)thedataintegrationschemethat we developed for this study enabled us to investigate interdependencies of themolecularpatternsbetweenthelayers.Thissetupenabledus,forthefirsttime,tomapthe response of cellular signalling networks to genomic variation, to dissect theinteractionoftranscript‐andproteinabundancechanges,andtoinvestigatetherelevanceofphQTLsforcomplexcellulartraits.





Page5

Figure1 |Experimentaldesignandstudyoverview.Yeastsegregantsderivedfromtwoparental strains, BY andRM,were grown and characterizedwith different omicsapproaches. Their transcriptome, proteome, and phosphoproteome were directlyquantifiedinsamplesobtainedfromthesamecultureflask.Residualtraitsrepresentingthedisparitiesbetweentranscriptandproteinlevels(i.e.,measuringpost‐transcriptionalregulation, light green), and those between protein and phosphopeptide levels (i.e.,measuringphosphorylationstatus,pink)werecomputed.QTLanalyseswereperformedtodeterminetheeffectofgeneticvariationoneachofthesemolecularlayers.





Page6

Results

Multi‐omics profiling of a yeast panel

TheBYxRMSaccharomyces cerevisiae yeast segregant panel used in this study resultsfromacrossofalaboratorystrain(BY4716)isogenictothereferencestrainS288C,andaderivativeofawildisolatefromavineyard(RM11‐1a).Thiscrosswaspreviouslyusedtostudythegeneticcontributiontomoleculartraits–includingRNAandproteinlevels–and to testnovel systemsgeneticsapproaches15–17.Wegrewthe twoparentalhaploidyeaststrainsand110oftheirrecombinantoffspringundertightlycontrolledconditions,withmultiple replicates for some of the strains (Fig. 1, Supplementary Data S1). Thetranscriptomesof150culturesweresequencedathighcoverage(38‐186x)allowingforthequantificationof5,429transcriptsinallsamples(SupplementaryDataS2).TheRNA‐seq datawas alsoused to infer the genotypes of the strains, using a strategy thatwedevelopedpreviously18(SupplementaryFigureS1,SupplementaryDataS3),significantlyimproving the localization of recombination sites compared to previous studies19. Insamplesobtainedfromthesameyeastcultures,weusedSWATH‐massspectrometry(MS)tomeasuretheabundancesof1,862proteinswithlessthan1.8%missingvaluesacrossallsamples(Fig.2a,SupplementaryDataS4).Thisrepresentsafour‐foldincreaseinthenumberofquantifiedproteinscomparedtopreviousstudiesinthesamecross8,11,16,17.WealsoquantifiedthephosphorylationstateoftheproteinsbySWATH‐MS;afterstringentfiltering,weobtainedabundancesof2,116phosphopeptidesfrom988proteinswithlessthan2.4%missingvaluesacrossallsamples(SupplementaryDataS5).





Page7

Figure 2 | Genetic control of molecular phenotypes. a, Overlap of quantifiedtranscriptome, proteome, and phosphoproteome. b, Broad‐sense heritability of traitsbelonging to the 402 genes for which measurements at each molecular level wereavailable.Broad‐senseheritabilitywasestimatedby comparing the variationbetweenreplicates from the same strain (which is non‐genetic) to the total variation betweenstrains.Whentheintra‐strainvariationissmallcomparedtotheinter‐strainvariation,itcan be concluded that the genetic contribution to trait variation is large under theexperimentalconditionstested20.c,ProportionofmoleculartraitsaffectedbyatleastoneQTLareshownseparatelyforeachmolecularlayer.ShadedareasindicatetheproportionoftraitsforwhichalocalQTLwasfound.As expected, we observed that in most cases, protein abundances were positivelycorrelatedwiththeirtranscriptlevels(averager=0.23),andmostphosphopeptideswerepositively correlated with their proteins of origin (average r=0.29) (SupplementaryFigureS2).However,eachlayercanalsobeaffectedindependentlyofthegeneticeffectsontheotherlayers.Inordertoidentifydirecteffectsthatactonamolecularlayer,we





Page8

computationally generated traits fromwhich effects acting on othermolecular layerswereremovedaspreviouslyproposed8.First,weestimated thecontributionofmRNAchanges to protein‐level changes by regressing the concentration of a given proteinagainsttheconcentrationofitsencodingmRNAacrossallstrains.Deviationsfromthisregression(residuals)caneitherresultfromnoiseinthedata,orfrompQTLeffectsthatareindependentoftranscript‐levelchanges.Weusedtheresidualsasestimatesofpost‐transcriptionalregulation,resultingin1857post‐transcriptional(pt)traits8.Likewise,weregressed phosphopeptide levels against levels of the proteins of origin and used theresidualsasestimatesofdifferentialphosphorylation,resultingin879phospho‐residual(phRes) traits.TheQTLsobtainedbymapping theseresidual traitsweretermedpost‐transcriptionalQTL(ptQTL)andphospho‐residualQTL(phResQTL),respectively.

QTLs have widespread effects on all quantified molecular layers

In order to quantify the fraction of trait variation that can be attributed to geneticdifferences,wequantifiedbroad‐senseheritabilitybyleveragingavailablereplicatesasproposed before20 (Fig. 2b). All five types of molecular traits outlined above hadheritabilitiesgreaterthanexpectedbychance(p


Page9

biologicalreasons,duetoincreasednoise(asdiscussedabove),oracombinationofthetwo.Next,weaskedwhetherlocaleQTLsandpQTLscouldbeattributedtochangesinthesequenceoftherespectivetranscript,whichmightinfluencetranscriptionandtranslationrates.WhencomparinggeneswithlocalQTLs(i.e.,localeQTLs,pQTLs,andptQTLs)withgenesthatareonlyaffectedbydistantQTLs,wefoundthattheformerhadanincreasednumberofpolymorphismsinnon‐codingregions(e.g.,5'untranslatedregions(UTRs)or3’UTRs;Fig.3a).TheexistenceoflocalptQTLs(129localptQTLsfor6.9%ofallpttraits)and the fact thatptQTLswere enriched forpolymorphismsoutside of coding regions,suggest that variants in non‐coding parts of the genome can influence protein levelsindependently of their coding transcripts8, for example through polymorphisms inribosomalbindingsitesaffectingtranslationinitiationorvariantsalteringmRNAcappingandlooping.Regionsinthegenomethataffectsignificantlymoretraitsthanexpectedbychancearereferred to asQTL hotspots22. Since these loci affect large numbers of traits, they areassumedtoactthroughmasterregulatorssuchastranscriptionfactorsorkinases7,23.Wetestedforregulatoryhotspotsasproposedbefore15(Methods)anddetectedbetween9and15significanthotspotsforeachmolecularlayer,withthelargestnumberofhotspotsbeingdetectedfortheeQTLlayer(Fig.3b,SupplementaryTableS2).PreviousworkfoundthatmostdistanteQTLsactfromwithinhotspots7.OurdatashowsthatthisisthecaseforQTLs of all five types ofmolecular traits considered here (Supplementary Figure S4).ManyofthedetectedhotspotshavebeenreportedaseQTLhotspotsforthisyeastcrossbefore.For someof themacausal genehasbeenvalidated,e.g.,HAP1 (chrXII:2), IRA2(chrXV:1), and MKT1 (chrXIV:1)15,22. While most of the hotspots affected multiplemolecular layers simultaneously, we also observed hotspots that predominantlyimpacted the transcriptome (e.g., chrV:2), the proteome (e.g., chrXII:1) or thephosphoproteome(e.g.,chrVIII:1).Thus,themolecularmechanismofaQTLmayhaveastronginfluenceonwhichmolecularlayerisprimarilyaffected.

Transmission of genetic effects from the transcriptome to the proteome

The deep coverage of the transcriptome and proteome in this study enabled us toinvestigatetowhatextenttranscriptlevelchangesaretransmittedtotheircorrespondingproteins. First, we observed that local eQTLs and local pQTLs were significantlyoverlapping(p


Page10

between the two sub‐populations. When comparing transcript fold changes with therespectiveproteinfoldchanges,wefoundthattheywerestronglycorrelated(eQTLswithFDR


Page11

functionsrelatedtocytoplasmictranslation(SupplementaryTableS6).Thisunexpectedfunctional similarity between buffered proteins and proteins subject to ‘protein only’effects raised the question whether the same proteins could be subject to bothphenomena.Indeed,wefoundthatgenesaffectedbyabufferedeQTLweremorelikelytoalsobeaffectedbya ‘proteinonly’QTL(219genes;p


Page12

Figure3 |Genomiceffectson transcriptandprotein levels.a,AveragenumberofSNPsinandaroundgenesaffectedbylocalQTLsattherespectivemolecularlayer.Thehorizontal dashed black line shows the average number of SNPs across all annotatedgenesintherespectiveregions(Methods).b,NumberoftraitsaffectedbyeachlocusatFDR


Page13

Protein complex stoichiometry controls QTL effects

Thebufferingofcytoplasmicribosomalproteinsisawell‐describedphenomenon:excessproteins not incorporated into ribosomes get degraded24. Recently, buffering ofcomponentsofotherproteincomplexeswas reported25–27, suggesting that theneed tomaintainproteincomplexstoichiometrydrivespost‐transcriptionalproteinabundanceregulation. Indeed, we observed extensive co‐regulation of genes whose productsparticipateincommoncomplexesonmultiplemolecularlayers:bothtranscriptlevelsandproteinlevelsofproteincomplexmemberswerecorrelatedacrossthestrains(averagePearson’s correlation on RNA and protein levels: r=0.8 and r=0.46, respectively).Furthermore,pairsofgenesofthesamecomplexwereaffectedbythesameeQTLorpQTLatasignificantlyhigherratethanpairsofgenesfromdifferentcomplexes.Whereasthisphenomenonwasparticularlystrongforribosomalproteins,westillobservedsignificantenrichment after excluding ribosomal proteins from the analysis (eQTLs: p


Page14

butonly6.5%and5.2%ofthecorrespondingtranscriptsandproteinsofthe402geneswhoseproductsweredetectedonallthreemolecularlayers(Fig.4a).Next,weaskedwhethergenomicvariationcouldresultincoordinatedeffectsondifferentphosphositesonthesameprotein.Indeed,phRestraitsofthesameprotein(i.e.,differentphosphosites on the same protein) had a higher chance to be targeted by the samephResQTLthanrandompairsofphosphosites(p


Page15

Local phResQTLs might be caused by genetic variants directly affecting thephosphorylationstateofaprotein,forexamplethroughthemodificationofresiduesclosetoakinaseorphosphatasebindingsite.Insupportofthatnotion,wefoundthatproteinsthatwereaffectedbyat leastone localphResQTLhadmoremissensepolymorphismsthanproteinsthatonlyhaddistantphResQTLs(p


Page16

availabilities30.EarlierworkhasshownthatpolymorphismsintheIRA2codingsequenceaffect theactivityof theRas/Pkapathwayand that theRMalleleof IRA2 inhibits thispathway more efficiently22. Several targets of phQTLs in the IRA2 hotspot werefunctionally connected to IRA2 including a local phQTL targeting Ira2 itself, a phQTLtargetingCdc25,whichpromotestheGTP‐boundformofRas2,andaphQTLtargetingtheRas2protein31. In addition,wedetectedphQTLs fornumerousdownstream targetsofRas/PkasignalingattheIRA2hotspot(SupplementaryTableS8).Hence,ourdatarevealswidespread modifications of key signaling molecules in the Ras/Pka pathway upongeneticvariationandcontributestoabetterunderstandingofthemolecularmechanismsthroughwhichthisQTLhotspotacts.AnotherexampleillustratingtheeffectofgeneticvariantsoncellularsignalingistheHAP1hotspot on Chromosome 12 (chrXII:2)15,17, which was previously shown to affect 19growthtraits20.Hap1hasbeenshowntobeinvolvedintheregulationofrespirationinresponsetooxygenandirondeprivation32.Again,ouranalysisrevealedthatthislocushasmuch more prominent effects on the phospho‐layer than on transcript or proteinabundances(Figure4a).Theintegrationofourmulti‐layeredQTLdatawithpreviouslypublishedkinase‐substratenetworks suggestedanewmodeof effect for thishotspot:boththeexpressionandphosphorylationofPsk2,aproteinkinaseandknownregulatorofcarbohydratemetabolism33,weresignificantlyregulatedbytheHAP1hotspot.Proteinswhose phosphorylation was affected by theHAP1 locus were enriched in previouslyreportedtargetsofthekinasePsk234.ThisenrichmentofphosphorylatedPsk2substratesstrongly implies a directed signaling cascade, from the HAP1 locus via altered Psk2activitytothePsk2substrates.The added value of integrating protein phosphorylation with transcript and proteinabundanceswasevenmoreapparentfromahotspotonChromosome8(chrVIII:1)withthe pheromone response gene GPA1 harboring the causal mutation23. The hotspotharboredonly69eQTLs,8pQTLs,and1ptQTL,but41phQTLs(SupplementaryTableS2,Fig. 3b).Many of the hotspot eQTL targets (41%) could be found downstream of thematingpheromoneresponsepathway,whichsupportsGPA1asacausalgene.However,severaltargetsofthishotspot,especiallythoseonthephospholevelsuchasRck2andRsc9,wereinvolvedintheosmoticstressresponse,whichisunrelatedtoGPA135–37.ThissuggeststhatthepolymorphismsinGPA1donotfullyexplaintheeffectsofthishotspot,promptingustolookforotherpotentialkeyeffectorsofthehotspot.Thegeneticmarkerwith themostphQTL targetswithinthehotspot is locatedwithin thecodingregionofSTE20.Ste20isthekeyactivatorofmultiplemitogen‐activatedproteinkinase(MAPK)pathways, including the mating pheromone response, but also invasive growth





Page17

regulation,regulationofsteroluptake,and,importantly,osmoticstressresponse35(Fig.5). Indeed, forup to60%of the target genesof thishotspotwe founda link toSte20and/orGpa1(SupplementaryTablesS9andS10).Further,mostofthephosphopeptidesandphRes traits targetedby thehotspotcorrespondedtoproteinsphosphorylatedbycomponentsofMAPKpathwaysdownstreamofSte20(51%;Fig.5,SupplementaryTablesS9 and S10). GO enrichment analysis of targets of the hotspot at the transcript layerrevealed an enrichment for biological processes under the influence of STE20(SupplementaryTableS11).Takentogether,theseresultssuggestthattheeffectsofthechrVIII:1 hotspot are due to the combined effects of the polymorphisms inGPA1 andSTE20.Furthermore,weprovidenewpotentialcomponentsoftheMAPKpathwaysthatareregulatedbythishotspotbutwerenotpreviouslyidentifiedasdownstreamtargetsofSte20(SupplementaryTableS12).TheexamplesofthesehotspotsillustratehowphQTLmappingprovidesinformationonsignalingnetworksthatisorthogonaltotranscriptandproteinabundancedata:geneticvariantsoftenaffectthephosphorylationstatesofgeneproductsthataredistinctfromthe genes affected by abundance changes (i.e. eQTL and pQTL targets). Furthermore,thoseproteinactivitychangesoftenactupstreamofQTLabundanceeffects.Overall,weshowthatbyquantifyingtheeffectsofsequencepolymorphismsonmulti‐layermolecularnetworksourintegratedapproachcanprovidecluestowardreconstructingthemoleculararchitectureunderlyingcomplextraitsandthechainofcausalitythroughmultiplelayersofregulation.





Page18

Figure5:Multi‐layereffectsofthehotspotaroundGPA1andSTE20onChromosome8.Gpa1andSte20playkeyrolesinthematingresponsepathway(left,throughtranscriptionfactor complex Ste12/Ste12) and filamentous growth regulation pathway (middle,through transcription factor complex Tec1/Ste12). Ste20 additionally regulates theosmotic stress response pathway (right, through transcription factors Msn2/Msn4).Depicted are genes that are known to have a direct or indirect connectionwithGPA1and/orSTE20.Genes thatwe found tobeaffectedby theGPA1/STE20hotspotonanymolecularlayerareindicatedwitharrowsintherespectivecolorofeachlayer.Thearroworientationindicatesthefoldchangedirection.ReferencesforeachknownconnectionaregiveninSupplementaryTableS9.





Page19

Discussion In this study, we present the first dataset that concurrently quantifies the effects ofnatural genomic variation on the transcript (eQTL), protein (pQTL), and proteinphosphorylation(phQTL)layers.Whereasprevious ‘omics’QTLstudieshaveprimarilyfocusedontheabundanceofbiomolecules,ourstudyintegratesabundancewiththestatesofbiomolecules.Hence,theconceptualadvanceofourstudyisthepossibilitytodiscoverthe consequences of genomic variation, for example the phosphorylation‐mediatedactivationofasignalingpathwaycomponent,throughthedifferentlayersofamolecularnetwork. The three hotspots containing IRA2, HAP1, and GPA1/STE20 exemplify thisnotion.Quantifyingtranscriptandproteinlevelsalongwiththestateoftheproteinsindicatedbyphosphorylationfromthesameyeastculturemaximizedthecomparabilityofthedata,thus facilitating data integration. This three‐layer molecular profiling enabled thefollowing findings: (i)RNA effects are generally transmitted to protein levels in a 1:1relationship.(ii)Thereare,however,numerousexceptionsandspecificclassesofgenesare subject to enhanced (e.g., mitochondrial ribosomes) and buffered (e.g., cytosolicribosomesandnucleolus)protein‐leveleffects.(iii)Themaintenanceofproteincomplexstoichiometryplaysamajorrole in thepost‐transcriptionaleffectsofgeneticvariants.(iv)Asubstantialfractionofphosphopeptides(44%)wereunderthedirectcontrolofatleast one QTL, independently of the abundance of its protein of origin. Further, weobserved multiple cases in which protein abundance effects were buffered(i.e.,repressed)atthelevelofphosphopeptides.(v)QTLsaffectmultiplephosphositesonthe sameproteinmoreoften thanexpectedby chance. (vi)Proteinphosphorylation ismuchmoreoftenaffectedby localmissensevariants thanprotein levelsand,thus, thestructural destabilization of proteins through segregating variants seems to be a rareevent and is presumably under strong negative selection. (vii)Variants affectingphosphorylation are typically in close proximity to the affected residue. (viii)Thecharacterization of phosphorylation events in addition to transcript and proteinabundances enabled the separation of upstream and downstream effects of geneticvariantsonsignalingpathways.Importantly,thesameQTLcanaffectdistinctsetsofgenesatdifferentmolecularlayers.(ix)QTLhotspotswererarelyexplainedbyvariantsdirectlyaffectingthetranscriptencodedatthatlocation;instead,ourdatasuggeststhatalteredproteinphosphorylation contributes significantly to the causalmechanismsof severalQTLhotspotsbytheactivationofpleiotropicdownstreameffects.





Page20

Thisstudycapturessomeoftheenormouscomplexityinhowgeneticvariantsinfluencesignal transmission between different molecular layers, despite all the data beingobtained from steady‐state standard growth conditions. Clearly, mappingphosphorylation states will be even more critical when aiming to understand theinteractionbetweengeneticvariantsandenvironmentalcues. Inaddition, futureworkshould address additional aspects of molecular state changes, including other PTMs,changesinproteinfolding,proteincomplexformation,andprotein‐ligandinteractions.Our findingshave implications for futureworkoncomplexhumandiseases.First,ourstudyshowsthat,atleastinyeast,quantifyingphosphoproteomesatlargescaleandwithhigh precision and reproducibility is feasible with state‐of‐the‐art proteomicstechnologies. Thus, genome‐wide association studies (GWAS) on phosphoproteomechangesinhumancellsmightbepossiblesoonaswell.Second,ourfindingsimplythatmapping phospho‐traits and molecular states in general will greatly contribute tounderstanding themechanisms throughwhich GWAS loci act. Third, phosphorylationstates aremore likely to harbor information aboutphysiological processes thanmereabundancesoftranscriptsorproteins(SupplementaryFigureS11c).Lastly,ourresultssuggestthatcausalcis‐variantsareoftennearbythealteredphosphosite,whichcanhelpprioritizingcandidatevariants.Inconclusion,ourstudysetsthestageforamoremechanisticunderstandingofgenomiceffectsonmulti‐layeredcellularnetworksandphysiologicaltraits.

Acknowledgements JGandCSreceivedfundingfromtheGermanScienceFoundation(DFGCRC680&CRC1310).MCZandCSreceivedfundingfromtheGermanFederalMinistryforScienceandEducation (BMBF) as part of the e:Med program (grant no. 01ZX1406). OTS receivedfundingfromtheGermanAcademicExchangeService(DAAD),theSwissNationalScienceFoundation,andtheHumanFrontiersScienceProgram.IBwassupportedbytheSwissNationalScienceFoundation(grantno.31003A_166435).WethankDr.RachelBremforprovidingthestrainsoftheBY/RMcross.

Author Contributions AB,RA,andMCZenvisionedthestudy.LG,MCZ,andCABperformedtheexperiments.JG,CLS,MCZ,OTS, IB, andLGperformed thedata analysis.All authors contributed to thewritingofthemanuscript.





Page21

Competing financial interests Inthecourseofthestudy,MCZbecameanemployeeofLesaffreInternational,France.

Materials and methods

Sample preparation

Allmediawerepreparedinasinglebatchtolimitexperimentalvariability.TheBYxRMyeaststraincollection,whichweobtainedfromRachelBrem,wasoriginallyderivedfromacrossbetweenthetwoparentalstrains,BY4716,anS288Cderivative(MATαlys2Δ0),andRM11‐1a(MATaleu2Δ0ura3Δ0ho::KAN)15.Asubsetof129strainswerepickedinrandom series of 16 (see Supplementary Data S1), pre‐cultured in in‐house madesyntheticdextrosemedium(S.D.,containingperliter:1.7gyeastnitrogenbasewithoutamino acids (Chemie Brunschwig), 5 g ammonium sulfate, 2% glucose (w/v), 0.03 gisoleucine,0.15gvaline,0.04gadenine,0.02garginine,0.02ghistidine,0.1gleucine,0.03 g lysine, 0.02 g methionine, 0.05 g phenylalanine, 0.2 g threonine, 0.04 gtryptophan,0.03gtyrosine,0.02guracil,0.1gglutamicacidand0.1gasparticacid),andthengrownin115mlfreshS.D.mediumat30ºCuntilamaximalopticaldensityat600nm(OD600)of0.8(+/‐0.1).Intotal,180cellculturesweresuccessfullygrowntoOD600of0.8andthensubdividedasfollowsforthetranscriptandproteomicanalysesrespectively.Ofthe115mlofcultures,15mlwerecollected,centrifugedat2000gfor3min at 4°C, transferred into an Eppendorf and snap frozen in liquid nitrogen fortranscriptomicanalysis.The remaining100mlwereprocessed forproteomicanalysisessentiallyasdescribedin34. Inshort,6.66mlof100%trichloroaceticacid(TCA)wasaddedtothe100mlculturemediatoafinalconcentrationof6.25%andthecellswereharvestedbycentrifugationat1500gfor5minat4°Candwashedthreetimeswithcoldacetone.Thecellpelletsweretransferredinto2‐mlEppendorftubesandfrozeninliquidnitrogen.

RNA sequencing

RNAextraction

TotalRNAwasisolatedfromdeepfrozenaliquotsofyeastpelletsusingtheRiboPure™RNA Purification Kit, yeast (Ambion), which includes a DNase treatment to eliminatecontamination.RNAqualitywasassessedusingRNAScreenTapeassay(Agilent).AllRNAswereofveryhighquality(SupplementaryDataS1,medianRIN9.8,minimalRIN9.1).RNA‐seqlibrarypreparation





Page22

cDNAlibrarieswerepreparedfrompoly(A)selectedRNAapplyingtheIlluminaTruSeqprotocolformRNAusingatotalof1μgRNApersampleand14PCRcycles.SequencingThecDNAlibrariesweresequencedonaHiSeq2000with20samplesperlane(8millionreadspersample).Thegeneratedreadswerestranded,single‐end,andhadalengthof100bp.

RNA‐seq based genotyping

The BYxRM yeast cross has been widely used for QTL mapping. Microarray‐basedgenotypeinformationofthesegregantsisavailable19.However,deepsequencingenablesamoreaccurategenotypingofrecombinantlines.Tothisendwehaveexploitedpublishedresequencingdataof theparental strains20 togetherwith theRNA‐seqdata generatedheretoinferthegenotypeofthesegregantsoftheBYxRMcrossusingamethodthatwepreviouslydeveloped18.

Resequencing data of the parental strains

Sequence variation information of the parental strains was obtained fromhttp://genomics‐pubs.princeton.edu/YeastCross_BYxRM/; only calls with a MQ≥30whereconsidered,whichrepresents42,769polymorphicsites.

Genotype inferring from RNA‐seq data

RNA‐seqdataweremappedtotheS.cerevisiaereferencegenome(SaCer3)usingTopHat2(Ref.38)withthefollowingoptions:--min-intron-length 10 --min-segment-intron 10 --b2-very-sensitive --max-multihits 1 --library-type

fr-secondstrand. Read group informationwas added, and BAM files were sortedusingPicardutilities(http://broadinstitute.github.io/picard/).RNA‐seqdatawerefurtherprocessedusingtheGATKpipeline(version3.4‐46‐gbc02625)followingthebestpracticeguide39.Firstreadscontainingexon‐exonjunctionweresplitusing the following options: -T SplitNCigarReads -rf ReassignOneMappingQuality -RMQF 255 -RMQT 60 -U

ALLOW_N_CIGAR_READS,thenthevariantswerecalledusingtheUnifiedGenotyperatthe42,769polymorphicsitesidentifiedintheresequencingdataoftheparentalstrains.Genotype calls where the GATK genotype score was below 40 (GQ≤40) or that werecoveredwithlessthan5reads(DP≤4)wereconsideredasmissingvalues.

Filtering and missing value imputing

For every polymorphic site between the parental strains, we compared thepolymorphisms in the segregants and the parental strains to infer which allele was





Page23

inherited.Wefurtherexcludedpolymorphisms(i)thatcouldnotbecalled(orcorrectlycalled)intheparentalstrainsbasedonRNA‐seqdata,(ii)thatcouldbecalledinlessthan70% of the segregants, and (iii) with a lower allele frequency of less than 20%. Aspreviouslydiscussed18,genotypescalleddifferingfromthetwodirectflankingmarkersinmore than one segregant (93 cases) probably denote erroneous genotype calls; thecorrespondingpolymorphismswereexcludedfromtheanalysis.Thisresultedin25,590polymorphismsthatwereconsideredasgeneticmarkers.Finallymissinggenotypevalueswereinferredfromthetwoneighboringpolymorphismsifthosewereeachwithin20kbofthepolymorphismofinterest,andwereinheritedfromthesameparentalstrain18(SupplementaryDataS3).

Assembling a set of markers for linkage analyses

Adjacentmarkerswiththesamesegregationpatternacrossallsegregantswerecollapsedintooneuniquemarker,resultinginasetof3,593uniquemappinggenotypicmarkers(SupplementaryDataS13).Thus,eachmarkerrepresentsagenomicintervalinwhichallpolymorphismsareinfulllinkagedisequilibriuminthecross.

Gene expression quantification and normalization

Inpreviouswork,wehaveshownthataccountingforindividualgenomevariationsforRNA‐seqalignmentimprovedgeneexpressionquantificationanddeflatedthenumberoffalsely detected local eQTLs18. Therefore we used the strategy we had previouslydevelopedtomapRNA‐seqreads.Itconsistsofgeneratingastrain‐specificgenome,foreachsegregant,againstwhichthecorrespondingreadsarealigned.First, we generated both strain‐specific genome sequences and strain‐specificannotations from the reference genome sequence (SaCer3), the reference genomeannotationandthegenomicvariationsinformation(VCFfiles)previouslycreatedusingRNA‐seq data. Then RNA‐seq readswere aligned to the corresponding strain specificgenomesusingSTAR(ver2.5.0a,40)withthe followingoptions:--alignIntronMin 10 --quantMode GeneCounts. The gene‐specific read counts (strand specific)generatedbySTARwereusedtoquantifygeneexpression.RNA‐seqcoveragewascomputedbydividingthesumofthelengthofallreadsbythesumofthelengththecodingregionsofthequantifiedtranscriptspersample.RawreadcountswerenormalizedusingrlogmethodofDESeq241.Normalizeddatawerefurthercorrectedforeffectsduetoculturebatchesusingthenon‐parametricempiricalBayesframeworkComBat42(SupplementaryDataS2).Normalizedandbatchcorrectedreadcountswerecorrectedforgenelengthasfollowsforgeneiinsamplej:





Page24

2 2 ∗ 1000 ,whereliisthelengthofthecodingregionofgenei,excludingintronicregions.

Proteomics

Sample preparation and phospho enrichment

Cellpelletswereresuspendedinlysisbuffercontaining8Murea,0.1MNH4HCO3,and5mM EDTA and cells were disrupted by glass bead beating (5 times for 5min at 4°C,allowingthesamplestocooldownbetweencycles).Thetotalproteinamountfromthepooled supernatantswasdeterminedbyBCAProteinAssayKit (Thermo,USA). Threemilligramsofextractedyeastproteinswerereducedwith5mMTCEPat37°Cfor30minandalkylatedwith12mMiodoacetamideatroomtemperatureinthedarkfor30min.Thesampleswerethendilutedwith0.1MNH4HCO3toafinalconcentrationof1Mureaand the proteins were digested with sequencing‐grade porcine trypsin (Promega,Switzerland)atafinalenzyme:substrateratioof1:100(w/w).Digestionwasstoppedbyaddingformicacidtoafinalconcentrationof1%.Peptidemixturesweredesaltedusing3ccreversephasecartridges(Sep‐PaktC18,Waters,USA)andaccordingtothefollowingprocedure:washingof columnwithonevolumeof100%methanol,washingwithonevolumeof50%acetonitrile,washingwith3volumesof0.1%formicacid,loadingacidifiedsample, reloading flow‐through,washingcolumnwithsamplewith3volumesof0.1%formicacid,andelutingsamplewithtwovolumesof50%acetonitrilein0.1%formicacid.Peptidesweredriedusingavacuumcentrifugeandresolubilizedin100µlof0.1%formicacid.Retentiontimestandardpeptides(iRT‐Kit,Biognosys,Switzerland)werespikedintothe samples before theywere analyzedbyLC‐MS for total protein abundances (“non‐enrichedsamples”).Theremaining95µlweresupplementedwith300µlofanovernightre‐crystalizedandclearedupphthalicacidsolutionpreparedbycarefullydissolving5gofphthalicacidin50mlof80%acetonitrilebeforeadding1.75mloftrifluoroaceticacid.Thesampleswere thenenriched forphosphopeptidesby incubating for1hourunderrotationwith1.25mgofTiO2resin(GLscience,Japan)pre‐equilibratedtwicewith500µlofmethanol,andtwicewith500µlofphthalicacidsolution.PeptidesboundtotheTiO2resinwerethenwashedtwicewith500µlphthalicacidsolution,thentwicewith80%acetonitrile with 0.1% formic acid, and finally twice with 0.1% formic acid. Thephosphopeptideswere eluted from the beads twicewith 150 µl of 0.3M ammoniumhydroxideatpH10.5andimmediatelyacidifiedagainwith50ulof5%trifluoroaceticacidto reach about pH 2.0. The enriched phosphopeptides were desalted on microspincolumns(TheNestGroup,USA)withtheprotocoldescribedabove,driedusingavacuum





Page25

centrifuge,andresolubilizedin10µlof0.1%formicacid.Again,retentiontimestandardpeptides(iRT‐Kit,Biognosys,Switzerland)werespikedintothesamplesbeforetheywereanalyzedbyLC‐MSfortotalpeptideabundances(“phosphopeptidessamples”).

LC‐MS data acquisition

ThepeptideconcentrationinallsampleswasmeasuredonaNanoDropatOD280andnormalizedtoallowinjection~1µgofmaterialintothemassspectrometer.Thesampleswere randomized and then either injected individually for SWATH‐MS acquisition orpooled and injected in technical duplicates for shotgun acquisition. The LC‐MSacquisitionswereperformedonanABSciex5600TripleTOFcoupledtoaNanoLC2DplusHPLCsystem.Theliquidchromatographicseparationandmassspectrometricacquisitionparameters were essentially as described earlier43. The peptide separation wasperformedona75µmdiameterPicoTip/PicoFritemitterpackedwith20cmofMagicC18AQ3resinusinga2‐35%bufferBat300nl/min(bufferA:2%acetonitrile,0.1%formicacid;bufferB:98%acetonitrile,0.1%formicacid).Forshotgunexperiments,themassspectrometerwasoperatedwitha“top20”method,witha500‐mssurveyscanfollowedbyamaximumof20MS/MSeventsof150mseach.TheMS/MSselectionwasset forprecursors exceeding 200 counts per second and charge states greater than 2. Theselected precursors were then added to a dynamic exclusion list for 20 s. Ions wereisolatedusingaquadrupole resolutionof0.7amuand fragmented in the collisioncellusingthecollisionenergyequation

0.0625 ∗ 3.5

withacollisionenergyspreadof15eV.ForSWATH‐MSacquisition,a100‐mssurveyscanwasfollowedbyaseriesof32consecutiveMS/MSeventsof100mseachwith25amuprecursorisolationwith1amuoverlap.Thesequentialprecursorisolationwindowset‐upwasasfollows:400–425,424–450,449–475,…,1174–1200m/z.Thecollisionenergyfor eachwindowwas determined basedon the collision energy for a putative doublychargedioncenteredintherespectivewindowusingthesameequationasabovewithacollisionenergyspreadof15eV.AlltheMSdatafileswerevisuallyinspectedandcuratedatthisstageforlowtotal ionchromatogram intensities,and thecorrespondingsamplesre‐injected, ifpossible.Thisresultedinafinalsetof179SWATHdatafilesfornon‐enrichedand179SWATHdatafilesfor phospho‐enriched samples (Supplementary Data S1) that were used for dataextraction. Similarly, 40 DDA files for non‐enriched and 30 DDA files for phospho‐enrichedsampleswereselectedfordatabasesearchingandlibrarygeneration.





Page26

LC‐MS database searching

The shotgun data was searched with Sorcerer‐Sequest (TurboSequest v4.0.3rev11runningonaSage‐NSorcererv4.0.4)andMascot(version2.3.0)againsttheSGDdatabase(release03Feb.2011,containing6,750yeastproteinentries,concatenatedwith6,750corresponding “tryptic peptide pseudo‐reverse” decoy protein sequences). For thesearch,weallowedforsemi‐trypticpeptidesanduptotwomissedcleavagesperpeptide.Forthenon‐enrichedsamples,weusedcarbamidomethylationasafixedmodificationoncysteineresiduesandoxidationasvariablemodificationonmethionineresidues.Forthephospho‐enriched samples, we additionally allowed for phosphoralytion as variablemodificationonserine,threonine,andtyrosineresidues.TheSequestandMascotsearchresultswereconvertedtopep.xmlandthencombinedusingiProphet(includedinTPPversion 4.5.2) both for the non‐enriched and for the phospho‐enriched samples. Bothsearch results were filtered at 1% FDR by decoy counting at the peptide spectrummatches (PSM) level, resulting in a total of 698,652 identified spectra, 26,893 uniquepeptides,and4,310proteinsforthenon‐enrichedsampleset;inthephospho‐enrichedsample set there were a total of 224,551 identified spectra, 16,515 unique peptides(thereof14,466uniquephosphopeptides),and2,333proteins(thereof1,911phospho‐proteins).Thosedatawerecompiledintotwospectralibraries(one‘non‐enriched’andone‘phospho‐enriched’)usingSpectraST(includedinTPP4.5.2)essentiallyasdescribedearlier44, including the specific splitting of the consensus spectrawhenMS/MS scansidentifyingthesamepeptidesequencewererecordedmorethan2minutesapart,alsodescribedearlier44.Those‘splitpeptideassays’weregivendifferentproteinentrynameslabeled Subgroup_0_ProteinX to Subgroup_N_ProteinX, respectively. The fragment ioncoordinatesforthepeptidescontainedthetop6mostintense(singlyordoublycharged)yorbfragmentionsforeachspectrum,excludingthoseintheSWATHprecursorisolationwindowforthecorrespondingpeptide.Thenon‐enrichedassaylibrarycomprisedassaysfor19,473peptides(thereof18,074proteotypicpeptidesmatchingatotalof3,119uniqueproteins). The phospho‐enriched assay library comprised assays for 14,339 peptides(thereof12,969phosphopeptidesequences)orassays for13,786proteotypicpeptides(thereof 12,678 proteotypic phosphopeptides, matching a total of 1,676 uniquephosphoproteins).The SWATH‐MSdata extractionwasperformedusing the iPortalworkflowmanager45callingOpenSWATH(openMSv.1.10)46andpyProphet47.Theprecursorswerethenre‐aligned across runsusingTRIC 48. The two resulting SWATH identification result filescontained a total of 18,273 identified peptides (thereof 16,922 proteotypic peptidesmatching a total of 2,940 proteins) for the non‐enriched datasets; in the phospho‐





Page27

enriched datasets there were 13,748 identified peptides (thereof 12,412phosphopeptides) or 13,218 proteotypic peptides (thereof 12,139 proteotypicphosphopeptidesmatchingatotalof2,247 uniquephosphoproteins).Afteralignment,weusedasetofin‐housescriptstocomparethechromatographicelutionprofilesofthevarious isobaric phosphopeptide isoformsmatching a same delocalized peptide form(peptide sequence + number of phosphorylations) within each single run and toeventuallygroupthoseco‐elutingphosphopeptideassaysintothepropercorrespondingnumber of phospho‐peak clusters (labeled _cluster0 to _clusterN, respectively). Thephospho‐peakclusterswerethenconsistentlyre‐numberedacrossrunsandthosewereused as input tomapDIA49 to select for the best suitable transitions and peptides forquantification.Thisresultedinthefinalpeptideandproteinquantificationmatricesforthenon‐enrichedandphospho‐enricheddatasetsthatwereusedforfurtherprocessing.

Preprocessing of proteome data

First,featuresdetectedafter7000sandthosecorrespondingtodecoys,reverseproteins,or not unique peptides were removed. We also removed fragments with oxidizedmethionineandtheircorrespondingnon‐oxidizedfragments.Next,nativeretentiontimeswere converted to iRTs51. Fragments corresponding to peptides,whose sequencewasexistingonly inthereferenceproteome(i.e.BY1416background)andnot inRM11‐1abackgroundwere excluded from subsequent analysis. Normalization of the fragment‐level data and aggregation into peptides and protein‐level data was performed usingmapDIA49withthefollowingoptions:NORMALIZATION = RT 10, MIN_CORREL = 0.3, MIN_FRAG_PER_PEP = 2, MIN_PEP_PER_PROT = 1andamaximumof20%missingdataforeachfragment.Finally,abundancedatawerefurthercorrectedforeffects due to culture batches and proteomics measurement batches using the non‐parametricempiricalBayesframeworkComBat42.

Preprocessing of phospho‐proteome data

Firstfragmentsdetectedafter6000sec,correspondingtonon‐phosphorylatedpeptides(0P), corresponding to non‐unique proteins, and oxidizedwere removed. As for non‐phosphorylated proteomics data, native retention times were replaced by iRT51 andfragments corresponding to peptides, which sequence was not existing in RM11‐1abackgroundwereexcludedfromsubsequentanalysis.Tonormalizethe fragment leveldata and to aggregate them at the phosphopeptide cluster level we used mapDIA49.Fragments derived from phosphopeptides belonging to the same phosphopeptidescluster were aggregated together. The following mapDIA options were used:NORMALIZATION = RT 10, MIN_CORREL = 0.3, MIN_FRAG_PER_PEP = 2





Page28

andamaximumproportionofmissingdataof 20% for each fragmentas for thenon‐phosphoproteomicsdata.Datawerecorrectedforculturebatchandwellasphospho‐proteomicsmeasurementsbatcheffectsusingComBat42.

Derivation of the post‐transcriptional traits (protein abundance regressed to RNA

levels) and phosphorylation levels (phosphopeptide abundance regressed to protein

abundance)

Inorder(i)toseparatethechangesinproteinabundanceduetoRNAchangesfromthepost‐transcriptionalspecificregulation,and(ii)todistinguishchangesinphosphopeptideabundanceduetoproteinabundancechangesfrommodificationofthephosphorylationlevels,wehavegeneratedregressedtraits(asalreadyproposedin8).Thesameprocedurehas beenapplied forboth traits andwill bedetailedbelowusing thephosphopeptideabundanceexample.First,foreachpairofcorrespondingtraits(i.e.aphosphopeptideanditsproteinoforigin)relativeabundancesacrosssamplewerenormalizedusingamodifiedtransformationtostandardscore(i.e.centeringandscaling).Tocomputemeanandstandarddeviationforthisnormalization,onlythevaluescorrespondingtothesamples,inwhichmeasurementswere available for both phosphopeptides and proteinwere used. Then, for each pair,normalizedphosphopeptidedatawereregressedonproteindatausingarobustlinearregressionusing aMM‐estimate52, initializedbyan S‐estimateusingHubber’sweightsfunctionandusinganM‐estimatorasfinalestimateusingaTukey'sbiweightfunctionasimplementedMM‐estimation option in the rlm function of theMASSRpackage53. Theresidualsoftheseregressionwerethenusedtoastraitinthesubsequencesanalyses.

QTL mapping

WeemployedapreviouslydevelopedQTLdetectionmethodbasedonRandomForest(RF),thatwasfurthermodifiedtoimproveitsperformance18,21.Tocorrectforpopulationsubstructureweincludepopulationstructureasacovariateinthemodel, as we previously proposed21, but in a slightly differentmanner. First, thegenotypematrixwasnormalizedasdescribedin54(seeequations(1)to(3)there).Thenwe carried out a singular value decomposition on the normalized genotype matrixfollowed by an eigenvector decomposition. We then selected those eigenvectorscorrespondingtothetopseveneigenvaluesascovariatesfortheQTLmapping(additionalpredictorsforgrowingtheRandomForests).Thesefirstsevenvectorsexplainedmorethan25%ofthegenotypevariance(SupplementaryDataS14).





Page29

We adapted the approach described in 18 by using a different score describing theimportanceofeachpredictorintheRandomForest(variableimportancemeasure,VIM).WhilepreviousworkreliedontheselectionfrequencyastheVIM,i.e.thenumberoftimesa predictor was used in a forest, to quantify the importance of each predictor, wecombined two previously published VIMs, RSS and PI55, to compute an informativecombined score servingas theVIM inour study.While theRSSdescribes the averagereductioninthesumofthesquaredresidualsaftersplittinganodewithapredictor,thePIreferstotherelativereductioninpredictiveaccuracyafterpermutingtheinformationfortherespectivepredictor.RSS andPI are combined in the followingway to generate amore robust scoreSi forpredictori:

max 0, ∗ max 0, WeaveragedreplicatesforthesamestraintoavoidthedetectionoffalsepositiveQTL.Consecutivemarkerslinkedtothesametraitand/ormarkersinhighLD(Pearson’sr>0.8)linkedtothesametraitwerecountedasasinglelinkageaswepreviouslydescribed18.

QTL hotspot detection

Toformallyidentifyhotspotsinourdataset,thegenomewasdividedinto40‐kbbins(293bins, bins at chromosome extremities could be bigger). As proposed before15, if thelinkageswererandomlydistributedacrossthegenome,thenumberoflinkagesineachbinwouldbeexpectedtofollowaPoissondistributionwiththemeanofΝ Ν⁄ .We use this distribution to estimate the highest number of linkages that a bin couldcontainataprobabilitylowerthan0.01.Thesenumberswere30foreQTL,13forpQTL,9forptQTL,11forphQTL,and4forphResQTL.BinswithasufficientnumberofQTLsatagivenmolecularlayerwereconsideredasQTL‐hotspots.ConsecutivebinsonthesamechromosomewithenoughQTLswerecombinedtosinglehotspots.

Integration of growth QTL

QTLaffectinggrowthundervariousconditionsweretakenfrom20.IfthereportedpeakpositionofthegrowthQTLwaswithin50kbofthemiddleofthepositionsofthelocithataffectedthemosttraitsoneachofthemolecular layers,weconsideredthemtobethesameQTL.

Identification of local QTL

QTLswereconsideredtobelocaltotheirtargetiftheQTLcontainedatleastonegeneticmarkerthathadacorrelationof0.8orhigherwithoneofthemarkersthataredirectlyup‐ordownstreamofthetargetgene(Pearson’sr).ForeQTL,thetargetgenecorresponds





Page30

to the gene, whose expression is investigated; for pQTL it corresponds to the geneencoding the protein, whose abundance is studied; for phQTL and phResQTL itcorrespondstothegeneencodingtheproteinofwhichthephosphopeptideispartof.

GO enrichment

GeneOntology(GO)enrichmentanalyseswereperformedusingtopGO.WeperformedFisher’sexacttestswiththe“weight01”algorithmandaminimalnodesizeof1056.WeusedannotationsfromSacCer3.

Broad‐sense heritability estimation

Broad‐senseheritabilityestimateswerecomputedbasedonreplicatemeasurementsofsome strains, as described elsewhere20. We used six different segregants with threereplicateseach,aswellastheparentswithsix(BY)andeight(RM)replicateseach. Inshort, the 'lmer' function fromthe lme4Rpackagewasusedtocreatea linearmixed‐effectsmodelwiththephenotypeas theresponseandthesegregant labelsas randomeffects57.Thevariancecomponentsσ²G(thevarianceduetogeneticeffects,i.e.,differentsegregants),andσ2E,(theerrorvariance),wereextracted,andbroad‐senseheritabilitywascalculatedasH²=σ²G/(σ²G+σ²E).Standarderrorswerecalculatedusingthedelete‐onejackknifeprocedure,asproposedpreviously20.Randomdistributionsofheritabilityestimates foreachmolecular layerweregeneratedbypermuting thestrain labelsandcompared to the real distributions using aMann‐Whitney test. In order to be able tocompareheritabilityestimatesbetweenthedifferentmolecularlevels,werestrictedtheanalysistothesetof402proteinswherewehadmeasurementsforexpression,protein,and(atleastone)phosphopeptide.

Polymorphisms in and around genes

ToinvestigatethecauseoflocalQTL,wecountedpolymorphismsintheupstreamregion,downstreamregion,3’and5’UTRs,codingsequenceandaminoacidsequenceofeachcodinggenebetweentheBYandRMgenomes.WeconsideredallSNPsreportedbyBloomet al. 20.We excluded all geneswith insertions and deletions from this analysis. UTR‐annotationsweredownloadedfromwww.yeastgenome.orginOctober2017.IftheUTRwasreportedmultipletimeswithdifferinglengths,weusedthelargestannotation.Theup‐anddownstreamregionsofagenespanned2kbeachandbeganattheouterbordersoftheUTRsrelativetothegeneofinterest.ThenumberofcodingpolymorphismswasdefinedasthenumberofaminoacidchangesinaproteinbetweenBYandRM.MultipleSNPsinthesamecodonwereonlycountedonce.





Page31

Figure3cshowsonlygeneswithavailableUTRannotationsandavailablemeasurementsontheproteinlevel.Geneswithindelsarenotincluded.Foreachphosphositeweidentifiedtheclosestpolymorphismintheaminoacidsequencespacebycomputingtheabsolutedifferenceofthepositionofthemodifiedserineandallpolymorphicaminoacidswithintheproteinandusingtheminimumofthosedistances.Note that phosphopeptides were only included in this study if they didn’t have anypolymorphisms.

Annotation of protein complexes

We used previously published annotations for protein complexes(ftp://ftp.ebi.ac.uk/pub/databases/intact/complex/current/complextab/saccharomyces_cerevisiae.tsv, 58). The annotated complex members included proteins and smallmoleculessuchasions.Weonlyconsideredthoseproteinsasmembersofcomplexesthatsharedacomplexwithatleastoneotherprotein.Toinvestigatetheinfluenceofthestoichiometryofproteincomplexesonproteinlevels,wecomputedthecorrelationbetweenlevelsofproteinsofthesamecomplex.Hereweonlyconsideredgenes(i)forwhichproteinandtranscriptlevelswereavailable,and(ii)whichwereannotatedforcomplexesthatdidnothaveanyoverlapwithothercomplexes.Nucleolar and ribosomal proteinswere excluded from this analysis, sincewe alreadyshowedthattheywereenrichedingeneswithbufferedproteinlevels(‘buffered’inFigure3b).Takingthemintoaccountmighthavebiasedtheresultsfromtheanalysisofcomplexproteinpairs. In totalwe considered286uniquepairs of proteins that each shared acomplex.Thesepairscorrespondedto188genes.WethenusedFisher’sexactteststodetermineifpairsofgenesthatshareacomplexwereaffectedbythesameQTLatasignificantlydifferentfrequencythanpairsofgenesthatdonotshareacomplex.TraitswereconsideredtobeaffectedbythesameQTLifanymarkerwassignificantforbothtraits.

Classification of QTL based on their effects on the transcriptome and proteome

ToanalyzehowQTLaffecttheproteome,weclassifiedeQTLintothreedistinctclasses:(i)eQTLwith ‘similar’effectsonthe transcriptomeandproteome,(ii) ‘buffered’eQTL,and(iii)‘enhanced’eQTL.WecomputedtheeffectEofaQTLatlocusionamoleculartraitjasalog2foldchangebysubtractingtheaverageofthelog2‐transformedtraitvaluestofallstrainswiththeRM‐alleleatthislocusfromtheaveragetraitvalueforallstrainswithBY‐allele.

̅ , ̅ ,





Page32

TheeQTLswereclassifiedbasedonthedifferenceoftheireffectsonthetargettranscriptandtheencodedprotein:

∆ | | where and aretheeffectsofeQTLionthemRNAandproteinlevels,respectively.Iftheabsolutedifference|∆ |wasbelow0.15theeQTLeffectwasclassifiedas‘similar’.If∆ was above +0.15, itwas classified as ‘buffered’ and if the effect differencewasbelow‐0.15(i.e.∆ 0.15)itwasclassifiedas‘enhanced’.ptQTLatFDR


Page33

References 1. EmilssonV,ThorleifssonG,ZhangB,LeonardsonAS,ZinkF,ZhuJ,CarlsonS,Helgason

A, Walters GB, Gunnarsdottir S, Mouy M, Steinthorsdottir V, Eiriksdottir GH,Bjornsdottir G, Reynisdottir I, Gudbjartsson D, Helgadottir A, Jonasdottir A,JonasdottirA,StyrkarsdottirU,GretarsdottirS,MagnussonKP,StefanssonH,FossdalR,KristjanssonK,GislasonHG,StefanssonT,LeifssonBG,ThorsteinsdottirU,LambJR, Gulcher JR, Reitman ML, Kong A, Schadt EE, Stefansson K. Genetics of geneexpressionanditseffectondisease.Nature(2008).

2. vanderSijde,M.R.,Ng,A.&Fu,J.Systemsgenetics:FromGWAStodiseasepathways.Biochim.Biophys.ActaBBA‐Mol.BasisDis.1842,1903–1909(2014).

3. Bali V, Bebok Z. Decoding mechanisms by which silent codon changes influenceproteinbiogenesisandfunction.Int.J.Biochem.CellBiol.(2015).

4. HenchozS,ChiY,CatarinB,HerskowitzI,DeshaiesRJ,PeterM.Phosphorylation‐andubiquitin‐dependentdegradationofthecyclin‐dependentkinaseinhibitorFar1pinbuddingyeast.GenesDev.(1997).

5. Ardito,F.,Giuliani,M.,Perrone,D.,Troiano,G.&Muzio,L.L.Thecrucialroleofproteinphosphorylationincellsignalinganditsuseastargetedtherapy(Review).Int.J.Mol.Med.40,271–280(2017).

6. Miller,M.E.&Cross,F.R.MechanismsControllingSubcellularLocalizationoftheG1CyclinsCln2pandCln3pinBuddingYeast.Mol.Cell.Biol.21,6292–6311(2001).

7. AlbertFW,BloomJS,SiegelJ,DayL,KruglyakL.Geneticsoftrans‐regulatoryvariationingeneexpression.eLife(2018).

8. Foss, E. J. et al. Genetic Variation Shapes Protein Networks Mainly through Non‐transcriptionalMechanisms.PLoSBiol.9,e1001144(2011).

9. Holdt, L.M. etal. QuantitativeTrait LociMapping of theMousePlasmaProteome(pQTL).Genetics193,601–608(2013).

10. Wu,L.etal.Variationandgeneticcontrolofproteinabundanceinhumans.Nature499,79–82(2013).

11. Picotti,P.etal.Acompletemass‐spectrometricmapoftheyeastproteomeappliedtoquantitativetraitanalysis.Nature494,266–270(2013).

12. Gowda GA, Djukovic D. Overview of mass spectrometry‐based metabolomics:opportunitiesandchallenges.MethodsMol.Biol.CliftonNJ(2015).

13. Chick,J.M.etal.Definingtheconsequencesofgeneticvariationonaproteome‐widescale.Nature534,500–505(2016).

14. Civelek,M.&Lusis,A.J.Systemsgeneticsapproachestounderstandcomplextraits.Nat.Rev.Genet.15,34–48(2014).

15. Brem,R., Yvert,G., Clinton,R.&Kruglyak,L.GeneticDissectionofTranscriptionalRegulationin.Nature411,41(2002).

16. Foss,E.J.etal.Geneticbasisofproteomevariationinyeast.Nat.Genet.39,1369–1375(2007).

17. Albert,F.W.,Treusch,S.,Shockley,A.H.,Bloom,J.S.&Kruglyak,L.Geneticsofsingle‐cell protein abundancevariation in large yeastpopulations.Nature506, 494–497(2014).

18. Clement‐Ziza,M.etal.Naturalgeneticvariationimpactsexpressionlevelsofcoding,non‐coding, and antisense transcripts in fissionyeast.Mol.Syst.Biol.10, 764–764(2014).





Page34

19. Brem RB, Kruglyak L. The landscape of genetic complexity across 5,700 geneexpressiontraitsinyeast.Proc.Natl.Acad.Sci.U.S.A.(2005).

20. Bloom,J.S.,Ehrenreich,I.M.,Loo,W.T.,Lite,T.‐L.V.&Kruglyak,L.Findingthesourcesofmissingheritabilityinayeastcross.Nature494,234–237(2013).

21. Michaelson,J.J.,Alberts,R.,Schughart,K.&Beyer,A.Data‐drivenassessmentofeQTLmappingmethods.BMCGenomics11,502(2010).

22. Smith,E.N.&Kruglyak,L.Gene–EnvironmentInteractioninYeastGeneExpression.PLoSBiol.6,e83(2008).

23. Yvert,G.etal.Trans‐actingregulatoryvariationinSaccharomycescerevisiaeandtheroleoftranscriptionfactors.Nat.Genet.35,57–64(2003).

24. Tsay,Y.F.,Thompson,J.R.,Rotenberg,M.O.,Larkin,J.C.&Woolford,J.L.Ribosomalprotein synthesis is not regulated at the translational level in Saccharomycescerevisiae:balancedaccumulationofribosomalproteinsL16andrp59ismediatedbyturnoverofexcessprotein.GenesDev.(1988).doi:10.1101/gad.2.6.664

25. Liu Y, Mi Y, Mueller T, Kreibich S,Williams EG, Van Drogen A, Borel C, FrankM,GermainPL,BludauI,MehnertM,SeifertM,EmmenlauerM,SorgI,BezrukovF,BenaFS,ZhouH,DehioC,TestaG,Saez‐RodriguezJ,AntonarakisSE,HardtWD,AebersoldR.Multi‐omicmeasurementsofheterogeneityinHeLacellsacrosslaboratories.Nat.Biotechnol.(2019).

26. LiuY,BorelC,LiL,MüllerT,WilliamsEG,GermainPL,BuljanM,SajicT,BoersemaPJ,ShaoW,FainiM,TestaG,BeyerA,AntonarakisSE,AebersoldR.SystematicproteomeandproteostasisprofilinginhumanTrisomy21fibroblastcells.Nat.Commun.(2018).

27. JüschkeC,DohnalI,PichlerP,HarzerH,SwartR,AmmererG,MechtlerK,KnoblichJA.Transcriptomeandproteomequantificationofatumormodelprovidesnovelinsightsintopost‐transcriptionalgeneregulation.GenomeBiol.(2014).

28. Garcia‐Barrio M, Dong J, Ufano S, Hinnebusch AG. Association of GCN1‐GCN20regulatorycomplexwith theN‐terminusofeIF2alphakinaseGCN2 is required forGCN2activation.EMBOJ.(2000).

29. Ben‐Levy R, Leighton IA, Doza YN, Attwood P, Morrice N, Marshall CJ, Cohen P.Identification of novel phosphorylation sites required for activation of MAPKAPkinase‐2.EMBOJ.(1996).

30. Tatchell,K.,Robinson,L.C.&Breitenbach,M.RAS2ofSaccharomycescerevisiae isrequiredforgluconeogenicgrowthandproperresponsetonutrientlimitation.Proc.Natl.Acad.Sci.82,3785–3789(1985).

31. Jian, D., Aili, Z., Xiaojia, B., Huansheng, Z. & Yun, H. Feedback regulation of Ras2guanine nucleotide exchange factor (Ras2‐GEF) activity of Cdc25p by Cdc25pphosphorylation in the yeast Saccharomyces cerevisiae. FEBS Lett. (2010).doi:10.1016/j.febslet.2010.11.006

32. VerdièreJ,CreusotF,GuérineauM.Regulationoftheexpressionofiso2‐cytochromecgeneinS.cerevisiae:cloningofthepositiveregulatorygeneCYP1andidentificationoftheregionofitstargetsequenceonthestructuralgeneCYP3.Mol.Gen.Genet.MGG(1985).

33. RutterJ,ProbstBL,McKnightSL.CoordinateregulationofsugarfluxandtranslationbyPASkinase.Cell(2002).

34. BodenmillerB,WankaS,KraftC,UrbanJ,CampbellD,PedrioliPG,GerritsB,PicottiP,Lam H, Vitek O, Brusniak MY, Roschitzki B, Zhang C, Shokat KM, Schlapbach R,Colman‐Lerner A, Nolan GP, Nesvizhskii AI, Peter M, Loewith R, von Mering C,





Page35

Aebersold R. Phosphoproteomic analysis reveals interconnected system‐wideresponsestoperturbationsofkinasesandphosphatasesinyeast.Sci.Signal.(2010).

35. Klipp, E. & Liebermeister, W. Mathematical modeling of intracellular signalingpathways.BMCNeurosci.7,S10(2006).

36. Mas, G. etal. Recruitment of a chromatin remodelling complex by theHog1MAPkinasetostressgenes.EMBOJ.28,326–336(2009).

37. Bilsland‐Marchesan,E.,Ariño,J.,Saito,H.,Sunnerhagen,P.&Posas,F.Rck2kinaseisasubstratefortheosmoticstress‐activatedmitogen‐activatedproteinkinaseHog1.Mol.Cell.Biol.20,3887–3895(2000).

38. KimD, PerteaG, Trapnell C, PimentelH, Kelley R, Salzberg SL. TopHat2: accuratealignmentoftranscriptomesinthepresenceofinsertions,deletionsandgenefusions.GenomeBiol.(2014).

39. VanderAuweraGA,CarneiroMO,HartlC,PoplinR,DelAngelG,Levy‐MoonshineA,JordanT,ShakirK,RoazenD,ThibaultJ,BanksE,GarimellaKV,AltshulerD,GabrielS,DePristoMA.FromFastQdatatohighconfidencevariantcalls:theGenomeAnalysisToolkitbestpracticespipeline.Curr.Protoc.Bioinforma.(2016).

40. DobinA,DavisCA,SchlesingerF,Drenkow J,ZaleskiC, JhaS,BatutP,ChaissonM,GingerasTR.STAR:ultrafastuniversalRNA‐seqaligner.Bioinforma.Oxf.Engl.(2012).

41. LoveMI,HuberW,AndersS.ModeratedestimationoffoldchangeanddispersionforRNA‐seqdatawithDESeq2.GenomeBiol.(2015).

42. JohnsonWE,LiC,RabinovicA.AdjustingbatcheffectsinmicroarrayexpressiondatausingempiricalBayesmethods.Biostat.Oxf.Engl.(2007).

43. SelevsekN,ChangCY,GilletLC,NavarroP,BernhardtOM,ReiterL,ChengLY,VitekO,Aebersold R. Reproducible and consistent quantification of the Saccharomycescerevisiae proteome by SWATH‐mass spectrometry. Mol. Cell. Proteomics MCP(2015).

44. Schubert OT, Gillet LC, Collins BC, Navarro P, Rosenberger G,WolskiWE, LamH,AmodeiD,MallickP,MacLeanB,AebersoldR.Buildinghigh‐qualityassaylibrariesfortargetedanalysisofSWATHMSdata.Nat.Protoc.(2015).

45. Kunszt, P. etal. iPortal: the swiss grid proteomics portal: Requirements and newfeaturesbasedonexperienceandusabilityconsiderations.Concurr.Comput.Pract.Exp.27,433–445(2015).

46. RöstHL,RosenbergerG,NavarroP,GilletL,MiladinovićSM,SchubertOT,WolskiW,CollinsBC,MalmströmJ,MalmströmL,AebersoldR.OpenSWATHenablesautomated,targetedanalysisofdata‐independentacquisitionMSdata.Nat.Biotechnol.(2014).

47. TelemanJ,RöstHL,RosenbergerG,SchmittU,MalmströmL,MalmströmJ,LevanderF.DIANA–algorithmicimprovementsforanalysisofdata‐independentacquisitionMSdata.Bioinforma.Oxf.Engl.(2015).

48. RöstHL,LiuY,D’AgostinoG,ZanellaM,NavarroP,RosenbergerG,CollinsBC,GilletL,Testa G, Malmström L, Aebersold R. TRIC: an automated alignment strategy forreproducibleproteinquantificationintargetedproteomics.Nat.Methods(2017).

49. Teo G, Kim S, Tsou CC, Collins B, Gingras AC, Nesvizhskii AI, Choi H. mapDIA:Preprocessing and statistical analysis of quantitative proteomics data from dataindependentacquisitionmassspectrometry.J.Proteomics(2015).

50. VizcaínoJA,CôtéRG,CsordasA,DianesJA,FabregatA,FosterJM,GrissJ,AlpiE,BirimM,ContellJ,O’KellyG,SchoeneggerA,OvelleiroD,Pérez‐RiverolY,ReisingerF,Ríos





Page36

D, Wang R, Hermjakob H. The PRoteomics IDEntifications (PRIDE) database andassociatedtools:statusin2013.NucleicAcidsRes.(2012).

51. EscherC,ReiterL,MacLeanB,OssolaR,HerzogF,ChiltonJ,MacCossMJ,RinnerO.UsingiRT,anormalizedretentiontimeformoretargetedmeasurementofpeptides.Proteomics(2012).

52. Yohai, V. J. High Breakdown‐Point and High Efficiency Robust Estimates forRegression.Ann.Stat.15,642–656(1987).

53. Venables,W.N.&Ripley,B.D.ModernAppliedStatisticswithS.(Springer,2002).54. PattersonN,PriceAL,ReichD.Populationstructureandeigenanalysis.PLoSGenet.

(2007).55. Liaw,A.&Wiener,M.ClassificationandregressionbyrandomForest.RNews2,18–

22(2002).56. Alexa,A.,Rahnenfuhrer,J.&Lengauer,T.Improvedscoringoffunctionalgroupsfrom

geneexpressiondatabydecorrelatingGOgraphstructure.Bioinformatics22,1600–1607(2006).

57. Bates, D.,Mächler,M., Bolker, B. &Walker, S. Fitting LinearMixed‐EffectsModelsUsinglme4.J.Stat.Softw.67,1–48(2015).

58. Meldal BH, Forner‐Martinez O, Costanzo MC, Dana J, Demeter J, Dumousseau M,DwightSS,GaultonA,LicataL,MelidoniAN,Ricard‐BlumS,RoechertB,SkyzypekMS,TiwariM, Velankar S,Wong ED,HermjakobH, Orchard S. The complex portal–anencyclopaediaofmacromolecularcomplexes.NucleicAcidsRes.(2015).

59. RCoreTeam.R:ALanguageandEnvironmentforStatisticalComputing.(RFoundationforStatisticalComputing,2013).



integration of transcriptome, proteome and …ruedi aebersold# affiliation: department of biology,...

Documents