inhibition of fat cell differentiation in 3t3 l1 pre adipocytes by all trans retinoic acid integrative analysis of transcriptomic and phenotypic data

Pfaffla,∗ a Animal Physiology and Immunology, School of Life Sciences Weihenstephan, Technical University of Munich, Freising, Germany b Institute of Computational Biology, Helmholtz Cent

jou rn al h om e p a g e :w w w e l s e v i e r c o m / l o c a t e/ b d q

Research paper

Inhibition of fat cell differentiation in 3T3-L1 pre-adipocytes by

all-trans retinoic acid: Integrative analysis of transcriptomic and

phenotypic data

Katharina Stoeckera, Steffen Sassb, Fabian J Theisb,c, Hans Haunerd,e, Michael W Pfaffla,∗

a Animal Physiology and Immunology, School of Life Sciences Weihenstephan, Technical University of Munich, Freising, Germany

b Institute of Computational Biology, Helmholtz Center Munich, German Research Center for Environmental Health, Oberschleißheim, Germany

c Department of Mathematics, Technical University of Munich, Garching, Germany

d Else-Kröner-Fresenius-Centre for Nutritional Medicine, ZIEL Research Center for Nutrition and Food Sciences, Technical University of Munich, Freising,

Germany

e Else Kröner-Fresenius-Centre for Nutritional Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany

a r t i c l e i n f o

Article history:

Received 8 September 2016

Received in revised form 8 November 2016

Accepted 15 November 2016

Available online xxx

Handled by Justin O’Grady

Keywords:

All-trans retinoic acid (ATRA)

Adipogenesis

3T3-L1 cells

MicroRNA–mRNA interaction

Regulation of biological pathways

Multiple linear regression models

a b s t r a c t

Theprocessofadipogenesisiscontrolledinahighlyorchestratedmanner,includingtranscriptionaland post-transcriptionalevents.Indeveloping3T3-L1pre-adipocytes,thisprogramcanbeinterruptedby all-transretinoicacid(ATRA).ToexaminethisinhibitingimpactbyATRA,wegeneratedlarge-scale tran-scriptomicdataonthemicroRNAandmRNAlevel.Non-codingRNAssuchasmicroRNAsrepresentafield

inRNAturnover,whichisveryimportantforunderstandingtheregulationofmRNAgeneexpression HighthroughputmRNAandmicroRNAexpressionprofilingwasperformedusingmRNAhybridisation microarraytechnologyandmultiplexedexpressionassayformicroRNAquantification.Afterquantitative measurementswemergedexpressiondatasets,integratedtheresultsandanalysedthemolecular regu-lationofinvitroadipogenesis.Forthispurpose,weappliedlocalenrichmentanalysisontheintegrative microRNA-mRNAnetworkdeterminedbyalinearregressionapproach.Thisapproachincludesthetarget predictionsofTargetScanMouse5.2and23pre-selected,significantlyregulatedmicroRNAsaswellas AffymetrixmicroarraymRNAdata.Wefoundthatthecellularlipidmetabolismisnegativelyaffected

byATRA.Furthermore,wewereabletoshowthatmicroRNA27aand/ormicroRNA96areimportant regulatorsofgapjunctionsignalling,therearrangementoftheactincytoskeletonaswellasthecitric acidcycle,whichrepresentthemostaffectedpathwayswithregardtoinhibitoryeffectsofATRAin 3T3-L1preadipocytes.Inconclusion,theexperimentalworkflowandtheintegrativemicroRNA–mRNAdata analysisshowninthisstudyrepresentapossibilityforillustratinginteractionsinhighlyorchestrated biologicalprocesses.FurthertheappliedglobalmicroRNA–mRNAinteractionnetworkmayalsobeused forthepre-selectionofpotentialnewbiomarkerswithregardtoobesityorfortheidentificationofnew pharmaceuticaltargets

©2016TheAuthor(s).PublishedbyElsevierGmbH.ThisisanopenaccessarticleundertheCCBY

license(http://creativecommons.org/licenses/by/4.0/)

1 Introduction

Dietary behaviour has a strong influence on body energy

metabolism.Whiteadiposetissue(WAT)isthemajorstoragedepot

forexcessenergyandisinvolvedinenergyhomeostasis.WATis

∗ Corresponding author at: Animal Physiology and Immunology, School of Life

Sciences Weihenstephan, Technical University of Munich, Weihenstephaner Berg 3,

Freising 85354, Germany.

E-mail address: michael.pfaffl@wzw.tum.de (M.W Pfaffl).

mainlycomposedofmatureadipocytesandpre-adipocytesaswell

asothercelltypes.Anexpansionofadiposetissueisusuallybecause

ofanincreaseinfatcellvolume,butinthecaseofsevereformsof obesity,itisalsocharacterisedbyfatcellhyperplasiacausedby preadipocytedifferentiation.Thisprocessisregulatedinahighly complexmannerandincludesmanyregulatorycomponents[1] However,theenlargementofbodyfatmassandanincreaseinbody massindex(BMI)dependsonachronicpositiveenergybalance[2] Adipogenesis, the process of fat cell differentiation can be affectedbyvariousendogenousand/orexogenousfactors Accord-ingly,nutrientsorothercompoundsaffectingadipogenesisexert

http://dx.doi.org/10.1016/j.bdq.2016.11.001

2214-7535/© 2016 The Author(s) Published by Elsevier GmbH This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ).

sizeandnumber,andthesecretionofhormonesregulatinghunger,

satietyor energyexpenditure [3].Fructose,nicotine,pesticides,

industrial chemicals or pharmaceuticals like estradiol are able

tomediatepro-adipogenicfunctionsinmammalsandarecalled

obesogeniccompounds[3,4].Ontheotherhand,anti-adipogenic

compoundsalsoexistinnature.All-transretinoicacid(ATRA),a

vitaminAderivate,iscapableofmodulatingdifferentiation[5–7]

Thereby,thedifferentiationofosteoblasts[8]andmyoblasts [9]

isinduced,whereas thedevelopmentof adipocytesis inhibited

[10–12].Theeffects of ATRAstrongly dependonits

concentra-tion, duration of action and target cell type [13] Retinol and

carotenoids,the precursorsof retinoic acid, are foundin many

foods,e.g.meat,milkandvegetables,andinhighconcentrations

incarrots.Afterabsorptionfromthediet, retinoicacidis

trans-portedtovarioustissuesincluding,WAT,andismetabolisedinside

thecells[5,7].TheeffectsofATRAonadipogenesishavebe

exam-inedincellculturemodelse.g.themouse3T3-L1pre-adipocyte

cellline[14].Inconfluentgrowth-arrested3T3-L1pre-adipocytes,

anadipogenicstimulusinitiatesalimitedclonalexpansionbefore

the cells enter the phase of terminal differentiation, including

lipidaccumulation[15–17].Forthis,thesequentialinductionof

thetranscriptionfactorsCebp˛,Cebpˇ[18]andPpar [19–21]is

necessary.Thereafter, adipocytespecificgenesandenzymesare

induced[22],andcellmorphologyischangedbyreorganisation

ofthecytoskeletonfroma fibroblast-likeappearancetowardsa

spherical shape [22,23] Recent data suggest that the

develop-mentofmatureadipocytesisinitiatedbytranscriptionalaswellas

post-transcriptionalchangesconveyedbysmallnon-codingRNAs

MicroRNAs(miRs)aresmallnon-codingRNAspecies,and

regula-tiontakesplacethroughboth,mRNAdegradationorsuppression

oftranslation[24,25].Thismechanismwasinitiallydiscoveredin

plantsandCaenorhabditiselegansbyLeeetal.andwasreportedto

beassociatedwiththeregulationofgeneexpressionandcell

differ-entiationandwithguardingorganismsagainstexternalnucleotide

sequencessuchasviruses,transposonsorparasites[26]

Thepresentstudyaimedtofurtherstudythemolecular

mech-anismsinducedbyATRA in3T3-L1mousepreadipocytesatthe

transcriptome and phenotypic levels in a high-resolution time

frame.Therefore,we usedthe methodofoil-red-Ostaining for

phenotyping and microarray technology, reverse-transcription

quantitativepolymerase chain reaction(RT-qPCR) and a

multi-plexedassayforexpressionprofilingatboththemRNAandmiR

levels.Pre-adipocytesweretreatedwithATRAinatimeframefrom

0hto288h(12days)post-treatment.In addition,togenerate a

comprehensivepictureofregulatorynetworksofthe

physiologi-calprocesses,wecombinedthesehigh-throughputtranscriptional

datasetstocreateatwo-levelregulatorymRNA-miRnetworkof

transcriptomicdata

2 Materials and methods

2.1 Cellcultureformaintenance

For all experiments, the mouse preadipocyte cell line

3T3-L1 (ATCC®/LGC Standards GmbH, Wesel, Germany) was used,

and cells werecultured as described by thesupplier Thecells

weremaintainedinT175flasks(NalgeneNuncInternational/Fisher

Scientific, Schwerte, Germany) withfibroblastmedium

consist-ingof500mlDulbecco’smodifiedEagle´ısmedium(DMEM)(LGC

Standards GmbH, Wesel, Germany), 10% new born calf serum

(PAN Biotech GmbH, Aidenbach, Germany) and 1%

penicillin-streptomycin (InvitrogenTM, Darmstadt, Germany) Cells were

subculturedevery3days.Followingthis, thecells werewashed

withphosphate-bufferedsaline (PBS) (PAA LaboratoriesGmbH,

Cölbe, Germany), and the cell layer was detached with 0.25% Trypsin–EDTA(LifeTechnologiesGmbH,Darmstadt,Germany).Cell culturewasperformedat37◦Cinahumidified5%CO2incubator 2.2 Experimentalcellculture

Initially, cells were resuspended in fibroblast medium and 1.8×105cellswereseededpersixwells(NalgeneNunc Interna-tional/Fisher Scientific, Schwerte, Germany) After5days, when pre-confluencewasreached,thefibroblastmediumwasrenewed and the cells were cultured for another 2days The fibroblast medium was replaced by differentiation medium (DMI), con-taining 10% foetal bovine serum (FBS) (PAN Biotech GmbH, Aidenbach, Germany), 1% penicillin–streptomycin, 0.5mM 3-Isobutyl-1-methylxanthin (IBMX) (Sigma Aldrich, Taufkirchen, Germany), 1␮M dexamethasone (DEX) (Sigma Aldrich) and

1␮g/ml insulin (Sigma Aldrich) in 500ml DMEM The 3T3-L1 preadipocytes were cultivated with DMI in the presence and absenceof3␮MATRA(SigmaAldrich,Taufkirchen,Germany).After

96h, DMIwasreplacedbygrowthmedium(±3␮MATRA) con-taining10%FBS,1%penicillin–streptomycinand1␮g/mlinsulin

in500mlDMEM.Themediumwasrenewedevery2days ATRA-treatedandưuntreated3T3-L1cellswereharvestedafter0,2,4,7,

10,24,48and96handagainafter288h(12days)

2.3 EvaluationofRNAquality Cellular RNA of ATRA-treated and untreated 3T3-L1 preadipocytes was extracted using the RNeasy Kit (Qiagen, Hilden,Germany),asdescribedbythesupplier.RNAwaseluted

in RNase-free water The RNA concentration and purity were determinedusingtheSpectrophotometerNanoDrop1000 (Nano-Dropproducts,Wilmington,USA)andBioanalyzer2100(Agilent Technologies,Mannheim,Germany)

2.4 cDNAsynthesis Forgeneexpressionprofiling,500ngoftotalRNApersample wasreversetranscribedintofirst-strandcDNA.ForcDNAsynthesis, theMoloneymurineleukaemiavirusreversetranscriptaseH- (M-MLVRTH-)(Promega,Mannheim,Germany),10mMdNTPsand

50␮Mhexamerprimerswereutilisedinatotalvolumeof31␮l Thereactionmixwasincubatedfor20minat21◦C,followingwhich thecDNAsynthesisstepwasperformedfor120minat48◦C.The reactionwasstoppedby2-minincubationat90◦C

ForthereversetranscriptionofmaturemiRs,weusedthe miS-criptRTKit(Qiagen),asdescribedbythesupplier.Thetotalvolume

ofeachreactionwas20␮l.Weused300ngtotalRNA.TheRT reac-tionswereperformedbyincubationfor1hat37◦C,andthecDNA reactionwasstoppedbyincubationat95◦Cfor5min.Thereafter, cDNAwasdiluted1:4withRNase-freewaterandstoredatư20◦C.

2.5 RT-qPCRformRNAquantification RT-qPCR was performed using a CFX384 Touch real-time detection cycler (Bio-Rad Laboratories, Munich, Germany), and theSsoFastEvaGreenSupermix(Bio-RadLaboratories)wasused for gene expression profiling Gene primers (200nM) and 1␮l

of first-strand cDNA were added to the master mix From the entire set of all quantified genes the most stable mRNAs were selected for data normalisation We appliedthe geomet-rical average of the following four stably expressed reference genes:Glyceraldehyde-3-phosphatedehydrogenase(Gapdh); Non-POU-domain-containing Octamer-binding protein (Nono); Beta Actin (Actb) [27]; and Importin 8 (Ipo8) We quantified the

proliferator-activated receptor gamma (Ppar); CCAAT/enhancer-binding

protein(C/EBP),alpha(Cebp˛);CCAAT/enhancer-bindingprotein

(C/EBP),beta(Cebpˇ);sterolregulatoryelementbinding

transcrip-tionfactor 1 (Srebf1)and retinoidX receptor andalpha (Rxr˛)

All primers were synthesized by Integrated DNA Technologies

(Leuven,Belgium)andprimersequencesaresummarisedinthe

SupportinginformationTableS1

Thefollowing cyclingconditionswereusedinCFX384Touch

real-timeDetectionCycler(Bio-Rad):afterinitialactivationfor30s

at98◦C,thecyclestepsofdenaturationfor5sat95◦Cand

anneal-ing/elongationfor20sat60◦Cwererepeatedfor39cycles.Melting

curveanalysiswasappliedtoconfirmtheintegrityofgenerated

RT-qPCRproductsbyasingleampliconpeak.Themeltingcurvewas

generatedfrom65◦Cto95◦Cwithanincrementof5◦C/5min

2.6 RT-qPCRformiRquantification

FormiRexpressionprofiling,weusedthemiScriptSYBRGreen

PCRKit(Qiagen,Hilden,Germany),asdescribedbythesupplier

Thetotalvolumeofeachreactionwas10␮l,including1␮l

tem-platecDNA,1␮l10×miScriptUniversalPrimeraswellas1␮lof

thespecific10×miScriptPrimerAssay.Thefollowingconditions

wereused:initialactivation15minat95◦C,denaturation15sat

94◦C,annealing30sat55◦Candextension30sat70◦C.Afterdata

collectionattheextensionstep,meltingcurveanalysiswas

per-formedinthesamemannerasthatmentionedforgeneexpression

profiling.Fornormalisation,wecalculatedthegeometricalaverage

ofallmeasuredmiRs,includingmiR-26b,miR-29a-5p,miR-29b-5p,

miR-93,miR-96,miR-103,miR–miR-146,miR-221andmiR-365,as

wellasthemiScriptPCRControlsRNU5a,RNU6b,RNU1a,SNORD25

andSNORA73a [28].Quantified miRsequencesarelistedinthe

SupportinginformationTableS2

2.7 Expressiondatanormalisation

AllinternalreferencegenesatthemRNAandmiRlevelwere

evaluatedwithgeNorm [29] and NormFinder [30] both partof

the gene expression analysis software suite GenEx 5.4.2

(Mul-tiD, Gothenburg, Sweden) All relative expression changes are

representedasfoldchangesaccordingtothefollowingformula:

FC=2−Ct[31]

2.8 MIQEcompliance

QuantitativePCRassayswerevalidatedaccordingtotheMIQE

guidelines[32](fordetailsseetheMIQEchecklistintheappendix)

In brief;sample andRNA integrity wereevaluated; exon

span-ningprimerdesign wasdoneusingPrimer 3web version;PCR

assayefficiency(96.2±5.6%)wasevaluatedusingdilutionseries;

relativegeneexpressiondatawereanalysedwithMIQEapproved

algorithmsusingGenEx5.4.2softwarepackage

2.9 MicroarraymRNAgeneexpressionprofiling

Using theGeneChip® MouseGene 1.0 ST Array(Affymetrix,

Santa Clara, United States), 28,853 well annotated genes are

detectable.UsingtheGeneChipArray,thegeneexpression

pro-filesinthepresenceandabsenceof3␮MATRAwereanalysedin

3T3-L1cellsateight differenttime points(0,2,4,7,10,24, 48

and96h).Therefore,threeindependentcellcultureexperiments

wereconducted,andRNAwasextractedasmentionedabove.In

total,100ngtotal RNAofeachsamplewasusedasthestarting

materialformicroarrayhybridisation[33].AnAffymetrix

process-ingprotocolwasutilised[34],and45arrayswereperformedby

theGenomicCoreFacilityoftheEuropeanMolecularBiology Lab-oratory(EMBL,Heidelberg,Germany).Fordatanormalisation,the GeneSpringGXSoftware(AgilentTechnologies,SantaClara,United States) [35] wasusedand therobust multichipanalysis (RMA) algorithmwasapplied.Thedatawereanalysedwithregardtothe qualityandquantityofthedata.Themicroarray datawere ver-ified byRT-qPCR, and thePearsoncorrelation coefficients were calculated.DatawereanalysedwiththeMultiexperimentViewer (Dana-Faber-Cancer-Institute,Boston,USA)[36].Datawere gener-atedaswellasanalysedfollowingtheMIAMEguidance[37] 2.10 MiRexpressionanalysis

FormiRexpressionprofiling,thenCounterMousemiR Expres-sionAssay(NanoString,Seattle,UnitedStates)wasused[38,39] Usingthisanalysissystem,600murineandmurine-associatedviral miRsaredetectable.Adigitalcolour-codedbarcodetechnologywas utilised,andeachbarcodewasattachedtoasingletarget-specific probecorrespondingtomiRofinterest.MiR-specificprobeswere designed againsttheannotations inmirBaseversion18 [40,41] Thesetarget-specificprobesweremixedtogetherwithcontrols; therefore, a multiplexedcodesetwas formed[38,39].ThemiR expressioninthepresenceandabsenceof3␮MATRAwas ana-lysedintwoindependentcellcultureexperimentsatninedifferent timepoints(0,2,4,7,10,24,48and96hand288h=12days).As

astartingamountfortheassays,100ngtotalRNAwasanalysed persample.Eachassaywasnormalisedonthebasis ofthe geo-metricmeanofthetop100expressedmiRs,andanormalisation factorrelativetothemeanofallsampleswascalculatedforeach assayusingthenSolverAnalysisSoftware(NanoString Technolo-gies).Thenormalisationfactorsforallassaysrangedbetween0.1 and10;thisrangeisalsorecommendedbyNanoStringTechnology itself.HeatmapsweregeneratedusingtheMultiexperimentViewer software[36]

2.11 Genomatixpathwaysystemanalysis(GePS) Foranalysisofenrichedbiologicalprocesses(GOterms)in DMI-treatedandDMI+3␮MATRA-treatedcells,weutilisedtheGePS Software (Genomatix,Munich,Germany).Based oninformation frompublicandproprietary databasesaswellasco-citationsin theliterature,thissoftwareenablesacomprehensiveanalysisof enrichedGOterms[42]

2.12 Geneontologyanalysis TheGOtermsdeterminedbyGePSwereanalysedbytheopen sourcesoftwareREVIGO (Laboratoryforinformation systemsat theRudjerBoskovicInstitute,Zagreb,Croatia)[43].REVIGO sum-marises lists of GOterms by excluding redundant terms using

aclusteringalgorithm.Onlynon-redundanttermsarevisualised

in clusters(rectangles) Related GOterms were summarisedin superclusterswiththesamecolourcode.Weselectedthe50most enriched GOtermsamong theATRAregulatedgenesover time (0–96h)

2.13 BuildingmiR–mRNAregulatorynetworks ForbuildingupanmiR–mRNAregulatorynetwork,weinitially setupaprediction-basednetworkbycombiningtheconserved tar-getpredictionsofTargetScan[44].We thenfitteda generalised linear model (GLM)onthe scaledexpressionprofiles of mRNA and its predicted miRs for each gene and computed the regu-larisationpathfortheelasticnetpenalty,whichisimplemented

intheGLMnetpackageforR[45].Using10-foldcross-validation,

correspond-ingmRNA regulation.Because we wereprimarily interested in

anti-correlativeeffectsofmiR–mRNAinteraction,weintroduced

anegativityconstraintbymodifyingtheimplementationinthe

GLMnetpackage[45].TheweightofthemiR–mRNArelation

cor-respondstotheabsolutecoefficientoftherespectivemiRinthe

GLM

Toidentifypathwaysthatarespecificallyaffectedbythe

net-work structure of the resulting miR–mRNA network, we first

calculatedap-valueofpathwayover-representationforeachgene

inthenetwork.Fromthis,wedefinedasetofneighbourgenesfor

eachgene.Thissetincludesallgenesthataretargetedbythesame

miRsasthegeneitself.WethenappliedFisher’sexacttesttotest

forover-representationofthepathwaygenesetamongthe

neigh-bourgenesetofeachgene.Weobtainedadistributionofp-values

foreverypathway.Thiswasdoneforallpathwaysderivedfromthe

KyotoEncyclopediaofGenesandGenomes(KEGG)[46].Following

this,werandomlypermutedthelabels100timesandtestedeach

timewhetherthep-valuedistributionoftheoriginalnetworkwas

significantlyshiftedtolowerp-valuescomparedwiththe

respec-tiverandom networkusinga two-sampleKolmogorov–Smirnov

test The visualisation of the pathways and the node

colour-ingaccordingtothepathwayover-representationwasperformed

usingCytoscape2.8.3(CytoscapeConsortium,SanDiego,USA)[47]

2.14 Oil-red-O-staining

Lipidaccumulation in ATRA-treated and −untreated 3T3-L1

cellswasdetectedbyoil-red-Ostaining[48].On day12 (288h)

afterstartingdifferentiation,themediumwasdiscardedandthe

cellswereairdriedfor20minandfixedatroomtemperatureby

adding10%formalininPBSfor15min.Afterfixation,thecellswere

airdriedagainandstainedwithafilteredoil-red-Oworking

solu-tionfor1hontheorbitalshaker.Fortheworkingsolution,the0.5%

oil-red-Oisopropanolstocksolutionwasdiluted3:2withwater

Stainedcellswerewashedthree timeswithdistilled water,the

wellswereairdriedandtheresidualcolouronthewellsideswas

removedwithanisopropanol-soakedcottonstick.The

six-well-plateswerescannedandstoredat4◦Cuntildyeextraction.The

dyewasextractedfromcellsbyadding1mlofthedyeextraction

solution.Thecellswereincubatedwiththesolutionfor1honthe

orbitalshaker,andtheabsorptionwasmeasuredat492nmwitha

platereader(TecanDeutschlandGmbH,Crailsheim,Germany)

3 Results

3.1 ATRAinhibitsadipogenesisinatime-dependentmanner

The well-known characteristic time-dependent phenotype

changesweredetectedduringdifferentiationof 3T3-L1cells,as

assessedbyoil-red-Ostaining(Fig.1a).Thecellswereculturedin

triplicatesfor24,48and96hinDMImedium(±3␮MATRA),and

DMImediumwaschangedtogrowthmedium(±3␮MATRA)after

theindicatedtimes.Allculturedcellswerestainedafter12days

(288h)ofdifferentiation.Thegrowthmediumwasrenewedevery

2days Cultures exposed only for the first 24h to 3␮M ATRA

showednosignificantinhibitionofdifferentiation.Lipid

accumu-lationwassimilartocontrolculturesthatwereexposedfor24hto

DMIonly.Theexposureto3␮MATRAfor48hresultedinthe

sup-pressionoflipidaccumulationby40%.Incubationofthecellswith

ATRAformorethan96hcausedfurtherinhibitionoflipid

accumu-lation.Therefore,incubationof3T3-L1preadipocyteswith3␮M

ATRAfor288h(12days)washighlyefficient;lipiddropletswere

onlyaccumulatedinafewcells.Thecapabilityfordifferentiation

incellstreatedwithATRAforalongterm(ATRAtreatmentpersists

from0hto288h)wassimilartothephenotypeof undifferenti-atedcells,whichwerecultivatedaftergrowtharrestforafurther

288h(12days)withfibroblastmedium(Fig.1a).Themorphologyof controlandATRA-treatedcellswasalsovisualisedmicroscopically

asshowninFig.1b.InATRA-treatedcultures,onlyafewmature adipocytesweredetectable,whereasthecontrolcultureswerefully differentiated

3.2 High-throughputgeneexpressionprofilinginATRA-treated and-untreatedcells

Usingmicroarraytechnology,geneexpressioninDMI-treated andDMI+3␮MATRA-treated 3T3-L1cells wasanalysedduring thefirst96h Thesquaredcorrelation coefficientsfor all corre-spondingtriplicates werecalculated (r2>0.95).Only transcripts withsignificantAffymetrixsignalvalues(>30)andwithatwofold up-or down-regulation ata minimum of onetime point were includedintheanalysis.Fig.1cdepictsanoverviewofthe rela-tivegeneexpressiontrend,eitherup-ordown-regulated,inthe control and ATRA-treated cells After 2h, 388 transcriptswere significantlydifferentiallyexpressedin3T3-L1preadipocytes.Of these,298 transcripts(77%) were up-regulatedby DMI and 62 transcripts(16%)weredown-regulatedbyDMI,whereasonly21 transcripts (5%) were induced by 3␮M ATRA and only seven transcripts(2%) wererepressedby3␮MATRA.After96h,1314 transcriptsweresignificantlyregulatedin3T3-L1preadipocytes

In total, 455 transcripts (35%) were up-regulated by DMI and

471(36%)weredown-regulatedbyDMI.Finally,148transcripts (11%)wereinducedby3␮MATRAand240transcripts(18%)were repressedbyATRA

3.3 High-throughputmiRexpressionprofilinginATRA-treated and-untreatedcells

Underthesamecultureconditions,miRexpressionin ATRA-treated and -untreated cells wasanalysed in two independent experiments.First,thequalityofmiRexpressiondatawas eval-uated, including calculation of an average squared correlation coefficientr2=0.88of thecorrespondingduplicatesand usinga two-wayanalysisofvariance(ANOVA)(criticalp-valuefor signif-icance:0.01)toselectsignificantlyregulatedmiRs.Furthermore, onlymiRsexhibitingstableNanoStringsignalvalues(>50)atall timepointsandwithatwofoldup-ordown-regulationinatleast onetimepointwereselectedforfurtheranalysis.Therewere23 miRsthatfulfilledthesecriteria(Fig.1d;theexpressionvaluesare presentedinTables1and2.These23selectedmiRswereusedfor thegenerationofthemiR–mRNAnetwork.Asproofoftheintegrity

ofourmiRexpressionassaydata,theexpressionlevelsofthree putativeadipogenesis-relevantmiRs(miR-103,miR-146and miR-221)andfivehighlyregulatedmiRs(miR-29a,miR29b,miR-365, miR93andmiR96)wereanalysedbyRT-qPCR(S1andS2Figures; thecorrespondingexpressionvaluesaresummarisedinthe Sup-porting information S3and S4Tables) The Pearsoncorrelation coefficientbetweenthenCountermiRexpressionassaydataand theRT-qPCRwashighlysignificant(r=0.80)(S3aFigure).Forthe calculation,weusedtheexpressionvaluesoftheabovementioned miRsoutofthemiRexpressionassays(Tables1and2)aswellas thecorrespondingexpressionvaluesoftheRT-qPCRevaluation(see theSupportinginformationS3andS4Tables)

3.4 InhibitionoffatcelldifferentiationbyATRAinvolvesthe repressionofadipogenicspecifictranscriptionfactors FormicroarraymRNAdatavalidation,wequantifiedthe expres-sionprofilesofadipogenesis-specifictranscriptionfactorssuchas Cebpˇ,Cebp˛,Ppar,Srebf1andRxr˛(Fig.2a–e).Theexpression

Fig 1.ATRA modulates the phenotype as well as the post-transcriptional mechanisms in 3T3-L1 cells (A) Oil-red-O-staining of 3T3-L1 cells: Quantification of lipid accumu-lation in DMI-treated (control) and DMI + 3 ␮M ATRA-treated 3T3-L1 cells After the indicated times, DMI (±3 ␮M ATRA) was replaced by growth medium without ATRA Only

in the labelled 96-h approach, DMI + 3 ␮M ATRA was changed to growth medium with ATRA for 288 h (12 d) (B) 3T3-L1 cell phenotypes: lipid accumulation in ATRA-treated and -untreated cells were visualised by microscopic images (C) Relative gene expression trend, either up- or down-regulated, in the control and ATRA-treated cells in the time frame 2 h to 96 h (compared with 0 h) (D) MicroRNA expression profiling: Expression changes in ATRA-untreated 3T3-L1 cells are presented relative to 0 h, whereas the expression changes in ATRA-treated 3T3-L1 cells are presented relative to the corresponding untreated samples An increase in microRNA expression is represented by accelerating red intensities, whereas decreasing ratios are represented by accelerating green intensities.

changes are presented relative tobaseline Compared withthe

expressionprofileofCebpˇinDMI-treatedcellsATRAhasno

influ-enceonCebpˇexpression after2h(Fig.2a) At288h(12days)

post-induction,Cebp˛(FC=86)(Fig.2b),Ppar (FC=37)(Fig.2c),

Srebf1(FC=7)(Fig.2d)aswellasRxr˛(FC=10)(Fig.2e)wasinduced

byDMIrelativeto0h,whereastheinductioninATRA-treatedcells

wasdownregulated(Cebp˛;FC=17,Ppar;FC=9, Srebf1;FC=2,

Rxr˛;FC=2).Themicroarraygeneexpressiondatawereconfirmed

byRT-qPCR,andwealsocalculatedthePearsoncorrelation

coeffi-cientbetweenthemicroarraydataandtheRT-qPCRdata(r=0.7)

(S3bFigure).ForthecalculationofthePearsoncorrelation

coeffi-cient,weusedtheexpressionvaluesoftheabovementionedgenes

fromthemicroarrayexperimentsaswellastheexpressionvalues

oftheRT-qPCRevaluation(S5–S8Tables)

3.5 RegulationofbiologicalprocessesinATRA-treated3T3-L1 cells

We next examined the effects of treatment and exposure time (0–96h) to 3␮M ATRA on biological processes in 3T3-L1 preadipocytes.AnoverviewoftheimpactofDMIaswellasofATRA exposureontheregulationofbiologicalprocessesispresentedin Fig.3a.TheeffectsofDMIonbiologicalprocessesduringtheearly stages(2,4and7hpost-induction)andthemidstages(10and24h post-induction)werestrongerthantheregulatoryeffectsofATRA

Inthelatestages(48and96hpost-induction)ofearly differen-tiation,theregulationofGOtermsbyDMIandATRAwasalmost identical.Insummary,at2hpost-induction,949GOtermswere significantlyregulatedbyDMI treatment,whereas only112GO

Table 1

Changes in microRNA expression in DMI-treated 3T3-L1 cells.

Fold changes are presented relative to 0 h (log 2 -transformed ratios).

Table 2

Changes in microRNA expression in 3 ␮M ATRA-treated 3T3-L1 cells.

microRNA 0 h ATRA 2 h ATRA 4 h ATRA 7 h ATRA 10 h ATRA 24 h ATRA 48 h ATRA 96 h ATRA 12d ATRA

Fold changes are presented relative to the corresponding untreated samples (log 2 -transformed ratio).

termswereregulatedbyATRAtreatment.Theregulatoryeffects

mediatedbyATRAgraduallyincreased,andinthelatestageofearly

differentiation(48and96hpost-induction)theregulatoryeffects

ofDMIandATRAweresimilar:451GOtermsweresubjectedto

treatmentwithDMIandeven471GOtermswerecontrolledby

ATRA(Fig.3a)

Themostaffectedsuperclustersin early-stagedifferentiation

werecellproliferation (bluerectangle)and tissue development

(redrectangle)(S4aFigure).Inmid-stagedifferentiation,themost

affectedsuperclustersweremulticellularorganismprocess(dark

purplerectangle),responsetowounding(lightpurplerectangles)

andcellproliferation(yellowrectangles)(S4bFigure).Inthelate

stagesofdifferentiation(48/96hpost-induction),themostaffected

superclusteruponATRAwasfoundtobecellularlipidmetabolism (Fig.3b)

3.6 Themetabolismof3T3-L1cellsisregulatedbyATRA

Inthenextstep,weanalysedmRNAgeneexpressionofkey reg-ulatorswithinenergymetabolism.Thegeneexpressionrelatedto triglyceridesynthesisin3T3-L1cellswasnegativelyinfluencedby ATRA.Indetail, thegenesthatwereinvolved inbeta-oxidation, thetricarboxylic acid cycle(TCA cycle), fatty acidbiosynthesis, fattyacidtransport,triglyceridesynthesis,lipidaccumulationas wellasdegradationoftriacylglycerolforenergyproductionwere down-regulated by ATRA (Table 3; Fig.4).In this context, the mRNAexpressionofacetyl-CoenzymeAdehydrogenase(Acadm), acetyl-CoAacyltransferase2(Acaa2),phosphofructokinase(Pfkp),

Fig 2. Quantitative analysis of mRNA expression by RT-qPCR of adipogenetic specific regulators in DMI-treated (red) and DMI + 3 ␮M ATRA-treated (blue) preadipocytes in

a time course study (0–288 h) ATRA has no effect on the gene expression of (A) Cebp␤, whereas the expression of (B) Cebp␣ (C) Ppar␥ (D) Srebf1 (E) Rxr␣ is inhibited by ATRA Expression changes are presented relative to 0 h.

malatedehydrogenase1(Mdh1)and pyruvatecarboxylase(Pcx)

wasinhibitedbyATRAinthelate stageofdifferentiation

Isoci-tratedehydrogenase2(Idh2)wasdown-regulatedbyATRAafter

24h.Thetranscriptionoffattyacidsynthase(Fasn)aswellasfatty

acid-binding proteins 4 and 5 (Fabp4; Fabp5) wasinhibited by

ATRA.Triglyceridesynthesisaswellaslipidaccumulationinvolves

theexpressionof1-acylglycerol-3-phosphateO-acyltransferase2

(Agpat2),lipin1(Lpin1),diacylglycerolO-acyltransferase2(Dgat2)

andperilipin(Plin)thatwerealsoinhibitedbyATRA.Inaddition,

thegeneexpressionpatternsubjectedtothedegradationof

tri-acylglycerolwasnegativelyregulatedbyATRA;theexpressionof

hormone-sensitivelipase(Lipe)andlipoproteinlipase(Lpl)wasalso

suppressedafter96h

3.7 TheregulationofbiologicalpathwaysinATRA-treated3T3-L1 preadipocytesrelativetotheobservedmiR-mRNAinteractions Usingamultiplelinearregressionmodel[49]combinedwith thederivedmiRandmRNAexpressiondataandthetarget predic-tionsofTargetScanMouse5.2[44,50],aninsilicobasedmiR–mRNA networkwasgenerated(Fig.5).Fig.6summarisesthemiR–mRNA interactions with specific pathway annotations and the corre-sponding expression profiles detected by our high-throughput transcriptomic data screening (Tables 1–4).The comprehensive miR–mRNAnetworksare visualisedin theSupporting informa-tionS5–S7Figures.Usingourinsilicobasedpathway-expression analysistools,wewereabletoshowthefollowing:1)Themost affectedpathwayssubjectedtoourdatasetsweregapjunction

sig-Fig 3. Regulation of biological processes in DMI-treated and DMI + 3 ␮M ATRA-treated 3T3-L1 cells (A) Regulated GO terms in DMI and ATRA treated 3T3-L1 cells in early stages (2, 4 and 7 h post-induction) and mid stages of differentiation (10 and 24 h), ATRA (blue) only has small effects on biological processes compared with DMI treatment (red) (B) Cluster of ATRA regulated GO terms in the late stages of early differentiation (48/96 h post-induction) GO terms were analysed by the Genomatix Software and the top 50 regulated GO terms per time were selected and visualised with Revigo The late regulation (48 and 96 h) of GO terms is visualised The cellular lipid metabolism

is most affected by ATRA, shown in the left yellow supercluster The size of the rectangles represents the level of significance, whereby the log10 p-values of the GO term enrichments are given at the bottom.

nalling(functionallocalityscore=2.86E-06),therearrangementof

theactincytoskeleton(functionallocalityscore=1.06E-10)andthe

regulationoftheTCAcycle(functionallocalityscore=5.84E-26).2)

MiR-27aandmiR-96areprobablythemostimportantco-actorsin

thesepathways,inwhichmiR-27aisinhibitedbyATRAand

miR-96isinducedbyATRA.3)Usinginsilico-basedtools,wewerealso

abletovisualisetheinteractionsofmiRswithmRNAswith

spe-cificpathwayannotation(mRNAsarevisualisedasrectanglesin

theSupportinginformationS5–S7Figures)subjectedtoourown

expressiondatasets.Therefore,insilico,wecouldvisualisethe

pre-dictedinteractionsofmiR-27aandthecorrespondingtranscripts

suchastheplatelet-derivedgrowthfactorreceptor,alpha

polypep-tide(Pdgfra),sonofsevenless1(Sos1),vav3oncogene(Vav3)and

LIMkinase 2(Limk2) Allmentioned genes are associated with gapjunctionsignalling onthebasisoftheinformationof KEGG pathways(KEGG:mmu04540)(Fig.6a)aswellasthe rearrange-mentofthecytoskeleton(Fig.6b)(KEGG-Pathway:mmu04810) Furthermore,wevisualisedtheinteractionofmiR-96withthe pre-dictedtranscriptsoftheWASproteinfamily,member2(Wasf2), theactinrelatedprotein2/3complex(Arp2/3)[51]andthe mus-cle and microspike RAS (Mras) All these mentioned genes are alsodirectlyinvolvedincytoskeletonregulation(KEGG-Pathway: mmu04810).Inadditiontothepotentialregulatoryeffectsof

miR-96insidetheactincytoskeleton(Fig.6b),Pcxanddihydrolipoamide S-acetyltransferase,theE2componentofthepyruvate dehydro-genasecomplex(Dlat),whicharekeyregulatorsoftheTCAcycle

Table 3

Changes in mRNA expression in ATRA-treated 3T3-L1 cells.

Dlat dihydrolipoamide S-acetyltransferase (E2 component of pyruvate dehydrogenase complex) 0 0 0 0,1 0,1 −0,3 −0,9

Fold changes are presented relative to the corresponding untreated samples (log 2 -transformed ratios).

Table 4

Changes in mRNA expression in DMI-treated 3T3-L1 cells.

Dlat dihydrolipoamide S-acetyltransferase (E2 component of pyruvate dehydrogenase complex) 0 0 0 −0,1 0,4 0,7 1,5 Pdgfra platelet derived growth factor receptor, alpha polypeptide −1,5 −2,8 −2,7 −2,1 −2,1 −2,4 −2,8

Fold changes are presented relative to 0 h (log 2 -transformed ratios).

(KEGG-Pathway:mmu00020)[52]areforecastedastargetsof

miR-96byTargetScanMouse(Fig.6c)

4 Discussion

ThereisincreasingevidencethatmiRsplayaveryimportantrole

inmetabolismandenergyhomeostasis.Therefore,thisregulatory

smallRNAspeciesmaybeinvolvedinthepathogenesisofdisorders

suchasthemetabolicsyndromeortype2diabetes.Inthisstudy,

weusedlarge-scaletranscriptomicmethodstoprofiletheglobal

expressionofmiRsandmRNAsin3T3-L1preadipocytesina

high-resolutiontimeframe.Wecombinedthesedatatogenerateglobal

molecularnetworksinordertoexplainthephenotypeof

ATRA-treated3T3-L1cells(Fig.1b)

Itisknownfromtheliterature[5,10,11]that3T3-L1cellsare

verysensitivetoATRAexposureduringtheearlystagesof

differen-tiation;however,simultaneously,theATRA-dependentinhibition

is reversible within the first 48h [11] This may imply that thelongertheincubationtime withATRA,thestrongerare the inhibitory effects of ATRA onfat cell differentiation.We could confirm this effectonphenotypicdevelopment (Fig 1a and b) Moreover,theseresultsareinaccordancewithourtranscriptomic data(Figs.1candFig.2)

In ourexperiments,a modulationof miRexpressionin DMI

aswellasinATRA-treatedcellswasobservedafter24hbutnot

atearlierstagesofdifferentiation,suggestingthatmiRsmodulate adipocytedifferentiationafterinductionofthisprogram(Fig.1d) Similartime-dependentmiRexpressionpatternsforDMI-induced differentiationof3T3-L1cellshavebeenreportedbyKajimotoetal [53]fortheTPA-induceddifferentiationofHL-60cells[54]aswell

asfortheneuronaldifferentiationofprimaryratcorticalcells[55] Moreover,ourmiRexpressiondata(Fig.1d)confirmthefindingsof

Fig 4.Schematic overview of metabolic processes that are affected by ATRA during adipogenesis in 3T3-L1 cells over 96 h The highlighted genes are direct targets of Ppar Gene expression changes in ATRA-treated 3T3-L1 cells are presented relative to the corresponding untreated samples The colour code reflects an up- or down-regulation greater than 1.5.

Xie(2009)[56]andKnelangen[57]thatmiR-221andmiR-125b-5p

aredown-regulatedduringdifferentiationof3T3-L1cells,whereas

miR-103andmiR-146bareup-regulated.Toourknowledge,our

studyisthefirstdetailedreportonthecomplexmodulationofmiR

expressionin3T3-L1preadipocytesbyATRA

SimilartothemRNAandmiRexpressionpattern,no adipocyte-specificGOtermswereenrichedinATRA-treatedcellsduringthe earlyandmidstagesofdifferentiation(seetheSupporting infor-mationS4aandS4bFigs.).Apossibleexplanationforthisfindingis thatononehand,itwasshownthatthemitoticclonalexpansionis