Transcription factor activity rhythms and tissue-specific ... · and lows in body temperature...

1

Transcription factor activity rhythms and tissue-specific chromatin interactions

explaincircadiangeneexpressionacrossorgans

JakeYeung1*, JérômeMermet1*,CélineJouffe3, JulienMarquis4,AlineCharpagne4,Frédéric

Gachon2,3,FélixNaef1

1Institute of Bioengineering, School of Life Sciences, Ecole Polytechnique Fédérale de

Lausanne(EPFL),Lausanne,CH-1015,Switzerland

2Faculty of Life Sciences, Ecole Polytechnique Fédérale de Lausanne (EPFL), CH-1015

Lausanne,Switzerland

3DepartmentofDiabetes andCircadianRhythms,Nestlé InstituteofHealth Sciences, CH-

1015Lausanne,Switzerland

4FunctionalGenomics,NestléInstituteofHealthSciences,CH-1015Lausanne,Switzerland

*Theseauthorscontributedequallytothiswork.

Correspondingauthor:

FélixNaef

InstituteofBioengineering

EcolePolytechniqueFédéraledeLausanne

CH-1015Lausanne

Switzerland

Tel:(0041)216931621

E-mail:[email protected]

.CC-BY-NC-ND 4.0 International licensecertified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which was notthis version posted October 23, 2017. . https://doi.org/10.1101/207787doi: bioRxiv preprint

https://doi.org/10.1101/207787

http://creativecommons.org/licenses/by-nc-nd/4.0/

2

Abstract

Temporalcontrolofphysiologyrequirestheinterplaybetweengenenetworksinvolvedin

daily timekeeping and tissue function across different organs. How the circadian clock

interweaveswith tissue-specific transcriptional programs is poorly understood.Herewe

dissected temporal and tissue-specific regulation at multiple gene regulatory layers by

examiningmouse tissueswithan intactordisruptedclockover time. Integratedanalysis

uncovered two distinct regulatory modes underlying tissue-specific rhythms: tissue-

specific oscillations in transcription factor (TF) activity, which were linked to feeding-

fastingcyclesinliverandsodiumhomeostasisinkidney;andco-localizedbindingofclock

and tissue-specific transcription factors at distal enhancers. Chromosome conformation

capture(4C-Seq)inliverandkidneyidentifiedliver-specificchromatinloopsthatrecruited

clock-bound enhancers to promoters to regulate liver-specific transcriptional rhythms.

Furthermore,this loopingwasremarkablypromoter-specificonthescaleof lessthanten

kilobases. Enhancers can contact a rhythmic promoter while looping out nearby

nonrhythmic alternative promoters, confining rhythmic enhancer activity to specific

promoters. These findings suggest that chromatin folding enables the clock to regulate

rhythmictranscriptionofspecificpromoterstooutputtemporaltranscriptionalprograms

tailoredtodifferenttissues.


https://doi.org/10.1101/207787


3

Introduction

Amammalianinternaltimingsystem,knownasthecircadianclock,orchestratestemporal

physiology in organs to anticipate daily environmental cycles (Dibner & Schibler 2015).

Individual cellswithinorganscontainamolecularoscillator that, togetherwithrhythmic

systemic signals suchashormones, temperature, and feedingbehavior, collectivelydrive

diurnal oscillations in gene expression and physiology (Lamia et al. 2008; Reinke et al.

2008;Vollmersetal.2012;Choetal.2012).Remarkably,thecircadianclockimpingeson

many gene regulatory layers, from transcriptional and posttranscriptional processes,

translationefficiency,totranslationalandposttranslationalprocesses(Mermetetal.2016).

Transcriptome analysis of large collections of mammalian cell types and tissues

havehighlightedthebreadthoftissue-specifictranscriptionalregulation(Yueetal.2014;

Merkinetal.2012).However,manyphysiologicalprocessesaredynamicatthetimescaleof

hoursandoftenundercircadiancontrol,suchashormonesecretion,drugandxenobiotic

metabolism, and glucose homeostasis (Takahashi et al. 2008). Therefore, unlocking the

temporaldimensiontotissue-specificgeneregulationisneededforanintegratedviewof

physiologicalcontrol.

Chronobiologystudieshaveshownthatdifferenttissuesutilizethecircadianclock

to drive tissue-specific rhythmic gene expression (Storch et al. 2002; Zhang et al. 2014;

Korenčičetal.2014),presumably toschedulephysiological functions tooptimal timesof

day.Indeed,geneticablationofthecircadianclockindifferenttissuescanleadtodivergent

pathologies, such as diabetes in pancreas-specific Bmal1 knockout (KO) and fasting

hypoglycemiainliver-specificBmal1KO,suggestingthattheclockinterweaveswithtissue-

specific transcriptional programs (Bass & Lazar 2016). But how diurnal and tissue-


https://doi.org/10.1101/207787


4

dependent regulatory landscapes interact to generate tissue-specific rhythms is poorly

understood.

Results

Contributionsoftissue,dailytime,andcircadianclocktoglobalvarianceinmRNAexpression

Toestimatetherespectivecontributionsoftissues,dailytime,andcircadianclocktoglobal

variance in gene expression, we analyzed available temporal transcriptomes across 11

tissuesinWTmice(Zhangetal.2014),andgeneratedtemporalRNA-Seqdataofliverand

kidney fromBmal1KOmiceandWTlittermates(SupplementalTableS1&Supplemental

TableS2,Methods).TheZhangetal.datasetwasobtainedunderdark-dark(DD),adlibitum

feeding, sampled every 2 hours. The liver and kidneyBmal1KO andWT datasets were

obtainedunderlight-dark,night-restrictedfeeding(LD)conditions,sampledevery4hours.

To avoid mixing different experimental designs (e.g. temporal resolution and

number of repeats, Deckard et al. 2013; Li et al. 2015), we analyzed these datasets

separately. We performed principal component analysis (PCA) on the entire set of

conditions (11 tissues times24 timepoints) toobtaina firstunbiasedoverview into the

contributionsoftissueandtime-specificvarianceinthedata.Thisshowedthatmostofthe

varianceconcerneddifferencesinexpressionbetweentissues(Figure1A&Supplemental

Figures S1A-D). Temporal variance, in particular 24h periodicity, was present among a

group of principle components carrying lower amounts of variance (Figure 1A &

Supplemental Figures S1E-G). Focusing on genome-wide temporal variationwithin each

tissue, we found that 24-hour rhythms constituted the largest contribution of temporal

variance, followedby12-hour rhythms,whichwere close tobackground levels formany


https://doi.org/10.1101/207787


5

tissues(Figure1B)(Hughesetal.2009).Wethusfocusedtherestofouranalysison24h

rhythms.

We analyzed the peak-to-trough amplitudes (hereafter also referred to as fold

change) of 24h rhythmic transcripts. This showed that metabolic tissues, notably liver,

brown fat, and skeletal muscle stand out as exhibiting far more (on the order of 100

transcripts) intermediate tohigh amplitude (between2 and10 fold) transcript rhythms.

Braintissuesshowvirtuallynorhythmictranscriptsabove4fold(Figure1C).Inliverand

kidneyofBmal1KOmice,thenumberofrhythmicmRNAswasreducedby3foldcompared

to WT littermates. This effect increased for larger amplitudes. Only few transcripts in

tissues of Bmal1 KO oscillated by more than 10 fold (Figure 1D). Thus, a functional

circadian clock is required for high amplitude transcript rhythms across diverse tissues,

while systemic signals regulate lower amplitude rhythms that persist in clock-deficient

liver(Hughesetal.2012;Atgeretal.2015;Sobeletal.2017)andkidney(Nikolaevaetal.

2012).

Combinatoricsofrhythmictranscriptexpressionacrosstissuesandgenotypes

Wereasonedthat identifyingsetsofgeneswithsharedrhythmsacrosssubsetsof tissues

wouldallowfindingunderlyingregulatorymechanisms.Wethereforedevelopedamodel

selection (MS) algorithm extending harmonic regression (Fisher 1929) to classify genes

into modules sharing rhythmic mRNA profiles across subsets of tissues (Figure 2A,

Methods). Phase-amplitude relationships (phase is defined as the time of the peak, and

amplitude as the log2 fold change) between genes and tissues are summarized using

complex-valuedsingularvaluedecomposition(SVD)(Figure2B,Methods).WeappliedMS

tothe11tissues,whichidentifiedgenemodulesinvolvingrhythmicmRNAaccumulationin

nearlyalltissues(tissue-wide)(Figure2C),insingletissues(tissue-specific),orinseveral


https://doi.org/10.1101/207787


6

tissues (tissue-restricted) (examples shown in Figure 2D & Supplemental Figure S2A &

SupplementalTableS3).

Thetissue-widemodulecontainedasetofbothclock-andsystem-drivenrhythmic

mRNAs,asdeterminedbycomparingBmal1KOdata in liverandkidney(Figure2C, left).

Moreover, these transcriptsoscillated insynchronyacrossall tissuesandpeakedat fixed

timesof day, albeit their amplitudes variedbetween tissues,withbrain regions showing

the smallest amplitudes (Figure 2C, right). The clock drove synchronized oscillations at

highamplitudes,notablyclockgenes(e.g.Arntl,Npas2,Nr1d1,2;notethatArntlandNr1d1,2

arealsonamedBmal1andRev-erba,brespectively),clockoutputgenes(e.g.Dbp,Nfil3),and

cellcycleregulators(Cdkn1aandWee1)(Gréchez-Cassiauetal.2008;Matsuoetal.2003).

Interestingly, clock genes Per1,2 continued to oscillate in Bmal1KO in multiple tissues,

extending previous studies in liver (Kornmann et al. 2007). Other clock-independent

oscillations included mRNAs of heat- and cold-induced genes, such as Hspa8 and Cirbp

(Morfetal.2012;Goticetal.2016),thatpeaked12hoursapartnearCT18andCT6(CT:

circadian time;CT0corresponds to subjectivedawnandstartof the restingphase;CT12

corresponds tosubjectiveduskandstartof theactivityphase), concomitantlywithhighs

andlowsinbodytemperaturerhythms(Refinetti&Menaker1992).

Tissue-restrictedmodulescontainedrhythmictranscriptsthatpeakedinsynchrony,

such as in liver and kidney, or with fixed offsets, such as the nearly 12 hours shifted

rhythmsinbrownfatandskeletalmuscle(SupplementalFigureS3A).Overall, transcripts

withlargeamplitudes(FC>8)oscillatedineitherafewtissues(3orless)ortissue-wide(8

ormore)(Figure2E).

To distinguish clock- and system-driven mRNA rhythms, we applied the MS

algorithmtotheliverandkidneytranscriptomesinWTandBmal1KOmice(Figure2F&


https://doi.org/10.1101/207787


7

SupplementalFigureS3B&SupplementalTableS4).Thisseparationidentifiedclock-and

system-driven modules that oscillated in liver but were flat in kidney (Figure 2F), as

exemplified by mRNAs of Lipg and Lpin1 (Supplemental Figure S2B). Indeed, both

transcriptsoscillatedinWTliverwithrobustamplitudes,peakingnearZT11,butwereflat

inkidney(ZT:Zeitgebertime;ZT0correspondstoonsetoflights-on;ZT12correspondsto

onsetoflights-off).However,inBmal1KO,Lpin1continuedtooscillate,whileLipgwasflat.

Summarizing,we found thatsharedclock-drivenmRNArhythms,whichcontained

coreclockandclock-controlledgenes,oscillatedwithsignificantly largeramplitudesthan

system-drivengenes(Figure2G,magentasolidversusdotted).Similarly,clock-drivenliver-

specificmRNArhythmsalsooscillatedathigheramplitudescomparedwithsystem-driven

mRNA rhythms (Figure 2G, red solid versus dotted). On the other hand, kidney-specific

clock- and system-driven transcripts oscillated with comparable amplitudes (Figure 2G,

bluesolidversusdotted),andwerelessnumerousoverall,whichcouldreflectthedistinct

cell typesconstitutingthekidney(Leeetal.2015).Theuncovereddiversityofclock-and

system-drivenmRNArhythmsinvolvingdistinctcombinationsoftissueshintsatcomplex

transcriptional or post-transcriptional regulation. Below, we examine transcription

regulatorsresponsiblefortissue-specificmRNArhythms.

OscillatoryTFactivityinonetissuebutnototherscandrivetissue-specificmRNArhythms

We focused on WT and Bmal1 KO liver and kidney to identify rhythmic TF activities

underlying clock- and system-driven tissue-specific mRNA rhythms. We first analyzed

liver-rhythmic genes driven by systemic signals (n=1395, MS; Figure 3A), which were

associated with feeding and fasting rhythms (GO analysis around the clock, Method).

Indeed, ribosomebiogenesiswasupregulatedmost stronglyduring the first six hours of

the feeding phase (from ZT12 to ZT18) (Jouffe et al. 2013; Chauvin et al. 2014), while


https://doi.org/10.1101/207787


8

insulinsignalingwasdownregulatedduringfirstsixhoursofthefastingphase(fromZT0to

ZT6) (Ravnskjaer et al. 2013), consistentwithdaily responses tonutrient fluctuations in

liver(Sintureletal.2017).

ToinferrhythmicTFactivitiesthatmayunderliethesemRNArhythms,weapplieda

penalizedregressionmodel(MARA)(Balwierzetal.2014)that integratesTFbindingsite

predictionsnearpromoterswithmRNAaccumulation.TFanalysisof thismodulenotably

identified TFs related to insulin biosynthesis and gluconeogenesis, such as MAFB

(Matsuokaetal.2003)andEGR1(Matsuokaetal.2003;Shenetal.2015),whoseactivities

peakedatZT11andZT3,respectively(Figure3B&SupplementalFigureS4A).Integrating

temporalactivitiesofcandidateTFswithRNA-Seqandourpreviouslydescribedtemporal

nuclearproteindataset(Wangetal.2017),we foundthatrhythmicactivityofMAFBand

EGR1wassupportedbyrhythmicmRNAabundancefollowedbyrhythmicnuclearprotein

abundance (Figure 3B, Supplemental Figure S4B), likely reflecting the delayed protein

abundanceaftermRNAaccumulation(Mermetetal.2016).

Next, we analyzed clock-driven transcripts oscillating specifically in the kidney

(n=156,MS;Figure3C),amongwhichsodiumionandorganicaniontransporterspeaked

near ZT12 andZT0, respectively. Theupregulation of sodium ion transporters in kidney

during the behaviorally active phase may underlie clock-dependent increase of sodium

excretion(Nikolaevaetal.2012).Similarly,theupregulationoforganicaniontransporters

during the resting phase may explain increased transport activity for precursors of

gluconeogenesis,suchaspyruvateandlactate,duringfasting(Ekbergetal.1999;Stumvoll

etal.1998).mRNAsthatpeakedduringtherestingphasemayberegulatedbyTFCP2,as

predictedbyTFanalysis(Figure3D&SupplementalFigureS4C).Inaddition,thepredicted

TFCP2 activity was anti-phasic with Tfcp2mRNA abundance, suggestive of a repressive


https://doi.org/10.1101/207787


9

activity,consistentwiththeabilityofTFCP2torecruithistonedeacetylaseHDAC1(Kimet

al.2016).

Finally, liver-specific clock-driven rhythmic transcripts (n=991, MS) were

comprised of genes associatedwith glucosemetabolism (enriched at ZT18), such asGck

and Ppp1r3b (Kelsall et al. 2009; Oosterveer & Schoonjans 2014), as well as lipid,

cholesterol, and bile acid metabolism genes (enriched at ZT2), such as Elovl3, Insig2,

Hsd3b7,andCyp8b1(Guillouetal.2010;LeMartelotetal.2009;Sayinetal.2013;Sheaet

al.2007)(Figure3E).PredictedactivityofELFoscillatedandpeakednearZT3inWTliver

but was flat in Bmal1 KO (Fang et al. 2014) (Figure 3F & Supplemental Figure S4D).

Interestingly, mRNA abundance of Elf1 as well as its nuclear protein abundance also

oscillated in WT, supporting Elf1 as a potential regulator of oscillating transcriptions

peakingnearmidday.Thus,theMSalgorithmseparatedgenesintophysiologicallyrelevant

modules, allowing reliable prediction of rhythmically active TFs regulating temporal

physiologyofrespectivetissues.

Co-localizedbindingofclockandliver-specificTFsdrivesliver-specificmRNArhythms

To further dissect liver-specific clock-driven rhythms, we reasoned that accessible

chromatin regions specific to the liver couldharbor regulatory sites for clockTFs,which

could then regulatemRNA rhythms liver-specifically. Comparing DNase I hypersensitive

sites(DHSs)inliverandkidney(DNase-SeqdatafromENCODE)(Yueetal.2014),wefound

thatliver-specificclock-drivengeneswereenrichedwithliver-specificDHSs(within40kb

frompromoters), compared to system-driven aswell as nonrhythmic genes (Figure 4A).

UsingTFbindingsitepredictionsunderlyingtheseliver-specificDHSs,weappliedMARAto

predictrhythmicTFactivities thatexplaingeneexpressionof thismodule(Supplemental

FigureS5A).InWTliver,thepredictedactivityofROREoscillatedwithrobustamplitudes


https://doi.org/10.1101/207787


10

andpeakednearZT21.ROREactivitybecamehighand flat inBmal1KO liver, consistent

withlossofREV-ERBexpressionandconsequentlyderepressionofREV-ERBtargetgenes

(Buggeetal.2012)(Figure4B,top).ActivityofE-boxinWTliverpeakedatZT7,consistent

withBMAL1:CLOCKactivity(Reyetal.2011),albeitwithweakeramplitudescomparedto

ROREactivity,likelyreflectingfewerE-boxtargetgenescomparedtoROREinthismodule.

InBmal1KOmice;E-boxactivitywaslowandflatinliver,asexpected.

Wehypothesized that cooperativityof liver-specific and clockTFsat liver-specific

DHSs can regulate liver-specificmRNA rhythms. Pairwise analysis of TF binding sites at

liver-specificDHSsfoundenrichmentofco-occurrencebetweenROREandliver-specificTF

motifs, FOXA2, ONECUT, and CUX2 (Figure 4C). Enrichment of both CUX2 and ONECUT

(alsonamedHNF6)isconsistentwithONECUT1bindingtobothONECUTandCUX2motifs

(Confortoetal.2015).mRNAsofgeneswithco-occurrenceofROREand liver-specificTF

motifs peaked near ZT0-ZT2, consistentwith peak RORE activity (near ZT21) preceding

peakmRNAabundanceofREV-ERB targets (SupplementalFigureS5B).AnalysisofChIP-

exodatasetstargetingFOXA2,ONECUT1,andREV-ERBainliver(Iwafuchi-Doietal.2016;

Wang et al. 2014; Zhang et al. 2015) confirmed co-localized TF binding at liver-specific

DHSsdistalfromclock-drivenlivermRNAssuchasInsig2andSlc4a4(Figure4D).Thus,co-

localizedbindingof liver-specificandclockTFsatdistal liver-specificDHSsmayregulate

liver-specificmRNArhythms.

Liver-specificchromatinloopsregulateliver-specificmRNArhythms

Totestwhetherdistallylocatedliver-specificDHSscancontactpromotersofclock-driven

liver-rhythmicgenes,weselectedthepromotersofMreg,Pik3ap1,andSlc44a1asbaitsfor

4C-SeqexperimentsinliverandkidneyharvestedatthetimeofpeakmRNAaccumulation

for the selected genes (Methods, Figure 5A & Supplemental Figure S6A & Supplemental


https://doi.org/10.1101/207787


11

FigureS7A).UpstreamofMreg,the4C-Seqsignal,whichmeasuresfrequencyofpromoter-

enhancercontacts(vandeWerkenetal.2012),decayedrapidlytobackgroundlevelinboth

liverandkidney(Figure5Btop).DownstreamofMreg,however,the4C-Seqsignalshowed

a tissue-dependent pattern, decaying slowly in the liver butmore rapidly in the kidney.

This difference in decay suggests increased frequency of promoter-enhancer contacts in

the liver compared to the kidney. Indeed, differential analysis identified liver-specific

chromatincontacts40kbdownstreamofthepromoter(Figure5Bbottom).Overlayingthe

contactdatawithDNase-Seq,wefoundthatliver-specificchromatincontactsdownstream

ofMregconnected liver-specificDHSswith theMregpromoter (Figure5C).Furthermore,

ChIP-exo showed co-localization of REV-ERBa and FOXA2 binding at liver-specific DHSs

contactingthepromoters(Figure5C).Bycontrast,accessibleregionsupstreamoftheMreg

promoterdidnotshowliver-specificchromatincontacts.The4C-Seqdatathussuggestthat

liver-specific chromatin loops can recruit clock-bound distal elements to promoters to

regulate liver-specific transcriptional rhythms. Other liver-specific rhythmic transcripts,

Pik3ap1andSlc44a1,alsodisplayedliver-specificchromatinloopsbetweenpromoterand

liver-specificopenchromatinregions(SupplementalFigureS6&SupplementalFigureS7),

corroboratingthatsuchtissue-specificloopingdrivestissue-specificmRNArhythms.

Precisepromoter-enhancercontactsunderlieliver-specificmRNArhythms

Totestwhetherdistinctchromatinloopswouldformatalternativenearbygenepromoters

with distinct temporal mRNA profiles, we searched for candidate genes where one

promoter was rhythmically transcribed while the alternative one was nonrhythmic

(Supplemental Figure S8). Slc45a3has two alternative transcripts using promoters 8 kb

apart,withtheshorteroscillatingintheliver(rhythmicpromoter,Slc45a3-short),whilethe

longernot(flatpromoter,Slc45a3-long).Inkidney,neitherSlc45a3-shortnorSlc45a3-long


https://doi.org/10.1101/207787


12

showedrobust transcriptrhythms(SupplementalFigureS9).Targeting theSlc45a3-short

promoterwith4C-Seq in liverandkidneyshowed liver-specific chromatin loopsat three

distal regions (two upstream, one downstream) (Figure 6A). Remarkably, these same

regionsdidnotformliver-specificchromatinloopswiththeSlc45a3-longpromoter(Figure

6B), suggesting thatpromoters8kbapartcancontactdistinctenhancers.Overlaying4C-

Seq with DNase-Seq, we found that these chromatin loops link liver-specific DHSs

specificallytotheSlc45a3-shortpromoter(Figure6C).Theseliver-specificDHSsarebound

byliver-specificTFs,FOXA2andONECUT1,andclockTF,REV-ERBa,asshowninChIP-Seq.

Taken together, the 4C experiments suggest that enhancers can contact a rhythmic

promoterwhileloopingoutnearbynonrhythmicalternativepromoters,confiningrhythmic

enhancer activity to specific promoters (Figure 6D). Furthermore, rhythmically active

enhancerscancontactpromotersinatissue-specificmanner.Thus,chromatinfoldingnot

only regulates tissue-specific rhythms, but also differentiates between closely spaced

promoterstocontrolrhythmictranscriptionwithspatialprecision.


https://doi.org/10.1101/207787


13

Discussion

Themammaliangenomeencodestranscriptionalprogramsthatallowthemolecularclock

to robustly oscillate across diverse tissue transcriptomeswhilemaintaining flexibility to

regulatedistinctclockoutputsindifferentcombinationsoftissues.Hereweidentifiedtwo

regulatorymodesunderlyingtissue-specific transcriptrhythms:(1)regulatorysequences

canrecruit individualTFsbearingrhythmicactivity;(2)coordinatedbindingofclockand

tissue-specificTFscangeneratetissue-specificrhythms.Moreover,wefoundthatclockand

tissue-specific TFs bound at distal enhancers can be recruited to promoters through

remarkablyprecisechromatinloops.

Several of our predictions of transcription regulators and regulated genes (e.g.

EGR1,Por,Upp2)corroboratedwithpreviousanalysesofindependentdatasets(Yanetal.

2008;Bozeketal.2009;Bhargavaetal.2015).Furtheranalysis incorporatingoutputsof

enhanceractivity, suchaseRNAs(Fangetal.2014),acrossmultiple tissuesmayuncover

additionalrhythmicallyactiveregulators.

Co-localized binding of clock and tissue-specific TFs at enhancers provides a

putative mechanism for the clock to regulate clock output genes in a tissue-specific

manner. Inmouse liver,clockTFscanco-localizewithmultiple liver-specificTFs,suchas

FOXA2 and ONECUT1, consistent with multiple liver TFs associating with liver-specific

DHSs(Iwafuchi-Doietal.2016).Ourfindingsarecurrentlybasedonsequence-specificDNA

bindingofTFs,comparisonoftissues,andChIP-Seqdatasets.Furthermechanisticbasisfor

the functional significance of co-localization could be gained, for example by using

inducible knockout models for tissue-specific regulators. Moreover, the observed co-

localization do not exclude other cooperative modes, such as tethering of REV-ERBa to

ONECUT1throughprotein-proteininteractions(Zhangetal.2015).


https://doi.org/10.1101/207787


14

Our4Canalysisshowedthatchromatinloopingmightmediateinteractionbetween

clock and tissue-specific transcriptional programs, by recruiting clock-bound distal

elementstopromoters inatissue-specificmanner.Remarkably,such loopscansurgically

discriminate between nearby promoters as close as 8 kb apart, suggesting a way to

separatethetemporalregulationofneighboringpromoters.Apreviousstudyapplying4C

techniques to probe the contact landscape of a core clock gene enhancer proposed that

cohesion-mediated promoter-enhancer looping can compartmentalize rhythmic gene

expression within genomic regions spanning 150 kb (Xu et al. 2016). Here, chromatin

interactions that differed between tissues were localized to a relatively small genomic

region(<10kb)proximaltothepromoters(<100kb).Futurestudiesintegratingtemporal

data across tissueswith large-scale promoter-enhancer networksmay reveal regulatory

sequences that encode promoter-enhancer compatibility and elucidate whether this

compatibilityistissue-specific(Li&Noll1994;Merlietal.1996;Zabidietal.2014;Nguyen

etal.2016).

Overall, this work proposed a role for newly identified rhythmic transcription

factors and tissue-specific chromatin interactions in regulating tissue-specific rhythmic

geneexpression.Whileourworkfocusedontranscriptionalmechanisms,studyingothers

mechanisms such as posttranscriptional, translational, and posttranslational processes

usingPRO-Seq,Ribo-Seq,andproteomicsdatamayprovideadditionalinsights.Expanding

our 24-hour analysis to 12-hour or other harmonics would broaden the view of tissue-

specific temporal gene expression, but may require experimental designs of higher

temporalresolution(Hughesetal.2009;Krishnaiahetal.2017).Tissuesregulatedynamic

physiological processes such as glucose homeostasis, lipid metabolism, and sodium

homeostasis at different times of day. Thus, integrating the temporal axis into tissue-


https://doi.org/10.1101/207787


15

specific gene regulation offers an integrated understanding of how tissue physiology

resonateswithdailycyclesintheenvironment.

MaterialsandMethods

Animalexperiments

8-14weeksoldC57Bl/6micehavebeenpurchasedfromCharlesRiverLaboratory.Bmal1

KOmicehavebeenpreviouslydescribed(Jouffeetal.2013).Without further indications,

mice are kept under 12 hours light/12 hours dark regimen and ad libitum feeding. All

animalcareandhandlingwasperformedaccordingtotheCantondeVaud(FredGachon,

authorizationnoVD2720)lawsforanimalprotection.

RNA-Seqexperimentsandanalysis

Processing

TocomplementthemouseliverWTandBmal1KORNA-Seqdata(GSE73554)(Atgeretal.

2015), transcriptomesof kidneys fromBmal1KOandWT littermates (12hours light/12

hoursregimen;night-restrictedfeeding)weremeasuredfollowingthesameprotocolasin

(Atger et al. 2015). mRNA levels were quantified using kallisto version 0.42.4 (mm10)

(Brayetal.2015).

GlobalTemporalVariance

Foreachtissue,weestimatedthecontributionoftemporalvarianceforeachgene,broken

downbyitsFouriercomponents.Wecalculatedthebackgroundlevelassumingtemporally

unstructured data (white noise), whose magnitude (strength of the white noise) was


https://doi.org/10.1101/207787


16

estimated from the mean of squared magnitudes of Fourier coefficients that were not

submultiplesof24hours(i.e.,themeanof48,16,9.6,6.9,5.3,4.4hourcomponents).

ModelSelection

We fitted harmonic regression models that integrated temporal gene expression across

different combinations of rhythms in different conditions (Atger et al. 2015). One

difference frompreviousmethodswas that forcomparingdifferentmodels,weusedag-

priorfortherhythmicparameters𝛽ratherthanBIC(Liangetal.2008),

β~ N(0, gσ! X!X!! .

The hyperparameter g controls the spread of the prior over themodels; as g increases,

simplermodels(suchastissue-widemodel)arefavoredovermorecomplexmodels(such

as model with many divergent rhythms). We set g=1000, which we found to maximize

temporal variations captured in the shared rhythms model while minimizing temporal

variationscapturedintheflatmodel.Thenumberofrhythmiccombinationskscalesasa

functionofthenumberofconditionsnas𝑘(𝑛) = 𝐵!!!whereBistheBellnumberusedin

combinatorialmathematics.Seesupplementalmethodsfordetails.

Complexsingularvaluedecomposition(SVD)representationofgeneandtissuemodule

Geneexpressionovertimeandacrosstissuescanberepresentedasa3-dimensionalarray.

However,sinceSVDofatensordoesnothaveall thepropertiesofamatrixSVD,wefirst

transformedthetimedomaintothefrequencydomaincorrespondingto24-hourrhythms

forallgenesgandconditionsc:

𝐸!,! = 𝐸!,!,!𝑒!"#

!∈!

where𝐸!,! isacomplexvaluerepresentingtheamplitudeandphaseofexpressionforgene

ginconditioncand 𝜔 = 2𝜋/24.


https://doi.org/10.1101/207787


17

The resulting matrix was decomposed using SVD and the first left -and right-singular

valueswere visualized in separate polar plots. To ensure the first component recovered

mostoftheoriginalsignal,theSVDrepresentationwasperformedseparatelyforeachgene

moduleidentifiedbymodelselection.

PredictingActivitiesofTranscriptionalRegulators

Predictionsoftranscriptionfactorbindingsite(TFBS)

ForTFBSpredictionsnearpromoters,weusedmotevoversion1.03(Arnoldetal.2012)to

scan +/- 500 bp around the promoter. We used promoters (Balwierz et al. 2009) and

weight matrices of transcription factors defined by SwissRegulon (Pachkov et al. 2013)

(http://swissregulon.unibas.ch/fcgi/sr/downloads).For distal regions, we scanned the

genome for TFBSs in 500bpwindows in genomic regionswithin 40 kb of an annotated

gene.

Penalizedregressionmodel

We applied a penalized regressionmodel as previously described (Balwierz et al. 2014)

using an L2 penalty for penalization, which allows a direct estimate of the standard

deviation. Rhythmic activities of transcription factor motifs were summarized using

complex-valuedsingularvaluedecomposition.Weprojectedtheactivitiestoanamplitude

andphaseandcalculatedthezscoreoftheamplitude.Weconsideredactivitieswithzscore

>1.25as rhythmicTFactivities.Timeofpeak temporalactivitiesof transcription factors

were subtracted by 3 hours, to account for an average 3 hour shift between peak

transcriptionandpeakmRNAaccumulation(LeMartelotetal.2012).

Enrichmentofpairsofmotifs

We applied log-linear models to test for statistical significance between pairs of motifs

acrossrhythmicversusnonrhythmicmodules.Foreachmotif,weorderedDHSsitesbythe


https://doi.org/10.1101/207787


18

posteriorsitecountofthemotif(decreasingorder)andconsideredthemotiftobepresent

intheDHSsiteifthesitecountwasinthetop300(Myšičkováetal.2012).Weconsidered

liver-specificDHSsitesthatwereannotatedtoaclock-dependentliver-rhythmicgeneorto

anonrhythmicgene.Foreachannotatedlabelandforeachpairofmotifs,weconstructeda

2 by 2 contingency table by counting the number of DHS sites that contain one of the

motifs,bothmotifs,ornone,resulting ina3-waycontingencytable(motif1,motif2,and

annotated label). We assessed whether the resulting contingency table was statistically

significant to a nullmodel,where the nullmodelwas the expected counts if the pair of

motifswerejointlyindependentontherhythmicity.

Chromatinconformationexperimentsandanalysis

C57Bl/6micewere sacrificed at ZT08 and ZT20 to extract liver and kidneys. Liver and

kidney nuclei were prepared as previously described (Ripperger & Schibler 2006) with

some minor changes. 4C-Seq assays were performed as in (Gheldof et al. 2012). See

supplementalmethodsfordetails.

Rawreadcountsforeachsamplewerenormalizedbylibrarysizebythesumoftheread

countsonthecis-chromosome(excluding10fragmentsaroundthebait).Readcountswere

log-transformedusingtheformula:

Y = log!"cp+ 1

wherep=500,thepseudocount.


https://doi.org/10.1101/207787


19

The weighted linear model was fit locally around each fragment f. A Gaussian window

centered on f was used to incorporate signal from neighboring fragments. The 4C-Seq

fragmentcountswasmodeledbythefragmenteffectiandtissueeffectj.

Y!,! = a! + b! + ϵ!,

Wheretheweightsofthelinearmodelisdefinedas:

W = w!w!,

Where:

w! isaGaussiansmoothingkernel(widthσ! = 2500bp,centeredonfragmentf).

w!is the sampleweight based on the number of non-zero values counts on fragment i,

specifically,weusedw! = (0.5, 1.5, 2.5)forfragmentswith(0, 1, 2)finitecountsoutofthe

tworeplicates.

Differentialcontactswereestimatedusingt-statistics:

Z =∆bσ

Whereσ!standsfortheregularizedsamplevariance:

σ!! =σ!

! +σ!"#! exp −

bb!

Where:

b=theestimatedsignalacrosssamples

b! = log!"(2)


https://doi.org/10.1101/207787


20

DataAccess

Raw and processed data generated in this study are available in the Gene Expression

Omnibus(GEO)databaseunderaccessionnumberGSE100457.

Acknowledgements

We thank Eric Paquet for critical reading and Saeed Omidi for help and discussions in

bioinformatics. This work was supported by Swiss National Science Foundation Grant

31003A-153340, EuropeanResearchCouncil Grant ERC-2010-StG-260667, and theEcole

Polytechnique de Lausanne. J.Y. benefits from the Natural Sciences and Engineering

ResearchCouncilofCanadaPostgraduateStudiesDoctoralscholarship.

AuthorContributions

Conceptualization, J.Y., J.M., andF.N;Formalanalysis, J.Y. andF.N.; Investigation, J.M.,C.J.,

J.M.,A.C.;Writing–OriginalDraft, J.Y., J.M.,andF.N.;Writing–Review&Editing, J.Y., J.M.,

F.G.,F.N.;Supervision,F.N.andF.G.;FundingAcquisition,F.N.andF.G.


https://doi.org/10.1101/207787


21

References

Arnold, P. et al., 2012. MotEvo: integrated Bayesian probabilistic methods for inferring

regulatorysitesandmotifsonmultiplealignmentsofDNAsequences.Bioinformatics

(Oxford, England), 28(4), pp.487–94. Available at:

http://bioinformatics.oxfordjournals.org/content/28/4/487.short [Accessed August

10,2015].

Atger, F. et al., 2015. Circadian and feeding rhythmsdifferentially affect rhythmicmRNA

transcription and translation inmouse liver.Proceedingsof theNationalAcademyof

Sciences, 112(47), pp.E6579–E6588. Available at:

http://www.pnas.org/lookup/doi/10.1073/pnas.1515308112 [Accessed June 13,

2016].

Balwierz,P.J.etal.,2014.ISMARA:automatedmodelingofgenomicsignalsasademocracy

of regulatory motifs. Genome research, 24(5), pp.869–84. Available at:

http://genome.cshlp.org/content/24/5/869.short[AccessedAugust3,2015].

Balwierz, P.J. et al., 2009. Methods for analyzing deep sequencing expression data:

constructing the human and mouse promoterome with deepCAGE data. Genome

Biology, 10(7), p.R79. Available at:

http://genomebiology.biomedcentral.com/articles/10.1186/gb-2009-10-7-r79

[AccessedDecember7,2016].

Bass, J. & Lazar, M.A., 2016. Circadian time signatures of fitness and disease. Science,

354(6315). Available at: http://science.sciencemag.org/content/354/6315/994.full

[AccessedMay24,2017].

Bhargava,A.,Herzel,H.&Ananthasubramaniam,B.,2015.Miningfornovelcandidateclock

genes inthecircadianregulatorynetwork.BMCSystemsBiology,9(1),p.78.Available


https://doi.org/10.1101/207787


22

at:http://www.biomedcentral.com/1752-0509/9/78[AccessedOctober2,2017].

Bozek,K.etal.,2009.RegulationofClock-ControlledGenesinMammalsW.W.Wasserman,

ed. PLoS ONE, 4(3), p.e4882. Available at:

http://dx.plos.org/10.1371/journal.pone.0004882[AccessedOctober28,2016].

Bray, N. et al., 2015. Near-optimal RNA-Seq quantification. Available at:

http://arxiv.org/abs/1505.02710[AccessedAugust17,2015].

Bugge,A.etal.,2012.Rev-erbαandRev-erbβcoordinatelyprotectthecircadianclockand

normal metabolic function. Genes & development, 26(7), pp.657–67. Available at:

http://www.ncbi.nlm.nih.gov/pubmed/22474260[AccessedMay20,2017].

Chauvin, C. et al., 2014. Ribosomal protein S6 kinase activity controls the ribosome

biogenesis transcriptional program. Oncogene, 33(4), pp.474–483. Available at:

http://www.nature.com/doifinder/10.1038/onc.2012.606 [Accessed September 26,

2016].

Cho,H.etal.,2012.RegulationofcircadianbehaviourandmetabolismbyREV-ERB-αand

REV-ERB-β. Nature, 485(7396), pp.123–127. Available at:

http://www.nature.com/doifinder/10.1038/nature11048 [Accessed November 26,

2016].

Conforto, T.L., Steinhardt, G.F. &Waxman, D.J., 2015. Cross Talk Between GH-Regulated

TranscriptionFactorsHNF6andCUX2inAdultMouseLiver.MolecularEndocrinology,

29(9), pp.1286–1302. Available at: https://academic.oup.com/mend/article-

lookup/doi/10.1210/me.2015-1028[AccessedApril27,2017].

Deckard, A. et al., 2013. Design and analysis of large-scale biological rhythm studies: a

comparison of algorithms for detecting periodic signals in biological data.

Bioinformatics, 29(24), pp.3174–3180. Available at:


https://doi.org/10.1101/207787


23

https://academic.oup.com/bioinformatics/article-

lookup/doi/10.1093/bioinformatics/btt541[AccessedOctober6,2017].

Dibner, C. & Schibler, U., 2015. Circadian timing of metabolism in animal models and

humans. Journal of Internal Medicine, 277(5), pp.513–527. Available at:

http://doi.wiley.com/10.1111/joim.12347[AccessedMarch27,2017].

Ekberg, K. et al., 1999. Contributions by Kidney and Liver to Glucose Production in the

PostabsorptiveStateandAfter60hofFasting.DIABETES,48.

Fang, B. et al., 2014. Circadian Enhancers CoordinateMultiple Phases of Rhythmic Gene

Transcription InVivo. Cell, 159(5), pp.1140–1152. Available at:

http://www.sciencedirect.com/science/article/pii/S0092867414013105 [Accessed

May19,2017].

Fisher,R.A.,1929.TestsofsignificanceinHarmonicanalysis.Proc.Roy.Soc.LondonSer.A,

125,pp.54–59.

Gheldof, N. et al., 2012. Detecting Long-Range Chromatin Interactions Using the

Chromosome Conformation Capture Sequencing (4C-seq) Method. In pp. 211–225.

Available at: http://link.springer.com/10.1007/978-1-61779-292-2_13 [Accessed

December8,2016].

Gotic, I. et al., 2016.Temperature regulates splicingefficiencyof the cold-inducibleRNA-

binding protein gene Cirbp.Genes&development, 30(17), pp.2005–17. Available at:

http://www.ncbi.nlm.nih.gov/pubmed/27633015[AccessedNovember8,2016].

Gréchez-Cassiau, A. et al., 2008. The circadian clock component BMAL1 is a critical

regulator of p21WAF1/CIP1 expression and hepatocyte proliferation.The Journalof

biological chemistry, 283(8), pp.4535–42. Available at:

http://www.ncbi.nlm.nih.gov/pubmed/18086663[AccessedMay19,2017].


https://doi.org/10.1101/207787


24

Guillou,H.etal.,2010.Thekeyrolesofelongasesanddesaturasesinmammalianfattyacid

metabolism:Insightsfromtransgenicmice.ProgressinLipidResearch,49(2),pp.186–

199.

Hughes, M.E. et al., 2012. Brain-Specific Rescue of Clock Reveals System-Driven

Transcriptional Rhythms in Peripheral Tissue A. Kramer, ed. PLoS Genetics, 8(7),

p.e1002835. Available at: http://dx.plos.org/10.1371/journal.pgen.1002835

[AccessedOctober2,2017].

Hughes, M.E. et al., 2009. Harmonics of Circadian Gene Transcription in Mammals G. S.

Barsh, ed. PLoS Genetics, 5(4), p.e1000442. Available at:

http://dx.plos.org/10.1371/journal.pgen.1000442[AccessedDecember19,2016].

Iwafuchi-Doi, M. et al., 2016. The Pioneer Transcription Factor FoxA Maintains an

Accessible Nucleosome Configuration at Enhancers for Tissue-Specific Gene

Activation.MolecularCell,62(1),pp.79–91.

Jouffe,C.etal.,2013.Thecircadianclockcoordinatesribosomebiogenesis.PLoSBiol,11(1),

p.e1001455.Availableat:http://www.ncbi.nlm.nih.gov/pubmed/23300384.

Kelsall, I.R., Rosenzweig, D. & Cohen, P.T.W., 2009. Disruption of the allosteric

phosphorylase a regulation of the hepatic glycogen-targeted protein phosphatase 1

improvesglucosetoleranceinvivo.CellularSignalling,21(7),pp.1123–1134.

Kim, J.S. et al., 2016. Reciprocal localization of transcription factors YY1 and CP2c in

spermatogonial stem cells and their putative roles during spermatogenesis. Acta

Histochemica,118(7),pp.685–692.

Korenčič,A.etal.,2014.Timingofcircadiangenesinmammaliantissues.ScientificReports,

4,pp.1349–1354.Availableat:http://www.nature.com/articles/srep05782[Accessed

August14,2016].


https://doi.org/10.1101/207787


25

Kornmann,B.etal.,2007.System-drivenandoscillator-dependentcircadiantranscription

inmicewithaconditionallyactive liverclock.PLoSbiology,5(2),p.e34.Availableat:

http://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.0050034

[AccessedJuly23,2015].

Krishnaiah, S.Y. et al., 2017. Clock Regulation of Metabolites Reveals Coupling between

Transcription and Metabolism. Cell Metabolism, 25(4), p.961–974.e4. Available at:

http://linkinghub.elsevier.com/retrieve/pii/S1550413117301717 [Accessed October

2,2017].

Lamia,K.A.,Storch,K.-F.&Weitz,C.J.,2008.Physiologicalsignificanceofaperipheraltissue

circadianclock.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesof

America, 105(39), pp.15172–7. Available at:

http://www.pnas.org/content/105/39/15172.short[AccessedMay1,2015].

Lee, J.W., Chou, C.-L. & Knepper, M.A., 2015. Deep Sequencing in Microdissected Renal

TubulesIdentifiesNephronSegment-SpecificTranscriptomes.JournaloftheAmerican

Society of Nephrology : JASN, 26(11), pp.2669–77. Available at:

http://www.ncbi.nlm.nih.gov/pubmed/25817355[AccessedJuly27,2016].

Li,J.etal.,2015.ConsiderationsforRNA-seqAnalysisofCircadianRhythms.InMethodsin

enzymology. pp. 349–367. Available at:

http://www.ncbi.nlm.nih.gov/pubmed/25662464[AccessedOctober6,2017].

Li, X. & Noll, M., 1994. Compatibility between enhancers and promoters determines the

transcriptional specificity of gooseberry and gooseberry neuro in the Drosophila

embryo. The EMBO journal, 13(2), pp.400–6. Available at:

http://www.ncbi.nlm.nih.gov/pubmed/8313885[AccessedApril26,2017].

Liang, F. et al., 2008.Mixtures ofg Priors for Bayesian Variable Selection. Journalof the


https://doi.org/10.1101/207787


26

American Statistical Association, 103(481), pp.410–423. Available at:

http://www.tandfonline.com/doi/abs/10.1198/016214507000001337 [Accessed

June11,2016].

Le Martelot, G. et al., 2012. Genome-Wide RNA Polymerase II Profiles and RNA

Accumulation Reveal Kinetics of Transcription and Associated Epigenetic Changes

DuringDiurnalCyclesA.Kramer,ed.PLoSBiology,10(11),p.e1001442.Availableat:

http://dx.plos.org/10.1371/journal.pbio.1001442[AccessedFebruary22,2017].

LeMartelot, G. et al., 2009. REV-ERBalpha participates in circadian SREBP signaling and

bile acid homeostasis. PLoS biology, 7(9), p.e1000181. Available at:

http://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.1000181

[AccessedAugust8,2015].

Matsuo,T.etal.,2003.ControlMechanismoftheCircadianClockforTimingofCellDivision

in Vivo. Science, 302(5643). Available at:

http://science.sciencemag.org/content/302/5643/255[AccessedMay19,2017].

Matsuoka, T. et al., 2003.Members of the largeMaf transcription family regulate insulin

genetranscriptionin isletbetacells.Molecularandcellularbiology,23(17),pp.6049–

62. Available at: http://www.ncbi.nlm.nih.gov/pubmed/12917329 [Accessed

November3,2016].

Merkin,J.etal.,2012.EvolutionarydynamicsofgeneandisoformregulationinMammalian

tissues. Science (New York, N.Y.), 338(6114), pp.1593–9. Available at:

http://www.sciencemag.org/content/338/6114/1593.full[AccessedMay25,2013].

Merli, C. et al., 1996. Promoter specificity mediates the independent regulation of

neighboring genes. Genes & development, 10(10), pp.1260–70. Available at:



https://doi.org/10.1101/207787


27

Mermet, J., Yeung, J. & Naef, F., 2016. Systems Chronobiology: Global Analysis of Gene

Regulation in a 24-Hour PeriodicWorld. Cold SpringHarbor perspectives in biology,

p.a028720.Available at: http://www.ncbi.nlm.nih.gov/pubmed/27920039 [Accessed

February15,2017].

Morf, J. et al., 2012. Cold-Inducible RNA-Binding Protein Modulates Circadian Gene

ExpressionPosttranscriptionally.Science,338(6105).

Myšičková,A. et al., 2012.Detectionof interacting transcription factors inhuman tissues

usingpredictedDNAbinding affinity.BMCGenomics, 13(Suppl 1), p.S2.Available at:

http://bmcgenomics.biomedcentral.com/articles/10.1186/1471-2164-13-S1-S2

[AccessedNovember3,2016].

Nguyen, T.A. et al., 2016. High-throughput functional comparison of promoter and

enhancer activities. Genome Research, 26(8), pp.1023–1033. Available at:

http://genome.cshlp.org/lookup/doi/10.1101/gr.204834.116 [Accessed August 11,

2016].

Nikolaeva,S.etal.,2012.Thecircadianclockmodulatesrenalsodiumhandling.Journalof

the American Society of Nephrology : JASN, 23(6), pp.1019–26. Available at:

http://www.ncbi.nlm.nih.gov/pubmed/22440902[AccessedOctober28,2016].

Oosterveer,M.H.&Schoonjans,K.,2014.Hepaticglucosesensingandintegrativepathways

in the liver. Cellular andMolecular Life Sciences, 71(8), pp.1453–1467. Available at:

http://link.springer.com/10.1007/s00018-013-1505-z [Accessed December 14,

2016].

Pachkov, M. et al., 2013. SwissRegulon, a database of genome-wide annotations of

regulatorysites:recentupdates.Nucleicacidsresearch,41(Database issue),pp.D214-

20. Available at: http://nar.oxfordjournals.org/content/41/D1/D214 [Accessed


https://doi.org/10.1101/207787


28

August3,2015].

Ravnskjaer, K. et al., 2013. Glucagon regulates gluconeogenesis through KAT2B- and

WDR5-mediatedepigeneticeffects.JournalofClinicalInvestigation,123(10),pp.4318–

4328. Available at: http://www.jci.org/articles/view/69035 [Accessed November 3,

2016].

Refinetti,R.&Menaker,M.,1992.Thecircadianrhythmofbodytemperature.Physiology&

Behavior,51(3),pp.613–637.

Reinke, H. et al., 2008. Differential display of DNA-binding proteins reveals heat-shock

factor 1 as a circadian transcription factor.Genes& development, 22(3), pp.331–45.

Availableat:http://genesdev.cshlp.org/content/22/3/331.short[AccessedAugust10,

2015].

Rey, G. et al., 2011. Genome-Wide and Phase-Specific DNA-Binding Rhythms of BMAL1

ControlCircadianOutputFunctionsinMouseLiverA.Kramer,ed.PLoSBiology,9(2),

p.e1000595.Availableat:http://dx.plos.org/10.1371/journal.pbio.1000595[Accessed

April4,2017].

Ripperger, J.A. & Schibler, U., 2006. Rhythmic CLOCK-BMAL1 binding to multiple E-box

motifsdrivescircadianDbptranscriptionandchromatintransitions.NatureGenetics,

38(3),pp.369–374.Availableat:http://www.nature.com/doifinder/10.1038/ng1738

[AccessedAugust16,2016].

Sayin, S.I. et al., 2013. Gut Microbiota Regulates Bile Acid Metabolism by Reducing the

Levels of Tauro-beta-muricholic Acid, a Naturally Occurring FXR Antagonist. Cell

Metabolism,17(2),pp.225–235.

Shea,H.C.etal.,2007.AnalysisofHSD3B7knockoutmice reveals thata3alpha-hydroxyl

stereochemistryisrequiredforbileacidfunction.ProceedingsoftheNationalAcademy


https://doi.org/10.1101/207787


29

of Sciences of the United States of America, 104(28), pp.11526–33. Available at:

http://www.ncbi.nlm.nih.gov/pubmed/17601774[AccessedDecember14,2016].

Shen, N. et al., 2015. The Constitutive Activation of Egr-1/C/EBPa Mediates the

DevelopmentofType2DiabetesMellitusbyEnhancingHepaticGluconeogenesis.The

American Journal of Pathology, 185(2), pp.513–523. Available at:

http://linkinghub.elsevier.com/retrieve/pii/S0002944014006105 [AccessedMay13,

2017].

Sinturel, F. et al., 2017. Diurnal Oscillations in Liver Mass and Cell Size Accompany

Ribosome Assembly Cycles. Cell, 169(4), p.651–663.e14. Available at:

http://linkinghub.elsevier.com/retrieve/pii/S0092867417304282 [Accessed May 4,

2017].

Sobel, J.A. et al., 2017. Transcriptional regulatory logic of the diurnal cycle in themouse

liver A. Kramer, ed. PLOS Biology, 15(4), p.e2001069. Available at:

http://dx.plos.org/10.1371/journal.pbio.2001069[AccessedApril18,2017].

Storch, K.-F. et al., 2002. Extensive and divergent circadian gene expression in liver and

heart. Nature, 417(6884), pp.78–83. Available at:

http://dx.doi.org/10.1038/nature744[AccessedApril10,2015].

Stumvoll, M. et al., 1998. Human kidney and liver gluconeogenesis: evidence for organ

substrate selectivity.American JournalofPhysiology -EndocrinologyandMetabolism,

274(5).

Takahashi, J.S. et al., 2008. The genetics of mammalian circadian order and disorder:

implications forphysiologyanddisease.NatureReviewsGenetics,9(10),pp.764–775.

Available at: http://www.nature.com/doifinder/10.1038/nrg2430 [Accessed June 9,

2017].


https://doi.org/10.1101/207787


30

Vollmers,C.etal.,2012.Circadianoscillationsofprotein-codingandregulatoryRNAsina

highly dynamic mammalian liver epigenome. Cell metabolism, 16(6), pp.833–45.

Availableat:http://www.sciencedirect.com/science/article/pii/S1550413112004573

[AccessedSeptember15,2014].

Wang,J.etal.,2017.NuclearProteomicsUncoversDiurnalRegulatoryLandscapesinMouse

Liver. Cell Metab, 25(1), pp.102–117. Available at:

http://www.ncbi.nlm.nih.gov/pubmed/27818260.

Wang, L. et al., 2014. MACE: model based analysis of ChIP-exo. Nucleic acids research,

42(20), p.e156. Available at: http://nar.oxfordjournals.org/content/42/20/e156

[AccessedJuly8,2015].

vandeWerken,H.J.G.etal.,2012.Robust4C-seqdataanalysistoscreenforregulatoryDNA

interactions. Nature Methods, 9(10), pp.969–972. Available at:

http://www.ncbi.nlm.nih.gov/pubmed/22961246[AccessedJune6,2017].

Xu, Y. et al., 2016. Long-Range Chromosome Interactions Mediated by Cohesin Shape

CircadianGeneExpressionA.Kramer,ed.PLOSGenetics,12(5),p.e1005992.Available

at:http://dx.plos.org/10.1371/journal.pgen.1005992[AccessedSeptember29,2017].

Yan, J. et al., 2008. Analysis of Gene Regulatory Networks in the Mammalian Circadian

Rhythm.PLOSComputBiol,4(10),pp.647–676.

Yue, F. et al., 2014. A comparative encyclopedia of DNA elements in themouse genome.

Nature,515(7527),pp.355–364.Availableat:http://dx.doi.org/10.1038/nature13992

[AccessedNovember19,2014].

Zabidi,M.A.etal.,2014.Enhancer–core-promoterspecificityseparatesdevelopmentaland

housekeeping gene regulation. Nature, 518(7540), pp.556–559. Available at:



https://doi.org/10.1101/207787


31

Zhang, R. et al., 2014. A circadian gene expression atlas in mammals: implications for

biology andmedicine. Proceedings of theNational Academy of Sciences of theUnited

States of America, 111(45), pp.16219–24. Available at:

http://www.pnas.org/content/111/45/16219.short[AccessedJuly21,2015].

Zhang,Y.etal.,2015.DiscretefunctionsofnuclearreceptorRev-erbαcouplemetabolismto

the clock. Science (New York, N.Y.), 348(6242), pp.1488–1492. Available at:

http://www.sciencemag.org/content/348/6242/1488.full[AccessedJune10,2015].


https://doi.org/10.1101/207787


Figure1–Contributionoftissue,dailytime,andcircadianclocktoglobalvariancein

mRNAexpression

(A)Principalcomponent(PC)analysisoftwodaystemporaltranscriptomesacross

11 WT tissues. PC1 and PC2 show clustering of samples by tissues; each point

representsatissuesample(seelegend)ataspecifictimepoint(notlabeled).Inset:

LoadingsforPC13andPC17fortheliversampleslabeledwithcircadiantime(CT),

showing temporal variationalonganellipticpath.Colors:CT;CT0 corresponds to

subjectivedawn;CT12correspondstosubjectivedusk.

(B) Fractions of temporal variance in each tissue explained by 24- and 12-hour

periods,obtainedbyapplyingspectralanalysisgenome-wideforeachtissue.Dotted

horizontallinesrepresenttheexpectedbackgroundlevel,assumingwhitenoise.

(C,D) Cumulative number of rhythmic genes (p<0.01, harmonic regression) with

log2foldchangelargerthanthevalueonthex-axis.(C)Analysison11WTtissues.

(D)Analysison4conditions:Bmal1KOmiceandWTlittermatesinliverandkidney.


https://doi.org/10.1101/207787


Figure 2 – Combinatorics of rhythmic transcript expression across tissues and

genotypes

(A) Schema for the model selection (MS) algorithm to identify rhythmic gene

expression modules across tissues. Temporal transcriptomes of different tissues

represented as a 3-dimensional array (left). Gene modules are probabilistically

assigned amongst different combinations of 24-hour rhythms across tissues (e.g.

tissue-specificortissue-widerhythmsschematicallyshownonright).

(B) Gene modules are summarized by the first component of complex-valued

singular value decomposition (SVD) to highlight phase (peak time shown as the

clockwise angle) and amplitude (log2 fold change shown as the radial distance)

relationshipsbetweengenes(genespace)andbetweentissues(tissuespace).SVD

representation is scaled such that the genes show log2 fold changes,while tissue

vectorsarescaledsuchthatthehighestamplitudetissuehaslengthof1andaphase

offsetof0hours.

(C-E)MSappliedto11WTtissues.

(F,G)MSappliedtoBmal1KOandWTlittermatesinliverandkidney.

(C)SVDrepresentationoftissue-widemRNArhythmsfromthe11tissues.Genesare

labeledassystem-driven(blue)orclock-driven(red)accordingtothecomparison

ofthecorrespondingtemporalprofilesinBmal1KOandWTlittermates(Methods).

(D) Examples of anti-phasic rhythms (brown fat and muscle, n=20, first SVD

component explains 81% of variance), and tissue-specific rhythms (liver, n=846,

first SVD component explains 59% of variance). Representative genes with large

amplitudesarelabeled.


https://doi.org/10.1101/207787


(E)Numberof transcriptsshowingrhythms(p-value<0.01,harmonicregression)

indifferentnumbersoftissues,infunctionofincreasingpeaktotroughamplitudes

on the x-axis. X-axis: average log2 fold change calculated from the identified

rhythmictissues.

(F)SVDrepresentationofclock- (top,n=991,83%ofvariance)andsystem-driven

(bottom,n=1395,84%ofvariance)liver-specificrhythms.

(G)Numberoftranscriptsshowingclock-(solid)orsystem-driven(dotted)rhythms

(p-value < 0.01, harmonic regression) in liver (red), kidney (blue), or both

(magenta).


https://doi.org/10.1101/207787


Figure3-OscillatoryTFactivity inonetissuebutnototherscandrivetissue-specific

rhythms

(A) Module describing system-driven liver-specific rhythms (n=1395, first SVD

component explains 84% of variance). Radial coordinate of the colored polygons

represent enrichment of the indicated GO terms at each time point, obtained by

comparingthegenesfallinginaslidingwindowof+/-3hourstothebackgroundset

ofall1395genesassignedtomodule(p-valuecomputedfromFisher’sexacttest).

(B)MAFBisacandidateTFforthemoduleinA.PredictedMAFBactivity(solidline),

nuclearproteinabundance(triangles),andmRNAaccumulation(dotted)oscillatein

WTandBmal1KO,withpeakmRNAprecedingpeaknuclearproteinandTFactivity.

Errorbarsinnuclearprotein,mRNA,andTFactivityshowSEM(n=2).

(C)Clock-drivenkidney-specificmodule(n=156,firstSVDcomponentexplains80%

ofvariance).Coloredpolygonsasin(A).

(D)TFCP2isacandidateTFforthemoduleinC.Thetemporalprofileofpredicted

TFCP2activity(solidline)isanti-phasicwithTfcp2mRNAaccumulation(dotted)in

WT,andbothareflat inBmal1KO.Errorbars inmRNAandTFactivityshowSEM

(n=2).

(E)Clock-drivenliver-specificmodule(n=991,firstSVDexplains83%ofvariance).

(F)ELFisacandidateTFforthemoduleinE.ThetemporalprofileofpredictedELF

activity (solid line) in WT matches that of nuclear protein abundance in liver

(triangles),andbotharedelayedcomparedtoElf1mRNAaccumulation(dotted).In

Bmal1 KO, ELF activity and Elf1mRNA are nonrhythmic. Error bars in nuclear

protein,mRNA,andTFactivityshowSEM(n=2).


https://doi.org/10.1101/207787


Figure 4 – Co-localized binding of clock and liver-specific TFs underlies liver-specific

mRNArhythms

(A) The fraction of genes containing liver-specific DNase-I hypersensitive sites

(DHSs) in the clock-driven liver-specific module is higher compared with both

nonrhythmic and system-driven liver-specific modules. Error bars and p-values

calculatedfrom10000bootstrapiterations.

(B) Predicted temporal activities of RORE (top) and E-box (bottom) TF motifs

located within liver-specific DHSs. Error bars show standard deviation of the

estimatedactivities.

(C)Co-occurrenceofROREwithallotherTFsintheSwissRegulondatabase(189TF

motifs).Positive log10oddsratios (ORs) representpairsofmotifsenriched in the

clock-driven liver-specific module compared to the flat module. P-values for the

motif pairs were calculated from chi-square tests applied to 3-way contingency

tables(Myšičkováetal.2012).Selectedpairsareinbold.

(D) DNase-I hypersensitivity in liver, kidney, and the corresponding differential

signal (in log2 fold change) near two representative genes (top: Insig2; bottom:

Slc4a4).RORE,ONECUT1,andFOXATFbindingmotifs(posteriorprobability>0.5,

MotEvo) co-occur at liver specific DHSs (red boxes). Predicted TF binding sites

correspond to experimentally observed TF binding in publicly available ChIP-exo

datasetsforREV-ERBa,ONECUT1,andFOXA2(bottom).


https://doi.org/10.1101/207787


Figure5–Liver-specificchromatinloopsregulateliver-specificmRNArhythms

(A)TemporalmRNAprofileforMreg,aclock-drivenliver-rhythmicgene.Errorbars

areSEM(n=2).

(B)4C-Seqprofiles(summaryfrom2replicates,eachpooling2differentmice)using

theMreg promoter as a bait in liver and kidney at ZT20. Data are shown in a

windowof+/-250kbfromthebait(top).Profilesofdifferentialcontactsbetween

liver and kidney (bottom) represented as signed log p-values (regularized t-test,

positivevaluesdenoteliver-enriched4Ccontacts).

(C) Tracks of differential 4C contacts (signed log p-values), log2 fold change of

DNase-Ihypersensitivitybetweenliverandkidney,andChIP-exoofREV-ERBaand

FOXA2. Regions of significant differential 4C contacts correspond to liver-specific

DNase-IhypersensitiveregionsandREV-ERBabindingsites.


https://doi.org/10.1101/207787


Figure6-Precisepromoter-enhancercontactsunderlieliver-specificmRNArhythms

(A,B)4C-Seqprofilesforthe(A)Slc45a3-shortand(B)Slc45a3-longisoformswithin

+/- 250 kb around baits targeting the two TSSs (top). Signed log p-values for

differential contacts between liver and kidney (bottom) as in Figure 5B. TSSs for

Slc45a3-shortandSlc45a3-longare8kbapart.Yellowarrowsdenote liver-specific

distalcontactsfoundattheSlc45a3-shortbutabsentattheSlc45a3-longTSS.

(C) Differential 4C contacts (signed log p-values), log2 fold change of DNase-I

hypersensitivity between liver and kidney, and ChIP-exo signal of REV-ERBa,

FOXA2, andONECUT1.Regionsof significantdifferential contacts inSlc45a3-short

correspondtoliver-specificDNase-Ihypersensitiveregions.

(D) Schematic model illustrating enhancer-promoter interactions in liver and

kidneythatmaygenerateliver-specificrhythms.Yellowcirclesillustrateliver-active

enhancers contacting the rhythmic promoter (red arrow) but not the alternative

nonrhythmic promoter (grey). In kidney, the enhancer is not accessible and both

promotersarenonrhythmic.


https://doi.org/10.1101/207787


PC13

PC17

A

B

Figure 1

C D

1

10

100

1000

# Ge

nes

LiverKidneyWTBmal1 KO

0.0

0.1

0.2

Liver

BFAT

Kidn

eyLu

ng Adr

Mus

Hear

tAo

rtaHy

poCe

re BSFrac

tion

of te

mpo

ral v

arian

ce Period [h] 12 24

1

10

100

1000

0 1 2 3 4 5Log fold change

# Ge

nes

LiverKidneyBFATLungMus

AdrAortaHeartCereBSHypo

−100 −50 0 50 100 150

−150

−100

−50

050

100

PC1

PC2

Brain regions

, ,, ,, ,,,, ,,, ,,,,,, ,,,,,, ʻ̒̒ʻʻ ʻʻʻʻ ʻ̒ʻʻ ʻʻʻʻʻ̒̒̒̒̒ʻ

Adrenal (Adr)

AortaBrown fat (BFAT)

Brain stem (BS)Cerebellum (Cere)HeartHypothalamus (Hypo)

KidneyLiver

Lung

Muscle (Mus) ,ʻ

0 1 2 3 4 5Log fold change22

18

2022

2426

28

3032

34

36

38

40

42

444648

50

52

54

56

5860

62

64


https://doi.org/10.1101/207787


1

10

100

1000

0 2 4 6

# Ge

nes

1

10

100

1000

0 2 4 6

# Ge

nes

gene

timetis

sue

...

Gene space Tissue space

A

F

G

...

Rgs16

Leap2Ube2u

12

24

Liver

12

24

Tnnt1Myl3

12

24

BFAT

Mus

12

24

6

12

18

0

6

12

18

0

x

Selection of modules

Complex-valued SVD

......

...

Tissue space

# Rhythmictissues

12−34−78−11

...

Liver-specific

Kidney-specificShared

Clock−drivenSystem−driven

1

0

1

0

6

0

4.5

0

5.5

0

Log2

fold

cha

nge

Scal

e fa

ctor

CT

618 6

18618

618

Time of day [h] Offset [h]

x

x

x

B

Nr1d1

ArntlNpas2

Dbp

Nr1d2

Per2

6

12

18

24

Tissue SpaceGene Space

Per3

Per1

Bhlhe41

Nfil3

Lonrf3

Cdkn1a

Wee1

Cirbp

Hspa8

Gene space

CT Offset [h]

�

�

�

��

�

�

�

�

�

�

Adr

AortaBFAT

BstemCereHeart

Hypo

Kidney

LiverLung

Mus

6

12

18

24

0

1

Scale

facto

r

Offset [h]

D

C

ZT Offset [h]

Rgs16Lipg

Gck6

12

18

24Pitx3 Liver WT

6

12

18

24

Mfsd2a Lpin1Tsku

Hsph1 Fbxo216

12

18

24Liver WTLiver Bmal1 KO

6

12

18

24

x

x

4

0

5

0

1

0

1

0

Log2

fold

cha

nge

Scal

e fa

ctor

E

Clock-drivenSystem-driven

Figure 2

2Avg log fold change2Avg log fold change

Log2

fold

cha

nge


https://doi.org/10.1101/207787


Gene: Elf1TF: ELF1

A

B

C

D

6

12

18

24

0

4

DNA replication initiationNegative response to insulinRibosome biogenesis

6

12

18

24

0

2.5

AAAAA

Organic anion transportSodium ion transport

Log

fold

chan

ge o

r -lo

g10(

p-va

lue)

Figure 3

Lpin1

Fgf21

Egr1

MafbJun

Nop56Pik3r1

Cebpb

Rpp38Nop58

Ddit3Erbb4

Mcm2

Cpeb2

Sorbs1

Mcm5Mcm4Mcm3

Sesn2Pparg

Kat2b

PmlPrkcd

Foxo1

PtprfPPP kkckkk

Mcm2McmMMm4m4

m333McmcmMcmm

PtprfPtptptpprprrfprpprfppprfprprfrfrfrfrfrfffrfrffrffffffff

gCpeb2b2CC

S

m5m5m5

SppppargpPparggpapPpPpPpPpPPPPPPPpargmlmlmmlmmmmmm

cdcdcdddccdccckckckckcddcc

McmcmMcMMPmmPmPmmPPmPmPmPPPPP

Slc16a1

Rhobtb1Trib2

Slc41a1

Slc6a4Prnp

Slc7a8Slc39a5

Slc22a4

Clcn2

Igf1r

Slc9a3

Slc12a6 Tfcp2Spred3Lrrc52

Angpt1

2Slc7a8Slc a8SSSSSS

55555 Angpt1AnnggpgpAngpAnA gAnAAAAAAAA

S

a444Slc6a6ac6a Slc22a4

22Clcncnnlc9a39a3c9SlcclcSSSlc9

TfTTed33333d3d3prredde

ZT ZT

Abun

danc

e or

mot

if acti

vity

(sca

led)

Gene: Tfcp2TF: TFCP2

Abun

danc

e or

mot

if acti

vity

(sca

led)

E

F

Abun

danc

e or

mot

if acti

vity

(sca

led)

mRNA Nuclear protein TF activity

Gene: MafbTF: MAFB

0

5

Cellular carbohydrate biosynthetic processCellular lipid metabolic processNucleotide metabolic process

Rgs16

Elovl3

Ppp1r3b

Tff3Upp2

Insig2

Nrg4Lipg

GckPor Abcg5

Hsd3b7Slc45a3Mreg

Pik3ap1

Slc4a4Cpeb4Slc44a1

Pklr

Elf1Colgalt2p1r3bbS44a1

PklrP rlr

lflflfEEEColgalt2gaalt2alt2lt22t2t2t222

Pik3Pikk3ikk3ikPikik3k3k33kk333a33333333333

llllllElElEEEEEEEEEEEEEE 6

12

18

24

WT Bmal1 KOWT Bmal1 KO

ZT

WT Bmal1 KO

−2−1

012

0 6 12 18 24 0 6 12 18 24ZT

−2

−1

0

1

0 6 12 18 0 6 12 18ZT

mRNA TF activity

−3−2−1

012

0 6 12 18 24 0 6 12 18 24ZT

mRNA Nuclear protein TF activity

2

Log

fold

chan

ge o

r -lo

g10(

p-va

lue)

2

Log

fold

chan

ge o

r -lo

g10(

p-va

lue)

2

AAAAAAAAAA


https://doi.org/10.1101/207787


−0.10

−0.05

0.00

0.05

0.10

0 12 24 0 12 24ZT

Activ

ity [A

.U.]

E-box

−0.2

−0.1

0.0

0.1

0 12 24 0 12 24ZT

Activ

ity [A

.U.]

ROREAFigure 4

D

ONECUT1FOXA2

RORE

1 kb mm9123,229,500 123,230,000 123,230,500 123,231,000 123,231,500 123,232,000 123,232,500 123,233,000

Insig21 -

0 _1 -

0 _1.5 -

0 _

34 -

0.3 _66 -

0.1 _44 -

0.6 _

ONECUTFOXA2

RORA

1 kb mm989,366,500 89,367,000 89,367,500 89,368,000 89,368,500 89,369,000 89,369,500 89,370,000

Slc4a4

1 -

0 _1 -

0 _1.8 -

-0.6 _

55 -

0.9 _11 -

0.1 _19 -

0.6 _

DNase-I

ChIPFOXA2

ONECUT1

REVERBa

TF motifs

DNase-I

ChIPFOXA2

ONECUT1

REVERBa

TF motifs

0.0

0.1

0.2

0.3

Clock−d

riven

System−dr

iven

Nonrhyt

hmic

Frac

tion

of g

enes

B

Liver KidneyWT Bmal1 KO

C

p < 0.001 FOXA2;ROREFOXD3;RORE

ONECUT;RORE RORE;CUX2

RORE;GATA6RORE;TFAP2

2

3

4

5

6

7

0.0 0.2 0.4Log10 odds ratio compared to bg

−log

(pv

alue)

10


https://doi.org/10.1101/207787


B

C

0.00.20.40.60.8

0

510

log 4

C sig

nal

−log

(pva

lue)

(liver

- kid

ney)

Distance relative to bait (kb)0 100 200-100-200

0 100 200-100-200

ZT20 WT

Mreg

LiverKidney

Log

mRN

A ab

unda

nce

� �Liver Kidney

Mreg

AFigure 5

100 kb mm972,050,000 72,100,000 72,150,000 72,200,000 72,250,000 72,300,000 72,350,000

1.7 -

-0.9 _20 -

-5 _ Mreg

PecrXrcc5

DNase-I

ChIP

4C

��

��

�

�

� � �

��

2

4

6

0 5 10 15 20

REVERBa

FOXA2

ZT10

10

2


https://doi.org/10.1101/207787


A

C

Slc45a3-short

0.0

0.5

1.0

−4

0

4

8

−200 −100 0 100 200

−200 −100 0 100 200Position relative to bait (kb)

0.0

0.5

1.0

1.5

−4

0

4

8

−200 −100 0 100 200

−200 −100 0 100 200Position relative to bait (kb)

D

Slc45a3-long

B

LiverKidney

Figure 6

100 kb mm9133,800,000 133,850,000 133,900,000

Slc41a1 Rab7l1 Nucks1Elk4

Mfsd41.5 -

-0.8 _20 -

-5 _20 -

-5 _

DNase-I

4C Short

LongSlc45a3-short

Slc45a3-long

ChIP

REVERBa

FOXA2

ONECUT1

log 4

C sig

nal

10−lo

g (p

value

)(liv

er -

kidne

y)10

log 4

C sig

nal

10−lo

g (p

value

)(liv

er -

kidne

y)10


https://doi.org/10.1101/207787


Transcription factor activity rhythms and tissue-specific ... · and lows in body temperature...

Documents

Transcript of Transcription factor activity rhythms and tissue-specific ... · and lows in body temperature...