Published online 19 Decembe 2016 Nucleic Acids Resea ch, 2017, Vol. 45, No. 7 3693–3706
doi: 10.1093/na /gkw1283
Inducible supe -enhance s a e o ganized based on
canonical signal-speci ic ansc ip ion ac o binding
elemen s
D´
o a Bojcsuk1, Ge gely Nagy2and Balin L. Balin 1,*
1Depa men o Biochemis y and Molecula Biology, Genomic Medicine and Bioin o ma ic Co e Facili y, Uni e si y o
Deb ecen, Deb ecen 4032, Hunga y and 2Depa men o Biochemis y and Molecula Biology, MTA-DE ‘Lendule ’
Immunogenomics Resea ch G oup, Uni e si y o Deb ecen, Deb ecen 4032, Hunga y
Recei ed Augus 12, 2016; Re ised Decembe 07, 2016; Edi o ial Decision Decembe 08, 2016; Accep ed Decembe 13, 2016
ABSTRACT
Supe -enhance s a e es ablished h ough he in e -
ac ions o se e al enhance s and a la ge numbe
o p o eins, including ansc ip ion ac o s and co-
egula o s; howe e , he o ma ion o hese in e -
ac ions is poo ly unde s ood. By e-analysing p e-
iously published es ogen ecep o alpha (ER␣)
ChIP-seq da a se s de i ed om he MCF-7 cell
line, we obse ed ha in he absence o s imu-
la ion, u u e supe -enhance s a e ep esen ed by
one o a ew ansc ip ion ac o binding e en (s)
and hese ex ao dina y enhance s possess a e-
sponse elemen la gely speci ic o he ER␣dime .
Upon ho monal s imula ion, hese p ima y binding
si es a e su ounded by a la ge amoun o ER␣and
he c i ical componen s o ac i e enhance s, such
as P300 and MED1, and oge he wi h neighbou ing
si es bound by newly ec ui ed ER␣, hey gene a e
he unc ional supe -enhance s. To u he alida e
he ole o canonical elemen s in supe -enhance
o ma ion, we in es iga ed some addi ional signal-
dependen ansc ip ion ac o s, con i ming ha ce -
ain, dis inguished binding elemen s ha e a gene al
o ganize unc ion. These esul s sugges ha ce -
ain signal-speci ic ansc ip ion ac o s guide supe -
enhance o ma ion upon binding o s ong esponse
elemen s. These indings may eshape he cu en
unde s anding o how hese egula o y uni s assem-
ble, highligh ing he in ol emen o DNA elemen s
ins ead o p o ein–p o ein in e ac ions.
INTRODUCTION
Es ogen ecep o alpha (ER␣) is a well-s udied mem-
be o he nuclea ecep o (NR) supe amily and is a
key ho mone- egula ed ansc ip ion ac o (TF) in h ee-
qua e s o b eas cance cases (1). The ac i a ion o ER␣
h ough 17-es adiol ( e e ed o as es adiol o E2) e-
sul s in he ansloca ion o he ecep o om he cy oplasm
o he nucleus and s eng hens he binding o he ecep-
o and i s collabo a ing ac o s o egula o y egions o
DNA. These e en s con ibu e o he al e ed, pa icula ly
inc eased exp ession o he speci ic a ge genes (2). ER␣
p ima ily binds he es ogen esponse elemen (ERE), ini-
ially desc ibed as a conse ed egula o y elemen o he
p omo e o Xenopus and chicken i ellogenin genes in 1986
(3). ERE is ypically an in e ed epea o he NR hal
si e (AGGTCA) ha is sepa a ed by a 3-bp-long space
(IR3). The e a e se e al a ia ions o his elemen , which
also binds he ligand-induced ER dime (4).
The mos widely used human cell line o s udy he beha -
io o ER␣is MCF-7. MCF-7 is a well-es ablished in i o
model o he in es iga ion o es ogen-dependen biologi-
cal p ocesses o b eas cance de elopmen , as his model
is an ER+b eas cance -de i ed cell line isola ed om he
pleu al e usion o a pa ien wi h me as a ic b eas cance
(5,6).
The e a e se e al pu a i e ER␣ ansc ip ion ac o bind-
ing si es (TFBSs) in he e e ence human genome, bu only a
po ion o hese si es is unc ionally ele an . Using gene al
bioin o ma ic mo i scans, mo e han 1 million o hese pu-
a i e si es we e iden i ied, while he numbe o op imal o
nea -op imal sequence mo i s was p edic ed o be app oxi-
ma ely 230 000 acco ding o p e ious s udies (7,8). Among
hese, only a small ac ion, 5105 elemen s, was bound by
ER␣in MCF-7 cells, while app oxima ely 11 000 addi ional
ER␣binding e en s we e iden i ied wi hou he p esence o
any ERE in he DNA. The disc epancy be ween he pu a-
i e ER␣binding si es and hose iden i ied in expe imen al
models may e lec he accessibili y o he si es. Ch oma in
accessibili y has been sugges ed o d i e, in gene al, he
si e selec ion o TFs and ch oma in egions wi h inc eased
DNase I hype sensi i i y we e indeed en iched in cell line-
and issue-speci ic TFBSs (9). The eme gence o hese acces-
sible si es may be p imed by cell line-speci ic ac o s. Mo e-
*To whom co espondence should be add essed. Tel: +36 52 411717 (Ex . 50015); Fax: +36 52 314 989; Email: [email p o ec ed].hu
C
The Au ho (s) 2016. Published by Ox o d Uni e si y P ess on behal o Nucleic Acids Resea ch.
This is an Open Access a icle dis ibu ed unde he e ms o he C ea i e Commons A ibu ion License (h p://c ea i ecommons.o g/licenses/by-nc/4.0/), which
pe mi s non-comme cial e-use, dis ibu ion, and ep oduc ion in any medium, p o ided he o iginal wo k is p ope ly ci ed. Fo comme cial e-use, please con ac
jou nals.pe [email protected]
3694 Nucleic Acids Resea ch, 2017, Vol. 45, No. 7
o e , p iming e en s a he ch oma in le el u he p epa e
he con o ma ional con ex o subsequen p o ein–DNA
and p o ein–p o ein binding e en s ha ul ima ely lead o
ansc ip ional ac i a ion.
The majo collabo a i e ac o s o ER␣a e Fo khead
box p o ein A1 (FoxA1) and ac i a o p o ein 2 gamma
(AP2␥)(10,11). FoxA1 plays a ole as a pionee ac o , and
i has been p oposed ha FoxA1 binds ∼50% o he egions
occupied by ER␣, e en in he absence o es adiol (12–14).
Mo eo e , FoxA1 is indispensable o any ER␣ ec ui men
(15). AP2␥also joins o ER binding si es and may be in-
ol ed in s abilizing ER␣-ch oma in in e ac ions (10).
Recen s udies ha e sugges ed ha enhance s may o m
a highe -o de s uc u e wi hin egula o y uni s, also called
supe -enhance s (SEs) (16). Se e al epo s ha e sugges ed
ha he exp ession o cell ype-speci ic genes is con olled
h ough hese ma kedly ac i e enhance g oups. Thus, el-
e an genes ha e been published wi h espec o di e en
ypes o cells, wi h genes de ining he na u e o a gi en
cell (17). In cance cells, SEs a e associa ed wi h key onco-
genes, such as c-MYC (16,18) and TAL1 (19); he e o e, he
genome-wide de e mina ion o SE-associa ed genes would
be use ul o mapping addi ional de e mina i e genes ha
may be in ol ed in umo pa hogenesis (20). Thus, an un-
de s anding o how hese SEs wo k would be use ul o
b idging he gaps in cell di e en ia ion and umo igenesis.
Mo eo e , an unde s anding o he mechanisms unde lying
he manne in which SEs in luence he gene ic p og ammes
o a cell could b idge he pheno ype–geno ype gap as he
ounda ion o he missing he i abili y.
Acco ding o a cu en app oach o HOMER bioin o -
ma ic analysis sui e (21), ac i e SEs a e de ined based on
he ele a ed local sequence ead densi y o such genomic
egions whe e he indi idual ChIP-seq (ch oma in immuno-
p ecipi a ion ollowed by high- h oughpu DNA sequenc-
ing) peaks o he key TFs a e close o each o he han 12.5
kb. Each ChIP-seq peak o g oup o peaks is gi en a ead
densi y sco e, so ed based on sco e alues and hose ha
a e loca ed on he egion o he g aph wi h a slope >1a e
conside ed SEs.
Al hough ele an da a ha e accumula ed conce ning
how hese SEs unc ion, li le is known ega ding he de-
ailed mechanisms ha d i e he ac i i ies o hese com-
plexes. SEs a e bound by mas e egula o s (17,22), and
hei p edic ion can also be done based on he simul aneous
p esence o BRD4 and/o he ac i e his one ma k H3K27ac
(23,24).
The Media o complex, which is a key componen o
ansc ip ion ini ia ion, is in ol ed in he looping o en-
hance s o ansc ip ion s a si es (25,26). The la ges mem-
be o he Media o complex, MED1, binds o se e al nu-
clea ecep o s, such as hy oid ho mone ecep o (TR), i-
amin D ecep o (VDR), pe oxisome p oli e a o -ac i a ed
ecep o gamma (PPAR␥), hepa ocy e nuclea ac o 4 al-
pha (HNF4␣), ER and glucoco icoid ecep o (GR), and
i is also a co- egula o o SEs (17,27). Ano he subuni o
he complex, MED14, also in e ac s wi h GR, HNF4␣and
PPAR␥(27). P e ious s udies ha e demons a ed ha he
educed le el o Media o membe s also a ec s he exp es-
sion o lineage-speci ic genes, e lec ing he absence o majo
TFs (28). This e ec is consis en wi h he ole o MED1 in
o ming a b idge be ween collabo a ing p omo e s and en-
hance s h ough DNA loop o ma ion (29).
In he p esen s udy, using p e iously published and pub-
licly a ailable ChIP-seq da a se s, we conduc ed a de ailed
examina ion o how ER␣-d i en SEs a e o med. We com-
pa ed he si es wi h he highes le el o ER␣binding ha
upon es adiol ea men , oge he wi h he newly ac i a ed
egions, we e o ganized in o SEs. Unexpec edly, we iden i-
ied a subse o he enhance s wi hin SE clus e s ha plays
dis inc oles in he o ganiza ion o hese unc ional uni s,
and only hese enhance s possess canonical EREs. Mo e-
o e , hese enhance s ha e a di e en p o ile o p o ein–
p o ein in e ac ions upon ligand ea men . Fu he exam-
ina ion o he in e dependence o ER␣, FoxA1 and AP2␥
showed ha di e en TFs ha e speci ic subse s o SEs, in
which he coope a i e binding e en s a e inciden al.
To u he alida e he ole o canonical elemen s in SE
o ma ion, we in es iga ed se e al addi ional NRs and TFs,
con i ming he pi o al ole o s ong binding si es in SE o -
ma ion.
MATERIALS AND METHODS
Da a selec ion
We in es iga ed se en TFs in i e di e en human and
mouse cell ypes: ER␣, FoxA1 and AP2␥in he MCF-7
cell line (10,30,31), and ogen ecep o (AR) in he p os a e
cance -de i ed LNCaP cell line (32), JUNB in p ima y bone
ma ow-de i ed mac ophages (BMDM) o he C57BL/6
mouse s ain (33), VDR in mouse in es inal epi helial cells
and RAR in he F9 mouse emb yonic es is ca cinoma cell
line (34). Con ol and ligand- ea ed ChIP-seq samples o
he abo e-men ioned TFs we e selec ed om he publicly
a ailable gene exp ession omnibus da abase (35) (Supple-
men a y Table S1).
We selec ed se en addi ional E2- ea ed ER␣ChIP-seq
samples (8 in o al, om which a leas 10 000 ER␣TFBSs
we e p edic ed) o u he compa ison (Supplemen a y Ta-
ble S2) (30,36–40). P300, BRD4, MED1 and H3K27ac
ChIP-seq and DNase-seq expe imen s we e also pe o med
as ollows (Supplemen a y Table S3) (41–43). Fo s eng h-
ening ou model, se e al u he se s samples o ER␣ChIP-
seq expe imen s we e included in ou analysis ha co e :
ehicle- ea ed, un ea ed samples, ehicle- and E2- ea ed
FoxA1 knock-down samples, amoxi en- and ul es an -
ea ed samples as well as a ime-cou se expe imen o E2-
ea men (Supplemen a y Table S3) (8,31,44–47).
Da a p ocessing
Raw sequence iles o he selec ed ChIP-seq samples we e
downloaded and p ocessed using a p e iously published
compu a ional pipeline wi h he hg19 and mm10 e e -
ence genomes (48). ChIP-seq peaks we e p edic ed using
MACS2 (49), and SEs we e p edic ed om he ligand-
ea ed samples using HOMER indPeaks (21). A i ac s,
based on he blacklis ed genomic egions o he Encyclo-
pedia o DNA Elemen s (50), we e emo ed om he peak
se s using BEDTools (51).
Reads pe kilobase pe million mapped eads (RPKM)
alues o bo h he con ol and ligand- ea ed samples we e
Nucleic Acids Resea ch, 2017, Vol. 45, No. 7 3695
calcula ed on he summi ±50 bp egion o he peaks de-
e mined om he co esponding ligand- ea ed samples.
Peaks wi h he highes ead densi y ( e e ed o as ‘mo he ’
peaks) o each u u e SE we e selec ed om he con ol
samples. The eme ging SE peaks ( e e ed o as ‘daugh e ’
peaks) we e iden i ied in he ligand- ea ed samples in a sub-
sequen analysis. The de ailed in o ma ion abou he si es
o p edic ed SEs, mo he and daugh e enhance s was p e-
sen ed in Addi ional ma e ial (h p://dx.doi.o g/10.13140/
RG.2.2.22940.49286).
Mo i en ichmen analysis was ca ied ou by indMo i -
sGenome.pl and was pe o med on he −50, 50 bp lank-
ing egion o he peak summi s. The indi idual and a e age
ead densi y o he di e en peak se s was de e mined by
anno a ePeaks.pl (21). Read dis ibu ion and a e age den-
si y hea maps we e displayed by Ja a T eeView (52). His-
og ams and box plo s we e pe o med using G aphPad
P ism e sion 6.00 o Windows (G aphPad So wa e, La
Jolla, CA, USA, www.g aphpad.com).
RESULTS
Pu a i e ER␣binding si es in MCF-7 cells
To de e mine he ER␣cis ome, we collec ed public ChIP-
seq da a se s ha included ligand- ea ed samples wi h he
highes IP e iciency a ailable. By e-analysing he op eigh
es adiol- ea ed MCF-7 samples de i ed om six labo a-
o ies (Supplemen a y Table S2), we iden i ied a su p is-
ingly la ge numbe o ER␣binding si es (∼189 000 in o-
al), al hough he consis ency o hese si es was a he low;
speci ically, in a leas wo samples, only 56 931 ER␣TF-
BSs could be p edic ed (Supplemen a y Figu e S1A and B).
Among hese, only 4387 TFBSs we e p esen in all o he
in es iga ed samples, and we named hese si es he ‘consen-
sus peak se ’ ( he consensus ER␣TFBSs we e lis ed in Ad-
di ional ma e ial). The obse ed he e ogenei y may e lec
gene ic al e a ions, he numbe o cell passages, he es a-
diol concen a ion used in he speci ic expe imen s, he in-
duc ion ime o he es adiol esponse o o he echnical
di e ences, e.g. he numbe o cells o he p o ocol used
o ChIP. Measu ing DNase I hype sensi i i y o iden i y
ch oma in egions accessible o TFs e ealed he consen-
sus peak se wi h he highes ch oma in accessibili y (Sup-
plemen a y Figu e S1C).
While he majo i y o he de ec ed ER␣binding si es a e
a iega ed, ER␣SEs showed a high deg ee o consis ency
wi hin he epo ed expe imen s, highligh ing he impo -
ance o hese egula o y egions wi hin he genome. No-
ably, mo e han 90% o he SEs con ained a leas one peak
om he consensus se (Supplemen a y Table S2), and, in-
e es ingly, hese peaks we e ypically p esen also in he
ehicle- ea ed samples (Figu e 1A).
P ima y enhance s designa e u u e supe -enhance s
A compa ison o he ehicle- and es adiol- ea ed sam-
ples (GSM614610 and GSM614611 ER␣ChIP-seq samples
we e used du ing he main analysis) e ealed ha in he ab-
sence o ligand, each SE was ep esen ed by one o a ew
peak(s) and ligand-induced SEs o med a ound hese ac i-
a ed genomic egions. In he selec ed sample, 392 highly
co e ed egula o y egions could be p edic ed by he SE
inde o HOMER (Supplemen a y Figu e S1D).
Figu e 1A concen a es on he new binding e en s bu
canno emphasize he inc ease on p e-occupied egions; he
exac co e age o he examples is shown on Supplemen a y
Figu e S1E. No ably, hese indi idual peaks could no be
p edic ed as SEs in he ehicle- ea ed samples, as he cu -
en ly used p edic ion c i e ia ely mainly on he p esence
o clus e s o enhance s. These peaks a e indi idual o a
ew sepa a e peaks, bu upon ER␣ac i a ion, a clus e o
peaks assembles and is clea ly iden i ied as SEs. Based on
his obse a ion, we conside ed ha hese p e e en ial p i-
ma y enhance s may play a key ole in he o ma ion o SEs,
such as egula o y uni s; he e o e, we u he in es iga ed
hese genomic egions.
We examined he size dis ibu ion o ER␣peaks (4042
peaks we e in es iga ed), plo ing each membe o ER␣SEs
based on hei ag densi y (in deciles) (Figu e 1B and C).
As a con ol, we used o he ER␣TFBSs ha do no clus e
in SEs (75 693 peaks) (Figu e 1D and E). The op 10% o
SE peaks had an ex emely high ag densi y compa ed wi h
he o he si es in bo h ehicle- and es adiol- ea ed samples
(Figu e 1B and C). Al hough he dis ibu ion o he ag den-
si y was simila unde bo h condi ions, a mo e han 5- old
inc ease in he median ag densi y was obse ed upon ligand
ea men . To exclude echnical issues as he cause o his
obse a ion, we pe o med he same analysis using ano he
se o ER␣ChIP-seq samples and obse ed he same esul s
(Supplemen a y Figu e S1F and G). Plo ing he ER␣en-
ichmen o all ER␣-bound single enhance s ha a e no
membe s o a SE e ealed one- en h o he ag densi y o
hese si es compa ed wi h he peaks wi hin SEs. These e-
sul s sugges ha he binding e en s ou side o SEs a e less
p onounced (Figu e 1D and E).
Acco ding o he assump ion ha p ima y enhance s
o m he basis o SEs, we e e ed o hese elemen s as
‘mo he enhance s’, and he subsequen ly appea ing sec-
onda y enhance s we e e e ed o as ‘daugh e enhance s’.
The de e mina ion o mo he and daugh e enhance s is de-
sc ibed in he da a p ocessing desc ip ion in he Ma e ials
and Me hods sec ion.
To globally de e mine whe he he binding o ER␣ o he
so-called mo he enhance s p ecedes binding o he sub-
sequen so-called daugh e enhance s wi hin he SEs, we
so ed all ER␣SEs based on he RPKM alues o he
mo he enhance s (calcula ed om he ehicle- ea ed sam-
ples). In Figu e 1F, mo he enhance s we e well sepa a ed,
and a ela i ely low co e age o he daugh e enhance s was
obse ed. Howe e , upon ligand ea men , he exis ence o
enhance s was de ec ed on he majo i y o ER␣SEs (Fig-
u e 1F and G, Supplemen a y Figu e S2A and B). We also
plo ed he a e age ag densi y o he ER␣SE peaks, indi-
ca ing ha mo he enhance s show e en highe signals in
he ehicle- ea ed samples han daugh e enhance s upon
es adiol ea men (Figu e 1H and I, Supplemen a y Fig-
u e S3A and B).
To alida e he p ecedence o mo he enhance occu-
pancy, we compa ed ehicle- ea ed and un ea ed ER␣
ChIP-seq samples (Supplemen a y Figu e S4A). These da a
sugges ha ehicle ea men o MCF-7 cells changing he
ER␣binding nei he on mo he no on daugh e enhance
3696 Nucleic Acids Resea ch, 2017, Vol. 45, No. 7
Figu e 1. MCF-7-speci ic es ogen ecep o alpha (ER␣) supe -enhance s a e ep esen ed by one (o a ew) enhance (s) e en in he absence o es adiol
ea men . (A) In eg a i e Genomics Viewe (IGV) snapsho o ER␣ChIP-seq co e age ep esen ing six ER␣supe -enhance s (SEs) upon ehicle and
es adiol (E2) ea men . The in e al scale is 50 in bo h cases. (Band C)ER␣ ag densi y upon ehicle o E2 ea men in he deciles de e mined based
on ER␣ ec ui men a peaks wi hin SEs. (Dand E)ER␣ ag densi y upon ehicle o E2 ea men in deciles de e mined based on ER␣ ec ui men a
peaks no o e lapping wi h SEs. The boxes ep esen he i s and hi d qua iles, he ho izon al lines indica e he median eads pe kilobase pe million
mapped eads (RPKM) alues, and he whiske s indica e he lowe and uppe ex emes pe decile. (Fand G)ER␣ ag densi ies o he mo he (M) and he
op 6 daugh e enhance s (1–6) wi hin he same supe -enhance egion in ehicle- and E2- ea ed MCF-7 cells. Enhance s we e e ically so ed based on
he RPKM alues o he mo he enhance s (in he i s column), and he indi idual enhance s wi hin a SE egion we e subsequen ly ho izon ally aligned
based on he ead en ichmen o he ehicle- ea ed samples. (Hand I) His og ams show he a e age ag densi y o mo he (392) and daugh e enhance s
(3650) in he p esence o absence o E2. (Jand K) His og ams show he a e age ag densi y o mo he (392) and daugh e enhance s (3650) in he p esence
ehicle- o E2- ea men and upon amoxi en- and ul es an - ea men . Samples de i ed om he same expe imen .
Nucleic Acids Resea ch, 2017, Vol. 45, No. 7 3697
egions. The p ocessed ime-cou se expe imen also sug-
ges ed ha mo he enhance s a e indeed occupied p io o
daugh e enhance s, and show a p olonged binding a la e
ime poin s, oo (Supplemen a y Figu e S4B–D). Thus,
he e is a di e ence be ween he dynamics o he occupancy
o he dis inc SE peak se s. Impo an ly, he p esence o
ER␣on mo he enhance s is ema kable a e ER␣an ag-
onis ea men s such as amoxi en and ul es an (Figu e
1J and K). Mo eo e , in FoxA1-deple ed s a e, as shown on
Supplemen a y Figu e S4E–G, he binding o ER␣is e-
duced, bu no abolished nei he on mo he no on daugh-
e enhance s.
As MED1 was epo ed o be a key componen o he
Media o complex b idging SEs wi h ansc ip ion s a
si es, we isualized he p esence o his componen wi hin
he ER␣SEs. MED1 is ec ui ed o ER␣-bound si es wi h
a high binding a ini y o he buds o ini ia o (s) o SEs,
namely, mo he enhance s, upon es adiol ea men (Fig-
u e 2A). We also in es iga ed he MED1 densi ies based on
he p e iously de ined and so ed deciles o ER␣peaks, and
as shown in Figu e 2B and C, Media o is p e e en ially e-
c ui ed o ER␣peaks wi h highe ag densi ies, pa icula ly
in es adiol- ea ed samples (Figu e 2C). Collec i ely, we
obse ed collabo a ing enhance s clus e ed in SEs, showing
a much highe ER␣occupancy han single enhance s, and
e en mo e p ominen en ichmen s accompanied by Media-
o binding we e obse ed wi hin SEs.
The ead dis ibu ion plo a ound he indi idual en-
hance s demons a ed ha mo he enhance s a e loca ed
in he mos accessible ch oma in egions wi h high le -
els o DNase I signal and show he highes MED1, P300,
H3K27ac and BRD4 co e age upon induc ion, sugges ing
ha he op ER␣enhance s (392 mo he enhance s in o-
al) ep esen he mos ac i e egula o y egions, while he
daugh e enhance s (3650 in o al) did no show his pa e n
(Figu e 2D, Supplemen a y Table S3).
Based on he p e ious obse a ion, namely, ha each
supe -enhance egion is indica ed by a ma kedly ac i e e-
gion, we assessed whe he mo he and daugh e enhance s
we e obse ed o o he TFs and cell ypes. Fi s , we exam-
ined he o he wo lineage-de e mining ac o s associa ed
wi h MCF-7 cells, namely, FoxA1 and AP2␥(Supplemen-
a y Figu e S2C–F). Second, we ex ended hese in es iga-
ions o o he NRs in di e en cell ypes (AR in LNCaP
cells, RAR in mouse F9 cells and VDR in mouse in es i-
nal epi helial cells) and o JUNB in mouse BMDMs (Sup-
plemen a y Figu e S2G–N). In all expe imen al sys ems
in es iga ed (excep o FoxA1, as desc ibed la e in his
manusc ip ), pa e ns simila o hose o ER␣in MCF-
7 cells we e obse ed, indica ing ha a mo e gene al phe-
nomenon occu s in he p esence o se e al TFs (Supplemen-
a y Figu e S2A–N, Figu e S3A–N).
Canonical elemen s d i e ansc ip ion ac o s
To be e unde s and he ac o s ha disc imina e ER␣
mo he and daugh e enhance s, we applied a mo i en ich-
men analysis o hese egions. While mo he enhance s
showed s ong canonical ERE mo i en ichmen om a el-
a i ely low numbe o a ge sequences, daugh e enhance s
we e no en iched o he ER␣dime -speci ic elemen s bu
we e en iched o NR hal si es and FoxA1 mo i s (Fig-
u e 3A and B and Supplemen a y Figu e S5A and B). In-
e es ingly, a he si es o mo he enhance s, a unique NR
di ec epea elemen wi h o e lapping hal si es was also
iden i ied. Conside ing he s ong P- alue (1e-200) and he
high en ichmen o ERE (62.2%) (compa ed wi h he back-
g ound, which was 3.37%), despi e he small numbe o a -
ge sequences (n =328), we concluded ha he high le el o
ER␣ ec ui men a mo he enhance s (e en in he absence
o es adiol) e lec s s ong canonical DNA elemen s. How-
e e , o he eme gence o daugh e peaks, o he TFs, such
as FoxA1 and ac i a o p o ein 1 (AP-1), ac in conce wi h
he inc eased le el o ER␣a e es adiol ea men .
To alida e hese obse a ions, we examined he mo i de-
pendence o binding equency. We plo ed he mo i sco es
o he emapped EREs as a unc ion o he deciles de i ed
om he ehicle- ea ed peak size dis ibu ion in he case o
mo he enhance s and om he es adiol- ea ed peak size
dis ibu ion in he case o daugh e enhance s and obse ed
ha in all cases, he ex emely high mo he peaks had sig-
ni ican ly s onge binding elemen s han he daugh e en-
hance s (deciles 8–10 ha e P<0.0001) (Figu e 3C). This
inding sugges s ha he s onges elemen s o accessible eu-
ch oma ic egions i esis ibly a ac he ER␣dime s. To de-
e mine whe he his e ec is also obse ed o he binding
o o he TFs, we in es iga ed hese co ela ions o o he
ypes o SEs.
Mo i en ichmen analysis a he si es o he mo he
and daugh e enhance s p edic ed om o he in es iga ed
samples showed he same phenomenon. Fo example, we
de ec ed be e mo i en ichmen o mo he enhance s in
MCF-7 o FoxA1 han ha o hei daugh e s, and his
e ec was also obse ed o AP2␥(Supplemen a y Fig-
u e S5C–F). In e es ingly, AP2␥daugh e enhance s ha -
bou ed he mo i s o he main collabo a ing pa ne s,
FoxA1, ER␣and AP-1, oge he wi h he CCCTC-binding
ac o (CTCF). In he case o AR in he LNCaP cell line,
FoxA1 is needed o he es ablishmen o mo he enhance s,
as shown by he ex emely high mo i en ichmen in he
mo he enhance s (Supplemen a y Figu e S5G and H).
Impo an ly, daugh e enhance s showed lowe IR3/ARE
(ARE: and ogen esponse elemen ) and FoxA1 mo i en-
ichmen , bu a composi e elemen o FoxA1 and AR was
also iden i ied, indica ing a igh collabo a ion be ween
hese TFs, which may be less a ec ed by o he egula o s.
In he case o JUNB, mo he enhance s o mouse
BMDM cells showed all he main speci ic mo i en ich-
men s, namely, AP-1, AP-1::IRF (IRF: in e e on eg-
ula o y ac o ) composi e elemen (AICE), cAMP e-
sponse elemen and he mo i o he lineage-de e mining
CCAAT/enhance -binding p o ein (C/EBP). These mo i s
showed weake en ichmen in daugh e enhance s, and u -
he mo i s o collabo a ing TFs, namely, pu ine- ich box-1
(PU.1), nuclea ac o kappa-ligh -chain-enhance o ac i-
a ed B cells (NF-B) and signal ansduce and ac i a o
o ansc ip ion (STAT) appea ed (Supplemen a y Figu e
S5I and J). The RAR and VDR mo he enhance s o mouse
F9 and epi helial cells, espec i ely, also showed s onge
speci ic mo i en ichmen han he co esponding daugh e
peaks, and GATA p o ein(s) may also occupy VDR mo he
enhance s. Howe e , daugh e enhance s also showed pu-
3698 Nucleic Acids Resea ch, 2017, Vol. 45, No. 7
Figu e 2. MED1 is ec ui ed o ER␣-bound si es wi h an ex emely high binding a ini y. (A) IGV snapsho o MED1 and ER␣ChIP-seq co e age,
ep esen ing six ER␣SEs upon ehicle and es adiol (E2) ea men . The in e al scale is 50 in bo h cases. (Band C) MED1 ag densi y upon ehicle o E2
ea men in deciles de e mined based on ER␣ ec ui men a peaks wi hin SEs. The boxes ep esen he i s and hi d qua iles, ho izon al lines indica e
he median RPKM alues, and whiske s indica e he lowe and uppe ex emes pe decile. Pai ed - es , * signi ican a P<0.05; ** a P<0.01; *** a P
<0.001; **** a P<0.0001. (D) Read dis ibu ion plo o ER␣, DNase I, MED1, P300, H3K27ac and BRD4 upon ehicle o E2 ea men , ela i e o
SE peaks in 2-kb ames. Mo he (392 in o al) and daugh e (3650 in o al) peaks a e so ed acco ding o ER␣ ag densi y.
a i e AP-1 and K ¨uppel-like ac o (KLF) binding si es
(Supplemen a y Figu e S5K–N).
All examined TFs playing oles in SE o ma ion showed
a phenomenon simila o ha o ER␣, namely, hese p i-
ma y egula o y egions possessed canonical elemen s spe-
ci ic o dominan TF(s), while he u he occupied egions
had ewe speci ic elemen s oge he wi h hei collabo a-
i e ac o s. The binding o canonical elemen s p ecedes,
and upon ea men , likely acili a es he occupa ion o he
nea by egions by dominan TFs and in e ac ing pa ne s.
In summa y, he p ima y ChIP-seq peaks, which we e
p esen p io o es adiol s imula ion, had signi ican ly
s onge binding elemen s han he ac i a ed peaks, sug-
ges ing ha ce ain elemen s ha e high DNA–p o ein in e -
ac ion a ini ies and ha he e is no need o coope a i e
binding wi h o he ac o s, e.g. FoxA1 o AP2␥. The e o e,
we examined he di e en ially en iched ER␣and FoxA1
binding on ER␣SEs. As shown in Figu e 3D and E, he
occu ence o ER␣on he mo he enhance s wi hin ER␣
SEs is a p edominan e en (P<0.0001), bu in e es ingly,
Nucleic Acids Resea ch, 2017, Vol. 45, No. 7 3699
Figu e 3. Canonical elemen s p o ide highe DNA-binding a ini y han non-canonical elemen s. (Aand B) Mo i en ichmen s o mo he and daugh e
enhance s. The P- alue and a ge and backg ound (Bg) pe cen ages a e included o each mo i . (C) Es ogen esponse elemen (ERE) mo i sco es in he
deciles de e mined based on ER␣ ec ui men upon ehicle (in he case o mo he enhance s) o E2 ea men (in he case o daugh e enhance s) wi hin
SEs. (Dand E)RPKM alueso ER␣and FoxA1 co e age a ER␣mo he and daugh e enhance s. In he case o panels C, D and E, he boxes ep esen
he i s and hi d qua iles, he ho izon al lines indica e he median RPKM alues, and he whiske s indica e he 10 h o 90 h pe cen ile anges pe decile.
Pai ed - es , * signi ican a P<0.05, ** a P<0.01, *** a P<0.001, **** a P<0.0001.
FoxA1 also shows a simila pa e n (P<0.0001). The e-
sul s sugges ed ha he binding a ini y o ER␣was de i-
ni ely highe a he si es o mo he enhance s, and FoxA1
showed a s onge binding a hese si es despi e he lack
o i s binding elemen s ha migh mean he impo ance o
p o ein–p o ein in e ac ions, namely he binding o ER␣.
ER␣, FoxA1 and AP2␥ o m dis inc SEs
P e ious s udies ha e sugges ed ha in e ac ions wi h col-
labo a ing pionee ac o s a e necessa y o ER␣ unc ion;
he e o e, we conduc ed a de ailed in es iga ion o how
ER␣, FoxA1 and AP2␥ ec ui men co ela es wi h each
o he a he SE egions and examined he ex en o he
o e lap be ween he SEs o hese MCF-7-speci ic TFs. Fo
his pu pose, we plo ed he FoxA1 and AP2␥ ag den-
si y o mo he and daugh e ER␣enhance s (Figu e 4A),
and he esul ing ead dis ibu ion hea map showed a pos-
i i e co ela ion be ween he h ee TFs ollowing es adiol
ea men , bu he ec ui men o he collabo a ing TFs was
much lowe han ha o ER␣. Examining he mo he and
daugh e FoxA1 peaks e ealed an inc ease in AP2␥bind-
ing bu no in FoxA1 and ER␣binding (Figu e 4B), and we
consis en ly ob ained a simila esul o AP2␥SE peaks
(Figu e 4C). Measu ing he ag densi y o he h ee TFs on
hei SEs showed ha ER␣is ypically much less equen ly
ec ui ed a he mo he (and daugh e ) peaks o he o he
wo TFs (Figu e 4D and E) compa ed wi h he ec ui men
o FoxA1 and AP2␥a he si es o ER␣SEs. This e ec was
also obse ed o he ec ui men o FoxA1 and AP2␥a
each o he ’s SEs (Figu e 4F–I), sugges ing ha SEs a e yp-
ically domina ed by a single TF and ha o he TFs a e likely
collabo a ing pa ne s in hei ac i i y. TF speci ici y was
also con i med by he o e lap o SEs de e mined o ER␣,
FoxA1 and AP2␥(Figu e 4J). The Venn diag am shows ha
o hese h ee p o eins, ER␣and FoxA1 ha e a mo e dis-
inc SE p o ile, while AP2␥has i s own, less p ominen ,
SE p o ile. This inding is consis en wi h he co e age e-
sul s shown in Figu e 4A–C, as AP2␥was induced a all SEs
examined. The obse ed dominances may e lec he canon-
ical elemen s because no only he ER␣-bound mo he en-
hance s bu also hose o FoxA1 and AP2␥showed s onge
speci ic mo i en ichmen compa ed wi h ha in he co -
esponding seconda y egions (Figu e 4A and B, Supple-
men a y Figu e S5A–F). On FoxA1- and AP2␥-d i en SEs
FoxA1 showed no no able change upon E2 ea men , bu
in e es ingly, i did show highe densi y on ER␣-domina ed
SEs (Figu e 4A–C). The o me is no su p ising as no sig-
ni ican change could p e iously been obse ed in FoxA1
binding upon E2 s imula ion (46), bu he la e is s iking
as we ound a se o genomic egions, whe e E2 could a ec
FoxA1 binding.
Focusing on some ER␣-d i en SEs, we obse ed ha
he e is ypically a leas one FoxA1-bound enhance pe SE
in he absence o es adiol; howe e , hese enhance s a he
3700 Nucleic Acids Resea ch, 2017, Vol. 45, No. 7
Figu e 4. ER␣, FoxA1 and AP2␥ o m dis inc supe -enhance s. (A–C) Read dis ibu ion plo o ER␣, FoxA1 and AP2␥upon ehicle o E2 ea men ,
ela i e o each o he ’s SE peaks in 2-kb ames. The numbe o mo he and daugh e peaks, so ed based on ag densi y, is indica ed. (D–I)ER␣, FoxA1
and AP2␥ ag densi y o each o he ’s mo he and daugh e enhance s in E2- ea ed MCF-7 cells. (J) A ea-p opo ional Venn diag am illus a es he
o e laps be ween he ER␣, FoxA1 and AP2␥SEs.
occasionally o e lap wi h mo he enhance s (Figu e 5and
Supplemen a y Figu e S6A and B). In con as , AP2␥did
no show en ichmen p io o ea men and eme ged o-
ge he wi h ER␣, ypically a mo he enhance s and speci ic
daugh e enhance s occupied by FoxA1 (Figu e 5and Sup-
plemen a y Figu e S6C and D). The e is a simila en ich-
men a ER␣mo he enhance s as o AP2␥, which sugges s
an ERE/ER␣-dependen FoxA1 and AP2␥ ec ui men by
p o ein–p o ein in e ac ions. Impo an ly, MED1 is speci ic
only o mo he enhance s, while P300 may also be ec ui ed
o egions bound by FoxA1. H3K27ac and BRD4, oge he
wi h he DNA accessibili y, also ollowed hese endencies,
hus showing ele a ed ac i i y upon ea men (Figu es 5,
6A, D, G, J, M).
By examining he p esence o DNase I, MED1, P300,
H3K27ac, BRD4 and collabo a ing TFs on FoxA1- and
AP2␥-d i en SEs, we obse ed a s iking phenomenon: as
ER␣ ec ui men is nei he induced a hese si es, P300,
H3K27ac and BRD4 no show en ichmen , while DNase I
signal and MED1 binding is induced upon ea men (Fig-
u e 6B and C, E and F, H and I, K and L, N and O, Sup-
plemen a y Figu e S7A and B). No ably, acco ding o he
en ichmen o H3K27ac and BRD4, he sligh dec ease in
P300 ec ui men did no a ec hese egions, as hey we e
al eady a leas as ac i e in he un ea ed s a e as he in-
duced ER␣SEs, and his ac i a ion did no change (Fig-
u e 6G–I). These esul s s ongly sugges ER␣-dependen
MED1/P300 ec ui men , ollowed by his one ace yla ion
Nucleic Acids Resea ch, 2017, Vol. 45, No. 7 3701
Figu e 5. Rec ui men o FoxA1 and AP2␥a ER␣supe -enhance s. IGV snapsho o ER␣ChIP-seq co e age, ep esen ing eigh ER␣SEs upon ehicle
( eh) o E2 ea men and he simul aneous p esence o FoxA1, AP2␥, MED1, P300, H3K27ac, BRD4 and DNase I upon ehicle and E2 ea men . The
in e al scales a e indica ed in he uppe le co ne s. Peaks, highligh ed in g ey, ep esen he si es o ER␣mo he enhance s, while dashed lines indica e
he loca ion o FoxA1 peaks.
and BRD4 binding (Figu e 6A, D, G, J, M), while a FoxA1-
and AP2␥-domina ed SEs, only MED1 ec ui men and el-
e a ed ch oma in accessibili y we e obse ed upon es a-
diol ea men (Figu e 6E and F). On FoxA1- and AP2␥-
domina ed SEs es adiol ea men esul s in MED1 bind-
ing. Howe e , because P300 is no u he ec ui ed, nei he
his one ace yla ion no he eading module o he ace yla ed
his ones, namely BRD4 will be u he ec ui ed o hese
si es (Figu e 6E, F, H, I, K, L, N, O and Supplemen a y
Figu e S7A and B). These e en s sugges ed ha in he case
o TF-speci ic SE ec ui men and he inc ease in MED1
binding, an inc ease in DNase I hype sensi i i y is no su -
icien o he ini ia ion o hese e en s. Howe e , i is impo -
an o no e ha AP2␥-domina ed SEs showed he mos en-
iched ac i e enhance ma ks (Figu e 6C, F, I, L, O), while
hose o FoxA1 we e he leas ac i e (Figu e 6B, E, H, K,
N). Mo eo e , ER␣bound in a sca e ed manne o FoxA1-
and AP2␥-speci ic SEs (Supplemen a y Figu e S6E–L).
DISCUSSION
In his s udy we i s desc ibe ligand-inducible supe -
enhance s. Due o he almos p omp biological esponse
(Supplemen a y Figu e S4A–D) and he la ge a ie y o
