Spring2017–EpigeneticsandSystemsBiologyDiscussionSession(EpigeneticsandDevelopment)MichaelK.Skinner–Biol476/576Week8(March2)

EpigeneticsofCellandDevelopmentalBiology

PrimaryPapers1. Hackett,etal.,(2013)Science.339:4482. Chen,etal.,(2016)Science354:4683. Barau,etal(2016)Science354:909

DiscussionStudent21–Ref#1above

• WhatstagesofdevelopmentandcellshavereducedDNAmethylation?• Whattechnologywasused?• Whatroledoes5hmChaveintheprocess?

Student22–Ref#2above

• WhatisXist?• WhatisXinactivationandmechanism?• HowwilllncRNAlocalizethechromosometonuclearmembrane?

Student23–Ref#3above

• WhatisDNMTandisoformidentified?• Whatisatransposonandeffect?• Whatistheepigeneticobservationontheregulationoftransposonrepression?

performedwithP-element–transposon–containingCg4-neo and DmTNP or hTh9 (Fig. 2, B and C,and fig. S7). These observations imply that theDmTNP and hTh9 proteins can nick DNA, in-dependent of having P-element termini, whichwould lead to elevated gene transfer via DNAlinearization. A similar observation was made forthe SET domain and mariner transposase fusiongene–containing protein (SETMAR or Metnaseprotein), but this protein is inactive for transpo-sition of HsMAR transposons (8, 9). However,most important, the presence of P-element terminion Cg4-neo enhanced the DNA integration ac-tivity of both DmTNP and hTh9 3 to 5 timesabove background, which suggested transposi-tional DNA integration. Many G418-resistant col-onies were also obtained (fig. S7) when Cg4-neowas transfected into a stable HEK293 cell lineinduced to express a tetracycline-inducible hu-man THAP9 gene (Fig. 2B).

To analyze the nature of the DNA integrationevents in the G418-resistant colonies, genomicDNA was isolated from individual colonies ob-tained fromDmTNP or hTh9 cotransfections withCg4-neo, and the sites ofP-element insertion werecharacterized by splinkerette polymerase chainreaction (PCR) (17, 18) followed by DNA se-quencing. DNA sequence analysis of PCR inte-gration sites identified distinct integration siteswith novel 8–base pair (bp) target-site duplica-tions (TSDs) for individual integration eventsinto the human genome that had occurred withboth DmTNP and hTh9 (Table 1 and tables S1and S2). Taken together, these data indicate thathuman THAP9 actively integrates geneticallymarked Drosophila P-element vectors into hu-man cells by transposition.

The studies reported here indicate that the hu-man THAP9 gene encodes an active DNA trans-posase that can mobilize Drosophila P-elementtransposons in Drosophila and human cells. Itwill be interesting to investigate the physiologicalrelevance of THAP9’s transposition function andto find out if any THAP9 recombination signalDNA elements can be found in the human ge-nome. This is the first report, beyond the V(D)Jrecombination system, of an active DNA trans-posase in the human genome. P element–liketransposons and THAP9-related genes are notrestricted to Drosophila or related insect speciesbut are widely distributed in eukaryotic genomeslike Ciona (sea squirt), zebrafish, chicken, andTrichomonas vaginalis (a parasitic protozoan)(7, 19). The THAP9 gene is absent and has ap-parently been lost from sequenced rodent genomes(6). Although many of the human transposase–related genes are derived from DNA transposons(43 of 47) (2), most have not been characterized,with the exception of the V(D)J recombinaseRAG1 and RAG2 (10, 11) and the SETMAR(Metnase) protein (8). It is possible that other hu-man genes of this class, besides THAP9, may alsoencode active DNA transposases.

References and Notes1. C. Feschotte, E. J. Pritham, Annu. Rev. Genet. 41, 331

(2007).2. E. S. Lander et al., Nature 409, 860 (2001).3. S. Schaack, C. Gilbert, C. Feschotte, Trends Ecol. Evol. 25,

537 (2010).4. H. H. Kazazian Jr., Science 303, 1626 (2004).5. C. Feschotte, Nat. Rev. Genet. 9, 397 (2008).6. S. E. Hammer, S. Strehl, S. Hagemann, Mol. Biol. Evol.

22, 833 (2005).7. H. Quesneville, D. Nouaud, D. Anxolabehere, Mol. Biol. Evol.

22, 741 (2005).

8. M. Shaheen, E. Williamson, J. Nickoloff, S.-H. Lee,R. Hromas, Genetica 138, 559 (2010).

9. S.-H. Lee et al., Proc. Natl. Acad. Sci. U.S.A. 102, 18075(2005).

10. S. D. Fugmann, A. I. Lee, P. E. Shockett, I. J. Villey,D. G. Schatz, Annu. Rev. Immunol. 18, 495(2000).

11. M. Gellert, Annu. Rev. Biochem. 71, 101 (2002).12. M. Roussigne et al., Trends Biochem. Sci. 28, 66

(2003).13. S. Hagemann, W. Pinsker, Mol. Biol. Evol. 18, 1979

(2001).14. A. Sabogal, A. Y. Lyubimov, J. E. Corn,

J. M. Berger, D. C. Rio, Nat. Struct. Mol. Biol. 17, 117(2010).

15. Y. M. Mul, D. C. Rio, EMBO J. 16, 4441 (1997).16. P. J. Southern, P. Berg, J. Mol. Appl. Genet. 1, 327

(1982).17. C. Horn et al., Nat. Genet. 39, 933 (2007).18. C. J. Potter, L. Luo, PLoS ONE 5, e10168 (2010).19. J. M. Carlton et al., Science 315, 207 (2007).20. E. L. Beall, D. C. Rio, Genes Dev. 10, 921 (1996).

Acknowledgments: We thank J. M. Taliaferro, J. Aspden,and M. Francis for comments and discussion; R. Rawat fortechnical assistance; and C. J. Potter ( Johns Hopkins MedicalSchool) for helpful suggestions about the splinkerettePCR experiments. This work was supported by NIH grantsR01GM48862, R01GM61987, R01 GM097352, R01GM104385,and R01GM094890. S.M. and D.C.R. conceived theexperiments. S.M., A.S., and D.C.R. performed cell cultureand molecular biological experiments and analyzed data. S.M.and D.C.R. wrote the paper. The authors declare no competingfinancial interests. Requests for materials should beaddressed to the corresponding author.

Supplementary Materialswww.sciencemag.org/cgi/content/full/339/6118/446/DC1Materials and MethodsFigs. S1 to S7Tables S1 and S2References (21, 22)

19 October 2012; accepted 27 November 201210.1126/science.1231789

Germline DNA DemethylationDynamics and Imprint Erasure Through5-HydroxymethylcytosineJamie A. Hackett,1,2 Roopsha Sengupta,1,2* Jan J. Zylicz,1,2,3* Kazuhiro Murakami,1,2*Caroline Lee,1,2 Thomas A. Down,1 M. Azim Surani1,2,3†

Mouse primordial germ cells (PGCs) undergo sequential epigenetic changes and genome-wideDNA demethylation to reset the epigenome for totipotency. Here, we demonstrate that erasureof CpG methylation (5mC) in PGCs occurs via conversion to 5-hydroxymethylcytosine (5hmC),driven by high levels of TET1 and TET2. Global conversion to 5hmC initiates asynchronously amongPGCs at embryonic day (E) 9.5 to E10.5 and accounts for the unique process of imprint erasure.Mechanistically, 5hmC enrichment is followed by its protracted decline thereafter at a rate consistentwith replication-coupled dilution. The conversion to 5hmC is an important component of parallelredundant systems that drive comprehensive reprogramming in PGCs. Nonetheless, we identify rareregulatory elements that escape systematic DNA demethylation in PGCs, providing a potentialmechanistic basis for transgenerational epigenetic inheritance.

Specification of primordial germ cells (PGCs)from epiblast cells at ~embryonic day (E)6.25 is linked with extensive epigenetic

reprogramming—including global DNA demeth-

ylation, chromatin reorganization, and imprinterasure—that is vital for generating totipotency(1, 2). The erasure of CpG methylation (5mC) isa key component of this program, but the dynamics

and underlying mechanisms of the process remainunclear (3). Here, we report a comprehensive anal-ysis of PGCs by combining immunofluorescence,genome-wide 5-(hydroxy)methylcytosine DNAimmunoprecipitation sequencing [(h)meDIP-seq],single-cell RNA sequencing (RNA-seq), bisulfitesequencing, and functional analyses to addressthe mechanistic basis of epigenetic reprogrammingin PGCs.

We investigated Tet expression by using single-cell RNA-seq, which revealed that Tet1 and Tet2are expressed in PGCs and peak between E10.5and E11.5 but that Tet3 is undetectable (Fig. 1A).Immunofluorescence (IF) showed that TET1 andTET2 are nuclear and expressed at significantlyhigher levels in PGCs than in neighboring somat-ic cells between E9.5 and E11.5 (Fig. 1B and figs.

1Wellcome Trust/Cancer Research UK Gurdon Institute, Uni-versity of Cambridge, Cambridge CB2 1QN, UK. 2WellcomeTrust/Medical Research Council Stem Cell Institute, Universityof Cambridge, Cambridge, UK. 3Department of Physiology,Development and Neuroscience, University of Cambridge,Cambridge, UK.

*These authors contributed equally to this work.†To whom correspondence should be addressed. E-mail:a.surani@gurdon.cam.ac.uk

25 JANUARY 2013 VOL 339 SCIENCE www.sciencemag.org448

REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

0

2

4

6

8

10

12

Tet1Tet2Tet3

Log 2

(RP

M)

ES

E7.5 so

maE6.5E7.5 E9.5E8.5

E10.5E11

.5E12

.5

PGC

DAPI/Stella 5mC Merge

E9.

5E

10.5

E11

.5E

12.5

DAPI/TG1 5hmC Merge

E9.

5E

10.5

E11

.5E

12.5

E10

.5

DAPI/TG1 TET1 Merge

DAPI/TG1 TET2 Merge

A

B

C D

Fig. 1. Global dynamics of 5mC, 5hmC, and TETs in PGCs. (A) Single-cellRNA-seq analysis of Tet1, Tet2, and Tet3 expression. Shown is log2 reads permillion (RPM). (B) Expression of TET1 and TET2 in E10.5 PGCs (arrowheads)

and soma. (C) Dynamics of DNA methylation (5mC) in PGCs shows 5mC era-sure between E9.5 and E11.5. (D) Kinetics of 5hmC in PGCs. TG1/STELLA markPGCs. Scale bars indicate 10 mm.

0

5

10

15

20

25

30

35

40

Dazl Sycp3 Mael

WT Tet1CD/Tet2Mut Tet1CD/Tet2

EmptyVec

PG

CLC

% M

ethy

latio

n

1kb1

1

1

1

1

E12

.5

E11

.5

E10

.5

Epi

SC

2

2

2

2

2

E13

.5

E13

.5

E12

.5

E11

.5

E10

.5

Epi

SC

Dazl Dazl

5mC 5hmC

Nor

mal

ised

tag

dens

ity

1kb

E10.5soma

E10.5PGC

E11.5PGC

E12.5PGC

E13.5PGC -5

0

5

10

15

20

25

30

-20

0

20

60

80

100

120

% 5

mC

(G

lu-q

PC

R)

5mC5hmC

Dazl

% 5hm

C (G

lu-qPC

R)

40

74%

88%

92%

IAP

EpiLC

-DOX

+DOX

Tet1

/2 m

iR P

GC

LC

LINE-1

81%

51%

69%

0102030405060708090

EpiLC NTmiR

Tet1/2miR

-DOX+DOX

PGCLC

0

10

20

30

40

50

EpiLC NTmiR

Tet1/2miR

PGCLC

0

10

20

30

40

50

EpiLC NTmiR

Tet1/2miR

PGCLC

% M

ethy

latio

n (b

isul

fite)

Dazl Sycp3 Mael

0.05

0.10

0.15

0TSS TES50%+5kb -5kb

5mC

0.05

0.10

0.15

0TSS TES50%+5kb -5kb

5hmC

Nor

mal

ised

tag

dens

ity

E10.5E11.5E12.5E13.5

0.05

0.25

0.10

0.15

0.20

0

0.35

0.30

+250bp-250bp Exon

5mC

0.05

0.25

0.10

0.15

0.20

0

0.30

+250bp-250bp Exon

5hmC

Nor

mal

ised

tag

dens

ity EpiSCE10.5E11.5E12.5E13.5

EpiSCE10.5E11.5E12.5E13.5

E10.5E11.5E12.5E13.5

A

B

C

D

E

F G

Fig. 2. Erasure of 5mC is coupled to 5hmC conversion. (A) Enrichment of 5mCand 5hmC in E10.5 to E13.5 PGCs and EpiSCs over internal exons. (B) Distribu-tions of 5mC and 5hmC relative to a metagene. TSS, transcription start site; TES,transcription end site. (C) Profiles of 5mC (blue) and 5hmC (red) at the Dazlpromoter. (D) Glu-qPCR showing quantitative levels of 5mC and 5hmC at a CCGG

site in the Dazl promoter. Error bars represent SEM. (E and F) DNA methylation(%) by bisulfite sequencing of –DOX or +DOX Tet1/ Tet2miR or NT miR PGCLCsat (E) gene promoters and (F) repeat elements. Open and solid circles representunmethylated andmethylated CpGs, respectively. (G) DNAmethylation in PGCLCsstably expressing catalytically active (WT) or mutant (Mut) TET1 and TET2.

www.sciencemag.org SCIENCE VOL 339 25 JANUARY 2013 449

REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

S1 and S2). This suggests that erasure of 5mC inPGCs could occur through conversion to 5hmCby TET1 and TET2 together (4, 5).

We pursued this possibility by IF and founda progressive reduction of 5mC in PGCs betweenE9.5 and E10.5, until it became undetecta-ble by E11.5 (Fig. 1C). The loss of 5mC oc-curs concurrently with a global enrichment of5-hydroxymethylcytosine (5hmC) in PGCs be-tween E9.5 and E10.5, suggesting a genome-scale conversion of 5mC to 5hmC (Fig. 1D). Theglobal conversion to 5hmC initiates asynchro-nously among PGCs from E9.5, perhaps reflect-ing developmental heterogeneity (figs. S3 to S5).Indeed, TET1 up-regulation also initially occursin a subset of PGCs from E9.5, which apparentlyalso exhibit lower 5mC signal (fig. S6). In con-trast to soma and embryonic stem (ES) cells (6),we observed that 5hmC exhibited a distinct local-ization in PGCs that coincided with 4 ,́6-diamidino-2-phenylindole (DAPI)–dense chromocenters,indicating that the conversion of 5mC to 5hmCincludes heterochromatic satellite regions (fig. S7).The enrichment of 5hmC in PGCs at E10.5 is fol-lowed by its progressive reduction, suggestingthat 5hmC is an intermediate toward demethyl-ation to unmodified cytosine (C) (Fig. 1D). Wechecked whether 5hmC is subsequently converted

to 5-formylcytosine or 5-carboxycytosine but foundno detectable enrichment of these derivatives inPGCs (fig. S8) (7).

To gain further insight into the dynamics of5mC to 5hmC conversion, we performed meDIP-seq and hmeDIP-seq in E10.5 to E13.5 PGCs(fig. S9). Because before E10.5 PGCs were high-ly limiting, we also profiled epiblast stem cells(EpiSCs), which are derived from the same post-implantation epiblast as nascent PGCs, and em-bryonic soma (E10.5) as references (fig. S10).Unlike bisulfite sequencing, our approach distin-guishes between 5mC and 5hmC but generatesa relative rather than a quantitative measure ofmodifications (6). We therefore initially exam-ined exonic sequences, which are highly methyl-ated and thus exhibit an informative dynamicrange of relative (h)meDIP signal when they be-come demethylated. We found significantly re-duced 5mC in E10.5 PGCs relative to EpiSCand soma and erasure by E11.5 (Fig. 2A andfigs. S11 to S13). The loss of 5mC in PGCs isparalleled by a strong exonic enrichment of 5hmC,indicating 5mC-to-5hmC conversion (Fig. 2Aand fig. S11). Once 5mC is converted to 5hmC, itis set on a pathway toward demethylation, be-cause there are no 5hmC maintenance mecha-nisms (6). Consistent with this, 5hmC undergoes

a progressive depletion during PGC develop-ment, which is delayed relative to loss of 5mC(Fig. 2, A and B).

Next, we examined methylation-dependentgenes such as Dazl, which are activated by pro-moter demethylation in PGCs (8, 9), and ob-served strong 5hmC enrichment coincident withloss of 5mC at their promoters (Fig. 2C and fig.S14). We confirmed that 5mC erasure is cou-pled to 5hmC enrichment at the Dazl promoterquantitatively, by using the glucosyltransferase–quantitative polymerase chain reaction (Glu-qPCR)assay (Fig. 2D). RNA-seq revealed that transcrip-tional activation of Dazl and other methylation-dependent germline genes initiates at E9.5 andincreases progressively until ~E11.5 (fig. S15).This represents an important functional readoutof the timing of DNA demethylation in PGCs.

To functionally link 5hmC to DNA demethyl-ation, we used in vitro PGC-like cells (PGCLC).PGCLCs are specified from epiblast-like cells(EpiLCs) and exhibit the fundamental proper-ties of migratory PGCs in vivo, including globalDNA demethylation and chromatin reorganiza-tion (fig. S10) (10). TET1 and TET2 are both activein PGCs, so we generated PGCLCs carrying a dox-ycycline (DOX)–inducible compound microRNA(miR) knockdown of Tet1 and Tet2 (T-KD). We

0

0.1

0.2

0.3

0.4

0.5

0.6

Nor

mal

ised

tag

dens

ity

EpiSC

E10.5

PGC

E11.5

PGC

E12.5

PGC

5mC5hmC

EpiSC

E10.5

PGC

E11.5

PGC

E12.5

PGC

gDMR

E10.5soma

E10.5PGC

E11.5PGC

E12.5PGC

E13.5PGC

-5

0

5

10

15

20

25

30

-10

0

10

20

30

40

50

60

% 5

mC

5mC5hmC

Kcnq1ot1

% 5hm

C

0

10

20

30

40

50

60Predicted (passive)Peg10Peg3

% D

NA

mod

ifica

tion

(Glu

-qP

CR

)

E10.5PGC

E11.5PGC

E12.5PGC

E13.5 PGC

Peg10Sgce

DMR

Nor

mal

ised

tag

dens

ity

1

1

E12

.5

1

E11

.5

1

E10

.5

1

Epi

SC

1kb

3

3

E12

.5

3

E11

.5

3

E10

.5

3

Epi

SC

5mC 5hmC1kb

Peg10Sgce

DMR

E10.5soma

E10.5PGC

E11.5PGC

E12.5PGC

E13.5PGC

-5

0

5

10

15

20

25

30

-10

0

10

20

30

40

50

605mC5hmC

Igf2r

% 5

mC

% 5hm

C

E10.5soma

E10.5PGC

E11.5PGC

E12.5PGC

E13.5PGC

-5

0

5

10

15

20

25

30

-10

0

10

20

30

40

50

605mC5hmC

Peg10

% 5

mC

% 5hm

C

E10.5soma

E10.5PGC

E11.5PGC

E12.5PGC

E13.5PGC

-5

0

5

10

15

20

25

30

-10

0

10

20

30

40

50

605mC5hmC

Peg3

% 5

mC

% 5hm

C

Kcnq1Kcnq1ot1

DMR

Kcnq1Kcnq1ot1

DMR

Nor

mal

ised

tag

dens

ity

1.5

E13

.5

E13

.5

E13

.5

E13

.5

1.5

E12

.5

1.5

E11

.5

1.5

E10

.5

1.5

Epi

SC

2

2

2

E12

.5

2

E11

.5

2

E10

.5

Epi

SC

1kb1kb5mC 5hmC

A

B C

D

E Fig. 3. Imprint erasure in PGCs. (A) Average (h)meDIP-seq en-richment of 5mC and 5hmC across all imprinted gDMRs in PGCs.EpiSC represents monoallelic methylation. (B and C) The 5mC and5hmC profiles of DMRs that undergo (B) early or (C) late 5mC era-sure and corresponding delay in 5hmC enrichment. (D) Glu-qPCRanalysis of 5mC and 5hmC levels at early (Kcnq1ot1 and Igf2r)and late (Peg10 and Peg3) reprogrammed DMRs. Error bars rep-resent SEM. (E) Rate of demethylation to unmodified cytosine.Shown is the predicted rate of passive demethylation and ob-served rates for Peg10 and Peg3.

25 JANUARY 2013 VOL 339 SCIENCE www.sciencemag.org450

REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

found that genes known to be demethylated inPGCs in vivo (8) also underwent DNA demeth-ylation upon specification of control uninduced(–DOX) T-KD PGCLCs and in nontargeting(NT) miR PGCLCs (+/−DOX). In contrast, in-duction of Tet1/Tet2 miR (+DOX) resulted in asubstantial inhibition of DNA demethylation inPGCLCs but did not reduce the efficiency oftheir specification (Fig. 2E and fig. S16).Knockdown of Tet1/Tet2 also inhibited DNAdemethylation at long interspersed nuclear ele-ment 1 (LINE-1) sites and prevented the limitederasure of 5mC that occurs at intracisternal-A-particles (IAP) (Fig. 2F). These findings are im-portant considering that both the maintenanceand de novo DNA methylation systems are re-pressed in PGCs and PGCLCs (10), which likelyaccounts for some direct passive demethylation.Moreover, constitutive overexpression of catalyt-ically active, but not catalytic mutant, TET1 andTET2 in PGCLCs promoted 5mC erasure to agreater extent (Fig. 2G). Thus, TET-mediated5hmC conversion is a key event toward DNA de-methylation in PGCs.

The reprogramming of gonadal PGCs in vivouniquely entails the complete erasure of genomicimprints (11). Analysis of imprinted gametic dif-ferentially methylated regions (gDMRs) (n = 21)in PGCs revealed that erasure of 5mC is coupledto a significant increase of 5hmC enrichment(Fig. 3A). However, the precise timing of 5mCerasure is imprinted locus-specific. For exam-ple, the DMRs at Kcnq1ot1 and Igf2r exhibit lossof 5mC by E10.5 relative to EpiSC (whichrepresent ~50% allelic 5mC) and erasure by

E11.5 (Fig. 3B), whereasPeg10 andPeg3 remainmethylated until E11.5 (Fig. 3C and fig. S17).Moreover, Kcnq1ot1 and Ig f2r are enriched in5hmC by E10.5, whereas 5hmC enrichment atPeg10 and Peg3 is delayed until E11.5, suggest-ing that conversion to 5hmC follows a definedtemporal order at imprinted DMRs, which dic-tates the timing of demethylation in PGCs. In-deed, we observed that other genomic regionsalso exhibited differential onset of 5mC erasure(compare Peg10 DMR versus exon, Fig. 3C).Glu-qPCR analysis confirmed that the Peg10and Peg3 DMRs maintained 5mC levels of 50%and 34%, respectively, in E10.5 PGCs, whereasKcnq1ot1 and Igf2rDMRswere already reducedto 21%and25%, respectively (Fig. 3D).Glu-qPCRalso established the quantitative enrichment of5hmC at imprinted DMRs in PGCs. The cumula-tive data suggest that conversion of 5mC to 5hmCby TET1 and TET2 is a general mechanism forthe erasure of imprints in PGCs.

Conversion of 5mC to 5hmC at exons, pro-moters, and gDMRs in PGCs was followed by aprotracted period of progressive 5hmC depletionbetween E11.5 and E13.5 (Figs. 2, A to D, and 3),suggesting a replication-coupled process (12).This prompted us to examine the rate of DNAdemethylation between E10.5 and E13.5 quanti-tatively by using Glu-qPCR. Because demethyl-ation commences asynchronously in PGCs, it isnecessary to examine loci that have not initiatedsubstantial 5mC erasure by E10.5, such as Peg10and Peg3. Because PGCs have an estimated cellcycle of ~16 hours between E10.5 and E13.5(13), we would predict a reduction of DNA mod-

ification of ~threefold per 24 hours (1.5 pop-ulation doublings) if the process is coupled toDNA replication. We observed that the rate ofdemethylation at Peg10 (P = 0.0022) and Peg3(P = 0.0019) fits highly significantly with thepredicted rate (Fig. 3E), suggesting that 5hmCmay be removed from these loci by replication-coupled dilution. We obtained similar results forthe Dazl promoter (P = 0.0014).

We next asked whether any promoters orregulatory elements can escape the comprehen-sive 5mC reprogramming in PGCs. We screenedfor CpG islands (CGI) that remain methylated infemale PGCs at E13.5, because these cells rep-resent the lowest point of global demethylation(fig. S18) (14). We identified 11 CGIs with sig-nificant 5mC enrichment in E13.5 PGCs (figs.S19 and S20). Validation by bisulfite sequencingshowed that the promoter CGIs of Vmn2r29 andSfi1 and the exonic CGI of Srrm2 were all meth-ylated in PGCs at E10.5 and maintained CpGmethylation throughout reprogramming (Fig. 4A).

To define the extent of 5mC erasure at single-base resolution, we performed whole-genome bi-sulfite sequencing (WGBS), which revealed thatglobal CpG methylation is reduced to 2.2% infemale E13.5 PGCs (fig. S21). However, we iden-tified 4730 loci that escape demethylation (>40%5mC) in PGCs, which are predominately repeatassociated (>95%). Resistant loci predominantlycorrespond to IAP elements, but the IAPLTR1subclass is significantly more methylated thanany other (fig. S22). IAPLTR1 is the most activeand hence hazardous IAP subclass to genomic in-tegrity, suggesting specific systems are mobilizedto maintain 5mC at IAPLTR1 during reprogram-ming to protect genome stability (15). We wereunable to determine any unique sequence char-acteristics of the 233 single-copy loci with >40%5mC, suggesting that positional context or chro-matin structure may contribute to their escapefrom reprogramming. Indeed, “escapees” wereoften adjacent to IAP elements or telomeric re-gions. Considered with the recent observationsthat many regulatory elements can evade zygotic5mC erasure (16, 17), our data suggest that rarebut potentially functionally relevant 5mC epiallelescould be inherited over multiple generations byevading erasure during both zygotic and PGCreprogramming.

We demonstrate here that comprehensiveDNA demethylation in PGCs, including imprinterasure, entails conversion of 5mC to 5hmC, like-ly redundantly by TET1 and TET2. In vivo 5hmCconversion initiates asynchronously in PGCs be-tween E9.5 and E10.5 and is largely complete byE11.5. The rate of progressive decline of 5hmCthereafter, both globally and at single-copy loci,is consistent with a replication-dependent mech-anism of demethylation toward unmodified cy-tosines (Fig. 4B). In parallel to 5hmC conversion,repression of the de novo (Dnmt3a/b) and main-tenance (Uhrf1) DNA methylation systems inPGCs prevents cyclical remethylation and simul-taneously renders PGCs permissive for direct

E10.5PGCE10.5PGC

E11.5PGC

E12.5PGC

Vmn2r29 Sfi1 Srrm2 DazlA

Passive 5mC depletion

Passive 5hmC depletion (Tet3)

Maternal

Paternal

BE

R

Zygote PGC

Passive 5hmC depletion (Tet1/Tet2) Passive 5mC depletion Active BER

B

Fig. 4. Inheritance of 5mC through reprogramming. (A) The Vmn2r29, Sfi1, and Srrm2 CGIs escapereprogramming in PGCs. Open and solid circles represent unmethylated and methylated CpGs, respec-tively.Dazl is representative of demethylation at most loci. (B) Model for the mechanisms and dynamics ofDNA demethylation in PGCs.

www.sciencemag.org SCIENCE VOL 339 25 JANUARY 2013 451

REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

passive 5mC depletion (fig. S23) (18), which maycontribute to the partial demethylation observed inTet1 and Tet2 knockdown PGCLCs. Thus, where-as in zygotes 5mC reprogramming is mechanis-tically compartmentalized into TET3-mediated5hmC conversion of the paternal genome and di-rect passive 5mC depletion on the maternal ge-nome (12, 19–21), both of these mechanismsoperate together in PGCs (Fig. 4B). In addition,up-regulation of the base excision repair (BER)pathway in PGCs may both protect against cu-mulative genetic damage and act as an auxiliaryactive demethylation mechanism, perhaps forspecific loci (22, 23). Reprogramming in PGCstherefore involvesmultiple redundantmechanismsto reset the epigenome for totipotency, which ac-counts for the apparent fertility (albeit subfertile)of mice lacking individual components, such asTet1 (24). The existence of multiple mechanismsmay also underpin the comprehensive nature ofDNA demethylation in PGCs (3). Nonetheless,some rare single-copy sites of CpG methylationescape from 5mC erasure (25), which may pro-

vide mechanistic avenues for investigations intotransgenerational epigenetic inheritance.

References and Notes1. M. A. Surani, K. Hayashi, P. Hajkova, Cell 128, 747

(2007).2. P. Hajkova et al., Nature 452, 877 (2008).3. J. A. Hackett, J. J. Zylicz, M. A. Surani, Trends Genet. 28,

164 (2012).4. M. Tahiliani et al., Science 324, 930 (2009).5. S. Ito et al., Nature 466, 1129 (2010).6. G. Ficz et al., Nature 473, 398 (2011).7. S. Ito et al., Science 333, 1300 (2011); 10.1126/science.

1210597.8. D. M. Maatouk et al., Development 133, 3411

(2006).9. J. A. Hackett et al., Development 139, 3623 (2012).10. K. Hayashi, H. Ohta, K. Kurimoto, S. Aramaki, M. Saitou,

Cell 146, 519 (2011).11. P. Hajkova et al., Mech. Dev. 117, 15 (2002).12. A. Inoue, Y. Zhang, Science 334, 194 (2011);

10.1126/science.1212483.13. P. P. Tam, M. H. Snow, J. Embryol. Exp. Morphol. 64, 133

(1981).14. C. Popp et al., Nature 463, 1101 (2010).15. C. Qin et al., Mol. Carcinog. 49, 54 (2010).16. S. A. Smallwood et al., Nat. Genet. 43, 811 (2011).

17. J. Borgel et al., Nat. Genet. 42, 1093 (2010).18. K. Kurimoto et al., Genes Dev. 22, 1617 (2008).19. T. P. Gu et al., Nature 477, 606 (2011).20. M. Wossidlo et al., Nat. Commun. 2, 241 (2011).21. K. Iqbal, S. G. Jin, G. P. Pfeifer, P. E. Szabó, Proc. Natl.

Acad. Sci. U.S.A. 108, 3642 (2011).22. P. Hajkova et al., Science 329, 78 (2010).23. S. Cortellino et al., Cell 146, 67 (2011).24. M. M. Dawlaty et al., Cell Stem Cell 9, 166 (2011).25. S. Guibert, T. Forné, M. Weber, Genome Res. 22, 633

(2012).

Acknowledgments: We thank N. Miller for fluorescence-activatedcell sorting analysis, F. Tang and W. Tang for experimentalsupport, and G-L. Xu for reagents. This work was funded by theWellcome Trust (RG49135, RG44593, and 083563) and theHuman Frontier Science Program. Sequencing data havebeen deposited in the Sequence Read Archive (SRA060914).

Supplementary Materialswww.sciencemag.org/cgi/content/full/science.1229277/DC1Materials and MethodsFigs. S1 to S24References (25, 26)

24 August 2012; accepted 16 November 2012Published online 6 December 2012;10.1126/science.1229277

Actin, Spectrin, and AssociatedProteins Form a PeriodicCytoskeletal Structure in AxonsKe Xu,1* Guisheng Zhong,1* Xiaowei Zhuang1,2†

Actin and spectrin play important roles in neurons, but their organization in axons anddendrites remains unclear. We used stochastic optical reconstruction microscopy to study theorganization of actin, spectrin, and associated proteins in neurons. Actin formed ringlike structuresthat wrapped around the circumference of axons and were evenly spaced along axonal shaftswith a periodicity of ~180 to 190 nanometers. This periodic structure was not observed indendrites, which instead contained long actin filaments running along dendritic shafts. Adducin,an actin-capping protein, colocalized with the actin rings. Spectrin exhibited periodic structuresalternating with those of actin and adducin, and the distance between adjacent actin-adducinrings was comparable to the length of a spectrin tetramer. Sodium channels in axons were distributedin a periodic pattern coordinated with the underlying actin-spectrin–based cytoskeleton.

Actin plays critical roles in shaping andmaintaining cell morphology, as well asin supporting various cellular functions,

including cell motility, cell division, and intra-cellular transport (1). In neurons, actin is essen-tial for the establishment of neuronal polarity,cargo transport, neurite growth, and stabilizationof synaptic structures (2–4). Despite its impor-tance, our understanding of actin structures inneurons remains incomplete. Electron micros-copy has shown detailed actin ultrastructure ingrowth cones and dendritic spines (5, 6), in which

actin is the dominant cytoskeletal protein, butlittle is known about the organization of actinin the axonal and dendritic shafts (4). Theseneurites contain a high density of different typesof cytoskeletal filaments, such as microtubulesand neurofilaments (6–8). Hence, the challengeof resolving the organization of actin in axonsand dendrites requires imaging tools with bothhigh spatial resolution and molecular specificity.

A prototypical actin-spectrin–based cytoskel-eton structure is found in red blood cells (erythro-cytes) (9, 10), where actin, spectrin, and associatedproteins form a two-dimensional (2D) polygonalnetwork (mostly composed of hexagons andpentagons) underneath the erythrocyte mem-brane (11, 12). Spectrin analogs have been foundin many other animal cells (9, 10), includingneurons (13, 14). These analogs play importantroles, ranging from regulation of the heartbeat

to stabilization of axons, formation of axon ini-tial segments and nodes of Ranvier, and stabi-lization of synapses in neurons (9, 10, 15). Anerythrocyte-like, polygonal lattice structure hasbeen observed for spectrin in the Drosophilaneuromuscular junction (16), and models similarto the erythrocyte cytoskeleton have also beenproposed for other systems (10). However, theultrastructural organization of spectrin in non-erythrocyte cells is largely unknown due to sim-ilar challenges in imaging.

Recent advances in superresolution fluores-cencemicroscopy (17, 18) allow resolutions downto ~10 nm to be achieved with molecular spec-ificity, providing a promising solution to theabove challenges. In particular, superresolutionstudies of neurons have provided valuable struc-tural and dynamic information of actin in den-dritic spines (19–22). In this work, we used asuperresolution fluorescence imaging method,stochastic optical reconstruction microscopy(STORM) (23–27), to study the 3D ultrastructuralorganization of actin and spectrin in neurons.

To image actin in neurons, we fixed culturedrat hippocampal neurons at various days in vitro(DIV) and labeled actin filaments with phalloi-din conjugated to a photoswitchable dye, AlexaFluor 647 (Invitrogen, Carlsbad, CA) (Fig. 1) (28).To identify axons and dendrites, we immuno-labeled MAP2, a microtubule-associated proteinenriched in dendrites, or NrCAM, a cell adhe-sion molecule found in the initial segments ofaxons (15), using a dye of a different color(Fig. 1) (28). In the conventional fluorescenceimages (Fig. 1, A, D, and F), MAP2 specificallystained dendrites, and NrCAM specifically la-beled the initial segments of axons, whereas actinwas found in both dendrites and axons.

Next, we used 3D STORM (27) with a dual-objective astigmatism-imaging scheme (28, 29)

1Howard Hughes Medical Institute (HHMI), Department ofChemistry and Chemical Biology, Harvard University, Cam-bridge, MA 02138, USA. 2Department of Physics, HarvardUniversity, Cambridge, MA 02138, USA.

*These authors contributed equally to this work.†To whom correspondence should be addressed. E-mail:zhuang@chemistry.harvard.edu

25 JANUARY 2013 VOL 339 SCIENCE www.sciencemag.org452

REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

southern Spain turns into desert, deciduousforests invade most of the mountains, and Me-diterranean vegetation replaces most of thedeciduous forests in a large part of the Mediter-ranean basin. Figure 3H illustrates the variationfrom areas without any changes, regardless ofscenario (stable white areas), to areas in whichchanges from scenario RCP2.6 already appear(red areas). As expected, the most-sensitive areasare those located at the limit between two biomes—for example, in the mountains at the transitionbetween temperate and montane forest or in thesouthern Mediterranean at the transition betweenforest and desert biomes. The map for 4700 yrB.P., in which the past changes were among thehighest (Fig. 3B), has the largest changes in thesouthwest, eastern steppe areas, and the moun-tains, but these changes are relatively sparse.Our analysis shows that, in approximately one

century, anthropogenic climate change withoutambitious mitigation policies will likely alterecosystems in the Mediterranean in a way thatis without precedent during the past 10 millen-nia. Despite known uncertainties in climatemodels, GHG emission scenarios at the levelof country commitments before the UNFCCCParis Agreement will likely lead to the sub-stantial expansion of deserts in much of south-ern Europe and northern Africa. The highlyambitious RCP2.6 scenario seems to be the onlypossible pathway toward more limited impacts.Only the coldest RCP2.6L simulations, whichcorrespond broadly to the 1.5°C target of theParis Agreement, allow ecosystem shifts to re-main inside the limits experienced during theHolocene.This analysis does not account for other hu-

man impacts on ecosystems, in addition to cli-mate change (i.e., land-use change, urbanization,soil degradation, etc.), which have grown inimportance after the mid-Holocene and havebecome dominant during the past centuries.Many of these effects are likely to become evenstronger in the future because of the expand-ing human population and economic activity.Most land change processes reduce natural veg-etation or they seal or degrade the soils, repre-senting additional effects on ecosystems, whichwill enhance, rather than dampen, the biomeshifts toward a drier state than estimated bythis analysis. This assessment shows that, with-out ambitious climate targets, the potential forfuture managed or unmanaged ecosystems tohost biodiversity or deliver services to society islikely to be greatly reduced by climate changeand direct local effects.

REFERENCES AND NOTES

1. S. I. Seneviratne, M. G. Donat, A. J. Pitman, R. Knutti,R. L. Wilby, Nature 529, 477–483 (2016).

2. W. Cramer et al., in Climate Change 2014: Impacts, Adaptation,and Vulnerability. Part A: Global and Sectoral Aspects.Contribution of Working Group II to the Fifth Assessment Reportof the Intergovernmental Panel on Climate Change, C. B. Fieldet al., Eds. (Cambridge Univ. Press, 2014), pp. 979–1037.

3. C.-F. Schleussner et al., Earth Syst. Dyn. Discuss. 6,2447–2505 (2015).

4. J. Guiot, D. Kaniewski, Front. Earth Sci. 3, 28 (2015).

5. F. Médail, N. Myers, in Hotspots Revisited: Earth’s BiologicallyRichest and Most Endangered Terrestrial Ecoregions,R. A. Mittermeier et al., Eds. (Conservation International,2004), pp. 144–147.

6. Materials and methods are available as supplementarymaterials on Science Online.

7. E. Xoplaki, J. F. González-Rouco, J. Luterbacher, H. Wanner,Clim. Dyn. 23, 63–78 (2004).

8. L. Maiorano et al., Philos. Trans. R. Soc. London Ser. B 366,2681–2692 (2011).

9. T. Keenan, J. Maria Serra, F. Lloret, M. Ninyerola, S. Sabate,Glob. Change Biol. 17, 565–579 (2011).

10. W. Thuiller, S. Lavorel, M. B. Araújo, Glob. Ecol. Biogeogr. 14,347–357 (2005).

11. B. Weninger et al., Doc. Praehist. 36, 7–59(2009).

12. D. Kaniewski, E. Van Campo, H. Weiss, Proc. Natl. Acad. Sci.U.S.A. 109, 3862–3867 (2012).

13. N. Roberts, D. Brayshaw, C. Kuzucuoglu, R. Perez, L. Sadori,Holocene 21, 3–13 (2011).

14. C. P. Kelley, S. Mohtadi, M. A. Cane, R. Seager, Y. Kushnir,Proc. Natl. Acad. Sci. U.S.A. 112, 3241–3246 (2015).

15. B. I. Cook, K. J. Anchukaitis, R. Touchan, D. M. Meko,E. R. Cook, J. Geophys. Res. 121, 2060–2074 (2016).

16. G. Middleton, J. Archaeol. Res. 20, 257–307 (2012).17. A. B. Knapp, S. W. Manning, Am. J. Archaeol. 120, 99–149

(2016).18. C. B. Yackulc, J. D. Nichols, J. Reid, R. Der, Ecology 96, 16–23

(2015).19. C. Roumieux et al., Ecol. Mediterr. 36, 17–24

(2010).20. I. Harris, P. D. Jones, T. J. Osborn, D. H. Lister, Int. J. Climatol.

34, 623–642 (2014).

ACKNOWLEDGMENTS

The authors are members of the Observatoire des Sciences del’Univers Pytheas Institute and the ECCOREV network. Thisresearch has been funded by Labex OT-Med (project ANR-11-LABEX-0061), the “Investissements d’Avenir” French governmentproject of the French National Research Agency (ANR),AMidex (project 11-IDEX-0001-02), and the EuropeanUnion FP7-ENVIRONMENT project OPERAs (grant 308393).We acknowledge the World Climate Research Programme’sWorking Group on Coupled Modelling, which is responsible forCMIP, and we thank the climate modeling groups (table S2) forproducing and making their model outputs available. For CMIP,the U.S. Department of Energy’s Program for Climate ModelDiagnosis and Intercomparison provides coordinating supportand led the development of software infrastructure, inpartnership with the Global Organization for Earth SystemScience Portals. R. Suarez and S. Shi have extracted andpreprocessed the model simulations. Holocene climatereconstructions are available at http://database.otmed.fr/geonetworkotmed/srv/eng/search - |54b9bf34-57ae-45ea-b455-9f90351e538f. Future climate projections are available athttp://cmip-pcmdi.llnl.gov/cmip5/.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/354/6311/465/suppl/DC1Materials and MethodsTable S1 and S2References (21–28)

6 July 2016; accepted 21 September 201610.1126/science.aah5015

GENE EXPRESSION

Xist recruits the X chromosome to thenuclear lamina to enablechromosome-wide silencingChun-Kan Chen,1 Mario Blanco,1 Constanza Jackson,1 Erik Aznauryan,1

Noah Ollikainen,1 Christine Surka,1 Amy Chow,1 Andrea Cerase,2

Patrick McDonel,3 Mitchell Guttman1*

The Xist long noncoding RNA orchestrates X chromosome inactivation, a process that entailschromosome-wide silencing and remodeling of the three-dimensional (3D) structure of theX chromosome. Yet, it remains unclear whether these changes in nuclear structure aremediated by Xist and whether they are required for silencing. Here, we show that Xist directlyinteracts with the Lamin B receptor, an integral component of the nuclear lamina, and thatthis interaction is required for Xist-mediated silencing by recruiting the inactive X to thenuclear lamina and by doing so enables Xist to spread to actively transcribed genes acrossthe X. Our results demonstrate that lamina recruitment changes the 3D structure of DNA,enabling Xist and its silencing proteins to spread across the X to silence transcription.

The Xist long noncoding RNA (lncRNA) ini-tiates X chromosome inactivation (XCI), aprocess that entails chromosome-wide tran-scriptional silencing (1) and large-scale re-modeling of the three-dimensional (3D)

structure of the X chromosome (2–4), by spread-ing across the future inactive X chromosomeand excluding RNA polymerase II (PolII) (1, 5).Xist initially localizes to genomic DNA regionson the X chromosome that are not actively tran-scribed (6–8), before spreading to actively tran-scribed genes (7–9). Deletion of a highly conservedregion of Xist that is required for transcrip-

tional silencing, called the A-repeat (10), leads toa defect in Xist spreading (7) and spatial ex-clusion of active genes from the Xist-coated nu-clear compartment (9). Whether these structuralchanges are required for, or merely a conse-quence of, transcriptional silencing mediatedby the A-repeat of Xist remains unclear (7, 9).Recently, we and others identified by meansof mass spectrometry the proteins that inter-act with Xist (11–13). One of these proteins isthe Lamin B receptor (LBR) (11, 13), a trans-membrane protein that is anchored in theinner nuclear membrane, binds to Lamin B,

468 28 OCTOBER 2016 • VOL 354 ISSUE 6311 sciencemag.org SCIENCE

RESEARCH | REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

and is required for anchoring chromatin to thenuclear lamina (14)—a nuclear compartment thathelps shape the 3D structure of DNA (15) andis enriched for silencing proteins (14, 16). Be-cause induction of XCI leads to recruitment ofthe inactive X-chromosome to the nuclear lamina(4), we hypothesized that the Xist-LBR interactionmight be required to shape nuclear structure andregulate gene expression during XCI.To test this, we knocked down LBR and mea-

sured the expression of five X chromosome genesand two autosomal genes before and after Xistinduction (supplementary materials, materialsand methods, note 1). Knockdown of LBR ledto a defect in Xist-mediated silencing of theseX chromosome genes but showed no effecton autosomal genes (Fig. 1A and figs. S1 andS2). We observed a similar silencing defect uponknockdown or knockout of LBR in differen-tiating female embryonic stem cells (fig. S3and note 2). This silencing defect is not merelycaused by disruption of the nuclear lamina be-cause knockdown of Lamin B1 or Emerin, addi-tional components of the nuclear lamina (14),had no effect on Xist-mediated silencing (Fig. 1Aand fig. S4).We hypothesized that the arginine-serine (RS)

motif of LBR might be required for interactingwith Xist (fig. S4A and note 3). A truncated LBRprotein containing a deletion of the RS motif(DRS-LBR) (fig. S5 and materials and methods)did not interact with Xist (~97% reduced bind-ing) (Fig. 1B and materials and methods) andfailed to rescue the silencing defect upon knock-down of LBR (Fig. 1C and figs. S1B and S6). Incontrast, deletion of seven of the eight trans-

membrane domains (DTM-LBR) (fig. S5) did notaffect Xist binding (Fig. 1B) and was able torescue the silencing defect upon knockdown ofLBR (Fig. 1C, figs. S1B and S6, and note 4). Toensure that DRS-LBR fails to silence X chromo-some genes because of its RNA binding ability,we fused three copies of the viral BoxB RNAaptamer, which binds tightly to the viral lN coatprotein (17), to the 3′ end of the endogenousXist RNA (Xist-BoxB) (fig. S7). Expression ofDRS-LBR-lN in Xist-BoxB cells rescued the si-lencing defect observed upon LBR knockdown(Fig. 1D). Together, these results demonstratethat the Xist-LBR interaction is required forXist-mediated transcriptional silencing.We identified three discrete LBR binding sites

(LBSs) that are spread across >10,000 nucleo-tides of the Xist RNA (Fig. 2A and materials andmethods, note 5). One LBR binding site (LBS-1)overlaps the ~900 nucleotide region of Xist re-quired for silencing (DA-repeat region) (Fig. 2A)(10). We tested LBR binding within a DA-repeatcell line (10) and found that LBR binding is dis-rupted across the entire Xist RNA (Fig. 2B), in-cluding the LBR binding sites that do not overlapthe DA-repeat region (fig. S8). Because the Xist-binding protein SMRT/HDAC1–associated re-pressor protein (SHARP, also called Spen) alsobinds within the DA-repeat region (Fig. 2A) (12, 18)and its binding is also disrupted in DA-Xist (Fig.2B) (12), we generated a mutant Xist that pre-cisely deletes a region within the LBR bindingsite that is not within the SHARP binding site(DLBS-Xist) (Fig. 2A). In DLBS-Xist, LBR bindingwas lost across the entire Xist RNA without af-fecting SHARP binding (Fig. 2B and fig. S8). DLBS-Xist fails to silence X chromosome transcriptionto a similar level, as observed in the DA-Xist (Fig.2C, figs. S1 and S9, and notes 6 and 7).To ensure that the observed silencing defect

in DLBS-Xist cells is due to LBR-binding aloneand not disruption of another unknown proteininteraction, we tested whether we could rescue

the observed silencing defect by reestablishingthe DLBS-LBR interaction. To do this, we gener-ated an endogenous DLBS-BoxB Xist RNA (mate-rials and methods) and confirmed that expressionof LBR-lN, but not LBR fused to a differentRNA-binding domain (MS2-coat protein) (19), wasable to rescue the silencing defect observed inDLBS-BoxB cells (Fig. 2D, fig. S10, and materialsand methods). In contrast, expression of othersilencing proteins fused to lN, such as SHARPand embryonic ectoderm development (EED),did not rescue the observed silencing defect (Fig.2D and fig. S11), demonstrating that the LBR-binding site that overlaps the DA-repeat regionof Xist is required for silencing.We hypothesized that the Xist-LBR interaction

might be required for recruiting the inactive Xchromosome to the nuclear lamina (4). To testthis, we measured the distance between the Xist-coated nuclear compartment and Lamin B1 inthe nucleus using RNA fluorescence in situ hy-bridization (FISH), X chromosome paint, andimmunofluorescence (fig. S12A and materialsand methods). Upon Xist induction in wild-typecells, we found that the Xist compartment over-laps Lamin B1 signal (~90% of cells) (Fig. 3A,figs. S12 and S13, and note 8). In contrast, uponLBR knockdown or knockout, DLBS-Xist, or DA-Xist cells, there was a clear separation betweenthe Xist-coated compartment and Lamin B1, dem-onstrating a >20-fold increase in distance rel-ative to wild-type Xist (Fig. 3 and figs. S12 andS13). Thus, recruitment of the inactive X chro-mosome to the nuclear lamina is directly me-diated by the Xist RNA through its interactionwith LBR.To determine whether LBR-mediated recruit-

ment of the X chromosome to the nuclear laminaleads to Xist-mediated transcriptional silencing,we replaced the Xist-LBR interaction with an-other protein that interacts with the nuclearlamina; specifically, we used our endogenousDLBS-BoxB Xist, which fails to interact with LBR,

SCIENCE sciencemag.org 28 OCTOBER 2016 • VOL 354 ISSUE 6311 469

1Division of Biology and Biological Engineering, CaliforniaInstitute of Technology, Pasadena, CA 91125, USA.2European Molecular Biology Laboratory–Monterotondo, ViaRamarini 32, 00015 Monterotondo (RM), Italy. 3Program inMolecular Medicine, University of Massachusetts MedicalSchool, Worcester, MA 01655, USA.*Corresponding author. Email: mguttman@caltech.edu

Atr

x E

xpre

ssio

n (+

Dox

/-D

ox)

0.0

0.2

0.4

0.6

0.8

1.0

1.2

WT siLBR sgLBRsiLMNB1 siEMD

NSNS

****

****

Xis

t RN

A e

nric

hmen

tLB

R IP

/Inpu

t

WT0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

∆RS

****

∆TM

NS

siLBR+∆RS

siLBR+WT

****

siLBR+∆TM

Atr

x E

xpre

ssio

n (+

Dox

/-D

ox)

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

****

NS

GFP-λN

LBR-λN

∆RS-λN

0.0

0.2

0.4

0.6

0.8

1.0

1.2

**** ****

Atr

x E

xpre

ssio

n (+

Dox

/-D

ox)

Xist-BoxB + siLBR

Fig. 1. LBR requires its RS motif to interact with Xist and silence tran-scription. (A) Atrx mRNA levels after Xist induction (+dox) relative to pre-Xist (–dox) levels upon knockdown of various nuclear lamina proteins. WT,scrambled small interfering RNA (siRNA) control; siEMD, Emerin knock-down; siLMNB1, Lamin B1 knockdown; sgLBR, knockdown of LBR by usingdCas9-KRAB (materials and methods). (B) Xist enrichment after immuno-precipitation of a 3x-FLAG–tagged full-length LBR (WT), DRS-LBR, or DTM-LBR(materials and methods). Error bars indicate SEM from three independent IP

experiments. (C) Relative Atrx mRNA expression upon knockdown of the en-dogenous LBR and expression of full length LBR (WT), DTM-LBR, or DRS-LBR.(D) Relative Atrx mRNA expression in Xist-BoxB cells after knockdown ofthe endogenous LBR and expression of green fluorescent protein (GFP)–lN(control), LBR-lN, or DRS-LBR-lN. NS, not significant. ****P < 0.001 relativeto [(A) and (B)] wild-type cells, (C) cells transfected with siRNAs alone [shownin (A)], or (D) cells transfected with GFP-lN by means of an unpaired two-sample t test. Error bars indicate SEM across 50 individual cells.

RESEARCH | REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

to create an interaction between Xist and LaminB1 (fig. S14A). We expressed a Lamin B1-lN fusionprotein and confirmed that in cells expressingDLBS-BoxB Xist, the Xist-compartment was re-cruited to the nuclear lamina to a similar extentas that observed in wild-type conditions (Fig. 3,A and B, and fig. S12). Tethering Xist to the nu-clear lamina rescues the Xist-silencing defectobserved in DLBS cells to the same extent asthat observed after rescuing directly with LBR-lN (Fig. 3C and fig. S14). Thus, Xist-mediatedrecruitment of the X chromosome to the nuclearlamina is required for Xist-mediated transcrip-tional silencing, and the function of LBR in Xist-mediated silencing is to recruit the X chromosometo the nuclear lamina.We considered the possibility that recruitment

to the nuclear lamina, a nuclear territory enrichedfor silenced DNA and repressive chromatin regu-lators (14, 16), may act to directly silence tran-scription on the X chromosome (20, 21). To testthis, we knocked down SHARP, which failsto silence transcription on the X chromosome(11, 12, 18, 22), and observed that the Xist-coatedcompartment was still localized at the nuclearlamina, demonstrating a comparable distancedistribution between Xist and Lamin B1 to thatobserved for wild-type Xist (Fig. 3, A and B, andfig. S12). Therefore, Xist-mediated recruitment ofthe X chromosome to the nuclear lamina doesnot directly lead to transcriptional silencing be-cause the X chromosome can still be transcribedeven when localized at the nuclear lamina.Instead, we considered the possibility that

LBR-mediated recruitment of the X chromosometo the nuclear lamina could reposition activegenes into the Xist-coated nuclear compartment,allowing Xist to spread across the X chromosome.Indeed, the Xist RNA gradually localizes to genesthat are actively transcribed before initiation ofXCI (7), but deletion of the A-repeat leads to a

470 28 OCTOBER 2016 • VOL 354 ISSUE 6311 sciencemag.org SCIENCE

∆LBS-BoxB

EED-λN

SHARP-λN

LBR-λN

GFP-λN

LBR-MCP

Atr

x E

xpre

ssio

n (+

Dox

/-D

ox)

0.0

0.2

0.4

0.6

0.8

1.0

1.2

LBS-2 LBS-3SHARP

LBR

PTBP1

Input

WTA B C D EF

LBS-10

15

0

15

0

15

0

15

∆LBS

∆A

Xist

Atr

x E

xpre

ssio

n (+

Dox

/-D

ox)

WT ∆LBS∆A0.0

0.2

0.4

0.6

0.8

1.0

1.2**** ****

***

NS NSNS

Xis

t enr

ichm

ent

IP/In

put

WT ∆LBS∆A0.0

0.2

0.4

0.6

0.8

1.0

1.2 IP-SHARP

IP-LBRNS

********

****

Fig. 2. LBR binds to precise regions of the Xist RNA that are required for silencing. (A) Cross-linkingimmunoprecipitation (CLIP) data plotted across the Xist RNA for LBR, SHARP, and PTBP1 proteins. Thevalues represent fold-enrichment at each position on Xist normalized to a size-matched input RNAcontrol. Input represents the total RNA control for the LBR sample. (Bottom) A schematic of the annotatedrepeat regions on the Xist RNA (WT) and the locations of the deleted regions in DA (nucleotides 1–937) andDLBS (nucleotides 898–1682). (B) Xist RNA enrichment level measured with quantitative reverse tran-scription polymerase chain reaction after immunoprecipitation of endogenous LBR or SHARP in wild-type,DA, or DLBS cells. Error bars indicate SEM from four independent immunoprecipitation experiments.(C) Relative Atrx mRNA expression in wild-type, DA, or DLBS-Xist cells. (D) Expression of DLBS-Xist with a3x-BoxB fusion (DLBS-BoxB) along with expression of GFP-lN (control), EED-lN, SHARP-lN, or LBR-lN. Asan additional control, we expressed LBR fused with the bacteriophage MS2 coat protein (LBR-MCP). Errorbars indicate SEM across 50 individual cells. NS, not significant. ***P < 0.005, ****P < 0.001 relative to wild-type cells [(B) and (C)], or cells transfected with GFP-lN (D) by means of an unpaired two-sample t test.

Atr

x E

xpre

ssio

n (+

Dox

/-D

ox)

LBR-MCP

LaminB1-λN

LBR-λN

∆LBS-BoxB

****

0.0

0.2

0.4

0.6

0.8

1.0

1.2

****

Wild type ∆ LBS siSHARP∆LBS-BoxB +Lamin B1-λN∆ AsiLBR

0.0Normalized distance Xist-Lamin B1

(d/nuclear radius)

Cum

ulat

ive

freq

uenc

y

0.0

0.2

0.4

0.6

0.8

1.0

Wild type∆LBS

siSHARP

siLBR∆A

∆LBS-BoxB +Lamin B1-λN

0.50.40.30.20.1

Fig. 3. Xist-mediated recruitment of DNA to thenuclear lamina is required for transcriptional silenc-ing. (A) Images of Xist (red), Lamin B1 (green), and4′,6-diamidino-2-phenylindole (DAPI) (blue) across dif-ferent conditions. Scale bars, 5 mm. (B) The cumulativefrequency distribution of normalized distances betweenXist and Lamin B1 across 40 individual cells acrossdifferent conditions. Dashed lines represents a secondindependent experiment. (C) Relative AtrxmRNAexpres-sion in DLBS-BoxB cells along with expression ofLBR-MCP (control), LBR-lN, or LaminB1-lN. Error barsindicate SEM across 50 individual cells. NS, not signif-icant. ****P < 0.001 relative to cells transfected withLBR-MCP bymeans of an unpaired two-sample t test.

RESEARCH | REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

defect in Xist spreading to these actively tran-scribed regions (7, 9). In DLBS-Xist cells or uponknockdown of LBR, we observed a strong deple-tion of Xist RNA localization across regions ofactively transcribed genes, comparable with thedefect observed in DA-Xist cells (Fig. 4, A and B,and figs. S15 and S16) (7). We found that XistRNA localization is even more strongly depletedover more highly transcribed genes (Fig. 4B).Knockdown of SHARP, which also binds the A-repeat, did not affect Xist localization (Fig. 4Band fig. S15). Synthetically tethering DLBS-BoxBto the nuclear lamina by using a Lamin B1-lNfusion enables Xist to spread to active genes toa similar level as that observed in wild-type con-ditions (Fig. 4B and figs. S15 and S16).To determine whether this spreading defect

is due to a failure to reposition actively tran-scribed genes into the Xist-coated compartment,

we measured the position of the genomic loci ofthree actively transcribed genes relative to theXist-coated compartment (Fig. 4C and materialsand methods, note 9). In DLBS cells or uponknockdown or knockout of LBR, the distancebetween the Xist compartment and the loci ofthese actively transcribed X chromosome genes(Gpc4, Mecp2, and Pgk1 loci) were comparablewith the distance between Xist and an auto-somal gene (Notch2 locus) (Fig. 4, D and E, andfigs. S17 and S18). Upon knockdown of SHARP,we found that these actively transcribed loci over-lapped the Xist compartment (~80% of cells)(fig. S17), comparable with the Xist genomiclocus itself (~90% of cells) (fig. S17). Because Xistcan still spread to active genes upon knockdownof SHARP, which is known to be required forthe exclusion of RNA PolII (11), our results dem-onstrate that spreading to active genes and ex-

clusion of RNA PolII are independent functionsthat are both required for chromosome-widetranscriptional silencing.Our results suggest a model for how Xist

shapes the 3D nuclear structure of the inactiveX chromosome to spread to active genes andsilence chromosome-wide transcription (Fig. 4Fand fig. S19). Xist initially localizes to the core ofthe X chromosome territory by localizing atDNA sites that are in close 3D proximity to itstranscriptional locus (7). These initial Xist lo-calization sites are generally inactive before Xistinduction (6, 7, 9). The Xist-coated DNA, likeother chromosomal DNA regions, will dynam-ically sample different nuclear locations (23)and, because Xist binds LBR, will becometethered at the nuclear lamina when it comesinto spatial proximity. This lamina associationis known to constrain chromosomal mobility

SCIENCE sciencemag.org 28 OCTOBER 2016 • VOL 354 ISSUE 6311 471

siLB

R∆ A

siS

HA

RP

∆LB

S

1 hr

0

100

0

100

0.5

1

Xis

t enr

ichm

ent

Enr

ichm

ent

ratio

WT

∆LBS

∆LBS vs WT

Active GenesInactive Genes

chrX:45,354,236-46,363,849

+100 kb

−100 kb +100 kbGene

∆LB

S v

s W

T

1/4

1

All Active GenesInactive Genes

−100 kb Gene

∆LB

S-B

oxB

+

Lam

inB

1-λN

vs

WT

SH

AR

P K

D v

s W

T

−100 kb +100 kbGene

1/4

1

Highest Active Genes

1/4

1

Xist

Chr X

LBR

Active

POL IIPOL IL Gene

Xist

Chr X

Inactive

SAF-ASHARP

SMRT

HDAC3

t

16 hr

Gpc4/Xist

**

0.0

0.2

0.4

0.6

Dis

tanc

e be

twee

n X

ist

and

Gpc

4(d

ista

nce/

cell

radi

us)

1 hr 3 hr 6 hr 16 hr

Xist induction time

siSHARP∆A

siLBR

∆LBS**

**

Dis

tanc

e be

twee

n X

ist

and

targ

et g

enes

(dis

tanc

e/ce

ll ra

dius

)

0.0

0.4

siSHARP ∆A siLBR ∆LBS

Mecp2 (chrX)

0.6

0.2

siSHARP ∆A siLBR ∆LBS

Pgk1 (chrX)

*** ****** ***

*****

Fig. 4. Recruitment to the nuclear lamina is required for Xist spreading to active genes. (A) Xist RNA localization asmeasured with RNA antisense purification (RAP)–DNA for wild type (top), DLBS-Xist (middle), and the smoothed foldchange (bottom) across a region of the X chromosome containing active (red) and inactive (blue) genes. The dashed lineindicates average Xist enrichment in wild-type cells. (B) Aggregate Xist enrichment relative to the genomic locations ofhighly active genes [dark red, reads per kilobase per million (RPKM) > 5], all active genes (red, RPKM > 1), and inactivegenes (blue) on the X-chromosome for DLBS, SHARP knockdown, and DLBS-BoxB + LMNB1-lN cells compared with wild-type cells. Shaded areas represent 95% confidence interval. (C) Images of Xist (red), Gpc4 locus (green), and DAPI (blue)across different cell lines (rows) after Xist induction for 1 or 16 hours. Yellow arrowheads indicate the genomic DNA locationof Gpc4. Scale bars, 5 mm. (D) The median distance from Gpc4 locus to the Xist-compartment after Xist induction for 1, 3,6, and 16 hours. Error bars represent the standard error of the median across 50 individual cells. **P < 0.01 relative to 1-hourinduction by means of an unpaired two-sample t test. (E) The median distance from the Mecp2 and Pgk1 locus to the Xistcompartment after Xist induction for 16 hours across different conditions. Error bars represent the standard error of themedian across 50 individual cells. **P < 0.01, ***P < 0.005 relative to siRNAs targeting SHARP (siSHARP) by means of an unpaired two-sample t test. (F) Amodel for how Xist-mediated recruitment to the nuclear lamina enables spreading to active genes and transcriptional silencing on the X chromosome.

RESEARCH | REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

(24) and by doing so would position the Xist-coated DNA away from the actively transcribedXist transcription locus. This would enable otherDNA regions on the X chromosome, which arephysically linked to these tethered regions, tobe brought into closer spatial proximity of theXist transcription locus. In this way, Xist and itssilencing factors can spread to these newlyaccessible DNA regions on the X chromosome.

REFERENCES AND NOTES

1. R. Galupa, E. Heard, Curr. Opin. Genet. Dev. 31, 57–66 (2015).2. E. Splinter et al., Genes Dev. 25, 1371–1383 (2011).3. S. S. Rao et al., Cell 159, 1665–1680 (2014).4. A. Rego, P. B. Sinclair, W. Tao, I. Kireev, A. S. Belmont, J. Cell

Sci. 121, 1119–1127 (2008).5. A. Wutz, Nat. Rev. Genet. 12, 542–553 (2011).6. C. M. Clemson, L. L. Hall, M. Byron, J. McNeil, J. B. Lawrence,

Proc. Natl. Acad. Sci. U.S.A. 103, 7688–7693 (2006).7. J. M. Engreitz et al., Science 341, 1237973 (2013).8. M. D. Simon et al., Nature 504, 465–469 (2013).9. J. Chaumeil, P. Le Baccon, A. Wutz, E. Heard, Genes Dev. 20,

2223–2237 (2006).10. A. Wutz, T. P. Rasmussen, R. Jaenisch, Nat. Genet. 30, 167–174

(2002).11. C. A. McHugh et al., Nature 521, 232–236 (2015).12. C. Chu et al., Cell 161, 404–416 (2015).13. A. Minajigi et al., Science 349, aab2276 (2015).14. Y. Gruenbaum, A. Margalit, R. D. Goldman, D. K. Shumaker,

K. L. Wilson, Nat. Rev. Mol. Cell Biol. 6, 21–31 (2005).15. A. Pombo, N. Dillon, Nat. Rev. Mol. Cell Biol. 16, 245–257

(2015).16. J. Kind, B. van Steensel, Curr. Opin. Cell Biol. 22, 320–325 (2010).17. J. Baron-Benhamou, N. H. Gehring, A. E. Kulozik, M. W. Hentze,

Methods Mol. Biol. 257, 135–154 (2004).18. A. Monfort et al., Cell Reports 12, 554–561 (2015).19. A. R. Buxbaum, G. Haimovich, R. H. Singer, Nat. Rev. Mol. Cell

Biol. 16, 95–109 (2015).20. L. E. Finlan et al., PLOS Genet. 4, e1000039 (2008).21. K. L. Reddy, J. M. Zullo, E. Bertolino, H. Singh, Nature 452,

243–247 (2008).22. B. Moindrot et al., Cell Rep. 10.1016/j.celrep.2015.06.053

(2015).23. W. F. Marshall et al., Curr. Biol. 7, 930–939 (1997).24. J. R. Chubb, S. Boyle, P. Perry, W. A. Bickmore, Curr. Biol. 12,

439–445 (2002).

ACKNOWLEDGMENTS

We thank K. Plath for extensive discussions; A. Collazo formicroscopy help; A. Shur, P. Quintero, and V. Grishkevich fortechnical help; M. Lai for analytical help; J. Engreitz, S. Quinodoz,M. Garber, I. Amit, and J. Rinn for comments on the manuscript;and S. Knemeyer for illustrations. Imaging was performedin the Biological Imaging Facility, and sequencing was performed inthe Millard and Muriel Jacobs Genetics and Genomics Laboratoryat the California Institute of Technology. C.-K.C. is supported by aNIH National Research Service Award training grant(T32GM07616). This research was funded by the New York StemCell Foundation, a NIH Director’s Early Independence Award(DP5OD012190), the Edward Mallinckrodt Foundation,Sontag Foundation, Searle Scholars Program, Pew-StewardScholars program, and funds from the California Institute ofTechnology. M.G. is a New York Stem Cell Foundation–RobertsonInvestigator. Sequencing data are available online from theNational Center for Biotechnology Information Gene ExpressionOmnibus (www.ncbi.nlm.nih.gov/geo) accession no. GSE80510(RAP data) and GSE86250 (CLIP data), and additional data andinformation are available at www.lncRNA.caltech.edu/data.php.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/354/6311/468/suppl/DC1Materials and MethodsSupplementary TextFigs. S1 to S19References (25–49)

5 December 2015; accepted 25 July 2016Published online 4 August 201610.1126/science.aae0047

VASCULAR DISEASE

Senescent intimal foam cellsare deleterious at all stagesof atherosclerosisBennett G. Childs,1 Darren J. Baker,2 Tobias Wijshake,2,3 Cheryl A. Conover,4

Judith Campisi,5,6 Jan M. van Deursen1,2*

Advanced atherosclerotic lesions contain senescent cells, but the role of these cellsin atherogenesis remains unclear. Using transgenic and pharmacological approachesto eliminate senescent cells in atherosclerosis-prone low-density lipoproteinreceptor–deficient (Ldlr–/–) mice, we show that these cells are detrimental throughoutdisease pathogenesis. We find that foamy macrophages with senescence markersaccumulate in the subendothelial space at the onset of atherosclerosis, where they drivepathology by increasing expression of key atherogenic and inflammatory cytokinesand chemokines. In advanced lesions, senescent cells promote features of plaqueinstability, including elastic fiber degradation and fibrous cap thinning, by heighteningmetalloprotease production. Together, these results demonstrate that senescent cells arekey drivers of atheroma formation and maturation and suggest that selective clearanceof these cells by senolytic agents holds promise for the treatment of atherosclerosis.

Atherosclerosis initiates when oxidized lipo-protein infiltrates the subendothelial spaceof arteries, often due to aberrantly elevatedlevels of apolipoprotein B–containing lipo-proteins in the blood (1). Chemotactic signals

arising from activated endothelium and vascu-lar smooth muscle attract circulating monocytesthat develop into lipid-loaded foamy macrophages,a subset of which adopt a proinflammatory pheno-type through a mechanism that is not fully un-derstood. The proinflammatory signals lead toadditional rounds of monocyte recruitment andaccumulation of other inflammatory cells (in-cluding T and B cells, dendritic cells, and mastcells), allowing initial lesions, often termed “fattystreaks,” to increase in size and develop intoplaques (2). Plaque stability, rather than abso-lute size, determines whether atherosclerosisis clinically silent or pathogenic because un-stable plaques can rupture and produce vessel-occluding thrombosis and end-organ damage.Stable plaques have a relatively thick fibrouscap, which largely consists of vascular smoothmuscle cells (VSMCs) and extracellular matrixcomponents, partitioning soluble clotting fac-tors in the blood from thrombogenic moleculesin the plaque (3). In advanced disease, plaquesdestabilize when elevated local matrix metallo-protease production degrades the fibrous cap,

increasing the risk of lesion rupture and subse-quent thrombosis.Advanced plaques contain cells with markers

of senescence, a stress response that entails a per-manent growth arrest coupled to the robust se-cretion of numerous biologically active moleculesand is referred to as the senescence-associatedsecretory phenotype (SASP). The senescencemarkers include elevated senescence-associatedb-galactosidase (SA b-Gal) activity and p16Ink4a,p53, and p21 expression (4, 5). However, whetherand how senescent cells contribute to athero-genesis remains unclear (6, 7). Human plaquescontain cells with shortened telomeres, whichpredispose cells to undergo senescence (8). Con-sistent with a proatherogenic role of senescenceis the observation that expression of a loss-of-function telomere-binding protein (Trf2) in VSMCsaccelerates plaque growth in the ApoE–/– mousemodel of atherosclerosis, although in vivo evi-dence for increased senescence in plaques wasnot provided. On the other hand, mice lackingcore components of senescence pathways, suchas p53, p21, or p19Arf (7, 9–11), show acceleratedatherosclerosis, implying a protective role for se-nescence. Studies showing that human and mousepolymorphisms that reduce expression of p16Ink4a

and p14Arf (p19Arf in mice) correlate with in-creased atheroma risk support this conclusion(7, 12, 13). Thus, whether senescent cells accel-erate or retard atherogenesis is unclear.We used genetic and pharmacological meth-

ods of eliminating senescent cells to examine therole of naturally occurring senescent cells at dif-ferent stages of atherogenesis. First, we verifiedthat senescent cells accumulate in low-densitylipoprotein receptor–deficient (Ldlr–/–) mice,a model of atherogenesis. We fed 10-week-oldLdlr–/– mice a high-fat diet (HFD) for 88 days.We then performed SA b-Gal staining, which

472 28 OCTOBER 2016 • VOL 354 ISSUE 6311 sciencemag.org SCIENCE

1Department of Biochemistry and Molecular Biology, MayoClinic, Rochester, MN 55905, USA. 2Department of Pediatricand Adolescent Medicine, Mayo Clinic, Rochester, MN 55905,USA. 3Department of Pediatrics, University of Groningen,University Medical Center Groningen, 9713 AV Groningen,Netherlands. 4Division of Endocrinology, Metabolism, andNutrition, Mayo Clinic, Rochester, MN 55905, USA. 5BuckInstitute for Research on Aging, Novato, CA 94945, USA.6Life Sciences Division, Lawrence Berkeley NationalLaboratory, Berkeley, CA 94720, USA.*Corresponding author. Email: vandeursen.jan@mayo.edu

RESEARCH | REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

◥DNA METHYLATION

The DNA methyltransferase DNMT3Cprotects male germ cells fromtransposon activityJoan Barau,1 Aurélie Teissandier,1,2 Natasha Zamudio,1* Stéphanie Roy,3

Valérie Nalesso,4,5,6 Yann Hérault,4,5,6 Florian Guillou,3 Déborah Bourc’his1†

DNA methylation is prevalent in mammalian genomes and plays a central role in theepigenetic control of development.The mammalian DNA methylation machinery is thoughtto be composed of three DNA methyltransferase enzymes (DNMT1, DNMT3A, and DNMT3B)and one cofactor (DNMT3L). Here, we describe the discovery of Dnmt3C, a de novo DNAmethyltransferase gene that evolved via a duplication of Dnmt3B in rodent genomes and waspreviously annotated as a pseudogene.We show that DNMT3C is the enzyme responsible formethylating the promoters of evolutionarily young retrotransposons in the male germ lineand that this specialized activity is required for mouse fertility. DNMT3C reveals the plasticityof themammalianDNAmethylation systemand expands the scope of themechanisms involvedin the epigenetic control of retrotransposons.

Genome defense via transcriptional silenc-ing of transposable elements has beenproposed to be a driving force for theevolution of DNAmethylation (1). Retro-transposons occupy half of mammalian

genomic space, and their control is of paramountimportance in the germ line: Their activity candamage the hereditary material with an impact

on fertility and the fitness of subsequent genera-tions (2). In mammals, after germline epigenet-ic reprogramming, small RNA-directed DNAmethylation establishes life-long epigenetic si-lencing of retrotransposons during the perinatalperiod of spermatogenesis (3). Piwi-interactingRNAs (piRNAs) are the cleavageproducts of retro-transposon transcripts and guide DNA methyla-

tion to the promoters of these elements throughhomology recognition (4, 5). Mammals have spe-cifically evolved a catalytically inactiveDNAmethyl-transferase (DNMT) cofactor,DNMT3L, which actsdownstream of the piRNA pathway (6, 7). Theinactivation of DNMT3L or PIWI-pathway pro-teins invariably results in hypomethylation andreactivation of retrotransposons, meiotic failure,azoospermia, and male sterility marked by smalltestis size (hypogonadism) (8).To gain insights into the biology of retrotrans-

poson silencing in the germ line, we screened acollection of hypogonadal male mice generatedthroughN-ethyl-N-nitrosourea (ENU)mutagenesisfor ectopic retrotransposon activity (fig. S1, A andB). Five independent positive lines were obtained,but all showed linkage to the same genomicinterval on chromosome 2 (fig. S1C), suggestingthat they shared a spontaneous, ENU-independentmutation. Whole-genome sequencing revealed ade novo insertion of an IAPEz element, a subclassof inracisternalA-particle (IAP) retrotransposon, inthe last intron of the Gm14490 gene (Fig. 1A andfig. S1, D to F). Gm14490 maps 9 kilobases (kb)

SCIENCE sciencemag.org 18 NOVEMBER 2016 • VOL 354 ISSUE 6314 909

Fig. 1. Gm14490 encodes a male germ cell–specific de novo DNAmethyltransferase, DNMT3C.(A) Structure of Gm14490 (Ensembl 2011) and po-sition of the IAPEz insertion (antisense orienta-tion). RACE and RNA-seq analysis of E16.5 andP10 testis identifies a long isoform with codingpotential (ATG, green triangle). (B) Gm14490 is de-tected in testis by reverse transcription–quantitativepolymerase chain reaction (RT-qPCR), but not ingerm cell–depleted testis from Dnmt3LKO/KO ani-mals. Tissues from 10-week-old mice, unless other-wise specified. (C) RT-qPCR of the two Gm14490isoforms during testis development. Predominantgerm cell populations are represented. Primordialgerm cells (PGC), spermatogonial stem cells (SSC),and spermatogonia (Spg). (D) DNMT3C showscharacteristics of DNMT3 proteins—conservedmethyltransferase (MTase) motifs and an ADD do-main, but no PWWP domain. (E) RNA-seq support-ing wild-type and mutant Dnmt3C IAP splicingevents in E16.5 testis. (F) LUminometric methyla-tion assay (LUMA) of global CpG methylation inDnmt-tKO ESCs that transiently express Dnmt3C-and Dnmt3B-3XFLAG alleles. Data are mean ±SD from three technical replicates in (B) and (C)and from three biological replicates in (F). nd, notdetectable.

1Institut Curie, PSL Research University, INSERM, CNRS,75005 Paris, France. 2Ecole des Mines, 75005 Paris, France.3PRC, INRA, CNRS, IFCE, Tours University, 37380 Nouzilly,France. 4Institut de Génétique et Biologie Moléculaire etCellulaire, Strasbourg University, 67404 Illkirch, France.5Centre National de la Recherche Scientifique, UMR7104,Illkirch, France. 6Institut National de la Santé et de laRecherche Médicale, U964, Illkirch, France.*Present address: The Walter and Eliza Hall Institute of MedicalResearch, Parkville, Victoria, Australia. †Corresponding author.Email: deborah.bourchis@curie.fr

RESEARCH | REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

downstream of the Dnmt3B gene and was an-notated as a nonfunctional tandemduplication ofDnmt3B, based on lack of transcription and recog-nizable open reading frames (ORFs) (9).We found that Gm14490 was exclusively ex-

pressed in male germ cells (Fig. 1B and fig. S2A).Using RNA sequencing (RNA-seq) and rapid am-plification of cDNA ends (RACE), we annotatedtwo transcript isoforms, whose expression wastightly regulated during spermatogenesis (Fig. 1,A and C). The short, noncoding isoform was ex-pressed in postnatal testes. However, the long iso-form (2.8 kb) possessed a 709-codon ORF, and itsexpression sharply peaked around embryonicday 16.5 (E16.5), coinciding with male germlinede novo DNAmethylation (5, 10). In comparison,Dnmt3B is expressed in germ cells, early embryos,and somatic tissues of both sexes (11, 12). Usingdiscriminating primers, we showed independentregulation ofDnmt3B and Gm14490 during sper-matogenesis (fig. S2B). The coding potential and

specific developmental regulation of Gm14490led us to reconsider it as a functional paralog ofDnmt3B rather than a pseudogene, and thus werenamed it Dnmt3C.The long Dnmt3C isoform encodes a protein

with an organization characteristic of DNMT3enzymes: six methyltransferase motifs (I, IV, VI,VIII, IX, and X) in C-terminal position and anN-terminal ATRX-DNMT3L-DNMT3A (ADD) do-main, which binds unmethylated lysine 4 resi-dues of histone H3 (H3K4) (Fig. 1D and fig. S2C)(13). However, DNMT3C lacks the Pro-Trp-Trp-Pro (PWWP) domain, which targets DNMT3 pro-teins to gene bodies through recognition ofH3K36 trimethylation (H3K36me3) (14, 15). Over-all, DNMT3C exhibits 70% identity with DNMT3B,while DNMT3A and DNMT3B are 46% identical.In hypogonadalmutants, the IAP insertion did notaffect Dnmt3C transcript levels but provided analternative splice acceptor site, which led to theexclusion of Dnmt3C last exon in favor of the

retrotransposon sequence in a chimeric Dnmt3C-IAPmRNA (Fig. 1E and fig. S2, D and E). Its pre-dicted translation product lacks motifs IX and X(fig. S2C), which are essential for the AdoMet-dependent methyltransferase fold and for thebinding of the methyl donor S-adenosyl methi-onine, respectively (16).To demonstrate that DNMT3C is catalytically

active, we performed an in vivo DNA methyla-tion assay. Dnmt3C is not expressed in mouseembryonic stem cells (ESCs) (Fig. 1B). By trans-fecting constructs driving Dnmt3C expressionin DNAmethylation-free ESCs (Dnmt1, Dnmt3A,and Dnmt3B triple-knockout; Dnmt-tKO) (17), weobserved a gain of CpG methylation (10 to 20%),similar to that observed upon Dnmt3B transfec-tion (Fig. 1F and fig. S2, F and G). The mutantDnmt3CIAP allele failed to raise CpGmethylationlevels, as didDnmt3C andDnmt3Bmutant alleleswith a missense mutation in the catalytic site(DNMT3C C507A and DNMT3BC658A, in whichcysteine at position 507 and 658 is replaced byalanine). An in vitroDNAmethylation assay usingthe DNMT3Cmethyltransferase domain showedconcordant results (fig. S2,H and I). These findingsdemonstrate that DNMT3C is an enzymaticallyactive member of the DNMT3 family of de novoDNA methyltransferases.Dnmt3CIAP/IAPanimalswere somaticallynormal,

and only males were sterile (fig. S3A). Hypo-gonadism was linked to azoospermia with inter-ruption of spermatogenesis at the pachytene stageof meiosis I, in the context of impaired chromo-some synapsis (Fig. 2, A to C). The developmentalphenotype ofDnmt3Cmutantmice was similar tothat observed in Dnmt3LKO/KO males (7, 18), sug-gesting that DNMT3C could be involved in trans-poson silencing during spermatogenesis. Indeed,the same set of retrotransposonswereup-regulatedinDnmt3CIAP/IAP andDnmt3LKO/KO testes at P20(postnatal day 20) (Fig. 2D and fig. S3B) (18). Longinterspersed nuclear elements (LINEs or L1s)showed the strongest reactivation, and morespecifically evolutionarily young subfamilies: typeA, T, and Gf transcripts were increased by 10-foldin Dnmt3CIAP/IAP testes, in association with ac-cumulation of L1-encodedORF1 proteins (Fig. 2E).Among endogenous retroviruses (ERVs), reac-tivation was specific to some ERVK families(MMERVK10C, IAPEz, and IAPEy). As in thecase of the Dnmt3L mutation, L1 and IAPEz de-repression was linked to a DNA methylationdefect in Dnmt3CIAP/IAP testes (Fig. 2F), despitenormal expression of piRNA/DNA methylationgenes and piRNA production during fetal sper-matogenesis (fig. S3, C to F). Finally, we con-firmedDNMT3C function by generating aDnmt3Cknockout mouse through CRISPR-Cas9–mediateddeletion (fig. S4A). TheDnmt3CKO allele recapitulatedtheDnmt3CIAP/IAP developmental andmolecularphenotypes in homozygous Dnmt3CKO/KO malesand failed to complement the Dnmt3CIAP allelein Dnmt3CIAP/KO compound heterozygous males(fig. S4, B to E).To assess the contribution of DNMT3C to

male germlinemethylation, we performedwhole-genome bisulfite sequencing (WGBS) in sorted

910 18 NOVEMBER 2016 • VOL 354 ISSUE 6314 sciencemag.org SCIENCE

Fig. 2. Phenotype of Dnmt3CIAP/IAP males. (A) Hypogonadism (6-week-old mice). Scale bars, 5 mm.(B) Severe germ cell loss in testis sections (11-week-old mice). Scale bars, 100 mm. (C) Impaired chro-mosome synapsis at meiosis as detected by immunofluorescence against synaptonemal complexproteins (SYCP1 and SYCP3). Scale bars, 4 mm. (D) RNA-seq heatmap shows overexpression of young L1and specific ERVK types in Dnmt3CIAP/IAP compared to Dnmt3CIAP/WT testis at P20. Annotations fromRepeatMasker. (E) Aberrant expression of L1-ORF1 proteins in Dnmt3CIAP/IAP germ cells (TRA98-positive)at P20. Scale bars, 50 mm. (F) L1A-5′UTR and IAPEz-5′LTR are hypomethylated in Dnmt3CIAP/IAP testisDNA at P20, as inDnmt3LKO/KO testis. Southern blot analysis aftermethyl-sensitive Hpa II digestion. Msp Iis used as a digestion control.

RESEARCH | REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

germ cells from testes at P10, when de novo DNAmethylation is completed. Overall CpG methyla-tion levels ofDnmt3CIAP/IAPmutant cells were notmarkedly different from the Dnmt3CIAP/WT con-trol (77.7 versus 78.5%). A slight decrease was onlyapparent when focusing on transposons (81.5versus 84.2%), and more specifically on LINEs,ERVK, and ERV1 (Fig. 3, A and B). Accordingly,there was only a limited number (264) of differen-tially methylated regions (DMRs) inDnmt3CIAP/IAP

versus Dnmt3C IAP/WT germ cells (fig. S5A); allreflected hypomethylation in the mutant, andmost overlapped with LINEs (34%) and ERVs(48%). RepeatMasker annotations further high-lighted that the same families that were tran-scriptionally derepressed were hypomethylatedin Dnmt3CIAP/IAP testes; namely, young L1s and

specific ERVKs (Fig. 2D and Fig. 3C). By com-parison, deletion of DNMT3L had a strongerand broader impact: Dnmt3LKO/KO germ cellsexhibited only 39% of global CpG methylation,and all genomic compartments and retrotrans-poson classes were affected (Fig. 3, D and E, andfig. S5B).The DNA methylation defect in Dnmt3C mu-

tants only reached 30% at themost for young L1s(Fig. 3C). We reasoned that DNMT3C selectivelyaffects transcriptionally active retrotransposoncopies, as these are targets of piRNA-dependentDNA methylation during fetal spermatogenesis(4, 5). Indeed, individual L1-A and -T elementswith a 5′ promoter (length >5 kb) and the highestsimilarity toward the consensus sequence showedthe greatestDNAmethylation loss inDnmt3CIAP/IAP

cells (Fig. 3F and fig. S4C). Additionally, DNMT3C-dependent DNAmethylation was not evenly dis-tributed across the length of these transcriptionallycompetent elements, but rather focalized to theirpromoters [5′ untranslated region (UTR)] (Fig.3G), in a pattern previously observed in piRNA-deficientMiliKO/KOmales (fig. S5D) (4). Older L1-Felements did not show such trends (Fig. 3F andfig. S5C). By contrast,Dnmt3L deficiency causeddemethylation of L1s independently of their ageand throughout their sequence (Fig. 3H).ERVKs exhibited milder methylation loss

(10%) than LINEs in Dnmt3C IAP/IAP cells (Fig.3C), and this was not related to the size or thesequence conservation of individual copies (fig.S5, C and E). ERVKs partially resist the genome-wide erasure of methylation that occurs in the

SCIENCE sciencemag.org 18 NOVEMBER 2016 • VOL 354 ISSUE 6314 911

Fig. 3. DNMT3Cmethylates evolution-arily young retrotransposon promo-ters. (A and B) Tukey box-plotrepresentation of CpGmethylation con-tent as determined by WGBS overdifferent (A) genomic compartmentsand (B) retrotransposon classes in con-trol Dnmt3C IAP/WT (purple) and mutantDnmt3C IAP/IAP (yellow) germ cells atP10. (C) Percentage of DNA methylationloss within individual retrotransposonfamilies in Dnmt3CIAP/IAP samples.(D and E) As in (A) and (B) but forDnmt3LKO/WT (green) and Dnmt3LKO/KO

(red) germ cells. (F) Plotting of DNAmethylation loss over individual copiesof L1 families according to genetic dis-tance from consensus sequences.(G andH) Metaplots of DNAmethylationover full-length L1s (>5 kb) comparing(G) Dnmt3CIAP/IAP and (H) Dnmt3LKO/KO

versus control samples. (I) Left: Heat-map of Dnmt3CIAP/IAP and control DNAmethylation levels across paternallyimprinted loci. Right: Methylation mapsat the Rasgrf1 locus.

RESEARCH | REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

fetal germ line (19, 20); this likely explains theirlimited dependency toward DNMT3C remethyla-tion activity. Nevertheless, DNMT3Cdependencywas still greatest over regulatory long-terminalrepeat (LTR) sequences ofMMERVK10C (fig. S5F)and IAPEz elements when highly conserved copieswere analyzed by bisulfite pyrosequencing (fig.S4E). Finally, among paternally methylated im-printed loci, only the Rasgfr1 imprinting controlregion (ICR)washypomethylated inDnmt3CIAP/IAP

germ cells (Fig. 3I). This ICR includes an ERVKLTR fragment, which acquires DNAmethylationin a piRNA-dependentmanner during spermato-genesis (21). This highlights again the genomicconvergence between piRNA and DNMT3C tar-geting in fetal male germ cells.DNMT3C is highly specialized at methylating

young retrotransposon promoters in the malegerm line. In comparison,DNMT3B is not involvedin germ cell development but establishes somaticmethylation patterns genome-wide in the earlyembryo (11, 22, 23). DNMT3Chas therefore evolvedits own regulatory pattern and genomic targets,which have diverged from the ancestral DNMT3Bcopy. Despite the strict requirement of DNMT3Cin the mouse, the Dnmt3C duplication is not uni-versal among mammals but occurred ~46millionyears ago in the last common ancestor of themuroid rodents (Fig. 4, A and B, and fig. S6, Aand B). These represent the largest mammaliansuperfamily, with 1518 species accounting for

28% of all extant mammals (fig. S5C) (24), andinclude the two primary models for biomedicalresearch, particularly in reproduction and endo-crinology: the mouse and the rat. The genomesofMuroidea carry a heavy burden in young trans-posons: 25% have integrated in the last 25 millionyears with currently thousands of active copies(25). In comparison, most human transposonsbecame extinct during that time (26).The lack of transposon methylation defects in

the germ cells of Dnmt3A or Dnmt3B mutantmice had been previously interpreted as a signof functional redundancy between these twoenzymes (23, 27). We show here that this func-tion relies insteadonanadditional denovoDNMT.ThemammalianDNAmethylationmembers shouldnow be considered as a quintet: DNMT1, 3A, 3B,3C, and 3L. The two most recent evolutionaryadditions are linked to reproduction: The euthe-rianDNMT3L cofactor stimulates germlinemeth-ylation genome-wide, and the muroid DNMT3Cenzyme methylates young retrotransposons dur-ing spermatogenesis. Future research should re-solve the biochemicalmechanism that designatesspecific sequences for DNMT3C-dependent DNAmethylation. Its PWWP-free statusmay prevent itfrom targeting H3K36me3-enriched gene bodiesand redirect it to retrotransposon promoters.In conclusion, our discovery of DNMT3C raisesa new set of challenges to the current views ofthe remarkable evolution of DNA methyltrans-

ferases, the regulation of the de novo methyl-ation process, and its tight links with the selectivepressure to epigenetically control transposons inthe germ line.

REFERENCES AND NOTES

1. J. A. Yoder, C. P. Walsh, T. H. Bestor, Trends Genet. 13,335–340 (1997).

2. J. L. Goodier, H. H. Kazazian Jr., Cell 135, 23–35 (2008).3. A. A. Aravin, D. Bourc’his, Genes Dev. 22, 970–975 (2008).4. S. A. Manakov et al., Cell Reports 12, 1234–1243 (2015).5. A. Molaro et al., Genes Dev. 28, 1544–1549 (2014).6. A. A. Aravin et al., Mol. Cell 31, 785–799 (2008).7. D. Bourc’his, T. H. Bestor, Nature 431, 96–99 (2004).8. N. Zamudio, D. Bourc’his, Heredity (Edinb) 105, 92–104

(2010).9. D. J. Lees-Murdock et al., Genomics 84, 193–204 (2004).10. H. Kobayashi et al., Genome Res. 23, 616–627 (2013).11. M. Okano, D. W. Bell, D. A. Haber, E. Li, Cell 99, 247–257

(1999).12. S. La Salle et al., Dev. Biol. 268, 403–415 (2004).13. S. K. T. Ooi et al., Nature 448, 714–717 (2007).14. T. Baubec et al., Nature 520, 243–247 (2015).15. A. Dhayalan et al., J. Biol. Chem. 285, 26114–26120

(2010).16. S. Klimasauskas, S. Kumar, R. J. Roberts, X. Cheng, Cell 76,

357–369 (1994).17. A. Tsumura et al., Genes Cells 11, 805–814 (2006).18. N. Zamudio et al., Genes Dev. 29, 1256–1270 (2015).19. P. Hajkova et al., Mech. Dev. 117, 15–23 (2002).20. S. Seisenberger et al., Mol. Cell 48, 849–862 (2012).21. T. Watanabe et al., Science 332, 848–852 (2011).22. G. Auclair, S. Guibert, A. Bender, M. Weber, Genome Biol. 15,

545 (2014).23. M. Kaneda et al., Nature 429, 900–903 (2004).24. J. Michaux, A. Reyes, F. Catzeflis, Mol. Biol. Evol. 18,

2017–2031 (2001).25. R. H. Waterston et al., Nature 420, 520–562 (2002).26. R. Cordaux, M. A. Batzer, Nat. Rev. Genet. 10, 691–703

(2009).27. Y. Kato et al., Hum. Mol. Genet. 16, 2272–2280 (2007).

ACKNOWLEDGMENTS

We thank members of D.B.’s laboratory, especially M. Greenbergand M. Walter. We are grateful to T. H. Bestor, V. Colot, E. Heard, andE. Mulugeta for advice; S. Fouchécourt, B. Fumel, L. Mourao, andB. Piegu for technical help; A. Bortvin and S. Tajima for antibodies;C. Jouhanneau for excellent mouse husbandry; E. Desale andPHENOMIN-TAAM for the mutagenesis program; and PICT-IBiSA(France–Bioimaging, ANR-10-INSB-04), the Flow Cytometry andAnimal Platforms of the Institut Curie. This research was supportedby grants from European Research Council (ERC-Cog EpiRepro),Agence Nationale pour la Recherche (ANR TranspoFertil, ANR-10-LABX-0030-INRT, and ANR-10-INBS-07 PHENOMIN). D.B.’slaboratory is part of the Laboratoire d’Excellence (LABEX) DEEP(11-LBX0044). J.B. was a recipient of a postdoctoral fellowship fromAgence Nationale de Sécurité Sanitaire (Anses). The Institut CurieICGex NGS platform was supported by ANR-10-EQPX-03 (Equipex)and ANR-10-INBS-09-08 (France Génomique Consortium) grantsand by the Canceropôle Ile-de-France. Mouse line distribution underuniform biological material transfer agreement. All sequencing datasets are available at the National Center for Biotechnology InformationGene Expression Omnibus database under GSE84141.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/354/6314/909/suppl/DC1Materials and MethodsFigs. S1 to S6Tables S1 to S7References (28–54)

8 July 2016; accepted 19 October 201610.1126/science.aah5143

912 18 NOVEMBER 2016 • VOL 354 ISSUE 6314 sciencemag.org SCIENCE

Fig. 4. Phylogenetic dis-tribution of Dnmt3C.(A) Locus organization andsynteny around the Dnmt3B(blue) and Dnmt3C (red)genes in human, rabbit,mouse, and rat genomes.(B) Dnmt3C distribution inthe mammalian phyloge-netic tree with a focus on theRodentia order. Species indetail have availablesequenced genomes. Redbranches: Muroidea familiesin which Dnmt3C was iden-tified; dashed branches:Muroidea families withoutavailable genome sequence.

RESEARCH | REPORTS

on

Febr

uary

21,

201

7ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from