www.nature.com/nsmb

EDITORIAL OFFICE nsmb@natureny.com75 Varick Street, Fl 9, New York, NY 10013-1917Tel: (212) 726 9331, Fax: (212) 679 0735Editor: Boyana KonfortiSenior Editor: Michelle MontoyaAssociate Editors: Inês Chen, Sabbi LallAssistant Editor: Maria HodgesCopy Editor: Carrie PatisProduction Editors: Sabina Eberle, Jamel WootenSenior Illustrator: Katie VicariIllustrator: Kimberly CaesarCover Design: Erin BoyleEditorial Assistant: Christina Polizoto

MANAGEMENT OFFICESNPG New York75 Varick Street, Fl 9, New York, NY 10013-1917Tel: (212) 726 9200, Fax: (212) 696 9006Publisher: Stephanie DimentExecutive Editor: Linda MillerChief Technology Officer: Howard RatnerHead of Nature Research & Reviews Marketing: Sara GirardMarketing Manager: Leah RodriguezAssistant Production Coordinator: Karen WilsonHead of Web Services: Anthony BarreraWeb Production Manager: Susan Kline

NPG LondonThe Macmillan Building, 4 Crinan Street, London N1 9XWTel: 44 207 833 4000, Fax: 44 207 843 4996Managing Director: Steven InchcoombePublishing Director: Alison MitchellEditor-in-Chief, Nature Publications: Philip CampbellMarketing Director: Della SarDirector of Web Publishing: Timo Hannay

NPG Nature Asia-PacificChiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843Tel: 81 3 3267 8751, Fax: 81 3 3267 8746Publishing Director — Asia Pacific: David SwinbanksAssociate Director: Antoine E. BocquetManager: Koichi NakamuraOperations Director: Hiroshi MinemuraMarketing Manager: Masahiro YamashitaAsia-Pacific Sales Director: Kate YoneyamaAsia-Pacific Sales Manager: Ken Mikami

DISPLAY ADVERTISING display@natureny.com (US/Canada) display@nature.com (Europe) nature@natureasia.com (Asia)Global Head of Advertising and Sponsorship: Dean Sanderson, Tel: (212) 726 9350, Fax: (212) 696 9482Global Head of Display Advertising: Andrew Douglas, Tel: 44 207 843 4975, Fax: 44 207 843 4996Asia-Pacific Sales Manager: Ken Mikami, Tel: 81 3 3267 8765, Fax: 81 3 3267 8746Display Account Managers:New England: Sheila Reardon, Tel: (617) 399 4098, Fax: (617) 426 3717New York/Mid-Atlantic/Southeast: Jim Breault, Tel: (212) 726 9334, Fax: (212) 696 9481Midwest: Mike Rossi, Tel: (212) 726 9255, Fax: (212) 696 9481West Coast South: George Lui, Tel: (415) 781 3804, Fax: (415) 781 3805West Coast North: Bruce Shaver, Tel: (415) 781 6422, Fax: (415) 781 3805Germany/Switzerland/Austria: Sabine Hugi-Fürst, Tel: 41 52761 3386, Fax: 41 52761 3419United Kingdom/Ireland: Jeremy Betts, Tel: 44 207 843 4968, Fax: 44 207 843 4749Scandinavia/Iceland/Spain/Portugal: Evelina Rubio-Hakansson, Tel: 44 207 843 4079, Fax: 44 207 843 4749France/Belgium/The Netherlands/Italy/Israel/Eastern Europe: Nicola Wright, Tel: 44 207 843 4959, Fax: 44 207 843 4749Greater China/Singapore: Gloria To, Tel: 852 2811 7191, Fax: 852 2811 0743

NATUREJOBS naturejobs@natureny.com (US/Canada) naturejobs@nature.com (Europe) nature@natureasia.com (Asia)US Sales Manager: Ken Finnegan, Tel: (212) 726 9248, Fax: (212) 696 9482European Sales Manager: Dan Churchward, Tel: 44 207 843 4966, Fax: 44 207 843 4596Asia-Pacific Sales Manager: Ayako Watanabe, Tel: 81 3 3267 8765, Fax: 81 3 3267 8746

SITE LICENSE BUSINESS UNITAmericas: Tel: (888) 331 6288 institutions@natureny.comAsia/Pacific: Tel: 81 3 3267 8751 institutions@natureasia.comAustralia/New Zealand: Tel: 61 3 9825 1160 nature@macmillan.com.auIndia: Tel: 91 124 2881054/55 npgindia@nature.comROW: Tel: 44 207 843 4759 institutions@nature.com

CUSTOMER SERVICE www.nature.com/helpSenior Global Customer Service Manager: Gerald CoppinFor all print and online assistance, please visit www.nature.com/helpPurchase subscriptions:Americas: Nature Structural & Molecular Biology, Subscription Dept., 342 Broadway, PMB 301, New York, NY 10013-3910. Tel: (866) 363 7860, Fax: (212) 689 9108Europe/ROW: Nature Structural & Molecular Biology, Subscription Dept., Macmillan Magazines Ltd., Brunel Road, Houndmills, Basingstoke RG21 6XS, United Kingdom. Tel: 44 1256 329 242, Fax: 44 1256 812 358Japan: Nature Structural & Molecular Biology, NPG Nature Asia-Pacific, Chiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843. Tel: 81 3 3267 8751, Fax: 81 3 3267 8746India: Nature Structural & Molecular Biology, NPG India, 3A, 4th Floor, DLF Corporate Park, Gurgaon 122002, India. Tel: 91 124 2881054/55, Fax: 91 124 2881052

REPRINTS reprint@natureny.comNature Structural & Molecular Biology Reprint Department, Nature Publishing Group, 75 Varick Street, Fl 9, New York, NY 10013-1917, USA.For commercial reprint orders of 600 or more, please contact:UK Reprints: Tel: 44 1256 302 923, Fax: 44 1256 321 531US Reprints: Tel: (212) 726 9278, Fax: (212) 679 0843

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

i

volume 16 number 4 APrIl 2009

Nature Structural & Molecular Biology (ISSN 1545-9993) is published monthly by Nature Publishing Group, a trading name of Nature America Inc. located at 75 Varick Street, Fl 9, New York, NY 10013-1917. Periodicals postage paid at New York, NY and additional mailing post offices. Editorial Office: 75 Varick Street, Fl 9, New York, NY 10013-1917. Tel: (212) 726 9331, Fax: (212) 679 0735. Annual subscription rates: USA/Canada: US$225 (personal), US$3,060 (institution). Canada add 7% GST #104911595RT001; Euro-zone: €287 (personal), €2,430 (institution); Rest of world (excluding China, Japan, Korea): £185 (personal), £1,570 (institution); Japan: Contact NPG Nature Asia-Pacific, Chiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843. Tel: 81 (03) 3267 8751, Fax: 81 (03) 3267 8746. POSTMASTER: Send address changes to Nature Structural & Molecular Biology, Subscriptions Department, 342 Broadway, PMB 301, New York, NY 10013-3910. Authorization to photocopy material for internal or personal use, or internal or personal use of specific clients, is granted by Nature Publishing Group to libraries and others registered with the Copyright Clearance Center (CCC) Transactional Reporting Service, provided the relevant copyright fee is paid direct to CCC, 222 Rosewood Drive, Danvers, MA 01923, USA. Identification code for Nature Structural & Molecular Biology: 1545-9993/04. Back issues: US$45, Canada add 7% for GST. CPC PUB AGREEMENT #40032744. Printed on acid-free paper by Dartmouth Journal Services, Hanover, NH, USA. Copyright © 2009 Nature Publishing Group. Printed in USA.

e d I to r I A l

345 Pressing information

n e w s A n d v I e w s

346 miR-9 and TLX: chasing tails in neural stem cellsAhmet M Denli, Xinwei Cao & Fred H Gage see also p 365

348 A tipping point for mistranslation and diseasePaul Schimmel & Min Guo see also p 353 and p 359

350 At the (3′) end, you’ll turn to meiosisAlberto Moldón & José Ayté

351 Yeast as budding stem cells?Inês Chen

352 reseArch hIghlIghts

A r t I c l e s

353 Bases in the anticodon loop of tRNAAlaGGC prevent misreading

Hiroshi Murakami, Atsushi Ohta & Hiroaki Suga see also p 348

359 A sequence element that tunes Escherichia coli tRNAAlaGGC to ensure

accurate decodingSarah Ledoux, Mikołaj Olejniczak & Olke C Uhlenbeck see also p 348

365 A feedback regulatory loop involving microRNA-9 and nuclear receptor TLX in neural stem cell fate determinationChunnian Zhao, GuoQiang Sun, Shengxiu Li & Yanhong Shi see also p 346

372 TRF2 functions as a protein hub and regulates telomere maintenance by recognizing specific peptide motifsHyeung Kim, Ok-Hee Lee, Huawei Xin, Liuh-Yow Chen, Jun Qin, Heekyung Kate Chae, Shiaw-Yih Lin, Amin Safari, Dan Liu & Zhou Songyang

Structural and kinetic analysis of a M. tuberculosis NAD+

synthetase gives insight into the coordination of catalysis.

(p 421)

Ribosomal proteins make assembly more cooperative by discriminating

against non-native conformations of the E. coli 16S rRNA. Ramaswamy

and Woodson use hydroxyl radical footprinting to reveal a conformational switch during

assembly of the 30S 5′ domain. Cover art by Priya Ramaswamy is an interpretation of a footprinting gel.

pp 438–445

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

iii

volume 16 number 4 APrIl 2009

nAture structurAl & moleculAr bIology

380 Polyglutamine disruption of the huntingtin exon 1 N terminus triggers a complex aggregation mechanismAshwani K Thakur, Murali Jayaraman, Rakesh Mishra, Monika Thakur, Veronique M Chellgren, In-Ja L Byeon, Dalaver H Anjum, Ravindra Kodali, Trevor P Creamer, James F Conway, Angela M Gronenborn & Ronald Wetzel

390 Bacterial frataxin CyaY is the gatekeeper of iron-sulfur cluster formation catalyzed by IscSSalvatore Adinolfi, Clara Iannuzzi, Filippo Prischi, Chiara Pastore, Stefania Iametti, Stephen R Martin, Franco Bonomi & Annalisa Pastore

397 The pathway of hepatitis C virus mRNA recruitment to the human ribosomeChristopher S Fraser, John W B Hershey & Jennifer A Doudna

405 Tertiary interactions within the ribosomal exit tunnelAndrey Kosolapov & Carol Deutsch

412 Acetylation by GCN5 regulates CDC6 phosphorylation in the S phase of the cell cycleRoberta Paolinelli, Ramiro Mendoza-Maldonado, Anna Cereseto & Mauro Giacca

421 Regulation of active site coupling in glutamine-dependent NAD+ synthetaseNicole LaRonde-LeBlanc, Melissa Resto & Barbara Gerratana

430 Precursor-product discrimination by La protein during tRNA metabolismMark A Bayfield & Richard J Maraia

438 S16 throws a conformational switch during assembly of 30S 5′ domainPriya Ramaswamy & Sarah A Woodson

b r I e f com m u n I c At I o n s

446 CK2α phosphorylates BMAL1 to regulate the mammalian clockTeruya Tamaru, Jun Hirayama, Yasushi Isojima, Katsuya Nagai, Shigemi Norioka, Ken Takamatsu & Paolo Sassone-Corsi

449 Distinct transcriptional outputs associated with mono- and dimethylated histone H3 arginine 2Antonis Kirmizis, Helena Santos-Rosa, Christopher J Penkett, Michael A Singer, Roland D Green & Tony Kouzarides

nAture structurAl & moleculAr bIology clAssIfIed

See back pages.

New insights into the bacterial CyaY protein may shed light on the function of human frataxin.

(p 390)

In vitro studies reveal a complex pathway initiated when polyQ

disrupts the structure of an N-terminal huntingtin region.

(p 380)

Precursor-product discrimination by the La protein is examined.

(p 430)

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

nature structural & molecular biology volume 16 number 4 APrIl 2009 345

adequate time for internal coverage by institutional public information or public affairs offices. Advance notice is particularly useful for large agencies, such as the National Institutes of Health, that participate in a large number of research programs and may therefore be contacted about any number of forthcoming research publications.

With backgrounds in the biological and physical sciences, the NPG press officers write the press releases for newsworthy articles published in Nature. They also work with the editors at the Nature research journals to compose press releases for articles that may have wide public appeal. (Usually, articles whose findings have a direct connection to a disease may garner attention, though it can be difficult to predict what will catch the eyes of science journalists.) Author contact details accompany the highlighted papers within a release, along with a list of all papers that are being published in that journal in that particular week. On the Tuesday before the articles will be published online, the press office e-mails a compiled release covering Nature and the Nature research journals for that week to over 3,000 science journalists and media organizations. In some special cases, the press office will also organize a press briefing, at which journalists can speak directly to the researchers about their work.

The goal for all this work by our press offices is to make the science we publish accessible to the public by directly connecting the researchers with the journalists who report upon their work, and to do so in a way that does not distort, hype or dilute the scientific message. Although these are important steps toward getting science to the public, a certain limitation is present at the receiving end, given that the media covers only certain types of stories (see the recent Nature Methods editorial (http://www.nature.com/nmeth/journal/v6/n3/abs/nmeth0309-181.html) for more details on this).

A bright spot aimed at opening up more science to journalists is the Science Media Centre (SMC, http://www. sciencemediacentre.org). The SMC is a nonprofit organization that has become a valuable resource for UK-based journalists looking to report on late- breaking science news stories, directing them to scientists who would be appropriate to comment on the story and providing them with the salient scientific points involved in the story, almost like our own press office but on a national scale. Perhaps with all the money being injected into US-based research in the immediate future—and to add another suggestion to the thousands of ideas out there for how this windfall should be spent and allocated—some money should be directed toward developing a US organization similar to the SMC, which would be perfectly in line with the current administration’s goals of open communication and “restoring science to its rightful place.” L

as we’ve expressed several times in past editorials, it is important for scientists to find a way to effectively communicate with the general public. This serves to emphasize the importance

of funding basic science research and to increase public awareness of scientific findings and of how they relate to and may improve everyday life and the world around us. NSMB, as one of the Nature Research Journals and part of Nature Publishing Group (NPG), already has an effective system in place to notify the media, and thus the public, of the research published in our pages.

For those who may not have noticed, NSMB recently made a relatively big splash in US and international news outlets with the publication of an article from Sui, Donis, Liddington, Marasco and colleagues (http://www.nature.com/nsmb/journal/v16/n3/abs/nsmb.1566.html) on the isolation of monoclonal antibodies that could recognize a variety of influenza strains, paving the way for a broad-spectrum therapy against the flu and giving hope for the development of a long-sought-after universal flu vaccine. If you haven’t seen some of the wonderful coverage the research received, don’t worry, we’ve been keeping track of it all (http://www.nature.com/nsmb/news.html).

Although the credit for this research and breakthrough goes to the authors, who performed and analyzed the experiments, a lot of credit for getting the word out should definitely go to the NPG press office (http://press.nature.com), which sends press releases for articles published by all Nature-branded journals and helps coordinate press coverage by members of the science media, providing another service to our authors (http://www.nature.com/authors/author_ benefits/index.html). With press offices based in the US and UK, our press officers serve as first contacts for journalists both local and abroad, and are wardens of our press and embargo policies.

The NPG press office is notified of every manuscript that is accepted for publication and is responsible for coordinating the press release of manuscripts once publication dates have been set. For the Nature research journals, research articles are published online every Sunday, and the news embargo lifts at the time of publication on the journals’ websites, at 1 p.m. US Eastern time (6 p.m. London time). The press office performs a variety of tasks as each article is prepared for publication. Importantly for our authors, the press office provides additional notification to the authors of a research paper the Tuesday before the particular Sunday a paper is scheduled to appear online, informing them of our embargo policies, which are strictly enforced.

In a recently added service, press officers also contact relevant funding agencies and home institutions involved with the work. This gives

Pressing informationOne aim in science communication is to make the general public more aware of the breakthroughs and insights basic science research provides. Our press office gives us one route to help achieve that goal.

e d i to r i a l

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

346 volume 16 number 4 APrIl 2009 nature structural & molecular biology

Ahmet M. Denli, Xinwei Cao and Fred H. Gage are at the Salk Institute for Biological Studies, 10010 North Torrey Pines Road, La Jolla, California 92037, USA. email: gage@salk.edu

targeted the TLX 3′ UTR. More importantly, both endogenous TLX mRNA and protein levels were inversely affected by manipulations of miR-9 levels in cultured mouse NSCs. So, could miR-9 overexpression recapitulate TLX deficiency phenotypes in NSCs? Indeed, similar to the proliferation defect caused by the loss of TLX, overexpression of miR-9 led to decreased BrdU incorporation. The

mRNA target pairs for further study. The study by Zhao et al.10 on page 365 of this issue takes advantage of microRNA target predictions in combination with prior data on TLX and miR-9, and hypothesizes that TLX is a miR-9 target.

Zhao and colleagues started by showing, using luciferase-based assays in HEK293 human embryonic kidney cells, that miR-9

Neural stem cells (NSCs) are defined by their ability to self-renew and to differentiate into all neural cell types1. A combination of intrinsic and extrinsic factors, including nuclear receptors and small RNAs, contribute to their function2. Nuclear receptors are a highly conserved and ancient superfamily of transcriptional regulators that have a central role in integrating developmental processes as well as physiological responses3. MicroRNAs are endogenously expressed small RNAs that negatively regulate downstream target mRNAs mainly through their 3′ untranslated regions (3′ UTRs) and have established roles during development as well as in adult organisms4.

miR-9 is one of the numerous conserved microRNAs that have altered expression levels between NSCs and their differentiated progenies, with a propensity to be upregulated in more mature cell types5. This expression pattern is opposite to that of the highly conserved orphan nuclear receptor TLX, which is expressed in the neuroepithelium of the embryonic mouse brain and in adult neurogenic regions6,7. TLX is essential for NSC proliferation7, and TLX-null mice have smaller brains and thinner cortex8. The Drosophila homolog, Tailless, is a gap gene product that is expressed in the embryonic brain and is required for brain development9.

Even though microRNA target predictions are still imperfect, they have been invaluable resources for picking candidate microRNA-

miR-9 and TLX: chasing tails in neural stem cellsAhmet M Denli, Xinwei Cao & Fred H Gage

Development and maintenance of an organism require the precise spatiotemporal orchestration of stem cell proliferation and differentiation. In neurogenesis, a microRNA and an orphan nuclear receptor comprise a negative feedback loop that regulates neural stem cell fate.

Figure 1 miR-9–TLX feedback loop in neural stem cells. Expression of a mature microRNA can be regulated at multiple levels of its life cycle. TLX binds downstream of miR-9 sequence at the mir-9-1 locus and represses miR-9 at the transcriptional level. Meanwhile, miR-9 targets TLX mRNA for destabilization and/or translational inhibition, reducing TLX protein levels. In proliferating cells (red arrows), TLX predominates in this loop and is expressed at higher levels than in differentiating cells, whereas the reverse expression trend is true for miR-9. During the differentiation of neural stem cells (blue arrows), miR-9 becomes predominant and its levels increase, inhibiting TLX expression. It is plausible that miR-9 targets include the TLX co-repressor HDAC5 and other hypothetical components (denoted ‘X’) of the TLX complex. It is also conceivable that other differentiation-related microRNAs (denoted ‘miR-A’ and ‘miR-B’) cooperate with miR-9 in repressing TLX expression.

Kat

ie V

icar

i

X mRNA

Cytoplasm

Nucleus

miR-9?

TLX

pri-miR-9-1

?

TLX mRNA

miR-9 miR-A? miR-B?An

HDAC5 mRNAmiR-9?

An

An

An

miR-9

mir-9-1 locus

pre-miR-9-1

pre-miR-9-1

TLXHDAC5

X?

n e w s a n d v i e w s

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

nature structural & molecular biology volume 16 number 4 APrIl 2009 347

Last but not least, the functional significance of individual microRNAs during development has been a topic of hot debate. The phenotypes caused by inhibiting microRNA function range from undetectable to quite severe13. Although it is tempting to classify microRNAs as a single group and try to assign fine-tuning roles to them, generalizations may be dangerous. First of all, we do not have data on a large enough set of microRNAs. Technically, a large proportion of studies use nongenetic methods to inhibit microRNAs. In these experimental setups, the level of reduction in microRNA function may not be sufficient to cause a detectable phenotype. Biologically, some 3′ UTRs are targeted by multiple microRNAs, and manipulation of microRNA combinations may be needed to uncover any effects. This is similar to the way that some protein-coding genes require secondary lesions to cause a detectable phenotype. Most important of all, functions of individual microRNAs depend on the genes they target, and the severities of their loss-of-function phenotypes may be as diverse as those of protein-coding genes. Thus, it is very likely that researchers analyzing the effects of some microRNA will readily observe identifiable phenotypes, whereas others will have to scratch their heads and ask, “What’s wrong with my cells?”

1. Zhao, C., Deng, W. & Gage, F.H. Cell 132, 645–660 (2008).

2. Shi, Y., Sun, G., Zhao, C. & Stewart, R. Crit. Rev. Oncol. Hematol. 65, 43–53 (2008).

3. Mangelsdorf, D.J. et al. Cell 83, 835–839 (1995).4. Ambros, V. & Chen, X. Development 134, 1635–1641

(2007).5. Krichevsky, A.M., King, K.S., Donahue, C.P., Khrapko, K.

& Kosik, K.S. RNA 9, 1274–1281 (2003).6. Yu, R.T., McKeown, M., Evans, R.M. & Umesono, K.

Nature 370, 375–379 (1994).7. Shi, Y. et al. Nature 427, 78–83 (2004).8. Land, P.W. & Monaghan, A.P. Cereb. Cortex 13, 921–931

(2003).9. Pignoni, F. et al. Cell 62, 151–163 (1990).10. Zhao, C., Sun, G., Li, S. & Shi, Y. Nat. Struct. Mol. Biol.

16, 365–371 (2009).11. Leucht, C. et al. Nat. Neurosci. 11, 641–648 (2008).12. Viswanathan, S.R., Daley, G.Q. & Gregory, R.I. Science

320, 97–100 (2008).13. Flynt, A.S. & Lai, E.C. Nat. Rev. Genet. 9, 831–842

(2008).14. Sun, G., Yu, R.T., Evans, R.M. & Shi, Y. Proc. Natl.

Acad. Sci. USA 104, 15282–15287 (2007).15. Shibata, M., Kurokawa, D., Nakao, H., Ohmura, T. &

Aizawa, S. J. Neurosci. 28, 10415–10421 (2008).16. Packer, A.N., Xing, Y., Harper, S.Q., Jones, L. &

Davidson, B.L. J. Neurosci. 28, 14341–14346 (2008).

overexpression. Although these data imply that TLX represses the mir-9-1 locus, more experiments will be needed to provide further insight into this regulation. Three loci encode miR-9, and the authors focused on the mir-9-1 locus. What happens at the other mir-9 loci, and what chromatin marks are present at these sites in TLX-knockdown and TLX-null cells? It will be interesting to assess by in situ hybridization whether miR-9 expression is increased in the neuroepithelium of TLX-null brains.

A number of microRNAs and their targets are expressed simultaneously in cultured stem cells and their in vivo counterparts during development and in adult organisms. Thus, stem cells provide a biologically relevant platform to define microRNA-target interactions. Zhao et al. provide a glimpse of the wealth of insight that can be gained from such an experimental system (Fig. 1). From a mechanistic perspective, their study raises a number of well-defined questions. What other co-repressors besides HDAC5 are at work on the mir-9-1 locus? Multiple components of a protein complex or a signaling pathway are sometimes targeted by an individual microRNA13, raising the possibility that miR-9 may target other components of the TLX complex. Aside from the cross-repression between TLX and miR-9, the cell cycle regulators Pten and p21 are repressed by TLX14. Moreover, miR-9 downregulates Foxg1, REST and Co-REST15,16 in the mouse as well as Her5 and Her9 in zebrafish11. These molecules and others yet to be identified may contribute greatly to the observed TLX and miR-9 phenotypes. Thus, it remains to be determined how important a role the miR-9–TLX loop plays in embryonic and adult brains. Considering the prevalence of feedback loops in developmental pathways, it is possible that miR-9 and TLX regulate a subset of each other’s downstream targets.

The emerging picture of TLX and miR-9 mutual regulation suggests that a balance exists between these two molecules in NSCs. Could there be common upstream activators of TLX and miR-9 expression creating this equipoise? It will be essential to gain insight into the molecular switch that perturbs this balance and dictates developmental timing.

authors went on to show that increased miR-9 levels accelerated differentiation, as is the case in zebrafish11. Interestingly, this effect was present only when cells were primed to differentiate. Moreover, overexpressing miR-9–insensitive TLX rescued these phenotypes, suggesting that the effects of miR-9 on NSCs are at least partially mediated by repression of TLX. Overexpression of microRNAs above physiological levels may lead to biologically irrelevant effects, so the authors also performed the converse experiment and showed that reducing mature miR-9 levels led to increased TLX expression and NSC proliferation.

Although these results pointed to a role for miR-9 in TLX regulation, an essential test of the hypothesis was to look at in vivo effects. In utero electroporation of miR-9 duplexes into the developing mouse brain reduced the abundance of TLX protein and decreased the number of proliferative cells in the ventricular zone. Furthermore, compared to control electroporated cells, a higher percentage of miR-9–overexpressing cells migrated into the cortical plate and proceeded through neural differentiation. This phenotype could be rescued by TLX overexpression. It is worth noting that not all cells that took up miR-9 and had lower TLX levels showed enhanced differentiation. Together with similar results from cell culture experiments, this may imply that miR-9 accelerates differentiation only in certain contexts, perhaps in cells that are primed to exit the cell cycle.

Feedback loops have emerged as a major theme in microRNA regulation, with one of the most recent examples coming from the lin-28–let-7 pair12. Zhao et al. noticed higher pri-miR-9 levels in TLX-null mice, and their analysis of the mir-9 locus highlighted the presence of multiple TLX binding sites downstream of the mature mir-9 sequence. The authors were able to pull down the miR-9-1 locus using antibodies to TLX. In addition, a TLX co-repressor HDAC5 and a repressive histone mark were associated with the mir-9-1 locus, whereas active histone marks were absent. Zhao et al. also used a luciferase-based assay to show that a fragment of the mir-9-1 locus containing the TLX target sites was repressed by TLX

n e w s a n d v i e w s

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

348 volume 16 number 4 APrIl 2009 nature structural & molecular biology

Paul Schimmel and Min Guo are at The Skaggs Institute for Chemical Biology, The Scripps Research Institute, Beckman Center, La Jolla, California,USA. e-mail: schimmel@scripps.edu

anticodon loop, at positions 32 and 38. These nucleotides interact across the loop and most commonly are C32-A38 or U32-A38 (ref. 7). But, surprisingly, they find that a simple transversion of an uncommon A32-U38 in tRNAAla

GGC that preserves complementarity— a U32-A38 pair for example—results in loss of translational fidelity. In particular, a tRNAAla

GGC isoacceptor that normally reads a GCC alanine codon but not the GUC valine codon can, with a U32-A38 pair, read the near-cognate GUC codon and thereby insert alanine at the GUC valine codon. The U32-A38 tRNAAla

GGC still reads the cognate GCC alanine codon, so the specificity of codon- anticodon recognition has been relaxed. Pre–steady state kinetic measurements rigorously showed that the U32-A38 tRNAAla

GGC manifested rapid GTP hydrolysis and was not rejected by GUC or other near-cognate codons4. Another construct, C32-A38 tRNAAla

GGC, read each cognate and near-cognate codon that was tested.

After aminoacylation, the aminoacyl-tRNA (aa-tRNA) binds to elongation factor EF-Tu, which is in complex with GTP. The aa-tRNA–EF-Tu–GTP complex is then directed to the ribosome. At this point, if the correct codon-anticodon interaction occurs, aa-tRNA is released to the ribosome with concomitant hydrolysis of GTP, leaving EF-Tu–GDP. Release of aa-tRNA to the A site is facilitated by the hydrolysis-induced conformational change in EF-Tu, because EF-Tu–GDP has a much weaker affinity for aa-tRNA. If, however, the codon-anticodon interaction is mismatched, the aa-tRNA–EF-Tu–GTP complex can simply dissociate from the ribosome or, alternatively, GTP hydrolysis is activated and the aa-tRNA and EF-Tu–GDP complex break up and fall off the ribosome6 (Fig. 1).

The independent work from the Suga3 and Uhlenbeck4 groups show that the course taken by aa-tRNA as it enters the ribosome is especially sensitive to two nucleotides in the

The genetic code is the greatest scientific discovery of the past 100 years. The code itself is an algorithm that relates nucleotide triplets to specific amino acids. That relationship is established by the 20 aminoacyl tRNA synthetases, which catalyze the aminoacylations of each tRNA with its cognate amino acid (the amino acid that corresponds to the anticodon triplet of the tRNA). tRNAs are typically composed of 74–93 nucleotides (76 nucleotides is the most common size) arranged into a cloverleaf with four helical stems and three loops, terminating in the universal single-stranded 3′-CCA76 sequence, with A76 being the amino acid attachment site (Fig. 1). The tRNA three-dimensional structure revealed an extraordinary L-shaped molecule1,2, with the four stems fused into two longer helices— a 12-bp minihelix containing the A76 amino acid attachment site, and a 9-bp hairpin helix harboring the anticodon triplet in the hairpin loop, where it binds to its complementary mRNA codon during protein synthesis. The L-shaped structure is fragile and delicate, being held together by a series of unusual interactions between conserved nucleotides in the loops that facilitate formation of a right-angled corner between the minihelix domain and the 9-bp anticodon-containing domain. In this issue of Nature Structural & Molecular Biology, two papers3,4 present strong evidence that the anticodon loop of the molecule is poised on a tipping point so that, given a nudge, it mistranslates the codon- anticodon interaction and the wrong amino acid is inserted into the growing polypeptide chain, resulting in a loss of translational fidelity.

Among the first achievements of the intense research that followed the discovery of the genetic code were the isolation of active ribosomes and the identification of two crucial sites for peptide synthesis5: the A site accepts the newly binding aminoacyl tRNA, whereas the P site harbors the growing peptide chain in the form of peptidyl-tRNA.

A tipping point for mistranslation and diseasePaul Schimmel & Min Guo

Two papers present strong evidence that the codon-anticodon interaction is poised on a tipping point so that, given a nudge, the tRNA can insert the wrong amino acid into the growing polypeptide chain, leading to translational fidelity loss.

Figure 1 A 32-38 pair in tRNAAlaGGC for translation fidelity. (a) The ribosome discriminates between

correct and incorrect aa-tRNAs according to the match between the anticodon and the mRNA codon in the A site. In the initial selection of the aa-tRNA–EF-Tu–GTP complex, the tRNA assumes a deformed conformation and a signal is transmitted from the codon-anticodon recognition site. The second proofreading selection is irreversibly separated by GTP hydrolysis, as the cognate tRNA moves rapidly to A site accommodation and peptidyl transfer, whereas a near-cognate tRNA dissociates from the ribosome with high probability. (b) The L-shaped arrangement of tRNA with the sequence of anticodon loop as in tRNAAla

GGC. (c) The 32-38 pair, residing close to the anticodon triplet, affects codon-anticodon decoding.

32

38

mRNA

5′

3′

GTP

50S

30S

mRNA

Binding

Rejection

Accommodation/peptidyl transfer

Rejection

GTPase activation

GDP

ACC

AUG G C

TΨC

D

32 38

3435

36

76

1

1025

7

65

50

6055

20 15

T Ψ C loopMinihelix domain

D loop

Anticodon loop

+

+

+ +

a

b c

n e w s a n d v i e w s

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

nature structural & molecular biology volume 16 number 4 APrIl 2009 349

toxic to E. coli, inferring that mistranslation causes cell death. More generally, work from another source of mistranslation—mutations in the editing centers of tRNA synthetases that cause misaminoacylations by tRNA synthetases—have been shown to be toxic in bacteria, cause pathologies in cultured mammalian cells11 and, even with a mild error-causing mutation, give rise to neurodegeneration and ataxia in mice12. Thus, it is not a long extrapolation to assume that lethal and disease-causing phenotypes provide powerful selective pressure on the 32-38 pair and that, in some instances, diseases might be found that arise from mutations at the 32-38 tipping point for translational infidelity. On this point, we wondered whether some of the mutations in mitochondrial tRNAs that have long been associated with pathologies in the human population might be at the 32 or 38 position. Five examples of mutations at the 32 or 38 position were found that are causally associated with chronic progressive external ophthalmoplegia, dementia, ataxia, encephalomyopathy, lethal infantile mito-chondrial myopathy and sensorineural hearing loss, among others (Table 1). Three of these mutant tRNAs express their phenotype in a heteroplasmic setting, that is, the mutation is operationally dominant, inferring a gain of function (such as misreading), a situation reminiscent of the pathology- causing dominance of an editing-defective tRNA synthetase expressed in mammalian cells11. Most of these mutations convert a C32-A38 pair to a C-G, as in one of the examples in the Suga and Uhlenbeck studies. Thus, it looks like the 32-38 tipping point for mistranslation really matters.

ACKNOWLEDGMENTSWe thank C. Florentz and S. Kelley for helpful comments on human mitochondrial tRNAs.

1. Kim, S.H. et al. Science 185, 435–440 (1974).2. Robertus, J.D. et al. Nature 250, 546–551 (1974).3. Murakami, H., Ohta, A. & Suga, H. Nat. Struct. Mol.

Biol. 16, 353–358 (2009).4. Ledoux, S., Olejniczak, M. & Uhlenbeck, O.C.

Nat. Struct. Mol. Biol. 16, 359–364 (2009).5. Warner, J.R. & Rich, A. J. Mol. Biol. 10, 202–211

(1964).6. Cochella, L., Brunelle, J.L. & Green, R. Nat. Struct.

Mol. Biol. 14, 30–36 (2007).7. Olejniczak, M. & Uhlenbeck, O.C. Biochimie 88,

943–950 (2006).8. Thompson, J. et al. Proc. Natl. Acad. Sci. USA 98,

9002–9007 (2001).9. Cabello-Villegas, J., Winkler, M.E. & Nikonowicz, E.P.

J. Mol. Biol. 319, 1015–1034 (2002).10. Ogle, J.M. et al. Science 292, 897–902 (2001).11. Nangle, L.A., Motta, C.M. & Schimmel, P. Chem. Biol.

13, 1091–1100 (2006).12. Lee, J.W. et al. Nature 443, 50–55 (2006).13. Scaglia, F. & Wong, L.J. Muscle Nerve 37, 150–171

(2008).

bond formation on different codons4. And, of particular significance, the Suga group carried out the work in a cell-free translation system, using MS to nail down the predicted substitution of alanine at a codon for valine3. All of these experiments ‘square off ’, that is, they are consistent among themselves.

How does the 32-38 base pair exert its effect? Here the picture is less clear, especially because the structural effects per se are subtle and involve figuring out the difference, for example, between transversions like U-A and A-U. Ideally, an NMR or X-ray structure of a charged tRNA–EF-Tu–GTP complex at the A site, with a specific 32-38 pair, could be compared with the identical bound tRNA with a transversion at 32-38. Lacking this sort of structural information, and wanting simply to see the response of the anticodon loop and the 32-38 pair to different local conditions, the NMR structure of the anticodon stem-loop of Escherichia coli tRNAPhe free in solution9 can be compared with the X-ray structure of the same tRNA anticodon stem-loop bound in the A site7,10. This tRNA in solution has a U32-A38 pair that, not surprisingly, is in the Watson-Crick conformation, where one of the two hydrogen bonds of the pair is between N6 of A and O4 of U. But when bound at the A site, the Watson-Crick pairing is lost, and the N6 is now connected by a single hydrogen bond to O2 of U. These and other changes can occur only with adjustments in the sugar-phosphate backbone and base-sugar torsional angles of the loop nucleotides, including the 32-38 pair, which is close to the site of the codon- anticodon interaction (Fig. 1). The consequences of these adjustments, which are needed for the codon-anticodon interaction, place the 32-38 nucleotides at a tipping point.

In the end, does translational infidelity actually matter? On that point, the Suga group showed that overexpression of U32-A38 tRNAAla

GGC is

Thus, the codon-anticodon interaction is on a tipping point. This functional sensitivity is further emphasized by the context of the surrounding sequences in the tRNA itself. For example, the many tRNAs with the commonly found U32-A38 and C32-A38 pairs do not misread codons. However, these tRNAs have a sequence context that is different from that of tRNAAla

GGC. These considerations show clearly that the codon-anticodon system of protein synthesis, with 20 amino acids and the collective set of roughly 60 isoacceptor tRNAs, has undergone a profound degree of refinement, not just for tRNA in general, but for each individual tRNA isoacceptor.

Whenever in vitro assays for peptide synthesis are used, several hazards and pitfalls can distort the results. An important consideration is the assays themselves that were used for measuring the effects of the misreading tRNAAla mutants. One lesson the field has learned is that different assays do not necessarily give the same picture. For example, when the conserved G2447 in the peptidyl transferase center of 23S ribosomal RNA is mutated to G2447A, the functional consequence is different according to the assay used for peptide bond formation. In essence, in an A site–limiting assay, the G2447A ribosomes were more than ten-fold less active than wild-type ribosomes. In contrast, in an A site– saturating assay, the G2447A and wild-type ribosomes were indistinguishable in their activities8. In the present case, the Suga and Uhlenbeck laboratories used different assays—a 13-mer–peptide synthesis assay in one instance, and a dipeptide synthesis assay in the other. Importantly, the Uhlenbeck study investigated many experimental parameters, including ribosome entry site binding for the tRNAAla–EF-Tu–GTP ternary complexes on different codons, GTP-hydrolysis rates on different codons and rates of peptide

Table 1 Human mitochondrial tRNAs with mutations at 32-38 and their clinical phenotypestRNA gene Mutation site Nucleotide

changeClinical features of related pathologies

Het/Homo Inheritance

MTTI (Ile) 4290 UG→CG EM Homo I

MTTN (Asn) 5692 CA→CG CPEO Het ND

MTTN (Asn) 5698 CA→UA CPEO, MM Het D

MTSS1(Ser) 7480 CA→CG MM, SNHL, dementia, ataxia

Het D

MTTT (Thr) 15923 CA→CG LIMM Homo I

MTTH (His) 12166 UA→CA Polymorphism

MTTH (His) 12172 UA→UG Polymorphism

CPEO, chronic progressive external ophthalmoplegia; EM, encephalomyopathy; LIMM, lethal infantile mitochondrial myopathy; MM, mitochondrial myopathy; SNHL, sensorineural hearing loss; Het, heteroplasmy; Hom, homoplasmy; D, de novo; ND, not determined; I, inherited. The mutation sites refer to the position in the mitochondrial DNA sequence, (http://www.mitomap.org). The information is extracted from Mamit-tRNA (http://mamit-trna.u-strasbg.fr/) and ref. 13.

n e w s a n d v i e w s

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

350 volume 16 number 4 APrIl 2009 nature structural & molecular biology

Alberto Moldón and José Ayté are in the Oxidative Stress and Cell Cycle Group, Universitat Pompeu Fabra, Barcelona, Spain. e-mail: jose.ayte@upf.edu

Furthermore, in elegantly described experiments, the authors identified two different polyadenylation signals: a proximal noncanonical and a distal canonical one. When the former is mutated, the normal regulation of crs1 is lost, preventing both splicing and accumulation of mature mRNA during meiosis. A new insight into the regulation of the meiotic gene expression program is that binding (or release) of a yet to be determined factor to these two elements in the 3′ end of the mRNA promotes its polyadenylation and splicing, concomitantly with an increase in mRNA stability; this process takes place exclusively during meiosis. However, the identity of the factor(s) that binds to this region is still unknown. In addition, it will be interesting to know whether these two signals can work in trans when placed in the 3′ end of a heterologous gene that otherwise is not subject to this mechanism of regulation.

During vegetative growth, Mmi1 selectively promotes the degradation of some meiosis- specific mRNAs by binding to a region called DSR (for ‘determinants of selective removal’), which is located at the 3′ end of those mRNAs. When cells start the meiotic program, Mmi1 is sequestered by Mei2, the master regulator of meiosis in fission yeast, avoiding the degradation of meiosis- specific genes3. Similarly, if Mmi1’s function is impaired in vegetatively growing cells, meiotic mRNAs are ectopically stabilized, causing cell death3. McPheeters and colleagues have also shown that crs1 mRNA is actively degraded by the exosome in a mmi1-dependent pathway. In meiosis, or in mmi1- or exosome-mutant backgrounds, crs1 mRNA is polyadenylated and spliced, and cells have increased crs1 levels. An important and unexpected turn in this story is the observation that 3′ end processing and splicing of crs1 are mechanistically linked. In vegetatively growing cells, crs1 mRNA is unspliced but also lacks polyadenylated tails. In meiosis, after the mRNA is completely transcribed and polyadenylated (and only after these two events have taken place), intron removal and stabilization of the mRNA is triggered, implying that splicing is not co-transcriptional.

described by Harigaya and co-workers, which ‘corrects’ the leakage of transcription of some meiosis-specific genes during vegetative growth by recruiting the exosome onto the newly synthesized messengers3; and (ii) splicing of meiosis- specific genes, which depends on the presence of meiosis-specific transcription factors4. This is the case for the forkhead transcription factor Mei4, which recruits the spliceosome to the promoters of some meiosis-specific genes, activating splicing at the onset of meiosis.

In a recent issue of Nature Structural & Molecular Biology, McPheeters and colleagues report a new mechanism that regulates the expression of a meiosis-specific cyclin, crs1, coupling transcription, 3′ end processing, stability and mRNA splicing to switch from a genetic program where the cyclin should not be expressed because of its ‘toxicity’ (during vegetative growth) to a different program where the cyclin is required (during meiosis)5. In yeasts, many meiotic proteins are toxic during vegetative growth, and their expression has to be turned off 3,6,7. This new report outlines a new mechanism to keep mitotically growing cells free of these toxic gene products.

McPheeters and colleagues show that crs1 RNA accumulation in meiosis is not due to increased transcription rates but, rather, depends on its mRNA processing: although the transcription of crs1 is active in cells undergoing vegetative growth (and, in fact, they show that it is higher in this stage than in meiosis), its mRNA is constantly degraded. In contrast, during meiosis the turnover is inhibited, and as a result, its mRNA is stabilized. This is achieved by two sequences located in the last exon at the 3′ end of the gene (at the 3′ untranslated region and at the polyadenylation site) that can regulate the expression of crs1 in both genetic programs. This is reminiscent of the Mmi1 pathway, involved in the selective elimination of meiosis-specific mRNAs during the mitotic cell cycle3. Importantly, this regulation seems to rely on structural properties of the mRNA rather than its specific sequence, because the distance between both sequences is crucial for normal function.

Cells are constantly receiving signals from the environment that can trigger a cellular response. In all organisms, signals such as nutrient deprivation, pheromones, confluence, toxins and so on can alter the status of the cell at various levels. In response to such signal transduction pathways, what determines the genetic program of a specific cell is the differential expression of clusters of genes that will define the cell phenotype. Genetic expression is regulated at multiple levels: chromatin structure, transcription rate, splicing, 3′ end processing, mRNA stability and nuclear export, translation and so on, but there are many gaps in our knowledge regarding the cross-regulation of all these processes and the contribution of each one to the maintenance program.

The fission yeast Schizosaccharomyces pombe is a unicellular eukaryote commonly used as a model organism to study several well- conserved molecular processes. S. pombe cells have a typical cell cycle but, under different conditions, cells can switch to different genetic programs. If cells are glucose deprived, they enter stationary phase; when grown under limiting nitrogen conditions, they can undergo meiosis and sporulation. Thus, fission yeast meiosis is an ideal system in which to study cell differentiation at the molecular level, on the basis of changing genetic expression programs1.

Some transcription factors in S. pombe have been described as master regulators of meiotic progression, inhibiting the expression of the previous cluster of genes and promoting the next one2. Accidental activation or inefficient downregulation of these transcription factors is usually lethal for the cell, leading to aberrant meiosis. To ensure the timely and orderly expression of different clusters of proteins, the steady-state level of the corresponding mRNAs is controlled at different levels. Until now, two mechanisms have been studied at the molecular level: (i) the mmi1 system,

At the (3′) end, you’ll turn to meiosisAlberto Moldón & José Ayté

Many cellular fates are determined by different genetic programs, but the regulation of cellular differentiation is still not well understood. Besides the possible control exerted by the activity and combination of transcription factors, there are multiple RNA processing mechanisms, ensuring differential gene expression.

n e w s a n d v i e w s

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

nature structural & molecular biology volume 16 number 4 APrIl 2009 351

ACKNOWLEDGMENTSWork in the Ayté laboratory is supported by grants from the Ministerio de Ciencia e Innovación,Spain (BFU2006-01785) and the Consolider-Ingenio (2007-0020).

1. Mata, J., Lyne, R., Burns, G. & Bahler, J. Nat. Genet. 32, 143–147 (2002).

2. Mata, J., Wilbrey, A. & Bahler, J. Genome Biol. 8, R217 (2007).

3. Harigaya, Y. et al. Nature 442, 45–50 (2006).4. Moldon, A. et al. Nature 455, 997–1000 (2008).5. McPheeters, D.S. et al. Nat. Struct. Mol. Biol. 16,

255–264 (2009).6. Averbeck, N., Sunder, S., Sample, N., Wise, J.A. &

Leatherwood, J. Mol. Cell 18, 491–498 (2005).7. Malapeira, J. et al. Mol. Cell. Biol. 25, 6330–6337

(2005).8. Manganelli, R., Dubnau, E., Tyagi, S., Kramer, F.R. &

Smith, I. Mol. Microbiol. 31, 715–724 (1999).

From bacteria to mammalian cells, there are many mechanisms to avoid overlapping expression of different gene clusters. For example, in Mycobacterium tuberculosis, 13 different σ subunits of RNA polymerase have been described to provide specificity to the transcription of gene clusters required to respond to different environmental inputs8. Eukaryotic cells have added several layers of complexity to this mechanism, regulating the differentiation pathways not only at the level of transcription. So far, these new mechanisms are shedding light on the molecular mechanisms underlying gene regulation during meiosis in fission yeast. But we can be assured that other processes will participate in this complex network.

Despite the solid data, the model presented by McPheeters et al. is puzzling: why do vegetatively growing cells keep synthesizing mRNAs just to have them be degraded upon meiosis-triggering signals? As in some signaling pathways, such as oxidative stress, this may allow cells to quickly respond to damaging insults from the environment. However, it does seem energetically wasteful, especially if we consider that yeasts trigger meiosis when they sense poor nutrient conditions. A quick change in media composition to induce meiosis is something that happens only to yeast cultures growing in the laboratory, while in nature, yeast cells use up nutrients slowly as they grow, resulting in a very gradual change in medium composition.

DNA replication is semiconservative, with each strand serving as a template for the synthesis of a complementary strand. Thus, after one round of replication and cell division, each daughter cell will have a DNA molecule consisting of an old strand and a newly synthesized one. Each of these strands will serve as a template in the next round of replication, to generate two DNA molecules with strands of different ‘ages’: one comprising an old, original strand paired to a newly minted one, and the other with the strand synthesized in the first round also paired to a newly synthesized strand. There is a chance that errors will arise during replication, and, if not repaired, such mutations will become fixed in the next round of replication and passed along to one of the daughter cells. Thus, the only strands that will be completely free of replication errors are the original ones (provided there are no recombination events). John Cairns proposed in 1975 that it would be advantageous for adult stem cells in metazoans to retain one of the original strands for each of their DNA molecules (that is, chromosomes), to avoid the accumulation of mutations—a concept known as the ‘immortal strand’ hypothesis. Whether such asymmetric DNA segregation indeed occurs in adult stem cells has been the subject of debate, with evidence both for and against it. There has also been speculation as to how cells would be able to segregate the sister chromatids in a nonrandom fashion during mitosis. Chromosomal segregation is mediated by spindle microtubules attached to the kinetochore, a structure that forms transiently on top of the centromeric DNA, so it has been proposed that differences in this region could mark the chromosome containing the immortal strand.

Now Thorpe, Bruno and Rothstein find that four kinetochore components (Ndc10, Ctf19, Mtw1 and Ask1) are indeed segregated asymmetrically in postmeiotic budding yeast (Proc. Natl. Acad. Sci. USA, in the press, doi:10.1073/PNAS.0811248106). This unicellular organism undergoes asymmetric cell division, with one mother cell and one bud being generated at each cell division. The authors fused candidate kinetochore proteins to yellow or cyan fluorescent protein (YFP or CFP), made a diploid yeast strain

containing both fusions, and then had those cells undergo meiosis to generate spores carrying the sequence for only one of the labeled constructs but containing both YFP- and CFP-fused proteins. The fate of the non-encoded protein as well as the encoded protein was then followed from the germinating spore through three generations via fluorescence microscopy.

In the first round of division, the fluorescence signal was on average twofold stronger in the mother cell than in the bud cell as shown in the microscopy images (mother and bud cells indicated by “m” and “b” on the top left, fluorescence in the bottom left,

with an enhanced view on the right), indicating an asymmetric segregation of the kinetochore protein. In the next round of cell replication, a similar behavior was observed for the mother cell, whereas the previous “bud” cell segregated the protein equally between itself and its daughter bud. The same thing happened in the third round of cell division,

thus defining a cell lineage or pedigree, as represented in the drawing on the right. Such an asymmetric segregation pattern was not observed with another nuclear protein, histone H2A, nor was it seen in vegetatively growing cells.

Budding yeast cells have a limited reproductive lifespan, and the mother-cell lineage stops generating new daughters after a certain number of cellular divisions. In fact, the mother cells accumulate extrachromosomal DNA circles and oxidized proteins, which may contribute to ensuring a longer reproductive lifespan for the daughter bud cells. So one may ask, why would the postmeiotic cells studied here segregate their kinetochores, and hence their chromatids, in such an asymmetric way? It also remains to be shown whether sister chromatids are indeed segregated in a nonrandom fashion in the progeny of a germinating yeast spore. But regardless of ‘why’, this work opens new avenues to explore ‘how’ asymmetric segregation can occur. It also shows that budding yeast, the geneticists’ darling organism, can be a valuable model for studying the establishment of cellular lineages and asymmetric cell division in adult stem cells.

Inês Chen

Yeast as budding stem cells?

m

b

n e w s a n d v i e w s

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

352 volume 16 number 4 APrIl 2009 nature structural & molecular biology

Written by Angela K. Eggleston, Joshua M. Finkelstein, Maria Hodges & Sabbi Lall

as the resection proceeds, ATR activation becomes dominant. This would alter the downstream kinase targets and help move the repair process along. (Mol. Cell 33, 547–558, 2009) AKE

nmr in living cellsAtomic-resolution structures of proteins in living human cells would link three-dimensional structural information to biological processes. In-cell NMR allows observations of the conformations and functions of proteins in living cells, but requires two problems to be overcome: how to label proteins so that they are detectable by NMR spectroscopy in human cells and how to collect an adequate data set without destroying the cell. Until now two approaches have been used to label proteins for NMR studies. The first is to grow bacteria on isotopically labeled medium. The second is to microinject labeled proteins into large cells such as Xenopus oocytes. Inomata et al. introduce a third technique: fusing cell- penetrating peptides to 15N-labeled proteins. In the presence of pyrenbutyrate, the labeled target protein is taken into the cell, where the peptide is removed. This approach was so successful that the authors used it to study the intracellular dynamics of ubiquitin. They found that ubiquitin in cells was more dynamic and less structured than ubiquitin examined in vitro. The other major difficulty with in-cell NMR is the limited lifetime of cells within the NMR sample tube. NMR data- collection experiments usually take between one and two days, but Sakakibara et al. have shortened this to two to three hours by implementing a nonlinear sampling scheme, which differs from standard NMR sampling methods. They demonstrated the technique with three different proteins, one of which (FKBP12) forms specific complexes with externally added immunosuppressants— meaning that this technique might be useful for drug screening. In addition¸ they solved the structure of a putative heavy-metal-binding protein, TTHA1718. (Nature 458, 102–105 and 106–109, 2009) MH

orchestrating mitosisThe anaphase-promoting complex APC/C is a key regulator, and regulation point, during mitosis, when events must happen in an ordered fashion to ensure that each daughter cell receives a correct set of chromosomes. By ubiquitinating a number of substrates once the chromosomes are properly aligned and attached to the spindle, the APC/C triggers the onset of chromosome separation. The APC/C is activated by Cdc20 but inhibited by a complex called the MCC. By purifying different forms of the APC/C from checkpoint-arrested human cells, Stark, Peters and colleagues have now been able to gain insight into the structure and regulation of the APC/C. The authors isolated the apo-APC/C and the MCC-associated form of APC/C and used single-particle EM to analyze their structures. By adding recombinant Cdc20 to the apo-APC/C, they could also examine the active form of the complex. The EM analyses show that the APC/C consists of a platform plus an ‘arc lamp’-like region that adopts varying conformations in the apo complex. Using antibodies, the authors mapped the positions of particular components, and they argue that the repetitive-looking ‘stem’ of the arc lamp may be formed by the tetratricopeptide repeats present in certain APC/C components. The region to which Apc4 maps has a ring-like feature that may correspond to Apc4’s predicted propeller-shaped WD40 repeats. MCC renders the APC/C more compact and conformationally stable, and seems to overlap with the region where density for Cdc20 resides, suggesting that its inhibition of the APC/C may result from directly interfering with Cdc20 binding. The authors also found that the MCC-bound form of the complex was slightly inhibited in its ability to bind the ubiquitin- conjugating enzyme UbcH10 but was more strikingly defective in substrate binding. These results provide important insights into the structural basis for the regulation of this complex. (Science 323, 1477–1481, 2009) SL

Digesting a switchWhen a DNA double-strand break (DSB) occurs, two related kinases, ATM and ATR, are activated, inducing a checkpoint that halts the cell cycle while the lesion is repaired. In mammals, ATM activation involves binding of the MRN complex to a DSB, whereas ATR activation involves binding of the single-stranded DNA-binding protein RPA to resected single-strand tails. As DSBs are often converted into tailed molecules, it was possible that a hand-off between ATM and ATR activities might occur. Shiotani and Zou have examined the basis for such a switch using nuclear extracts from HeLa cells. They find that ATM activation requires pairing of the ends of double-stranded DNA. The presence of single-strand overhangs (SSOs) reduces ATM activation by interfering with binding of the MRN complex. ATM activation is dependent on its binding to the double strand–single strand junction and not the SSO of a tailed DNA; furthermore, this junction needs to be at the end and not internal to the molecule. In vivo, SSOs are covered by RPA, which recruits ATR-interacting protein (ATRIP). As a linear duplex is progressively digested by an exonuclease to make an SSO in vitro, ATM activation is inhibited, while ATR activation is enhanced. But is it the single-strand tail itself, or activation of ATR by the RPA-ATRIP complex, that downregulates ATM? When the level of ATR was suppressed, ATM activation was still inhibited in damaged cells, suggesting that DSB resection was sufficient. Thus, although ATM is activated by resected DSBs,

Don’t you know you’re toxic?Some transition metals, such as iron and copper, are essential for the catalytic activity of key metalloproteins, whereas others—for example, cadmium or silver—can be very toxic to both prokaryotic and eukaryotic cells. The acquisition and cellular uptake of transition-metal ions and their intracellular transport to specific protein targets are extremely complex processes, and the mechanisms by which cells import sufficient quantities of ‘essential’ transition-metal ions while keeping other, more toxic, transition metals out of the cell are not fully understood. Lewinson et al. recently reported the use of a cell-based ‘metal tolerance’ assay to characterize 18 prokaryotic P-type ATPases, proteins that actively transport metal ions across cellular membranes. The authors were able to definitively classify five of the P-type ATPases as ‘exporters’: two could restore zinc and cadmium tolerance to a metal-sensitive strain of Escherichia coli, and the other three increased tolerance to exogenously supplied copper and silver ions. None of the identified ATPases could nonselectively export all four transition-metal ions. The authors also identified a transition-metal ‘importer,’ HmtA, that was able to mediate the cellular uptake of copper and zinc—‘essential’ transition metals—but not cadmium or silver. Unraveling the exact mechanism by which HmtA achieves the observed selectivity for copper and zinc will require additional work, but it is noteworthy that HmtA homologs are present in several pathogens, suggesting that it may play a role in virulence. (Proc. Natl. Acad. Sci. USA, doi:10.1073/pnas.0900666106; published online 5 March 2009) JMF

r e s e a r c h h i g h l i g h t s

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Bases in the anticodon loop of tRNAAlaGGC

prevent misreadingHiroshi Murakami1, Atsushi Ohta2 & Hiroaki Suga1,2

The bases at positions 32 and 38 in the tRNA anticodon loop are known to have a specific conservation depending upon theanticodon triplets. Here we report that evolutionarily conserved pairs of bases at positions 32 and 38 in tRNAAla

GGC preventmisreading of a near-cognate valine codon, GUC. The tRNAAla

GGC molecules with the conserved A32-U38 and C32-G38 pairsdo not read GUC, whereas those with three representative nonconserved pairs, U32-U38, U32-A38 and C32-A38, direct themisincorporation of alanine at this valine codon into the peptide chain. Overexpression of the nonconserved tRNAAla

GGC inEscherichia coli is toxic and prevents cell growth. These results suggested that the bases at positions 32 and 38 in tRNAAla

GGC

evolved to preserve the fidelity of the cognate codon reading.

Decoding fidelity of translation relies on accurate selection of anaminoacyl-tRNA (aa-tRNA) whose anticodon base-pairs with thecognate codon encoded in the mRNA. It is well known that thethird base pair in the codon-anticodon interaction tolerates awobble pair represented by a U�G interaction, in addition to thecanonical Watson-Click base pair, however, this decoding does notgenerally alter the identity of the amino acid in the peptide, andthus its fidelity is maintained1. On the other hand, if a U�G mispairat the first or second base pair occurs, such a codon-anticodoninteraction does accompany an amino acid alteration, and thereforethis kind of miscoding is generally prohibited. Still, some examplesof misreading of the first codon have been reported in literature.For instance, in Saccharomyces cerevisiae, amber (UAG) mutationsgenerated by UV irradiation are read by Gln-tRNAGln

CUG, causingtermination to be suppressed by glutamine incorporation2–6. InE. coli, mutation of the AGC (serine) codon to GGC (glycine)codon at the catalytic Ser68 residue in b-lactamase is suppressed bythe endogenous Ser-tRNASer

GCU, although the probability of mis-reading resulting in glycine incorporation was estimated to be lessthan 1 in 1,000 (ref. 7).

It is also known that a base mutation or mutations near the junctionof arms in the tRNA cloverleaf structure diminish decoding fidelity.One of the well-known cases is the G24A mutation in theD-stem of tRNATrp

CCA, the so-called Hirsh suppressor tRNA, whichmisreads CGG (arginine) and UGA (see Fig. 1a for a reference of thebase position in a tRNA structure, tRNAAla

GGC)8,9. It was recently shownthat the Hirsh suppressor tRNATrp

CCA elevates the rates of both GTPhydrolysis and accommodation independently from the codon-anticodon interaction, and thus the misreading described aboveoccurs9. These experiments suggest that, remotely, this base in the

tRNA body has a crucial role in controlling the decoding event.Similarly, artificial mutations introduced into the C27-G43 Watson-Crick base pair in the anticodon stem of tRNATrp

CUG increased thefrequency of misreading of the first position wobble10,11. For instance,tRNATrp

CUG bearing the G27-A43 mispair misread the UAG amber codon40 times more frequently than the wild-type pair. Taken together withother biochemical data, it was postulated that such mutations possiblyalter the angle of the junction of the anticodon stem and the centraltRNA L-shaped structure, increasing the frequency of wobble reading10.

Some bases in tRNA anticodon loop are also known to con-tribute to the maintenance of decoding fidelity. Although a typicalexample is base modifications in the anticodon loop that disruptcodon recognition12,13, here we focus on sequence variations in theanticodon loop. For instance, E. coli has tRNAGly with threeisoacceptors for GGN (N can be any base) codons, whereasMycoplasma mycoides has only tRNAGly

UCC for reading these codons.It turns out that the difference in the sequence of the anticodonloop between E. coli tRNAGly

UCC and M. mycoides tRNAGlyUCC is a base

at position 32, in which the former has U32 whereas the latter hasC32, both pairing with A38. Notably, the U32C mutation intro-duced into E. coli tRNAGly

UCC made it capable of reading all fourglycine codons14,15. This suggests that the base at position 32 in theanticodon loop influences the tolerance of the U34�U and U34�Cmispairs in codon-anticodon recognition. As described earlier,however, this misreading does not accompany an amino acidalteration. Hence, the study described above does not explain theimportance of these bases at positions 32 and 38 in decodingfidelity. Nevertheless, this work prompted us to investigate whetherthe conservation of positions 32 and 38 contributes to the ability oftRNAs to correctly decode cognate codons in E. coli.

Received 29 October 2008; accepted 13 February 2009; published online 22 March 2009; doi:10.1038/nsmb.1580

1Research Center for Advanced Science and Technology, University of Tokyo, Meguro-ku, Tokyo, Japan. 2Department of Chemistry and Biotechnology, GraduateSchool of Engineering, University of Tokyo, Bunkyo-ku, Tokyo, Japan. Correspondence should be addressed to H.M. (hmura@rcast.u-tokyo.ac.jp) or H.S.(hsuga@rcast.u-tokyo.ac.jp).

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 5 3

ART IC L E S

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

RESULTSAn evolutionary bias of the 32-38 pair in tRNAAla

GGC

First, we used the tDNA database to look for evolutionary bias in the32-38 pair. Prior to our study, a 1982 report on the comparison of42 kinds of E. coli and bacterial phage tRNA sequences focusing on theanticodon stem-loop region, proposed that some of the base pairs inthe anticodon stem and the bases at positions 37 and 38 might show apreference for certain nucleotides depending upon the base at position36, which forms the first base pair in the codon-anticodon inter-action16. This finding has led to ‘the extended anticodon hypothesis’,which posits that these bases evolved to optimize translation efficiencyand, possibly, decoding fidelity. Furthermore, this hypothesis wasexperimentally verified by suppression of the amber codon by mutanttRNATrp

CUA, and of the ochre codon by mutant tRNAGluUUA, demonstrat-

ing that tRNAs with an extended anticodon sequence showed thehighest suppression efficiency17–20.

More recently, 5,601 bacterial tRNA sequences were extracted fromthe tDNA database and used to analyze the statistical conservation ofbases at the 32 and 38 positions21. Certain tRNAs have a specificsubset of combinations that differ from those of other tRNAs. Forinstance, 99% of tRNAAla

GGC contain either A32-U38 (77%) or C32-G38(22%), whereas the bases contained in bacterial tRNAs in general havefrequencies of 52% for C32-A38, 17% for U32-A38, 11% for U32-U38and 8% for C32-C38. Notably, tRNAAla

GGC derivatives with non-conserved pairs such as U32-U38 and U32-A38 dissociate from the

A site of the E. coli ribosome four to ten times more slowly than thosecontaining A32-U38 (ref. 22). Moreover, the U32C mutation oftRNAGly

CCC, which is 98% conserved with the U32-A38 pair, increasesthe affinity of the tRNA not only to the cognate codon but also to thenear-cognate codons involving third position mismatches21. Theseresults imply that the 32-38 pair influences the affinity of tRNAs in theA site; however, again, these third position mismatches in the codon-anticodon interaction do not alter the amino acid, so it is unclearwhether the evolutionary force driving this bias in the 32-38 pair,which depends on the anticodon triplet, arises from the need tocontrol efficiency in translation or decoding fidelity.

We also independently searched the bacterial tDNA database23 toassess the sequence bias in the 32-38 pair in 84 nonredundantsequences of bacterial tRNAAla

GGC. We found a trend similar to thatpreviously described21 (Fig. 1). Note that no bacterial tRNAAla

GGC

contains U32-A38 and C32-A38 pairs, whereas archeal tRNAAlaGGC has

the U32-A38 pair (18% out of 17 nonredundant sequences) but,again, no C32-A38 pair. Because the above sequence bias in the 32-38pair possibly determines the translation efficiency of Ala-tRNAAla

GGC, weanalyzed the difference in translation efficiency of in vitro transcriptsbetween E. coli tRNAAla

GGC (Fig. 1a) with the conserved A32-U38 pair(wild type) or the C32-G38 pair, as well as the nonconservedU32-U38, U32-A38 and C32-A38 pairs (Fig. 1b). For simplicity, werefer the former and latter sets of tRNAAla

GGC as conserved andnonconserved tRNAAla

GGC, respectively.

No change in the decoding efficiency of the GCC cognate codonTo assess the translation efficiency of each tRNAAla

GGC variant, we usedan E. coli cell-free translation system that was specially reconstitutedfor this experiment. In this system, the native tRNAs were entirelysubstituted with in vitro transcripts of four tRNAs (tRNAfMet

CAU,tRNATyr

GUA, tRNAAspGUC and tRNALys

CUU; we refer to the mixture of thesetRNAs as ‘tRNA mix’) along with a tRNAAla

GGC, referred to as thewPURE system (w stands for ‘withdrawn’). To validate whether this

Frequency (%)E. coli tRNA

a b

Conserved

Nonconserved

38

73

19

5

0

0

43

76

1

2724

32

Figure 1 Structure of wild-type tRNAAlaGGC and its variants (a) wild-type

E. coli tRNAAlaGGC. (b) Frequency in the occurrence of the 32-38 pair in

84 nonredundant sequences of bacterial tRNAAlaGGC. The tRNAAla

GGC with the

A32-U38 and C32-G38 pairs are referred to as conserved tRNAAlaGGC, whereas

those with the U32-U38, U32-A38 and C32-A38 pairs are referred to as

nonconserved tRNAAlaGGC.

Natural tRNAs

a

b

d

c

mRNA

Peptide

5′ 3′

Anticodon32 38

loop

Codon

(A)

(B)

(A)

(B)

(A)

(C)

tRNA mix

(32-38)(32-38)

LanesLanes 1

– A-U U-U U-A C-AC-G

– A-U A-U U-U U-A C-A

+++++–+

1 2 3 4 5 6 7

C-G

2 3 4 5 61

– –

––

–

–

+

+

2 3

A-U

+

––

4

Ala

Lanes

tRNAGGC

tRNAGGC

(32-38)tRNAGGCAla

Leu

Figure 2 Decoding efficiency of the GCC codon by tRNAAlaGGC with the

conserved or nonconserved 32-38 pair. (a) Sequences of mRNA and peptide

used in this study. The GCC (alanine) codon was placed at the fifth position.

(b) Tricine SDS-PAGE analysis of the peptide expressed in the presence of

tRNA mix and wild-type tRNAAlaGGC in the wPURE system. The tRNA mix

consists of in vitro transcripts of tRNAfMetCAU, tRNATyr

GUA, tRNAAspGUC and tRNALys

CUU.

The peptide was expressed at 37 1C for 15 min in the presence of 0.2 mMproteinogenic amino acids (except aspartate) and 50 mM [14C]aspartate.

Arrows indicate alanine-containing peptide (A) and [14C]aspartate (B).

(c) Tricine SDS-PAGE analysis of the peptide in the presence of tRNA mix

and each tRNAAlaGGC variant in the wPURE system. (d) Tricine SDS-PAGE

analysis of the competitive decoding of the GCC codon by Ala-tRNAAlaGGC

and Leu-tRNALeuGGC in the wPURE system. The competition contained 3 mM

tRNALeuGGC and each tRNAAla

GGC variant to a concentration of 3 mM. Arrows

indicate Ala-peptide (A) and Leu-peptide (C).

ART IC L E S

35 4 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

wPURE system was able to function like the ordinary PURE system24

for the expression of a model peptide consisting of amino acidsassigned by the above tRNAs, a 13-mer peptide, MKKKADYKDDDDK(italicized residues indicate a Flag peptide sequence), was expressedfrom the corresponding mRNA (Fig. 2a) in the presence of wild-typetRNAAla

GGC and [14C]Asp in both systems. We determined theexpression level of the peptide by the intensity of the radioactiveband following tricine SDS-PAGE, showing that the wPURE systemfunctioned like the ordinary PURE system for the expression ofthis peptide (Fig. 2b, lane 1 versus lane 4). Most importantly, theexpression was tRNAAla

GGC dependent (lanes 3 and 4). MALDI-TOFanalysis of the peptide expressed in the wPURE system also confirmedthe accuracy of expression (data not shown), indicating that correctreading of the GCC codon could be achieved by tRNAAla

GGC.We then tested the tRNAAla

GGC variants (Fig. 1) in the wPURE systemfor the decoding ability of the respective tRNAs to the GCC cognatecodon. It should be noted that because E. coli alanyl-tRNA synthetase(AlaRS) does not recognize the anticodon loop25–27, all the tRNAAla

GGC

variants were alanylated by AlaRS with virtually the same efficiency(Supplementary Fig. 1 online). Thus, the observed translation effi-ciency is likely to reflect the intrinsic decoding ability of eachtRNAAla

GGC to the GCC codon. Unexpectedly, we observed no differencein incorporation efficiency (Fig. 2c).

To avoid exhausting the energy source of translation, we terminatedthe reaction described above after 15 min (Supplementary Fig. 2online); however, it was still possible that the difference in the decodingability of each tRNAAla

GGC was so small that the apparent translationefficiency was not sensitive enough to reflect to the actual value undersuch conditions. We therefore performed an additional experimentto rule out this possibility. Because E. coli leucinyl-tRNA synthetase(LeuRS) does not recognize the anticodon loop of tRNALeu (refs. 28–30), LeuRS charged leucine on the engineered tRNALeu carryingthe anticodon loop sequence of E. coli wild-type tRNAAla

GGC (Supple-mentary Figs. 1 and 3 online). In fact, when we added tRNALeu

GGC to

the wPURE system instead of tRNAAlaGGC, translation of the same

mRNA took place smoothly (Fig. 2d, lane 1). Notably, this leucine-containing peptide (Leu-peptide) appeared as a faster-migrating bandthan the alanine-containing peptide (Ala-peptide) band in tricineSDS-PAGE (Fig. 2d, lanes 1 and 2). MALDI-TOF analysis alsorevealed a molecular mass consistent with the Leu-peptide (data notshown), indicating that the single substitution of alanine to leucine inthis peptide altered its migration properties. Thus, this feature allowedus to use tricine SDS-PAGE to conveniently assess the expression levelof the individual peptides in competition assays between tRNAAla

GGC

and tRNALeuGGC. We observed no appreciable difference in the intensities

between the Ala- and Leu-peptides generated by any of tRNAAlaGGC

variants competing with tRNALeuGGC (lanes 3–7). These experiments

clearly showed that the conserved and nonconserved tRNAAlaGGC

variants were able to decode the GCC cognate codon with similarefficiencies. We thus suspected that the evolutionary conservation ofthe 32-38 pair in tRNAAla

GGC arose for a different reason(s).

The 32-38 pair controls misreading of GUC near-cognate codonAs sequence variation in the 32-38 pair did not affect decodingefficiency, we turned our investigation toward its decoding fidelity.The wobble pairing at the second G35 in tRNAAla

GGC to a near-cognatevaline codon, GUC, would be expected to alter the amino acidincorporation from valine to alanine. We therefore prepared anothermRNA template based on the previously used mRNA in which theGCC codon was substituted with a GUC codon, and tested whethermisreading by tRNAAla

GGC would result in this substitution (Fig. 3a).We first monitored the background incorporation of valine into the

GUC codon in the wPURE system, which lacks the in vitro transcripts.In the absence of the tRNA mix, mRNA translation did not occur at all(Fig. 3b, lane 2); however, addition of the tRNA mix stimulated theexpression of peptide (Fig. 3b, lane 3). Even though the isolatedbackground-level peptide was present only in trace amounts, MALDI-TOF analysis revealed that it was consistent with the molecular mass

mRNA 5′ 3′Peptide

Natural tRNAs

Anticodonloop

Codon

(32-38)

(32-38)

(32-38)

Lanes

Lanes

Lanes

+ –––

– – –

– –+ +

+tRNA mix

Lanes

(A) (C)

(C)

(D)

(C)

(D)(B)(B)

Calculated mass:

200

150

100

50

0

Val-peptide m/z 1655.795 [M+H]+

Ala-peptide m/z 1627.763 [M+H]+

Val-peptideCodon:GUC

Ala-peptideCodon:GUC

Obs.=1656.419

Obs.=1628.287

1600

1,200 1,400 1,600m/z

1,800 2,000

1680

1600 1680

Inte

nsity

(a.

u.)

200

150

100

50

0

Inte

nsity

(a.

u.)

tRNAValGAC tRNAAla

GGC

tRNA AlaGGC

tRNALeuGAC

tRNAAlaGGC

1 2 3 4–

2

A-U U-U U-A C-AC-G

32 38

1 3 4 5 6

2A-U U-U U-A C-AC-G

A-U U-U U-A C-AC-G

3 µM

tRNALeuGAC

0.3 µM

1 3 4 5

109876

a

b

d e

c

Figure 3 Influence of the sequence variation of the 32-38 pair in tRNAAlaGGC on misreading of GUC codon. (a) Sequences of mRNA and peptide used in this

study. The GUC (valine) codon was placed at the fifth position. (b) Tricine SDS-PAGE analysis of the peptide expressed in the presence of tRNA mix and the

in vitro transcript of tRNAValGAC in the wPURE system. Other conditions were the same as Figure 2b. Arrows indicate Val-peptide (A) and [14C]aspartate (B).

(c) Tricine SDS-PAGE analysis of the peptide in the presence of tRNA mix and each tRNAAlaGGC variant in the wPURE system. Arrows indicate Ala-peptide (C)

(and Val-peptide in lane 1) and [14C]aspartate (B). (d) MALDI-TOF analysis of the peptides expressed above. The Val-peptide (codon: GUC) was obtained

from the expression sample in lane 4 in Figure 3b with aspartate instead of [14C]aspartate. The Ala-peptide (codon: GUC) was obtained from the expression

sample in lane 6 in Figure 3c with aspartate instead of [14C]aspartate. Inset, expansion of the region between 1,600 and 1,680 m/z of the MS spectra.

(e) Competitive decoding of the GUC codon by tRNAAlaGGC variants and tRNALeu

GAC. In lanes 1–5, 3 mM tRNALeuGAC and 3 mM each tRNAAla

GGC variant were used;

in lanes 6–10, 0.3 mM tRNALeuGAC and 3 mM each tRNAAla

GGC variant were used. Arrows indicate Ala-peptide (C), Leu-peptide (D).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 5 5

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

of the valine-containing peptide (Val-peptide) as a major peak (datanot shown). This suggests that the background expression can beattributed to a trace amount of tRNAVal

GAC contaminating the wPUREsystem. On the other hand, addition of the in vitro transcript oftRNAVal

GAC to the wPURE system markedly elevated the expression levelof peptide (Fig. 3b, lane 4).

We then tested whether alanine misincorporation at the GUCcodon could be induced by addition of tRNAAla

GGC variants to thewPURE system. The presence of wild-type or C32-G38 tRNAAla

GGC

slightly increased the background expression, presumably owing tomisreading of the GUC codon resulting in alanine incorporation intothe peptide chain (Fig. 3c, lanes 1–3). Unexpectedly, the presence ofnonconserved tRNAAla

GGC (U32-U38, U32-A38 and C32-A38) substan-tially increased the expression level (Fig. 3c, lanes 4–6, respectively).MALDI-TOF analysis of the isolated peptide showed a single majorpeak of molecular mass corresponding to the Ala-peptide (Fig. 3d).This result clearly shows that the background incorporation at theGUC codon by the contaminated tRNAVal

GAC was completely competedout by the nonconserved tRNAAla

GGC.Even though the nonconserved tRNAAla

GGC misreads GUC effectivelyin the wPURE system, in E. coli the cognate tRNAVal

GAC coexistsendogenously and thus competes out such a misreading event. There-fore, it was necessary to assess how effectively misreading occurredunder the competitive conditions. Because the Val-peptide andthe Ala-peptide had nearly the same migration pattern in tricineSDS-PAGE (Fig. 3b,c), it was difficult to quantitatively assess thecompetition. Instead, we engineered a tRNALeu containing the nativeanticodon loop sequence of E. coli tRNAVal

GAC (Supplementary Fig. 3c)and used it as a competitor against each tRNAAla

GGC variant. As expectedon the basis of previous experiments28–30, LeuRS charged leucineonto the engineered tRNALeu

GAC (Supplementary Fig. 1) and theresulting Leu-tRNALeu

GAC decoded the mRNA GUC codon, yieldingthe Leu-peptide. Because the Leu-peptide migrated faster than theAla-peptide in tricine-SDS-PAGE, we could readily visualize the degreeof competition (Fig. 3e).

When we added an equal amount of each tRNAAlaGGC variant and

tRNALeuGAC to the wPURE system, only the Leu-peptide band was

observed in all cases, suggesting that each Ala-tRNAAlaGGC variant was

completely competed out by Leu-tRNALeuGAC (Fig. 3e, lanes 1–5).

However, when we reduced the concentration of the tRNALeuGAC to

one-tenth that of tRNAAlaGGC, a faint but clearly visible Ala-peptide band

appeared in the presence of the nonconserved tRNAAlaGGC (Fig. 3e,

lanes 6–10). Particularly, the frequency of misreading of GUC byAla-tRNAAla

GGC containing the C32-A38 pair reached approximately30% (Fig. 3e, lane 10). This result clearly indicates that the 32-38 pairin tRNAAla

GGC controls misreading of the near-cognate GUC codon.

Overexpression of the nonconserved tRNAAlaGGC is toxic in E. coli

The above in vitro experiments clearly demonstrated that the non-conserved tRNAAla

GGC misreads the near-cognate GUC codon involvingthe G35�U wobble pair. We wondered whether this misreading eventcould occur in vivo, so that the nonconserved tRNAAla

GGC acts as atoxigenic tRNA. We transformed E. coli BL21 cells with a vector thatcould overexpress each conserved or nonconserved tRNAAla

GGC variantunder the control of an arabinose promoter (Supplementary Fig. 4online). The tranformed cells were grown individually on either 0.2%(w/v) glucose (negative control) or 0.2% (w/v) arabinose on LB agarplates at 42 1C. Before induction of tRNA expression, all cells appearedas healthy as the untransformed control cells (Fig. 4a). Upon induc-tion, cells expressing the conserved tRNAAla

GGC showed no change ingrowth, whereas those expressing the nonconserved tRNAAla

GGC became

unhealthy (Fig. 4b). Particularly, those expressing the nonconservedtRNAAla

GGC with U32-A38 or C32-A38 were unable to grow. TheseU32-A38 and C32-A38 pairs were never found in the tRNAAla

GGC

sequence database, indicating that the sequence bias of the 32-38pair in tRNAAla

GGC probably appeared to avoid formation of toxigenictRNAs in vivo.

It should be noted that at 37 1C most cells appeared to behealthy, with the exception of those cells expressing the nonconservedtRNAAla

GGC with C32-A38, which grew slightly more slowly (data notshown). This temperature sensitivity may suggest that the frequency ofthe misreading of the GUC codon by tRNAAla

GGC with the nonconservedC32-A38 pair is not marked because the codon is predominantly readcorrectly by the cognate tRNAVal

GAC. However, in some proteins theresulting valine to alanine substitution would cause them to be lessstable, resulting in loss of function at 42 1C. This probably led to theobserved temperature-dependent cell growth. Nonetheless, ourdemonstration clearly shows that the nonconserved tRNAAla

GGC istoxic in vivo and is therefore not conserved in the repertoire offunctional tRNAs.

DISCUSSIONHere we provide in vitro evidence that the nonconserved tRNAAla

GGC

(Fig. 1) misreads its near-cognate valine codon, GUC, resulting inmisincorporation of alanine into the valine site of the peptide chain(Fig. 3). In contrast, misreading of this codon by the conservedtRNAAla

GGC (Fig. 1) is minimal and thus is readily competed out bythe cognate tRNALeu

GAC (Fig. 3). This observation is also valid in vivo,where overexpression of the nonconserved tRNAAla

GGC is toxic, whereasthat of the conserved tRNAAla

GGC is not (Fig. 4). These results imply thatthe reason for the evolutionary force selecting the 32-38 pair intRNAAla

GGC is to secure the decoding fidelity.Fidelity of aa-tRNA selection in the ribosome relies on two

mechanistic steps, so-called initial selection and proofreading, whichoccur before and after GTP hydrolysis, respectively31,32. In the initialselection step, incorrect tRNA is rejected by rapid dissociation of theternary complex of aa-tRNA–EF-Tu–GTP from the A site and thesluggish rate of GTP hydrolysis33,34. Even though GTP hydrolysisoccasionally occurs for the incorrect aa-tRNA, in the next proof-reading step the slow accommodation rate of the incorrect aa-tRNA tothe peptidyl-transferase center results in its rejection, and thereforeincorrect reading of the noncognate codon is avoided34. It is likely thatthe sequence variation of 32-38 pair in tRNAAla

GGC also influences eitheror both steps of aa-tRNA selection. It was reported that the non-conserved tRNAAla

GGC(U32-U38 or U32-A38) binding to the cognateGCC codon has a slower dissociation rate from the A site thanthe conserved tRNAAla

GGC(A32-U38 or C32-G38)22. Therefore, an

0.2 % Glucosea b

0.2 % Arabinose

No tRNA No tRNAA-U A-U

C-G C-G

U-U U-UU-A U-A

C-A C-A

Figure 4 Overexpression of the conserved or nonconserved tRNAAlaGGC in

E. coli (BL21). Each tRNAAlaGGC variant was cloned under the control of the

arabinose promoter. LB plates contained 100 mg ml–1 ampicillin in the

presence of 0.2% (w/v) glucose (a) or 0.2% (w/v) arabinose (b) and were

incubated at 42 1C overnight.

ART IC L E S

35 6 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

explanation for the increase in the frequency of misreading of GUCby such nonconserved tRNAAla

GGC is also due to their slow dissociationrate from the ribosome. Recently, various kinetics measurementswere performed for misreading of near-cognate codons, includingGUC, by the conserved tRNAAla

GGC(A32-U38, C32-G38) and non-conserved tRNAAla

GGC(U32-A38 or C32-A38)35. The apparent rate ofpeptide bond formation in misreading of the GUC codon by the twononconserved tRNAAla

GGC(U32-A38 or C32-A38) is elevated to the levelof that which occurs during reading of the cognate GCC codon.Clearly, this result is consistent with our finding that the nonconservedtRNAAla

GGC tends to misread the near-cognate GUC codon.Structures of the anticodon loop with various 32-38 pairs have been

modeled in silico based on the available crystal structures36. The U32-A38 and C32-A38 pairs, belonging to the largest structural family I,form noncanonical structures involving bifurcated hydrogen bonds. Incontrast, the U32-U38 pair, categorized in family II, forms a single,noncanonical hydrogen bond. Structures for the A32-U38 and C32-G38 pairs, in family III, cannot yet be predicted because of insufficientavailable structural information. It should be noted that families I andII combined constitute about 93% of bacterial tRNAs36, implying thatthese base pairs evolved to maximize the decoding ability of tRNAs onthe ribosome. In the present study, we have shown that, paradoxically,the family I tRNAAla

GGC with U32-A38 or C32-A38 and the family IItRNAAla

GGC with U32-U38 misread GUC codon. Consequently, the rarefamily III pairs, A32-U38 and C32-G38, are found in the naturallyoccurring tRNAAla

GGC. This suggests that the decoding fidelity oftRNAAla

GGC is tuned by selecting uncommon 32-38 pairs during theevolution. Presumably, similar unique sequence biases that tunedecoding fidelity can be found in many regions of the tRNA bodysequence37. More extensive sequence analyses of tRNAs and biochem-ical studies on such evolutionarily biased variants will be important toreveal the mechanism of decoding fidelity in translation.

METHODSMaterials. We prepared all of the tRNAs by in vitro run-off transcription using

T7 RNA polymerase38, and the DNA templates of mRNAs (5¢-CGAAG CTAAT

ACGAC TCACT ATAGG GCTTT AATAA GGAGA AAAAC ATGAA GAAGA

AGNNN GACTA CAAGG ACGAC GACGA CAAGT AAGCT TCG -3¢,where NNN indicates GCC or GUC, and the underlined sequence encodes

the T7 promoter) by PCR using Taq DNA polymerase (Supplementary

Methods online).

Translation. We performed batch translation using the PURE system without

the tRNA mixture (wPURE system) according to described protocols39–42. The

translation mixture contained 50 mM HEPES-K+, pH 7.6, 20 mM creatine

phosphate, 100 mM potassium glutamate, 14 mM magnesium acetate, 2 mM

EDTA, 2 mM spermidine, 1 mM DTT, 2 mM ATP, 2 mM GTP, 1 mM UTP,

1 mM CTP and 10 mM 10-formyl-5,6,7,8-tetrahydrofolic acid. The translation

was carried out with 0.02 mM DNA template of mRNA and a 200 mM

concentration of 19 kinds of proteinogenic amino acids without aspartate

and 50 mM [14C]Asp. Natural tRNA extract (1.5 mg ml–1 at final concentration,

Roche) was added in the control experiment. In vitro transcripts of tRNAfMet,

tRNATyr, tRNAAsp (5 mM each tRNA at final concentration) and tRNALys

(40 mM at final concentration) were added instead of natural tRNA extract in

other all experiments. The concentrations of tRNAAlaGGC variants, tRNAVal

GAC, the

engineered tRNALeuGGC and tRNALeu

GAC are described in the figures. The reaction

was carried out in a total volume of 2 ml at 37 1C for 15 min and the products

were analyzed by tricine SDS-PAGE.

Mass spectroscopy measurements of peptides. For MS analysis, we performed

the reactions (5 ml) with a 200 mM concentration of 20 proteinogenic amino

acids. The products were precipitated with 50 ml of acetone, dissolved in

2.5 ml of water and then immobilized with 2.5 ml of Flag–M2 agarose (Sigma).

After the resin was washed twice with 50 ml of W buffer (50 mM Tris-HCl,

pH 8.0, 150 mM NaCl), the immobilized peptides were eluted with 2.5 ml of

0.2% (v/v) trifluoroacetic acid (TFA), desalted with Zip tips C18 (Millipore)

and eluted with 1.5 ml of a 50% (v/v) acetonitrile, 0.1% (v/v) TFA solution

saturated with the matrix (R)-cyano-4-hydroxycinnamic acid. Mass measure-

ments were performed using MALDI-TOF (Autoflex, Bruker) in the positive

mode and externally calibrated with Substance P (average 1,348.66 Da),

Bombesin (average 1,620.88 Da), ACTH clip 1–17 (average 2,094.46 Da) and

Somatostatin 28 (average 3,149.61 Da) as standards.

Construction of plasmids. The DNA fragment was amplified by Pyrobest DNA

polymerase (Takara) from pUC18 using primers (pUCHin.F33, 5¢-GCAAG

CTTGC TCTTC CGCTT CCTCG CTCAC TGA-3¢, and pUCNotPst.R44,

5¢-CCGCT GCAGA CGCGG CCGCG CCTGA TGCGG TATTT TCTCC

TTAC-3¢) and the product was digested with PstI and HindIII. The annealed

DNA fragment (5¢-GATCC TTAGC GAAAG CTAAG GATTT TTTTT A-3¢ and

5¢-AGCTT AAAAA AAATC CTTAG CTTTC GCTAA GGATC TGCA-3¢)containing rrnC terminator was cloned in the PstI–HindIII site of the product

DNA. The resulting plasmid was named pMUC. The DNA region that contains

the araC gene and the PBAD promoter of pBAD–GFPuv (BioRad) was amplified

by PCR using primers (araNot.F35, 5¢-ACGCG GCCGC GCATA ATGTG

CCTGT CAAAT GGACG-3¢, and araEcoPst.R43, 5¢-CCGCT GCAGC AGAAT

TCCCA AAAAA ACGGG TATGG AGAAA CAG-3¢). After NotI-PstI digestion,

the fragment was cloned into the NotI-PstI site of pMUC. The resulting

plasmid was named pMUCA. Template DNA of tRNAAlaGGC variants were

amplified using primers (EcoT7.F26, 5¢-GCGAA TTCTA ATACG ACTCA

CTATA G-3¢, and AlaPst.R35, 5¢-GCGCT GCAGT GTTAT TGGTG GAGCT

AAGCG GGATC-3¢) from the corresponding PCR products described

above and digested with EcoRI and PstI, and then cloned into EcoRI-PstI site

in pMUCA. We confirmed the sequence between NotI-HindIII site by

sequence analysis.

Overexpression of the tRNAAlaGGC variant in E. coli. The plasmids were

transformed into BL21 (Invitrogen) and spread on LB agar plates containing

100 mg ml–1 ampicillin and 4% (w/v) glucose. The plates were incubated at

37 1C overnight and the colonies were cultivated in LB medium containing

100 mg ml–1 ampicillin and 4% (w/v) glucose at 37 1C overnight. The cultures

were diluted by 10� volume of LB medium and streaked on LB agar plates

containing 100 mg ml–1 amplicillin and 0.2% (w/v) glucose or 0.2% (w/v)

arabinose. The plates were incubated at 42 1C overnight.

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ACKNOWLEDGMENTSWe thank O.C. Uhlenbeck and S. Ledoux for their invaluable discussion.This work was supported by grants from the Japan Society for the Promotionof Science (JSPS) Grants-in-Aid for Scientific Research (S) (16101007) to H.S.,a Young Scientists (A) (20681022) to H.M., a JSPS Fellowship (19-1722) to A.O.,a research and development project of the Industrial Science and TechnologyProgram in the New Energy and Industrial Technology DevelopmentOrganization (NEDO) to H.S., the Industrial Technology Research Grant Programin NEDO (05A02513a) to H.M., and the Takeda Science Foundation.

AUTHOR CONTRIBUTIONSThis study was designed by H.M., A.O. and H.S.; all of the experiments wereperformed by H.M.; the paper was written by H.M. and H.S.

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Crick, F.H. Codon–anticodon pairing: the wobble hypothesis. J. Mol. Biol. 19, 548–555(1966).

2. Calderon, I.L., Contopoulou, C.R. & Mortimer, R.K. Isolation of a DNA fragment that isexpressed as an amber suppressor when present in high copy number in yeast. Gene29, 69–76 (1984).

3. Pure, G.A., Robinson, G.W., Naumovski, L. & Friedberg, E.C. Partial suppression of anochre mutation in Saccharomyces cerevisiae by multicopy plasmids containing anormal yeast tRNAGln gene. J. Mol. Biol. 183, 31–42 (1985).

4. Lin, J.P., Aker, M., Sitney, K.C. & Mortimer, R.K. First position wobble in codon-anticodon pairing: amber suppression by a yeast glutamine tRNA. Gene 49, 383–388(1986).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 5 7

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

5. Weiss, W.A. & Friedberg, E.C. Normal yeast tRNACAGGln can suppress amber codons and is

encoded by an essential gene. J. Mol. Biol. 192, 725–735 (1986).6. Weiss, W.A., Edelman, I., Culbertson, M.R. & Friedberg, E.C. Physiological levels of

normal tRNACAGGln can effect partial suppression of amber mutations in the yeast

Saccharomyces cerevisiae. Proc. Natl. Acad. Sci. USA 84, 8031–8034 (1987).7. Toth, M.J., Murgola, E.J. & Schimmel, P. Evidence for a unique first position codon-

anticodon mismatch in vivo. J. Mol. Biol. 201, 451–454 (1988).8. Hirsh, D. Tryptophan transfer RNA as the UGA suppressor. J. Mol. Biol. 58, 439–458

(1971).9. Cochella, L. & Green, R. An active role for tRNA in decoding beyond codon:anticodon

pairing. Science 308, 1178–1180 (2005).10. Schultz, D.W. & Yarus, M. tRNA structure and ribosomal function. I. tRNA nucleotide

27–43 mutations enhance first position wobble. J. Mol. Biol. 235, 1381–1394(1994).

11. Schultz, D.W. & Yarus, M. tRNA structure and ribosomal function. II. Interactionbetween anticodon helix and other tRNA mutations. J. Mol. Biol. 235, 1395–1405(1994).

12. Takai, K. & Yokoyama, S. Roles of 5-substituents of tRNA wobble uridines in therecognition of purine-ending codons. Nucleic Acids Res. 31, 6383–6391 (2003).

13. Agris, P.F., Vendeix, F.A. & Graham, W.D. tRNA’s wobble decoding of the genome:40 years of modification. J. Mol. Biol. 366, 1–13 (2007).

14. Lustig, F. et al. Codon discrimination and anticodon structural context. Proc. Natl.Acad. Sci. USA 86, 6873–6877 (1989).

15. Lustig, F. et al. The nucleotide in position 32 of the tRNA anticodon loop determinesability of anticodon UCC to discriminate among glycine codons. Proc. Natl. Acad. Sci.USA 90, 3343–3347 (1993).

16. Yarus, M. Translational efficiency of transfer RNA’s: uses of an extended anticodon.Science 218, 646–652 (1982).

17. Raftery, L.A. & Yarus, M. Site-specific mutagenesis of Escherichia coli gltT yields aweak, glutamic acid-inserting ochre suppressor. J. Mol. Biol. 184, 343–345(1985).

18. Yarus, M., Cline, S.W., Wier, P., Breeden, L. & Thompson, R.C. Actions of the anticodonarm in translation on the phenotypes of RNA mutants. J. Mol. Biol. 192, 235–255(1986).

19. Raftery, L.A. & Yarus, M. Systematic alterations in the anticodon arm make tRNAGlu-Suoc a more efficient suppressor. EMBO J. 6, 1499–1506 (1987).

20. Smith, D., Breeden, L., Farrell, E. & Yarus, M. The bases of the tRNA anticodon loopare independent by genetic criteria. Nucleic Acids Res. 15, 4669–4686 (1987).

21. Olejniczak, M. & Uhlenbeck, O.C. tRNA residues that have coevolved with theiranticodon to ensure uniform and accurate codon recognition. Biochimie 88,943–950 (2006).

22. Olejniczak, M., Dale, T., Fahlman, R.P. & Uhlenbeck, O.C. Idiosyncratic tuning oftRNAs to achieve uniform ribosome binding. Nat. Struct. Mol. Biol. 12, 788–793(2005).

23. Sprinzl, M., Horn, C., Brown, M., Ioudovitch, A. & Steinberg, S. Compilation of tRNAsequences and sequences of tRNA genes. Nucleic Acids Res. 26, 148–153 (1998).

24. Shimizu, Y. et al. Cell-free translation reconstituted with purified components.Nat. Biotechnol. 19, 751–755 (2001).

25. Hou, Y.M. & Schimmel, P. A simple structural feature is a major determinant of theidentity of a transfer RNA. Nature 333, 140–145 (1988).

26. Francklyn, C. & Schimmel, P. Aminoacylation of RNA minihelices with alanine. Nature337, 478–481 (1989).

27. Tamura, K., Asahara, H., Himeno, H., Hasegawa, T. & Shimizu, M. Identity elements ofEscherichia coli tRNAAla. J. Mol. Recognit. 4, 129–132 (1991).

28. Normanly, J., Ogden, R.C., Horvath, S.J. & Abelson, J. Changing the identity of atransfer RNA. Nature 321, 213–219 (1986).

29. Asahara, H. et al. Recognition nucleotides of Escherichia coli tRNALeu and its elementsfacilitating discrimination from tRNASer and tRNATyr. J. Mol. Biol. 231, 219–229(1993).

30. Tukalo, M., Yaremchuk, A., Fukunaga, R., Yokoyama, S. & Cusack, S. The crystalstructure of leucyl-tRNA synthetase complexed with tRNALeu in the post-transfer-editing conformation. Nat. Struct. Mol. Biol. 12, 923–930 (2005).

31. Rodnina, M.V. & Wintermeyer, W. Fidelity of aminoacyl-tRNA selection on theribosome: kinetic and structural mechanisms. Annu. Rev. Biochem. 70, 415–435(2001).

32. Rodnina, M.V. & Wintermeyer, W. Ribosome fidelity: tRNA discrimination, proofreadingand induced fit. Trends Biochem. Sci. 26, 124–130 (2001).

33. Pape, T., Wintermeyer, W. & Rodnina, M.V. Complete kinetic mechanism of elongationfactor Tu-dependent binding of aminoacyl-tRNA to the A site of the E. coli ribosome.EMBO J. 17, 7490–7497 (1998).

34. Pape, T., Wintermeyer, W. & Rodnina, M. Induced fit in initial selection andproofreading of aminoacyl-tRNA on the ribosome. EMBO J. 18, 3800–3807 (1999).

35. Ledoux, S., Olejniczak, M. & Uhlenbeck, O.C. A sequence element that tunesEscherichia coli tRNAAla

GGC to ensure accurate decoding. Nat. Struct. Mol. Biol. advanceonline publication, doi:10.1038/nsmb.1581 (22 March 2009).

36. Auffinger, P. & Westhof, E. Singly and bifurcated hydrogen-bonded base-pairs in tRNAanticodon hairpins and ribozymes. J. Mol. Biol. 292, 467–483 (1999).

37. Saks, M.E. & Conery, J.S. Anticodon-dependent conservation of bacterial tRNA genesequences. RNA 13, 651–660 (2007).

38. Milligan, J.F., Groebe, D.R., Witherell, G.W. & Uhlenbeck, O.C. Oligoribonucleotidesynthesis using T7 RNA polymerase and synthetic DNA templates. Nucleic Acids Res.15, 8783–8798 (1987).

39. Shimizu, Y., Kanamori, T. & Ueda, T. Protein synthesis by pure translation systems.Methods 36, 299–304 (2005).

40. Goto, Y. et al. Reprogramming the translation initiation for the synthesis of physiolo-gically stable cyclic peptides. ACS Chem. Biol. 3, 120–129 (2008).

41. Kawakami, T., Murakami, H. & Suga, H. Messenger RNA-programmed incorporation ofmultiple N-methyl-amino acids into linear and cyclic peptides. Chem. Biol. 15, 32–42(2008).

42. Sako, Y., Goto, Y., Murakami, H. & Suga, H. Ribosomal synthesis of peptidase-resistantpeptides closed by a nonreducible inter-side-chain bond. ACS Chem. Biol. 3, 241–249(2008).

ART IC L E S

35 8 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

A sequence element that tunes Escherichia coli tRNAAlaGGC

to ensure accurate decodingSarah Ledoux1, Miko$aj Olejniczak1,2 & Olke C Uhlenbeck1

Mutating the rare A32-U38 nucleotide pair at the top of the anticodon loop of Escherichia coli tRNAAlaGGC to a more common

U32-A38 pair results in a tRNA that performs almost normally on cognate codons but is unusually efficient in reading near-cognate codons. Pre–steady state kinetic measurements on E. coli ribosomes show that, unlike the wild-type tRNAAla

GGC, themisreading mutant tRNAAla

GGC shows rapid GTP hydrolysis and no detectable proofreading on near-cognate codons. Similarly,tRNAAla

GGC mutated to contain C32-G38, a pair that is found in some bacterial tRNAAlaGGC sequences, was able to decode only the

cognate codons, whereas tRNAAlaGGC containing a more common C32-A38 pair was able to decode all cognate and near-cognate

codons tested. We propose that many of the phylogenetically conserved sequence elements present in each tRNA have evolvedto suppress translation of near-cognate codons.

Numerous biochemical experiments suggest that the 45 elongatoraminoacyl-tRNAs (aa-tRNAs) in E. coli act as equivalent substratesof the translational machinery. More than 20 different E. coliaa-tRNAs were found to bind elongation factor Tu (EF-Tu) withsimilar affinities1, and 8 show nearly identical rates of dissociationfrom the A site and the P site of encoded E. coli ribosomes2. Recentexperiments have shown that ten different E. coli aa-tRNAs havenearly identical ternary complex binding affinities to the ribosomalentry site and similar rates of GTP hydrolysis and peptide bondformation during decoding3. Despite their uniform functionalproperties, aa-tRNAs are quite different from one another chemi-cally. Phylogenetic analysis of tRNA sequences from 145 bacteriawith fully sequenced genomes indicates that each tRNA isoacceptorhas a unique set of consensus residues distributed throughout themolecule4. In addition, each tRNA species contains different typesand numbers of post-transcriptional modifications in the anti-codon loop and the tertiary core5. Several experiments have shownthat when these consensus residues are mutated or when one ormore of the modifications are removed, the uniform functionalproperties of the aa-tRNA are lost. For example, removing all thepost-transcriptional modifications from aa-tRNAs weakens thebinding to the ribosomal A or P sites of several aa-tRNAs2. Basepair changes in the T-stem of individual tRNAs can either weakenor strengthen their binding affinity to EF-Tu6. Removal of selectedmodifications or changes in the sequence within the body of asuppressor tRNA can also either increase or decrease its ability todecode a termination codon in vivo7–9. These experiments suggestthat the overall chemical composition of every aa-tRNA has been‘tuned’ by evolution such that each aa-tRNA functions equivalentlyin the decoding process.

Although the emerging data support the view that tRNA sequencesare idiosyncratically tuned for uniform behavior during decoding, itdoes not explain the underlying reason why this has occurred. It isunclear what the evolutionary disadvantage would be if the differentaa-tRNAs showed a range of affinities for the ribosome or proceededthrough the decoding pathway at different rates. One possibility is thatthe uniform behavior is related to the need for aa-tRNAs to undergoaccurate decoding. Each aa-tRNA must read its cognate codons, but itmust not efficiently read the structurally similar near-cognate codonscontaining a single-nucleotide mismatch. Numerous experiments haveshown that the introduction of certain tRNA mutations or theremoval of an individual post-transcriptional modification can leadto misreading10–14 or translational frameshifting15–18 in vivo. However,a mechanistic understanding of this phenomenon is limited to theG24A mutation of E. coli tRNATrp, which substantially promotesmisreading of several near-cognate codons19. Here we evaluatedmutations in the anticodon loop of E. coli tRNAAla

GGC that are knownto stabilize binding to ribosomes for their ability to read near-cognatecodons. To achieve this, we used kinetic and thermodynamic assaysthat measure different steps in the decoding process.

RESULTSMutating A32-U38 has little effect on cognate decodingtRNAAla

GGC is the minor alanine isoacceptor in E. coli that selectivelyreads its complementary GCC and wobble GCU codons20. The majorisoacceptor tRNAAla

UGC is capable of reading all four alanine codons, sotRNAAla

GGC is not essential for growth, although its deletion causes aslow-growth phenotype in minimal media21. One of the distinctivestructural features of tRNAAla

GGC is the A-U pair at positions 32 and 38at the top of the anticodon loop (Fig. 1). This combination of residues

Received 31 October 2008; accepted 20 February 2009; published online 22 March 2009; doi:10.1038/nsmb.1581

1Department of Biochemistry, Molecular Biology and Cell Biology, Northwestern University, Evanston, Illinois 60208, USA and 2Institute of Bioorganic Chemistry,Polish Academy of Sciences, Poznan, Poland. Correspondence should be addressed to O.C.U. (o-uhlenbeck@northwestern.edu).

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 5 9

ART IC L E S

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

is rare in bacterial tRNAs, being present only in tRNAAlaGGC and

tRNAProGGG (ref. 5). Recent experiments measuring the binding of

E. coli tRNAAlaGGC to the A site of ribosomes encoded with a comple-

mentary GCC codon showed that the identity of the nucleotide pair atpositions 32 and 38 modulates the tRNA binding affinity22. Whereasthe wild-type tRNAAla

GGC demonstrated an A site binding affinitysimilar to other deacylated elongator tRNAs2, replacement of theA32-U38 pair by U-A, U-U or A-A pairs caused the binding affinityof tRNAAla

GGC to be four- to ten-fold stronger22. Introduction of theC32-G38 pair, which is present in tRNAAla

GGC sequences of some otherbacteria, had no effect on A site binding. This suggested that the rareA32-U38 pair and its phylogenetic alternative, C32-G38, have evolvedto weaken the tRNAAla

GGC binding affinity for ribosomes to ensure thatits affinity is similar to that of other tRNAs.

Because the binding affinity of aa-tRNAs to the ribosomal A sitedoes not measure a step in the normal decoding process, discerningthe relevance of the 32-38 pair for tRNAAla

GGC function requires assaysthat measure decoding directly. As mutated tRNAAla

GGC sequences aremost easily tested using unmodified tRNA transcripts, we firstcompared the decoding properties of unmodified tRNAAla

GGC to pre-vious data obtained for its fully modified counterpart. We assayedunmodified Ala-tRNAAla

GGC on E. coli ribosomes programmed with a27-nucleotide derivative of the initiation region of T4 gp32 mRNAdisplaying the cognate GCC codon in the A site and an AUG codon inthe P site. As previously described in greater detail3, we used threedifferent assays to evaluate the ability of Ala-tRNAAla

GGC to undergodecoding. First, we determined the Kd of a catalytically inactive ternarycomplex bound to the entry site of E. coli ribosomes containingtRNAfMet in the P site23 (Fig. 2a). Second, we determined the rate ofGTP hydrolysis by the ternary complex at several encoded ribosomeconcentrations to deduce kGTPmax, the GTPase rate at saturatingribosome concentrations (Fig. 2b,c). Finally, we measured kpep, theobserved rate of peptide bond formation between fMet-tRNAfMet andAla-tRNAAla

GGC (Fig. 2d). The unmodified tRNAAlaGGC and the previously

assayed modified tRNAAlaGGC showed similar Kd values (2.3 nM

versus 1.7 nM, respectively), kGTPmax (31 s–1 versus 45 s–1, respectively)and kpep (1.7 s–1 versus 2.0 s–1, respectively) when determinedunder identical reaction conditions3. Thus, unlike with several othertRNAs2,10,24, the post-transcriptional modifications have only a smalleffect on the decoding process of tRNAAla

GGC in the conditions used inthese in vitro experiments. This is likely to reflect the fact that nativetRNAAla

GGC has no modifications in the anticodon loop and only fivemodifications in the tertiary core, which do not directly contact theribosome20,25,26 (Fig. 1). Although removal of the modifications in thetertiary core of tRNA can destabilize tRNA structure, these effectsare minimal in the buffer containing 10 mM MgCl2. However, inbuffers containing lower MgCl2 concentrations, such as the high-fidelity buffers often used in translation studies, transcripts are notfully folded27–30.

We next compared the ability of the unmodified wild-type tRNA,tRNAAla

GGC (wt), to decode its cognate codons with a tightly bindingdouble mutant where the wild-type A32-U38 pair was changed to aU32-A38 pair, tRNAAla

GGC (UA)22. Both of these tRNAs were effective indecoding the GCC and GCU codons, but the ternary complexcontaining tRNAAla

GGC (UA) bound to the perfectly complementaryGCC codon approximately two-fold more tightly and the wobbleGCU codon about four-fold more tightly than tRNAAla

GGC (wt)(Table 1). However, tRNAAla

GGC (UA) showed kGTPmax and kpep valuesthat were indistinguishable from tRNAAla

GGC (wt) on both cognatecodons (Table 1). Thus, it seems that, although mutating the A32-U38 pair to U32-A38 in tRNAAla

GGC slightly increases the affinity of theternary complex for the ribosome, it does not affect the subsequentkinetic steps of decoding under the conditions used here.

The A32-U38 pair in tRNAAlaGGC prevents misreading

We assayed the binding affinities of the wild-type and mutanttRNAAla

GGC ternary complexes to the ribosomal entry site using thetwo near-cognate alanine codons GCA and GCG, which introduce anA-G or G-G mismatch into the third position of the codon-anticodonhelix. As would be expected on the basis of previous studies compar-ing wild-type tRNA binding to near-cognate codons19,31, both ternarycomplexes bound the mismatched codons much more weakly than thecognate codons (Fig. 2a and Table 1). However, the ternary complexcontaining tRNAAla

GGC (UA) bound the near-cognate codons at leastfive-fold more tightly than complexes containing tRNAAla

GGC (wt).Hence, the stabilizing effect of mutating the A32-U38 base pair toU-A is similar or even slightly greater on the near-cognate codons thanwas observed with the cognate codons. The rate of ternary complexassociation to ribosomes is the same for both cognate and near-cognate codons32, so it is likely that the stabilizing effect of themutation is due to a slower dissociation rate of the ternary complexoff the ribosome. This would result in tRNAAla

GGC (UA) being lessaccurate in the initial selection step of decoding33. This suggests thatone reason the A32-U38 pair in tRNAAla

GGC has evolved to be so wellconserved is to destabilize ternary complex binding to ribosomes andthereby improve the accuracy of the initial selection steps of decoding.

To assess whether the 32-38 pair also influences translation accuracyin the subsequent steps of decoding, it was first necessary to determinehow well the tRNAAla

GGC (wt) transcript can decode a near-cognatecodon. It was possible to obtain values of kGTP at several ribosomeconcentrations and estimate a kGTPmax using the near-cognate GCAcodon, despite the fact that the ternary complex binds more weakly toribosomes containing mismatched codons (Fig. 2b,c). The value ofkGTPmax was 2.4 s–1, 13-fold slower than the value obtained for thecognate GCC codon. The formation of dipeptide bond on the near-cognate GCA codon occurs much more slowly than kpep ¼ 1.7 s–1

Figure 1 Secondary structure of E. coli tRNAAlaGGC. The nucleotides in bold

with post-transcriptional modifications were not modified in the tRNAs used

for this study. Residues in smaller type are present in E. coli tRNAAlaGGC but

are not conserved among all bacterial tRNAAlaGGC. Positions 32 and 38 in the

anticodon loop are numbered.

ART IC L E S

36 0 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

obtained with the cognate codon, but we could only estimate kpep to beless than 0.05 s–1 (Fig. 2d; see Methods). The slower values of kGTPmax

and kpep have been explained by an induced-fit mechanism in whichthe mismatched codon-anticodon interaction causes incorrect adapta-tion of the tRNA to the ribosome and thereby prevents the ribosomalconformational changes needed to promote rapid catalysis33–35.

The tighter binding tRNAAlaGGC (UA)

showed markedly different behavior fromtRNAAla

GGC (wt) in decoding the near-cognateGCA codon. tRNAAla

GGC (UA) showed akGTPmax value of 27 s–1, substantially fasterthan the value of 2.4 s–1 observed withtRNAAla

GGC (wt) and essentially the same rateas observed on its cognate codon (Fig. 2b,cand Table 1). Similarly, the kpep value of 2.1s–1 on the GCA near-cognate codon was alsosignificantly accelerated, such that it is nearlyequal to the value determined using thecognate GCC codon (Fig. 2d and Table 1).The fact that the fraction of dipeptide formedreached the same level with near-cognatecodons as with cognate codons indicatesthat the mutant tRNA is not considerablyrejected off the ribosome in the presence ofthe near-cognate codon. In other words,tRNAAla

GGC (UA) seems to evade the proof-reading process by stimulating the forward-reaction rates so that it efficiently makesdipeptide on the near-cognate codon.

The difference in initial selection rates fortRNAs on cognate versus near-cognate codonsis increased in buffers containing low concen-trations of MgCl2 and with polyamines33,36, sowe asked whether tRNAAla

GGC (UA) can alsodecode a near-cognate codon in such a high-fidelity buffer (Supplementary results online).Although the Kd of the different ternary com-plexes could not be determined even on thecognate codons in this buffer, owing to poorstability of the ribosome/ternary complexadduct in the filter-retention assay, we wereable to measure the rates of GTP hydrolysis andpeptide bond formation. The apparent rate ofGTP hydrolysis determined in high-fidelitybuffer with 2 mM ribosomes showed that,as in the 10 mM MgCl2 buffer, tRNAAla

GGC

(wt) has a fast rate of hydrolysis on the cognate codon and a muchslower rate on the near-cognate GCA codon, whereas tRNAAla

GGC (UA) hasa similar, fast rate of GTP hydrolysis on both codons (SupplementaryFig. 1 online). However, unlike in the 10 mM MgCl2 buffer (Fig. 2b),the extent of GTP hydrolysis achieved at long incubation times wasonly 20%, reflecting the fact that a substantial fraction of the EF-Tu/GTP

Table 1 Thermodynamic and kinetic parameters for different tRNAAlaGGC on cognate and near-cognate codons

Kd (nM)a kGTPmax (s–1)b kpep (s–1)a

Codon A-U U-A A-U U-A A-U U-A C-G C-A

GGC 2.3 ± 0.40 1.0 ± 0.23 31 ± 13 27 ± 11 1.7 ± 0.21 1.8 ± 0.69 1.5 ± 0.70 1.4 ± 0.27

GCU 5.9 ± 0.93 1.3 ± 0.24 23 ± 3.3 25 ± 3.1 1.4 ± 0.17 1.4 ± 0.16 1.5 ± 0.36 1.8 ± 0.15

GCG B1,000c 175 ± 30

GCA B1,000c 210 ± 52 2.4 ± 0.23 26 ± 18 o0.05d 2.1 ± 0.34 o0.05d 3.4 ± 0.64

GUC o0.05d 1.4 ± 0.36 o0.05d 1.6 ± 0.35

ACC o0.05d 0.49 ± 0.16 o0.05d 2.8 ± 0.79

Error values indicate s.e.m.aValues are the average of at least three independent experiments. bValues were determined based on curves fit to at least four apparent kGTP values determined at different ribosome concentrations.cEstimated value as precise Kd determination exceeded the limits of accurate measurement. dEstimated limit (see Methods).

1

0.8

a b

c d

0.6

Frac

tion

boun

d

0.4

0.2

0

1.5

1

0.5

0

k GT

P a

ppar

ent (

s–1)

k GT

P a

ppar

ent (

s–1)20

15

10

5

0

1

tRNAGGC (wt) GCC

0.8

0.6

Frac

tion

GT

P h

ydro

lyze

dFr

actio

n fM

et-A

la d

ipep

tide

0.4

0.2

0.35

0.3

0.25

0.2

0.15

0.1

0.05

0

00.001 0.01 0.1 1 10

0.001 0.01 0.1 1 10

1000.1 1 10 100Ribosome concentration (nM) Time (s)

Ribosome concentration (µM)

Ribosome concentration (µM)0 2 4

Time (s)

1,000 104

0 1 2 3 4 5

Ala tRNAGGC (UA) GCCAla tRNAGGC (wt) GCAAla tRNAGGC (UA) GCAAla

Figure 2 Comparison of tRNAAlaGGC (wt) to tRNAAla

GGC (UA) on the GCC cognate and GCA near-cognate

codons. (a) Equilibrium dissociation curves of catalytically inactive ternary complexes binding to the

ribosomal entry site. (b) Time course of GTP hydrolysis at a ribosome concentration of 1.7 mM.

(c) Ribosome saturation curve of GTP hydrolysis. (d) Time course of dipeptide formation between

fMet-tRNAfMet and Ala-tRNAAlaGGC. Dipeptide formation for tRNAAla

GGC (wt) on the GCA codon could

not be fit to a simple exponential, so no line was drawn.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 6 1

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

did not have an aa-tRNA stably bound. This presumably arises fromthe poor folding of the transcript tRNA in this buffer. Experimentsmeasuring the rate of peptide bond formation in the high-fidelitybuffer showed that tRNAAla

GGC (wt) rapidly formed dipeptide in thepresence of the cognate codon and not the GCA near-cognate codon,whereas tRNAAla

GGC (UA) could efficiently form dipeptide on bothcodons (Supplementary Fig. 2 online). Because the data collected inthe high-fidelity buffer resembled the data collected in the 10 mMMgCl2 buffer in which tRNA folding was not compromised, we usedthe 10 mM MgCl2 buffer for the remainder of the experiments.

To determine whether tRNAAlaGGC (UA) is also capable of misreading

other near-cognate codons, we used the kpep assay to monitor the rateof misincorporation at ACC (threonine) and GUC (valine) codons,which form mismatches at the first and second codon positions,respectively. Similarly to the results with the mismatched GCA(alanine) codon, tRNAAla

GGC (UA) was able to misread both near-cognate codons with rates and extents of reaction similar to thecognate GCC codon, whereas tRNAAla

GGC (wt) showed slow rates (Fig. 3and Table 1). This indicates that tRNAAla

GGC (UA) has lost its ability toperform accurate decoding on any near-cognate codon.

To determine whether misreading was a phenomenon specific tothe U32-A38 pair, we tested two other mutant tRNAAla

GGC moleculesusing the kpep assay with the two cognate codons and the three near-cognate codons (Table 1). We mutated tRNAAla

GGC to contain twoother 32–38-nucleotide pairs: one (C32-A38) is commonly found inbacterial tRNAs other than tRNAAla

GGC, and another (C32-G38) isconserved in 22% of known bacterial tRNAAla

GGC sequences but ispresent in only 1.2% of all bacterial tRNAs5. tRNAAla

GGC (CA), repre-senting a 32-38 pair present in 52% of all bacterial tRNAs5, was able toread cognate codons normally, but it also rapidly and efficientlymisread all three near-cognate codons, similarly to tRNAAla

GGC (UA).In contrast, tRNAAla

GGC (CG) behaved similarly to tRNAAlaGGC (wt) in

effectively reading the cognate codons, but showed slow rates of kpep

on near-cognate codons. These results are consistent with the fact thattRNAAla

GGC in some bacteria contains the rare C32-G38 pair in place ofA32-U38, but none contains the C32-A38 or U32-A38 pairs5.

DISCUSSIONThe A32-U38 pair was originally identified as a sequence element thatdestabilized binding of tRNAs to the ribosomal A site22. It was

hypothesized that the purpose of this pair in bacterial tRNAAlaGGC and

tRNAProGGG was to off-set the stabilizing effect of their GC-rich codon-

anticodon pairs so that they would act similarly to other tRNAs whendecoding their cognate codons. Although this view may be correct,experiments presented here measuring decoding on near-cognatecodons using pre–steady state kinetics make it clear that a crucialrole of this base pair is to prevent misreading. When this A32-U38 pairis mutated to a more common 32-38 pair, the resulting ternarycomplex can not only bind ribosomes somewhat more tightly thanthe wild-type tRNA but can also stimulate GTP hydrolysis and peptidebond formation equally well on both cognate and near-cognatecodons. As these effects would reduce the ability of tRNAAla

GGC todistinguish cognate from near-cognate codons in both the initialselection and proofreading steps of decoding, it is likely that theA32-U38 pair was selected to maintain translational accuracy.

Once bound to the ribosome, it is astonishing how welltRNAAla

GGC (UA) can function on near-cognate codons. Both themaximal rate of GTP hydrolysis and the rate and extent of peptidebond formation are indistinguishable from what is observed fortRNAAla

GGC (wt) with its cognate codon. In other words, in theseassays the ribosome does not detect that a mismatched codon-anticodon complex has formed, and it allows peptide bond for-mation to occur normally without any proofreading. This abilityof tRNAAla

GGC (UA) to efficiently read near-cognate codons farexceeds the in vitro effects of error-inducing antibiotics37,38.Although the G24A mutation of E. coli tRNATrp also showssubstantial misreading in vitro, it shows no difference in bindingto near-cognate codons on the ribosome as a ternary complex andstill shows substantially reduced levels of peptide bond formationon mismatched codons, indicating that some proofreadingoccurs19. tRNAAla

GGC (UA) can rapidly undergo GTP hydrolysisand peptide bond formation on the near-cognate GCA codon,even in the low-magnesium, high-fidelity buffer in which theunmodified transcript is not as well folded as in the standard10 mM MgCl2 buffer.

Several of the results presented here have been confirmed using anin vitro translation assay with purified components to prepare oligo-peptides from a defined mRNA39. Transcripts of tRNAAla

GGC containingone of the more common 32-38 pairs (C-A, U-A or U-U) wereeffective at incorporating alanine at a GUC (valine) codon, whereastRNAAla

GGC (wt) and tRNAAlaGGC (CG) were not. It is notable that, when a

competitor tRNA with an anticodon cognate to the GUC codon wasadded to the reaction, misincorporation by the tRNAAla

GGC mutants wasstrongly suppressed, presumably because the competitor ternary com-plex can bind its cognate codon much more tightly than the tRNAAla

GGC

mutants. Effective competition by correctly matched tRNAs probablyalso explains why expression of the misreading tRNAAla

GGC (CA) inE. coli has only modest effects on bacterial growth39.

The 32-38 pair modulates the binding of the Ala-tRNAAlaGGC ternary

complex to the ribosomal entry site with a trend similar to how itmodulates binding of the deacylated tRNAAla

GGC to the ribosomal Asite22. It is likely that in both cases the explanation of the sequencespecificity lies in the structure of the anticodon loop, because the32-38 pair of tRNAPhe present in high-resolution crystal structuresdoes not seem to interact directly with the 30S ribosome in eithercomplex35,40. As discussed previously41, the A32-U38 pair may form astable Watson-Crick pair that in turn allows U33 and A37 to form abase pair, resulting in a 3-nucleotide anticodon loop. The observedweaker binding of the wild-type A32-U38 pair would then be due tothe energy required to break these base pairs to rearrangethe loop into a more open conformation upon codon binding. An

0.35

0.30

0.25

0.20

0.15

Frac

tion

fMet

-Ala

dip

eptid

e

0.10

0.05

0

Time (s)

0.001 0.01 0.1 1 10

tRNAGGC (wt) GCCAla

tRNAGGC (UA) GCCAla

tRNAGGC (wt) GUCAla

tRNAGGC (UA) GUCAla

tRNAGGC (wt) ACCAla

tRNAGGC (UA) ACCAla

Figure 3 Time course of peptide bond formation for tRNAAlaGGC (wt) and

tRNAAlaGGC (UA) on the cognate GCC codon (taken from Fig. 2d) and the

mismatched ACC and GUC codons. Only data that can be fit to a simple

exponential is fit to a line (see Methods).

ART IC L E S

36 2 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

alternative, less specific explanation for the destabilizing effect of theA32-U38 pair may be that, compared to other nonconserved nucleo-tide pairs, this particular pair is in some way less able to stabilize thecodon-anticodon helix through stacking interactions.

A different explanation is required to account for how tRNAAlaGGC

(UA) can efficiently stimulate its rapid forward rates on near-cognatecodons once it is bound to ribosomes. Although no high-resolutionX-ray structure of the ternary complex bound to ribosomes isavailable, medium-resolution cryo-EM structures suggest that thestructures of tRNA and possibly EF-Tu are distorted when the ternarycomplex binds to a cognate codon in the entry site26,42,43. It has beenproposed that, when a mismatched codon is present, the alteredstructure of the codon-anticodon helix prohibits this distortion inthe ternary complex, leading to weaker binding and rejection afterGTP hydrolysis34. Presumably, tRNA mutations that promote mis-reading have altered distortability or dynamics that allow them to fitinto the ribosome correctly despite the mismatched codon, asdescribed in the original ‘waggle’ theory44. For example, the misread-ing G24A mutation of tRNATrp lies close to a major site of distortionin the ribosome-bound ternary complexes19,26,43. Although noobvious distortion of this complex is observed in the region of the32-38 pair, the resolutions of the structures are low. In addition, it isunclear whether each ternary complex will distort identically as aresult of their differing tRNA sequences. However, the fact thattRNATrp and tRNAAla

GGC use different positions to avoid the sameinaccurate decoding phenotype highlights the idea that each tRNAis tuned by different elements.

Although it seems that one important selective pressure on tRNAsequences seems to be to perform equivalently in translating theircognate codons, experiments presented here highlight the fact thattRNA consensus sequences can also maintain translational accuracy.Whereas the A32-U38 (or C32-G38) consensus element in tRNAAla

GGC

functions in tuning both ternary complex affinity and decodingaccuracy, it is uncertain whether this will always be the case. Mutatingother tRNAAla

GGC consensus elements does not seem to greatly affectribosome binding22, but these mutants’ ability to misread remains tobe tested. It is possible that the extensive and complex tRNA sequencerequirements associated with each anticodon reflect the apparent needfor tRNA to show a characteristic deformability to ensure accuratedecoding. Other elements, such as post-transcriptional modificationsand the identity of the amino acid, are likely to be important for howthe aa-tRNA functions on the ribosome, similarly to how the nature ofthe amino acid is important for aa-tRNA binding to EF-Tu45. In fact,this has recently been shown to be the case for proline, which has aslower rate of dipeptide formation if esterified to tRNAPhe rather thantRNAPro (ref. 46). If this is the case, mutations of tRNA consensuselements may not always directly affect aa-tRNA function on cognatecodons but may instead affect their ability to avoid translating near-cognate codons.

METHODSMaterials. We prepared tight-coupled 70S ribosomes from E. coli MRE600 cells

as described47. Final ribosome pellets were resuspended in ribosome binding

buffer (RB buffer: 50 mM HEPES, pH 7.0, 30 mM KCl, 70 mM NH4Cl, 10 mM

MgCl2 and 1 mM DTT) and were stored and activated as described2. The

mRNAs used were derivatives of the initiation region of the T4 gp32 mRNA

with the following sequence: 5¢-GGCAAGGAGGUAAAAAUGXXXGCACGU-

3¢, where XXX indicates the codon complementary to the anticodon of the A

site tRNA and the codon 3¢ of the A site has been changed from GCA to AAA

for all mRNAs with an alanine codon in the A site.

We prepared EF-Tu (H84A) as described3. Escherichia coli tRNAAlaGGC was

transcribed from templates generated by primer extension of overlapping DNA

oligonucleotides (IDT) and purified via denaturing PAGE. We performed 3¢[32P] labeling and aminoacylation as previously described48 with typical

aminoacylation yields of 70% for all tRNAs including tRNAfMet.

Ternary complex binding assay. We measured equilibrium binding of ternary

complexes to the entry site of the ribosome as described3. Ternary complex was

formed by first converting 0.6 mM EF-Tu (H84A) to its GTP-bound form by

incubating it with 100 mM GTP, 3 mM phosphoenolpyruvate and 12 U ml–1

pyruvate kinase in RB buffer at 37 1C for 20 min. The GTP-activated EF-Tu

(H84A) was incubated with 3¢ [32P]-labeled Ala-tRNA on ice for 20 min.

A final concentration of o1 nM ternary complex was incubated at 20 1C for

2 min with 0.5–1,300 nM ribosomes, programmed with an excess of mRNA

and tRNAfMet. We separated ribosome-bound ternary complex from free

ternary complex by filtering the sample over nitrocellulose (Whatman

0.45 mm) and positively charged nylon (Amersham 0.45 mm) membranes in

duplicate and washing with ten-fold excess RB buffer. Further washing did not

affect the amount of ternary complex retained on the nitrocellulose filter.

Because filter saturation made data collection with ribosome concentrations

above 1,300 nM unfeasible, we estimated the Kd values for weakly binding

complexes assuming that the extent of binding would reach the same level as

the more tightly bound cognate complexes. Data were quantified using a

phosphorimager (Molecular Dynamics), and binding constants were deter-

mined by fitting the data to a single Michaelis-Menten binding isotherm using

KaleidaGraph software (Synergy Software).

Kinetics experiments. We determined the rate of GTP hydrolysis as

described3,19. Briefly, 300 nM ternary complex was formed with EF-Tu, g-32P

GTP and Ala-tRNA on ice. We removed excess g-32P GTP by filtration through

two P30 spin columns (Bio-Rad) equilibrated with RB buffer. Equal volumes of

ternary complex and programmed ribosomes were mixed for set times in a

KinTek quench flow apparatus and quenched with 40% (v/v) formic acid to

determine apparent GTP-hydrolysis rates at each ribosome concentration

ranging from 0.5–4 mM. Hydrolyzed free 32Pi was separated from g-32P GTP

by thin-layer chromatography (TLC) using PEI cellulose plates run in 0.5 M

KH2PO4. We determined the apparent rates of hydrolysis at each ribosome

concentration by fitting the fraction of GTP hydrolyzed over time to a single-

exponential curvefit. The apparent rates were then plotted over the range of

ribosome concentrations tested to extrapolate the maximal rate of GTP

hydrolysis at a saturating ribosome concentration.

We determined the rate of peptide bond formation as described3,48. Equal

volumes of 50 nM ternary complex containing EF-Tu, GTP and 3¢ [32P]-labeled

Ala-tRNA was mixed with 500 nM ribosomes programmed with excess mRNA

and fMet-tRNAfMet in the P site using a Kintek quench-flow apparatus.

Reactions were quenched in 5 mM sodium acetate, pH 4.5, 100 mM EDTA.

We analyzed samples by S1 nuclease digestion followed by separating cleaved

[32P]-AMP, [32P]–Ala-AMP and [32P]–fMet-Ala-AMP on PEI cellulose TLC

plates in glacial acetic acid/1 M NH4Cl/H2O (5:10:85). The fraction of fMet-Ala

dipeptide formed was calculated compared to the total signal for deacyl,

aminoacyl and dipeptidyl tRNA present. The data for the fraction of dipeptide

formed over time were fit to a single-exponential curvefit to determine the rate

of peptide bond formation.

In experiments that measured the time course of peptide bond formation of

tRNAAlaGGC (wt) or tRNAAla

GGC (CG) on the mismatched codons GCA, GUC or

ACC, little dipeptide formed in the first second but then increasing amounts of

product appeared in the time period up to 10 s (Figs. 2d and 3). At longer

incubation times, the amount of product slowly continued to increase until as

much as 25% dipeptide was formed after 5 min (data not shown). This

may indicate that, in addition to a slow rate of peptide bond formation

on mismatched codons, tRNAAlaGGC shows a rate of rejection that is unusually

slow compared those of to tRNAPhe, tRNATrp and tRNAUGCAla (refs. 19,31,33).

However, because the kinetic curve could not be fit by a simple exponential,

it is also possible that the slow rate of dipeptide formation in these experi-

ments is the result of multiple binding events or even EF-Tu–independent

binding. As a result, we have only estimated a limit for kpep at o0.05 s–1 in

these cases.

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 6 3

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

ACKNOWLEDGMENTSWe thank H. Suga and H. Murakami for discussions and for sharing theirunpublished data. M.O. is supported by the ‘‘Homing’’ grant from theFoundation for Polish Science. This work was funded by the US NationalInstitutes of Health grant GM037552 (to O.C.U.)

AUTHOR CONTRIBUTIONSS.L. and M.O. performed the experiments; S.L., M.O. and O.C.U. contributed tothe design of the study and preparation of the manuscript.

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Louie, A., Ribeiro, N.S., Reid, B.R. & Jurnak, F. Relative affinities of all Escherichiacoli aminoacyl-tRNAs for elongation factor Tu-GTP. J. Biol. Chem. 259, 5010–5016(1984).

2. Fahlman, R.P., Dale, T. & Uhlenbeck, O.C. Uniform binding of aminoacylated transferRNAs to the ribosomal A and P sites. Mol. Cell 16, 799–805 (2004).

3. Ledoux, S. & Uhlenbeck, O.C. Different aa-tRNAs are selected uniformly on theribosome. Mol. Cell 31, 114–123 (2008).

4. Saks, M.E. & Conery, J.S. Anticodon-dependent conservation of bacterial tRNA genesequences. RNA 13, 651–660 (2007).

5. Sprinzl, M., Horn, C., Brown, M., Ioudovitch, A. & Steinberg, S. Compilation of tRNAsequences and sequences of tRNA genes. Nucleic Acids Res. 26, 148–153 (1998).

6. Sanderson, L.E. & Uhlenbeck, O.C. The 51–63 base pair of tRNA confers specificity forbinding by EF-Tu. RNA 13, 835–840 (2007).

7. Yarus, M., Cline, S., Raftery, L., Wier, P. & Bradley, D. The translational efficiency oftRNA is a property of the anticodon arm. J. Biol. Chem. 261, 10496–10505 (1986).

8. McClain, W.H., Schneider, J., Bhattacharya, S. & Gabriel, K. The importance of tRNAbackbone-mediated interactions with synthetase for aminoacylation. Proc. Natl. Acad.Sci. USA 95, 460–465 (1998).

9. Saks, M.E. et al. An engineered Tetrahymena tRNAGln for in vivo incorporation ofunnatural amino acids into proteins by nonsense suppression. J. Biol. Chem. 271,23169–23175 (1996).

10. Kruger, M.K., Pedersen, S., Hagervall, T.G. & Sørensen, M.A. The modification of thewobble base of tRNAGlu modulates the translation rate of glutamic acid codons in vivo.J. Mol. Biol. 284, 621–631 (1998).

11. Nasvall, S.J., Chen, P. & Bjork, G.R. The modified wobble nucleoside uridine-5-oxyacetic acid in tRNAPro

cmo5UGG promotes reading of all four proline codons in vivo. RNA10, 1662–1673 (2004).

12. Claesson, C. et al. Glycine codon discrimination and the nucleotide in position 32 ofthe anticodon loop. J. Mol. Biol. 247, 191–196 (1995).

13. Tsai, F. & Curran, J.F. tRNA2Gln mutants that translate the CGA arginine codon as

glutamine in Escherichia coli. RNA 4, 1514–1522 (1998).14. Diaz, I. & Ehrenberg, M. ms2i6A deficiency enhances proofreading in translation.

J. Mol. Biol. 222, 1161–1171 (1991).15. Bjork, G.R. et al. Transfer RNA modification: influence on translational frameshifting

and metabolism. FEBS Lett. 452, 47–51 (1999).16. Qian, Q. & Bjork, G.R. Structural alterations far from the anticodon of the tRNAGGG

Pro ofSalmonella typhimurium induce +1 frameshifting at the peptidyl-site. J. Mol. Biol.273, 978–992 (1997).

17. Urbonavicius, J., Qian, Q., Durand, J.M., Hagervall, T.G. & Bjork, G.R. Improvement ofreading frame maintenance is a common function for several tRNA modifications.EMBO J. 20, 4863–4873 (2001).

18. Herr, A.J., Atkins, J.F. & Gesteland, R.F. Mutations which alter the elbow region oftRNA2

Gly reduce T4 gene 60 translational bypassing efficiency. EMBO J. 18,2886–2896 (1999).

19. Cochella, L. & Green, R. An active role for tRNA in decoding beyond codon:anticodonpairing. Science 308, 1178–1180 (2005).

20. Mims, B.H., Prather, N.E. & Murgola, E.J. Isolation and nucleotide sequence analysisof tRNAAla

GGC from Escherichia coli K-12. J. Bacteriol. 162, 837–839 (1985).21. Gabriel, K., Schneider, J. & McClain, W.H. Functional evidence for indirect recognition

of G�U in tRNAAla by alanyl-tRNA synthetase. Science 271, 195–197 (1996).

22. Olejniczak, M., Dale, T., Fahlman, R.P. & Uhlenbeck, O.C. Idiosyncratic tuning oftRNAs to achieve uniform ribosome binding. Nat. Struct. Mol. Biol. 12, 788–793(2005).

23. Daviter, T., Wieden, H.J. & Rodnina, M.V. Essential role of histidine 84 in elongationfactor Tu for the chemical step of GTP hydrolysis on the ribosome. J. Mol. Biol. 332,689–699 (2003).

24. Hagervall, T.G., Ericson, J.U., Esberg, K.B., Li, J.N. & Bjork, G.R. Role of tRNAmodification in translational fidelity. Biochim. Biophys. Acta 1050, 263–266 (1990).

25. Yusupov, M.M. et al. Crystal structure of the ribosome at 5.5 A resolution. Science292, 883–896 (2001).

26. Valle, M. et al. Cryo-EM reveals an active role for aminoacyl-tRNA in the accommoda-tion process. EMBO J. 21, 3557–3567 (2002).

27. Sampson, J.R. & Uhlenbeck, O.C. Biochemical and physical characterization of anunmodified yeast phenylalanine transfer RNA transcribed in vitro. Proc. Natl. Acad.Sci. USA 85, 1033–1037 (1988).

28. Nobles, K.N., Yarian, C.S., Liu, G., Guenther, R.H. & Agris, P.F. Highly conservedmodified nucleosides influence Mg2+-dependent tRNA folding. Nucleic Acids Res. 30,4751–4760 (2002).

29. Maglott, E.J., Deo, S.S., Przykorska, A. & Glick, G.D. Conformational transitions of anunmodified tRNA: implications for RNA folding. Biochemistry 37, 16349–16359(1998).

30. Serebrov, V., Vassilenko, K., Kholod, N., Gross, H.J. & Kisselev, L. Mg2+ binding andstructural stability of mature and in vitro synthesized unmodified Escherichia colitRNAPhe. Nucleic Acids Res. 26, 2723–2728 (1998).

31. Kothe, U. & Rodnina, M.V. Codon reading by tRNAAla with modified uridine in thewobble position. Mol. Cell 25, 167–174 (2007).

32. Gromadski, K.B., Daviter, T. & Rodnina, M.V. A uniform response to mismatches incodon-anticodon complexes ensures ribosomal fidelity. Mol. Cell 21, 369–377(2006).

33. Pape, T., Wintermeyer, W. & Rodnina, M. Induced fit in initial selection and proof-reading of aminoacyl-tRNA on the ribosome. EMBO J. 18, 3800–3807 (1999).

34. Sanbonmatsu, K.Y. Alignment/misalignment hypothesis for tRNA selection by theribosome. Biochimie 88, 1075–1089 (2006).

35. Ogle, J.M., Murphy, F.V., Tarry, M.J. & Ramakrishnan, V. Selection of tRNA by theribosome requires a transition from an open to a closed form. Cell 111, 721–732(2002).

36. Gromadski, K.B. & Rodnina, M.V. Kinetic determinants of high-fidelity tRNA discri-mination on the ribosome. Mol. Cell 13, 191–200 (2004).

37. Pape, T., Wintermeyer, W. & Rodnina, M.V. Complete kinetic mechanism of elongationfactor Tu-dependent binding of aminoacyl-tRNA to the A site of the E. coli ribosome.EMBO J. 17, 7490–7497 (1998).

38. Pape, T., Wintermeyer, W. & Rodnina, M.V. Conformational switch in the decodingregion of 16S rRNA during aminoacyl-tRNA selection on the ribosome. Nat. Struct.Biol. 7, 104–107 (2000).

39. Murakami, H., Ohta, A. & Suga, H. Bases in the anticodon loop of tRNAAlaGGC prevent

misreading. Nat. Struct. Mol. Biol. advance online publication, doi:10.1038/nsmb.1580 (22 March 2009).

40. Selmer, M. et al. Structure of the 70S ribosome complexed with mRNA and tRNA.Science 313, 1935–1942 (2006).

41. Olejniczak, M. & Uhlenbeck, O.C. tRNA residues that have coevolved with theiranticodon to ensure uniform and accurate codon recognition. Biochimie 88,943–950 (2006).

42. Stark, H. et al. Ribosome interactions of aminoacyl-tRNA and elongation factor Tu inthe codon-recognition complex. Nat. Struct. Biol. 9, 849–854 (2002).

43. Li, W. et al. Recognition of aminoacyl-tRNA: a common molecular mechanism revealedby cryo-EM. EMBO J. 27, 3322–3331 (2008).

44. Soll, D. & RajBhandary, U. tRNA: Structure, Biosynthesis, and Function, (ASM,Washington DC, 1995).

45. LaRiviere, F.J., Wolfson, A.D. & Uhlenbeck, O.C. Uniform binding of aminoacyl-tRNAsto elongation factor Tu by thermodynamic compensation. Science 294, 165–168(2001).

46. Pavlov, M.Y. et al. Slow peptide bond formation by proline and other N-alkylaminoacids in translation. Proc. Natl. Acad. Sci. USA 106, 50–54 (2009).

47. Powers, T. & Noller, H.F. A functional pseudoknot in 16S ribosomal RNA. EMBO J 10,2203–2214 (1991).

48. Ledoux, S. & Uhlenbeck, O.C. [3¢-32P]-labeling tRNA with nucleotidyltransferase forassaying aminoacylation and peptide bond formation. Methods 44, 74–80 (2008).

ART IC L E S

36 4 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

A feedback regulatory loop involving microRNA-9and nuclear receptor TLX in neural stem cellfate determinationChunnian Zhao, GuoQiang Sun, Shengxiu Li & Yanhong Shi

MicroRNAs have been implicated as having important roles in stem cell biology. MicroRNA-9 (miR-9) is expressed specificallyin neurogenic areas of the brain and may be involved in neural stem cell self-renewal and differentiation. We showed previouslythat the nuclear receptor TLX is an essential regulator of neural stem cell self-renewal. Here we show that miR-9 suppressesTLX expression to negatively regulate neural stem cell proliferation and accelerate neural differentiation. Introducing a TLXexpression vector that is not prone to miR-9 regulation rescued miR-9–induced proliferation deficiency and inhibited precociousdifferentiation. In utero electroporation of miR-9 in embryonic brains led to premature differentiation and outward migration ofthe transfected neural stem cells. Moreover, TLX represses expression of the miR-9 pri-miRNA. By forming a negative regulatoryloop with TLX, miR-9 provides a model for controlling the balance between neural stem cell proliferation and differentiation.

One of the most important issues in stem cell biology is to understandthe molecular mechanisms underlying stem cell self-renewal anddifferentiation. Neural stem cells are a subset of undifferentiatedprecursors that retain the ability to proliferate and self-renew andhave the capacity to give rise to both neuronal and glial lineages1–4.Although the functional properties of neural stem cells have beenstudied extensively, how their self-renewal and differentiation isregulated is not completely understood.

Accumulating evidence indicates that both transcriptional andpost-transcriptional regulation mechanisms are important forregulating genes essential for neural stem cell self-renewal andneurogenesis. miRNAs are a recently identified large family of20–22-nucleotide noncoding RNAs involved in numerous cellularprocesses, including development, proliferation and differentia-tion5,6. Thus, miRNAs are potentially key post-transcriptionalregulators in stem cell self-renewal and differentiation. Distinctsets of miRNAs have been shown to be specifically expressed inembryonic stem cells7,8. Loss of Dicer1 causes embryonic lethalityand loss of stem cell populations9,10. Argonaute family members,key components of the RNA-induced silencing complex (RISC), arerequired for maintaining germline stem cells in various species11.These observations together support a role for miRNAs in stem cellbiology. Several brain-specific miRNAs have recently been identi-fied. Among these miRNAs, miR-9 is expressed specifically inneurogenic regions of the brain during neural development andin adulthood12–15. Whether miR-9 has a role in neural stem cellself-renewal and differentiation remains to be determined.

We have shown that TLX is an essential regulator of neural stem cellself-renewal16. TLX maintains adult neural stem cells in an undiffer-entiated and self-renewable state, in part through transcriptionalrepression of its downstream target genes, encoding p21 and phos-phatase and tensin homolog (Pten), by complexing with histonedeacetylases17. Recently, TLX-positive neural stem cells have beenshown to have a role in spatial learning and memory18. In additionto its function in adult brains, TLX is also involved in neuraldevelopment by regulating cell-cycle progression in neural stem cellsof the developing brain19–21. TLX is therefore a key regulator that actsto establish the undifferentiated and self-renewable state of neuralstem cells, although aspects of its regulation are enigmatic.

Here we demonstrate that miR-9 suppresses TLX expressionthrough the 3¢ untranslated region (UTR) of TLX mRNA, which, inturn, regulates neural stem cell proliferation and differentiation.Increased expression of miR-9 led to reduced mouse neural stemcell proliferation and accelerated neural differentiation, whereas anti-sense knockdown of miR-9 led to increased neural stem cell prolifera-tion. Introducing a TLX expression vector lacking the endogenousTLX 3¢ UTR rescued proliferation deficiency induced by miR-9overexpression and reversed miR-9–promoted precocious differentia-tion. These results suggest that miR-9 regulates neural stem cellproliferation and differentiation, at least in part, through targetingTLX mRNA via its 3¢ UTR. In utero electroporation of miR-9 intoventricular zone neural stem cells in embryonic mouse brains trig-gered premature differentiation and outward migration of the trans-fected cells, similar to that induced by electroporation of the TLX

Received 1 April 2008; accepted 17 February 2009; published online 29 March 2009; doi:10.1038/nsmb.1576

Department of Neurosciences, Center for Gene Expression and Drug Discovery, Beckman Research Institute of City of Hope, Duarte, California, USA. Correspondenceshould be addressed to Y.S. (yshi@coh.org).

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 6 5

ART IC L E S

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

small interfering RNA (siRNA)21. Furthermore, TLX binds to the 3¢genomic sequences of miR-9-1 to inhibit its expression. MiR-9 andTLX thus form a feedback loop to regulate the switch between neuralstem cell proliferation and differentiation.

RESULTSMiR-9 represses TLX expression by targeting its 3¢ UTRWe hypothesized that TLX is targeted by miRNAs to regulate itsexpression. Using the miRanda (http://www.microrna.org)22 andTargetScan (http://genes.mit.edu/targetscan)23 algorithms, miR-9 waspredicted to have a target site in the TLX 3¢ UTR. This target site isconserved in human, mouse, dog and chicken TLX (SupplementaryFig. 1a online). As TLX is specifically expressed in vertebrate fore-brains and is an essential regulator of neural stem cell self-renewal, wefirst asked whether the candidate TLX-targeting miR-9 is expressedin the brain and, specifically, whether this miRNA is expressed inneural stem cells or in their differentiated progeny. Northern blottingrevealed that miR-9 is expressed specifically in the brain (Supplemen-tary Fig. 1b), consistent with previous reports12,13,24. The size of themiRNA was as expected for the mature miR-9 (22 bp). MiR-9 is alsoexpressed in neural stem cells (d0, Fig. 1a). Notably, the expression ofmiR-9 is upregulated during neural differentiation (Fig. 1a), incontrast to the reduced expression of TLX (Fig. 1b).

To validate whether miR-9 targets TLX, we made a luciferasereporter construct with the mouse 1.4-kb TLX 3¢ UTR containingthe predicted miR-9 target site and flanking sequences inserted intothe 3¢ UTR of a Renilla luciferase reporter gene in a siCHECK vector.Increasing amounts of RNA duplexes of mature miR-9 were trans-fected into human embryonic kidney HEK 293 cells along with thecorresponding reporter gene. We observed dose-dependent repressionof the reporter gene in miR-9–treated cells (Fig. 1c). In addition to thesynthetic RNA duplexes, miR-9 was also expressed using a microRNAexpression vector, MDH1-PGK-GFP2 (ref. 25), into which we hadcloned a 489-nucleotide (nt) miR-9-1 genomic sequence including the89-nt miR-9 hairpin precursor and the 200-nt genomic sequenceflanking each side of the precursor. The miR-9 expression vectorrepressed luciferase reporter gene activity (Fig. 1d), similarly to the

miR-9 RNA duplexes (Fig. 1c), suggesting that miR-9 suppresses TLXexpression through its 3¢ UTR. Base-pairing between the miRNA seedsequence and its target gene is needed for miRNA-mediated repressionof the target mRNA26,27. To test whether the predicted miR-9 targetsite in TLX 3¢ UTR is crucial for repression of TLX expression bymiR-9, we introduced point mutations disrupting complementarity inthe predicted miR-9 target site into TLX 3¢ UTR in the luciferasereporter construct. Mutation of the miR-9 target site abolished therepression by miR-9 (Fig. 1e). A miR-9 mutant with compensatorymutations repressed a luciferase reporter gene with a mutated TLX 3¢UTR (Fig. 1e). These results strongly suggest that miR-9 represses TLXexpression through the predicted target site in TLX 3¢ UTR.

Next we tested whether miR-9 targets TLX in neural stem cells. Wetransfected mature miR-9 RNA duplexes into neural stem cellsand examined TLX expression by western and northern blotting.TLX protein and mRNA levels were markedly reduced in miR-9–transfected cells (Fig. 1f,g), indicating that miR-9 downregulatesTLX expression.

MiR-9 regulates neural stem cell fate determinationTo examine whether miR-9 regulates neural stem cell proliferation, wetransfected neural stem cells with increasing concentrations of miR-9RNA duplexes and measured cell proliferation by 5-bromodeoxyuridine (BrdU) labeling of dividing cells. Transfection of miR-9 ledto dose-dependent inhibition of cell proliferation (Fig. 2a,b) with aminimal effect on cell death (Supplementary Fig. 1c,d). Accordingly,reduced expression of TLX and increased expression of p21, encodedby a target gene that is repressed by TLX17, was observed in miR-9–transfected cells in a dose-dependent manner (Fig. 2c). This geneexpression profile is consistent with miR-9–induced inhibition ofneural stem cell proliferation (Fig. 2a,b), suggesting that miR-9negatively regulates neural stem cell proliferation, presumably throughdownregulation of TLX signaling.

To determine whether overexpression of miR-9 regulates neuralstem cell differentiation, we transfected neural stem cells with miR-9RNA duplexes and cultured them under varying conditions. Ascomplete withdrawal of epidermal growth factor (EGF) and fibroblastgrowth factor (FGF) from the medium led to considerable cell death(data not shown), we cultured neural stem cells in N2-supplementedmedia with low EGF and FGF concentrations (1 ng ml–1), whichallowed cell viability with minimal cell proliferation. Over a 7-d timecourse, no difference in neuronal and glial differentiation could be

d0a

c d e

f g

bd3

1 1.7

100

050

25 WT 3′ UTR

TLX

GAPDH

Mut 3′ UTR100 (nM)

80

Rel

ativ

e lu

cife

rase

act

ivity

Rel

ativ

e lu

cife

rase

act

ivity

60

40

20

C miR-9

TLX

GAPDH

TLX

GAPDH

miR-9

miR-9

miR-9mutant

miR-9mutant

miR-9miR-9mutant

1 2 C

0

100

80

60

40

20

0

Rel

ativ

e lu

cife

rase

act

ivity

100

80

60

40

20

0

4.9 5.7

d5 d7 d0 d3 d5 d7

miR-9

U6relativemiR-9 levels

Figure 1 miR-9–directed repression of TLX expression. (a) miR-9 expression

in adult mouse neural stem cells during a differentiation time course. Day 0

(d0) represents the undifferentiated neural stem cell state. U6 was included

as a loading control. Relative miR-9 levels, normalized to U6 levels, are

indicated under the blots, with the miR-9 level in d0 designated as 1.

(b) Western blot analysis of TLX expression in the same differentiation time

course. GAPDH was included as a loading control. (c) miR-9–mediated

repression of the luciferase reporter gene upstream of the TLX 3¢ UTR. The

TLX 3¢ UTR reporter gene was co-transfected with increasing amounts of

miR-9 RNA duplexes or control siRNA duplexes into HEK 293 cells.

(d) Similar reporter assays were performed in cells transfected with the

MDH1-PGK-GFP2 control vector (1) or the miR-9-1 expression vector (2).

(e) Mutation of the miR-9 target site in TLX 3¢ UTR abolished miR-9–

mediated repression. Luciferase reporter gene under the control of wild type

(WT) or mutant (Mut) TLX 3¢ UTR was transfected along with control siRNA(C), a miR-9 mutant with mutations in the seed region complementary to

the mutant TLX 3¢ UTR or wild-type miR-9 into HEK 293 cells. Error bars

indicate s.d. (f,g) miR-9–mediated repression of TLX expression in neural

stem cells revealed by western blot (f) and RT-PCR (g) analyses. A miR-9

mutant with mutations in the seed region was included as a control.

ART IC L E S

36 6 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

detected between control RNA and miR-9–transfected cells (data notshown), suggesting that overexpression of miR-9 alone is not sufficientto trigger neuronal or glial differentiation. However, when neural stemcells were induced for differentiation using forskolin or retinoic acid,transfection of miR-9 promoted both astroglial and neuronal differ-entiation, leading to an increase in the percentage of glial fibrillaryacidic protein (GFAP)-positive astrocytes and Tuj1-positive neurons atday 3 of differentiation (Fig. 2d,e), mimicking day 5 of differentiationin control cells (data not shown). These results indicate that miR-9accelerates differentiation of neural stem cells that have been primedfor differentiation.

To determine whether the effect of miR-9 transfection on neuralstem cell proliferation and differentiation is mediated through TLX, westably transduced neural stem cells with a vector expressing a version ofTLX lacking its 3¢ UTR (TLXD3¢ UTR). Transfection of miR-9 hadno effect on the expression of TLXD3¢ UTR, although miR-9 down-regulated endogenous TLX expression levels (Fig. 3a). Expression of

TLXD3¢ UTR led to a 1.24-fold increase in neural stem cell prolifera-tion. Co-transfection of TLXD3¢ UTR and miR-9 substantially reversedthe proliferative deficiency induced by miR-9 (Fig. 3b,c). Furthermore,whereas transfection of miR-9 increased astroglial differentiation incontrol neural stem cells, we detected no appreciable increase inastrocyte differentiation in TLXD3¢ UTR-transduced cells uponmiR-9 treatment (Fig. 3d). These results strongly suggest that miR-9regulates neural stem cell proliferation and differentiation, at least inpart, by inhibiting TLX expression through its 3¢ UTR.

We further investigated the role of miR-9 in neural stem cellproliferation using 2¢-O-methyl antisense RNA oligonucleotides assmall RNA inhibitors28,29. We synthesized 2¢-O-methyl antisenseoligonucleotide against miR-9 and transfected it into neural stemcells, with 2¢-O-methyl antisense oligonucleotide against green fluor-escent protein (GFP) included as a negative control. Treatment of

40Mergeda

d e

b

c

BrdU

3530

Brd

U+

cel

ls (

%)

0 nM

:miR

-950

nM

100

nM20

0 nM

Con

trol

miR

-9 GFA

P+

cel

ls (

%)

Con

trol

miR

-9

25

2015105

00 50 200100

0 50 200

TLX

TLX

p21

Actin

GAPDH

100

miR-9 (nM)

miR-9 (nM)

0 50 200100

30 *

*

**

*

25

1 2 1 2

20

10

15

5

0

Tuj1

+ c

ells

(%

)

30

25

20

10

15

5

0

miR-9 (nM)

Figure 2 Overexpression of miR-9 regulates neural stem cell proliferation

and differentiation. (a) Cell proliferation in miR-9–transfected neural stem

cells as revealed by BrdU labeling (green). Merged panels show BrdU

staining along with phase-contrast images. A miR-9 mutant with mutations

in the seed region was included as a control, with a total miRNA

concentration of 200 nM in each transfection. (b) Quantification of BrdU-

positive (BrdU+) cells in miR-9–treated neural stem cells. Error bars

indicate s.d. *P ¼ 0.0008 by one-way Anova. (c) Above, western blot

analysis of TLX expression in miR-9–transfected neural stem cells. GAPDH

was included as a loading control. Below, RT-PCR analysis of TLX and p21

in miR-9–transfected neural stem cells. Actin was included as a loading

control. (d) Overexpression of miR-9 promotes glial differentiation. Control

RNA or miR-9–transfected cells were induced into differentiation for 3 d

and immunostained with a GFAP-specific antibody (green). Nuclear DAPI

staining is shown in blue. (e) Overexpression of miR-9 promotes neuronaldifferentiation. Control RNA or miR-9–transfected neural stem cells were

induced to differentiate for 3 d and immunostained with a Tuj1-specific

antibody (red). Nuclear DAPI staining is shown in blue. For both d and e,

error bars indicate s.d. of the mean. * indicates P ¼ 0.03 (d) and

0.018 (e) by the Student’s t-test. About 4,000 cells were quantified

for d and e, respectively.

*

**

*

35–miR-9

miR-9a

b

c

d

+miR-9

–miR-9+miR-9

25

Brd

U+

cel

ls (

%)

+m

iR-9

–miR

-9+

miR

-9–m

iR-9

GFA

P+

cel

ls (

%)

15

10

5

01 2C TLX∆3′ UTR

C TLX∆3′ UTR

TLX∆3′ UTR

TLX∆3′ UTR

MergedBrdU

C

TLX

∆3′ U

TR

C

+– +–

TLX total

Actin

3 4

20

25

15

10

5

0

20

30

Figure 3 TLXD3¢ UTR rescues miR-9–induced neural stem cell proliferation

deficiency. (a) RT-PCR analysis of TLXD3¢ UTR and total TLX expression in

control neural stem cells (C) and TLXD3¢ UTR–expressing cells treated with

control RNA (–miR-9) or miR-9 (+miR-9). Actin was included as a loading

control. (b) Control (C) or TLXD3¢ UTR–expressing cells were transfected

with control RNA (–miR-9) or miR-9 followed by BrdU labeling (green).

Merged panels show BrdU staining along with phase-contrast images.

(c) Quantification of BrdU-positive (BrdU+) cells in control (C) and

TLXD3¢ UTR–expressing cells treated with control RNA (–miR-9) or miR-9.

Error bars indicate s.d. of the mean. * indicates P ¼ 0.003 by the Student¢st-test. ** indicates P ¼ 0.04 by the Student’s t-test. (d) Quantitation of

GFAP-positive (GFAP+) cells in control (C) and TLXD3¢ UTR–expressing cells

treated with control RNA (–miR-9) or miR-9. Error bars indicate s.d. of the

mean. * indicates P ¼ 0.02 by the Student’s t-test. GFP siRNA was

included as the control RNA.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 6 7

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

antisense oligonucleotides against miR-9 led to substantial knockdownof miR-9 mature forms (Fig. 4a). The expression of TLX wasupregulated in miR-9 antisense RNA-treated neural stem cells, alongwith decreased expression of p21 (Fig. 4b). BrdU-labeling analysisrevealed that knockdown of miR-9 led to an increase in cell prolifera-tion (1.37-fold, Fig. 4c,d), consistent with enhanced cell proliferationin TLXD3¢ UTR–transduced neural stem cells (Fig. 3b,c).

miR-9 stimulates neural differentiation in the brainDuring development, neural stem cells reside in the ventricular zoneand migrate out into the cortical plate upon differentiation. Todetermine whether miR-9 influences neural stem cell proliferationand differentiation in vivo, we introduced miR-9 RNA duplexesinto neural stem cells in the ventricular zone of mouse brains atembryonic day E13.5 by in utero electroporation. Electroporatedbrains were analyzed at E15.5. Cells that had taken up miR-9 werelabeled green by coexpression of GFP. Transfection of miR-9 led to amarked decrease of cells that were positivelylabeled for Ki67, a proliferative marker(Fig. 5a), and a substantial increase in thenumber of cells that migrated from the ven-tricular zone to the cortical plate (Fig. 5b,c),suggesting that miR-9 negatively regulatesneural stem cell proliferation and acceleratesneural differentiation.

Immunostaining revealed that transfectionof miR-9 led to decreased TLX expression(Fig. 5d and Supplementary Fig. 2 online).The miR-9–transfected cells that migrated tothe cortical plate lost the neural progenitormarker nestin (Fig. 5d). Instead, these cellsexpressed the neuronal marker double cortin(DCX, Fig. 5d), indicating neuronal differ-entiation. In contrast, the control smallRNA–transfected cells that remained in theventricular zone were nestin-positive andlacked DCX expression (Fig. 5d). Theseresults are similar to those obtained fromin utero electroporation of a TLX siRNA21.Furthermore, in utero electroporation ofTLXD3¢ UTR along with miR-9 rescued theprecocious migration induced by miR-9transfection (Supplementary Fig. 3 online).These results strongly suggest that miR-9regulates neural stem cell differentiationthrough targeting TLX expression in vivo.

In addition to neuronal differentiation, wealso examined whether miR-9 has a role inglial differentiation in vivo. Gliogenesis occurs

from late embryonic stage through to the early postnatal stage30. Weintroduced miR-9 RNA duplexes into the ventricular zone of E14.5brains by electroporation and analyzed glial differentiation at E17.5by staining for GFAP. We detected increased GFAP staining inmiR-9–transfected cells compared to that in miR-9 mutant–transfectedcells (Supplementary Fig. 4 online), suggesting that miR-9 over-expression also promotes astroglial differentiation in vivo.

Regulation of miR-9 gene expression by TLXThree genes, miR-9-1, miR-9-2 and miR-9-3, encode miR-9 inthe mouse genome. Both miR-9-1 and miR-9-2 are expressed inmammalian brains14, whereas miR-9-3 expression has not beendetected in vertebrate brains31. To determine whether miR-9 geneexpression is affected by TLX expression, we performed reverse-transcription PCR (RT-PCR) to assess the expression levels of miR-9-1 and miR-9-2 pri-miRNAs in brains of wild-type and TLX-nullmice. Notably, expression of both miR-9-1 and miR-9-2 pri-miRNAs

*C miR-9

miR-9

BrdU Mergeda

b

c d70

60

50

40

30

20

10

CmiR-9antisense

0

TLXp21

GAPDH

U6

antisense

C miR-9antisense

miR

-9an

tisen

se Brd

U+

cel

ls (

%)

Con

trol

Figure 4 miR-9 antisense RNA promotes neural stem cell proliferation.

(a) 2¢-O-methyl miR-9 antisense RNA knocks down the mature form of

miR-9, as analyzed by northern blot analysis. 2¢-O-methyl GFP antisense

RNA was included as a negative control (C) in all sections. U6 was included

as a loading control. (b) Expression of TLX and p21 in 2¢-O-methyl miR-9

antisense RNA-treated neural stem cells analyzed by RT-PCR. GAPDH was

included as a loading control. (c) Neural stem cells were transfected with

control RNA and 2¢-O-methyl miR-9 antisense RNA. The transfected cells

were labeled by BrdU staining (green). Merged panels show BrdU staining

along with phase-contrast images. (d) Quantification of BrdU-positive (BrdU+)

cells in control (C) and 2¢-O-methyl miR-9 antisense RNA-treated neural stem

cells. Error bars indicate s.d. * indicates P ¼ 0.03 by the Student’s t-test.

*

*1.2a

d

b c 20

15

10

Control-GFP

miR-9-GFP

5

C miR-90

1

0.8

Fol

d di

ffere

nce

in %

of

Ki6

7+G

FP

+/G

FP

+ c

ells

in V

Z

GF

P+

cel

ls in

CP

(%

)

0.6

0.4

0.2

C miR-9

TLXGFP GFP NestinMerge DCXGFP MergeMerge

0

Figure 5 In utero electroporation of miR-9 in embryonic neural stem cells. (a) In utero electroporation

of miR-9 decreased cell proliferation in the ventricular zone (VZ) of embryonic brains. Proliferating cells

were labeled by Ki67. Percentage of Ki67-positive cells relative to GFP-positive cells (Ki67+GFP+/

GFP+) in miR-9–electroporated brains was calculated and normalized with the percentage of Ki67-positive cells in control RNA (C)–electroporated brains. Error bars indicate s.d. * indicates P ¼ 0.002

by the Student’s t-test. (b) Electroporation of miR-9 led to precocious outward cell migration.

Transfected cells express the GFP marker. Control indicates control RNA; DCX, double cortin;

VZ, ventricular zone; CP, cortical plate. Images on the left panels show 10� magnification; middle and

right images show 20� magnification. (c) Quantification of control RNA (C) and miR-9–electroporated

cells (GFP+ cells) that migrated to the cortical plate (CP). Error bars indicate s.d. of the mean.

* indicates P ¼ 0.02 by the Student’s t-test. (d) Immunostaining of cells from control-GFP or

miR-9-GFP–electroporated brains.

ART IC L E S

36 8 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

was upregulated in brains of TLX-null mice (Fig. 6a), suggesting thatexpression of miR-9 precursors can be repressed by TLX.

Sequence analysis revealed several consensus TLX binding sites inthe flanking regions of the miR-9-1 (Fig. 6b) and miR-9-2 genes (datanot shown). We used chromatin immunoprecipitation (ChIP) tofurther explore whether TLX regulates miR-9 expression by directbinding to miR-9 genomic sequences. MiR-9-1 was chosen for thisanalysis because it is induced more substantially in TLX-null brains(Fig. 6a). We designed three pairs of primers for ChIP analysis, withone pair of primers covering both TLX binding sites 1 and 2 (TLX-1/2). The other two pairs of primers were designed for TLX binding sites3 (TLX-3) and site 4 (TLX-4) individually. ChIP assays revealed thatTLX binds to the consensus TLX binding sites, TLX-1/2 and TLX-3,that are downstream of miR-9-1 genes and upstream of a previouslycharacterized REST binding site32,33, whereas no binding was detectedon TLX-4, which is downstream of the REST binding site (Fig. 6c).

Consistent with TLX binding to miR-9-1 genomic sequences, wealso detected the TLX-interacting transcriptional co-repressor HDAC5on the TLX binding sites, TLX-1/2 and TLX-3, in the miR-9-1genomic locus (Fig. 6d). Furthermore, these sites are associatedwith the repressive chromatin marker, trimethylated histone H3 lysine9 (H3K9me3), but are not associated with active chromatin markers,acetylated histone H3 (AcH3) and trimethylated histone H3 lysine 4(H3K4me3)34,35 (Fig. 6d).

To validate the regulation of miR-9-1 gene expression by TLX, wecloned the 1.2-kb miR-9-1 downstream genomic sequence that con-tains the consensus TLX binding sites, TLX-1/2 and TLX-3, andinserted it downstream of a Renilla luciferase reporter gene in thesiCHECK vector. Co-transfection of TLX with the reporter gene inneural stem cells led to a 2.2-fold repression of the reporter activity(Fig. 6e). Mutation of the TLX binding sites substantially relieved therepression mediated by TLX (Fig. 6e). These results together suggestthat TLX represses miR-9 expression by binding to the consensus TLXbinding sites in the 3¢ genomic sequence of miR-9-1.

DISCUSSIONWe show here that microRNA miR-9 and nuclear receptor TLX form afeedback regulatory loop to regulate neural stem cell proliferation anddifferentiation. TLX is highly expressed in neural stem cells but isrepressed upon differentiation16; in contrast, the level of the miR-9mature form is increased upon differentiation13,24,36. The temporalrelationship between miR-9 and TLX expression resembles thatbetween miR-124 and its target genes lamc1, itgb1 and REST 32,37,38.In both instances, when miRNA expression is low, the targets tend tobe expressed at high levels. Conversely, the expression of these targetsis downregulated as the miRNAs accumulate. These data support the

hypothesis that miRNAs induced during differentiation function toensure proper cell fate transitions by suppressing leftover stem cellmaintenance transcripts in stem cells39.

This study demonstrates that miR-9 has an important role in neuralstem cell proliferation and differentiation and that TLX is a key targetof miR-9 in neural stem cells. Every miRNA could have multiple targetgenes23,26 and, indeed, several target genes have been predicted andsome tested for miR-9, including those encoding the transcriptionfactors REST, FoxG1, Senseless and Hairy/E(spl), and components ofthe FGF signaling pathway32,33,40–42. One of the questions addressedhere is whether the cell proliferation and differentiation effectmediated by miR-9 in neural stem cells is directly related to repressionof TLX expression. Transfection of miR-9 into neural stem cells thatare stably transduced with a TLX-expressing vector lacking the 3¢ UTRshowed that such ectopically expressed TLX rescued the proliferativedeficiency induced by overexpression of miR-9 and compromisedmiR-9–induced precocious differentiation. The result of this rescueexperiment suggests that miR-9 regulates neural stem cell proliferationand differentiation through repression of TLX expression. AlthoughTLX is an important target gene of miR-9, other targets may also havea role in miR-9 function in neural stem cells.

Recent evidence suggests that miRNAs often act as fine-tuningdevices rather than as primary gene regulators43. Consistent with thisconcept, we failed to induce neural differentiation by overexpressingmiR-9 alone in cultured neural stem cells. Instead, we detected anaccelerated differentiation program when differentiation of neuralstem cells was induced in culture or in E13.5 brains, where activeendogenous neurogenesis occurs. Furthermore, it has been suggestedthat inhibiting a miRNA may not generate a strong or even detectablephenotype, as expression of its target genes are already repressed at thetranscriptional level, whereas overexpressing a miRNA in cells whereits target genes are highly expressed may render the action of themiRNA more detectable37. In accordance with this theory, we failed todetect a change in cell differentiation in miR-9 antisense RNA–treatedneural stem cells (data not shown). However, we were able to detect aprecocious differentiation program upon miR-9 overexpression inneural stem cells that were primed for differentiation.

In addition to being a direct target of miR-9, TLX also transcrip-tionally inhibits miR-9 genes, suggesting a negative feedback loopbetween TLX and miR-9 that, perhaps, allow rapid transition fromneural stem cells to differentiated cells. In neural stem cells, TLX isexpressed at relatively high levels. During differentiation, as TLX levelsdecrease, miR-9 expression accumulates. In turn, miR-9 suppressesTLX expression post-transcriptionally to further promote neuraldifferentiation. This regulatory loop may represent a key mechanismto sense the intricate balance between cell proliferation and

WTa

c

d

b e

TLX–/–

miR-9-1

miR-9-11340

1840

5411

5750

2211

2545

1365/1799 5338

5338

5338 5511

2419

RE

RE

RE

TLX-1/2

TLX-4

miR-9-1 TLX-3

miR-9-1

miR-9-2

GAPDH

Input IgG anti-TLX

TLX-1/2

Input IgG1 IgG2 AcH3 H3K4me3 HDAC5 H3K9me3

TLX-3

TLX-1/2

3,500

*2,500

1,500

1,000

Rel

ativ

e lu

cife

rase

act

ivity

500

miR-9-1-luc WT MT

0

2,000

3,000

TLX-3

TLX – + – +

TLX-4

Figure 6 Regulation of miR-9 pri-miRNA expression by TLX. (a) RT-PCR

analysis of miR-9 pri-miRNAs, miR-9-1 and miR-9-2, in wild-type and

TLX-null brains. GAPDH was included as a loading control. (b) Schematics

of the miR-9-1 gene with consensus TLX binding sites (TLX-1/2, 3, 4) and

REST-responsive element (RE). The numbers are relative to miR-9 hairpin

precursor ending site. The locations of the PCR primers for ChIP assays are

indicated by arrows. (c) ChIP assays show binding of TLX to the consensus

TLX binding sites downstream of the miR-9-1 gene. (d) ChIP assays show

binding of HDAC5 to TLX binding sites TLX-1/2 and TLX-3. The association

of these sites with H3K9me3, but not with AcH3 and H3K4me3, was

shown in the same assays. (e) TLX represses miR-9-1 reporter activity.

Luciferase activity of the wild-type (WT) or mutant (MT) miR-9-1 reporter

gene was measured in the absence (–) or presence (+) of the transfected

TLX expression vector in mouse neural stem cells. Error bars indicate s.d.

of the mean. * indicates P ¼ 0.009 by the Student’s t-test.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 6 9

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

differentiation and to confer cell fate determination in a timelymanner. Overall, our study suggests that the brain-specific miRNAmiR-9 has a key role in vertebrate brain development. miR-9 providesa novel strategy to control neural stem cell fate determination byforming a feedback regulatory loop with TLX.

METHODSNeural stem cell culture and differentiation. We isolated mouse neural stem

cells from adult mouse forebrains using Percoll gradient as described16 and

cultured them in DMEM F12 medium with 1 mM L-glutamine, N2 supplement

(Gibco-BRL), 20 ng ml–1 EGF, 20 ng ml–1 FGF2 and 50 ng ml–1 heparin for

proliferation. For differentiation, we exposed neural stem cells to DMEM F12

media with N2 supplement, 5 mM forskolin and 0.5% (v/v) FBS or 1 mM

retinoic acid and 0.5% (v/v) FBS.

miR-9 expression vector and reporter construct. We amplified the miR-9-1

gene by PCR from the genomic locus of mouse miR-9-1, which contains the

89-nt hairpin sequence and 200 nt of genomic sequences flanking each side of

it. We inserted the 489-nt DNA fragment into a miRNA expression vector,

MDH1-PGK-GFP2, to generate the miR-9 expression construct. For the

reporter construct, we subcloned DNA fragments encoding the mouse TLX

3¢ UTR (1,813 bp to 3,232 bp) into psiCHECK 2 (Promega) to make the TLX3¢UTR-siCHECK construct. We created the mutant miR-9 target site by site-

directed mutagenesis in the TLX 3¢ UTR-siCHECK vector. We mutated the

wild-type miR-9 binding site 5¢-AACCAAAG-3¢ to 5¢-TTGGTTTC-3¢. To make

the miR-9-1 reporter construct, we inserted the mouse miR-9-1 downstream

genomic sequence (1,340 bp to 2,546 bp downstream of the miR-9 hairpin

structure), which contains the consensus TLX binding sites, TLX-1/2 and

TLX-3, downstream of a Renilla luciferase reporter gene in the psiCHECK2

vector. The miR-9-1 mutant reporter construct has the consensus TLX binding

site 5¢-AAGTCA-3¢ mutated to 5¢-AGATCA-¢3 at TLX-1/2 and TLX-3 sites by

sequential site-directed mutagenesis.

BrdU labeling and immunostaining analysis. We seeded mouse neural stem

cells at 1 � 105 cells per ml in four-well chamber slides. We added BrdU to cells

48 h after seeding and pulsed for 1–2 h. The BrdU-treated cells were fixed and

acid-treated, followed by immunostaining analysis with BrdU-specific anti-

body16. We performed immunostaining using antibodies to BrdU (Accurate;

diluted 1:1,000), Ki67 (Thermo Scientific; 1:400), nestin (Pharmingen; 1:1000),

Tuj1 (Covance; 1:6,000), DCX (Santa Cruz; 1:300) and GFAP (Advance

Immuno; 1:500).

Transfection, western blotting and reporter assays. We transfected plasmid

DNA or DNA-RNA mix using Transfectin (Bio-Rad). We transfected RNA

duplexes using SilentFect (Bio-Rad). For 50 nM, 100 nM or 200 nM final

concentration of miR-9, 0.5 ml, 1 ml or 2 ml of 50 mM RNA duplexes and 1 ml

SilentFect were mixed in 50 ml media, incubated at room temperature

(20–25 1C) for 20 min and added dropwise to cells in a 24-well plate with

450 ml medium to a total volume of 500 ml. The transfected cells were

harvested 48 h after transfection and subjected to subsequent analyses. The

wild-type miR-9 RNA duplex sense sequence is 5¢-ucu uug guu auc uag cug

uau ga-3¢. The mutant miR-9 RNA duplex sense sequence is 5¢-uga aac caa

auc uag cug uau ga-3¢. We carried out western blotting of TLX and GAPDH

using rabbit anti-TLX antibody (1:1,000) and rabbit anti-GAPDH antibody

(Santa Cruz, 1:1,000). We measured Renilla luciferase activity 48 h after

transfection, normalized it with firefly luciferase or the b-galactosidase

internal control and expressed it as relative luciferase activity.

Northern blotting of miRNAs. We extracted total RNA from tissues or

cultured cells by Trizol. We separated 8 mg of RNA on a 10% polyacrylamide

gel containing 8 M urea and transferred the RNA electrophoretically to nylon

membranes. Membranes were cross-linked by UV irradiation and hybridized

overnight with 32P-labeled oligonucleotide probes. We quantified miRNA

signals using Phosphor Imager (Molecular Dynamics). DNA probes for north-

ern blotting include miR-9-antisense probe (as): 5¢-CAT ACA GCT AGA TAA

CCA AAG A-3¢ and U6-as: 5¢-TAT GGA ACG CTT CTC GAA TT-3¢.

Reverse-transcription polymerase chain reaction analysis. We prepared

cDNA from total RNA using the Omniscript Reverse Transcription kit (Qiagen)

for RT-PCR analyses. Primers for RT-PCR include TLX forward primer: 5¢-GTC TTT ACA AGA TCA GCT GAT G-3¢, reverse primer: 5¢-ATG TCA CTG

GAT TTG TAC ATA TC-3¢; GFAP forward primer: 5¢-GCT ACA TCG AGA

AGG TCC GC-3¢, reverse primer: 5¢-GTC TCT TGC ATG TTA CTG GTG-3¢;Tuj-1 forward primer: 5¢-CTG GAG CGC ATC AGC GTA TAC-3¢, reverse

primer: 5¢-ATC TGC TGC GTG AGC TCA GG-3¢; p21 forward primer: 5¢-ATG

TCC AAT CCT GGT GAT GTC CG-3¢, reverse primer: 5¢-TCA GGG TTT TCT

CTT GCA GAA GA-3¢; GAPDH forward primer: 5¢-CAT CAC CAT CTT CCA

GGA GC-3¢, reverse primer: 5¢-GCT GTA GCC GTA TTC ATT GTC-3¢; actin

forward primer: 5¢-ACC TGG CCG TCA GGC AGC TC-3¢, reverse primer: 5¢-CCG AGC GTG GCT ACA GCT TC-3¢.

In utero electroporation. We performed all animal experiments in accordance

with City of Hope and National Institutes of Health guidelines and regulations.

We carried out in utero electroporation as described21. We injected 37.5 pmol of

miR-9 or control RNA duplex into the lateral ventricles of embryos along with

0.625 mg of pActin-EGFP reporter plasmid using electroporator CUY-21

(Protech International). The electroporated mice were allowed to survive for

2 d. Brains of embryos were collected and analyzed as described21.

Chromatin immunoprecipitation assays. We performed ChIP assays using the

EZ-ChIP kit (Upstate) with precleaned chromatin from 2 � 106 mouse neural

stem cells and 5 mg antibody for each ChIP assay. Antibodies used include

antibodies for TLX, H3K4me3 (Cell Signaling Technology), AcH3 (Cell Signal-

ing Technology), HDAC5 (Santa Cruz technology) and H3K9me3 (Abcam).

Primers for ChIP assays include miR-9-1 TLX-1/2 forward primer: 5¢-GGT AGG

GGT GGT GGG GAT GAA-3¢, reverse primer: 5¢-TCT AGG ATG CCC AAG

AAC TTG CT-3¢; miR-9-1 TLX-3 forward primer: 5¢-GCT GGG ACA CTG

GGG ATG CTA GA-3¢, reverse primer: 5¢-AGG AGA GAT CCA TGG AGA TAT

C-3¢; miR-9-1 TLX-4 forward primer: 5¢-TCC AGG CAG ACA TCC TGC ACT

AC-3¢, reverse primer: 5¢-CCT GGT TCT TAG GGA TAC TTC AC-3¢.Additional in utero electroporation procedures and methods for cell-death

analysis are available in Supplementary Methods online.

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ACKNOWLEDGMENTSWe thank J. Rossi and J. Zaia for their critical comments on the manuscript,C.-Z. Chen (Stanford University), H.F. Lodish and D.P. Bartel (MassachusettsInstitute of Technology) for providing the microRNA expression vector MDH1-PGK-GFP2, and Q. Lu (City of Hope) for providing the pEF-pUb-EGFP plasmid.This work was supported by the US National Institutes of Health, NationalInstitute of Neurological Disorders and Stroke grant R01 NS059546 (to Y.S.).

AUTHOR CONTRIBUTIONSY.S., C.Z. and G.S. designed the project; C.Z., G.S. and S.L. performed theexperiments; Y.S. and C.Z. analyzed the data and wrote the manuscript.

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. McKay, R. Stem cells in the central nervous system. Science 276, 66–71(1997).

2. Alvarez-Buylla, A. & Temple, S. Stem cells in the developing and adult nervous system.J. Neurobiol. 36, 105–110 (1998).

3. Gage, F.H., Kempermann, G., Palmer, T.D., Peterson, D.A. & Ray, J. Multipotentprogenitor cells in the adult dentate gyrus. J. Neurobiol. 36, 249–266 (1998).

4. Weiss, S. & van der Kooy, D. CNS stem cells: where’s the biology (a.k.a. beef)?J. Neurobiol. 36, 307–314 (1998).

5. Ambros, V. The functions of animal microRNAs. Nature 431, 350–355 (2004).6. Bartel, D.P. MicroRNAs: genomics, biogenesis, mechanism, and function. Cell 116,

281–297 (2004).7. Houbaviy, H.B., Murray, M.F. & Sharp, P.A. Embryonic stem cell-specific MicroRNAs.

Dev. Cell 5, 351–358 (2003).8. Suh, M.R. et al. Human embryonic stem cells express a unique set of microRNAs. Dev.

Biol. 270, 488–498 (2004).9. Bernstein, E. et al. Dicer is essential for mouse development. Nat. Genet. 35,

215–217 (2003).

ART IC L E S

37 0 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

10. Wienholds, E., Koudijs, M.J., van Eeden, F.J., Cuppen, E. & Plasterk, R.H. ThemicroRNA-producing enzyme Dicer1 is essential for zebrafish development. Nat.Genet. 35, 217–218 (2003).

11. Carmell, M.A., Xuan, Z., Zhang, M.Q. & Hannon, G.J. The Argonaute family: tentaclesthat reach into RNAi, developmental control, stem cell maintenance, and tumorigen-esis. Genes Dev. 16, 2733–2742 (2002).

12. Lagos-Quintana, M. et al. Identification of tissue-specific microRNAs from mouse.Curr. Biol. 12, 735–739 (2002).

13. Krichevsky, A.M., King, K.S., Donahue, C.P., Khrapko, K. & Kosik, K.S. A microRNAarray reveals extensive regulation of microRNAs during brain development. RNA 9,1274–1281 (2003).

14. Deo, M., Yu, J.Y., Chung, K.H., Tippens, M. & Turner, D.L. Detection of mammalianmicroRNA expression by in situ hybridization with RNA oligonucleotides. Dev. Dyn.235, 2538–2548 (2006).

15. Kapsimali, M. et al. MicroRNAs show a wide diversity of expression profiles in thedeveloping and mature central nervous system. Genome Biol. 8, R173 (2007).

16. Shi, Y. et al. Expression and function of orphan nuclear receptor TLX in adult neuralstem cells. Nature 427, 78–83 (2004).

17. Sun, G., Yu, R.T., Evans, R.M. & Shi, Y. Orphan nuclear receptor TLX recruits histonedeacetylases to repress transcription and regulate neural stem cell proliferation. Proc.Natl. Acad. Sci. USA 104, 15282–15287 (2007).

18. Zhang, C.L., Zou, Y., He, W., Gage, F.H. & Evans, R.M. A role for adult TLX-positiveneural stem cells in learning and behaviour. Nature 451, 1004–1007 (2008).

19. Roy, K. et al. The Tlx gene regulates the timing of neurogenesis in the cortex.J. Neurosci. 24, 8333–8345 (2004).

20. Stenman, J.M., Wang, B. & Campbell, K. Tlx controls proliferation and patterning oflateral telencephalic progenitor domains. J. Neurosci. 23, 10568–10576 (2003).

21. Li, W. et al. Nuclear receptor TLX regulates cell cycle progression in neural stem cellsof the developing brain. Mol. Endocrinol. 22, 56–64 (2008).

22. John, B. et al. Human microRNA targets. PLoS Biol. 2, e363 (2004).23. Lewis, B.P., Shih, I.H., Jones-Rhoades, M.W., Bartel, D.P. & Burge, C.B. Prediction of

mammalian microRNA targets. Cell 115, 787–798 (2003).24. Sempere, L.F. et al. Expression profiling of mammalian microRNAs uncovers a subset

of brain-expressed microRNAs with possible roles in murine and human neuronaldifferentiation. Genome Biol. 5, R13 (2004).

25. Chen, C.Z., Li, L., Lodish, H.F. & Bartel, D.P. MicroRNAs modulate hematopoieticlineage differentiation. Science 303, 83–86 (2004).

26. Lewis, B.P., Burge, C.B. & Bartel, D.P. Conserved seed pairing, often flanked byadenosines, indicates that thousands of human genes are microRNA targets. Cell 120,15–20 (2005).

27. Brennecke, J., Stark, A., Russell, R.B. & Cohen, S.M. Principles of microRNA-targetrecognition. PLoS Biol. 3, e85 (2005).

28. Meister, G., Landthaler, M., Dorsett, Y. & Tuschl, T. Sequence-specific inhibition ofmicroRNA- and siRNA-induced RNA silencing. RNA 10, 544–550 (2004).

29. Hutvagner, G., Simard, M.J., Mello, C.C. & Zamore, P.D. Sequence-specific inhibitionof small RNA function. PLoS Biol. 2, e98 (2004).

30. Qian, X. et al. Timing of CNS cell generation: a programmed sequence of neuron andglial cell production from isolated murine cortical stem cells. Neuron 28, 69–80(2000).

31. Kim, J. et al. A microRNA feedback circuit in midbrain dopamine neurons. Science317, 1220–1224 (2007).

32. Conaco, C., Otto, S., Han, J.J. & Mandel, G. Reciprocal actions of REST and amicroRNA promote neuronal identity. Proc. Natl. Acad. Sci. USA 103, 2422–2427(2006).

33. Wu, J. & Xie, X. Comparative sequence analysis reveals an intricate network amongREST, CREB and miRNA in mediating neuronal gene expression. Genome Biol. 7, R85(2006).

34. Li, B., Carey, M. & Workman, J.L. The role of chromatin during transcription. Cell 128,707–719 (2007).

35. Kouzarides, T. Chromatin modifications and their function. Cell 128, 693–705(2007).

36. Krichevsky, A.M., Sonntag, K.C., Isacson, O. & Kosik, K.S. Specific microRNAsmodulate embryonic stem cell-derived neurogenesis. Stem Cells 24, 857–864(2006).

37. Cao, X., Pfaff, S.L. & Gage, F.H. A functional study of miR-124 in the developingneural tube. Genes Dev. 21, 531–536 (2007).

38. Visvanathan, J., Lee, S., Lee, B., Lee, J.W. & Lee, S.K. The microRNA miR-124antagonizes the anti-neural REST/SCP1 pathway during embryonic CNS development.Genes Dev. 21, 744–749 (2007).

39. Farh, K.K. et al. The widespread impact of mammalian microRNAs on mRNArepression and evolution. Science 310, 1817–1821 (2005).

40. Bredenkamp, N., Seoighe, C. & Illing, N. Comparative evolutionary analysis of theFoxG1 transcription factor from diverse vertebrates identifies conserved recognitionsites for microRNA regulation. Dev. Genes Evol. 217, 227–233 (2007).

41. Li, Y., Wang, F., Lee, J.A. & Gao, F.B. MicroRNA-9a ensures the precise specification ofsensory organ precursors in Drosophila. Genes Dev. 20, 2793–2805 (2006).

42. Leucht, C. et al. MicroRNA-9 directs late organizer activity of the midbrain-hindbrainboundary. Nat. Neurosci. 11, 641–648 (2008).

43. Hornstein, E. & Shomron, N. Canalization of development by microRNAs. Nat. Genet.38, S20–S24 (2006).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 7 1

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

TRF2 functions as a protein hub and regulates telomeremaintenance by recognizing specific peptide motifsHyeung Kim1,3, Ok-Hee Lee1,3, Huawei Xin1,3, Liuh-Yow Chen1, Jun Qin1, Heekyung Kate Chae1,Shiaw-Yih Lin2, Amin Safari1, Dan Liu1 & Zhou Songyang1

In mammalian cells, the telomeric repeat binding factor (TRF) homology (TRFH) domain–containing telomeric proteins TRF1and TRF2 associate with a collection of molecules necessary for telomere maintenance and cell-cycle progression. However, thespecificity and the mechanisms by which TRF2 communicates with different signaling pathways remain largely unknown. Usingoriented peptide libraries, we demonstrate that the TRFH domain of human TRF2 recognizes [Y/F]XL peptides with the consensusmotif YYHKYRLSPL. Disrupting the interactions between the TRF2 TRFH domain and its targets resulted in telomeric DNA-damage responses. Furthermore, our genome-wide target analysis revealed phosphatase nuclear targeting subunit (PNUTS) andmicrocephalin 1 (MCPH1) as previously unreported telomere-associated proteins that directly interact with TRF2 via the [Y/F]XLmotif. PNUTS and MCPH1 can regulate telomere length and the telomeric DNA-damage response, respectively. Our findingsindicate that an array of TRF2 molecules functions as a protein hub and regulates telomeres by recruiting different signalingmolecules via a linear sequence code.

Telomere dysfunction has been implicated in cancer and aging1–9.Mammalian chromosomal ends contain long tracts of duplex telomererepeats with 3¢ single-stranded G overhangs10. The telosome/shelterincomplex, which includes TRF1, TRF2, TRF1-interacting nuclearfactor 2 (TIN2), RAP1 (also known as TERF2IP), TPP1 (forTINT1/PIP1/PTOP) and protection of telomeres 1 (POT1), helps tomaintain telomere integrity by protecting the telomeres from chromo-somal abnormalities and DNA-damage responses due to telomerereplication, recombination and erosion11,12. Both TRF1 and TRF2contain a TRFH domain, which mediates homodimerization, and amyb domain, which directly binds the telomeric double-strandedDNA13,14. In addition to telomeric DNA, TRF1 and TRF2 alsoassociate with various proteins involved in telosome assembly, telo-mere-length regulation, DNA replication, repair, end joining, recom-bination and cell-cycle control11,12,15. Consistent with the essentialroles of TRF1 and TRF2, homozygous inactivation of either generesulted in early embryonic lethality in mice16,17. In cultured cells,impairment of TRF2 function (for example, dominant negativeexpression of TRF2DBDM, which lacks the basic and myb domains)led to DNA-damage responses18,19, telomere loop deletion20 oranaphase bridging21. However, the mechanisms of TRF2-mediatedinteraction and the direct targets of TRF2 remain elusive.

One known target of TRF2 in telomere maintenance is the exo-nuclease Apollo22,23. Indeed, biochemical and structural analysesrevealed a direct interaction between the TRFH domain of TRF2(TRF2TRFH) and a short Apollo peptide sequence (500-LALK

YLLTPVNFFQA-514)24. Notably, TRF1TRFH and TRF2TRFH seem toharbor distinct binding specificities, suggesting differential recruitmentof distinct proteins by different TRFH domains. However, the specificdeterminant for TRF2TRFH recognition and the identities of other TRF2targets remain unknown. Here we investigated the specificity ofTRF2TRFH and demonstrated that TRF2TRFH is a protein domain thatrecognizes specific peptides with the [Y/F]XL motif. Through proteo-mic analyses, we identified several [Y/F]XL motif–containing proteinsthat can directly interact with TRF2 and mediate telomere-lengthcontrol and end protection. Our results indicate that an arrayof TRF2 molecules at the telomeres serves as a protein hub fortelomeric signaling.

RESULTSDetermining the binding specificity of TRF2TRFH

We reasoned that TRF2TRFH might represent a modular protein-protein interaction domain whose specificity could be studied usingthe oriented peptide library technique25. We therefore synthesized anoriented peptide library with the sequence KGXXXX[FYWH]X[ILV]XPXN (where X is any amino acid other than cysteine).Because Tyr504, Leu506 and Pro508 of Apollo are essential for itsinteraction with TRFH24, we partially fixed the corresponding posi-tions (P0, P+2 and P+4) in the library (as indicated by squarebrackets) (Fig. 1a). Peptide mixtures that specifically associated withglutathione S-transferase (GST)-TRF2TRFH fusion proteins were iso-lated and sequenced. Among the four aromatic residues partially fixed

Received 10 September 2008; accepted 10 February 2009; published online 15 March 2009; doi:10.1038/nsmb.1575

1Verna and Marrs McLean Department of Biochemistry and Molecular Biology, Baylor College of Medicine, Houston, Texas 77030, USA. 2Department of SystemBiology, The University of Texas M.D. Anderson Cancer Center, Houston, Texas 77054, USA. 3These authors contributed equally to this work. Correspondence should beaddressed to Z.S. (songyang@bcm.tmc.edu).

37 2 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

ART IC L E S

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

at position P0, TRF2TRFH preferred tyrosine and phenylalanine to alesser extent) for binding (Fig. 1b). At the P+2 position, leucine (butnot isoleucine or valine) was selected. Additional selections at otherpositions were also evident, including tyrosine at P�3, lysine at P�1and arginine at P+1. Indeed, a synthesized consensus peptide YRL(KGYYHKYRLSPLN) bound the TRFH domain of TRF2 with highaffinity (190 nM) (Fig. 1c). Furthermore, alanine substitution of theYXL motif in the YRL peptide resulted in its loss of TRF2 interaction,whereas alanine substitution of the P+1 residue arginine or the P+4residue proline reduced its affinity by more than ten-fold (Fig. 1d).These results indicate that TRF2TRFH recognizes specific peptidesequences with the core motif of [Y/F]XL.

TRF2–[Y/F]XL interaction is crucial for telomere maintenanceThese findings suggest that TRFH domains may recruit differentsignaling molecules via a linear sequence code, in a manner similarto other protein-protein interaction modules such as the SH3 andWW domains26,27. In addition, disruption of TRF2TRFH interactionwith its cellular targets may trigger telomere dysfunction. To furtherinvestigate the biological importance of the TRFH–[Y/F]XL motifinteraction, we expressed the TRF2 consensus peptide in tandemrepeats (2�YRL) in human HTC75 cells. We reasoned that thistandem repeat peptide should occupy the two [Y/F]XL binding siteson a TRFH dimer and act as dominant negatives to inhibit endo-genous TRF2 activity by competing for TRF2TRFH binding. Indeed,whereas the chromatin association of either TRF1 or TRF2 remainedintact (Supplementary Fig. 1 online), expression of this peptide didelicit DNA-damage responses, as measured by p53 binding protein 1(53BP1)-containing telomere dysfunction–induced foci (TIF)18,28

(Fig. 2a,b). In contrast, expression of the control peptide (2�YRA)

that does not bind TRF2 had no effect, underlining the important roleof the TRFH–YXL motif interaction in telomere maintenance.

On the basis of the TRF2–Apollo crystal structure, TRF2 Phe120,sandwiched between the P+4 proline of the Apollo peptide and TRF2Arg109, is essential for the TRF2TRFH–YRL interaction24. We thereforegenerated the F120A mutant form of TRF2 (TRF2 FA, Fig. 2c) andfound that it expressed at a level comparable to wild-type TRF2expression (Fig. 2d). Consistent with our biochemical and structuralanalyses, the F120A mutation led to a dramatic reduction in YRLpeptide binding (Fig. 2d). Furthermore, although TRF2 FA stillretained its ability to interact with wild-type TRF2 (Fig. 2e), itsexpression in HTC75 cells increased the percentage of TIF-containingcells compared to that of cells expressing wild-type TRF2 (8% versus2%; Fig. 2f–h). Notably, the percentage of TRF2 FA–expressing cellsthat contained TIF was similar to that of TRF2DBDM-expressing cells(Fig. 2g).

The mechanism of how TRF2DBDM expression triggers telomericDNA-damage responses has remained poorly understood18,29. Vastlyoverexpressed TRF2DBDM can lead to the displacement of endogenousTRF2 from telomeres21. In our experiments, modestly overexpressedTRF2DBDM could associate with telomeres and did not drasticallyalter the chromatin association of endogenous TRF2 (SupplementaryFig. 2a,b online). We reasoned that the intact TRFH domain onTRF2DBDM allowed it to act in a dominant negative manner,preventing multiple [Y/F]XL motif–containing proteins from bindingto endogenous TRF2. Indeed, alanine mutation of Phe120 onTRF2DBDM abolished the effect of TRF2DBDM expression in TIFassays (Fig. 2f,g). Collectively, these experiments indicate that theassociation of TRFH with different [Y/F]XL motif–containing targetsis crucial for TRF2-mediated telomere protection in human cells.

K G X X X X F X I X P X N

Apollo

Position

Library

R G L A L K Y L L T P V N

–4 –3 –2 –1 0 +1 +2 +3 +4 +5

Y LH VW

+a

Peptide sequencing and decoding

Bound peptides

GST-TRFHdomain beads

Orientedpeptide library

0

1

2

3

AR NDE Q GH I LK MFPT YVS W AR NDE Q GH I LK MFPT YVS W

AR NDE Q GH I LK MFPT YVS W

AR NDE Q GH I LK MFPT YVS W AR NDE Q GH I LK MFPT YVS W

AR NDE Q GH I LK MFPT YVS W

AR NDE Q GH I LK MFPT YVS W

AR NDE Q GH I LK MFPT YVS W

AR NDE Q GH I LK MFPT YVS W

b0

0 5 40

c

d

0.65P+5: AKGYYHKYRLSPANA

NBP+2: AKGYYHKYRASPLNA

>15P0: AKGYYHKARLSPLNA

2.7P+1: AKGYYHKYALSPLNA

0.20P+3: AKGYYHKYRLAPLNA

0.63P–4, P–3: AKGAAHKYRLSPLNA

0.24P–2: AKGYYAKYRLSPLNA

3.5P+4: AKGYYHKYRLSALNA

0.19YRL: AKGYYHKYRLSPLNA

TRF2Peptide Protein

Sel

ectiv

ity

2.5

1.5

0.5

0

1

2

3

Sel

ectiv

ity

2.5

1.5

0.5

0

1

2

3

Sel

ectiv

ity

2.5

1.5

0.5

0

1

2

32.5

1.5

0.5

0

1

2

32.5

1.5

0.5

0

1

2

32.5

1.5

0.5

0

1

2

32.5

1.5

0.5

0

1

2

3.53

2.5

1.5

0.5

0

1

2

32.5

1.5

0.5

P–4 P–3 P–2

P–1 P0 P+1

P+2 P+3 P+5

TRF2 concentration (µM)353025201510

TRF2-YRLKd = 190nM

FP

val

ue

160

120

80

40

Figure 1 The TRFH domain of TRF2 recognizes short peptide sequences. (a) A schematic representation of the oriented peptide library design. X indicates

any amino acids except for cysteine. (b) The specificity of TRF2TRFH as revealed by oriented peptide library analyses. A selectivity value of Z1 indicates

preference for a particular amino acid. (c) The consensus TRF2 peptide YRL (AKGYYHKYRLSPLNA) binds TRF2TRFH with high affinity as measured by

fluorescence polarization (FP). Error bars indicate s.e.m. (n ¼ 3). (d) The YXL motif on the consensus peptide is crucial for TRF2TRFH interaction. The

affinities (mM) of alanine-substituted peptides for TRF2TRFH were measured by FP. NB, not bound.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 7 3

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Identification of TIN2, PNUTS and MCPH1 as TRF2TRFH targetsThe identification and analysis of specific TRFH binding partnersshould greatly facilitate the understanding of TRF2 function andtelomere maintenance. To this end, we performed a genome-widesearch of potential TRF2-interacting partners based on our peptidelibrary data of TRF2TRFH (Supplementary Fig. 3 online). From thislist, we selected several human proteins that have been implicated insignal transduction and RNA or DNA regulation for further analysis(Fig. 3a and Supplementary Fig. 4 online). On the basis of theirsequences, we synthesized [Y/F]XL peptides and measured theiraffinities to TRF2TRFH. Three of the peptides (from TIN2, PNUTSand MCPH1, respectively) bound TRF2TRFH with affinities in the lowmicromolar range (Supplementary Fig. 4), and we followed upfurther on these interactions. Notably, TIN2 and PNUTS were alsoidentified in our large-scale immunoprecipitation and MS analysis ofTRF2 (Fig. 3b) and RAP1 protein complexes30,31. As in Apollo, TIN2(known to associate with TRF2) also harbors a TRFH domain bindingmotif (FNL) (Supplementary Fig. 4). Indeed, TRF2TRFH can interactwith the TIN2 FNL peptide with a modest affinity that is dependenton the FXL motif (Supplementary Fig. 5a–c online).

It has been suggested that TIN2 may bind TRF2 through twodistinct regions, the high-affinity TIN2 N-terminal region and thelow-affinity region containing the TRFH binding motif24. The TRFH–TIN2 interaction was not stable enough to survive co-immunopreci-pitation24. To further explore the association of the TIN2 FNL motifwith TRF2 in vivo, we studied the TRF2–TIN2 interaction in live cells

through the bimolecular fluorescence complementation (BiFC)assay32,33. Here we tagged TRF2 and TIN2, respectively, with theN-terminal half of Venus yellow fluorescent protein (YFP) (YFPn-TRF2) and the C-terminal domain of YFP (YC-TIN2). These proteinswere stably expressed in HTC75 cells for fluorescence complementationanalysis. Consistent with the notion of multiple domains mediating theTRF2–TIN2 interaction, the FNL motif mutant TIN2AA showedreduced association with TRF2 (Supplementary Fig. 5d). The TRF2–TIN2 interaction is unlikely to be mediated through TRF1, because theTRF1 binding mutant TIN2AA can still interact with TRF2 and TRF1does not interact directly with TRF2. In addition, interaction of TRF2FA to endogenous TIN2 was decreased (Fig. 4), indicating that theFNL motif contributes to the TRF2–TIN2 interaction in vivo.

Other TRF2-associated proteins identified include PNUTS, a nucleartargeting subunit of the protein phosphatase PP1 (ref. 34), and themicrocephaly syndrome protein MCPH1 (also known as BRIT1), aBRCT domain–containing protein that functions in DNA-damageresponses35–38. Neither protein has been shown previously to interactwith telomere proteins. To confirm the TRF2–PNUTS and TRF2–MCPH1 interactions, we first carried out co-immunoprecipitationexperiments using antibodies against the endogenous TRF2 andRAP1 complexes. Anti-RAP1 immunoprecipitation brought downendogenous TRF2 and endogenous PNUTS and MCPH1 (Fig. 3c,d).In addition, both PNUTS and MCPH1 could be targeted to telomeres(Fig. 3e). Flag-tagged PNUTS co-stained with about 10% of the TRF2foci, whereas Flag-MCHP1 co-localized with endogenous TRF2.

0

5

10

15

20

25

30P < 0.005

a b

c

53BP1 TRF2 Merged

Vec

tor

SF

B-

2×Y

RL

SF

B-

2×Y

RA

TIF

(%

)

2×YRA

2×YRL

Vecto

r

B

*

*

495449244451

TRFH Myb

TRF2

TRF2 F120A (FA)

TRF2 TRFH

TRF2 ∆B∆M

TRF2 ∆B∆M FA

GST WB

Flag WB

d

GSTfusion

SFB-2×YRL Mock SFB-2×YRL Mock

TRF2 FA

TRF2TRF2

FA

TRF2TRF2

FA

TRF2TRF2

FA

TRF2

Streptavidinpull down

Input (4%)

0

2

4

6

8

10g

h

GST WB

e

TRF2 FA

TRF2TRF2

FA

TRF2TRF2

FA

TRF2TRF2

FA

TRF2

Pull downInput (4%)

GST-TRF2 GST-TRF2 GSTGST

Flagtagged

Flag WB

TRF2 TRF2FA

TRF2∆B∆M

TRF2∆B∆M

FA

f

TRF2

TRF2FA

TRF2∆B∆M

TRF2∆B∆M

FA

53BP1 TRF1 Merged

TIF

(%

)

TRF2 WB

Actin WB Actin

Endo. TRF2∆B∆M-YFPnTRF2-YFPn

TRF2TRF2

FA

∆B∆M∆B∆M

FA

Figure 2 The TRF2-[YF]XL interaction is important for telomere maintenance. (a) Telomere dysfunction

induced foci (TIF) analysis of HTC75 cells expressing the consensus YRL peptide. SFB-2�YRL– or

SFB -2�YRA–expressing cells were immunostained with anti-53BP1 (green) and TRF2 antibodies (red).(b) Quantification of TIF-positive cells. Only cells with Z7 53BP1 foci that colocalized with TRF2 foci

were scored. Error bars indicate s.e.m. (n ¼ 3). P-value was determined by the Student’s t-test. (c) A

diagram of different TRF2 constructs used in this study. (d) Wild-type TRF2, but not the TRF2 FA

mutant, interacted with the SFB-2�YRL peptide in cells. GST-tagged TRF2 or TRF2 FA was coexpressed

with SFB-tagged YRL peptide in 293T cells. SFB-2XYRL peptides were pulled down from cell extracts

using streptavidin beads. The amount of co-precipitated GST-tagged protein was detected by anti-GST

western blots. (e) TRF2 FA maintained its ability to dimerize with TRF2. Flag-tagged TRF2 or the

TRF2FA mutant was coexpressed with GST-TRF2 in 293T cells. GST-TRF2 was pulled down from cell

extracts using glutathione beads. The amount of co-precipitated Flag-tagged TRF2 protein was detected

by anti-Flag western blotting (WB). (f) TIF analysis of HTC75 cells expressing wild-type or mutant TRF2 proteins. Cells expressing vector alone, or YFPn-

tagged TRF2, TRF2 FA (F120A), TRF2DBDM or TRF2DBDM-FA were stained with anti-53BP1 (green) and TRF1 (red) antibodies. YFPn, N-terminal fragment

of YFP. (g) Quantification of data from f. Only cells with Z7 53BP1 foci that colocalized with TRF1 foci were scored. Error bars indicate s.e.m. (n ¼ 3).

(h) Western blotting analysis of the expression levels of endogenous TRF2 and different TRF2 constructs used in f and g.

ART IC L E S

37 4 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

YXL-dependent interaction between PNUTS/MCPH1 and TRF2Next, we synthesized YXL motif–containing peptides based onPNUTS and MCPH1 sequences. The two peptides bound TRF2TRFH

(but not TRF1TRFH) in vitro with Kd values of 1.1 mM and 0.42 mM,respectively (Fig. 4a), demonstrating the specificity of the interaction.Moreover, alanine scanning analysis of the PNUTS or MCPH1 peptide

confirmed the importance of the YXLXPmotif in mediating TRF2 binding (Fig. 4a).

To determine whether TRF2–PNUTS or TRF2–MCPH1 associa-tion is dependent on the YXL motif in vivo, we generatedalanine substitution mutants of PNUTS (PNUTS-AA) and MCPH1(MCPH1-AA). Flag-tagged wild-type and mutant PNUTS or MCPH1was then coexpressed with TRF2 in 293T cells for co-precipitationexperiments. GST-TRF2 was able to specifically pull down wild-type

1

230 VVVSDKYNLKPIPLKRQ

1

324 ETFEEKYRLSPTLSSTK

220160120

605040

80

RAP1

100

70

30

b

a c

e

PNUTS

MCPH1

TF2S PP1D ZnF

BRCTBRCT BRCT83 648 835

94090743335714773

TRF2Mre11NBS1PNUTSRad50

Anti-Flag IP

MW(kDa)

Inpu

t (2.

5%)

RAP1 IP

IgG C

tr

PNUTS

TRF2

RAP1

TIN2

dIn

put (

1%)

RAP1 IP

IgG C

tr

MCPH1

TRF2

RAP1

Flag-PNUTS

Flag-MCPH1

YXL motif protein TRF2 Merged

NBNBETFEEKYRASPTLSST

0.42NBETFEEKYRLSPTLSST

4.8NBVSDKYNLKAIPLKRQ

>30NBVSDKANLKPIPLKRQ

NBNBVSDKYNAKPIPLKRQ

1.1NBVSDKYNLKPIPLKRQ

TRF2TRF1Peptide ProteinPNUTS

PNUTSAA

PNUTSPNUTS AA PNUTS

Input (2%) Pull down

Flag WB

GST WB

GST-TRF2

GST

a b cGST TRF2 GST TRF2

Flag WB

GST WB

GST

GST-TRF2

Input (20%) Pull down

MCPH1-SFB

PNUTS WB

RAP1 WB

Flag WB

TIN2 WB

POT1 WB

MCPH1 WB

Flag-TRF2 Flag-TRF2 FA

d e

0

5

10

15

20

25

% Y

FP

Pos

itive

s TRF2PNUTS

TRF2

Yc YFPn

YFP

PNUTS

f

MCPH1 WB

TRF2 WB

PNUTS WB

YRL

Input(10%)

RAP1 IP

-

YRA

1 10 100[Peptide](µM) 1 10 100

IgG IP

TIN2 WB

GSTTRF2

FA

TRF2TRF2

FA

TRF2TRF2

FA

TRF2TRF2

FA

TRF2GST

MCPH1A

A

MCPH1

Vecto

r

MCPH1A

A

MCPH1

Vecto

r

MCPH1A

A

MCPH1

Vecto

r

MCPH1A

A

MCPH1

Vecto

r

IgG C

tr

Flag IP

Inpu

t (10

%)

YC-PNUTSAA

YC-TIN

2

YC-PNUTSYC

IgG C

tr

Flag IP

Inpu

t (8%

)

Figure 4 TRF2 interacts with PNUTS and MCPH1 through the YXL motif. (a) YXL peptides derived from PNUTS or MCPH1 bind the TRFH domain of TRF2

(but not TRF1). NB, not bound. Affinities are measured in micromoles. (b) TRF2 interacts with PNUTS through its YXL motif. GST and GST-tagged TRF2 or

TRF2 FA (F120A) proteins were coexpressed with Flag-tagged PNUTS or PNUTS-AA (Y236A L238A) in 293T cells. GST fusion proteins were pulled down

using glutathione beads. The amounts of co-precipitated Flag-tagged PNUTS proteins were detected by western blotting. (c) TRF2 interacts with MCPH1

through its YXL motif. GST alone and GST-tagged TRF2 were coexpressed with SFB-tagged MCPH1 or MCPH1AA in 293T cells. Co-immunoprecipitation (IP)was performed as described in b. (d) The TRF2 FA mutant showed reduced interaction with endogenous PNUTS, MCPH1 and TIN2. Cells expressing Flag-

tagged TRF2 or TRF2 FA were immunoprecipitated with anti-Flag antibodies and blotted with different antibodies as indicated. (e) TRF2 interacts with

PNUTS in live cells. Venus YFP N-terminal fragment tagged TRF2 (TRF2-YFPn) was coexpressed with YFP C-terminal fragment (YC) alone or YC-tagged

TIN2, PNUTS or PNUTSAA in 293T cells. The percentages of YFP-positive cells were measured by flow cytometry. Error bars indicate s.e.m. (n ¼ 3).

(f) Differential binding of PNUTS and MCPH1 to TRF2. Hela cell nuclear extracts were first incubated with increasing concentrations of the YRL peptides,

and then the TRF2 complex was immunoprecipitated with anti-RAP1 antibodies. IgG was used as a control. The amounts of co-precipitated PNUTS and

MCPH1 proteins were detected by western blotting. Western blotting data were rearranged from the same gel.

Figure 3 TRF2 specifically interacts with YXL-

containing proteins PNUTS and MCPH1.

(a) Domain structures of PNUTS and MCPH1.

TF2S, transcription elongation factor S-II like

domain; PP1D, phosphatase PP1 binding

domain; ZnF, zinc finger; BRCT, BRCA1 C-

terminal domain. (b) Co-immunoprecipitation

(IP) and MS identified PNUTS as a Flag-TRF2–

associating protein in Flag-TRF2–expressing Hela

cells. (c,d) Endogenous TRF2–RAP1 complex

associates with endogenous PNUTS (c) and

MCPH1 (d). The TRF2–RAP1 complex was

immunoprecipitated from Hela nuclear extracts

using anti-RAP1 antibodies, followed by western

blotting analyses with the indicated antibodies.(e) PNUTS and MCPH1 foci colocalized with

endogenous TRF2. Cells expressing Flag-tagged

PNUTS or MCPH1 were co-immunostained with

anti-TRF2 (red) and anti-Flag (green) antibodies.

Arrows indicate colocalized spots.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 7 5

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

PNUTS but not PNUTS-AA (Fig. 4b). Similarly, TRF2 interactionwith MCPH1-AA was considerably impaired compared to its inter-action with wild-type MCPH1 (Fig. 4c), strongly supporting thenotion that both PNUTS and MCPH1 bind TRF2 through theYXL motif. To confirm whether the TRF2–PNUTS interaction ismediated through TRF2TRFH, we coexpressed Flag-tagged PNUTSwith GST-tagged wild-type TRF2 or TRF2 FA. In these cells, TRF2,but not TRF2 FA, could bring down PNUTS, even though the F120Amutation did not affect TRF2 dimerization (Figs. 2e and 4b). More-over, the ability of Flag–TRF2 FA to precipitate endogenous PNUTSand MCPH1 was greatly reduced (Fig. 4d). These observationsindicate that the TRF2–PNUTS and the TRF2–MCPH1 interactionsare indeed mediated through the TRFH domain.

We then went on to investigate the TRF2–PNUTS and TRF2–MCPH1 interactions in live cells through the BiFC assay, usingthe YFPn-TRF2 constrcut described above and PNUTS or MCPH1tagged with with the C-terminal domain of YFP (YC-PNUTS andYC-MCPH1, respectively). Whereas YFP-positive cells were virtuallyabsent in control cells, about 14% of the cells coexpressing YFPn-TRF2and YC-PNUTS were YFP positive (Fig. 4e), indicating an inter-action between TRF2 and PNUTS in vivo. This number is comparableto the percentage of YFP-positive cells in cells coexpressing YFPn-TRF2 and YC-TIN2 (20%), which served as a positive control33.However, coexpression of YFPn-TRF2 and the YC-PNUTS-AAmutant did not result in increased fluorescence complementationover background (Fig. 4e). Similarly, MCPH1, but not MCPH1-AA,complemented YFPn-TRF2 in BiFC assays (Supplementary Fig. 6online). These data provide further support that the YXL motifis crucial for the TRF2–PNUTS and TRF2–MCPH1 interactionsin cells.

The identification of multiple TRF2TRFH

targets raises the possibility that these pro-teins may compete for TRF2 binding in cells.To test this, we investigated whether the YRLpeptide could cause differential displacement

of endogenous PNUTS and MCPH1 from TRF2. Indeed, increasingthe concentration of the YRL peptide reduced the association of bothPNUTS and MCPH1 with TRF2 in the anti-RAP1 immunoprecipi-tates (Fig. 4f). Notably, the PNUTS–TRF2 interaction seemed to bemore sensitive than the MCPH1–TRF2 interaction in this peptide-titration experiment. This difference in sensitivity can be correlatedwith the affinities of their corresponding YXL peptides for the TRFHdomain (that is, the PNUTS peptide binds more weakly than theMCPH1 peptide). These observations open up the possibility that theaffinity of the YXL motif may affect the outcome of competitionbetween different TRFH binding proteins.

PNUTS and MCPH1 regulate telomere length and end protectionWe next studied the telomeric function of PNUTS and MCPH1.Expression of a C-terminal truncation mutant of PNUTS (PNUTSDC,residues 1–337, without its phosphatase-interacting domain) in telo-merase-positive HTC75 cells resulted in modest telomere elongation(Fig. 5a and Supplementary Fig. 7 online) but had little effect on TIFformation (data not shown), indicating a role for PNUTS in telomere-length maintenance but not DNA-damage responses. It should benoted that PNUTSDC may not act as an ideal dominant negativeprotein, so that the telomeric activity of PNUTS may have beenunderestimated in these assays.

Because MCPH1 has been implicated in DNA damage–responsepathways35–38, we hypothesized that the TRF2–MCPH1 interactionmight regulate DNA-damage responses at the telomeres. To test this,we used a mutant form of TPP1 (TPP1DC22), whose expressionresults in elevated TIF formation33. Consistent with our hypothesis,knocking down MCPH1 by two different short hairpin RNA (shRNA)sequences inhibited the TPP1DC22-induced TIF response (Fig. 5b–d

Telosome and telomeraseTelomerase

Telomere length

Peptide motif binding modules

DNA-damage repair

TelomereDNA

4

4.5

5

5.5

6

6.5

7

7 21 35 46 55 68

VectorPNUTSPNUTS∆C

Population doubling

Ave

rage

telo

mer

e le

ngth

(kb

)

YFPn-MCPH1Endo MCPH1 MCPH1 WB

TRF2 WB

RAP1 WB

Actin WB

Flag WB

+ + + + + +Flag-TPP1∆C22: – –

Control sh: + + + +– – – –

YFPn-MCPH1AA: + +– – – – – –YFPn-MCPH1: + +– – – – ––

MCPH1 sh: + + + +– – ––

a b

Moc

kM

ock

MCPH1

MCPH1A

A

TPP1∆C22

0

10

20

30

40

50

60

%T

IF

Control sh MCPH1 shP < 0.001 P = 0.005

c

53BP1

Merged

ControlshRNA

Control shTPP1∆C22

MCPH1 shTPP1∆C22

TRF2

Control shTPP1∆C22

MCPH1 shTPP1∆C22

Control shTPP1∆C22

MCPH1 shTPP1∆C22

MCPH1shRNA

YFPn-MCPH1 YFPn-MCPH1AAd

Peptide motif binding modules

Telomere lengthTelomerase

Telosome and telomerase

TelomereDNA

DNA-damage repaire

Figure 5 MCPH1 and PNUTS regulate DNA-

damage response and telomere length,

respectively, at the telomeres. (a) A comparison of

the average telomere length in cells expressing

vector alone, full-length PNUTS and the PNUTS

C-terminal deletion mutant (PNUTSDC, residues

1–337). (b) Stable shRNA knockdown of

endogenous MCPH1 in the indicated cells. Flag-

TPP1DC22–expressing HTC75 cells that also

stably coexpressed different combinations of

shRNA constructs and RNAi-resistant MCPH1

proteins were generated. Whole-cell extracts were

prepared from these cells for western blotting.

Anti-actin antibodies were used for loading

controls. (c) Wild-type MCPH1 but not MCPH1-AA mutants rescued the effects of MCPH1

knockdown on TIF formation. Error bars indicate

s.e.m. (n ¼ 10). P-value was determined by the

Student’s t-test. (d) Immunostaining pictures of

the data in c. Mock, TPP1DC22-expressing cells

or TPP1DC22- and RNAi-resistant MCPH1

coexpressing cells were infected with retroviruses

expressing either a control or MCPH1 shRNA1.

The cells were fixed and immunostained using

anti-TRF2 and anti-53BP1 antibodies. (e) A

model of TRF2 signaling via different [Y/F]XL

motif–containing proteins.

ART IC L E S

37 6 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

and Supplementary Fig. 8 online), indicating a requirement ofMCPH1 for foci formation in response to DNA damage at thetelomeres. Furthermore, an RNA interference (RNAi)-resistant formof wild-type MCPH1, but not MCPH1-AA, rescued the effect ofMCPH1 knockdown (Fig. 5b–d), suggesting that the TRF2–MCPH1interaction modulates the function of MCPH1 at the telomeres. Thesedata collectively demonstrate that PNUTS and MCPH1 are physiolo-gical targets of TRF2 and are likely to function in distinct pathways.

DISCUSSIONThe ability of TRF1 and TRF2 to bind telomeric sequences andthereby help to organize telomere chromatin structure and recruitother proteins to the telomeres has long been appreciated. Recentstudies have further hypothesized that TRF1 and TRF2 may serve asmolecular platforms for the recruitment and assembly of the telomereinteraction network (‘telomere interactome’)12. However, the mechan-isms by which this ever-expanding list of TRF1- and TRF2-interactingproteins contribute to TRF protein function remain unclear. The datapresented here support the model that TRFH domains representtelomere-specific domains that recognize linear peptide sequencemotifs, in a manner similar to that of many known protein modulessuch as the SH3 and WW domains. These sequences would effectivelyserve as molecular glue, allowing for the telomeric association ofvarious signaling molecules and enabling TRF1 and TRF2 to functionas hubs at the telomeres.

The long, repetitive DNA sequences at the telomere end enable itsassociation with arrays of TRF1 and TRF2 molecules to accommodatetemporal, combinatorial and perhaps developmental regulation ofdiverse signaling cascades (Fig. 5e). For example, our data indicatethat TRF2 arrays can function as a telomeric hub via TRF2TRFH andrecruit at least four [Y/F]XL motif proteins: TIN2, Apollo, MCPH1and PNUTS. The TRF2–TIN2 interaction regulates telosome forma-tion and telomerase recruitment39,40, whereas the TRF2–Apollo andTRF2–MCPH1 interactions regulate DNA damage–repairresponses22,23 (Fig. 5e). In addition, the TRF2–PNUTS associationmodulates telomere length. Because each TRF2 homodimer containstwo [Y/F]XL motif binding sites, it is possible that two different TRFHtargets can be recruited to the same TRF2 homodimer, promotingcommunication between two distinct signaling branches. In this sense,TRF2 arrays serve not merely as a hub but as a structural platform. Itwill be important to identify other proteins that directly interact withTRF2 and to dissect the function of and cross-talk between the TRF2targets. To this end, the specificity of TRFH domains for particularlinear sequences as determined by our peptide library experimentsshould prove highly valuable in predicting possible targets. In thisstudy, we have successfully identified MCPH1 and PNUTS as newtargets of the TRF2 TRFH domain.

It has been proposed that telomeres are protected by a singlelarge protein complex formed by the six core telomere proteins:TRF1, TRF2, RAP1, TIN2, TPP1 and POT1 (ref. 11). The TIN2–TRF1 and TIN2–TRF2 interactions are required to build such acomplex. However, TIN2 contains a [Y/F]XL motif, and the TIN2binding pockets on TRF1 and TRF2 overlap with the bindingpockets of other [Y/F]XL targets of TRF1 and TRF2, such asPINX1, Apollo and PNUTS (data not shown). As a result, wesuggest that the telosome/shelterin complex may be one of severalcomplexes (possibly competing with each other) at the telomeres atany given time. Consistent with this notion, we found that TRF2complexes are heterogeneous30. A large fraction of TRF2 and RAP1was detected in distinct peaks from the telosome. In addition, adistinct TRF2–RAP1 complex has been implicated in telomere

nonhomologous end joining41,42. Furthermore, interactions of theTRFH mutant TRF2 FA with [Y/F]XL proteins MCPH1, PNUTSand TIN2 were compromised, but its interaction with RAP1 wasnot (Fig. 4d). The ability of a TRFH domain to recruit differenttargets indicates a much more ‘proactive’ role for TRF2 in deter-mining the assortment of complexes at the telomeres. Our findingspoint to new avenues into which the function of TRFH-containingproteins can be probed and offer new clues regarding the mechan-isms of telomere dysfunction relevant to cancer and aging.

METHODSProtein expression and purification. We expressed human TRF2TRFH (resi-

dues 42–245) and TRF1TRFH (residues 65–267) as GST fusion proteins in

E. coli BL21(DE3) using the pGEX vector. The GST-fusion proteins were

purified with glutathione agarose beads and eluted with elution buffer (20 mM

glutathione, 20 mM Tris-HCl pH 7.3, 100 mM NaCl, 0.2 mM EDTA and 20%

(v/v) glycerol).

Vectors and antibodies. We cloned cDNAs encoding human wild-type and

mutant TRF2 and TIN2, and mouse wild-type and mutant PNUTS and

MCPH1, into a pBabe-based or pcl-based retroviral vector (Flag or YFP-

fragment tagged) for generating stable cell lines or for expression in 293T cells.

For expression of GST fusion proteins in human cells, we cloned wild-type and

mutant TRF2 into pDEST-27 (Invitrogen). MCPH1 and sequences encoding

tandem repeats of the YRL (2�YRL) or YRA (2�YRA) peptides were cloned

into SFB-tagged pBabe-based vectors43, where SFB stands for S-, Flag- and

streptavidin binding tag43. TRF2 mutants included TRF2 FA (F120A), TRF2

DBDM (residues 45–454)21 and TRF2 DBDM FA. YXL mutants included TIN2

AA (F258A and L260A), PNUTS AA (Y236A and L238A) and MCPH1 AA

(Y330A and L332A). PNUTS DC contains residues 1–337.

The antibodies used were monoclonal and polyclonal anti-Flag (Sigma),

anti-Flag-HRP (Sigma), anti-GST-HRP (Amersham), anti-hTRF2 (CalBio-

chem), polyclonal antibodies from Bethyl laboratories against RAP1, TIN2

(ref. 33), POT1N40 and PNUTS, anti-53BP1 (ref. 40), anti-MCPH1 (ref. 35)

and monoclonal anti-TRF1 (Genetex).

Oriented peptide library screening. We synthesized the oriented peptide

library (KGXXXX[HFYW]X[ILV]XPXN, where X is any amino acid other than

cysteine) as described44. The peptide libraries (0.5–1.0 mg) were incubated with

saturated GST-TRFH beads (150 ml) for 15–30 min at room temperature

(251C), and washed with 1� PBS (10 ml). The bound peptides were then

eluted by acetic acid, dried and resuspended in double-distilled H2O for Edman

peptide sequencing (Tufts University Proteomic Core). We calculated the

selectivity value in Figure 1 by two steps. First, the amount of each amino

acid at a given degenerate position was divided by its amount from the control

‘GST alone’ experiment. Second, the ratio from the first step was normalized

such that the sum of the ratios at a given degenerate position was equal to 19

(the number of total amino acids included at each degenerate position). The

resulting number from step 2 became the selectivity value. If no an amino acid

was selected, the ratio in step 2 would be 1. Therefore, a selectivity value of Z1

indicates preference.

Peptide synthesis, fluorescence polarization and affinity measurements. We

synthesized the peptides by solid-phase synthesis using an automated multiple

peptide synthesizer (INTAVIS Bioanalytical Instruments AG) and standard 9H-

flouren-9-ylmethoxycarbonyl chemistry. The synthesized peptides were incu-

bated overnight with 2 equivalent of fluorescein isothiocyanate (FITC) in

pyridine/dimethylformamide/dichloromethane (50:29:21, v/v). The FITC-

labeled peptides were then cleaved overnight from the resin with trifluoroacetic

acid (TFA)/tri-isopropyl silane/water (95:2.5:2.5, v/v/v). The final peptides were

precipitated with cold diethyl ether, washed twice with cold diethyl ether and

stored at �20 1C.

The purified GST-tagged TRFH domain proteins were serially diluted in

binding buffer (50 mM Tris-HCl, pH 8.0, 50 mM NaCl or 50 mM KCl plus

15mM NaCl, 5% (v/v) glycerol, and 1 mM DTT) and incubated with FITC-

labeled peptides (50 nM) at room temperature for 5–30 min. Fluorescence

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 7 7

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

polarization was subsequently measured in a 384-well plate using a Victor V

plate reader (Perkin Elmer).

Immunoprecipitation, western blotting and immunofluorescence. For large-

scale immunoprecipitations, we prepared nuclear extracts from HelaS cells

stably expressing Flag-tagged human TRF2. We purified the TRF2 complex

using anti-Flag M2 agarose beads (Sigma) and analyzed the sample by MS

sequencing as reported45.

We carried out co-immunoprecipitation studies as described45. Glutathione

agarose beads (Molecular Probes) and streptavidin-agarose beads (Fluka) were

used to pull down GST fusion proteins and SFB-tagged proteins, respectively.

We detected tagged proteins by western blotting using anti-Flag horseradish

peroxidase (HRP) or anti-GST HRP antibodies. We also detected Flag-tagged

proteins and endogenous TRF2 with anti-Flag polyclonal and anti-TRF2

monoclonal antibodies.

We carried out indirect immunofluorescence studies on a Deltavision

deconvolution microscope and a Nikon TE200 microscope45. We performed

TIF assays using anti-53BP1 antibodies together with anti-TRF2 (ref. 18) or

anti-TRF1 antibodies40.

Subcellular fractionation. We performed subcellular fractionation as

described46. Briefly, HTC75 cells were trypsinized, washed with PBS and

resuspended in hypotonic buffer with protease inhibitors. We then lysed the

cells by adding Trition X-100 to a final concentration of 0.1% (v/v) on ice. After

a 5-min incubation, we collected the nuclei by low-speed centrifugation

(1,300g, 4 min). The supernatant was clarified by high-speed centrifugation

(10,000g, 10 min) and collected as the cytoplasmic fraction, S1. Isolated nuclei

were washed once with buffer A, and lysed with buffer A (3mM EDTA, 0.2mM

EGTA, 1mM DTT and protease inhibitors) on ice for 10 min. Soluble nuclear

fractions (S2) were separated from chromatin (P) by centrifugation at 1,700g

for 4 min. The chromatin pellet (P) was washed once with buffer A and

collected under the same centrifugation conditions.

Bimolecular fluorescence complementation. We performed BiFC as

described33. Briefly, The Venus YFP N-terminal domain (residues 1–155) was

fused to TRF2 to construct TRF2-YFPn. The YFP C-terminal domain (Yc,

residues 156–239) was fused to MCPH1, TIN2 or PNUTS. These vectors were

either introduced into HTC75 cells by retroviral infection or co-transfected into

293T cells. We then collected the cells for flow cytometry analysis on a Guava

PCA cytometer.

Short hairpin RNA knockdown and rescue. We used two different shRNA

sequences (shRNA1 and shRNA2) to knockdown MCPH1 in human cells.

shRNA1 (5¢-GGATACAGTGGAAGTGTTAAA-3¢) was cloned into the lenti-

viral vector pGIPZ (Openbiosystems) and shRNA2 (5¢-AGGAAGTTG

GAAGGATCCA-3¢) was cloned into a retroviral vector36. We infected

human cells with shRNA-expressing retroviruses, selected with puromycin,

and used them for different experiments described here. To construct a

MCPH1 retroviral vector that was resistant to shRNA1, were replaced the

corresponding nucleotides sequences on MCPH1 with 5¢-GGATACAGCGG

GAGCGTTAAA-3¢. For rescue experiments, cells that expressed RNAi-

resistant MCPH1 were established first and subsequently infected with

retroviruses expressing MCPH1 shRNA1.

TRF assay. As previously described40, we used retroviruses encoding the pBabe

vector, Flag-PNUTS or Flag–PNUTSDC to establish stable HTC75 cells. The

cells were selected in puromycin and passaged for genomic DNA extraction for

the telomere restriction fragment assay40.

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ACKNOWLEDGMENTSWe thank S.Y. Jung and Q. He for technical help and M. Lei (University ofMichigan) for the GST-TRF2TRFH fusion proteins. We thank J. Penningtonand T. Palzkill for peptide synthesis. Work in the laboratories of Z.S. and D.L. issupported by awards from the US National Institutes of Health, the US Depart-ment of Defense, American Heart Association, the Welch foundation and theAmerican Cancer Society. Z.S. is funded by the Leukemia and Lymphoma Society.

AUTHOR CONTRIBUTIONSH.K., O.-H.L., H.X. and L.-Y.C. designed and performed most of the experiments;D.L. and J.Q. did the telomere length and MS experiment, respectively; A.S. andH.K.C. provided technical support. S.-Y.L. provided the MCHP1 reagents; D.L.and Z.S. wrote the paper.

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Blackburn, E.H. Switching and signaling at the telomere. Cell 106, 661–673 (2001).2. Hackett, J.A. & Greider, C.W. Balancing instability: dual roles for telomerase and

telomere dysfunction in tumorigenesis. Oncogene 21, 619–626 (2002).3. Granger, M.P., Wright, W.E. & Shay, J.W. Telomerase in cancer and aging. Crit. Rev.

Oncol. Hematol. 41, 29–40 (2002).4. Harrington, L. & Robinson, M.O. Telomere dysfunction: multiple paths to the same end.

Oncogene 21, 592–597 (2002).5. Collins, K. & Mitchell, J.R. Telomerase in the human organism. Oncogene 21,

564–579 (2002).6. Maser, R.S. & DePinho, R.A. Connecting chromosomes, crisis, and cancer. Science

297, 565–569 (2002).7. Hahn, W.C. & Weinberg, R.A. Modelling the molecular circuitry of cancer. Nat. Rev.

Cancer 2, 331–341 (2002).8. Blasco, M.A. Mammalian telomeres and telomerase: why they matter for cancer and

aging. Eur. J. Cell Biol. 82, 441–446 (2003).9. Cech, T.R. Beginning to understand the end of the chromosome. Cell 116, 273–279

(2004).10. Henderson, E., Hardin, C.C., Walk, S.K., Tinoco Jr, I. & Blackburn, E.H. Telomeric DNA

oligonucleotides form novel intramolecular structures containing guanine-guaninebase pairs. Cell 51, 899–908 (1987).

11. de Lange, T. Shelterin: the protein complex that shapes and safeguards humantelomeres. Genes Dev. 19, 2100–2110 (2005).

12. Songyang, Z. & Liu, D. Inside the mammalian telomere interactome: regulation andregulatory activities of telomeres. Crit. Rev. Eukaryot. Gene Expr. 16, 103–118(2006).

13. Broccoli, D., Smogorzewska, A., Chong, L. & de Lange, T. Human telomeres containtwo distinct Myb-related proteins, TRF1 and TRF2. Nat. Genet. 17, 231–235 (1997).

14. Bilaud, T. et al. Telomeric localization of TRF2, a novel human telobox protein. Nat.Genet. 17, 236–239 (1997).

15. Verdun, R.E. & Karlseder, J. Replication and protection of telomeres. Nature 447,924–931 (2007).

16. Iwano, T., Tachibana, M., Reth, M. & Shinkai, Y. Importance of TRF1 for functionaltelomere structure. J. Biol. Chem. 279, 1442–1448 (2004).

17. Celli, G.B. & de Lange, T. DNA processing is not required for ATM-mediated telomeredamage response after TRF2 deletion. Nat. Cell Biol. 7, 712–718 (2005).

18. Takai, H., Smogorzewska, A. & de Lange, T. DNA damage foci at dysfunctionaltelomeres. Curr. Biol. 13, 1549–1556 (2003).

19. Denchi, E.L. & de Lange, T. Protection of telomeres through independent control ofATM and ATR by TRF2 and POT1. Nature 448, 1068–1071 (2007).

20. Wang, R.C., Smogorzewska, A. & de Lange, T. Homologous recombination generatesT-loop-sized deletions at human telomeres. Cell 119, 355–368 (2004).

21. van Steensel, B., Smogorzewska, A. & de Lange, T. TRF2 protects human telomeresfrom end-to-end fusions. Cell 92, 401–413 (1998).

22. van Overbeek, M. & de Lange, T. Apollo, an Artemis-related nuclease, interacts withTRF2 and protects human telomeres in S phase. Curr. Biol. 16, 1295–1302 (2006).

23. Lenain, C. et al. The Apollo 5¢ exonuclease functions together with TRF2 to protecttelomeres from DNA repair. Curr. Biol. 16, 1303–1310 (2006).

24. Chen, Y. et al. A shared docking motif in TRF1 and TRF2 used for differentialrecruitment of telomeric proteins. Science 319, 1092–1096 (2008).

25. Songyang, Z. et al. SH2 domains recognize specific phosphopeptide sequences.Cell 72, 767–778 (1993).

26. Songyang, Z. Recognition and regulation of primary-sequence motifs by signalingmodular domains. Prog. Biophys. Mol. Biol. 71, 359–372 (1999).

27. Yaffe, M.B. et al. A motif-based profile scanning approach for genome-wide predictionof signaling pathways. Nat. Biotechnol. 19, 348–353 (2001).

28. d’Adda di Fagagna, F. et al. A DNA damage checkpoint response in telomere-initiatedsenescence. Nature 426, 194–198 (2003).

29. Karlseder, J., Broccoli, D., Dai, Y., Hardy, S. & de Lange, T. p53- and ATM-dependentapoptosis induced by telomeres lacking TRF2. Science 283, 1321–1325(1999).

30. Liu, D., O’Connor, M.S., Qin, J. & Songyang, Z. Telosome, a mammalian telomere-associated complex formed by multiple telomeric proteins. J. Biol. Chem. 279,51338–51342 (2004).

31. O’Connor, M.S., Safari, A., Liu, D., Qin, J. & Songyang, Z. The human Rap1 proteincomplex and modulation of telomere length. J. Biol. Chem. 279, 28585–28591(2004).

32. Hu, C.D. & Kerppola, T.K. Simultaneous visualization of multiple protein interactionsin living cells using multicolor fluorescence complementation analysis. Nat. Biotech-nol. 21, 539–545 (2003).

33. Chen, L.Y., Liu, D. & Songyang, Z. Telomere maintenance through spatial control oftelomeric proteins. Mol. Cell. Biol. 27, 5898–5909 (2007).

ART IC L E S

37 8 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

34. Allen, P.B., Kwon, Y.G., Nairn, A.C. & Greengard, P. Isolation and characterization ofPNUTS, a putative protein phosphatase 1 nuclear targeting subunit. J. Biol. Chem.273, 4089–4095 (1998).

35. Lin, S.Y., Rai, R., Li, K., Xu, Z.X. & Elledge, S.J. BRIT1/MCPH1 is a DNA damageresponsive protein that regulates the Brca1-Chk1 pathway, implicating checkpointdysfunction in microcephaly. Proc. Natl. Acad. Sci. USA 102, 15105–15109 (2005).

36. Rai, R. et al. BRIT1 regulates early DNA damage response, chromosomal integrity, andcancer. Cancer Cell 10, 145–157 (2006).

37. Alderton, G.K. et al. Regulation of mitotic entry by microcephalin and its overlap withATR signalling. Nat. Cell Biol. 8, 725–733 (2006).

38. Wood, J.L., Singh, N., Mer, G. & Chen, J. MCPH1 functions in an H2AX-dependent butMDC1-independent pathway in response to DNA damage. J. Biol. Chem. 282,35416–35423 (2007).

39. O’Connor, M.S., Safari, A., Xin, H., Liu, D. & Songyang, Z. A critical role for TPP1 andTIN2 interaction in high-order telomeric complex assembly. Proc. Natl. Acad. Sci. USA103, 11874–11879 (2006).

40. Xin, H. et al. TPP1 is a homologue of ciliate TEBP-b and interacts with POT1 to recruittelomerase. Nature 445, 559–562 (2007).

41. Bae, N.S. & Baumann, P.A. RAP1/TRF2 complex inhibits nonhomologous end-joiningat human telomeric DNA ends. Mol. Cell 26, 323–334 (2007).

42. Price, C.M. wRAPing up the end to prevent telomere fusions. Mol. Cell 26, 463–464(2007).

43. Kim, H., Chen, J. & Yu, X. Ubiquitin-binding protein RAP80 mediates BRCA1-dependent DNA damage response. Science 316, 1202–1205 (2007).

44. Songyang, Z. & Cantley, L.C. The use of peptide library for the determination of kinasepeptide substrates. Methods Mol. Biol. 87, 87–98 (1998).

45. Liu, D. et al. PTOP interacts with POT1 and regulates its localization to telomeres.Nat. Cell Biol. 6, 673–680 (2004).

46. Mendez, J. & Stillman, B. Chromatin association of human origin recognition complex,Cdc6, and minichromosome maintenance proteins during the cell cycle: assemblyof prereplication complexes in late mitosis. Mol. Cell. Biol. 20, 8602–8612(2000).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 7 9

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Polyglutamine disruption of the huntingtin exon 1N terminus triggers a complex aggregation mechanismAshwani K Thakur1,2,4, Murali Jayaraman1,2,4, Rakesh Mishra1,2, Monika Thakur1,2, Veronique M Chellgren3,In-Ja L Byeon1, Dalaver H Anjum1, Ravindra Kodali1,2, Trevor P Creamer3, James F Conway1,Angela M Gronenborn1 & Ronald Wetzel1,2

Simple polyglutamine (polyQ) peptides aggregate in vitro via a nucleated growth pathway directly yielding amyloid-likeaggregates. We show here that the 17-amino-acid flanking sequence (HTTNT) N-terminal to the polyQ in the toxic huntingtinexon 1 fragment imparts onto this peptide a complex alternative aggregation mechanism. In isolation, the HTTNT peptide is acompact coil that resists aggregation. When polyQ is fused to this sequence, it induces in HTTNT, in a repeat-length dependentfashion, a more extended conformation that greatly enhances its aggregation into globular oligomers with HTTNT cores andexposed polyQ. In a second step, a new, amyloid-like aggregate is formed with a core composed of both HTTNT and polyQ.The results indicate unprecedented complexity in how primary sequence controls aggregation within a substantially disorderedpeptide and have implications for the molecular mechanism of Huntington’s disease.

There are nine known expanded CAG repeat diseases, in whichexpansion of a disease protein’s polyQ sequence beyond a thresholdrepeat length causes progressive neurodegeneration through apredominantly gain-of-function mechanism1. In Huntington’s dis-ease, the repeat-length threshold is about 37 glutamines2. A majorchallenge to understanding disease mechanisms has been to dis-cover physical properties of polyQ proteins that show repeat-lengthdependence in this threshold regime and that therefore might serveas a link in the progression from genetics to disease. PolyQ-containing aggregates are ubiquitously observed in these diseases1,and aggregation rates of polyQ sequences increase as repeat lengthincreases3, mirroring correlations between the repeat lengthand the risk of disease and age of onset1. These observations ledto the hypothesis that repeat length–dependent aggregation ofpolyQ is the triggering event in the mechanism of expanded CAGrepeat diseases.

Not all data support this hypothesis, however. In particular, in celland animal models, disease progression is not always correlated withaggregate burden, as measured by inclusions revealed by light micro-scopy4. There are also inconsistent reports of the nature of polyQaggregates. Thus, whereas simple polyQ peptides follow a nucleatedgrowth polymerization mechanism with direct formation of amyloid-like aggregates3,5–8, aggregation products of the polyQ-containingdisease protein huntingtin (HTT) exon 1 include, in addition toamyloid fibrils9, oligomeric and protofibrillar structures10,11 that manyfeel are more relevant to disease pathology12.

Although the human huntingtin (HTT) gene encodes a protein ofmore than 3,500 amino acids, expression of the first exon of the genein cell and animal models is sufficient to replicate much of thepathology of Huntington’s disease1, and there is growing evidencethat proteolytic release of a fragment containing exon 1 is required fortoxicity13. The amino acid sequence of the translation product ofhuman HTT exon 1, which includes the polyQ sequence, is shown inTable 1. As polyQ repeats are the only apparent common feature ofthe nine expanded polyQ repeat disease proteins1, we have extensivelystudied a series of simple polyQ peptides that contain flanking lysineresidues added for solubility5–8,14. We found that these peptidesaggregate via a nucleated growth polymerization mechanism inwhich the critical nucleus is a rarely populated form of the mono-mer5–8. In these peptides, increases in aggregation rates for peptideswith longer polyQ repeat lengths are associated with more favorableequilibrium constants for nucleus formation5. We also showed pre-viously that the proline-rich flanking sequence (oligoPro) on the C-terminal side of the polyQ in exon 1 reduces aggregation kinetics andaggregate stability but does not fundamentally change the aggregationmechanism14. Its effect is also directional; oligoPro added to the Nterminus of polyQ has no impact on aggregation14.

Intrigued by these oligoPro effects, we turned our attention to the17-amino-acid sequence at the N terminus of the HTT protein, justupstream from the polyQ segment. In this paper, we describe detailedanalysis of the in vitro aggregation mechanism of chemically synthe-sized peptide models for human HTT exon 1 that include this

Received 5 May 2008; accepted 30 January 2009; published online 8 March 2009; doi:10.1038/nsmb.1570

1Department of Structural Biology and 2Pittsburgh Institute for Neurodegenerative Diseases, University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania15260, USA. 3Department of Molecular and Cellular Biochemistry, University of Kentucky, Lexington, Kentucky 40536, USA. 4These authors contributed equally tothis work. Correspondence should be addressed to R.W. (rwetzel@pitt.edu)

38 0 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

ART IC L E S

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

17-amino-acid sequence (HTTNT). We show that addition of HTTNT

to polyQ causes marked changes in the aggregation mechanism,intermediates and products. We also show that this behavior isgrounded in a kind of reciprocal cross-talk between the polyQ andHTTNT segments, both of which are intrinsically unfolded proteinsequences15. This work reveals a complex pathway featuring variousaggregate structures and suggests an unanticipated degree of confor-mational communication between adjacent disordered elements inintrinsically unfolded protein sequences. These data are consis-tent with previous reports showing modulation of polyQ aggregationby flanking sequences in general16–21 and the HTT N terminusin particular22,23.

RESULTSThe role of the HTT N terminus in exon 1 aggregationTo explore possible effects of the HTT exon 1 N-terminal 17 aminoacids on polyQ aggregation, we generated the peptide HTTNTQ35

(Table 1), subjected it to a disaggregation procedure required toensure the absence of pre-existing aggregates24 and studied its aggre-gation in PBS buffer at 37 1C. We found that 3 mM HTTNTQ35

undergoes aggregation significantly (P o 0.001 at t ¼ 10.5 h) morerapidly than 3 mM Q35 (Fig. 1a). In addition, the peptideHTTNTQ36P10 aggregates somewhat less rapidly than HTTNTQ35 but

much more rapidly than Q35 (Fig. 1a); this shows that, although theaggregation-suppressing ability of oligoPro also operates within theexon 1 context, the enhancing effect of HTTNT is dominant over thesuppressing effect of oligoPro. As discussed above, when oligoProwas placed C-terminal to polyQ, the aggregation rate was dimi-nished compared with that of the polyQ sequence alone (compare25 mM Q35P10 with 25 mM Q35 in Figure 1a), but oligoPro placedN-terminal to polyQ had no effect14 (data not shown). In contrast, theHTTNTeffect did not seem to depend on where the HTTNT was placed.Thus, Q35HTTNT shows an aggregation rate that is much faster thanthat of Q35 and not significantly (P 4 0.01 at t ¼ 0.75 h) different tothat of HTTNTQ35 (Fig. 1a).

Despite the apparent dominant effect by the HTTNT sequence,the kinetics of a series of HTTNTQN peptides continued to show astrong polyQ repeat–length dependence, as shown previously forboth simple polyQ peptides25 and for recombinantly producedexon 1 peptides9. Thus, whereas HTTNT, HTTNTQ3 and HTTNTQ15

all aggregated sluggishly, HTTNTQ25 aggregated over a period of1–2 d and, as discussed above, HTTNTQ35 aggregated within a fewhours (Fig. 1b). Although some of the peptides examined inFigure 1b contain a F17W mutation to allow certain fluorescenceexperiments (described below), this mutation did not appreciablyaffect aggregation properties. Control experiments in an

Table 1 Amino acid sequences of exon 1–related peptides

Name or identifier

Sequence

Human HTT exon 1 MATLEKLMKA FESLKSF--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQ----- PPPPPPPPPP-HTT-Ca Q15 KK QQQQQQQQQQ QQQQQ----- ---------- ---------- KK Q20 KK QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- KK Q29 KK QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQ- ---------- KK Q30 KK QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ ---------- KK Q35 KK QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQ----- KK Q35P10 MATLEKLMKA FESLKSF--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQ----- KK HTTNTQ35 MATLEKLMKA FESLKSF--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQ----- KK Q35HTTNT KK QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQ----- MATLEKLMKA FESLKSF HTTNT MATLEKLMKA FESLKSF-amide HTTNTQ3(F17W) MATLEKLMKA FESLKSW--- QQQ FRET-HTTNTQ3 ¥ATLEKLMKA FESLKSW--- QQQ HTTNTQ15(F17W) MATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQ----- ---------- ---------- KK HTTNTQ25(F17W) MATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQQQQQQ QQQQQ----- ---------- KK HTTNTQ30P6 MATLEKLMKA FESLKSF--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ ---------- PPPPPP---- KK HTTNTQ30P6(F17W) MATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ ---------- PPPPPP---- KK HTTNTQ20P10 MATLEKLMKA FESLKSF--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK HTTNTQ20P10(F11A F17A) MATLEKLMKA AESLKSA--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK HTTNTQ20P10(M1X M8X) XATLEKLXKA FESLKSF--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK HTTNTQ20P10(M1X M8X F11A F17A)

XATLEKLXKA AESLKSA--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK

HTTNTQ20P10(F17W) MATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK HTTNTQ20P10(F11W) MATLEKLMKA WESLKSF--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK

20 10FRET-HTTNTQ P ¥ATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK HTTNTQ37P10(F17W) MATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQQQ--- PPPPPPPPPP KK FRET-HTTNTQ37P10 ¥ATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQQQ--- PPPPPPPPPP KK HTTNTK2Q36 MATLEKLMKA FESLKSF-KK QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQQ---- KK Biotinyl-Q29 BKK QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQ- ---------- KK aHTT-C, PQLPQPPPQA QPLLPQPQPP PPPPPPPPGP AVAEEPPLHR P; ¥, nitrotyrosine; X, methionine sulfoxide; B, biotin.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 8 1

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

HTTNTQ20P10 background show that replacement of either Phe11or Phe17 with tryptophan in the HTTNT segment yields aggrega-tion kinetics similar to those of the corresponding wild-typepeptide (Fig. 1c).

Previously, we found that early stages of the aggregation of simplepolyQ peptides show a modest concentration dependence consistentwith a nucleated growth polymerization mechanism and a criticalnucleus of one5, as shown for Q30 in Figure 1d. In contrast, the initialstages of aggregation of the exon 1–related sequence HTTNTQ30P6

(previously, we found no appreciable difference between a Pro6 and aPro10 sequence14) yields a log-log plot of initial kinetics versusconcentration5 (Fig. 1d) with a slope ofapproximately 1, corresponding to a calcu-lated critical nucleus (n*) of about �1. Then* ¼ �1 value indicates a rate/concentrationrelationship for the early aggregation kineticsof HTTNTQ30P6 that is consistent with a non-nucleated, ‘downhill’ aggregation mechan-ism26 for oligomer formation, without akinetic barrier to spontaneous aggregation.This analysis suggests that polyQ peptidescontaining the HTTNT sequence sponta-neously aggregate by an entirely differentmechanism than that of simple polyQ pep-tides, a conclusion that is further supportedby additional data presented below.

Previously, we found that when simplepolyQ monomers undergo spontaneousaggregation in aqueous solution, the earliestobservable aggregates have fibril-associatedproperties similar to those of the final pro-duct5. In contrast, when the HTTNTQ30P6

peptide aggregates, electron micrographs showed that the initialproducts were oligomeric and protofibrillar (Fig. 2). Later in thetime course, various other aggregated structures appeared that weremore fibril-like (Fig. 2). A relationship between the formation of sucholigomeric products and non-nucleated, downhill aggregation kinetics(Fig. 1d) fitting a classical colloidal coagulation model has beensuggested previously for other protein-aggregation reactions27,28.

HttNT conformation and exon 1 aggregation kineticsThere are several possible explanations for the aggregation-enhancingability of the HTTNT sequence. (i) As addition of lysine residues to a

a

h j k l

i

b c f

gd

e

Figure 2 Electron micrographs of various HTTNT-related aggregates. HTTNTQ30P6 was incubated inPBS at 37 1C and sampled at 0 h (a), 15 min (b), 2.5 h (c–e), 5.5 h (f,g), 24 h (h,i), 48 h (j) and

100 h (k). HTTNTQ3 (F17W) was incubated in PBS at 37 1C for 800 h (l). All samples were transferred

directly from reaction mixture to freshly glow-discharged carbon-coated grids and stained with 1% (v/v)

uranyl acetate. Scale bar ¼ 50 nm.

0

0

20

40

60

80

% m

onom

er

HTTNT, 5 µMQ35, 3 µMQ35, 25 µMHTTNTQ35, 3 µMHTTNTK2Q36, 3 µMQ35P10, 25 µMQ35HTTNT, 3 µMHTTNTQ36P10, 3 µM

100

120

20 40 60Time (h)

80 100 120

a

HTTNT

HTTNTQ25HTTNTQ15HTTNTQ3HTTNTQ35

00

20

40

60

80

% m

onom

er

100

10 20 30Time (h)

40 50 60

b

F17WF11WWTF11A F17AM1X M8XM1X M8X F11A F17A

0

0

20

40

60

80

% m

onom

er

100

200 400 600 800Time (h)

c HTTNTQ30P6Q30

–6.5

–18

–16

–14

–12

log

slop

e

–6.0 –5.5 –5.0 –4.5 –4.0 –3.5log [peptide]

d

Figure 1 Aggregation kinetics of huntingtin exon

1 mimic peptides exploring various polyQ repeat

lengths. (a) Basic HTTNT effect. HPLC

sedimentation assay following aggregation of

HTTNT (5 mM, R2 ¼ 0.746, s.d. ¼ ±2.3),

HTTNTQ35 (3 mM, R2 ¼ 0.986, s.d. ¼ ±4.6), Q35

(25 mM, R2 ¼ 0.993, s.d. ¼ ±3.8; 3 mM, R2 ¼0.688, s.d. ¼ ±1.8), Q35HTTNT (3 mM, R2 ¼0.996, s.d. ¼ ±3.0), HTTNTQ36P10 (3 mM, R2 ¼0.992, s.d. ¼ ±4.3), Q35P10 (25 mM, R2 ¼0.973, s.d. ¼ ±0.8), HTTNTK2Q36 (3 mM, R2 ¼0.981, s.d. ¼ ±1.0). (b) Role of polyQ repeat

length on 5 mM peptides. HPLC sedimentation

assay following aggregation of HTTNT (R2 ¼0.746, s.d. ¼ ±2.3), HTTNTQ3 (F17W) (R2 ¼0.966, s.d. ¼ ±2.1), HTTNTQ15 (F17W) (R2 ¼0.971, s.d. ¼ ±3.2), HTTNTQ25 (F17W) (R2 ¼0.992, s.d. ¼ ±3.3), HTTNTQ35 (R2 ¼ 0.993,

s.d. ¼ ±4.3). (c) Role of HTTNT mutations in a

HTTNTQ20P10 peptides (Table 1) incubated at

B6 mM: wild-type HTTNT (R2 ¼ 0.992, s.d. ¼±3.1); F17W (R2 ¼ 0.997, s.d. ¼ ±2.1); F11W

(R2 ¼ 0.991, s.d. ¼ ±3.7); F11A F17A (R2 ¼0.987, s.d. ¼ ±2.5); M1X M8X (R2 ¼ 0.998,

s.d. ¼ ±1.4); M1X M8X F11A F17A (R2 ¼0.628, s.d. ¼ ±3.5). (d) Role of HTTNT sequence

in nucleation of aggregation: concentration

dependence of early aggregation rates for Q30

(slope ¼ 2.57, R2 ¼ 0.9987, s.d. ¼ ±0.026)

and HTTNTQ30P6 (slope ¼ 1.20, R2 ¼ 0.9445,

s.d. ¼ ±0.150). All reactions were conducted in

PBS at 37 1C. X, methionine sulfoxide.

ART IC L E S

38 2 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

sequence sometimes discourages aggregation29, HTTNT might enhanceexon 1 aggregation compared to our simple polyQ peptides because itreplaces the Lys-Lys pair at the N terminus of the model polyQpeptides (see introduction and Table 1). However, the aggregationkinetics of 3 mM HTTNTK2Q35, an exon 1 analog containing a Lys-Lyspair inserted between HTTNT and Q35, were not significantly (P4 0.01at t ¼ 0.3 h) different from the kinetics of 3 mM HTTNTQ35 but at thesame time were significantly (P o 0.001 at t ¼ 10.5 h) faster than thekinetics of 3 mM Q35 (Fig. 1a). In addition, mutations that reduce thehydrophobic character of HTTNT without altering charged residuesabrogated the rapid aggregation kinetics in an exon 1 peptide (Fig. 1c),confirming that the HTTNT segment provides some positive, sequence-specific contribution to aggregation rather than simply replacing aLys-Lys aggregation suppressor. (ii) An obvious alternative possibility isthat HTTNT itself might be highly aggregation prone. Unexpec-tedly, however, we found that a peptide consisting solely of theHTTNT sequence aggregated sluggishly under our standard conditions(Fig. 1a,b). (iii) Another possible explanation is that the combinationof all or part of HTTNT with the first portion of the polyQ sequencemight create a new aggregation sequence motif with much more robust

aggregation kinetics than either polyQ or HTTNT alone. However,HTTNT peptides containing short to intermediate lengths of polyQ alsoaggregated slowly (Fig. 1b), and robust aggregation kinetics ensued forlonger polyQ constructs even when HTTNT was attached to the polyQC terminus (Fig. 1a), which creates an entirely different sequence at theQn-HTTNT junction; these data suggest that polyQ abutted to HTTNT

does not simply create a powerful, novel, linear aggregation motif. TheHTTNTK2Q35 data (Fig. 1a) also argues against a novel aggregationmotif because, if there were such a motif at the junction of the HTTNT

and polyQ sequences, the highly charged Lys-Lys insertion would beexpected to disrupt it. (iv) It is also possible that the HTTNT sequencemight normally exist as a well-behaved oligomeric species, such as adimer or trimer, that can accelerate aggregation by concentrating and/or orientating the polyQ elements. However, size-exclusion chromato-graphy (SEC) mobility showed that the HTTNT peptide migrates as amonomer (Fig. 3), and SEC elution profiles (data not shown) showedno evidence of higher assembly states. Likewise, CD spectra of HTTNT

did not change with respect to concentration (Fig. 4a), consistent witha non-associating system. (v) Another possibility is that, in analogy to arecent report of the ability of expanded polyQ to destabilize a foldedprotein domain20, the HTTNT segment may exist in a folded and/orcompact state that resists aggregation but unfolds or extends whenattached to expanded polyQ, enhancing its aggregation; disruption ofthe folded state is a common trigger for globular protein aggregation30.We explored this last hypothesis in detail, as described below.

Arguing against this postulated polyQ-induced unfolding mechan-ism for the HTTNT effect is the fact that most peptides the size ofHTTNT do not fold into stable, globular structures but, rather, aredisordered. Notably, however, analytical SEC suggests that the HTTNT

sequence is actually relatively compact in solution. A series of simplepolyQ peptides yielded migration rates in SEC that fit a straight line

10

0

–10

–20

[θ] m

illid

egre

es.c

m2 .

dmol

–1

–30

–40190 200 210 220

Wavelength (nm)230 240 250 260

[θ] m

illid

egre

es.c

m2 .

dmol

–1

–60

–40

–20

0

20

40

60

80

190 200 210 220Wavelength (nm)

230 240 250 260

a b Figure 4 Concentration-dependent CD spectra of

HTTNT. (a) HTTNT in aqueous buffer (see Methods)

at 35 1C in concentrations of 3.8 mM (——),

7.5 mM (������) and 18.9 mM (— —). ContinLL58

predicts significant secondary structure: 12%

unordered, 4% b-strand, 20% turn, 8% poly-

proline type II helix and 55% a-helix. (b) HTTNT

in the presence of 10% (v/v) trifluoroethanol (TFE)

at 37 1C in concentrations of 7.5 mM (——),

18.9 mM (������) and 94 mM (— —). In the

presence of this relatively low TFE concentration,

the HTTNT adopts an a-helical structure, as

evidenced by the negative bands at 208 nm

and 222 nm. The development of structure is

protein-concentration dependent, suggesting an

oligomeric state under these conditions.

0.4a

b

0.35

0.3

0.25Kav

0.2

0.15

0.13.1 3.3 3.5

log MW3.7 3.9

BalPro14

HTTNT

Q15

Q20

Q29

Q35

AprIns

20Q3 Q20P10

HTTNT context

Q37P10

***

Q3 / urea

22

24

26

28

FR

ET

dis

tanc

e (Å

)

30

32

34

36

Figure 3 State of expansion of the HTTNT peptide in solution. (a) Fractional

migration (Kav) versus log molecular weight (MW) of various peptides in size-

exclusion chromatography. The straight line is fitted to the Kav values for the

simple polyQ peptides Q15, Q20, Q29 and Q35. a-helix–rich peptide Bal31

and the polyproline type II rich peptide Pro14 are extended. Insulin (Ins),

aprotinin (Apr) and HTTNT are relatively compact. (b) Average HTTNT end-to-

end separation calculated from FRET measurements for mutants FRET-

HTTNTQ3, FRET-HTTNTQ20P10 and FRET-HTTNTQ37P10, compared with their

F17W analogs. Also included is the value for FRET-HTTNTQ3 studied in 6 M

urea in PBS. The dotted line shows the average end-to-end distance

(34.5 ± 4 A) between residues 1 and 17 calculated from polymer theory for

a peptide in statistical coil. Asterisks indicate statistical significance of each

measurement with respect to that for HTTNTQ3 in PBS (*, P o 0.01;

**, P o 0.001).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 8 3

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

(Fig. 3a). As expected, peptides that are predicted to be extended andthus have larger hydrodynamic radii, such as an alanine-rich peptidestrongly favoring an a-helical structure31 and a Pro14 peptide favoringpolyproline type II helix32, migrated faster than polyQ peptides ofequivalent length. Insulin and aprotinin, which show smaller hydro-dynamic radii caused by their compact structures, migrated—asexpected—more slowly than the polyQ peptides of equivalent length.Within this data set, the HTTNT peptide also migrated substantiallymore slowly than expected for a peptide of its molecular weight,suggesting that—like insulin and aprotinin—HTTNT has a relativelysmall hydrodynamic radius.

To confirm the compactness of the HTTNT peptide, we conducted afluorescence-based resonance energy transfer (FRET) experiment inwhich we replaced Met1 of HTTNT with the resonance energy acceptornitrotyrosine, and Phe17 with the resonance energy donor trypto-phan33 in the context of an HTTNTQ3 sequence (‘FRET-HTTNTQ3’).Compared with HTTNTQ3 containing only the F17W replacement, thetryptophan fluorescence of the FRET-HTTNTQ3 peptide decreased byabout 50% (data not shown), corresponding to a calculated averageseparation between donor and acceptor groups of 24 ± 0.5 A (Fig. 3b).This is substantially shorter than the average end-to-end distance of34.5 ± 4 A calculated34 for a 17-residue peptide in an extendedstatistical coil conformation. This theoretical value is supported by aFRET analysis of the HTTNTQ3 peptide in 6 M urea, which yielded aseparation of 33.9 ± 0.5 A (Fig. 3b). Thus, in agreement with the SECdata, HTTNT in the HTTNTQ3 context in native buffer seems to bemarkedly collapsed.

Given this evidence for a collapsed structure in an isolated HTTNT

peptide, we investigated whether it was possible for expanded polyQsequences to disrupt that collapsed state in peptides dissolved in PBS.We found that for the FRET–exon 1 mimic peptide containing a Q20

repeat, the average separation between nitrotyrosine and tryptophanin the HTTNT sequence did not significantly change (P 4 0.01).

However, the corresponding separation for the Q37 repeat peptideexpanded to 32 ± 1 A (Po 0.01), a value approaching the 34-A rangeobtained both by measurement of HTTNTQ3 in urea and by calcula-tion based on an assumed statistical coil configuration (Fig. 3b). Thedata are thus consistent with a mechanism in which the HTTNT

segment of exon 1 is normally in a collapsed state that is resistant toaggregation but that, when connected to an expanded polyQ sequence,becomes extended and labile to aggregation—an effect analogous towhat has previously been observed for a globular protein20.

HTTNT has no stable secondary structureAlthough several features of the above results are consistent withHTTNT (in the absence of a connected expanded polyQ) being acompact domain, the short length of HTTNT makes it unlikely (but notimpossible) that it possesses a unique, folded structure. To probe thesecondary and tertiary structure of HTTNT, we applied two solutionmethods. The first, CD spectral analysis, proved equivocal. The CDspectrum of HTTNT at 35 1C (Fig. 4a) and the difference spectrumresulting from subtraction of a 35 1C spectrum from a 5 1C spectrum(data not shown) lack strong secondary-structure features, suggestingthe absence of a stable structure. At the same time, deconvolutionanalysis of the 35 1C spectrum predicts substantial a-helical structure(Fig. 4, see legend), consistent with projections based on amino acidsequence23,35. This interpretation of the CD spectrum is problematic,however, because the protein data sets used in programs such asContinLL are thought to be unsuitable for deconvoluting CD spectraof short peptides36.

0

0.1

0.2

0.3

0.4

0.5

PO

ND

R s

core

0.6

0.7

0.8

0.9

1.0

Exon 1

0 50 100 150 200 250 300Residue number

350 400 450 500 550 600

Figure 6 PONDR analysis of the first 600 amino acids of the humanhuntingtin sequence. Segments with low PONDR scores are predicted to be

stably folded and high scores (near 1) disordered. Short segments spiking

below a PONDR score of 0.5 are predicted to be MoRFs (see text).

Calculated using the VX-LT version of PONDR. Access to PONDR was

provided by Molecular Kinetics (http://www.pondr.com/).

MdαN(i,i)

dαN(i,i+1)

dNN(i,i+1)

dαN(i,i+2)

+0.20.0

–0.2

4.10

4.15

4.20

4.25

4.30

4.35

1 H (

p.p.

m.)

1 H (

p.p.

m.)

1HN (p.p.m.)

4.40

4.45

4.50

4.55

4.60

8.50 8.45 8.40 8.35 8.30 8.25 8.20

8.50 8.45 8.40 8.35 8.30 8.25 8.20

8.50

8.45

8.40

8.35 Glu5-Leu4

Leu7-Met8Leu7-Lys6

Ala2 Hα-Thr3 HN

Thr3 Hα-Glu5 HN

M8

Ser13

Lys9Leu7

Leu14

Leu4

Thr3

Lys6Glu5 Ala10 Glu12Lys15

Ser16

Phe11

Phe17

8.30

8.25

8.20

∆δ Hα

(p.p.m.)

MA1 5 10 15

A S SF FT L L LE EK K Ka

b

Figure 5 Proton NMR analysis of HTTNT. (a) Summary of NOE and

secondary 1H chemical shift (Dd Ha) data observed for HTTNT at 800 MHz,

5 1C in 20 mM phosphate buffer, pH 7.2. The relative intensities of the

interproton NOEs daN(i,i), daN(i,i+1), dNN(i,i+1) and daN(i,i+2) are depicted by the

thickness of the lines. The Ha secondary chemical shift values of HTTNT

(Dd Ha) were calculated by subtracting random coil values54 from the Ha

chemical shifts of HTTNT. (b) Two-dimensional proton TOCSY and NOESY

NMR spectra. Superposition of the Ha-HN (above) and HN-HN (below)

regions of the TOCSY (black) and NOESY (red) spectra show sequential

daN(i,i+1) and dNN(i,i+1) connectivities. The intraresidue Ha(i)-HN(i) cross-

peaks are labeled with residue name and number, and sequential Ha(i)-

HN(i+1) and HN(i)-HN(i+1) NOE cross-peaks are connected for consecutive

residues. The only observed, very small nonsequential Ha(i)-HN(i+2) NOE

cross-peak connecting the Thr3 Ha and Glu5 HN protons is marked with a

red circle and labeled in red above.

ART IC L E S

38 4 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

The CD dichotomy was clarified by high-resolution NMR analysis(Fig. 5), which strongly suggests the absence of stably folded structure.Two-dimensional proton TOCSY and NOESY NMR analyses showthat HTTNT adopts predominantly unfolded, random-coil conforma-tions characterized by small spectral dispersion, small secondarychemical shifts and strong sequential Ha(i) � HN(i + 1) NOEs withfew sequential HN(i) � HN(i + 1) or medium-range NOEs. A single,weak, medium-range NOE cross-peak connects the Thr3 Ha and Glu5HN protons (Fig. 5b). The region around these residues shows thelargest negative Ha secondary chemical shifts and the most HN(i) �HN(i + 1) NOE cross-peaks (Fig. 5a), indicating transient existence ofa few residues in the a-helix quadrant of F,C space. Despite this slightpropensity, there is clearly no stable a-helix in this peptide in solutionunder physiological conditions.

Thus, in the absence of an expanded polyQ, HTTNT adopts aconformation that lacks appreciable secondary and tertiary structuralfeatures but at the same time exists in a collapsed state. Results fromthe sequence-analysis algorithm PONDR (Molecular Kinetics; http://www.pondr.com/) are consistent with this, predicting the HTTNT

sequence to be a molecular recognition feature (MoRF)37 capable ofengaging in coupled folding and binding interactions15 (Fig. 6). Suchsequences tend to be collapsed coils in native buffer15, often taking upa-helical conformations when complexed with their binding partners.Consistent with this, HTTNT has essentially no a-helical structure inisolation (see above) but takes on a substantial amount of a-helicalstructure in the presence of low concentrations of trifluoroethanol, asrevealed by CD (Fig. 4b).

A multistep aggregation mechanismWhen the assembly of simple polyQ peptides into amyloid-likeaggregates is monitored by multiple analytical techniques, the datafrom all measures track closely, suggesting the absence of reactionintermediates5. In contrast, HTTNT-containing exon 1 peptides showdifferent aggregation curves depending on the analytical method(Figs. 7 and 8). In particular, thioflavin T (ThT) binding (D),commonly used for measuring the presence of amyloid-like struc-ture38, generated reaction curves that were delayed compared to the

346

348

350

Em

issi

on m

ax. (

nm)

352

354

356

0

20

40

60

% r

eact

ion

80

100

0 25 50

Time (h)

M

Reaction mix:

Aggregates:

1 4 8 15 25 40 70

75 100 125

00

10

20

30

40

50

2 4Time (h)

% r

eact

ion

6 8 10

a

b

c

1,550 1,600Amide I frequency (cm–1)

1,650 1,700

6

Gln C=O stretch

310-helix turns, β-sheetα-helixTurns,unordered

M 14 24 40 50 60 80 100 200 300 400

β-aggregates

β-sheetGln NH2 def.

5

4

3

2

1

+ ––

340

345

350

Em

issi

on m

ax. (

nm)

355

360

0 0

10

20

30

40

50

60

70

80

90

100

% a

ggre

grat

ion

100 200 300 400Time (h)

500 600 700 800 9000 100 200 300 400

Time (h)500 600 700 800 900

0.0

0.2

0.4

Elo

ngat

ion

rate

(fm

ol h

–1)

0.6

0.8

1.0

a

b

c d

e

Figure 8 Time course of aggregation of HTTNTQ20P10 by multiple analyses. (a) Trypsin sensitivity

of either monomer (t ¼ 0) or aggregates isolated by centrifugation at either 42 h or 700 h (see

Methods). (b) Properties of isolated aggregates. Fluorescence emission maxima of tryptophan

residues in the mutant peptides F11W (——�——, R2 ¼ 0.994, s.d. ¼ ±0.5) and F17W

(����m����, R2 ¼ 0.916, s.d. ¼ ±1.2); elongation rate constants for biotinyl-Q29 for isolated

aggregates adherent to microtiter plate wells (—&—, R2 ¼ 0.748, s.d. ¼ ±0.19). (c) Overall

aggregation kinetics of wild-type peptide monitored by the HPLC sedimentation assay (----E----, R2 ¼ 0.992, s.d. ¼ ±3.1) and by ThT fluorescence (—D—,

R2 ¼ 0.974, s.d. ¼ ±6.0). (d) Dot blot of non-incubated monomer (M) and isolated aggregates developed with the anti-polyQ MW1 antibody. (e) FTIR

spectra of aggregates. Monomeric Q15 (1); aggregates of HTTNTQ20P10 (F17W) isolated at 45 h (2), 120 h (3), and 120 d (4); aggregates of HTTNTQ36P10

isolated at 7 d (5); aggregates of Q30 isolated at 30 d (6). Amide I frequency values normally assigned to secondary-structural features59 and glutamine side

chains60 are shown above with bars.

Figure 7 Time course of aggregation of HTTNTQ30P6 (F17W) by multiple

analyses. (a) Fluorescence-emission maximum of the tryptophan residue at

position 17 in resuspended aggregates isolated from reaction of HTTNTQ30P6

(F17W) (’) or HTTNTQ3 (F17W) (D). The emission maximum of monomeric

peptide (�) is plotted as being equivalent to that of initial aggregates,

because this is the result obtained for the F17W mutant of the shorter, less

rapidly aggregating HTTNTQ20P10. The HTTNT aggregation reaction was

carried out to 800 h, at which time W17 remained completely solvent

exposed (not shown). (b) Time course monitored by HPLC sedimentation

assay (~), R2 ¼ 0.983, s.e.m. ¼ ±6.0), ThT fluorescence (—n—, R2 ¼0.994, s.d. ¼ ±3.9) and right-angle light scattering (—J—, R2 ¼ 0.983,

s.d. ¼ ±6.0). Inset, first 10 h. The fits of the HPLC sedimentation assay

(~) in b are shown with a beaded solid line. (c) Dot blots of HTTNTQ30P6

(F17W) time points using the antibody MW1. Above, unfractionated aliquots

of the reaction mixture (Numbers above indicate time in hours; M, non-incubated monomer). Below, equivalent masses of isolated aggregates (no

material in the ‘M’ column in this row).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 8 5

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

curves from the HPLC sedimentation (~) and light-scattering (J)assays (Fig. 7b and 8c). Thus, for the HTTNTQ30P6 peptide (Fig. 7) at6.25 h, the difference between the ThT value (D) and the HPLC value(~) is statistically significant (P o 0.01); at later time points (19 h,120 h) the difference is insignificant (P 4 0.01). The observation ofthe ThT lag followed by a burst suggests that the initial aggregates arenot amyloid-like, whereas later aggregates are. FTIR analysis alsosuggests that the initial aggregation product is qualitatively differ-ent—showing more coil and less b-structure—compared to aggregatesisolated later in the time course (Fig. 8e). This is also consistent withthe EM data (Fig. 2).

Details of the structures of these intermediates are revealed byfurther analysis of HTTNTQ30P6 (Fig. 7) and HTTNTQ20P10 (Fig. 8)aggregates isolated from ongoing reactions. In particular, dot blotanalysis of these aggregates showed that polyQ in the initial aggregatesis readily accessible to an antibody against a linear polyQ epitope butis masked in the later aggregates (Figs. 7c and 8d). In addition, theinitially formed aggregates are poor templates for recruiting polyQ-containing peptides, whereas the later, fibrillar aggregates are capableof seeding polyQ elongation (Fig. 8b). Seeding activity is a featureparticularly associated with amyloid-like aggregates39. Furthermore,whereas a tryptophan residue inserted in place of phenylalanine atposition 11 or 17 was solvent exposed both in the monomeric peptideand in the initial aggregates, it became less solvent accessible (based onits fluorescence maximum) as the aggregation reaction proceeded,with a time course that parallels the disappearance of polyQ antibodybinding epitopes (Figs. 7a (’); 8B (�,m)). In contrast, when theisolated HTTNT peptide itself aggregated slowly, the tryptophanresidue of the F17W mutant remained fully solvent exposed in theaggregate fraction, even after 800 h (Fig. 7a (D)); this is consistentwith the apparently inability of aggregates of HTTNT lacking a longpolyQ track to progress beyond the oligomer stage, in agreement withEM analysis (Fig. 2l). Although the initial aggregates of HTTNTQN

peptides seem to be less tightly packed than the later aggregates, theyare sufficiently structured that they are protected from proteolyis bytrypsin, in contrast to soluble, monomeric HTTNT (Fig. 8a).

Thus, the initial aggregates in this HTTNT-mediated mechanism arenonfibrillar oligomers, with their HTTNT segments composing the corebut their polyQ segments remaining unstructured and available forantibody binding. Subsequently, the polyQ elements also becomeintegrated into the aggregate core structure, which takes on a morefibrillar character both in the polyQ and HTTNT elements. Growth intolarger fibrillar assemblies, as seen in EM, take place at later incubationtimes. FTIR analysis (Fig. 8e) showed that the initial aggregates (beforeburial of Trp17) contain a substantial amount of coil and turnconformations in addition to b-structure. Notably, FTIR spectra of

all aggregates isolated after Trp17 burial fea-ture a single b-sheet band at B1,626 cm�1

and are indistinguishable from each other andfrom aggregates of simple polyQ (Fig. 8e).

Notably, at a time when the major changesin aggregate structure were occurring, as evi-denced by ThT binding, tryptophan fluores-cence shift, FTIR, EM and polyQ antibodybinding, the vast bulk of the exon 1 peptide(480%) did not pellet after centrifugation(Figs. 7 and 8) and migrated as a single,monomeric (Fig. 3a) species in SEC (M.J.and R.W., unpublished data). Coincidentwith the timing of the changes in aggregatestructure, there was a marked increase in the

rate of monomer loss, suggesting that a nucleation event had occurred.

DISCUSSIONThe results presented here are consistent with the model shown inFigure 9. In this model, the HTTNT peptide has an intrinsic tendencyto collapse into an aggregation-resistant compact coil state, but anattached, expanded polyQ sequence induces in this segment a moreextended state. When the HTTNT sequence is extended, it becomessusceptible to formation of a metastable, micelle-like aggregate, withthe HTTNT segment making up the loosely packed core and theflanking polyQ sequence excluded from the core. Although this initialaggregation reaction is non-nucleated, amyloid nuclei seem to risestochastically from among these initial, prefibrillar aggregates. Withthe emergence of amyloid fibril structure, the remaining monomericfraction—which represents the vast majority of the exon 1 moleculesat the time when these postulated nucleation events are taking place—fuels the ongoing fibril-elongation reaction. Although many details ofthis proposed mechanism remain to be worked out, it seems thatlonger polyQ sequences favor aggregation in two ways: first, bydisrupting the HTTNT compact coil and thereby facilitating initialHTTNT aggregation; and, second, by improving the efficiencyof the nucleation events proposed to occur within the prefibrillaraggregate population.

In many respects, the mechanism shown in Figure 9 resembles the‘conformational conversion’ model for spontaneous amyloid growthproposed for yeast prion amyloid formation40. In that model, rela-tively loosely structured spherical oligomers form first, then convertinto a more structured oligomer that can grow by recruiting otheroligomers or monomers. In Figure 9, we envision that the nucleationprocess for exon 1 aggregation consists of the stochastic rearrangementof some oligomers or protofibrils into amyloid-like structures capableof rapidly propagating via monomer addition. That is, although someprefibrillar aggregates readily dissociate back to monomers (M.J. andR.W., unpublished data), those aggregates that successfully undergothe nucleation process seem to be required intermediates, under theseexperimental conditions and, therefore, to be on-pathway41 to amy-loid formation. The kinetics of the only known alternative route toamyloid formation—the previously described5 nucleated growthmechanism—are too slow to provide the observed rapid fibril forma-tion (A.K.T. and R.W., unpublished data). Providing further supportfor the scheme shown in Figure 9, and extracting details on themechanism and kinetics of nucleation process and other assemblysteps are the focus of current work.

Amino acid sequence has been viewed as having two major roles ininfluencing protein aggregation and amyloid formation30. First,globular proteins become more amyloidogenic when their folded

Figure 9 Mechanism of HTTNT-mediated exon 1 aggregation. The HTTNT domain (green) unfolds in a

polyQ repeat length–dependent fashion and, once unfolded, self-aggregates without a nucleation barrier

to form oligomers with cores comprising HTTNT and not polyQ (red). The next identified aggregates

involve both HTTNT and polyQ in amyloid-like structure; oligoPro (black) is not incorporated into the

core. This drawing is schematic and is not meant to imply any details of aggregate structure, except

that final aggregates are rich in b-sheet, are fibrillar and involve both HTTNT and polyQ. Although the

initial formation of oligomers shows non-nucleated, downhill kinetics, it is likely that a nucleation event

takes place stochastically within the prefibrillar aggregate population, as shown in brackets, to trigger

rapid amyloid growth.

ART IC L E S

38 6 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

structures are destabilized, for example, by point mutations. Second,amyloidogenic proteins possess specific primary-sequence elementsthat constitute the core regions of the amyloid structure, and muta-tions within these regions can abrogate or enhance amyloid formation.Our studies on HTT exon 1 aggregation, however, reveal other meansby which sequence can influence aggregation rate. Recently, wereported that an oligoPro sequence on the C-terminal side of apolyQ tends to decrease aggregation rate and aggregate stability,apparently by favoring more aggregation-resistant conformations14.Here we show that adjacent elements of intrinsically unfolded poly-peptides can engage in an ‘aggregation two-step’ in which one elementsupplies rapid initial aggregation kinetics (but is insufficient to formamyloid or provide stability) whereas a second, connected element(which in isolation aggregates relatively slowly) provides the means toa highly stable amyloid structure. In a further twist, we show that adegree of compactness in the unstructured HTTNT sequence providesprotection against aggregation and that the previously describeddestabilizing ability of a connected polyQ20,42 disrupts this protectivestructure and opens the door to robust aggregation. Together, theseresults suggest that the rules by which primary sequences inform thedirection and kinetics of protein aggregation may be much morecomplex than previously believed29.

Perhaps 25% of the proteome consists of sequences that do not existas ordered globular or fibrous structures but rather are intrinsicallyunfolded43. The results in this paper, together with our previousstudies on the proline-rich sequence of HTT exon 114, illustrate thedramatic effects on peptide conformation and aggregation impartedthrough subtle sequence effects within a single disordered polypeptide.In particular, the ability of disordered polyQ sequences to influencethe folding stability of adjacent domains20,42 is an unprecedentedphenomenon in protein science whose limits, and still obscuremechanism, are yet to be elucidated. What is particularly unexpectedabout our data on the effect of expanded polyQ on HTTNT isthe implied ability of the compact coil state of HTTNT to resistaggregation; classically, compact but disordered ‘molten globule’ statesof globular proteins have been considered to be aggregation-prone44.PolyQ destabilization of adjacent domains may account for theobservation that longer polyQ tracts are often associated with reduc-tions in protein activity7, a trend suggesting the possibility that someloss of function45 might accompany the toxic gain of function that isgenerally thought to dominate polyQ diseases1.

We believe HTTNT derives its unusual impact on aggregationmechanism and rate because of its resemblance to MoRF sequences37,which exist as condensed coils poised at the cusp of foldedness,designed by nature for coupled folding and binding processes in thecell15. Although this designation for HTTNT is based solely onPONDR analysis (Fig. 6), the biophysical properties of HTTNT areentirely consistent with this notion, and HTT is well known to possessmany interaction partners46.

Our results add to a growing literature on interactions betweenpolyQ and its flanking sequences in aggregation reactions. After initialreports on the ability of flanking sequences to modulate aggregation ofpolyQ disease proteins in cells47, several papers16–19,21 appeareddescribing aggregation by a flanking sequence that facilitates aggrega-tion of the polyQ portion, sometimes as a clearly defined initialstep19,21 and sometimes initiated by the destabilizing influence of anadjacent expanded polyQ20.

Our results put into a molecular biophysics context several recentcell-based studies of the role of the HTTNT sequence in exon 1aggregation and toxicity. Using models in which exon 1 fragmentsare expressed in mammalian or yeast cells, several groups have shown

that the presence or absence of HTTNT, as well as mutations within oradjacent to HTTNT, introduce complex alterations in subcellularlocalization, aggregate formation and/or cytotoxicity or growth retar-dation22,23,35,48. Studies showing that binding of HTTNT by a specificimmunoglobulin fragment inhibits exon 1 aggregation and toxicity inseveral cell models49 are consistent with our data showing that HTTNT

aggregation is the first step in exon 1 aggregation and that thisdepends on destabilization of a compact state in HTTNT. Despite itslack of a-helical structure, the ability of HTTNT to form a-helixstructure in response to its environment (Fig. 4b) is consistent bothwith suggestions that HTTNT may mediate exon 1 targeting tomembrane fractions23,35 and with our hypothesis that HTTNT is aMoRF sequence that mediates coupled folding and binding to proteintargets. The ability of HTTNT to markedly alter the polyQ aggregationmechanism and the ability of HTTNT mutations (Fig. 1c) and bindingfactors49 to abrogate or modulate this effect are likely to contribute tosome observed cellular effects. In fact, our recognition that polyQsequences may aggregate by entirely different mechanisms, dependingon flanking sequences, suggests that at least some of the diversity ofaggregate morphologies observed in exon 1 cell and animal mod-els11,50 may ultimately be explained by the biophysical character of theexon 1 sequence itself.

METHODSMaterials. We obtained all HTT-related peptides, as well the Pro14 and

alanine-rich peptide ‘Bal’31 in nonpurified form from the small-scale custom

peptide synthesis facility of the Keck Biotechnology Center at Yale University

(http://keck.med.yale.edu/ssps/). We purified peptides by reverse-phase HPLC

and confirmed structures by MS on an Agilent 1100 electrospray MSD24.

Purified peptides were routinely freshly disaggregated before use, as described24.

Porcine insulin and aprotinin were from Sigma-Aldrich. Acetonitrile, hexa-

fluoroisopropanol (99.5% (v/v), spectrophotometric grade) and formic acid

were from Acros Organics, trifluoroacetic acid (99.5% (v/v), Sequanal Grade)

was from Pierce and trifluoroethanol was from Sigma-Aldrich.

General methods. The sedimentation assay24, the ThT and 901 light-scattering

assays51 and the nucleation kinetics analysis5,24 have been described. We

isolated aggregates for analysis by centrifuging a reaction aliquot at 20,817g

in an Eppendorf centrifuge at 4 1C for 30 min, washing the pellet two or three

times with PBS, resuspending it in buffer and determining the aggregate

concentration by an HPLC analysis of a dissolved aliquot, as described24.

Electron microscopy. We took aliquots of the ongoing aggregation reaction

mixture at different time points for EM visualization. We placed 3 ml of sample

on a freshly glow-discharged carbon-coated grid, adsorbed for 2 min, and then

washed the grid with deionized water before staining the protein with 2 ml of

1% (w/v) uranyl acetate and blotting. Grids were imaged on a Tecnai T12

microscope (FEI) operating at 120 kV and 30,000� magnification and

equipped with an UltraScan 1000 CCD camera (Gatan) with post-column

magnification of 1.4�.

Circular dichroism. We collected CD spectra using a Jasco J-810 spectrapo-

larimeter with a 1-mm pathlength quartz cuvette. HTTNT samples in 20 mM

Tris-TFA, pH 7.2. The spectra were collected immediately after thawing (from

�80 1C) the disaggregated peptide samples. Scans were made at 20 nm per min

with steps of 0.5 nm and an averaging time of 8 s. The reported spectra are

averages taken over four scans.

Proton nuclear magnetic resonance. The HTTNT sample for NMR experi-

ments contained 40 mM peptide in 20 mM sodium phosphate buffer, pH 7.2,

0.02% (w/v) sodium azide, 6% 2H2O. We carried out NMR experiments on a

Bruker Avance 800 MHz NMR spectrometer, equipped with a 5-mm z-axis

gradient cryoprobe. The water solvent peak was suppressed using the WATER-

GATE W5 pulse sequence52. We acquired two dimensional homonuclear

NOESY and TOCSY53 data at 5 1C, using a 1.5-s recycle delay and mixing

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 8 7

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

times of 200 ms and 60 ms. Complete backbone and side chain proton

assignments of HTTNT were obtained. Ha secondary chemical shifts (Dd Ha)

were calculated by subtracting sequence-corrected random coil values54.

Analytical size-exclusion chromatography. We conducted SEC experiments

with a Superdex peptide HR10/30 (Pharmacia Biotech) column on a Bio-Rad

(Biologic Duo flow) chromatograph using a flow rate of 1 ml min�1, detecting

at 215 nm at ambient temperature (23 1C). Peptides were suspended in PBS

(polyQ peptides and HTTNT after disaggregation) at 20–100 mM and 100 ml

were injected. The Kav values and the plot between Kav versus logMW were

calculated as described in the GE Healthcare handbook Gel Filtration: Principles

and Methods available on the GE Healthcare web site (http://www.gelifesciences.

com/protein-purification).

Fluorescence resonance energy transfer. FRET peptides (Table 1) were

purified and disaggregated. The peptide solution in TFA in water, pH 3.0,

obtained after the ultracentrifugation step of the disaggregation procedure was

immediately adjusted to 5 mM concentration in 10 mM PBS, pH 7.4, and its

fluorescence was determined. Peptide solutions were confirmed to be aggregate-

free at the end of the FRET measurements by EM, HPLC, ThT and 901 light-

scattering measurements. Fluorescence spectra were recorded at room tempera-

ture (23 1C) with excitation and emission slit widths at 5 nm. Raw data obtained

were buffer subtracted and energy transfer efficiencies (E) were calculated from

fluorescence intensities at 353.5 nm from the emission spectra of the F17W and

doubly labeled FRET peptide. The D-A distance (r) for each peptide was

determined from the E value and the Ro value (26 A) for the nitrotyrosine/

tryptophan FRET pair33,55. The measurements were carried out in triplicate and

the mean value (r) is reported with s.d.

We determined peptide concentrations underpinning the FRET calculations

in two different ways, obtaining similar results. In one method, we used amino

acid composition analysis (Keck Biotechnology Center, Yale University) to

calibrate a stock solution of the peptide which was then used as an HPLC

standard for future determinations24. In the second method, we used the UV

absorption peaks of tryptophan at 280 nm and of nitrotyrosine at 381 nm to

normalize concentrations of F17W and doubly labeled FRET peptides using the

extinction coefficients of tryptophan e280 nm ¼ 5,600 M�1 cm�1 and nitrotyr-

osine e381 nm ¼ 2,200 M�1 cm�1, respectively56.

Dot blots. We carried out dot blots as described21. Aggregate samples were

harvested, resuspended and quantified as described in ‘General methods’, and

aliquots containing 400 ng of aggregates were transferred to a nitrocellulose

membrane. In parallel, a portion of the unfractionated aggregation reaction

mixture was transferred to nitrocellulose membrane at various time intervals.

Blots were incubated overnight with TBST buffer (10 mM Tris-HCl, pH 7.5,

150 mM NaCl, 0.1% (v/v) Tween-20, 0.05% (w/v) sodium azide) containing

5% (w/v) BSA, washed three times with TBST and incubated with a 10 nM

solution of purified MW1 antibody57 (a gift from J. Ko and P. Patterson) for

2 h. After washing with TBST to remove unbound material, blots were

incubated 2 h with a 1:15,000 dilution of a peroxidase conjugate of anti-mouse

IgG (whole molecule) (Sigma, A4416) and then washed four times with TBST.

Blots were visualized with enhanced chemiluminescence solution (Pierce

#A4416) following the manufacturer’s instructions.

Tryptophan fluorescence emission of aggregates. Aggregates were isolated at

different times as described in ‘General methods’, then resuspended in 300 ml

150 mM NaCl, 10 mM phosphate, pH 7.4, and analyzed on a Perkin Elmer

luminescence LS50B spectrometer. The samples were excited at 280 nm and

emission was scanned between 290 nm and 550 nm. The wavelength at the

emission intensity maximum was recorded.

Fourier transform infrared spectroscopy. We carried out Fourier transform

infrared spectroscopy of various samples using an MB series spectrophotometer

with PROTA software (ABB Bomem). We harvested protein aggregates by

centrifugation at 20,817g and washed the pellet three times with PBS. Spectra of

resuspended aggregates were recorded at 4-cm�1 resolution (400 scans at room

temperature). Spectra were corrected for the residual buffer absorption by

subtracting the buffer-alone spectrum interactively until a flat baseline was

obtained between 1,700–1,800 cm�1. Second-derivative spectra for the Amide I

region were calculated from the primary spectrum by using PROTA software.

For the Q15 monomer sample, we purified the peptide by reverse-phase

HPLC with an aqueous acetonitrile gradient in 5 mM HCl to avoid exposure of

the peptide to TFA, which gives a large peak in FTIR. We pooled and

lyophilized the peak fractions and dissolved the powder in 50 ml of 1 mM

HCl, and then centrifuged the samples for 1 h at 435,680g. A 25-ml aliquot was

carefully removed and mixed with 25 ml of a 2� PBS buffer and centrifuged for

30 min at 435,680g, and the supernatant was subjected to analysis. This

material seemed to contain about 20% by mass of amyloid-like polyQ

aggregates, based on ThT analysis, presumably because of its high concentration

(2.4 mM) and the abbreviated and modified disaggregation protocol necessi-

tated by the demands of the experiment.

Trypsin sensitivity of aggregates. For the monomer control, HTTNT was

disaggregated and dissolved in TFA in water, pH 3, then added to a solution

of trypsin (SEQUENZ-Trypsin, Worthington Biochemical Corporation) in

100 mM Tris-HCl, pH 7.0, to yield a 1:10 ratio of trypsin to peptide in

50 mM Tris. This solution was incubated at room temperature and monitored

by injection of aliquots onto a RP-HPLC-MS system (Agilent 1100), which

indicated efficient cleavage after the HTTNT lysine residues (data not shown).

Aggregates were harvested and quantified as described in ‘General methods’,

then incubated with trypsin at 1:10 (w/w) (trypsin: peptide) in 50 mM Tris,

pH 7.0, at room temperature. LC-MS of digest centrifugation supernatants

yielded no material. All the material was found in pellet fractions and was

undigested (data not shown).

Microplate elongation assay. The elongation of biotinylated Q29 (B-Q29) on

HTTNTQ20P10 aggregates harvested at different times was done as described6.

Data analysis. For all reaction profiles, data sets were fit in Sigma Plot to either

three-parameter equations (exponential decay, exponential rise to maximum or

sigmoidal) or linear regression. Reported R2 values and s.d. are from the Sigma

Plot fits. Many data sets were obtained in duplicate, and those that were not are

representative of multiple experiments. The tight error bars for the HPLC

sedimentation assay data obtained in duplicate (Fig. 1a, all data sets but for

HTTNT and HTTNTQ36P10), along with the excellent curve fits throughout

Figure 1, give us confidence in the data sets where only single replicates were

taken (Fig. 1b,c). Duplicate data points in Figure 7 were derived from two

experiments run at different times, and some points (0.6 h, 4 h: ThT, LS; 28 h,

69 h: HPLC, ThT, LS) are represented by only one replicate. The larger error

bars for ThT and LS readings at the later time points are typical for late-stage

amyloid formation reactions as fibrils grow larger. P-values were calculated

using the two-tailed Student’s t-test.

ACKNOWLEDGMENTSThe authors acknowledge J. Ko and P. Patterson (California Institute ofTechnology) for the gift of the MW1 antibody, and T. Fullam (Allegheny College)for providing a set of aggregation kinetics data. We also acknowledge thefollowing funding sources that contributed to the work described here: NIH R01AG019322 (R.W.); Huntington’s Disease Society of America postdoctoralfellowship (V.M.C.); NSF MCB-0444049 (T.P.C.); Petroleum Research Fund/American Chemical Society 43138-AC4 (T.P.C.); grant #4100026429 from theCommonwealth of Pennsylvania (A.M.G.).

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Bates, G.P. & Benn, C. The polyglutamine diseases. in Huntington’s Disease (eds.Bates, G.P., Harper, P.S. & Jones, L.) 429–472 (Oxford University Press, Oxford,2002).

2. Andreson, J.M. et al. The relationship between CAG repeat length and age of onsetdiffers for Huntington’s disease patients with juvenile onset or adult onset. Ann. Hum.Genet. 71, 295–301 (2007).

3. Wetzel, R. Misfolding and aggregation in Huntington’s disease and other expandedpolyglutamine repeat diseases. in Protein Misfolding Diseases: Current and EmergingPrinciples and Therapies (eds. Dobson, C.M., Kelly, J.W. & Ramirez-Alvarado, M.)(Wiley, New York, in the press).

4. Arrasate, M., Mitra, S., Schweitzer, E.S., Segal, M.R. & Finkbeiner, S. Inclusion bodyformation reduces levels of mutant huntingtin and the risk of neuronal death. Nature431, 805–810 (2004).

ART IC L E S

38 8 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

5. Chen, S., Ferrone, F. & Wetzel, R. Huntington’s disease age-of-onset linkedto polyglutamine aggregation nucleation. Proc. Natl. Acad. Sci. USA 99,11884–11889 (2002).

6. Bhattacharyya, A.M., Thakur, A.K. & Wetzel, R. Polyglutamine aggregation nucleation:thermodynamics of a highly unfavorable protein folding reaction. Proc. Natl. Acad. Sci.USA 102, 15400–15405 (2005).

7. Wetzel, R. Chemical and physical properties of polyglutamine repeat sequences. inGenetic Instabilities and Neurological Diseases (eds. Wells, R.D. & Ashizawa, T.)517–534 (Elsevier, San Diego, 2006).

8. Slepko, N. et al. Normal-repeat-length polyglutamine peptides accelerate aggregationnucleation and cytotoxicity of expanded polyglutamine proteins. Proc. Natl. Acad. Sci.USA 103, 14367–14372 (2006).

9. Scherzinger, E. et al. Huntingtin-encoded polyglutamine expansions form amyloid-likeprotein aggregates in vitro and in vivo. Cell 90, 549–558 (1997).

10. Poirier, M.A. et al. Huntingtin spheroids and protofibrils as precursors in polyglutaminefibrilization. J. Biol. Chem. 277, 41032–41037 (2002).

11. Wacker, J.L., Zareie, M.H., Fong, H., Sarikaya, M. & Muchowski, P.J. Hsp70 andHsp40 attenuate formation of spherical and annular polyglutamine oligomers bypartitioning monomer. Nat. Struct. Mol. Biol. 11, 1215–1222 (2004).

12. Caughey, B. & Lansbury, P.T. Protofibrils, pores, fibrils, and neurodegeneration:separating the responsible protein aggregates from the innocent bystanders. Annu.Rev. Neurosci. 26, 267–298 (2003).

13. Graham, R.K. et al. Cleavage at the caspase-6 site is required for neuronal dysfunctionand degeneration due to mutant huntingtin. Cell 125, 1179–1191 (2006).

14. Bhattacharyya, A. et al. Oligoproline effects on polyglutamine conformation andaggregation. J. Mol. Biol. 355, 524–535 (2006).

15. Dyson, H.J. & Wright, P.E. Intrinsically unstructured proteins and their functions. Nat.Rev. Mol. Cell Biol. 6, 197–208 (2005).

16. Masino, L. et al. Characterization of the structure and the amyloidogenic properties ofthe Josephin domain of the polyglutamine-containing protein ataxin-3. J. Mol. Biol.344, 1021–1035 (2004).

17. de Chiara, C., Menon, R.P., Dal Piaz, F., Calder, L. & Pastore, A. Polyglutamine is notall: the functional role of the AXH domain in the ataxin-1 protein. J. Mol. Biol. 354,883–893 (2005).

18. Bulone, D., Masino, L., Thomas, D.J., San Biagio, P.L. & Pastore, A. The innterplaybetween PolyQ and protein context delays aggregation by forming a reservoir ofprotofibrils. PLoS ONE 1, e111 (2006).

19. Ellisdon, A.M., Thomas, B. & Bottomley, S.P. The two-stage pathway of ataxin-3fibrillogenesis involves a polyglutamine-independent step. J. Biol. Chem. 281,16888–16896 (2006).

20. Ignatova, Z. & Gierasch, L.M. Extended polyglutamine tracts cause aggregation andstructural perturbation of an adjacent b-barrel protein. J. Biol. Chem. 281,12959–12967 (2006).

21. Ignatova, Z., Thakur, A.K., Wetzel, R. & Gierasch, L.M. In-cell aggregation of apolyglutamine-containing chimera is a multistep process initiated by the flankingsequence. J. Biol. Chem. 282, 36736–36743 (2007).

22. Duennwald, M.L., Jagadish, S., Muchowski, P.J. & Lindquist, S. Flanking sequencesprofoundly alter polyglutamine toxicity in yeast. Proc. Natl. Acad. Sci. USA 103,11045–11050 (2006).

23. Rockabrand, E. et al. The first 17 amino acids of Huntingtin modulate its sub-cellularlocalization, aggregation and effects on calcium homeostasis. Hum. Mol. Genet. 16,61–77 (2007).

24. O’Nuallain, B. et al. Kinetics and thermodynamics of amyloid assembly using a high-performance liquid chromatography-based sedimentation assay. Methods Enzymol.413, 34–74 (2006).

25. Chen, S., Berthelier, V., Yang, W. & Wetzel, R. Polyglutamine aggregation behavior invitro supports a recruitment mechanism of cytotoxicity. J. Mol. Biol. 311, 173–182(2001).

26. Ferrone, F. Analysis of protein aggregation kinetics. Methods Enzymol. 309, 256–274(1999).

27. Modler, A.J. et al. Polymerization of proteins into amyloid protofibrils shares commoncritical oligomeric states but differs in the mechanisms of their formation. Amyloid 11,215–231 (2004).

28. Bieschke, J. et al. Small molecule oxidation products trigger disease-associatedprotein misfolding. Acc. Chem. Res. 39, 611–619 (2006).

29. Rousseau, F., Schymkowitz, J. & Serrano, L. Protein aggregation and amyloidosis:confusion of the kinds? Curr. Opin. Struct. Biol. 16, 118–126 (2006).

30. Wetzel, R. Mutations and off-pathway aggregation. Trends Biotechnol. 12, 193–198(1994).

31. Marqusee, S., Robbins, V.H. & Baldwin, R.L. Unusually stable helix formation in shortalanine-based peptides. Proc. Natl. Acad. Sci. USA 86, 5286–5290 (1989).

32. Schuler, B., Lipman, E.A. & Eaton, W.A. Probing the free-energy surface for proteinfolding with single-molecule fluorescence spectroscopy. Nature 419, 743–747(2002).

33. Wu, P. & Brand, L. Resonance energy transfer: methods and applications. Anal.Biochem. 218, 1–13 (1994).

34. Fitzkee, N.C. & Rose, G.D. Reassessing random-coil statistics in unfolded proteins.Proc. Natl. Acad. Sci. USA 101, 12497–12502 (2004).

35. Atwal, R.S. et al. Huntingtin has a membrane association signal that can modulatehuntingtin aggregation, nuclear entry and toxicity. Hum. Mol. Genet. 16, 2600–2615(2007).

36. Whitmore, L. & Wallace, B.A. DICHROWEB, an online server for protein secondarystructure analyses from circular dichroism spectroscopic data. Nucleic Acids Res. 32,W668–W673 (2004).

37. Mohan, A. et al. Analysis of molecular recognition features (MoRFs). J. Mol. Biol. 362,1043–1059 (2006).

38. LeVine, H. Quantification of b-sheet amyloid fibril structures with thioflavin T. MethodsEnzymol. 309, 274–284 (1999).

39. O’Nuallain, B., Williams, A.D., Westermark, P. & Wetzel, R. Seeding specificity inamyloid growth induced by heterologous fibrils. J. Biol. Chem. 279, 17490–17499(2004).

40. Serio, T.R. et al. Nucleated conformational conversion and the replication of con-formational information by a prion determinant. Science 289, 1317–1321 (2000).

41. Kodali, R. & Wetzel, R. Polymorphism in the intermediates and products of amyloidassembly. Curr. Opin. Struct. Biol. 17, 48–57 (2007).

42. Bevivino, A.E. & Loll, P.J. An expanded glutamine repeat destabilizes native ataxin-3structure and mediates formation of parallel b-fibrils. Proc. Natl. Acad. Sci. USA 98,11955–11960 (2001).

43. Bracken, C., Iakoucheva, L.M., Romero, P.R. & Dunker, A.K. Combining prediction,computation and experiment for the characterization of protein disorder. Curr. Opin.Struct. Biol. 14, 570–576 (2004).

44. Hammarstrom, P. et al. Structural mapping of an aggregation nucleation site in amolten globule intermediate. J. Biol. Chem. 274, 32897–32903 (1999).

45. Cattaneo, E. et al. Loss of normal huntingtin function: new developments in Hunting-ton’s disease research. Trends Neurosci. 24, 182–188 (2001).

46. Kaltenbach, L.S. et al. Huntingtin interacting proteins are genetic modifiers ofneurodegeneration. PLoS Genet. 3, e82 (2007).

47. Nozaki, K., Onodera, O., Takano, H. & Tsuji, S. Amino acid sequences flankingpolyglutamine stretches influence their potential for aggregate formation. Neuroreport12, 3357–3364 (2001).

48. Steffan, J.S. et al. SUMO modification of Huntingtin and Huntington’s diseasepathology. Science 304, 100–104 (2004).

49. Colby, D.W. et al. Potent inhibition of huntingtin aggregation and cytotoxicity by adisulfide bond-free single-domain intracellular antibody. Proc. Natl. Acad. Sci. USA101, 17616–17621 (2004).

50. Wanderer, J. & Morton, A.J. Differential morphology and composition of inclusions inthe R6/2 mouse and PC12 cell models of Huntington’s disease. Histochem. Cell Biol.127, 473–484 (2007).

51. Chen, S., Berthelier, V., Hamilton, J.B., O’Nuallain, B. & Wetzel, R. Amyloid-likefeatures of polyglutamine aggregates and their assembly kinetics. Biochemistry 41,7391–7399 (2002).

52. Liu, M. et al. Improved WATERGATE pulse sequences for solvent suppression in NMRspectroscopy. J. Magn. Reson. 132, 125–129 (1998).

53. Bax, A. & Davis, D.G. MLEV-17-based two-dimensional homonuclear magnetizationtransfer spectroscopy. J. Magn. Reson. 65, 355–360 (1985).

54. Schwarzinger, S. et al. Sequence-dependent correction of random coil NMR chemicalshifts. J. Am. Chem. Soc. 123, 2970–2978 (2001).

55. Lakowicz, J.R. Principles of Fluoresence Spectroscopy 954 (Kluwer, New York, 2006).56. Tcherkasskaya, O. & Ptitsyn, O.B. Direct energy transfer to study the 3D structure of

non-native proteins: AGH complex in molten globule state of apomyoglobin. ProteinEng. 12, 485–490 (1999).

57. Ko, J., Ou, S. & Patterson, P.H. New anti-huntingtin monoclonal antibodies: implica-tions for huntingtin conformation and its binding proteins. Brain Res. Bull. 56,319–329 (2001).

58. Sreerama, N. & Woody, R.W. Estimation of protein secondary structure from circulardichroism spectra: comparison of CONTIN, SELCON, and CDSSTR methods with anexpanded reference set. Anal. Biochem. 287, 252–260 (2000).

59. Jackson, M. & Mantsch, H.H. The use and misuse of FTIR spectroscopy in the deter-mination of protein structure. Crit. Rev. Biochem. Mol. Biol. 30, 95–120 (1995).

60. Venyaminov, S. & Kalnin, N.N. Quantitative IR spectrophotometry of peptide com-pounds in water (H2O) solutions. I. Spectral parameters of amino acid residueabsorption bands. Biopolymers 30, 1243–1257 (1990).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 8 9

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Bacterial frataxin CyaY is the gatekeeper of iron-sulfurcluster formation catalyzed by IscSSalvatore Adinolfi1, Clara Iannuzzi1,2, Filippo Prischi1,3, Chiara Pastore1,5, Stefania Iametti4,Stephen R Martin1, Franco Bonomi4 & Annalisa Pastore1,6

Frataxin is an essential mitochondrial protein whose reduced expression causes Friedreich’s ataxia (FRDA), a lethalneurodegenerative disease. It is believed that frataxin is an iron chaperone that participates in iron metabolism. We have testedthis hypothesis using the bacterial frataxin ortholog, CyaY, and different biochemical and biophysical techniques. We observe thatCyaY participates in iron-sulfur (Fe-S) cluster assembly as an iron-dependent inhibitor of cluster formation, through binding to thedesulfurase IscS. The interaction with IscS involves the iron binding surface of CyaY, which is conserved throughout the frataxinfamily. We propose that frataxins are iron sensors that act as regulators of Fe-S cluster formation to fine-tune the quantity of Fe-Scluster formed to the concentration of the available acceptors. Our observations provide new perspectives for understandingFRDA and a mechanistic model that rationalizes the available knowledge on frataxin.

Friedreich’s ataxia (FRDA) is a relentlessly progressive neurodegen-erative disease which leads to the death of the affected individuals.Although classified as rare, this recessive pathology has B1 carrier inevery 120 individuals1. Discovery of the responsible gene in 1996established that FRDA is caused by deficiency of frataxin, a smallessential protein highly conserved from bacteria to humans2. Ineukaryotes, frataxin is nuclearly encoded, translated in the cytoplasmand then imported into mitochondria, where it is finally matured3.

The cellular function of frataxin remains controversial. It is com-monly accepted that frataxin is involved in iron metabolism: partialdepletion of frataxin has been shown to increase mitochondrial ironlevels and to decrease the activity of Fe-S cluster proteins (reviewed inref. 1). In vitro, frataxins from different species bind both Fe2+ and Fe3+

with a defined stoichiometry, although with relatively low affinity andspecificity4–7. Bioinformatics, genetic and biochemical evidence hasshown that frataxin binds to ferrochelatase5,8–10, and to essentialcomponents of the Fe-S cluster machinery11–13, thus implicating theprotein in heme metabolism and Fe-S cluster formation. Two hypoth-eses have been suggested to explain the exact role of frataxin in theseprocesses. According to one, frataxin and its orthologs are ironchaperones whose role is to provide the iron necessary for Fe-S andheme assembly14. The second, based on the capacity of the protein toform iron-loaded multimers in vitro, proposes an iron-storage functionto scavenge the toxic iron in a sheltered, but readily available, form15.

We have challenged these hypotheses and studied the effects offrataxin on the efficiency of enzymatic Fe-S cluster reconstitution,using complementary biochemical and biophysical techniques. We

used as a model the well-defined and characterized bacterial system inwhich the genes encoding the protein components of the Fe-S clustermachinery (but not frataxin) are grouped in the isc operon. Addi-tionally, bacterial frataxin orthologs (the CyaY proteins) consist ofonly a conserved globular domain, thus avoiding possible complica-tions caused by the presence of the mitochondrial import signal.

Fe-S cluster formation is a complex enzymatic reaction, still notentirely understood, that requires several steps: conversion of cysteineinto alanine with formation of a persulfide, its transfer to an acceptorand the formation of the cluster. We followed the effect of CyaY on thekinetics of the reaction in vitro using purified proteins. In this assay,the cluster is formed under strict anaerobic conditions on thescaffold protein IscU, either chemically, using sulfide and ferricammonium citrate as sources of sulfur and iron, respectively, orenzymatically13,16,17. In the latter case, the desulfurase IscStransfers sulfur from cysteine onto IscU, with the concomitant uptakeof iron to form a Fe-S cluster. IscU further transfers the Fe-S cluster toits final acceptors.

Using this assay, we show that Escherichia coli frataxin is not merelyan iron chaperone with a neutral involvement in the enzymaticprocess but an integral part of the cluster-assembly machinery.Bacterial frataxin works as a molecular regulator able to inhibit,depending on the extent to which it is iron saturated, formation of2Fe-2S clusters. This function does not involve large aggregates offrataxin, which could store iron in a bioavailable form. Our workprovides a new perspective on the cellular role of frataxin and suggestsa molecular mechanism to explain FRDA.

Received 24 October 2008; accepted 13 February 2009; published online 22 March 2009; doi:10.1038/nsmb.1579

1National Institute for Medical Research, The Ridgeway, London, UK. 2Dipartimento di Biochimica e Biofisica, II Universita’ degli Studi di Napoli, Napoli, Italy.3Dipartimento di Biologia Molecolare, University of Siena, Siena, Italy. 4DISMA, University of Milan, Milan, Italy. 5Present address: Department of Biological Sciences,Columbia University, New York, New York, USA. 6Temporary address: Unite de Virologie et Immunologies Moeculaire, Intitut National de la Recherche Agronomique,Jouy-en-Josas, France. Correspondence should be address to A.P. (apastor@nimr.mrc.ac.uk).

39 0 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

ART IC L E S

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

RESULTSCyaY is an inhibitor of enzymatic Fe-S cluster formationWe implemented an in vitro reconstitution assay by adapting pub-lished protocols18–21. Specific care was paid to choosing the appro-priate concentrations of this complex, multicomponent system and toascertain the state of folding of each component (see the Resultssection of Supplementary Methods online).

Fe-S reconstitution on IscU was first monitored by absorbancespectroscopy: 2Fe-2S clusters are characterized by maxima at 400 nmand 456 nm and a shoulder at 510 nm, whereas 4Fe-4S clusters have abroad absorption centered at B390 nm18 (Fig. 1a). Formation of Fe-Sclusters was detected as an increase of absorbance in the range of400–550 nm and by the appearance of a brown-red color, as previouslydescribed. As expected, the IscS-mediated enzymatic reaction was bothfaster and more efficient than the non-enzymatic reaction (Fig. 1b).

When CyaY was added to the enzymatic reaction, we observed twomain effects: the absorption spectrum at the end of the reactionchanged shape, with a marked attenuation of the band at 456 nm andalmost complete disappearance of the shoulder at 510 nm (Fig. 1a).The absorbance ratio between the bands at 400 nm and456 nm increased progressively as the reaction proceeded. In thepresence of CyaY, the kinetics of the process were also distinctly slower(Fig. 1b and Supplementary Fig. 1 online). The brown-red color wasobserved also in this case, although it was less intense than in theabsence of CyaY.

To check whether the presence of CyaY could facilitate iron delivery,as would be expected for an iron chaperone, we compared the kineticswhen iron was added to the reaction solution to those obtained whenCyaY was preloaded with iron. We incubated a CyaY solution with aniron excess, either anaerobically (for Fe2+) or aerobically (for Fe3+),and then added the CyaY solution anaerobically to the enzymaticmixture. Apart from a small lag phase that may reflect iron availability,the kinetics with iron-preloaded CyaY were practically indistinguish-able from those obtained by adding iron independently (Fig. 1b). Thissuggests that, if the effect of CyaY is linked to its iron bindingproperties, it does not depend on whether the protein takes ironfrom the solution or is the carrier, as expected from the weak affinitiesof CyaY both for Fe2+ and Fe3+ (ref. 22).

To check whether the observed effect was due to the presence of anyagent able to bind iron and interfere with the enzymatic activity in anonspecific way, we repeated the experiments replacing CyaY with anexcess of either citrate or calmodulin. The first is a well-known ironchelator; the second has been seen to form iron-loaded aggregatesin the presence of excess iron (S.A. and A.P., unpublished data). Inneither case did we observe interfering effects on the kinetics of theenzymatic reaction (data not shown).

Taken together, this evidence suggests that the role of CyaY is notsimply that of an iron chaperone but that of an inhibitor.

CyaY specifically inhibits formation of 2Fe-2S clustersWe also used CD spectroscopy to follow the effect of CyaY on Fe-Scluster formation. This technique is complementary to absorbancespectroscopy as it avoids complications due to the overlappingabsorption spectra of other iron-bound components. It is alsosensitive to the cluster nuclearity: 2Fe-2S and 4Fe-4S clusters canclearly be distinguished by CD. 2Fe-2S gives intense contributionsin the range 300–700 nm. 4Fe-4S clusters, with the exception of2[4Fe-4S] bacterial ferredoxins23, have weak bands that can barely bedistinguished from the baseline at the concentrations used here.

Enzymatic cluster formation on IscU showed clear evidence forformation of a 2Fe-2S cluster with maxima at 340 nm, 430 nm and510 nm, and minima at 370 nm and 560 nm, in agreement with theliterature20,21,24 (Fig. 1c). In the presence of CyaY, the rate of 2Fe-2Scluster formation was much slower and the reaction efficiency greatlyreduced, as shown by the strong drop (480%) in the plateau intensity(Fig. 1d). Minor differences in the kinetics of the different experi-ments were due to several factors. Different enzyme preparations hadslightly different activities. CD measurements were overall slightlyfaster because, by this technique, we followed only one pathway andused larger IscU–IscS ratios. This was possible because, for technicalreasons, the reaction could be initiated directly in the spectrophot-ometer, thus the early events of the reaction were not lost. This couldnot be done for the absorbance measurements, for which we had totransfer the cuvette from the chamber to the spectrophotometer. Theinhibitory effect of CyaY seemed to be stronger when monitored byCD than by absorbance spectroscopy. This is probably because CDmonitors only formation of 2Fe-2S clusters on IscU, whereas absor-bance also detects other iron-containing species.

0.35a

b

c

d

0.30

0.25

10

8

6

4

2

0300 400 500 600 700

–2

–4

0.15

Abs

orba

nce

mde

g

Abs

orba

nce

mde

g

0.10

0.05

0.00

0.09

0.08

7

6

CyaY

No CyaY

CyaYNo CyaY

CyaYNo CyaY

CyaY added at 180 s

5

4

3

2

1

0

010

020

030

0

Time (s)

400

500

600

–1

Enzymatic reactions

Fe2+-preloaded CyaY

Chemical reaction

No CyaY

CyaY0.07

0.06

0.05

0.04

0.03

0.02

0.010

600

1,20

01,

800

Time (s)2,

400

3,00

03,

600

4,20

04,

800

0.20

300 350 400 450Wavelength (nm)

Wavelength (nm)500 550 600

Figure 1 Fe-S cluster reconstitution on IscU followed by absorbance and

CD measurements. (a) Comparison of the final absorbance spectra in the

absence and presence of CyaY. (b) Comparison between the kinetics of

cluster transfer on IcsU followed by absorbance at 456 nm as achieved

chemically or enzymatically in the absence and in the presence

of CyaY. The experiments were carried out using 50 mM IscU, 25 mM

Fe(NH4)2(SO4)2, 250 mM cysteine, 3 mM DTT and either 250 mM lithium

sulfide (for the non-enzymatic reaction) or 1 mM IscS (for the enzymatic

reaction). When present, the CyaY concentration was 5 mM. Iron was

dded both in the mixture or preloaded on CyaY as 2-molar excess.

(c) Comparison of the CD spectra (as expressed in mdeg) recorded in the

region 300–700 nm in the absence and in the presence of CyaY. The

spectra were recorded at plateau. (d) Comparison of the kinetics of

enzymatic cluster formation as followed by recording the CD spectra as a

function of time in the absence of CyaY, in the presence of CyaY and byadding iron-free CyaY 180 s after starting the reaction. The two experiments

were carried out using 50 mM IscU, 40 mM ferric ammonium citrate,

250 mM cysteine, 3 mM DTT and 2 mM IscS (for the enzymatic reaction).

When present, the CyaY concentration was 5 mM. In the experiment with

CyaY, additional 10 mM cysteine solution was added once the reaction

had reached a plateau. The alanine concentration was dosed by amino

acid analysis.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 9 1

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

To confirm that inhibition is due to CyaY and is independentof the order in which the components are added, we recordedkinetics in which CyaY was injected into the mixture only afterstarting the reaction (Fig. 1d). Again, we observed no appreciabledifferences in the presence of CyaY preloaded with iron or whenFe2+ was used instead of Fe3+ (data not shown).

These results indicate that CyaY has a strong and specific inhibitoryeffect on the formation of 2Fe-2S clusters.

Dissecting Fe-S cluster assembly into its componentsDuring Fe-S cluster assembly, IscS-mediated formation of a cluster onIscU is followed by cluster transfer from IscU to the final acceptor. Thesecond process is known to be slower than the first24. To assess whichstep is affected by CyaY, we dissected the pathway as follows. First, wecarried out non-enzymatic (chemical) cluster formation on IscU inthe presence and in the absence of CyaY and followed the process byCD (Fig. 2a). The absence of any observable effect indicates that CyaYintervenes in the enzymatic reaction.

We then followed the enzymatic reaction in the absence of IscU,using isc-encoded ferredoxin (Fdx) as the final acceptor. This

protein can form Fe-S clusters in the presence of IscS or otherdesulfurases, also in the absence of IscU23. (Fig. 2b). We observed astrong decrease in the kinetics in the presence of CyaY. The effectwas comparable with that observed in the presence of IscU as ascaffold (Fig. 1d), indicating that the inhibitory effect of CyaYinvolves IscS and not IscU.

To confirm these results and to assess further the role of IscS, weanalyzed cluster transfer from chemically reconstituted holo IscU (thatis, Fe-S cluster–loaded IscU) to Fdx in the presence and in the absenceof CyaY (Fig. 2c). Transfer is known to occur when the two proteinsare mixed because of the lower affinity of IscU for the cluster21.Because the CD signal of cluster-loaded Fdx is more intense than thatof cluster-loaded IscU24 (Supplementary Fig. 2 online), the reactionled to an overall increase of the signal intensity. CyaY had little, if any,effect on the rate of cluster transfer.

To further circumscribe the mode of action of CyaY, we testedwhether CyaY acts on the IscS desulfurase activity by assessing theefficiency of enzymatic cysteine-to-alanine conversion. We started thereaction in the presence of CyaY (5 mM) using low cysteine concen-trations (10 mM). When the reaction reached a plateau, we injectedadditional cysteine to reach a total added concentration at 20 mM,causing the reaction to start again and to proceed with the same rateobserved after the first addition (Fig. 2d). We collected aliquots of themixture at several time points. The alanine concentration, as estimatedby amino acid analysis, increased steadily and at the two plateaus was

0–1

0

1

2

3

4

mde

gm

deg

mde

gA

bsor

banc

e

5

6

Non-enzymatic IscU reconstitution

Enzymatic Fdx reconstitution

Cluster transfer from chemicallyloaded IscU to Fdx

CyaYNo CyaY

CyaYNo CyaY

CyaYNo CyaY

No CyaY

CyaY

7a

b

c

d

–1

0

1

2

3

4

5

6

7

1

2

3

4

5

6

7

100 200 300 400

Time (s)

500 600

010

020

030

040

0

Time (s)

500

600

0 200 400 600

Time (s)

800

0.02

0.03

0.04

0.025

0.035[Ala]

[Ala]

0.045 Dosage of Ala produced

03,

000

2,40

01,

800

1,20

060

0

Time (s)3,

600

Figure 2 Dissecting the pathway of cluster reconstitution in the absence and

in the presence of CyaY. (a) Kinetics of chemical reconstitution of IscU. The

reaction mixtures contained, in addition to 3 mM DTT and 250 mM cysteine,

which were common to all measurements in the figure, 40 mM Li2S, 40 mM

iron ammonium citrate and, when present, 5 mM CyaY. (b) Enzymatic

reconstitution of Fdx as the final acceptor in the absence of IscU. The

reaction mixtures contained 48 mM Fdx, 1 mM IscS, 80 mM ferric

ammonium citrate and, when present, 5 mM CyaY. (c) Cluster transfer from

chemically reconstituted holo IscU to Fdx as the final acceptor. The reaction

was started by adding Fdx (48 mM) to chemically preloaded IscU (50 mM)

and, when present, an excess of CyaY (50 mM) to enhance its effect, if any.

(d) Dosage of alanine production during enzymatic Fe-S assembly, as

followed by absorbance spectra recorded at 456 nm. The experiments were

carried out in the absence and in the presence of 5 mM CyaY using 10 mM

cysteine, 50 mM IscU, 25 mM Fe(NH4)2(SO4)2, 3 mM DTT and 1 mM IscS.In the experiment with CyaY, additional 10 mM cysteine solution was added

once the reaction had reached a plateau. The alanine concentration was

dosed by amino acid analysis.

66IscS

55

a b c d eNo CyaY

CyaY50

Flu

ores

cenc

e

15N

(p.

p.m

.)

45

105 Asp76His7

Trp61

0.12

0.10

0.08

Abs

orba

nce

0.06

0.04

0.02

Time (s)

0.000 1,200 2,400

No CyaYCyaY_D31KCyaY_E19KD22KCyaY_W61RCyaY_H7KD76KCyaY

3,600 4,800

Glu18Glu19

D22Asp31

110

2821

23

3022

4335

42

3433

39

29

44

46

115

120

125

130

13510 9 8

1H (p.p.m.)7

40

35

0 2.5 7.5[IscS] (µM)

10 12.55

GST

Mar

kers

GST cont

rol

GST-Cya

Y

pull-d

own

MW(kDa)

45

31

21

14

Figure 3 Interaction of CyaY with IscS. (a) GST pull-down assay using E. coli crude lysate. GST-saturated beads were used as a control. The band of IscU

was expected around the14-kDa region, where we instead identified lysozyme. (b) Titration of labeled IscU* with IscS in the absence and in the presence ofan excess of CyaY, as followed by fluorescence. (c) Comparison of the NMR HSQC spectra of 15N-labeled CyaY recorded at 25 1C and 800 MHz in the

absence (red) and in the presence (black) of unlabeled IscS (at a protein ratio of 1:0.8). Residues affected by the titration are marked. (d) Mapping the

observed effects on the CyaY structure (PDB 1EW4)6. The backbone of the protein is shown in blue (helical regions) and red (b-sheet). The side chains of

the residues affected are shown in yellow. The positions affected and mutated are shown in green. (e) Effect of CyaY mutations of residues involved in IscS

interaction on cluster reconstitution as followed by absorbance at 456 nm.

ART IC L E S

39 2 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

10.0 ± 0.8 mM and 18.9 ± 0.4 mM, respectively, showing that all thecysteine initially added has been converted into alanine.

These results indicate that inhibition of 2Fe-2S cluster formationresults from a direct effect of CyaY on IscS and is independent of thenature of the acceptor. CyaYdoes not inhibit the desulfurase activity ofIscS because the enzyme continues converting its substrate intoalanine until it is completely consumed.

CyaY forms a specific complex with IscSOur results point to a direct interaction between CyaY and IscS,which is in agreement with previous glutathione S-transferase(GST) pull-down studies using E. coli extracts containing over-expressed isc operon proteins13. To strengthen these results, wetested whether a pull-down also ‘fished out’ the endogenousproteins: we bound GST-tagged CyaY to a glutathione Sepharosecolumn and then passed an E. coli extract through the column.Under these conditions, we detected the presence of endogenousIscS but not IscU (Fig. 3a).

Because IscU also interacts with IscS, we tested whether CyaYcompetes with IscU for the same site on IscS using fluorescencespectroscopy. As all three proteins contain tryptophan residues, weattached a fluorophore, AlexaFluor532, to the more reactive sulfhydrylgroup on IscU. The monophasic curve obtained by titrating labeledIscU (IscU*) with IscS confirmed the presence of a direct interactionbetween IscU and IscS with a dissociation constant consistent with thevalue previously reported (1 mM)20 (Fig. 3b). The fluorescenceintensities obtained by individually titrating IscU* with IscS in thepresence of an excess of CyaY (apo or Fe3+ loaded) were super-imposable with those obtained in the absence of CyaY, suggesting anabsence of competition.

We next used NMR to map the surface of interaction on CyaY.When iron-free and iron-preloaded 15N labeled CyaY was titratedaerobically or anaerobically with unlabeled IscU, no spectralperturbations were observed (data not shown). In contrast, titra-tion of CyaY with IscS produced two effects: a progressive dis-appearance of the whole CyaY spectrum, as expected fromformation of the large molecular complex of CyaY with the IscSdimer (90 kDa), and a chemical shift perturbation that specificallyaffected a limited number of peaks (Fig. 3c). The residues stronglyaffected already at a 1:0.5 CyaY:IscS molar ratio were Asp22,Asp23, Ser28, Asp29, Glu33, Ile34, Phe43 and Glu44. Otherresidues affected at higher CyaY:IscS molar ratios (1:0.75) were

Leu21, Asn35, Leu39, Thr42, Gly46, Lys48, Thr64 and Gln98.These residues all map onto the protein surface that contains theiron binding sites6 (Fig. 3d).

The effect of CyaY on IscS is therefore the consequence of a directinteraction between the two molecules that does not competewith IscU binding.

CyaY mutations affect cluster reconstitution kineticsTo explore the role of CyaY residues in or around the surface of IscSbinding, we produced different mutants, chosen among those alreadywell characterized and known to have an effect in vivo or in vitro.CyaY_D22K, CyaY_D31K, CyaY_E19K D22K and CyaY_E18K E19KD22K affect residues that are completely or partially conserved and areknown to be involved in the main iron binding site6,25,26. Theequivalent mutations in yeast frataxin led to progressively severephenotypes in vivo27. We also tested a CyaY_W61R mutant becausethe equivalent position in human frataxin is associated with a severeFRDA case28. Although not immediately perturbed in our titrationwith IscS, this residue is in a region contiguous to the iron bindinginterface. A mutant of residues not perturbed by IscS titration andfar away from the iron binding surface (CyaY_H7K D76K)4 was usedas a control. We had previously confirmed that these mutants retaintheir fold4,29.

With the exception of the negative control CyaY_H7K D76K, whichbehaved like the wild type, the mutants showed progressively fasterkinetics of Fe-S formation on IscU, with initial rates similar to thatobserved in the absence of CyaY (Fig. 3e). CyaY_W61K has acomparatively smaller effect. These data validate the surfacefor CyaY–IscS interaction and confirm a role of CyaY in Fe-Scluster reconstitution.

The effect of CyaY is iron-concentration dependentAs the surface of CyaY interacting with IscS is also involved in ironbinding, we tested whether CyaY’s inhibitory effect could be sensitiveto variations in iron concentration. The initial reaction rates dependon the iron concentration, both in the presence and in the absence ofCyaY, which is expected because iron is a substrate in the overallcluster-formation reaction (Fig. 4a). However, the difference in ratesat the same Fe2+ concentration in the absence (closed symbols) and inthe presence (open symbols) of CyaY becomes much more marked athigher iron concentrations.

0.25 0.08

0.09

0.07

0.06

0.05

0.04

0.03

0.02

0 2 4 6 8 10 12 140.01

[Fe2+]100 µM

25 µM

10 µM5 µM

Abs

orba

nce

Initi

al r

ate

0.15

0.05

00

600

1,20

01,

800

2,40

0

Time (s)[CyaY] (µM)3,

000

3,60

04,

200

4,80

0

0.1

0.2

0.3Closed symbols: CyaYa bOpen symbols: no CyaY

Figure 4 The effect of iron and CyaY concentration on the kinetics of Fe-S

cluster formation. (a) Comparison of the kinetics followed by absorbance at

456 nm at increasing Fe2+ concentrations. (b) Effect of CyaY concentration

on the kinetics of cluster formation, keeping the other components fixed.

The Fe2+ concentration is 25 mM. The experiments were carried out using50 mM IscU, 250 mM cysteine, 3 mM DTT and 1 mM IscS.

600

500

400

Iron-inducedaggregates

Monomer

CyaY in 20 mM Tris-HCl, 100 mM NaClCyaY+Fe2+ (1:20) in 20 mM Tris-HCl, 100 mM NaClCyaY+Fe2+ (1:20) in 20 mM HEPES, no salt

300

mA

U

200

100

4 8 10 12

Volume (mI)

16 18 20146

0

Figure 5 Gel-filtration profiles to test the state of aggregation of CyaY under

the conditions used for cluster reconstitution. Minor differences in the

elution volumes of the monomer are likely to be due to minor differences in

iron loading of the individual species, which affect their stokes radii.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 9 3

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

As an independent control, we measured the rate of clusterformation as a function of CyaY concentration, while keeping theiron and enzyme concentrations constant (Fig. 4b). We observed adecrease in the rate of cluster formation with increasing CyaYconcentrations. The effect saturated at 5 mM CyaY, much lower thanthe IscU concentration used in our assays but close to the concentra-tion of IscS, confirming that CyaY affects IscS rather than IscU.

These results indicate that the CyaY effect depends on the amountof iron present in the environment.

CyaY inhibition does not require formation of multimerFrataxin has been suggested to act as a ferritin-like iron scavenger byforming large, spherical polymers of distinct stoichiometry15. We usedgel filtration to test whether oligomerization of CyaY could beimportant for the observed effects (Fig. 5). Freshly prepared iron-free CyaY eluted as a monomer, as previously described4. Under theconditions of the kinetic measurements, CyaY eluted again as amonomer. We did not observe species of large molecular weight,even when the protein was treated with an Fe2+ excess to createconditions favorable for oligomers formation4. Detectable amounts ofoligomers could be observed only when the experiment was carriedout at low ionic strength, in the absence of DTT and using HEPESrather than Tris-HCl.

These results indicate that the effect of CyaY is not linked toformation of oligomeric species. It must be completely attributedto the monomeric form of CyaY and its stoichiometric interactionwith IscS.

DISCUSSIONWe have investigated the effect of frataxin in the enzymology of Fe-Scluster biogenesis using purified proteins from E. coli. In using thebacterial system, we postulated that, given the high homology andstructural conservation within the frataxin family, the main features ofthe mechanism(s) by which frataxins functions must be the same, albeitwith species-specific adaptations. This assumption is supported bycomplementation studies in yeast using different frataxin orthologs30,31.

We reasoned that, if frataxins were iron chaperones14—that is,molecules that simply escort iron to its final destination—the presenceof CyaY should either enhance the enzymatic rates (if the rate-limitingstep of the reaction depended on iron delivery) or otherwise have noeffect. We observed instead that, far from facilitating iron delivery, thepresence of CyaY inhibits the reaction.

By dissecting the complex pathway of Fe-S cluster formation, wehave shown conclusively that IscS is indispensable for the CyaY actionand that the effect of CyaY is mediated through a direct interactionwith IscS, in agreement with a previous report13. Although we cannotin principle exclude an interaction of CyaY also with IscU in thecontext of the tertiary complex (IscU–IscS–CyaY), we see that CyaYhas little or no effect on the kinetics of the processes involving IscU orother scaffold proteins. CyaY does not alter the IscS desulfuraseactivity but is an inhibitor of cluster formation. Finally, we haveshown that CyaY does not compete with the IscU binding site on IscS,and its effect does not depend on the specific acceptor.

We can rationalize why a function of frataxin as an inhibitor of Fe-Sformation was not observed earlier. In vivo experiments, althoughcrucial for the identification of the hallmarks of the disease, could notprovide details on the mechanism of the process. Most of the previousin vitro work has, on the other hand, compared the effect of frataxinon the chemical reaction, thus missing the most important compo-nent, the desulfurase enzyme4,32. The only report that compares theeffect of CyaY on the enzymatic reaction was carried out by incubating

IscS with the substrate (cyteine) for 2 h before adding Fe3+-preloadedCyaY13. Under these conditions, cysteine is almost completelyconverted into alanine, as we have observed (Fig. 2d).

The mapped IscS interface includes CyaY residues that are highlyconserved and that have been implicated in iron binding, indicatingthat our overall conclusions can be generalized, although there mightbe some degree of species variability. Accordingly, we observed aniron-dependent effect of CyaY on Fe-S cluster formation, in agreementwith the observation that, in yeast, binding of the frataxin orthologYfh1 with the Isu1–Nfs1 complex (equivalent to IscU–IscS) is irondependent11. Notably, an iron-dependent effect of the activity offerrochelatase, the enzyme involved in another frataxin pathway, hasalso been reported10.

We propose that frataxins function as iron sensors and suggest amechanism for their action (Fig. 6). By a negative-regulation mechan-ism, frataxins would act as gate keepers of Fe-S assembly by finetuning the quantity of Fe-S clusters formed to match the concentra-tion of the apo acceptors. Frataxins would have low affinity for theIscS–IscU system at normal iron levels. At any even small ironimbalance (that is, an excess of iron as compared to final acceptors),the affinity of the protein for IscS would increase. One may wonderwhy such a regulation mechanism is needed, considering that theisc operon is under the control of the transcription factor IscR33.The necessity of regulation both at the transcriptional and post-translational levels can be explained by considering that IscR regulatesthe whole operon, and its action will require more time than theimmediate response that a component that interacts directly on thecentral component of the machinery, the enzyme, could have bysensing the iron concentrations.

In FRDA, and even more so in knockout models, where thephenotype is exacerbated, such a regulatory mechanism would beabsent. As a regulator that tunes the quantity of Fe-S cluster formed tothe availability of the apo acceptors, any reduction or depletion offrataxin levels would upset this equilibrium and lead to an imbalancein the amount of the Fe-S clusters produced with respect to the apoacceptors. Even a small iron accumulation could result in the forma-tion of Fe-S clusters at a rate incompatible with the concentration offinal acceptors. Fe-S clusters are labile species that cannot existwithout an acceptor or carrier. They would therefore fall apart,producing free iron, which in turn would give rise to Fentonchemistry. The FRDA phenotype, including the early damage ofFe-S cluster proteins observed both in mammalian and yeast knock-outs, could be a direct, early consequence of this process8,34. Theresulting free Fe3+, which is highly insoluble, would first form smalland amorphous nanoparticles and eventually precipitate, generatingthe large deposits detected in vivo8,9,34.

Our model provides an answer to several observations that arecurrently unexplained. Neither the molecular chaperone nor theferritin-like hypotheses14,15 fully account for the sequence conserva-tion26 and the essential nature of frataxins34. Our CyaY mutations ofconserved or semiconserved residues explain the severe phenotype ofequivalent mutants in yeast27,35. One of the main difficulties with theferritin-like hypothesis is the presence of a mitochondrial ferritin36.Iron transport in mitochondria, where large amounts of citrate andother iron transporters are present, is not a satisfactory answer.An iron-sensor role instead explains frataxin’s low affinity for ironand the long-term accumulation of the FRDA symptoms: a sensorrequires weak affinities. Even small frataxin concentrations would besufficient for cell viability, but frataxin deficiency would eventuallytrigger long-term catastrophic effects. It also becomes clear whyfrataxin is associated with oxidative stress and why time-dependent

ART IC L E S

39 4 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

intramitochondrial iron accumulation in frataxin-deficient organismsis observed after the onset of the pathology and after inactivation ofthe Fe-S–dependent enzymes. Finally, a role for the CyaY monomerrather than an aggregate is in agreement with previous work showingthat an oligomerization-deficient mutant of Yfh1 can still participatein Fe-S cluster biogenesis or heme assembly37.

We believe that, although more work is needed to establish thedetails of the molecular mechanism of IscS inhibition exerted by CyaYand provide a direct description of the eukaryotic system, our workopens an entirely new perspective to understanding the role offrataxins. This will hopefully promote new studies to clarify its linkswith the FRDA pathology.

METHODSProtein production. We overexpressed and purified the proteins, all from

E. coli, as previously described4,6,26. Fdx was obtained courtesy of L.E. Vickery

(University of California at Irvine). We prepared cluster-free Fdx by acidic

precipitation of holo Fdx24. We checked the purity of all proteins by SDS-PAGE

and by MS of the final product.

Absorbance experiments. We performed cluster reconstitution in an anaerobic

chamber (Belle) under a nitrogen atmosphere. We followed the reaction by

absorbance spectroscopy using a Cary 50 Bio (Varian) spectrophotometer.

Variations in the absorbance at 456 nm were measured as a function of time.

Unless otherwise specified, we incubated 50 mM solutions of purified IscU in

sealed cuvettes typically using 3 mM DTT and 25 mM Fe(NH4)2SO4 for 30 min

in 50 mM Tris-HCl buffer, pH 7.5, and 150 mM NaCl. Subsequently, we added

1 mM IscS and 250 mM cysteine to start the reaction. Chemical reconstitution

was carried out under similar conditions but replacing 1 mM IscS with 250 mM

Li2S as a source of sulfur. We studied the effects of iron and CyaY concentra-

tions by varying them individually in the range 5–100 mM and 0–12.5 mM,

respectively, with the other components fixed.

The control experiments of the effects of other iron carriers were done using

sodium citrate or calmodulin (50 mM).

To assess the effect of preloading CyaY with iron, we mixed CyaY (200 mM)

anaerobically with two equivalents of Fe(NH4)2SO4 (20 mM Tris-HCl, pH 7.5,

and 150 mM NaCl) and incubated the mixture for 1 h before adding an aliquot

to the enzymatic mixture to reach a final concentration of CyaY (10 mM)

preloaded with two equivalents of Fe2+.

To check the effect of the CyaY mutants, we added 5 mM of these proteins to

the enzymatic mixture before starting the reaction. Other controls are described

in the ‘‘Results’’ section of Supplementary Methods.

Frataxin

a

b

c

IscS dimer

IscU Final acceptor

+

+

+

In the presence of normal iron concentrations

Cluster-loadedfinal acceptor

Iron-loadedfrataxin

Slowerrates

No ratecontrol

Surplus ofFe-S clusters

Insoluble ironprecipitates

Fenton

chemistry

In conditions of frataxin deficiency

In the presence of any iron surplus

Cys Ala

Cys Ala

Cys Ala

Circular dichroism experiments. We obtained anaerobic conditions for CD

studies by using septum-capped 1-cm quartz cuvettes, stainless steel cannules

and anaerobic syringes for sample transfer. Cluster reconstitution was mon-

itored by following the increase of the CD signal at 435 nm using a Jasco

J-715 spectropolarimeter.

We diluted concentrated protein stocks to their final concentrations

(20–50 mM IscU and 4–5 mM CyaY or CyaY mutants) into 50 mM Tris-HCl,

150 mM NaCl, pH 7.5–8.0, containing 3 mM DTT. Cysteine was added to a

final concentration of 250 mM, followed by ferric ammonium citrate at a final

concentration less than or equal to that of IscU (40–50 mM). The reaction was

typically started by the addition of IscS at a final concentration of 1–2 mM.

The experiments in Figure 2 were carried out as follows: chemical recon-

stitution of IscU was performed with 3 mM DTT, 250 mM cysteine, 40 mM Li2S

and 40 mM ferric ammonium citrate. We probed the effect of CyaY on cluster

formation using Fdx as the final acceptor in the absence of IscU under similar

conditions to those used for IscU (1 mM IscS, 250 mM cysteine and 3 mM

DTT), but the final reaction mixtures contained 40–50 mM of Fdx, a twofold

molar excess of ferric ammonium citrate with respect to Fdx and, when

present, 5 mM CyaY. Transfer of the cluster from holo IscU to Fdx was followed

using 50 mM of chemically reconstituted IscU in the presence and in the

absence of CyaY (40 mM). For this experiment, IscU was chemically recon-

stituted by adding in small aliquots 500 mM ferric ammonium citrate and

500 mM Li2S to maximize the yield.

GST pull-down. We equilibrated GST-beads of Glutathione-Sepharose (500 ml)

in a buffer containing 20 mM Tris-HCl, 100 mM NaCl and 2 mM

b-mercaptoethanol and incubated them with an excess of purified GST-CyaY

in a final volume of 2 ml for 1 h at 4 1C. As a control we used GST. After

extensive washing with the same buffer, the saturated beads were mixed

overnight with E. coli crude lysate (DH5a strain). Potential protein partners

bound to the beads were eluted with 1 ml of 1 M NaCl in 50 mM Tris-HCl

buffer and separated by 12% SDS-PAGE. To ensure that no protein would

be retained on the beads, we used harsher conditions for the control:

GST-saturated beads were eluted with 20 mM glutathione in 50 mM

Tris-HCl buffer. The gels were stained by Novex colloidal Coomassie blue for

4 h. Stained bands were cut out, processed and analyzed by MS.

Fluorescence and nuclear magnetic resonance measurements. We performed

fluorescence experiments at 25 1C on a Jasco fluorimeter with excitation at

465 nm and emission at 546 nm. We kept the concentration of the species being

titrated constant throughout the titration. A 0.6 mM solution of IscU in 20 mM

Tris-HCl buffer, pH 7.0, and 150 mM NaCl was reacted for 1 h with a fourfold

excess of AlexaFluor 532 fluorescent probe (Invitrogen). We separated the

labeled product from the free fluorophore on a PD10 gel-filtration column and

eluted with 20 mM Tris-HCl buffer, pH 8.0, containing 150 mM NaCl and

20 mM b-mercaptoethanol. Labeled IscU (2 mM) was titrated with IscS (up to a

5-molar excess) in the absence and in the presence of CyaY (200 mM).

We recorded NMR spectra at 25 1C on a Varian spectrometer operating at

800 MHz 1H frequencies equipped with a 5 mm cryoprobe. All proteins were in

20 mM Tris-HCl, pH 8.0, 150 mM NaCl and 20 mM b-mercaptoethanol to

which 10% D2O was added. Iron-preloaded CyaY was obtained by adding

Fe2+ or Fe3+ (at protein:ion ratios of 1:2 or 1:6, respectively). Fe2+ was

added anaerobically.

Alanine dosage. We started enzymatic IscU reconstitution in the presence of

CyaY (5 mM) as in other absorbance assays (1 mM IscS, 50 mM IscU, 3 mM

Figure 6 Schematic model of the molecular mechanism of frataxin in the

cell. (a) At normal iron concentrations, the Fe-S clusters are assembled by

the IscS–IscU complex and passed on to their final acceptors. (b) Any

excess of iron as compared to the number of final acceptors will be

rebalanced by slowing down the reaction to match the concentration of final

acceptors and avoid unnecessary overproduction of Fe-S clusters. (c) When

frataxin is absent or produced in insufficient quantities, as in FRDA, there is

no regulation. Fe-S clusters will be produced irrespectively of whether they

can be transferred to an acceptor. Any iron excess will result in a surplus of

Fe-S clusters, which, being highly unstable, will fall apart, generating Fenton

reactions. Fe3+ will precipitate and form insoluble aggregates.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 9 5

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

DTT) but adding 20 mM cysteine in total, with a starting concentration of

10 mM. When the reaction reached a plateau, further cysteine was added to give

a total concentration of 20 mM. We collected aliquots in triplicates at different

reaction times (360, 720, 1,500, 2,200, 2,800 and 3,400 s). We then quenched

the reaction on each aliquot by adding 20% (w/v) trichloroacetic acid (TCA)

and collected samples. After keeping the samples on ice for 10 min, we left them

at –20 1C overnight to allow slow protein precipitation. The solution was

centrifuged at 11,000g for 5 min. We quantified the alanine content in the

supernatant by amino acid analysis.

Analysis of the oligomerization state of CyaY. We probed the oligomerization

state of CyaY during cluster reconstitution by gel filtration. CyaY (20 mM) was

incubated in a solution containing 3 mM DTT, 25 mM Fe(NH4)2(SO4)2,

50 mM Tris-HCl buffer, pH 7.5, 150 mM NaCl and 250 mM cysteine, that is,

the same composition used for cluster reconstitution except for the absence of

IscS and IscU. We incubated the solution at room temperature (25 1C) for 1 h

and injected it into the gel-filtration column. The experiment was repeated after

removing DTT and cysteine to eliminate reducing agents. In a separate

experiment, we incubated 20 mM CyaY with 100 mM Fe(NH4)2(SO4)2 in

20 mM HEPES, pH 7.4. These samples were loaded on an analytical Superdex

75 HR 10/30 column (Amersham Biosciences).

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ACKNOWLEDGMENTSWe dedicate this work to the memory of Margie Nair. We are thankful to R.A.G.Williams, F. Foury, M. Pandolfo and H. Puccio for stimulating discussions,P. Temussi for moral support, L. Temussi for technical discussions and theMill Hill NMR Centre for technical support. The project was supported by PURfunds (F.B. and S.I., University of Milan).

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Pandolfo, M. Iron and Friedreich’s ataxia. J. Neural Transm. Suppl. 70, 143–146(2006).

2. Campuzano, V. et al. Friedreich’s ataxia: autosomal recessive disease caused by anintronic GAA triplet repeat expansion. Science 271, 1423–1427 (1996).

3. Koutnikova, H. et al. Studies of human, mouse and yeast homologues indicate amitochondrial function for frataxin. Nat. Genet. 16, 345–351 (1997).

4. Adinolfi, S., Trifuoggi, M., Politou, A.S., Martin, S. & Pastore, A. A structural approachto understanding the iron-binding properties of phylogenetically different frataxins.Hum. Mol. Genet. 11, 1865–1877 (2002).

5. He, Y. et al. Yeast frataxin solution structure, iron binding & ferrochelatase interaction.Biochemistry 43, 16254–16262 (2004).

6. Nair, M. et al. Solution structure of the bacterial frataxin ortholog, CyaY: mapping theiron binding sites. Structure 12, 2037–2048 (2004).

7. Cook, J.D. et al. Monomeric yeast frataxin is an iron-binding protein. Biochemistry 45,7767–7777 (2006).

8. Foury, F. & Cazzalini, O. Deletion of the yeast homologue of the human gene associatedwith Friedreich’s ataxia elicits iron accumulation in mitochondria. FEBS Lett. 411,373–377 (1997).

9. Lesuisse, E. et al. Iron use for haeme synthesis is under control of the yeast frataxinhomologue (Yfh1). Hum. Mol. Genet. 12, 879–889 (2003).

10. Yoon, T. & Cowan, J.P. Frataxin-mediated iron delivery to ferrochelatase in the final stepof heme biosynthesis. J. Biol. Chem. 279, 25943–25946 (2004).

11. Gerber, J., Muhlenhoff, U. & Lill, R. An interaction between frataxin and Isu1/Nfs1 thatis crucial for Fe/S cluster synthesis on Isu1. EMBO Rep. 4, 906–911 (2003).

12. Ramazzotti, A., Vanmansart, V. & Foury, F. Mitochondrial functional interactionsbetween frataxin and Isu1p, the iron-sulfur cluster scaffold protein, in Saccharomycescerevisiae. FEBS Lett. 557, 215–220 (2004).

13. Layer, G., Ollagnier-de Choudens, S., Sanakis, Y. & Fontecave, M. Iron-sulfur clusterbiosynthesis: characterization of Escherichia coli CYaY as an iron donor for theassembly of [2Fe-2S] clusters in the scaffold IscU. J. Biol. Chem. 281,16256–16263 (2006).

14. Yoon, T. & Cowan, J.A. Iron-sulfur cluster biosynthesis characterization of frataxin as aniron donor for assembly of [2Fe-2S] clusters in ISU-type proteins. J. Am. Chem. Soc.125, 6078–6084 (2003).

15. Adamec, J. et al. Iron-dependent self-assembly of recombinant yeast frataxin: implica-tions for Friedreich ataxia. Am. J. Hum. Genet. 67, 549–562 (2000).

16. Johnson, D.C., Dean, D.R., Smith, A.D. & Johnson, M.K. Structure, function, andformation of biological iron-sulfur clusters. Annu. Rev. Biochem. 74, 247–281(2005).

17. Nuth, M., Yoon, T. & Cowan, J.A. Iron-sulfur cluster biosynthesis: characterization ofiron nucleation sites for assembly of the [2Fe-2S]2+ cluster core in IscU proteins.J. Am. Chem. Soc. 124, 8774–8775 (2002).

18. Agar, J.N., Krebs, C., Frazzon, J., Hanh Huynh, B. & Dean, D.R. IscU as a scaffold foriron-sulfur cluster biosynthesis: sequential assembly of [2Fe-2S] and [4Fe-4S] clustersin IscU. Biochemistry 39, 7856–7862 (2000).

19. Yuvaniyama, P., Agar, J.N., Cash, V.L., Johnson, M.K. & Dean, D.R. NifS-directedassembly of a transient [2Fe-2S] cluster within the NifU protein. Proc. Natl. Acad. Sci.USA 97, 599–604 (2000).

20. Urbina, H.D., Silberg, J.J., Hoff, K.G. & Vickery, L.E. Transfer of sulfur fromIscS to IscU during Fe/S cluster assembly. J. Biol. Chem. 276, 44521–44526(2001).

21. Mansy, S.S. Wu, G., Surerus, K.K. & Cowan, J.A. Iron-sulfur cluster biosynthesis.Thermatoga maritima IscU is a structured iron-sulfur cluster assembly protein.J. Biol. Chem. 277, 21397–21404 (2002).

22. Bou-Abdallah, F., Adinolfi, S., Pastore, A., Laue, T.M. & Chasteen, D.N. Iron binding &oxidation kinetics in frataxin CyaY of Escherichia coli. J. Mol. Biol. 341, 605–615(2004).

23. Bonomi, F., Pagani, S. & Kurtz, D.M. Jr. Enzymatic synthesis of the 4Fe-4S clusters ofClostridium pasteurianum ferredoxin. Eur. J. Biochem. 148, 67–73 (1985).

24. Bonomi, F., Iametti, S., Ta, D. & Vickery, L.E. Multiple turnover transfer of [2Fe2S]clusters by the iron-sulfur cluster assembly scaffold proteins IscU and IscA. J. Biol.Chem. 280, 29513–29518 (2005).

25. Musco, G. et al. Towards a structural understanding of Friedreich’s ataxia: the solutionstructure of frataxin. Structure 8, 695–707 (2000).

26. Pastore, C., Franzese, M., Sica, F., Temussi, P. & Pastore, A. Understanding thebinding properties of an unusual metal binding protein: a study of bacterial frataxin.FEBS J. 274, 4199–4210 (2007).

27. Foury, F., Pastore, A. & Trincal, M. Acidic residues of yeast frataxin have an essentialrole in Fe-S cluster assembly. EMBO Rep. 8, 194–199 (2007).

28. Labuda, M., Poirier, J. & Pandolfo, M. A missense mutation (W155R) in an Americanpatient with Friedreich Ataxia. Hum. Mutat. 13, 506–509 (1999).

29. Correia, A.R., Adinolfi, S., Pastore, A. & Gomes, C. Conformational stability of humanfrataxin and effect of Friedreich’s ataxia related mutations on protein folding. Biochem.J. 398, 605–611 (2006).

30. Bedekovics, T., Gajdos, G.B., Kispal, G. & Isaya, G. Partial conservation of functionsbetween eukaryotic frataxin and the Escherichia coli frataxin homolog CyaY. FEMSYeast Res. 7, 1276–1284 (2007).

31. Cavadini, P., Gellera, C., Patel, P.I. & Isaya, G. Human frataxin maintains mito-chondrial iron homeostasis in Saccharomyces cerevisiae. Hum. Mol. Genet. 9,2523–2530 (2000).

32. Huang, J., Dizin, E. & Cowan, J.A. Mapping iron binding sites on human frataxin:implications for cluster assembly on the ISU Fe-S cluster scaffold protein. J. Biol.Inorg. Chem. 13, 825–836 (2008).

33. Schwartz, C.J. et al. IscR, an Fe-S cluster-containing transcription factor, repressesexpression of Escherichia coli genes encoding Fe-S cluster assembly proteins. Proc.Natl. Acad. Sci. USA 98, 14895–14900 (2001).

34. Cossee, M. et al. Inactivation of the Friedreich ataxia mouse gene leads to earlyembryonic lethality without iron accumulation. Hum. Mol. Genet. 9, 1219–1226(2000).

35. Leidgens, S. The function of yeast frataxin in iron-sulfur cluster biogenesis: asystematic mutagenesis of the solvent-exposed side chains of the beta-sheet platform.PhD thesis Louvain la Neuf (Belgium) (2008).

36. Levi, S. et al. A human mitochondrial ferritin encoded by an intronless gene. J. Biol.Chem. 276, 24437–24440 (2001).

37. Aloria, K., Schilke, B., Andrew, A. & Craig, E.A. Iron-induced oligomerization of yeastfrataxin homologue Yfh1 is dispensable in vivo. EMBO Rep. 5, 1096–1101(2004).

ART IC L E S

39 6 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

The pathway of hepatitis C virus mRNA recruitmentto the human ribosomeChristopher S Fraser1,6, John W B Hershey2 & Jennifer A Doudna1,3–5

Eukaryotic protein synthesis begins with mRNA positioning in the ribosomal decoding channel in a process typically controlledby translation-initiation factors. Some viruses use an internal ribosome entry site (IRES) in their mRNA to harness ribosomesindependently of initiation factors. We show here that a ribosome conformational change that is induced upon hepatitis C viralIRES binding is necessary but not sufficient for correct mRNA positioning. Using directed hydroxyl radical probing to monitor theassembly of IRES-containing translation-initiation complexes, we have defined a crucial step in which mRNA is stabilized uponinitiator tRNA binding. Unexpectedly, however, this stabilization occurs independently of the AUG codon, underscoring theimportance of initiation factor–mediated interactions that influence the configuration of the decoding channel. These resultsreveal how an IRES RNA supplants some, but not all, of the functions normally carried out by protein factors during initiationof protein synthesis.

Protein synthesis in all cells begins with binding and positioning of anmRNA on the small ribosomal subunit. In eukaryotes, the 5¢ 7-methylguanosine cap structure on most mRNAs triggers an assembly oftranslation factors that recruit the 40S ribosomal subunit. This highlyregulated process, leading to 60S subunit joining and formation of anactive 80S ribosome, requires at least 12 initiation factors composed ofroughly 28 polypeptides1,2. Some viral mRNAs lack a 5¢ cap andinstead include a structured RNA sequence, the IRES, in the 5¢untranslated region (UTR), which functions in place of some or allof the canonical initiation factors3,4. Some of the most detailedinformation to date for this type of mechanism has come from thestudy of the IRES found in the hepatitis C virus (HCV) mRNA5,6. Anefficient 40S subunit initiation complex on the HCV IRES requiresonly two initiation factors, eIF2 and eIF3 (ref. 7). The eIF2 complexrecruits the initiator tRNA (Met-tRNAi), whereas the much larger eIF3complex enhances formation of the 40S subunit initiation complex onthe HCV IRES, in part by stabilizing the eIF2–GTP–Met-tRNAi

complex (ternary complex) on the 40S subunit8–10.In both initiation pathways, mRNA recruitment and decoding

occur in the mRNA binding channel, which is situated between thehead, body and platform of the 40S subunit. This binding regioncomprises the channel through which the mRNA enters the 40Ssubunit, the ribosome decoding sites (aminoacyl (A) site, peptidyl(P) site and exit (E) site) and the exit channel through which themRNA leaves the 40S subunit11 (Fig. 1a). The entry channel isoccluded in empty 40S subunits, leading to the proposal that a

conformational change is required for mRNA loading12–14. Indeed,cryo-EM–derived structures of 40S–HCV IRES complexes revealed astructural change in 40S subunits bound to the wild-type IRESwhereby the mRNA entry channel seemed to be more open relativeto that of unbound 40S subunits13. Domain II of the HCV IRES(Fig. 1b) was shown to be responsible for the 40S structural change,correlating with toeprinting data that indicated a requirement ofdomain II for mRNA entry into the binding channel in the absenceof initiation factors15,16. Subsequent cryo-EM reconstructions of the40S subunit bound to the cricket paralysis viral IRES or to initiationfactors eIF1 and eIF1A showed similar conformational changes in the40S subunit head, hinting at a common mechanism of mRNA loadingby viral IRESs and cap-dependent initiation factors14,17.

Another clue to the mechanism of mRNA loading came from thediscovery that a subunit of eIF3, eIF3j, binds stably to 40S subunitsonly in the absence of mRNA9,18. Directed hydroxyl radical probingshowed that the C terminus of eIF3j lies in the 40S mRNA entrychannel, where it presumably disfavors mRNA binding in the absenceof other initiation factors19. Upon initiator tRNA recruitment, as partof the ternary complex, eIF3j is displaced and mRNA binding isenhanced. How viral IRESs trigger this crucial switch, leadingto proper positioning of the viral mRNA on the 40S subunit,is unknown.

To address this question, we used directed hydroxyl radical probingand ribosome toeprinting of reconstituted translation-initiation com-plexes to determine the steps required for HCV IRES–mediated

Received 29 September 2008; accepted 3 February 2009; published online 15 March 2009; doi:10.1038/nsmb.1572

1Howard Hughes Medical Institute, University of California, Berkeley, California 94720, USA. 2Department of Biochemistry and Molecular Medicine, School ofMedicine, University of California, Davis, California 95616, USA. 3Department of Molecular and Cell Biology, 4Department of Chemistry, University of California,Berkeley, California 94720, USA. 5Physical Biosciences Division, Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA. 6Present address: Sectionof Molecular and Cellular Biology, College of Biological Sciences, University of California, Davis, California 95616, USA. Correspondence should be addressed to J.A.D.(doudna@berkeley.edu).

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 9 7

ART IC L E S

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

mRNA positioning in the 40S decoding channel. We established thatmRNA and the C terminus of eIF3j do not bind simultaneously in theribosome entry channel and that eIF3j displacement signals mRNAentry and binding in the channel. Using this as an assay for initiationcomplex formation, we show that the 40S conformational changeinduced by the IRES domain II is necessary but not sufficient formRNA entry into the decoding channel. In addition, both the mRNAstrand downstream of the AUG codon and the ternary complex arerequired for the mRNA to displace eIF3j. Notably, this effect does notrequire AUG recognition by the initiator tRNA, implying that correctmRNA positioning is a function of ribosome conformation ratherthan mRNA-tRNA base pairing. Our results support an orderedpathway for HCV IRES–mediated translation initiation in which aneIF3- and IRES-stabilized conformational state of the 40S subunitfavors viral mRNA entry into the decoding channel but only afterternary complex binding.

RESULTSIRES domain II promotes mRNA entry into the binding channelPrevious studies showed that initiation factors eIF1, eIF1A, eIF2 andeIF3 are necessary for stable mRNA positioning in the ribosomaldecoding channel, as monitored by toeprinting and reduced eIF3jaffinity to the 40S subunit19,20. To test whether the HCV IRESsimilarly favors mRNA binding and eIF3j displacement from themRNA binding channel, we monitored 18S ribosomal RNA (rRNA)cleavages induced by bromoacetamidobenzyl-EDTA/Fe (BABE-Fe)-modified eIF3j proteins19 (Fig. 1c). As shown previously, eIF3jproteins containing a single BABE-Fe moiety at several positionsnear the C terminus lead to site-specific rRNA cleavages in themRNA entry channel (helix 34) in complexes containing only the40S subunit and eIF3j. Addition of the wild-type HCV IRES mRNAto this complex largely prevented these cleavages (Fig. 1d and

Supplementary Fig. 1b online). This observation could indicatethat the C terminus of eIF3j dissociates from the 40S mRNA bindingcleft upon HCV mRNA association. Consistent with this idea, weobserved no cleavage of the mRNA segment of the HCV IRES RNA inIRES–40S–eIF3j complexes, suggesting that eIF3j is no longer in thechannel (data not shown).

To distinguish between competitive binding of eIF3j and HCVmRNA and HCV mRNA–induced protection of 18S rRNA cleavagesites, we mapped the 40S subunit position on the IRES mRNA bytoeprinting21–23. Consistent with previous results7,15, the associationof the HCV mRNA with the 40S subunit induces a toeprint atnucleotides +20 and +21 downstream of the AUG codon, reflectingthe leading edge of the 40S subunit on the mRNA (Fig. 2a). Thistoeprint was inhibited upon addition of increasing concentrations ofeIF3j, indicating that the association of mRNA with the entry channelrequires eIF3j displacement. Notably, the toeprint located at nucleo-tides +3 and +4 is still evident, even at a large molar excess of eIF3j(5–40 mM), suggesting that eIF3j influences only mRNA associationwith the entry channel and not HCV mRNA association with the restof the 40S subunit. This effect probably explains our previous dataindicating that a short mRNA can bind concurrently with eIF3j19.

As the conformational changes induced by domain II of the HCVIRES have been suggested to facilitate mRNA entry into the bindingchannel, we tested an HCV mRNA lacking this domain (denoted asHCVDII in figures) in our cleavage assay. Previous studies showed thatthis truncated form of the HCV IRES binds with similar affinity to the40S subunit but fails to induce efficient translation initiation7,24. Wefound that the cleavages induced at nucleotides 1486–1491 in helix34 by BABE-Fe–modified C-terminal positions on eIF3j were restored(Fig. 1d), suggesting that domain II is necessary for eIF3j displacementfrom, and mRNA binding to, the decoding channel. However, it is notpossible from these data to determine the order in which these events

Figure 1 Directed hydroxyl radical probing of

18S rRNA from BABE-Fe–eIF3j–40S–HCV IRES

complexes. (a) The 40S subunit structure based

on a cryo-EM reconstruction13 viewed from the

subunit interface with landmarks indicated:

A, A-site; P, P-site; E, E-site; bk, beak; b, body;

pt, platform; h, head. (b) The 5¢ UTR of the HCV

mRNA consists of four domains (I–IV); the IRES

domains (II–IV) with subdomains (a–f) of domain

III are indicated. (c) Representation of eIF3j,

indicating the positions of cysteine mutations

used for BABE-Fe attachment (above). Modeled

positions of eIF3j amino acids in the Thermus

thermophilus 30S crystal structure, adapted from

a previous publication19 (below). The boxed areaprovides a detailed view of the mRNA entry

channel and A-site with helices 18, 32 and 34

indicated. Nucleotides cleaved in these helices

for each experiment are shown in Supplementary

Fig. 1a. (d) Primer-extension analysis of 18S

rRNA cleaved by BABE-Fe–modified eIF3j.

Sequencing lanes are indicated (C, T, A and G).

Other control lanes include 40S subunits in

the absence or presence of EDTA/Fe, mock-

derivatized eIF3j (�cys+EDTA/Fe), in the

absence (lane 7) or presence of wild-type (HCV;

lane 8), or domains III–IV (HCVDII; lane 9), of

the HCV IRES RNA. Other lanes include eIF3j derivatized with BABE-Fe at the positions indicated, either in the absence (lanes 10, 13, 16 and 19) or

presence (lanes 11, 14, 17, and 20) of HCV IRES, or HCVDII IRES (lanes 12, 15, 18 and 21). 18S rRNA nucleotide positions of cleavage sites are

indicated. Colored circles indicate components added in each reaction as depicted in the cartoon. The deletion of domain II (HCVDII) is represented

by a dotted line.

mRNAexit channel

a b

c

d

30S

eIF3j

h32

h18

h34

CLane

1486–1491

1502–1504

1523–1526

eIF3j

40S HCV

+ OR

HCV∆II

1508

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

T AG 40S

ED

TA/F

e–c

ys+

ED

TA/F

e–c

ys+

HC

V+

ED

TA/F

e–c

ys+

HC

V∆I

I+E

DTA

/Fe

S15

2CS

152C

+H

CV

S15

2C+

HC

V∆I

IS

217C

S21

7C+

HC

VS

217C

+H

CV

∆II

T23

5CT

235C

+H

CV

T23

5C+

HC

V∆I

IA

241C

A24

1C+

HC

VA

241C

+H

CV

∆II

1521 217 235 241 258

IIId

IIIe

IV

IIIf

IIIcIIIa

IIIb

I

5′

3′

bkh

AP E pt

b

mRNAentry channel

40S

II

ART IC L E S

39 8 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

occur. We next tested an otherwise wild-type version of the HCV IRESmRNA truncated after the AUG codon to determine whether therRNA cleavage sites induced by BABE-Fe–modified eIF3j are generatedwhen the mRNA strand does not extend into the eIF3j binding site onthe 40S subunit. Although this shortened mutant associates with the40S subunit (data not shown), the rRNA cleavage sites were essentiallyunchanged relative to those observed for 40S–eIF3j complexes(Fig. 2b). This suggests that, in addition to the conformationalchanges induced by domain II, the presence of mRNA in the entrychannel and A-site of the 40S subunit is required to promote eIF3jdisplacement. It should be noted that domain II does not stablyassociate with the 40S subunit in the absence of other domains of theHCV IRES.

eIF3 regulates HCV mRNA entry into the binding channelWhereas the C-terminal half of eIF3j is located in the mRNAbinding channel of the 40S subunit, its N terminus interacts withthe eIF3b subunit of eIF325. If maintained in the presence of the40S–eIF3–HCV IRES mRNA complex, this interaction shouldincrease the local concentration of eIF3j on the 40S subunit andthus enhance eIF3j’s ability to compete for binding to the entry

channel with mRNA. To test this possibility,we analyzed BABE-Fe–modified eIF3j–induced cleavage of 18S rRNA in reconsti-tuted 40S–eIF3–HCV IRES mRNA com-plexes. This was achieved by the additionof BABE-Fe–modified eIF3j together withan eIF3 complex purified without endogen-ous eIF3j attached (eIF3Dj). In contrast tocomplexes containing the eIF3j subunitalone, those containing intact eIF3 yieldedsimilar cleavage patterns to those observedin the absence of the HCV IRES mRNA(Fig. 3a). Similar experiments usingHCVDII produced small, but reproduciblyenhanced, cleavage intensities in rRNA helix34 relative to wild-type HCV IRES mRNA(Fig. 3a). These results suggest that the eIF3complex enables eIF3j to compete moreeffectively with HCV IRES mRNA for bind-ing to the mRNA entry channel, particularlyin the absence of the IRES domain II–induced 40S conformational change. Thesedata raise the question of how the HCVmRNA can efficiently associate with themRNA entry channel in the presence ofeIF3j and the eIF3 complex during theprocess of translation initiation.

Previously, we showed that ternary com-plex association with the 40S subunitenhances the affinity of a short, unstructuredmRNA for the 40S subunit in the presence ofeIF3j (ref. 19), hinting at a role for the ternarycomplex in displacing eIF3j during mRNAloading. To test this possibility, we used site-directed hydroxyl radical probing to deter-mine the effect of the ternary complex onHCV IRES mRNA association in the 40Ssubunit mRNA entry channel in the presenceof eIF3j and intact eIF3Dj. Unfortunately, itwas not possible to use the cleavage site that

we have described in helix 34 for this purpose because association ofthe ternary complex alone protects helix 34 from eIF3j-inducedcleavage (Supplementary Fig. 2a online). This occurs even thougheIF3j is still present in the mRNA entry channel, as indicated by theunchanged cleavage intensity of helix 18 by BABE-Fe–modified eIF3jin the presence of ternary complex (Supplementary Fig. 2b). Cleavageof helix 18 by BABE-Fe–modified eIF3j was unchanged by theassociation of eIF3Dj with the 40S subunit in the absence (Supple-mentary Fig. 2c), or presence (Supplementary Fig. 2d), of the HCVmRNA, and this allowed visualization of this region to determinewhether any ternary complex–specific changes occur during HCVmRNA association. Upon recruitment of the ternary complex to the40S subunit in the presence of HCV mRNA and eIF3Dj, cleavageintensities at all nucleotide positions in helix 18 in the mRNA entrychannel were diminished (Fig. 3b). This observation implies that,once the ternary complex associates, even in the presence of intacteIF3, HCV IRES mRNA enters the entry channel and eIF3j isdisplaced. Notably, this effect was not seen in complexes containingHCVDII (Fig. 3b).

One explanation for these results is that AUG recognition by theinitiator tRNA stabilizes mRNA on the 40S subunit. To test this, we

1C

a b

T A G HC

VH

CV

+40

S

2 3 4 5 6 7 8 9 10 Lane

Entry channel Decoding site

A P E

A P E

40S

C T A G 40S

ED

TA/F

e

S15

2C

S21

7C

T23

5C

A24

1CA

241C

+H

CV

∆OR

F

T23

5C+

HC

V∆O

RF

S15

2C+

HC

V∆O

RF

S21

7C+

HC

V∆O

RF

–cys

+E

DTA

/Fe

–cys

+H

CV

∆OR

F+

ED

TA/F

e

+3/4

GUA

3′

3′

5′ (HCV)

5′ (HCV)GUA

Exit channel

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Lane

1486–1491

1523–1526

1502–1504

1508

ATG +3/4

+20/21

+20/21

Entry channel Decoding site Exit channel

40S

40S

HCV∆ORF

elF3j +

+elF3j

Figure 2 Toeprinting analysis of the 40S–HCV–eIF3j complexes. (a) Lanes C, T, A and G depict

sequencing lanes corresponding to HCV mRNA, with the AUG codon indicated. Toeprinting reactions of

HCV mRNA in the absence (lane 5) or presence (lane 6) of 40S subunits is shown. Additional reactions

including 40S subunits in the presence of 5 mM (lane7), 10 mM (lane 8), 20 mM (lane 9), or 40 mM(lane 10) eIF3j are indicated. The positions of toeprints that correspond to 40S–HCV complexes are

indicated (+3/4 and +20/21). Numbering is from the A (+1) of the AUG codon. Cartoons depicting the

40S–HCV complexes formed are also indicated. (b) Primer-extension analysis of 18S rRNA cleaved by

BABE-Fe–modified eIF3j in the absence or presence of HCV IRES RNA truncated after the AUG

codon (HCVDORF). Sequencing and control lanes are indicated, as described in Figure 1d. The lanes

corresponding to eIF3j in the absence (lanes 9, 11, 13 and 15) or presence (lanes 10, 12, 14

and 16) of HCVDORF are indicated. As described in Figure 1d, the colored circles correspond to

the components added in each reaction.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 3 9 9

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

mutated the AUG codon and surrounding nucleotides (ATCATG toAAAAAA) in domain IV of the HCV IRES and analyzed complexescontaining this construct by site-directed hydroxyl radical probing.Unexpectedly, in the absence of the AUG codon, ternary complexrecruitment still promotes HCV mRNA binding in the 40S subunitentry channel, as determined by the dissociation of eIF3j (Fig. 3b).Taken together, these data show that, in the presence of intact eIF3,association of ternary complex with the 40S subunit stabilizes HCVIRES mRNA in the entry channel in a process that is independent ofAUG codon recognition by the initiator tRNA.

eIF3 binding triggers structural changes in the 40S subunitHow do eIF3 and the ternary complex stabilize mRNA in the 40Ssubunit decoding channel irrespective of AUG codon recognition? Animportant clue to a possible mechanism came from observations ofdistinct conformational states of the 40S subunit by using cryo-EM tocompare empty 40S subunits to 40S subunits complexed withviral IRES mRNA or initiation factors eIF1 and eIF1A13,14,17. Thesestructures showed similar motions of the head relative to thebody of the 40S subunit, altering the architecture of the mRNAbinding channel, which could thereby affect access to the mRNAbinding channel.

To test whether these characterized structural changes are detectableusing our site-directed hydroxyl radical probing assay, we formed eachcomplex and determined the cleavage pattern from hydroxyl radicalsgenerated from eIF3j. To visualize its association with the 40S sub-unit, the HCV mRNA was truncated at the AUG codon to avoideIF3j dissociation, as discussed above. We used 18S rRNA cleavagesresulting from BABE-Fe–generated hydroxyl radicals from differentsites on eIF3j to estimate changes in the three-dimensional structureof the 40S subunit, as previously validated for the bacterial ribo-some26. Distance changes between the location of the BABE-Femoiety and a specific nucleotide in the rRNA result in altered

cleavage intensities, with more intense cleavages representing reduceddistances between BABE-Fe and the affected sites.

We compared the rRNA cleavage patterns generated in parallelexperiments using two different BABE-Fe–modified eIF3j variants(Fig. 1c). Cleavage sites to be monitored were selected in the regionof the 40S subunit beak (Fig. 1a) because the conformation of thisstructure changes substantially upon IRES or initiation factor bind-ing13,14,17. Consistent with the cryo-EM structures, 18S rRNA cleavagepatterns generated from BABE-Fe positions in eIF3j changed upon theassociation of saturating amounts of eIF1 and eIF1A or the HCVmRNA with the 40S subunit (Fig. 4a,b). Specifically, for both com-plexes, we observed reduced cleavage intensities at nucleotides 1268–1270, 1295–1298 and 1305–1307 in helices 32 and 33 generated fromposition 152 in eIF3j (Fig. 4a,b, compare lanes 9 and 10). However, incontrast to the recruitment of HCV mRNA, eIF1 and eIF1A associa-tion with the 40S subunit resulted in increased cleavage intensities ofnucleotides 1279–1284 generated from BABE-Fe tethered to aminoacid positions 152 and 217 on eIF3j (Fig. 4a,b, compare lanes 9 and10). Therefore, although eIF1 and eIF1A, or the HCV mRNA, inducesimilar structural changes in the 40S subunit, these data obtained insolution imply some differences in the way these components alter the40S subunit structure, particularly surrounding the mRNA entrychannel and A-site. It should be noted that the observed cleavagedifferences could also result partly or entirely from direct interactionsbetween eIF3j with eIF1 and eIF1A. We have previously shown that theinteraction between eIF3j and eIF1A is anticooperative19, whichprobably indicates altered conformations of eIF3j and eIF1A on the40S subunit.

We next tested whether the eIF3 complex alters the conformation ofthe 40S subunit upon binding. To this end, we compared 18S rRNAcleavages generated from BABE-Fe tethered to positions in eIF3j in theabsence or presence of the eIF3 complex. Association of the eIF3complex seems to alter the cleavage pattern in a similar, but distinct,

C

a b

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 181 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 2221201918

T A G C T A G 40S

40S

S21

7CS

217C

+H

CV

+el

F3∆

j+T

C

S21

7C+

HC

V∆I

I+el

F3∆

j+T

CS

217C

+H

CV

∆AU

G+

elF

3∆j+

TC

T23

5CT

235C

+H

CV

+el

F3∆

j+T

C

T23

5C+

HC

V∆I

I+el

F3∆

j+T

CT

235C

+H

CV

∆AU

G+

elF

3∆j+

TC

A24

1CA

241C

+H

CV

+el

F3∆

j+T

C

A24

1C+

HC

V∆I

I+el

F3∆

j+T

C

A24

1C+

HC

V∆A

UG

+el

F3∆

j+T

C

S21

7C

A24

1C

S21

7C+

HC

V+

elF

3∆j

S21

7C+

HC

V∆I

I+el

F3∆

jT

235C

T23

5C+

HC

V+

elF

3∆j

T23

5C+

HC

V∆I

I+el

F3∆

j

A24

1C+

HC

V+

elF

3∆j

A24

1C+

HC

V∆I

I+el

F3∆

j

ED

TA/F

e

ED

TA/F

e

–cys

+E

DTA

/Fe

–cys

+H

CV

+el

F3∆

j+T

C+

ED

TA/F

e

–cys

+E

DTA

/Fe

–cys

+H

CV

∆AU

G+

elF

3∆j+

TC

+E

DTA

/Fe

–cys

+H

CV

+el

F3∆

j+E

DTA

/Fe

–cys

+H

CV

∆II+

elF

3∆j+

ED

TA/F

e

–cys

+H

CV

∆II+

elF

3∆j+

TC

+E

DTA

/Fe

Lane

601–609

626–628

636–639

616

1523–1526

40S

elF3j

elF3∆jelF3j

elF3∆jHCV

OROR

++++

OR

HCV∆AUG HCV∆II

40S

TC

HCV HCV∆II

1502–1504

1486–1491

Lane

1508

Figure 3 Effects of eIF3 and eIF2–Met-tRNAi on

directed hydroxyl radical probing of 18S rRNA

with BABE-Fe–eIF3j. (a) Lanes include 40S

subunits in the absence or presence of EDTA/Fe

and mock-derivatized eIF3j (�cys+EDTA/Fe) in

the absence or presence of HCV constructs and

eIF3 complex without endogenous eIF3j (eIF3Dj).

Lanes corresponding to eIF3j derivatized with

BABE-Fe at the positions indicated in the

absence (lanes 10, 13 and 16), or presence, of

eIF3Dj and wild-type HCV IRES (HCV; lanes 11,

14 and 17), or domain III of the HCV IRES

(HCVDII; lanes 12, 15 and 18) are indicated.

(b) Lanes include 40S subunits in the absence or

presence of EDTA/Fe and mock-derivatized eIF3j(�cys+EDTA/Fe) in the absence or presence of

HCV constructs and other initiation factors, as

indicated. Lanes corresponding to BABE-Fe–

modified eIF3j at the positions indicated in the

absence (lanes 11, 15 and 19) or presence of

eIF3Dj, eIF2–Met-tRNAi (ternary complex; TC)

and HCV (lanes 12, 16 and 20), HCVDAUG

(lanes 13, 17 and 21), or HCVDII (lanes 14, 18

and 22) are indicated. For each gel, sequencing

lanes (C, T, A and G) and cleavage-nucleotide

positions in the 18S rRNA are indicated. Colored

circles correspond to the components added, as

depicted in the cartoons. Relevant mutations in

each HCV IRES construct are represented by a

dotted line.

ART IC L E S

40 0 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

manner compared to the different complexes described above (Fig. 4c).In particular, the cleavage intensities of 18S rRNA nucleotides 1279–1284 generated by eIF3j modified with BABE-Fe at positions 152 and217 were enhanced by the addition of eIF3 or eIF1 and eIF1A to the40S subunit (Fig. 4a,c, compare lanes 9 and 10). However, in contrastto the cleavage patterns observed upon addition of the HCV IRES oreIF1 and eIF1A, the cleavage intensities of nucleotides 1295–1298 and1305–1307 by hydroxyl radicals generated from position 152 in eIF3jremained unchanged upon addition of eIF3 (Fig. 4a,c, compare lanes11 and 12). Notably, the eIF3 complex, eIF1 and eIF1A, or the HCVIRES mRNA all reduced the cleavage intensities of nucleotides 1268–1270 in helix 32 that are generated from eIF3j modified with BABE-Feat position 152 (Fig. 4a–c, compare lanes 9 and 10). Thus, the eIF3complex stabilizes a specific 40S subunit conformation that has somesimilarities to the structures previously reported for HCV mRNA oreIF1 and eIF1A bound to the 40S subunit. However, these conforma-tional changes are not sufficient to favor mRNA binding to thedecoding channel, as detected by eIF3j displacement or 40S toeprint-ing20, in the absence of additional initiation factors. Furthermore, we

also observed differences in cleavage-intensity changes between com-plexes containing eIF1 and eIF1A, or eIF3j in helix 18 (SupplementaryFig. 2b), confirming the similar, but unique, conformations that theseinitiation factors promote in the 40S subunit.

Ternary complex binding induces a 40S structural changeA reasonable explanation for ternary complex–induced stability ofmRNA, irrespective of AUG recognition, may be that there areadditional conformational changes in the 40S subunit induced bythe ternary complex itself. To test this, we again compared rRNAcleavage patterns generated from BABE-Fe–modified eIF3j in thelocation of the 40S subunit beak upon recruitment of the ternarycomplex to the 40S subunit in the absence of other initiation factors.Changes in the cleavage intensities from hydroxyl radicals generatedfrom BABE-Fe tethered to position 152 in eIF3j were apparent (Fig. 5,compare lanes 9 and 10). Cleavage intensities increased markedly atnucleotide positions 1268–1270 (helix 32) and 1279–1284 (helix 33),and increased slightly at nucleotides 1295–1298 (helix 33). These datasuggest a conformational change upon ternary complex recruitment in

C T A G 40S

ED

TA

/Fe

–cys

+E

DT

A/F

e

–cys

+eI

F1/

1A+

ED

TA

/Fe

S15

2C

S15

2C+

eIF

1/1A

S21

7C

S21

7C+

eIF

1/1A

a

1268–1270

1279–1284

1295–1298

1305–1307

1318–1321

C T A G 40S

ED

TA

/Fe

–cys

+E

DT

A/F

e

S15

2C

S21

7C

c

1268–1270

1279–1284

1295–1298

1305–1307

1318–1321

C T A G 40S

ED

TA

/Fe

–cys

+E

DT

A/F

e

–cys

+H

CV

+E

DT

A/F

e

S15

2C

S21

7C

S21

7C+

HC

V∆O

RF

S15

2C+

HC

V∆O

RF

–cys

+eI

F3∆

j+E

DT

A/F

e

S15

2C+

eIF

3∆j

S21

7C+

eIF

3∆j

b

1268–1270

1279–1284

1295–1298

1305–1307

1318–1321

eIF1A

eIF1

40S

eIF3j + eIF3j

eIF3∆j

+

40S

HCV∆ORF

+eIF3j

1 2 3 4 5 6 7 8 9 10 11 12 Lane 1 2 3 4 5 6 7 8 9 10 11 12 Lane 1 2 3 4 5 6 7 8 9 10 11 12 Lane

40S

Figure 4 Effects of eIF1, eIF1A, HCV and eIF3 on directed hydroxyl radical probing of 18S rRNA from BABE-Fe–eIF3j. (a) Primer-extension analysis of 18S

rRNA cleaved by BABE-Fe–modified eIF3j in the absence (lanes 9 and 11) or presence of eIF1 and eIF1A (lanes 10 and 12). (b) Analysis of 18S rRNA

cleaved by BABE-Fe–modified eIF3j in the absence (lanes 9 and 11) or presence of HCVDORF (lanes 10 and 12). (c) Analysis of 18S rRNA cleavage by

BABE-Fe–modified eIF3j in the absence (lanes 9 and 11) or presence of eIF3Dj (lanes 10 and 12). In each gel, the sequencing lanes are indicated (C, T, A

and G). Other lanes include 40S subunits in the absence or presence of EDTA/Fe and mock-derivatized eIF3j (�cys+EDTA/Fe) in the absence or presence of

HCV and other initiation factors, as indicated. Cleavage-nucleotide positions in the 18S rRNA are indicated, and colored circles correspond to the

components added in each reaction, as depicted in the cartoons.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 0 1

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

which nucleotides in the 40S subunit beak move closer to the locationof amino acid 152 in eIF3j (or into a more favorable environment orposition for attack).

Notably, the cleavage intensities generated from eIF3j position 152at nucleotides 1305–1307, or cleavage of nucleotides 1279–1284generated from eIF3j position 217 in helix 33 did not change,suggesting that eIF3j does not move on the 40S subunit. Moreover,analysis of the cleavage pattern and intensity in helix 18 located insidethe mRNA entry channel indicated that this region of the 40S subunitalso does not change in relation to the location of eIF3j upon ternarycomplex recruitment (Supplementary Fig. 2b). These data reveal thatthe ternary complex promotes a structural rearrangement of the beakof the 40S subunit that would be likely to alter the conformation of themRNA binding channel, especially in the A-site of the 40S subunit.This change may contribute to the association of the mRNA with the40S subunit in a manner that is independent of AUG codon recogni-tion by the initiator tRNA. Notably, although a thermodynamicallystable complex between the ternary complex and 40S subunit persistsin the absence of any other initiation factors, it is expected thatother initiation factors are required for accelerating the rate ofcomplex formation.

DISCUSSIONIn this study, we investigated the streamlined initiation mechanismused by the HCV IRES mRNA to examine the mechanism of mRNApositioning in the 40S mRNA binding channel. On the basis ofsite-directed hydroxyl radical probing and toeprinting, we establishedthat eIF3j and HCV IRES mRNA binding to the 40S subunit mRNAentry channel is mutually exclusive. This enabled us to use thedissociation of eIF3j from the mRNA entry channel as an indicator

of mRNA binding to the 40S subunit entry channel. A proposedmodel for the pathway of HCV mRNA association with the 40Ssubunit is presented in Figure 6. In agreement with previous data7,13,the conformational change in the 40S subunit induced by domain II ofthe HCV IRES is required for mRNA positioning in the bindingchannel. This conformational change alone is not sufficient to allowthe HCV mRNA to associate with the entry channel when both eIF3jand eIF3 are present. Instead, eIF3j displacement and consequentassociation of the HCV mRNA with the entry channel require ternarycomplex recruitment to the 40S subunit. Of note, the toeprintingcompetition experiment shown in Figure 2a requires a large molarexcess of eIF3j to compete with HCV mRNA binding to the 40Ssubunit. This is probably required because the HCV mRNA is tetheredto the 40S subunit through domains II and III, increasing its localconcentration, whereas the absence of the interaction with the eIF3complex reduces the local concentration of eIF3j. Although it ispossible that the large excess of eIF3j used for toeprinting may bindnonspecifically to the 40S subunit, this seems unlikely because thephysiological local concentration of eIF3j on the 40S subunit wouldprobably be high owing to its interaction with the eIF3 complex.

Unexpectedly, our data show that the stabilizing effect of the ternarycomplex on HCV mRNA association does not require base-pairingbetween the AUG codon and the anticodon of the initiator tRNA. Thisfinding contradicts toeprinting data that indicated a strong require-ment of the AUG codon for HCV mRNA association with the entrychannel7. However, toeprinting enables detection of only the mostthermodynamically stable complexes, such as those enhanced by thecodon-anticodon interaction. In addition, mutation of the authenticAUG codon reduces protein synthesis by only 50–60% in vitro27. Theremaining activity may be due to the use of the ACG codon situatedtwo codons downstream of the AUG codon. Because toeprinting is notpossible with the mutant AUG construct7, we are not able todetermine whether our construct actually allows a codon-anticodoninteraction to occur at the ACG codon. Therefore, although our data

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Lane

A24

1C+

TC

A24

1CT

235C

+T

CT

235C

S21

7C+

TC

S21

7CS

152C

+T

CS

152C

–cys

+T

C+

ED

TA/F

e–c

ys+

ED

TA/F

eE

DTA

/Fe

40S

C T A G

1268–1270

1279–1284

1295–1298

1305–1307

1318–1321

40S

elF3j

TC

+

Figure 5 The effect of eIF2–Met-tRNAi on directed hydroxyl radical probing

of 18S rRNA with BABE-Fe–eIF3j. Primer-extension analysis of 18S rRNA

cleaved by BABE-Fe–modified eIF3j. The sequencing lanes are indicated

(C, T, A and G). Other lanes include 40S subunits in the absence or

presence of EDTA/Fe and mock-derivatized eIF3j (�cys+EDTA/Fe) in

the absence (lane7) or presence of eIF2-Met-tRNAi (TC; lane 8). Lanes

corresponding to eIF3j derivatized with BABE-Fe at the positions indicated

in the absence (lanes 9, 11, 13 and 15), or presence (lanes 10, 12, 14 and

16), of TC are indicated. Nucleotide positions of cleavage sites in the 18S

rRNA are indicated, and colored circles correspond to the components

added in each reaction, as depicted in the cartoon.

3′ 3′

3′5′5′5′

GUAGUA GUA

Domain II–induced conformation

(mRNA entry channel opening)

elF2–GTP–Met-tRNAi(elF3j dissociation

from entry channel)

Figure 6 A model for HCV IRES association with the mRNA binding channel

of the 40S subunit. Following the association of the HCV IRES with the 40S

subunit, domain II is required to promote an open conformation of the

mRNA entry channel. The stable association of eIF3 with this complexplaces eIF3j in the mRNA entry channel, preventing the stable binding of

HCV mRNA with the A-site and entry channel. The subsequent recruitment

of eIF2–Met-tRNAi is necessary to shift the equilibrium to favor the stability

of HCV mRNA in the entry channel over that of eIF3j.

ART IC L E S

40 2 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

suggest that AUG recognition by the initiator tRNA is not required foreIF3j dissociation in our assay, we cannot rule out the weakerinteraction between the initiator tRNA and the ACG codon, whichmay be essential for mRNA stabilization and eIF3j dissociation.Notably, recent data indicate that eIF3j can dissociate mRNA fromthe 40S subunit during ribosome recycling but only in the presence ofeIF1 (ref. 20). As eIF1 does not affect the affinity of eIF3j binding tothe 40S subunit19, this is probably due to the effect of eIF1 indissociating the P-site–bound tRNA, which would subsequently shiftthe equilibrium toward eIF3j binding to the entry channel over that ofmRNA. However, a detailed kinetics analysis is required before the roleof eIF3j in mRNA dissociation can be made, as it is plausible that eIF3jinfluences only mRNA association.

In this study, we also investigated the mRNA binding propensitiesof different 40S subunit conformations induced upon association ofthe HCV IRES or initiation factors. In agreement with cryo-EMstudies13,14, our data provide evidence that the HCV IRES or eIF1and eIF1A promote similar but distinct conformational changes in the40S subunit. Notably, we show that the eIF3 complex also induces aconformational change in the 40S subunit that has some similarities tothose induced by the HCV mRNA or eIF1 and eIF1A. Such eIF3-mediated effects on the 40S subunit may provide an explanation for itsinvolement in mRNA recruitment to the 40S subunit in addition to itsrole in recruiting the cap binding complex in vitro and in vivo8,9,28.A recent report also suggested a possible conformational change in themRNA entry channel when eIF3 associates with the 40S subunit29.However, those experiments were completed in the absence of eIF3jbut instead contained poly(U) to provide affinity for the eIF3 complexto the 40S subunit. Therefore, it is not possible to determine fromthose experiments what relative effects eIF3 and poly(U) have on thestructure of the 40S subunit.

Ternary complex association with the 40S subunit without othercomponents induces a conformational change in the head of the 40Ssubunit, which may have a role in promoting eIF3j dissociation and/ormRNA stability with the entry channel. Notably, this conformationalchange is not sufficient to promote HCV mRNA association with theentry channel because the ternary complex cannot stabilize HCVmRNA in the entry channel in the absence of domain II. Therefore,because eIF1, eIF1A, eIF3 and the ternary complex enable AUGrecognition on an unstructured mRNA30, it is tempting to speculatethat eIF1 and eIF1A probably provide the necessary conformationalchange in the 40S subunit to allow mRNA positioning in the bindingchannel, as proposed14. On the basis of these findings, we propose thateither the HCV IRES, or eIF1 and eIF1A, can induce a similarconformational state of the 40S subunit that favors mRNA loadinginto the decoding channel with consequential displacement of eIF3jupon ternary complex binding. This conclusion explains how theHCV IRES functionally replaces eIF1 and eIF1A during translationinitiation on the human ribosome.

METHODSSample purification and modification. We purified human eIF1, eIF1A, eIF2,

eIF3j, eIFD3j and 40S ribosomal subunits as described19,31. Initiator tRNA was

transcribed in vitro, and purified and charged in vitro using a purified tRNA

synthetase, as previously described19,32. We conjugated bromoacetamidobenzyl-

EDTA/Fe (BABE-Fe; Dojindo Molecular Technologies) to single-cysteine eIF3j

proteins according to a published protocol19,33. All HCV mRNA numbering is

according to a previous publication34. Wild-type HCV mRNA (40–372) and

HCVDII (120–372) used in probing experiments were prepared as described34.

We generated the derivative HCV construct (40–344; HCVDORF) using the

QuikChange mutagenesis kit (Stratagene). The HCV construct used for

toeprinting experiments included the firefly luciferase open reading frame

cloned into the wild-type HCV mRNA between the BamHI and HindIII

restriction sites located in the wild-type HCV construct34. This resulted in a

short nucleotide linker (5¢-GGATCCTC-3¢) following nucleotide 372 of the

original construct before the ATG of the luciferase gene. The HCV construct

containing the mutated AUG codon was generated using QuikChange muta-

genesis of the HCV construct, including the firefly luciferase open reading

frame. The resulting construct mutates the initiation codon 5¢-ATCATG-3¢ to

5¢-AAAAAA-3¢. We verified all constructs by sequencing, and we produced

HCV RNA by in vitro transcription and purified it by denaturing acrylamide

gel electrophoresis as described34.

Directed hydroxyl radical probing. We formed complexes containing either

mock-derivatized eIF3j (-Cys) or BABE-Fe–eIF3j bound to salt-washed 40S

subunits and carried out radical probing as described19,35,36. Specifically, each

probing reaction was carried out in 50-ml incubations in buffer A (50 mM

HEPES, pH 7.5, 50 mM KCl, 2 mM magnesium acetate). Reactions contained

16 pmols (320 nM) 40S subunits, 72 pmols (1.44 mM) eIF1 and eIF1A,

35 pmols (700 nM) eIF3j, 23 pmols eIF3Dj (460 nM) and 72 pmols HCV

mRNA (1.44 mM), as indicated in the figure legends. For reactions containing

eIF2 and charged initiator tRNA, we first incubated eIF2 with 1 mM GMP-PNP

(Sigma-Aldrich) in buffer A for 5 min at 30 1C. We then added a two-fold

excess of charged initiator tRNA in buffer A supplemented with a final

free concentration of 1 mM magnesium acetate, and incubated the reaction

at 30 1C for 10 min. Subsequent probing reactions included 30 pmols

eIF2–GMP-PNP and 60 pmols initiator tRNA. We detected 18S rRNA cleavage

by BABE-Fe–eIF3j using reverse transcription and denaturing gel electrophor-

esis as described19,35,36.

Toeprinting. We determined the position of 40S subunits on the HCV mRNA

by a primer-extension inhibition assay (toeprinting), as described21–23,37 with

minor modifications. 40S subunits (16 pmols; 400 nM), HCV-luciferase mRNA

(6 pmols; 150 nM) and 10 pmols (250 nM) of a 5¢ end–labeled 32P-labeled

primer (5¢-GCGCCGGGCCTTTCTTTATG-3¢) complementary to nucleotides

18–37 of firefly luciferase were incubated in 40 ml reactions in buffer

A supplemented with 1 mM DTT. Reactions were incubated at 37 1C for

10 min and then placed on ice for 5 min. We then added 4 ml of 10� extension

mix (80 mM magnesium acetate, 10 mM DTT, 10 mM dNTPs, 5 U SuperScript

III reverse transcriptase (Invitrogen)) and incubated the reaction at 30 1C for 15

min. The reaction was then cooled on ice, followed by RNA extraction and

analysis by denaturing gel electrophoresis as described19.

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ACKNOWLEDGMENTSWe thank D. King at the University of California, Berkeley, for expert MS analysisof modified proteins. We gratefully acknowledge members of the Doudnalaboratory for discussions and comments on the manuscript. In particular, wewould like to thank R. Spanggord for advice on hydroxyl radical probing andF. Siu for advice on RNA transcription protocols. This work was supported inpart by a grant from the US National Institutes of Health to J.A.D. and J.W.B.H.

AUTHOR CONTRIBUTIONSC.S.F. performed the experiments; C.S.F., J.W.B.H. and J.A.D. designedexperiments and wrote the manuscript.

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Pestova, T.V., Lorsch, J.R. & Hellen, C.U.T. The mechanism of translation initiation ineukaryotes. in Translational Control in Biology and Medicine (eds. Mathews, M.B.,Sonenberg, N. & Hershey, J.W.B.) 87–128 (Cold Spring Harbor Laboratory Press, ColdSpring Harbor, NY, 2007).

2. Fraser, C.S. & Doudna, J.A. Quantitative studies of ribosome conformational dynamics.Q. Rev. Biophys. 40, 163–189 (2007).

3. Doudna, J.A. & Sarnow, P. Translation initiation by viral internal ribosome entry sites. inTranslational Control in Biology and Medicine (eds. Mathews, M.B., Sonenberg, N. &Hershey, J.W.B.) 129–153 (Cold Spring Harbor Laboratory Press, Cold Spring Harbor,NY, 2007).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 0 3

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

4. Elroy-Stein, O. & Merrick, W.C. translation initiation via cellular internal ribosome entrysites. in Translational Control in Biology and Medicine (eds. Mathews, M.B., Sonen-berg, N. & Hershey, J.W.B.) 155–172 (Cold Spring Harbor Laboratory Press, ColdSpring Harbor, NY, 2007).

5. Pisarev, A.V., Shirokikh, N.E. & Hellen, C.U. Translation initiation by factor-indepen-dent binding of eukaryotic ribosomes to internal ribosomal entry sites. C. R. Biol. 328,589–605 (2005).

6. Fraser, C.S. & Doudna, J.A. Structural and mechanistic insights into hepatitis C viraltranslation initiation. Nat. Rev. Microbiol. 5, 29–38 (2007).

7. Pestova, T.V., Shatsky, I.N., Fletcher, S.P., Jackson, R.J. & Hellen, C.U. A prokaryotic-like mode of cytoplasmic eukaryotic ribosome binding to the initiation codon duringinternal translation initiation of hepatitis C and classical swine fever virus RNAs. GenesDev. 12, 67–83 (1998).

8. Trachsel, H., Erni, B., Schreier, M.H. & Staehelin, T. Initiation of mammalian proteinsynthesis. II. The assembly of the initiation complex with purified initiation factors.J. Mol. Biol. 116, 755–767 (1977).

9. Benne, R. & Hershey, J.W. The mechanism of action of protein synthesis initiationfactors from rabbit reticulocytes. J. Biol. Chem. 253, 3078–3087 (1978).

10. Ji, H., Fraser, C.S., Yu, Y., Leary, J. & Doudna, J.A. Coordinated assembly of humantranslation initiation complexes by the hepatitis C virus internal ribosome entry siteRNA. Proc. Natl. Acad. Sci. USA 101, 16990–16995 (2004).

11. Spahn, C.M. et al. Structure of the 80S ribosome from Saccharomyces cerevisiae–tRNA-ribosome and subunit-subunit interactions. Cell 107, 373–386 (2001).

12. Taylor, D.J., Frank, J. & Kinzy, T.G. Structure and function of the eukaryotic ribosomeand elongation fractors. in Translational Control in Biology and Medicine(eds. Mathews, M.B., Sonenberg, N. & Hershey, J.W.B.) pp. 59–85 (Cold SpringHarbor Laboratory Press, Cold Spring Harbor, NY, 2007).

13. Spahn, C.M. et al. Hepatitis C virus IRES RNA-induced changes in the conformation ofthe 40S ribosomal subunit. Science 291, 1959–1962 (2001).

14. Passmore, L.A. et al. The eukaryotic translation initiation factors eIF1 and eIF1Ainduce an open conformation of the 40S ribosome. Mol. Cell 26, 41–50 (2007).

15. Kolupaeva, V.G., Pestova, T.V. & Hellen, C.U. An enzymatic footprinting analysis of theinteraction of 40S ribosomal subunits with the internal ribosomal entry site of hepatitisC virus. J. Virol. 74, 6242–6250 (2000).

16. Otto, G.A. & Puglisi, J.D. The pathway of HCV IRES-mediated translation initiation.Cell 119, 369–380 (2004).

17. Spahn, C.M. et al. Cryo-EM visualization of a viral internal ribosome entry site bound tohuman ribosomes: the IRES functions as an RNA-based translation factor. Cell 118,465–475 (2004).

18. Unbehaun, A., Borukhov, S.I., Hellen, C.U. & Pestova, T.V. Release of initiation factorsfrom 48S complexes during ribosomal subunit joining and the link between establish-ment of codon-anticodon base-pairing and hydrolysis of eIF2-bound GTP. Genes Dev.18, 3078–3093 (2004).

19. Fraser, C.S., Berry, K.E., Hershey, J.W. & Doudna, J.A. eIF3j is located in the decodingcenter of the human 40S ribosomal subunit. Mol. Cell 26, 811–819 (2007).

20. Pisarev, A.V., Hellen, C.U. & Pestova, T.V. Recycling of eukaryotic postterminationribosomal complexes. Cell 131, 286–299 (2007).

21. Hartz, D., McPheeters, D.S., Traut, R. & Gold, L. Extension inhibition analysis oftranslation initiation complexes. Methods Enzymol. 164, 419–425 (1988).

22. Anthony, D.D. & Merrick, W.C. Analysis of 40 S and 80 S complexes with mRNA asmeasured by sucrose density gradients and primer extension inhibition. J. Biol. Chem.267, 1554–1562 (1992).

23. Kozak, M. Primer extension analysis of eukaryotic ribosome-mRNA complexes. NucleicAcids Res. 26, 4853–4859 (1998).

24. Kieft, J.S., Zhou, K., Jubin, R. & Doudna, J.A. Mechanism of ribosome recruitment byhepatitis C IRES RNA. RNA 7, 194–206 (2001).

25. ElAntak, L., Tzakos, A.G., Locker, N. & Lukavsky, P.J. Structure of eIF3b RNArecognition motif and its interaction with eIF3j: structural insights into the recruitmentof eIF3b to the 40 S ribosomal subunit. J. Biol. Chem. 282, 8165–8174 (2007).

26. Joseph, S. & Noller, H.F. Directed hydroxyl radical probing using iron(II) tethered toRNA. Methods Enzymol. 318, 175–190 (2000).

27. Reynolds, J.E. et al. Unique features of internal initiation of hepatitis C virus RNAtranslation. EMBO J. 14, 6010–6020 (1995).

28. Jivotovskaya, A.V., Valasek, L., Hinnebusch, A.G. & Nielsen, K.H. Eukaryotic transla-tion initiation factor 3 (eIF3) and eIF2 can promote mRNA binding to 40S subunitsindependently of eIF4G in yeast. Mol. Cell. Biol. 26, 1355–1372 (2006).

29. Pisarev, A.V., Kolupaeva, V.G., Yusupov, M.M., Hellen, C.U. & Pestova, T.V. Ribosomalposition and contacts of mRNA in eukaryotic translation initiation complexes. EMBO J.27, 1609–1621 (2008).

30. Pestova, T.V. & Kolupaeva, V.G. The roles of individual eukaryotic translation initiationfactors in ribosomal scanning and initiation codon selection. Genes Dev. 16,2906–2922 (2002).

31. Fraser, C.S. et al. The j-subunit of human translation initiation factor eIF3 is requiredfor the stable binding of eIF3 and its subcomplexes to 40 S ribosomal subunits in vitro.J. Biol. Chem. 279, 8946–8956 (2004).

32. Pestova, T.V. & Hellen, C.U. Preparation and activity of synthetic unmodified mamma-lian tRNAi(Met) in initiation of translation in vitro. RNA 7, 1496–1505 (2001).

33. Spanggord, R.J., Siu, F., Ke, A. & Doudna, J.A. RNA-mediated interaction between thepeptide-binding and GTPase domains of the signal recognition particle. Nat. Struct.Mol. Biol. 12, 1116–1122 (2005).

34. Kieft, J.S. et al. The hepatitis C virus internal ribosome entry site adopts an ion-dependent tertiary fold. J. Mol. Biol. 292, 513–529 (1999).

35. Lomakin, I.B., Kolupaeva, V.G., Marintchev, A., Wagner, G. & Pestova, T.V. Position ofeukaryotic initiation factor eIF1 on the 40S ribosomal subunit determined by directedhydroxyl radical probing. Genes Dev. 17, 2786–2797 (2003).

36. Culver, G.M. & Noller, H.F. Directed hydroxyl radical probing of RNA from iron(II)tethered to proteins in ribonucleoprotein complexes. Methods Enzymol. 318,461–475 (2000).

37. Pisarev, A.V., Unbehaun, A., Hellen, C.U. & Pestova, T.V. Assembly and analysis ofeukaryotic translation initiation complexes. Methods Enzymol. 430, 147–177 (2007).

ART IC L E S

40 4 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Tertiary interactions within the ribosomal exit tunnelAndrey Kosolapov & Carol Deutsch

Although tertiary folding of whole protein domains is prohibited by the cramped dimensions of the ribosomal tunnel, dynamictertiary interactions may permit folding of small elementary units within the tunnel. To probe this possibility, we used a b-hairpinand an a-helical hairpin from the cytosolic N terminus of a voltage-gated potassium channel and determined a probability offolding for each at defined locations inside and outside the tunnel. Minimalist tertiary structures can form near the exit port ofthe tunnel, a region that provides an entropic window for initial exploration of local peptide conformations. Tertiary subdomainsof the nascent peptide fold sequentially, but not independently, during translation. These studies offer an approach for diagnosingthe molecular basis for folding defects that lead to protein malfunction and provide insight into the role of the ribosome duringearly potassium channel biogenesis.

Protein folding begins with the birth of the peptide within theribosomal tunnel1–8, but the dimensions of this tunnel, 100 A inlength and 10–20 A in width9–12, limit the dimensions of a foldedpeptide that can fit inside the tunnel. For example, the completefolding of a tertiary monomeric functional T1 domain of thevoltage-gated potassium channel Kv1.3, which is approximately27 � 27 � 25 A13,14 (Fig. 1a), is precluded1. However, we may askwhether smaller subdomains can fold inside the ribosomal tunnel.Although compact secondary structures form within the tunnel1–8,including regions of the T1 domain1, subdomain formation inside thetunnel has not been investigated explicitly. The elementary tertiaryfolding unit has not been established.

The T1 domain is in the cytoplasmic N terminus of voltage-gatedpotassium (Kv) channels. It is a recognition domain with a crucial rolein the folding and oligomerization of Kv channel proteins, enablingformation of the tetrameric channel15–17, its axonal targeting18 andpossibly the opening and closing of the permeation pathway forpotassium ions13,19–21. We have previously shown that T1 domainstetramerize while still attached to ribosomes22 and that the whole T1domain folds into its tertiary structure only after the primary sequenceemerges from the ribosomal tunnel and the linker between T1 and thefirst transmembrane segment has been synthesized1.

T1 comprises several subdomains, including a b-hairpin, b1-b2,and an a-helical hairpin, a4-a5 (Fig. 1). Can these subdomains foldinside the ribosomal tunnel? In this paper, we assess the extent andrelative ease of T1 subdomain folding and the compartment inwhich this folding occurs. We also address a key mechanisticissue in protein folding in general, namely, whether subdomainsof nascent peptides fold sequentially in the tunnel or in a concer-ted fashion upon emergence, and whether the folding of onenascent subdomain depends on the synthesis and/or folding ofanother subdomain.

Here we establish that tertiary intrapeptide interactions occurwithin the tunnel and that stable hairpins form upon emergencefrom the tunnel. In addition, folding of a C-terminal hairpin dependson the presence and status of the N-terminal b-hairpin. Moreover, thenascent peptide–tunnel complex may be more dynamic than pre-viously thought, and a distal portion of the tunnel (near the exit port)provides an entropic window for exploration of conformational spaceby the nascent peptide.

RESULTSSubdomain foldingTo study elementary tertiary folding of T1 subdomains at differentstages of biogenesis, we chose two subdomains: a pair of antiparallelb-strands, a b-hairpin and an a-helical hairpin (Fig. 1, red and blue,respectively). According to the T1 crystal structure of Kv1.2 (ref. 13),which is 95% identical in sequence to the T1 domain of human Kv1.3,these subdomains are located at the N- and C-terminal regions,respectively, of the T1 domain (Fig. 1a). We engineered a pair ofcysteines (Cys53 and Cys66, denoted 53C66C) into the b-hairpin anda different pair (Cys125 and Cys149, denoted 125C149C) into the a-helical hairpin. In each case, the cysteines are exposed in the foldedmonomer and are thus available to cross-linking reagents under theconditions of our assay1,23. Moreover, these cysteines are predicted tobe within 4–6 A of each other if the subdomain is folded similarly tothat modeled in the crystal structure of the mature T1 domain13,14.

To cross-link a cysteine pair, we used orthophenyldimaleimide(PDM), a bifunctional cross-linking reagent with an intermaleimidedistance of B6 A. If the engineered cysteines in each of the T1subdomains shown in Figure 1b come within 4–7 A of each other,then they can be cross-linked with PDM. To distinguish cross-linkedfrom non–cross-linked products, we use methoxypolyethyleneglycol maleimide (PEG-MAL) and methoxypolyethylene glycol

Received 26 June 2008; accepted 30 January 2009; published online 8 March 2009; doi:10.1038/nsmb.1571

Department of Physiology, University of Pennsylvania, Philadelphia, Pennsylvania 19104-6085, USA. Correspondence should be addressed to C.D.(cjd@mail.med.upenn.edu).

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 0 5

ART IC L E S

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

thiol (PEG-SH). Addition of the PEG moiety to a protein (pegylation)shifts its molecular mass by Z10 kDa. A free SH group can be labeledwith PEG-MAL24, and a free peptidyl-maleimide can be labeled withPEG-SH23. A cross-linked subdomain will fail to undergo a gel shiftwith either pegylation reagent, PEG-MAL or PEG-SH. A probability ofcross-linking, Pxlink, can be calculated, as described1. We consider thatthe extent of cross-linking reflects tertiary interactions and, dependingon the location of the engineered cysteine pairs, the probability oftertiary folding.

To determine the cellular compartment (for example, ribosomaltunnel versus cytosol) in which subdomain folding occurs, wegenerated double-cysteine constructs as nascent peptides (still attached

to the ribosome) at various distances from the peptidyl transferasecenter (PTC; Fig. 1b) using different enzyme restriction sites (Sup-plementary Table 1 online) that eliminate the stop codon. Thisproduces different chain lengths between the PTC and the cysteineat the C terminus of each subdomain. The cross-linking assay wasperformed on each of these double-cysteine constructs. In addition,similar assays were carried out on constructs containing only a singlecysteine of each cysteine pair to calculate the probability of cross-linking, Pxlink

1 (see Methods). Each mRNA for the engineered nascentpeptide was translated in a cell-free, membrane-free rabbit reticulocytesystem with added 35S-methionine and then subjected to the cross-linking assay1 (see Methods). After labeling with PDM, samples werefirst treated with SDS and then pegylated to assay for either freeavailable cysteines or free available peptidylmaleimides.Figure 2a shows data for the cross-linking assays of the b-hairpin

(gels in columns 1–3) and the a-hairpin (gels in columns 4–6) fromconstructs with different numbers of amino acids between the PTCand the C-terminal cysteine, DPTC (calculated as the number of thelast residue included by the restriction site minus the residue numberof the engineered cysteine; for example, 91 – 66 ¼ 25 for the firstconstruct shown in Fig. 2a). Regardless of the DPTC and numberof cysteines in the construct, both N- and C-terminal cysteines wereavailable for pegylation (lanes 1) and were completely labeled by PDM(lanes 2). Moreover, all PDM-modified single-cysteine constructsshow a similar efficiency of labeling with PEG-SH, regardless ofDPTC (lane 3 for all single-cysteine constructs). For a DPTC of 52(outside tunnel), both the b-hairpin and the a-hairpin double-cysteine nascent peptides that were modified with PDM producednegligible gel shifts with PEG-SH (lanes 3, second rows, columns 1and 4, respectively). This indicates that no free peptidylmaleimideswere present—that is, the cysteines are cross-linked and subdomainsare mostly folded. In contrast, double-cysteine peptides with DPTC of19 residues or 25 residues (deeper in the tunnel) were not cross-linked

Glu149a b

Gln66

Arg53

NH2

NH2

125 149

53 66

β1

α4 α5

β2∆ PTC

∆ PTC

lle125

α4α5

β1

β2

Figure 1 The T1 domain and experimental design. (a) Hairpin structures.

The monomeric T1 domain (taken from ref. 13 for Kv1.2) is shown with

the indicated subdomains: a b-hairpin (red) comprised of b1 and b2, and ana-helical hairpin (blue) comprised of a4 and a5. Hairpin terminal residues

Arg53, Gln66, Ile125 and Glu149 are shown as space-filling atoms and

are equivalent to residues 34, 47, 106 and 130, respectively, at the

homologous hairpin termini in Kv1.2. (b) Amino acid sequence of the T1

Kv1.3 b-hairpin (red) and a-helical hairpin (blue) with secondary-structure

assignments derived from the highly identical Kv1.2 structure in a (ref. 14).

Engineered cysteines 53C and 66C (b-hairpin) and 125C and 149C

(a-helical hairpin) in Kv1.3 are highlighted in yellow. The PTC is indicated

by the vertical black bar at the right.

2

53C66Ca bPEG-MAL

PEG-SHPDM

∆PTC 25

∆PTC 52

∆PTC 19

∆PTC 52

∆PTC 322 ∆PTC 239

∆PTC 25 ∆PTC 19

PEG-MAL

2

2 3

1

1 2 31 2 31

2 31 2 312 3

66C

66 1

0

1

0

1

0

45

31

22

31

22

66 1

045

149C

β-hairpin α-hairpin

1

1

0

1

00

2

1

0

1

0

1

0

PEG-SHPDM

53C 66C 125C149C 125C 149C

1

0

2

2 3

1

1 2 31 2 3

+ ++++

–––

–+ +

+++

–––

–+ +

+++

–––

–+ +

+++

–––

–+ +

+++

–––

–+ +

+++

–––

–

1

2 31 2 31 2 31

0

1

0

1

0

1

0

1

0

Figure 2 Cross-linking and accessibility assays for b- and a-hairpins. (a) Intramolecular cross-linking assay for indicated cysteine pairs (gels in columns 1

and 4) and single cysteines (gels in columns 2,3,5 and 6) for b-hairpin (gels in columns 1–3) and a-hairpin (gels in columns 4–6). The number of amino

acids from the PTC to the C-terminal cysteine of the pair, or to the single cysteine, is indicated to the left of each row of gels as the DPTC. Engineered

restriction sites XbaI@91 and KpnI@118 were used to produce DPTCs of 25 and 52, respectively, for the b-hairpin constructs. An engineered restriction

site, BstEII@168, and a native site, XbaI@201, were used to produce DPTCs of 19 and 52, respectively, for the a-hairpin constructs. Numbers to the right

of each gel represent doubly (2) or singly (1) pegylated or unpegylated (0) protein. In all cases, the background in lane 2 is appreciably less than that of the

other lanes owing to 495% labeling of the peptide with PDM, which leads to unpegylated protein that accumulates entirely in band 0. (b) Accessibility

assays for C-terminal cysteines. Nascent peptides were pegylated (see Methods) for 4 h and 6 h (lanes 1 and 2, respectively) for 66C (left column of gels)

or 149C (right column of gels) in constructs with the indicated number of amino acids from the PTC to the C-terminal cysteine (DPTC). Long peptides

(DPTC ¼ 322 and 239) residing outside the tunnel are shown in the top row of gels; short peptides (DPTC ¼ 25 and 19) residing within the tunnel are

shown in the bottom row of gels. For the b-hairpin PTC 25 construct, the 4- and 6-hour samples are shown in separate panels to indicate that each was

fractionated on a different gel under identical electrophoresis conditions. Numbers to the left of each gel are molecular weight standards; numbers to the

right of each gel represent singly (1) pegylated or unpegylated (0) protein.

ART IC L E S

40 6 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

and therefore not folded. In this case, the PDM-treated peptidescontained free peptidylmaleimides that were pegylated with PEG-SHto produce gel shifts (bands 1 and 2 in lanes 3, first row, columns 1and 4). Single-cysteine constructs were not cross-linked, yet weremodified by PDM to contain a free maleimide that was pegylated byPEG-SH (band 1 in lanes 3, first row, columns 2, 3, 5 and 6). Cross-linking data for additional constructs with DPTC values of 27 and 39for the b-hairpin and 24 and 29 for the a-hairpin are shown inSupplementary Figure 1 online.

Location of the subdomain in the tunnelThe interpretation of these cross-linking data requires that we knowthe location of the subdomain, inside or outside of the tunnel, whichdepends on both DPTC and the secondary structure of the peptidebetween the C-terminal cysteine of the subdomain and the PTC. Todetermine the subdomain location, we used an accessibility assay(Fig. 2b and Supplementary Fig. 2 online). This assay also relies on amass-tagging strategy using PEG-MAL2,24, but without the SDS pre-treatment used in the cross-linking assay (see Methods)1. The finalextent of PEG-MAL labeling of a cysteine in the last 20 A of the tunnelis monotonically dependent on the distance of the modifiable cysteinefrom the PTC2. Accessibility is therefore an accurate estimator ofdistance from the PTC2 (see Discussion). For each subdomain, weengineered a single cysteine, corresponding to the C-terminal cysteineof each cysteine pair, into the nascent peptide, which was treated withPEG-MAL, and determined the final extent of labeling. For b- anda-hairpins located outside the ribosomal tunnel—that is, DPTC is 322and 239, respectively (top gels)—the fraction of pegylated protein isZ0.88 (Fig. 2b). In contrast, the fraction of pegylated protein isr0.14 for constructs where the C-terminal cysteines of the b- anda-subdomains were placed at 25 residues and 19 residues, respectively,from the PTC (bottom gels). These cysteines, located inside thetunnel, are 80 A or less from the PTC2.

Relationship between subdomain location and foldingTo determine the tunnel location where hairpin folding occurs, wecombined the results from the cross-linking and accessibility assays for

each hairpin. In Figure 3a, we plot the probability of cross-linking(Pxlink) against the fraction of accessible peptide (Facc). The midpointof each PF curve, F50, is the fraction accessible when Pxlink is 50%of the maximum (dotted lines). Thus, we can compare Pxlink atdifferent distances from the PTC and the relative ease of foldingfor each subdomain at the same location. Both subdomains appa-rently begin to fold before they emerge completely from the ribo-somal tunnel, and maximum folding occurs after the subdomainsemerge from the tunnel. Pxlink saturates below 100% completionowing to rapid binding of PDM molecules to each cysteine23. TheF50 values are 0.57 ± 0.10 for the b-hairpin and 0.37 ± 0.06 for thea-hairpin, values that are not significantly different (P ¼ 0.075,Z-test), suggesting that the relative ease of folding of the two differenthairpins is similar.

Is subdomain folding affected by the interaction of the ribosomaltunnel with the subdomain, or is it due solely to the intrinsicproperties of the peptide? To address this question, we first releasedthe nascent peptides from the ribosome using RNase, isolated thereleased peptide and assayed it for cross-linking. A nascent peptide inwhich the C terminus of the b-hairpin sequence was located 25residues from the PTC was released and gave nearly maximal cross-linking for the hairpin (Supplementary Fig. 3 online), suggesting thatit folded completely. In contrast, a released peptide in which the Cterminus of the a-hairpin sequence was 19 residues from the PTCcross-linked only 55% of the maximum value attained while attachedto the ribosome. Notably, when this sequence was first positionedmore distally in the tunnel, for example, 24 residues and 29 residuesfrom the PTC, and then released, the a-hairpin apparently foldedmore completely, at 71% and 76% of the maximum, respectively.This behavior suggests that several factors may determine the prob-ability of folding of the a-hairpin: the location of the a-hairpinsequence in the tunnel before release (so that the tunnel promotes afolding-competent state of the peptide), the total length of the nascentpeptide, and/or the number and nature of the residues between thehairpin C terminus and the PTC. In addition, the maximally foldeda-hairpin present in the crystal structure may represent only one ofseveral hairpin structures. This a-hairpin behavior contrasts with the

1.0a

b

1.0

0.8

α βP x

link

P x

link

0.8

0.6

0.6F acc

0.4

0.4

0.2

0.2

0.0

1.062C75C

62C75C

∆PTC > 285

59C72C86C99C

101C125C

101C125C

0.8

0.6

0.4

0.2

0.0

0.0

1.00.80.6F acc

0.40.20.0

Figure 3 Accessibility-dependent probability of cross-linking. (a) Probability

of cross-linking, Pxlink, as a function of the fraction of accessible peptide,

Facc. Data are derived from cross-linking and accessibility assays

(exemplified in Figure 2) for b-hairpins (red squares) and a-helical hairpins

(blue triangles) and fit to a sigmoidal function with two parameters

(SigmaPlot 8.0). Data are means ± s.e.m. for at least triplicate

measurements. The dotted lines indicate F50 values as defined in the text.

The b-hairpin has F50 ¼ 0.37 ± 0.06, whereas the a-helical hairpin has F50

¼ 0.57 ± 0.10. These midpoints are not significantly different (P ¼ 0.075,

Z-test). The DPTC values are (increasing order of Facc) 21, 25, 27, 29, 39,

322, 52 and 86 (last two occur at the same Facc) for the b-hairpin and 3,

19, 24, 29, 40, 52 and 239 for the a-hairpin. (b) Probability of cross-

linking of 62C75C (red squares) and 101C125C (blue triangles). A curve

was drawn through all the data points, but no particular function is

intended. The shaded region indicates a region in which peptides can becross-linked (an entropic window). All filled symbols represent cysteine pairs

that lie far outside the tunnel on a long tether (DPTC values 4285). Pxlink

for these constructs is 0.2–0.3. Pxlink for both 86C99C (filled green circle)

and 59C72C (filled inverted black triangle) is B0.3, in agreement with

62C75C (filled red square) at this same location. The DPTC values are

(increasing order of Facc) 30, 35, 37, 43, 67 and 313 for the 62C75C

constructs and 27, 38, 41, 43, 48 and 263 for the 101C125C constructs.

Data are means ± average error or ± s.e.m. for two to four replicate samples,

except for points at DPTC 48, 67 and the 86C99C construct (green filled

circle) at DPTC 289. For all other points, errors are either clearly visible or

within the symbol.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 0 7

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

folding requirements for the b-hairpin, which is constrained by thetunnel but apparently folds completely once it is released.

Two additional conclusions may be drawn from the PF curvesshown in Figure 3a. First, folding of the upstream b-hairpin doesnot require the presence and/or folding of the a-hairpin, as evidencedby maximum folding for the DPTC 52 peptide, which lacks thea-hairpin. Second, tertiary folding of both hairpins is prohibited inshort nascent intermediates with relatively inaccessible C-terminalcysteines (DPTC o 21). However, the a-hairpin shows a signi-ficant amount of folding (P ¼ 0.04, Pxlink ¼ 0.26, 34% of itsmaximum) when DPTC ¼ 24, that is, when the C-terminal cysteineis inside the tunnel (Facc ¼ 0.25), suggesting a dynamic equilibriumbetween folded and unfolded species (Supplementary Results online).

Evidence for an entropic windowThe PF curves shown in Figure 3a make two predictions. First, atlocations far outside the tunnel (DPTC 4 239), the hairpin sub-domains, as well as the whole T1 domain1, are folded. We cantherefore predict from the crystal structure of the T1 domain whichside chains in the folded T1 are too distant from one another to beintramolecularly cross-linked (414 A1) and should therefore show alow Pxlink. Second, at tunnel locations within 80 A of the PTC, nofolding occurs. Thus, pairs of distant cysteines engineered anywherealong the nascent peptide, including nonhairpin sequences, shouldalso show a low Pxlink within these 80 A. Each prediction derives froma different constraint on tertiary folding. In the first case, the wholestably folded T1 domain has a network of intramolecular interactionsthat restricts its mobility, decreasing the degrees of freedom allowedfor any given stretch of primary sequence within the folded T1. In thesecond case, the tunnel itself is too small to allow intramoleculartertiary interactions in the nascent peptide. To test these predictions,we designed two separate double-cysteine constructs, 62C75C and101C125C, in the region adjacent to the b-hairpin and a-hairpin,respectively. The cysteines in the 62C75C and 101C125C pairs areseparated by the same number of amino acids, respectively, as in the53C66C (b-hairpin) and 125C149C (a-hairpin) pairs. In contrast tointrapair proximities (6–7 A) for 53C66C and 125C149C, both62C75C and 101C125C are farther apart in the folded T1 monomer,B25 A and B17 A, respectively.

As predicted, cross-linking efficiencies for 62C75C (Fig. 3b, redsquares) and 101C125C (blue triangles) are low when these constructswere placed either deep inside or completely outside of the tunnel(Facc r 0.5 or Z 0.9, respectively), consistent with the lack of cross-linking reported for two cysteines located 14 A apart in the wholefolded T1 domain1.

However, between these two regions of restricted conforma-tional entropy there is a region where some cross-linking ispermitted (Fig. 3b, shaded area), even though in each case theintroduced cysteines within the pair are distant from one anotherin the mature T1 structure. We used chain lengths that position62C75C and 101C125C in the more distal regions of the tunnel, atDPTC values of 30, 35, 37, 43, 67, 313 and 27, 38, 41, 43, 48, 263,respectively. Short constructs (Fig. 3b, open symbols) with Facc of0.6–0.95 cross-linked to give a Pxlink of Z 0.6, suggesting that thisregion of the tunnel is dynamic and allows the peptide to exploreconformational space, therefore constituting an entropic window(shaded region) for sampling potential folding partners. For theseconstructs, both accessibility measurements and calculations basedon secondary structure between the C-terminal cysteine and thePTC indicate that these C-terminal cysteines are located either justwithin the tunnel at the exit port or outside in the immediatevicinity of the tunnel. To confirm findings in the restricted region,we engineered two constructs with DPTC 4 285. We engineered86C99C between the b- and a-hairpin sequences. These cysteinesare 14 A apart in the mature folded structure. We also engineered59C72C in the b-hairpin sequence. These cysteines are 25 A apartin the mature folded structure. Pxlink for both 86C99C (Fig. 3b,filled green circle) and 59C72C (filled inverted black triangle) isB0.3, in agreement with 62C75C (filled red square) at this samelocation. All filled symbols represent cysteine pairs far outside thetunnel on a long tether, which have Pxlink of 0.2–0.3. To confirmfindings in the interaction-permissive region, we engineered aconstruct containing 59C72C that poises this cysteine pair in thevicinity of the exit port but outside the tunnel. The Pxlink for thisconstruct was 40.6 (data not shown). These findings permit amore precise interpretation of the cross-linking results for hairpinresidues 53C66C and 125C149C, which are in close proximity inthe T1 domain (Fig. 3a). Although these residues in each hairpininitially interact as part of the exploration of conformationalspace allotted by the tunnel’s entropic window, the hairpins foldinto stable structures only in the context of the longer, foldedT1 domain.

Cooperative subdomain foldingThe a-hairpin folds when the N-terminal region of the T1 domain,including the b-hairpin, has already been synthesized and folded1

(Fig. 3a). This folded N terminus may provide a topological template.

NH2

a

b

NH2

1.0

0.8

Pxl

ink

0.6

0.4

0.2

125C

149C

53C66

C

53C66

C65D

125C

149C

125C

149C

65D

125C

149C

∆N0.0

N122M

PTCFull length

Truncated

Mutant

T65D

β1 β2

β1 β2

α4 α5

α4 α5

α4 α5

Figure 4 T1 domain mutants. (a) A schematic of T1 constructs with an

N-terminal deletion or a T65D mutation is shown along with the full-length

peptide for comparison. The T65D mutation is shown as a black circle.

A truncated peptide (DN) begins at N122M. All constructs were cut with a

BstEII restriction enzyme to give a DPTC value of 239 for the 125C149C

constructs and 322 for the 53C66C constructs. (b) Nascent peptides were

translated, isolated and assayed for cross-linking (see Methods). The Pxlink

values for these N-terminal deletion and T65D mutants are depicted as

white bars. The gray bars represent data shown in Figure 3a for nascent

peptides 125C149C and 53C66C. The results for the a-helical hairpins

are indicated by the rightmost and leftmost sets of bars. The results for the

b-hairpin are shown in the middle set of bars. Pxlink data are means ± s.e.m.

for triplicate measurements.

ART IC L E S

40 8 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

This hypothesis predicts that folding of the a-hairpin sequence shouldbe sensitive to deletion of its preceding T1 residues. To explore thispossibility, we truncated the T1 domain at position 122 (Fig. 4), whichyields an N-terminally deleted nascent peptide (DN) containing thea-hairpin sequence starting 3 residues after the methionine start site.For a long tether (DPTC 239), Pxlink is significantly decreased by 46%(P ¼ 0.004; Fig. 4). It seems, therefore, that the extent of a-hairpinfolding depends on the presence of the N-terminal T1 domain. Inaddition to these long-tethered DN constructs with a folding domainoutside of the tunnel, we investigated shorter DN constructs,DN-DPTC24 and DN-DPTC29, in which the folding domains residenear the end of the tunnel. The truncated DN-DPTC24 folded lessefficiently than the nontruncated peptide (Pxlink is 0.06 ± 0.12,n ¼ 2 versus 0.26 ± 0.03, n ¼ 3, respectively, (SupplementaryFig. 4 online)). Truncated DN-DPTC29, which is also near the endof the tunnel and provides a larger dynamic range, has a Pxlink of0.30 ± 0.02 (n ¼ 2) versus 0.46 ± 0.03 (n ¼ 5) for the nontruncatedconstruct. For both DN-DPTC24 and DN-DPTC29 constructs lackingthe N-terminal b-hairpin, folding of the a-hairpin was less efficient,again consistent with folding of a C-terminal hairpin depending on thepresence and status of the N-terminal b-hairpin sequence. In contrast,the b-hairpin apparently still folded in the absence of the C-terminalregion of T1, including the a-hairpin sequence. A 118-residue nascentpeptide that includes the b-hairpin sequence (DPTC ¼ 86), but notthe a-hairpin sequence, folds as much as the longer nascent peptides(DPTC ¼ 322 (Fig. 3a)) that include both b- and a-hairpins.

The second strategy we used to probe the dependence of a-hairpinformation on the b-hairpin was to use a Kv1.3 mutation, T65D.Residue 65 lies in the b-hairpin at the intersubunit interface of the T1tetramer. Kv1.3 with a T65D mutation does not tetramerize andonly achieves one-half as much tertiary folding as wild-type T1(refs. 23,25). We therefore used the T65D mutation to investigatetwo issues. Does the T65D mutation impair tertiary folding of theb-hairpin? Does the T65D mutation affect a-hairpin formation? Weintroduced the T65D mutation into both a 53C66C background(DPTC 322) and a 125C149C background (DPTC 239) that placesthe b- and a-hairpins far outside the tunnel. Pxlink in the T65D mutantis 0.46 for the b-hairpin and 0.26 for the a-hairpin, each significantlylower than wild-type Pxlink (P ¼ 0.003 and 0.01, respectively) (Fig. 4).The a-hairpin is more disrupted than the b-hairpin, consistent withaccumulated disruption for the more C-terminal location of thea-hairpin. We suggest that propagated disruption from the site ofmutation all along the folded T1 interface is responsible fordefective quaternary structure formation of the T65D mutant T1domain. Folding of the a-hairpin depends on the presence and statusof the b-hairpin.

DISCUSSIONSubdomain foldingAlthough it is too cramped inside the tunnel to accommodate theentire T1 domain of Kv channels in its fully folded state1, consistentwith previous investigations of other proteins6,26,27 and the geometricanalysis of Moore and co-workers28, tertiary interactions do occur inthe tunnel, particularly in its last 20 A. This region supports helixformation1–4,29, specifically, folding of the a5 helix of the T1 domainstudied herein3. Why is this region permissive for tertiary interactions?First, the dimensions of the tunnel in this region may be wider30 thanthe 20 A estimated from the crystal structures of the ribosome, whichlack both nascent peptides and attendant chaperones. Second, theribosomal tunnel may be more dynamic than previously thought(whereas Steitz’s estimate of the diameter of the largest sphere that can

fit inside a peptideless tunnel, 13.7 A31, can accommodate the b-hairpin, an increase of less than two-fold is required to accommodatethe a-hairpin). Moreover, the entrance of the ribosomal tunnelundergoes conformational changes during peptide elongation32 andthe tunnel is believed to have a gate33–35. Third, tertiary- andsecondary-structure formation may be coupled36 and thus potentiatedin this distal portion of the tunnel that favors secondary folding3,29.An a-helix requires less room than a hairpin of similar length andamino acid composition, and thus an a-helix may form along the first80 A of the tunnel, independent of tertiary folding, but the conversemay not hold. For the a4-a5 hairpin, tertiary structures in the last 20A of the tunnel may be favored by coupled secondary-structureformation. Moreover, chaperone proteins hovering at the exit portof the tunnel may facilitate tertiary and/or coupled folding.

Although these two subdomains differ in size and length of thehairpin, and the number of hydrophobic interactions along theirrespective hairpin interfaces, our cross-linking assay suggests thatthe ease of folding of the b- and the a-hairpin are not significantlydifferent. The b-hairpin evidently folds despite the absence of thea-hairpin. Similarly, the a-hairpin apparently folds in the absence ofthe b-hairpin in a truncated N-terminally deleted (DN) mutant.However, folding of the a-hairpin is not independent of the b-hairpin.The a-hairpin in the DN mutant folds less than the a-hairpin in thefull-length peptide, which contains both b- and a-subdomains. More-over, mutation of b-hairpin residue 65 (T65D) causes a decrease infolding of both the b- and a-hairpins. In the full-length mature Kv1.3,T65D prevents both quaternary and complete tertiary folding of theT1 domain and, consequently, function of the Kv1.3 channel23,25. Thisis expected because tertiary folding and tetramerization of T1 arecoupled25. We now understand that the basis for this defect in tertiaryfolding involves the propagated disruption of subdomain formation.Our findings underscore the usefulness of this approach in diagnosingthe molecular basis for folding defects.

Information about the extent of folding at discrete locations alongthe tunnel is embedded in the cross-linking-accessibility (PF) curves inFigure 3a. The locations are deduced from a cysteine-accessibilityassay. This assay is based on the monotonic dependence of thefraction pegylated for a given cysteine in an all-extended tape measureon the distance of that cysteine from the PTC2. We have compared thecysteine-accessibility results obtained for a-hairpin constructs withthose obtained for this all-extended tape measure. The accessibilityof the C-terminal cysteine in the tunnel, in a-hairpin constructs, issimilar to the accessibility of cysteines in the tape measure withthe same DPTC, demonstrating that these cysteines reside at thesame location in the tunnel. This similarity is expected, given thatthe segment of the peptide between the C-terminal cysteine of thea-hairpin (149C) and the PTC is completely extended3. Thus, themeasured accessibility of 149C is not an artifact of secondary-structureformation in the region downstream from 149C, nor is it an artifactof blocked access to 149C by the upstream a4 and a5 helicesin the tunnel.

Dynamics of the nascent peptide–tunnel complexOur results suggest that the nascent peptide–tunnel complex may bemore dynamic than originally believed. At one location (DPTC 24),the level of cross-linking suggests that the a-hairpin exists in atime-averaged equilibrium between an extended a4-a5 species anda tertiary hairpin. Dynamics are also manifest in the distal portionof the tunnel that provides an entropic window for initial explorationof local peptide regions for folding. This represents a restricted timeand space in which the nascent peptide visits potential tertiary

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 0 9

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

conformational states. It is possible that this region, the last 20 A nearthe exit port, is lined by chaperone proteins, which operationallyextend the length of an 80 A tunnel. Estimates of the length of thetunnel vary from 80 A to 112 A2,9,10,12,28. Regardless, this region showsa monotonically increasing accessibility with increasing distance fromthe PTC2,30. If this increased accessibility represents a widening of thetunnel, this will allow an increased number and mobility of watermolecules and ions, and therefore an increased driving force forburying hydrophobic residues at a tertiary folded interface (hydro-phobic effect37). Both a4-a5 and b1-b2 hairpins contain hydrophobicresidues that are in close contact all along their folded interfaces13. Thefinal tertiary hairpin fold is stabilized by constraints of the additionalsecondary and tertiary interactions of the whole folded T1 domainoutside the tunnel.

Our results highlight three principles that are likely to have criticalroles in protein folding during biogenesis. The first is that intra-molecular tertiary interactions occur before the nascent peptide hasfully emerged from the ribosomal tunnel, allowing for a regulatoryrole of folding by the ribosome and its attendant chaperones. Thesecond is that the nascent peptide in the exit port of the tunnel isstructurally dynamic and can therefore explore a myriad of conforma-tions before fully emerging from the ribosome. This may enable thepeptide to fold more efficiently than it would in a less confined space.The third is that tertiary interactions are influenced by subdomainfolding elsewhere in the nascent peptide. We propose that these threeprinciples constrain and shape the earliest stages of protein foldingand may be used to help us to understand the molecular basis forfolding defects that underlie protein malfunction.

METHODSConstructs and in vitro translation. We used standard methods of bacterial

transformation, plasmid DNA preparation and restriction enzyme analysis. The

nucleotide sequences of all mutants were confirmed by automated cycle

sequencing performed by the DNA Sequencing Facility at the University of

Pennsylvania School of Medicine on an ABI 377 Sequencer using Big dye

terminator chemistry (A0BI). We sequenced all mutant DNAs throughout

the entire coding region. Engineered cysteines, restriction enzyme sites and

N-terminal deletions were introduced into pSP/Kv1.3/cysteine-free22 using the

QuikChange Site-Directed Mutagenesis Kit (Stratagene) as described23.

We synthesized capped cRNA in vitro from linearized templates using Sp6

RNA polymerase (Promega). Linearized templates for Kv1.3 translocation

intermediates were generated using several restriction enzymes to produce

DNA constructs of different lengths lacking a stop codon. cRNAs were

translated in vitro with [35S]Methionine (2 ml per 25 ml translation mixture,

B10 mCi ml�1 Express (Dupont/NEN Research Products)) for 1 h at 22 1C in a

rabbit reticulocyte lysate (2 mM final DTT concentration) according to the

Promega Protocol and Application Guide.

Cross-linking assay. As described23, we added translation reaction (10–20 ml)

to 500 ml PBS (calcium-free DPBS (GIBCO) containing 4 mM MgCl2, pH 7.3)

with 2 mM DTT. The suspension was centrifuged using a TLA100.3 Beckman

rotor at 70,000 rpm for 20 min at 4 1C through a sucrose cushion (120 ml,

containing 0.5 M sucrose, 100 mM KCl, 50 mM HEPES, 5 mM MgCl2, 2 mM

DTT, pH 7.5). The pellet was resuspended in 50 ml or 500 ml PBS. Orthophe-

nyldimaleimide (PDM; Sigma), 0.5 mM, was added to those samples to be

labeled, whereas a control sample was treated identically but in the absence of

PDM, at B0 1C for 30 min. No reducing agent was present in these

incubations. PDM samples used for subsequent MAL-pegylation were

quenched with 10 mM b-mercaptoethanol at room temperature (24 1C) for

10 min. Control samples, untreated with PDM, were treated identically. A third

sample was labeled with PDM but reserved for treatment with PEG-SH. Thiol

reducing agents must be avoided in PDM samples subsequently treated with

PEG-SH, otherwise free maleimides will be modified and further assay with

PEG-SH will be blocked. Samples were centrifuged at 70,000 rpm (TLA100.3

Beckman rotor) at 4 1C for 20 min, resuspended in 50 ml PBS containing 1%

(w/v) SDS and incubated at room temperature for 20 min. Those samples

designated for pegylation with PEG-MAL (MW 5000; Shearwater) were treated

with 10 mM b-mercaptoethanol to prevent oxidation, which inhibits pegyla-

tion. Samples destined for pegylation with PEG-SH (MW 5000; Shearwater)

received 50 ml PBS containing SDS only. All SDS-treated samples were diluted

with either 50 ml PBS containing PEG-MAL to give a final PEG-MAL

concentration of 20 mM and 5 mM b-mercaptoethanol or 50 ml PBS contain-

ing PEG-SH to give a final concentration of 20 mM PEG-SH. Both the

pegylation and PDM reactions reached a maximum, constant level within 60

min and o15 min, respectively, at 4 1C23.

Single- and double-cysteine constructs were treated identically, as described

above. Data derived from single-cysteine constructs served as controls to

calculate the efficiency of PEG-SH pegylation needed to determine the

probability of cross-linking and folding in the double-cysteine constructs.

Accessibility assay. We added translation reaction (10–20 ml) to 500 ml PBS

(calcium-free DPBS (GIBCO), containing 4 mM MgCl2, pH 7.3) with 2 mM

DTT. The suspension was centrifuged at 70,000 rpm (TLA100.3 Beckman

rotor) for 20 min at 4 1C through a sucrose cushion (120 ml, containing 0.5M

sucrose, 100 mM KCl, 50 mM HEPES, 5 mM MgCl2, 2 mM DTT, pH 7.5). The

pellet was resuspended in 100 ml PBS with 50 mM b-mercaptoethanol and

treated with 1 mM PEG-MAL (SunBio) at 4 1C for 4–6 h. To quench the

pegylation reaction each sample was treated with DTT to neutralize PEG-MAL

(200:1 ratio), incubated at room temperature for 15 min and centrifuged at

70,000 rpm (TLA100.3 Beckman rotor) for 20 min at 4 1C. The relative

accessibility of each C-terminal cysteine indicates whether the subdomain is

inside or outside of the tunnel.

Gel electrophoresis and fluorography. We treated all samples with RNase

(20 mg ml�1) before subjecting them to electrophoresis using the NuPAGE

system and precast Bis-Tris 10% or 12% gels and MOPS running buffer. Gels

were soaked in Amplify (Amersham) to enhance 35S fluorography, dried

and exposed to Kodak X-AR film at –70 1C. Typical exposure times were

16–30 h. We quantified gels using a Molecular Dynamic PhosphorImager,

which detects c.p.m. that are not necessarily visualized in autoradiograms

exposed for 16–30 h.

Analysis of pegylation ladders. For any given construct, radioactive protein

incubated with PEG-MAL or PEG-SH was detected as distinct bands on

NuPAGE gels and quantified using a PhosphorImager (Molecular Dynamics).

The data were analyzed as follows. For each lane, j, of the gel, the fraction of

total protein molecules with exactly i pegylated cysteines was calculated as

Wj(i) ¼ c.p.m.(i)/Sc.p.m.(i) (equation (1), where c.p.m.(i) is the counts per

minute in the ith bin). If each cysteine is assumed to label to the same final

extent, the fraction Fj of individual cysteines pegylated in the jth lane is

SiWj(i)/N (equation (2), where N is the total number of cysteines in the

protein molecule).

In this study, we apply the analysis previously described1 to determine

the probability of cross-linking cysteines that reside inside the ribosomal

tunnel as well as those residing outside. Specifically, we compare the labeling

in single-cysteine constructs and double-cysteine constructs and can estimate

the cross-linking efficiency. The cross-linking efficiency is used to determine the

probability of a pair of cysteines being cross-linked by PDM: Pxlink ¼ 1 �(2F3,AB)/(F3,A + F3,B), where F3,A and F3,B are fractions of PDM-treated single

cysteine (A or B) mutants labeled with PEG-SH, and F3,AB is the fraction of

PDM-treated double cysteine (A and B) mutant labeled with PEG-SH.

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ACKNOWLEDGMENTSWe thank R. Horn, J. Frank and U. Hartl for careful reading of the manuscriptand R. Horn for helpful discussion. This research was funded by the US NationalInstitutes of Health grant GM 52302 to C.D.

AUTHOR CONTRIBUTIONSA.K. performed the experiments; A.K. and C.D. designed the research, interpretedthe results and wrote the manuscript.

ART IC L E S

41 0 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Kosolapov, A., Tu, L., Wang, J. & Deutsch, C. Structure acquisition of the T1 domain ofKv1.3 during biogenesis. Neuron 44, 295–307 (2004).

2. Lu, J. & Deutsch, C. Secondary structure formation of a transmembrane segment in Kvchannels. Biochemistry 44, 8230–8243 (2005).

3. Tu, L., Wang, J. & Deutsch, C. Biogenesis of the T1–S1 linker of voltage-gated K+channels. Biochemistry 46, 8075–8084 (2007).

4. Woolhead, C.A., McCormick, P.J. & Johnson, A.E. Nascent membrane and secretoryproteins differ in FRET-detected folding far inside the ribosome and in their exposure toribosomal proteins. Cell 116, 725–736 (2004).

5. Mingarro, I., Nilsson, I., Whitley, P. & von Heijne, G. Different conformations of nascentpolypeptides during translocation across the ER membrane. BMC Cell Biol. 1, 3(2000).

6. Kowarik, M., Kung, S., Martoglio, B. & Helenius, A. Protein folding during cotransla-tional translocation in the endoplasmic reticulum. Mol. Cell 10, 769–778(2002).

7. Hardesty, B. & Kramer, G. Folding of a nascent peptide on the ribosome. Prog. NucleicAcid Res. Mol. Biol. 66, 41–66 (2001).

8. Matlack, K.E. & Walter, P. The 70 carboxyl-terminal amino acids of nascent secretoryproteins are protected from proteolysis by the ribosome and the protein translocationapparatus of the endoplasmic reticulum membrane. J. Biol. Chem. 270, 6170–6180(1995).

9. Ban, N., Nissen, P., Hansen, J., Moore, P.B. & Steitz, T.A. The complete atomicstructure of the large ribosomal subunit at 2.4 A resolution. Science 289, 905–920(2000).

10. Nissen, P., Hansen, J., Ban, N., Moore, P.B. & Steitz, T.A. The structural basis ofribosome activity in peptide bond synthesis. Science 289, 920–930 (2000).

11. Menetret, J.F. et al. The structure of ribosome-channel complexes engaged in proteintranslocation. Mol. Cell 6, 1219–1232 (2000).

12. Beckmann, R. et al. Architecture of the protein-conducting channel associated withthe translating 80S ribosome. Cell 107, 361–372 (2001).

13. Minor, D.L. et al. The polar T1 interface is linked to conformational changes that openthe voltage-gated potassium channel. Cell 102, 657–670 (2000).

14. Kreusch, A., Pfaffinger, P.J., Stevens, C.F. & Choe, S. Crystal structure of thetetramerization domain of the Shaker potassium channel. Nature 392, 945–948(1998).

15. Li, M., Jan, Y.N. & Jan, L.Y. Specification of subunit assembly by the hydrophilicamino-terminal domain of the Shaker potassium channels. Science 257, 1225–1230(1992).

16. Shen, N.V., Chen, X., Boyer, M.M. & Pfaffinger, P. Deletion analysis of K+ channelassembly. Neuron 11, 67–76 (1993).

17. Xu, J., Yu, W., Jan, J.N., Jan, L. & Li, M. Assembly of voltage-gated potassiumchannels. Conserved hydrophilic motifs determine subfamily-specific interactionsbetween the a-subunits. J. Biol. Chem. 270, 24761–24768 (1995).

18. Gu, C., Jan, Y.N. & Jan, L.Y. A conserved domain in axonal targeting of Kv1 (Shaker)voltage-gated potassium channels. Science 301, 646–649 (2003).

19. Cushman, S.J. et al. Voltage dependent activation of potassium channels is coupled toT1 domain structure. Nat. Struct. Biol. 7, 403–407 (2000).

20. Kurata, H.T. et al. Amino-terminal determinants of U-type inactivation of voltage-gatedK+ channels. J. Biol. Chem. 277, 29045–29053 (2002).

21. Wang, G. & Covarrubias, M. Voltage-dependent gating rearrangements in the intracel-lular T1–T1 interface of a K+ channel. J. Gen. Physiol. 127, 391–400 (2006).

22. Lu, J., Robinson, J.M., Edwards, D. & Deutsch, C. T1–T1 interactions occur in ERmembranes while nascent Kv peptides are still attached to ribosomes. Biochemistry40, 10934–10946 (2001).

23. Kosolapov, A. & Deutsch, C. Folding of the voltage-gated K+ channel T1 recognitiondomain. J. Biol. Chem. 278, 4305–4313 (2003).

24. Lu, J. & Deutsch, C. Pegylation: a method for assessing topological accessibilities inKv1.3. Biochemistry 40, 13288–13301 (2001).

25. Robinson, J.M. & Deutsch, C. Coupled tertiary folding and oligomerization of the T1domain of Kv channels. Neuron 45, 223–232 (2005).

26. Kolb, V.A., Makeyev, E.V. & Spirin, A.S. Folding of firefly luciferase during translation ina cell-free system. EMBO J. 13, 3631–3637 (1994).

27. Makeyev, E.V., Kolb, V.A. & Spirin, A.S. Enzymatic activity of the ribosome-boundnascent polypeptide. FEBS Lett. 378, 166–170 (1996).

28. Voss, N.R., Gerstein, M., Steitz, T.A. & Moore, P.B. The geometry of the ribosomalpolypeptide exit tunnel. J. Mol. Biol. 360, 893–906 (2006).

29. Lu, J. & Deutsch, C. Folding zones inside the ribosomal exit tunnel. Nat. Struct. Mol.Biol. 12, 1123–1129 (2005).

30. Lu, J., Kobertz, W.R. & Deutsch, C. Mapping the electrostatic potential within theribosomal exit tunnel. J. Mol. Biol. 371, 1378–1391 (2007).

31. Steitz, T.A. A structural understanding of the dynamic ribosome machine. Nat. Rev.Mol. Cell Biol. 9, 242–253 (2008).

32. Gabashvili, I.S. et al. The polypeptide tunnel system in the ribosome and its gating inerythromycin resistance mutants of L4 and L22. Mol. Cell 8, 181–188 (2001).

33. Nakatogawa, H. & Ito, K. The ribosomal exit tunnel functions as a discriminating gate.Cell 108, 629–636 (2002).

34. Berisio, R. et al. Structural insight into the role of the ribosomal tunnel in cellularregulation. Nat. Struct. Biol. 10, 366–370 (2003).

35. Tu, D., Blaha, G., Moore, P.B. & Steitz, T.A. Structures of MLSBK antibiotics bound tomutated large ribosomal subunits provide a structural explanation for resistance.Cell 121, 257–270 (2005).

36. Daggett, V. & Fersht, A.R. Is there a unifying mechanism for protein folding? TrendsBiochem. Sci. 28, 18–25 (2003).

37. Chandler, D. Interfaces and the driving force of hydrophobic assembly. Nature 437,640–647 (2005).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 1 1

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Acetylation by GCN5 regulates CDC6 phosphorylationin the S phase of the cell cycleRoberta Paolinelli1,2,4,5, Ramiro Mendoza-Maldonado2,5, Anna Cereseto1 & Mauro Giacca2,3

In eukaryotic cells, the cell-division cycle (CDC)-6 protein is essential to promote the assembly of pre-replicative complexes in theearly G1 phase of the cell cycle, a process requiring tight regulation to ensure that proper origin licensing occurs once per cellcycle. Here we show that, in late G1 and early S phase, CDC6 is found in a complex also containing Cyclin A, cyclin-dependentkinase (CDK)-2 and the acetyltransferase general control nonderepressible 5 (GCN5). GCN5 specifically acetylates CDC6 at threelysine residues flanking its cyclin-docking motif, and this modification is crucial for the subsequent phosphorylation of the proteinby Cyclin A–CDKs at a specific residue close to the acetylation site. GCN5-mediated acetylation and site-specific phosphorylationof CDC6 are both necessary for the relocalization of the protein to the cell cytoplasm in the S phase, as well as to regulate itsstability. This two-step, intramolecular regulatory program by sequential modification of CDC6 seems to be essential forproper S-phase progression.

The exact duplication of the genome once per cell division is aprerequisite for allowing proper cell proliferation. In eukaryoticcells, this is primarily achieved by exercising tight control over theprocess of initiation of DNA replication. In particular, regulation isachieved by temporally separating the assembly of the pre-replicativecomplex (pre-RC) at origins of DNA replication from the actual startof DNA synthesis1. Origin ‘licensing’ is sequential with the binding ofthe six-subunit origin recognition complex (ORC) to origin DNA; thisdetermines the recruitment of the CDC6 and CDT1 factors, which inturn promote the loading of the putative DNA replicative helicaseminichromosome maintenance protein (MCM) complex2,3. Thelicensed origins are only subsequently triggered to initiate DNAreplication by the concerted actions of the S phase–promoting kinases(CDC7 and CDKs4). To prevent re-replication, origins are thusregulated in such a way that once they have fired they are broughtback to an unlicensed state, and re-licensing is inhibited untilcompletion of mitosis5.

Multiple lines of evidence accumulated over the last few years haveindicated that the CDC6 protein is essential for both the initiation ofDNA replication and the regulation of licensing in Saccharomycescerevisiae, Schizosaccharomyces pombe, Xenopus laevis and mammals(reviewed in refs. 6–8). This protein is indispensable for the formationand maintenance of pre-RCs and for the loading of MCM onto originsof replication9–12. Both the S. cerevisiae Cdc6p and S. pombe Cdc18homologs are CDK substrates, and their phosphorylation at the G1-Sboundary leads to ubiquitin-mediated proteolysis of the two proteins.In S. pombe, the degradation of Cdc18p seems to be a straightforwardmechanism to restrict the frequency of origin usage to once per cell

cycle, because overexpression of Cdc18p induces re-replication andmutants lacking CDK consensus sites are more efficient than wild-typeCdc18p at this process (reviewed in ref. 6). Human CDC6 is alsophosphorylated by CDKs at three specific serine residues in itsN-terminal domain, at positions 54, 74 and 106; these seem to bethe only functional CDK sites in CDC6 (refs. 13–15). The protein,however, is not destroyed during S phase, but remains presentthroughout S, G2 and M phase, and it is eventually degraded, bythe nonmitotic form of the anaphase-promoting complex/cyclosome(APC/C) E3 ubiquitin ligase, only after entering the G1 phase of thesubsequent cell cycle; the same event also occurs in cells that haveexited the cell cycle13–17. Recent work has indicated that, in late G1phase for cells exiting quiescence or at the end of mitosis in cyclingcells, CDKs containing Cyclin E phosphorylate CDC6 and that thismodification prevents APC/C-mediated proteolysis of the protein18.Prevention of CDC6 degradation would be a mechanism to ensurepre-RC assembly and origin licensing during a ‘window of opportu-nity’ when other APC/C substrates that are inhibitory for theseprocesses, such as geminin and Cyclin A, are instead degraded.

CDC6 phosphorylation by Cyclin E–CDKs provides a molecularexplanation for pre-RC assembly and origin licensing in the G1 phase.At this time point, CDC6 is exclusively found in the nucleus and bindsto chromatin19. Once the cells enter the S phase, however, at least partof the endogenously expressed protein is relocalized to the cyto-plasm13,14,16,20,21, an occurrence that is particularly evident when theprotein is overexpressed22. The relocalization of CDC6 is temporallycoincident with the appearance of Cyclin A, and it seems to bespecifically dependent on its phosphorylation by Cyclin A–containing

Received 12 May 2008; accepted 4 March 2009; published online 3 April 2009; doi:10.1038/nsmb.1583

1Molecular Biology Laboratory, Scuola Normale Superiore, AREA della Ricerca del CNR, Pisa, Italy. 2Molecular Medicine Laboratory, International Centre for GeneticEngineering and Biotechnology (ICGEB), Trieste, Italy. 3Department of Biomedicine, Faculty of Medicine, University of Trieste, Italy. 4Present address: IFOM-Instituteof Molecular Oncology-Foundation, Milan, Italy. 5These authors contributed equally to this work. Correspondence should be addressed to M.G. (giacca@icgeb.org).

41 2 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

ART IC L E S

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

CDKs, because it is induced by overexpression of Cyclin A but not ofCyclin E (ref. 14). These observations point to the existence of subtledifferences in the phosphorylation pattern that the different CDKsimpinge on CDC6 in subsequent phases of the cell cycle, suggestingthe existence of additional mechanisms that regulate the intracellularfate and stability of CDC6.

Resulting from a search for human pre-RC components that mightinteract with cellular histone acetyltransferases (HATs), here wedescribe our serendipitous discovery that human CDC6 associateswith, and is acetylated by, GCN5. In yeast, this protein represents theenzymatic subunit of the transcriptional regulatory complex SAGA(Spt-Ada-Gcn5 acetyltransferase)23 and seems essential for cell-cycleprogression24–27. We show that, in human cells, CDC6 takes part inthe formation of a complex with Cyclin A–CDK2 that also includesGCN5, that this HAT acetylates CDC6 during the early S phaseof the cell cycle and, finally, that this event is crucial for proper cell-cycle progression.

RESULTSGCN5 acetyltransferase binds and acetylates CDC6While searching for cellular HATs that interact with the pre-RCmembers, we observed that recombinant glutathione S-transferase(GST)-CDC6 associated with a HeLa cell nuclear HAT activity.Notably, we also observed that GST-CDC6, but not a GST control,was itself a substrate for acetylation (Fig. 1a). To verify whetherendogenous CDC6 was also acetylated, HeLa cell extracts were immu-noprecipitated using an anti–acetyllysine antibody and immunoblottedusing an anti-CDC6 antibody. A 63-kDa band corresponding to

endogenous CDC6 protein was readily detected (Fig. 1b); analogousfindings were also obtained in 293T, U2OS and T98G human cell lines(data not shown). To further confirm the in vivo acetylation of CDC6,we transfected 293T cells with Flag-tagged CDC6, treated them withtrichostatin A (TSA), an inhibitor of cellular deacetylases, carried outimmunoprecipitation of cell lysates with an anti-Flag antibody andimmunoblotted with an anti–acetyllysine antibody. We detected acety-lated Flag-CDC6 in the anti-Flag immunoprecipitates (SupplementaryFig. 1a online).

Next we observed that GST-CDC6 is a specific substrate ofrecombinant GCN5 in an in vitro HAT assay (SupplementaryFig. 1b). By using a series of GST-CDC6 fusion proteins (Fig. 1c),we concluded that the target region for GCN5 acetylation lies betweenamino acids 91 and 110 of CDC6 (Fig. 1d) and that the three lysinespresent in this region (at positions 92, 105 and 109; Fig. 1e) were alltargets for GCN5-mediated acetylation (Fig. 1f). Factor acetylation byHATs is often concomitant with the specific binding of the enzyme toits substrate. Indeed, this was also found to be the case for GCN5 andCDC6; the interaction between the two proteins requires the integrityof the N-terminal region of CDC6 and the C terminus of GCN5(Supplementary Fig. 1c–h).

By analyzing the level of acetylated CDC6 in asynchronous celllysates, we estimated that B18% of total CDC6 was acetylated insidethe cells (Supplementary Fig. 2a–d online). This level was clearlyenhanced by the overexpression of enzymatically active GCN5 but notof its catalytically inactive mutant GCN5mut28; this enhancement washigher than that obtained by cell treatment with trichostatin A (TSA)(Supplementary Fig. 2e). Similar to endogenous CDC6, acetylation of

1–90

CDC6

Destrbox

Cy dockmotif

ATPaseORChom NES

NLS KEN boxA B

Ac-GST-CDC6

Ac-Histones

WB anti-CDC6Ac-CDC6(endog.) *

*

*

* **

*

NEa

b

c

d

e

f

g

GSTGST-C

DC6

IP a

nti-A

c-Ly

s

IP C

ontro

l

WCL

91–561111–561186–561

Ac-GCN5 CDC6 proteins

HA-GCN5HA-GCN5mut

Ac-CDC6 (transf.)GCN5/GCN5mut

CDC6/CDC6(K3R) (transf.)

CDC6/CDC6(K3R) (transf.)GCN5/GCN5mut

Tubulin

WB anti-Ac-LysWB anti-HA WB anti-Flag

WB anti-FlagWB anti-HAWB anti-tubulin

IP anti-Flag

WCL

Vector

Flag-CDC6 Flag-CDC6(K3R)

GCN5

GST

CDC6

Ac Ac AcP

92 105

106

109

K

Cy dockmotif

K K

K

K

K

R

R

R

R K S

S

S

S

S

S

S

K

K

R

R

R R

K R

K

R

K92R

K105RK109R

K92R K105R

K105R K109R

K92R K105R K109R (K3R)

GSTCDC6

1–90

91–5

61

GSTCDC6

K92RK10

5R

K109R

K92R K

105R

K105R

K10

9R

K92R K

105R

K109R

(K3R

)

111–

561

186–

561

++

+

–– –

––– ––

++––

Leu zip

Figure 1 CDC6 is acetylated by GCN5 on lysines 92, 105 and 109 both in vitro and in vivo. (a) Recombinant CDC6 is acetylated by a nuclear HAT in vitro.

HeLa nuclear extract (NE) was incubated with GST-CDC6 or GST in the presence of [14C]-labeled acetyl-CoA and purified histones. Ac, acetylated.

(b) Endogenous CDC6 is acetylated. Immunoblot shows the acetylation of endogenous CDC6 after immunoprecipitation (IP) from HeLa cells lysates with an

anti–acetyllysine (Ac-Lys) antibody and subsequent detection with a specific anti-CDC6 antibody. Immunoprecipitation with an unrelated antibody was used

as a control. The band marked by an asterisk (*) in the whole-cell lysates (WCL) represents an additional band detected with this specific antibody, as also

reported by others39. (c) Schematic representation of the main functional domains of CDC6 and of the mutants used for acetyltransferase assays. Destr box

and KEN box, destruction boxes; NLS, nuclear localization signals; Cy dock motif, RXL or cyclin-docking motif; ATPase ORC hom, ATPase ORC homology

domain; A and B, Walker A and B motifs; Leu Zip, leucine zipper domain; NES, nuclear export signals. Dashed box indicates the region involved in GCN5-

dependent acetylation. (d) CDC6 fragment (residues 91–110) is acetylated by GCN5. An acetylation assay was performed incubating recombinant GCN5 and

the GST-CDC6 deletion mutant. The positions of the GST-fused CDC6 proteins tested are marked by an asterisk (*); protein levels were comparable for allmutants (data not shown). Acetylation was low or undetectable for the GST-CDC6 fragments encompassing either the C terminus (residues 186–561) or

the N terminus (residues 1–90) of the protein. Fragment 91–561 was acetylated at a level similar to wild-type CDC6, whereas acetylation dropped to

background levels in fragment 111–561. (e) Schematic representations of the CDC6 fragment (91–110), with the indication of lysine and serine residues

and of the point mutants used in HAT assays. (f) GCN5 acetylates CDC6 lysines 92, 105 and 109. All the singly mutated recombinant proteins scored

positive for acetylation by GCN5. Acetylation of the double mutants was reduced, whereas the triple mutant (K3R) was not acetylated. (g) Acetylation

of CDC6 by GCN5 requires integrity of lysines 92, 105 and 109. Three upper immunoblots show acetylation of Flag-tagged CDC6 or the CDC6 K3R

mutant after immunoprecipitation from 293T cell lysates using an anti-Flag antibody, followed by detection with an anti-Ac-Lys or anti-Flag antibodies.

Co-immunoprecipitated HA-tagged versions of GCN5 were detected with an anti-HA antibody. Notice that the K3R mutant, although insensitive to GCN5

expression, was still acetylated inside the cells, indicating that other HATs might also modify CDC6 in vivo at different lysine residues.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 1 3

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

transfected wild-type CDC6 was also markedly increased in responseto wild-type GCN5 but not to GCN5mut (Fig. 1g). In contrast, theCDC6 K3R mutant, which was not acetylated by GCN5 in vitro, wasnot sensitive to GCN5 overexpression in vivo, although it still co-immunoprecipitated with the enzyme.

Collectively, these results clearly indicate that CDC6 lysines 92, 105and 109 are specific targets for GCN5 acetylation both in vitro andinside the cells.

CDC6 acetylation by GCN5 affects Ser106 phosphorylationThe acetylated lysine residues of CDC6 frame the cyclin-docking motifof the protein20. In particular, Lys105 is adjacent to Ser106, one of thethree CDC6 serine residues (54, 74 and 106) that are specificallyphosphorylated by the CDKs13,14. We therefore questioned whetherGCN5-dependent acetylation might affect CDC6 phosphorylation. Weinitially observed that the overexpression of GCN5 increased the levelsof CDC6 phosphorylated at Ser106 (pSer106), as detected either afterimmunoprecipitation (Fig. 2a) or by straight western blotting(Fig. 2b). In the same lysates, no modification was observed forCDC6 phosphorylated at Ser54 (pSer54). In contrast to wild-typeGCN5, the expression of GCN5mut did not increase CDC6 pSer106

levels, but it led to a substantial accumulation of the protein inside thecells while decreasing the levels of Cyclin E and Cyclin A, suggestingperturbed cell-cycle progression (Fig. 2b).

These results initially disclosed an unexpected link between GCN5-mediated CDC6 acetylation and the specific phosphorylation of theprotein on Ser106. To further explore this issue, we analyzed the levelsof phosphorylation of transfected wild-type CDC6, of the K3R mutantand of an additional mutant we constructed bearing a serine-to-alanine substitution at position 106 (S106A; Fig. 2c). Notably, theK3R mutant was not phosphorylated on Ser106, similarly to theS106A mutant and unlike wild-type CDC6 (Fig. 2d). Both mutants,however, were still phosphorylated on Ser54. The anti–pSer106-CDC6antibody was still able to recognize the K3R mutant when phosphory-lated in vitro by Cyclin A–CDK, thus indicating that the K3Rmutation per se did not impair epitope recognition (SupplementaryFig. 3 online).

To further define the role of GCN5 in mediating endogenous CDC6Ser106 phosphorylation, we knocked down GCN5 by RNA inter-ference29. GCN5 depletion markedly increased the levels of totalCDC6 (Fig. 2e) and selectively inhibited CDC6 Ser106 phosphoryla-tion, while leaving Ser54 phosphorylation unaltered (Fig. 2f). In the

same siRNA-treated cell lysates, the level ofMCM3 protein, a factor also modified byacetylation30, was unaltered. In addition,silencing of GCN5 did not decrease the levelsof CDK2–Cyclin A nor it impaired phosphor-ylation of other CDK2–Cyclin A targets, suchas retinoblastoma-1 (Rb)31 (SupplementaryFig. 4a online).

Collectively, these data are consistent withthe possibility that GCN5-mediated acetyla-tion perturbs the levels of CDC6 proteinby specifically affecting its phosphorylationon Ser106.

GCN5 complexes with Cyclin A–CDK2and acetylates CDC6 in S phaseWe analyzed the levels of CDC6 acetylationin human glioblastoma T98G cells, whichaccumulate in G0 upon serum starvationand then synchronously enter G1 after re-addition of serum32,33 (Fig. 3a). The levels ofGCN5 were reduced in cell lysates fromserum-starved cells as compared to asynchro-nous cells; however, they started to increase6 h after serum addition (early G1), peaked at20 h (early S) and decreased at 24 h (midS-phase) (Fig. 3b). The levels of CDC6 werealmost undetectable under serum starvationand started to progressively increase from16 h after serum re-addition onward, inaccordance with earlier observations18. Ofinterest, acetylated CDC6 showed a clearenrichment at 20 h (early S), followed by areturn to basal levels at 24 h (mid S). Thus,the amounts of GCN5 and acetylated CDC6both peak in early S phase. Similar findingswere also obtained in synchronized HeLacells (data not shown).

Next we monitored the levels of the twoCDC6 phospho-isoforms during cell-cycle

HA-GCN5

CDC6

HA-GCN5 – + –HA-GCN5mut – – +

Vector + – –

HA-GCN5 + + ++ – –Flag-CDC6

–– +Flag-CDC6(K3R)–– +Flag-CDC6(S106A)

CDC6 pS106

CDC6 pS54

GCN5

CDC6 & mutants

Tubulin

CDC6 pS106

CDC6 pS54

Cyclin E

Cyclin A

GCN5/GCN5mut

Tubulin

Vectora d

e

f

b

c

IP anti-CDC6 pS106WB anti-CDC6

WB anti-CDC6

CDC6

92 105

106

109

Cy dockmotif

K

Ac Ac AcP

KS K

R R

A

RK3R

S106A

WB anti-CDC6 pS106

WB anti-CDC6 pS54

CDC6 pS54WCL

WCL

IP anti-CDC6 pS54

IP anti-CDC6 pS106

Moc

ksi-

Luc

si-GCN5

Moc

ksi-

Luc

si-GCN5

CDC6 pS106

WB anti-Cyclin E

WB anti-Cyclin A

WB anti-HA

WB anti-tubulin

WB anti-CDC6 pS106

Mock siGCN5

WB anti-GCN5

WB anti-CDC6

WB anti-CDC6

WB anti-CDC6

WB anti-GCN5GCN5

MCM3

Tubulin

WB anti-MCM3

WB anti-tubulin

WB anti-tubulin

GCN5

CDC6

Tubulin48h 72h150 nM 300 nM

48h 72h

WB anti-CDC6 pS54

WB anti-HA

WB anti-tubulin

WB anti-Flag

WB anti-CDC6

WB anti-HA WCL

WCL

WB anti-tubulin

––+

+

CDC6 pS106

CDC6

GCN5

Tubulin

Figure 2 Specific CDC6 phosphorylation on Ser106 depends on GCN5-mediated CDC6 acetylation.

(a) GCN5 overexpression increases specific phosphorylation of CDC6 on Ser06 (CDC6 pS106). The

upper immunoblot shows the levels of endogenous CDC6 pSer106 after immunoprecipitation with an

anti-CDC6 pSer106 antibody. The additional immunoblots show the levels of total CDC6, HA-GCN5

and a–tubulin in whole-cell lysates. (b) Expression of catalytically inactive GCN5 impairs CDC6 Ser106

phosphorylation and increases total CDC6 levels. Immunoblots were performed on whole-cell lysates

from cells transfected with wild-type GCN5 or GCN5mut. (c) Schematic representations of the CDC6

fragment encompassing residues 91–110 and of the K3R and S106A point mutants. (d) Integrity of

lysines 92, 105 and 109 is essential for CDC6 phosphorylation on Ser106. The upper immunoblot

shows immunodetection of CDC6 pSer106 after co-transfection of GCN5 with wild-type CDC6 or withCDC6 K3R or S106A. The additional immunoblots show the levels of CDC6 pSer54, total CDC6

and GCN5, along with endogenous a-tubulin. (e) Depletion of endogenous GCN5 leads to CDC6

accumulation. Immunodetection of total CDC6 protein levels after RNAi, as indicated. (f) The CDC6

pSer106, but not the CDC6 pSer54, level decreases upon GCN5 knockdown. The upper immunoblots

show the levels of CDC6 phosphorylated on either Ser54 or Ser106 after cell treatment with an anti-

GCN5 siRNA, or with an irrelevant siRNA anti-luciferase. The lower immunoblots show additional

results from whole-cell lysates using the indicated antibodies.

ART IC L E S

41 4 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

progression. After 6 h (late G1) phosphorylation of Ser54 was low,whereas from 16 h onward it markedly increased (Fig. 3c). Phosphor-ylation of Ser106 was less pronounced at 16 h as compared tophosphorylationof Ser54, and pSer106 levels peaked at 20–24 h.Notably, CDC6 is known to be phosphorylated in the S phase in aCyclin A–CDK2–dependent manner13,14,16,20,21. Indeed, in the T98Gsynchronization, Cyclin E levels peaked at 6 h (early G1) and16 h (late G1), whereas Cyclin A started to appear at 16 h and itslevels increased at 20 h (early S) and 24 h (mid S) (Fig. 3c). Thesefindings are thus consistent with the possibility that CDC6 Ser106phosphorylation might be attributable to Cyclin A–CDK2. Indeed, weobserved that Ser106 phosphorylation substantially increased uponCyclin A (but not Cyclin E) overexpression, whereas the levels ofpSer54 remained unaltered (Supplementary Fig. 4b).

We proceeded to investigate whether the interaction between CDC6and GCN5 might vary during cell-cycle progression. As concludedfrom co-immunoprecipitation experiments using extracts from syn-chronized T98G cells, binding between CDC6 and GCN5 was max-imal at 20 h after serum addition, the same time point at which bothGCN5 levels and CDC6 acetylation peaked (Fig. 3d). Both pSer54 andpSer106 CDC6 bound GCN5 at this time point. However, the amountof co-immunoprecipitated GCN5 was higher in the anti-pSer106immunoprecipitates. This was even more clear at 24 h after serumaddition (mid S), when co-immunoprecipitation was detected usingthe only the anti-pSer106 but not the anti-pSer54 antibody.

Taken together, the cell-cycle experiments showed that: (i) CDC6associated with GCN5 in early S phase; (ii) at the same time point, theprotein became acetylated; (iii) binding between GCN5 and CDC6was preferential for the CDC6 pSer106 phospho-isoform; (iv) acetyla-tion, phosphorylation and GCN5 binding were concomitant with theexpression of Cyclin A; (v) the overexpression of Cyclin A selectivelyincreased phosphorylation of CDC6 on Ser106. As GCN5 oftenparticipates in the formation of multicomponent protein complexesin different species23,34,35, we tested whether it might also associatewith Cyclin A–CDK2. Indeed, both GCN5 and CDC6 were co-immunoprecipitated together with CDK2 by an antibody againstCyclin A, as was CDK2 by an antibody against GCN5 (SupplementaryFig. 5a and 5b online, respectively).

The observations above are consistent with the possibility thatCDC6 phosphorylation on Ser106 might preferentially occur onacetylated CDC6. Accordingly, we found that higher levels of acety-lated CDC6 were associated with pSer106 CDC6 rather than withpSer54 CDC6 or total CDC6 (Supplementary Fig. 6a,b online).

CDC6 modifications regulate its subcellular localizationIn human cells, CDC6 is phosphorylated in a Cyclin A–CDK2–dependent manner during the S phase and then translocated to thecytosol and subsequently degraded6,13,14,20,21. Confocal immunofluor-escence analysis performed in nonsynchronized HeLa cells using amonoclonal anti-CDC6 antibody revealed that in about 55% of thecells endogenous CDC6 was nuclear, whereas in about 45% of cells theprotein had a cytoplasmic localization (Fig. 4a). However, most(490%) of the asynchronous cells that expressed Cyclin A, anS-phase marker, belonged to the subset of cells with exclusivecytoplasmic localization of CDC6 (Fig. 4b). In keeping with thesefindings, CDC6 was localized in the cytoplasm in more than 90% ofthe cells in which Cyclin A had been overexpressed (Fig. 4c).

We examined whether the acetylation of the factor might influenceits subcellular localization. Indeed, we found that cell treatment withTSA increased the number of cells with exclusive cytoplasmic localiza-tion of CDC6 (68% versus 45%, P o 0.01; Fig. 4c). Analogous resultswere obtained by overexpression of active GCN5 (71% of cells withcytoplasmic localization among those positive for the expression ofthe transfected protein, P o 0.01), but not of GCN5mut (Fig. 4d).Flow cytometry experiments on the cells transfected with wild-typeGCN5 showed an appreciable increase in the number of S-phasecells, similar to that obtained by overexpression of Cyclin A (Supple-mentary Fig. 7 online). Finally, we knocked down GCN5 expressionby RNA interference; this treatment increased the number ofcells in the G1 phase (Fig. 4e) and determined a prevalentlynuclear CDC6 localization (Fig. 4f).

Next we explored the subcellular localization of transfected CDC6S106A, which cannot be phosphorylated on Ser106, and of K3R,which cannot be acetylated and is thus unphosphorylated on Ser106.We observed that transfected CDC6 had a similar distribution toendogenous CDC6 (B55% of the cells with nuclear localization;

Figure 3 CDC6 acetylation is cell cycle

dependent. (a) T98G cells synchronization. Flow

cytometry profiles of asynchronous cells (AS),

cells blocked in G0 by serum starvation (0 h) or

cells at different times after serum stimulation,

with the deduced cell-cycle phase indicated.

(b) Acetylated CDC6 peaks in early S phase. The

upper four immunoblots show the levels of the

indicated proteins in whole-cell extracts from

cells synchronized at different time points. The

lower blot shows the levels of acetylated CDC6

after immunoprecipitation with an anti-CDC6

antibody followed by western blotting with an

anti-Ac-Lys antibody. (c) Phosphorylation of CDC6

on Ser106 is concomitant with the appearance ofCyclin A, at the same time point at which CDC6

is acetylated. The immunoblots show the levels

of the indicated proteins in whole-cell lysates

of synchronized cells, with the exception of

CDC6 pSer106, which was detected after

immunoprecipitation with an anti-pSer106-CDC6

antibody followed by western blotting with an antibody against total CDC6. (d) CDC6 co-immunoprecipitates with GCN5 in early S phase. The antibodies

used for immunoprecipitation (IP) and immunodetection (WB) are indicated on the right side. The experiments were performed with asynchronous T98G cell

populations (AS) or with cells synchronized at the indicated time points after serum addition, as in a.

ASa

T98Gcells

72 h

Serumstarvation

Add serum

0 h

28 h

6 h(early G1)

20 h(early S)

24 h(mid S)

16 h(late G1-S)

cTime after

6 16 20 24 28WB anti-CDC6

WB anti-CDC6 pS54

WB anti-CDK2

WB anti-Cyclin E

WB anti-Cyclin A

WB anti-tubulin

IP anti-CDC6 pS106WB anti-CDC6

serum addition (h):

CDC6

CDC6 pS54

CDC6 pS106

CDK2

Cyclin E

Cyclin A

Tubulin

bTime after

AS 0 6 16 20 24 28WB anti-GCN5

WB anti-CDC6

WB anti-tubulin

WB anti-Ac-Lys IP anti-CDC6WB anti-CDC6

WB anti-Cyclin A

serumaddition (h):

GCN5Cyclin A

CDC6Tubulin

Ac-CDC6

CDC6

d Time after

AS 6 20 24IP anti-CDC6WB anti-CDC6

WB anti-CDC6IP anti-GCN5

IP anti-CDC6 pS106

IP anti-CDC6 pS54

WB anti-GCN5

WB anti-CDC6WB anti-GCN5

WB anti-CDC6WB anti-GCN5

serumaddition (h):

CDC6

CDC6

CDC6 pS106

GCN5

GCN5

CDC6 pS54

GCN5

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 1 5

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Fig. 5a). Similarly to what was observed for endogenous CDC6, thisdistribution was modified by cell treatment with TSA, at which CDC6became cytoplasmic in 465% of the cells (P o 0.01). Notably, boththe S106A and the K3R mutants were strictly nuclear, and theirlocalization was altered neither by TSA treatment (Fig. 5a) nor byGCN5 overexpression (Supplementary Fig. 8 online). Both mutants,however, were still able to co-immunoprecipitate endogenous CyclinA and hemagglutin-tagged GCN5, with similar efficiency to the wild-type protein (data not shown). Taken together, these observationsclearly indicate that both acetylation and Ser106 phosphorylation ofCDC6 are required for its relocation to the cytoplasm.

Next we checked whether the exclusive nuclear localization of theCDC6 Ser106 and K3R mutants might reflect the association of theseproteins with chromatin. Whole-cell lysates from cells transfected withwild-type CDC6 or with the two mutants were fractionated toseparate the cytosolic, nuclear soluble and nuclear insoluble compart-ments19. The wild-type protein was found to localize in the insolublenuclear compartment (B50% of total) and in the nucleosolic andcytoplasmic fractions (15% and 35% respectively; see Fig. 5b forwestern blots and Fig. 5c for quantification). In sharp contrast, and inagreement with the immunofluorescence data, less than 10% of either

mutant was detected in the cytoplasm. Of interest, however, thesubnuclear distribution of the two mutants was markedly different,in that the S106A protein was found in both the nucleosolic andchromatin fractions (30% and 63% respectively), whereas the K3Rmutant was detectable only in the chromatin pellet (92%). Thesefindings are thus consistent with the conclusion that CDC6 acetylationis essential for the detachment of CDC6 from chromatin, whereasSer106 phosphorylation later promotes relocalization of the protein tothe cytoplasm. In keeping with this conclusion, when lysates fromnontransfected cells were partitioned in the same manner, endogenousCDC6 pSer54 was found distributed in both the cytoplasmic andnuclear insoluble compartments, similarly to the distribution of totalCDC6. In contrast, CDC6 pSer106 was present exclusively in thecytoplasmic fraction (Supplementary Fig. 9 online).

Effects of CDC6 modifications on cell-cycle progressionNext we examined whether the change in chromatin association andsubcellular localization of the two CDC6 mutants perturbed normalcell-cycle progression. Overexpression of wild-type CDC6 or some ofits other mutants does not exert a strong effect in primary cells,because multiple mechanisms (including CDT1 inhibition by geminin

IF a

nti-C

DC

6(e

ndog

enou

sC

DC

6)

IF anti-CDC6(endogenous CDC6)

IF anti-CDC6(endogenous CDC6)

+HA-GCN5

+HA-GCN5mut

IF anti-CDC6(endogenous

CDC6)

HeLa cells

si-Luc ++–

–si-GCN5

GCN5

Tubulin

si-Luc si-GCN5

WB anti-tubulin

WB anti-GCN5

IF anti-GCN5(endogenous

GCN5) Merge

IF anti-Cyclin A(endogenous Cyclin A)

IF anti-HA(GCN5/GCN5mut)

Merge

Merge

IF a

nti-C

DC

6(e

ndog

enou

sC

DC

6)

Num

ber

of C

DC

6+

cells

(%

)

Num

ber

of C

DC

6+

cells

(%

)

Cel

l-cyc

le d

istr

ibut

ion

(%)

Num

ber

of C

yclin

A–

posi

tive

cells

(%

)N

umbe

r of

HA

-pos

itive

cel

ls (

%)

100

75

50

25

N C N C N C

N

Mock

si-Luc

si-GCN5

C

0

100

75

50

25

0

100

75

50

25

0

100

75

50

25

0

100

G2-M

si-Lu

c

si-GCN5

SG0/G1

80

60

40

20

0

N C

100

75

50

25

0

N C

100

75

50

25

0

N C

100

75

50

25

0

Num

ber

of C

DC

6+

cells

(%

)N

umbe

r of

CD

C6+

cells

(%

)N

umbe

r of

CD

C6+

cells

(%

)

N C

100

75

50

25

0

Num

ber

ofH

A-p

ositi

ve c

ells

(%

)

N C

100

75

50

25

0

a

b f

d

c e

Figure 4 GCN5-dependent CDC6 acetylation regulates its subcellular localization. (a) Subcellular localization of endogenous CDC6. The image shows HeLa

cells immunostained with a monoclonal anti-CDC6 antibody. The graph shows the percentage of nuclear (N) and cytoplasmic (C) subcellular distribution of

CDC6-positive cells. The histograms report the results (mean ± s.e.m., indicated by error bars) of at least three different experiments. Scale bar, 10 mm.

(b) Cells expressing Cyclin A show cytoplasmic localization of endogenous CDC6. The pictures show co-immunostaining of endogenous CDC6 (green) andendogenous Cyclin A (red). (c) Cytoplasmic CDC6 localization is enhanced by Cyclin A overexpression or treatment with TSA. Data are presented as in a. The

efficiency of transfection in the experiments with Cyclin A was 480%. (d) Cytoplasmic CDC6 localization is enhanced by GCN5 HAT activity. The pictures

show the co-immunostaining of endogenous CDC6 (green) and of transfected HA-tagged GCN5 or GCN5mut (red) in HeLa cells. (e) GCN5 depletion

accumulates cells in G1. Cells were treated with an anti-GCN5 siRNA (reduction of GCN5 level to o10%; blots shown above). The flow cytometry profiles

shown below left show the DNA content of the siRNA-treated cells; the distribution is shown in the histogram to the right. The arrow shows the increase in

the number of G1-phase cells after GCN5 knockdown. (f) GCN5 depletion forces CDC6 nuclear localization. The pictures show the co-immunostaining of

endogenous CDC6 (red) and endogenous GCN5 (green) on cells after RNAi with anti-GCN5 and control siRNAs.

ART IC L E S

41 6 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

and p53-dependent DNA-damage response) prevent cellular DNAre-replication36,37. However, overexpression of CDC6 in p53�/� cellshas been shown to escape checkpoint inhibition and to induce, tosome extent, DNA re-replication38. We therefore analyzed the effect ofour mutants in the p53-null H1299 human lung carcinoma cell line.At 24 h after transfection, we treated cells with bromodeoxyuridine(BrdU) to selectively label cells in active DNA synthesis; after 1 h, wevisualized BrdU incorporation and DNA content by flow cytometry.As shown in Figure 5d and quantified in Figure 5e, the overexpressionof both wild-type CDC6 and the S106A mutant determined areduction in the number of cells incorporating BrdU; in sharpcontrast, the K3R mutant markedly increased the number of cellsinvolved in DNA synthesis (from 43% to 67%), while reducing thenumber of cells in G1 (from 44% to 30%) and, most notably, in G2-M

(from 16% to 3%). This result is consistent with the conclusion thatthe overexpression of the K3R mutant, which is tightly chromatin-bound, specifically impairs S-phase progression without affectingentry into the S phase. We observed no detectable re-replicationwith any of the mutants.

Cytoplasmic transport of human CDC6 correlates with degradationof the protein6,13,14,20,21. We therefore wanted to assess the stability ofthe constitutively nuclear CDC6 S106A and K3R mutants relative towild-type CDC6. After treatment of asynchronous cells with cyclo-heximide (CHX) to block protein synthesis, the half-life of transfectedwild-type CDC6 was less than 1 h, similar to endogenous CDC6 andanalogous with published findings39. In contrast, both the S106A andK3R mutants were remarkably more stable; in particular, after 4 h ofCHX treatment, more than 75% of K3R was still present inside

Flag-CDC6a

c d

b

100

75

Num

ber

of c

ells

(%

)

50

25

0

-BrdU

BrdU

FL1

H (

Brd

U)

FL1

H (

Brd

U)

0

104

103

102

101

100

104

103

102

101

100

200 400FL2A (PI)

CDC6 CDC6(S106A) CDC6(K3R)

600 0 200 400FL2A (PI)

600

G1 G2-M G1 G2-M

BrdU BrdU BrdU

G1 G2-MG1 G2-M G1 G2-M

BrdU BrdU

+BrdU Mock

100

75

50

25

0

100

75

50

25

0

100

75

50

25

0

100

75

50

25

0

100

75

50

25

0N C N C N C N C N C N C

IF a

nti-F

lag

(tra

nsfe

cted

CD

C6)

Flag-CDC6(K3R)

Flag-CDC6(K3R)

Tubulin

ORC2

Flag-CDC6(S106A)

Flag-CDC6(S106A)

Flag-CDC6

WB anti-Flag

WB anti-tubulin

WB anti-ORC2

e100

80

60

Cel

l cyc

le d

istr

ibut

ion

(%)

40

20

0G0/G1 BrdU G2/M

MockCDC6CDC6(S106A)CDC6(K3R)

f Time after0 1 2 4

WB anti-CDC6

WB anti-tubulin

WB anti-Flag

WB anti-tubulin

WB anti-Flag

WB anti-tubulin

WB anti-Flag

WB anti-tubulin

CHX addition (h):

CDC6 (endog.)

Tubulin

Tubulin

Tubulin

Tubulin

Flag-CDC6

Flag-CDC6(S106A)

Flag-CDC6(K3R)

g

75

% C

DC

6 pr

otei

n

50

25

100

00 1 2

CHX (h)

3 4

Flag-CDC6

CDC6(endog.)

Flag-CDC6(S106A)Flag-CDC6(K3R)

WCL Cyt Sol Ins

100

80

% o

f tot

al

60 InsSolCyt

40

20

0

CDC6

CDC6(S10

6A)

CDC6(K3R

)

Figure 5 Characterization of the CDC6 K3R and S106A mutants. (a) CDC6 K3R and CDC6 S106A have an exclusively nuclear localization which is not

modified by TSA treatment. The pictures show HeLa cells immunostained with an anti-Flag antibody after transfection with wild-type CDC6 or the two

mutants. The graphs show the nuclear (N) and cytoplasmic (C) subcellular distribution of Flag-positive cells by percentage (mean ± s.e.m. of at least three

independent experiments). Scale bar, 10 mm. (b) CDC6 K3R and CDC6 S106A localize to the insoluble nuclear compartment. Whole-cell lysates (WCL) fromuntransfected cells (for CDC6, Orc2 and a-tubulin) or transfected with wild-type CDC6, K3R and S106A were fractionated to generate a cytosolic (Cyt), a

soluble nuclear (Sol) and an insoluble nuclear (Ins) fraction, in which protein levels were assessed by western blotting. (c) Quantification of the relative

amount of protein in the cytosolic, nuclear soluble and nuclear insoluble compartments in the experiment shown in b. (d) Cell-cycle distribution and BrdU

incorporation in H1299 cells expressing wild-type CDC6 or the CDC6 K3R and S106A mutants. The pictures show flow cytometry profiles (x axis, propidium

iodide (PI) staining; y axis, BrdU incorporation) of cells transfected with either an empty vector (Mock) or with plasmids expressing the indicated proteins

(efficiency of transfection 490%) and then pulse-labeled with BrdU. Above left, cells not treated with BrdU; above middle, untransfected cells. The regions

corresponding to the G1 (BrdU-negative, 2n-DNA content), G2-M (BrdU-negative, 4n-DNA content) and S-phase (BrdU-positive, from 2n- to 4n-DNA

content) gates are indicated. (e) Quantification of cell-cycle distribution after transfection of the indicated CDC6 mutants (mean ± s.e.m. of three

independent experiments). Experiments were performed as in d. (f) Stability of endogenous wild-type CDC6, transfected wild-type CDC6 and the K3R

and S106A mutants. After transfection, H1299 cells were treated with cycloheximide (CHX) for the indicated time periods and total-cell lysates were

analyzed by western blotting. (g) Quantification of the experiment shown in f. The amount of each protein is expressed as a percentage of the initial level.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 1 7

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

the cells (Fig. 5f,g). Thus, by promoting CDC6 detachment fromchromatin and relocalization to the cytoplasm, acetylation facilitatesCDC6 degradation. This conclusion fully concurs with the observationthat GCN5 depletion and GCN5mut overexpression both markedlyincreased the total amount of CDC6 (Fig. 2e and 2b, respectively).

DISCUSSIONOur data show that CDC6 is acetylated in vivo by the GCN5 HAT atthree lysines that frame the Cyclin-docking motif in the N terminus ofthe protein, a region that is not directly involved in pre-RC forma-tion38. Acetylation regulates the levels of CDC6 pSer106, because theoverexpression of GCN5 increases phosphorylation at this residue,whereas the triple mutant in the three acetylated lysines is no longerphosphorylated. During the cell cycle, acetylation of CDC6 occurs inthe early S phase, when the levels of both Cyclin A and GCN5 peak. Atthis time point, both GCN5 and CDC6 are found in a complex withCyclin A and CDK2. Cyclin A–mediated phosphorylation of CDC6 onSer106 requires prior acetylation of CDC6. Finally, cell treatment witha deacetylase inhibitor or overexpression of GCN5 force cytoplasmicrelocalization of both endogenous and transfected CDC6, an effectthat is also obtained by transfection of Cyclin A. Consistently, CDC6proteins bearing mutations at either Ser106 or at the three lysines thatare acetylated are exclusively nuclear and their stability is increased.

Work originally performed in the Xenopus in vitro replicationsystem indicated that Cyclins E and A have specialized roles duringthe transition from G0 to S phase40. Whereas Cyclin E stimulates pre-RC assembly, Cyclin A activates DNA synthesis by replication com-plexes that are already assembled, while also inhibiting the assembly ofnew complexes. Thus, Cyclin E opens a ‘window of opportunity’ forpre-RC assembly that is closed by Cyclin A40. Recent work indicatesthat an essential mechanism that permits the opening of this windowis the specific phosphorylation of CDC6 by Cyclin E, which preventsdegradation of the protein by the APC/C and thus permits pre-RCassembly18. This conclusion has been reached mainly by using anantibody against CDC6 phosphorylated on Ser54. Our work extendsthese findings further by showing that, when the cells enter the Sphase, CDC6 pSer54 is found in a complex that also includes GCN5,Cyclin A and CDK2. At this moment, GCN5 specifically acetylatesCDC6. This modification determines its release from chromatinand allows its further phosphorylation on Ser106, followed byrelocalization of the protein to the cytoplasm and its eventualdegradation. The finding that the K3R mutant, which is not acety-lated, is still normally phosphorylated on Ser54 but not at all on

Ser106 and is specifically chromatin bound is fully consistent with thisconclusion. Equally consistent are the observations that overexpres-sion of the K3R mutant leads to a marked increase in the number ofBrdU-positive cells and a substantial decrease of both G1 and, mostremarkably, G2-M cells and that both the K3R and S106A mutants areconsiderably more stable than the wild-type protein. A model sum-marizing these findings is drawn schematically in Figure 6.

Our results uncover the events that occur at a step that follows pre-RC assembly and origin firing, namely the steps that coincide withCyclin A’s appearance. Work performed in X. laevis using an XCdc6protein mutated in the three CDK phosphorylation sites has indeedsuggested that XCdc6 phosphorylation by CDKs is not essential foreither regulated binding of XCdc6 to chromatin nor for the subsequentloading of the MCMs, thus suggesting that Cdc6 phosphorylationmight be required at later stages of the replication process41. Indeed,our work indicates that acetylation and subsequent specific phosphor-ylation of CDC6 on Ser106 are essential to allow detachment of theprotein from chromatin, relocalization to the cytoplasm and degrada-tion, and that these events are essential to ensure proper cell-cycleprogression. Consistent with this conclusion, experimental evidenceobtained in both X. laevis and human cells has shown that over-expression of CDC6 in G2 cells inhibits mitosis by inducing acheckpoint pathway involving Chk1 (refs. 42,43), and that this propertyis modulated by the phosphorylation of CDC6 at particular residues43.

The role of CDC6 subcellular localization has roused much con-troversy in recent years. Several authors have reported convincinglythat in the G1 phase the protein is exclusively nuclear, whereas duringthe S phase most of it relocalizes to the cytoplasm13,14,16,20,21; howeverother data have suggested that these effects might be a peculiarity of anexogenously overexpressed protein, and that most of the endogenouslyexpressed protein would remain essentially nuclear throughout the cellcycle19,22. Our results in fact challenge this latter conclusion, andclearly indicate that human endogenous CDC6 (stained by an anti-CDC6 monoclonal antibody) is found in both the nucleus and thecytoplasm in nonsynchronous cells. Consistent with the reportedbinding and kinase activity of Cyclin A–CDK2 (refs. 14,20,41), inmost of the cells that express endogenous Cyclin A, CDC6 is foundonly in the cytoplasm (Fig. 4b), and Cyclin A overexpression forcesthe cytoplasmic relocalization of endogenous CDC6 (Fig. 4c), as alsoreported by other laboratories14,18. CDC6 acetylation has an essentialrole in determining the cytoplasmic relocalization of the protein. Thisconclusion is supported by the observation that the overexpression ofGCN5 (but not that of its catalytically inactive mutant) forces the

CDC6 NUCLEAR CDC6 CYTOPLASMIC

Relocalization to thecytoplasm

Degradation

APC/C

CDC6

CDK2

P P P PAc Ac Ac

Ac Ac Ac

PPP

CDK2

Cyclin E

Cyclin E

Cyclin A

6 h 16 h 20 h 24 h 28 h

G0 G1 G1-S S

Cyclin ACDC6 CDC6

GCN5

Release fromchromatin

GCN5

CDC6

Ser54, Ser74 phosphorylationby Cyclin E–CDK2

Ser106 phosphorylationby Cyclin A–CDK2

Lys92, Lys105, Lys109acetylation by GCN5

Figure 6 Model showing the regulation of CDC6

by sequential modification by acetylation and

phosphorylation in early S phase. In G0 cells,

CDC6 is not phosphorylated and is continuously

degraded by the APC/C. Upon entry into the cell

cycle, phosphorylation of CDC6 on Ser54 by

Cyclin E–CDKs opens a ‘window of opportunity’,

during which degradation of the protein is

prevented and assembly of the pre-RC is thus

allowed. Upon S-phase entry, CDC6 is specifically

acetylated by GCN5; this modification determines

the release of the protein from chromatin and

permits its further phosphorylation on Ser106, a

modification that is carried out by Cyclin A–CDKs.

CDC6, GCN5, Cyclin A and CDK2 indeed interactin early S-phase cells. Phosphorylation of

CDC6 on Ser106 determines the relocalization

of the protein to the cytoplasm, followed by

its degradation.

ART IC L E S

41 8 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

cytoplasmic relocation of both endogenous and transfected CDC6,whereas the GCN5 knockdown has the opposite effect. In a consistentmanner, both the K3R mutant (which is not acetylated) and the S106Amutant (which is not phosphorylated on Ser106) have a strictlynuclear localization, irrespective of GCN5 overexpression (Fig. 5).These data reinforce the conclusion that the cytoplasmic relocalizationof CDC6 is strictly dependent on the consequential acetylation andphosphorylation of CDC6 on Ser106.

The results presented in this manuscript also underscore the role ofGCN5 as a general cell-cycle regulator44. Indeed, this HAT seems tohave a key role in regulating the expression of several cell cycle–relatedgenes, such as Cyclin A, Cyclin D3, PCNA and CDC25B. Theobservation that GCN5 also regulates the function of one keyregulator of pre-RC formation and controls licensing extends thisconcept further. The finding that CDC6 is specifically acetylated byGCN5 and that this HAT finely tunes its function does not exclude thepossibility that the protein might also be the substrate of other HATs.Indeed, recent work performed in Xenopus has shown that CDC6might also be an in vitro substrate for the HBO1 acetyltransferase45,the same HAT that also interacts with ORC1 and MCM2 (refs. 46,47).

Finally, our findings support the recently coined concept that proteinacetylation and phosphorylation might occur in different proteins astwo closely interconnected modifications that are part of a multistepregulatory program48. Examples of other factors in which thesemodifications are strictly related and, in some instances, sequential,include p53 and p73, Rb, FOXO1 and MYC49–53. It will be interestingto understand the exact changes, resulting from CDC6 acetylation, inthe molecular structure of the protein and how these changes mightimpinge on the subsequent site-specific phosphorylation.

METHODSPlasmids. We constructed the expression vectors pFlag-CDC6 and pGEX20T-

CDC6 by PCR amplification of the CDC6 cDNA from the pcDNA3-CDC6

vector (a gift from C. Pelizon, Wellcome/CRC Institute, Cambridge, UK) and

subcloned them into pFlagCMV 2 (Stratagene) and pGEX20T vectors, respec-

tively. The pGEX2T-GCN5 short isoform (GCN5 S) expressing vector was a

gift from M. Benkirane. pGEX-2T-GCN5 deletion mutants were obtained by

PCR amplification of GCN5 cDNA with primers specific for all the deleted

versions. pcDNA3-HA-GCN5 was prepared by subcloning the GCN5 cDNA

into the pcDNA3-HA vector (Invitrogen). The expression vector pcDNA3-

HA-GCN5mut containing the catalytically inactive GCN5 S mutant (Y260A

F261A)28 was constructed by recombinant PCR. Different versions of CDC6

(residues 1–60, 1–90, 1–185, 1–363, 91–561, 111–561, 186–561) and GCN5 S

(1–189, 190–270, 271–383, 384–476) deleted mutants were obtained by PCR

amplification and cloned into the pGEX vectors. The pGEX-CDC6 K3R and the

pFlag-CDC6 KR and S106A point mutants were constructed using recombinant

PCR starting from each original vector. The pCMX-cyclin A and pCMX-cyclin E

vectors were a gift from J. Pines.

Cell culture, cell synchronization and cell-cycle analysis. HeLa, T98G, U2-OS

and HEK 293T cell lines were maintained in DMEM, the H1299 cell line was

maintained in RPMI and both media were supplemented with Glutamax (Life

Tecnologies) and 10% (v/v) FBS (Life Tecnologies). T98G cells were synchro-

nized by serum starvation32. Cells were analyzed for their cell-cycle profile

(DNA content) by incorporation of propidium iodide (Sigma) and analyzed by

flow cytometry on a FACSCalibur (Becton Dickinson). Cell-cycle profile

distributions were determined with the Modfit LT 3.0 software (Verity Software

House). Treatments with TSA (Sigma) were performed by adding the drug

(250 ng ml–1) overnight. BrdU incorporation experiments were performed on

transiently transfected cells at 48 h after transfection. Cells were pulsed for 1 h

with BrdU (final concentration 10 mM), and BrdU-positive cells were detected

by using a fluorescein isothiocyanate (FITC)-conjugated anti-BrdU antibody

(Becton Dickinson). Cells were collected and analyzed by double-flow cytome-

try analysis on a FACSCalibur (Becton Dickinson) to simultaneously determine

the cell-cycle profile (DNA content) by incorporation of propidium iodide and

the S-phase cell population by incorporation of BrdU. Cell-cycle profile

distributions were determined with the CellQuestPro (BD Biosciences) and

Modfit LT 3.0 software.

Analysis of protein stability. We performed protein-stability experiments on

cells transiently transfected with Flag-CDC6–expressing vectors. At 24 h after

transfection, cycloheximide (Sigma) was added at a final concentration of

30 mg ml–1. Cell lysates were obtained at different time points, and protein

levels were assessed by immunoblotting using the ECL system followed by

densitometric analysis.

Antibodies and biochemical fractionation. Antibodies to CDC6 (sc-9964),

CDC6 (sc-8341), CDC6 (sc-13136), CDC6 pSer54 (sc-12920), CDC6 pSer106

(sc-12922), Cyclin E (sc-481), Cyclin A (sc-571), GCN5 (sc-6303 and sc-

20698), MCM3 (sc-9850) and hemagglutinin (HA) (sc-805) were from Santa

Cruz Biotechnology; anti-Flag M2 (F1804) and anti–a-tubulin (T6074) were

from Sigma; anti-CDK2 (610145) was from BD Transduction Laboratories;

anti-ORC2 was from MBL; anti-acetyllysines antibodies (#9441 and #06-933)

were from Cell Signaling Technology and Upstate Biotechnology, respectively;

anti-Rb antibodies were from BD Pharmingen (#554136, anti-Rb), Cell

Signaling (#9308S, anti-phospho S807/S811 Rb) and Invitrogen (#44-582G,

anti-phospho T821 Rb). We prepared whole-cell extracts in HNNG buffer

(15 mM HEPES pH 7.5, 250 mM NaCl, 1% (v/v) NP-40, 5% (v/v) glycerol,

1 mM PMSF) supplemented with 20 mM sodium butyrate (Sigma), 10 mM

NaF (Sigma) and protease inhibitors cocktail tablet (Roche). Immunoblots

were carried out with 30–50 mg of whole-cell lysates. Lysates for immuno-

precipitation (1–2 mg ml–1 of protein) were incubated overnight with the

appropriate amount of antibody (1–2 mg) at 4 1C. Immunocomplexes were

collected with protein A/G plus agarose beads (Santa Cruz Biotechnology),

protein A trisacryl beads (Pierce) or anti-Flag M2-conjugated agarose beads

(Sigma), washed in HNNG buffer and treated with DNase I (Gibco BRL) for

15 min at room temperature (22–24 1C). Beads were sequentially washed at

4 1C with HLNG buffer (as HNNG but with LiCl), TE buffer and finally

resuspended in Laemli sample buffer. Proteins were separated by 10%

SDS-PAGE (Invitrogen) and detected by immunoblotting using the enhanced

chemiluminescence systems (ECL, Amersham Bioscience). Cytosolic (S2-Cyt),

nucleosolic (S3-Sol) and chromatin-bound, nuclear insoluble (P3-Ins) frac-

tions were prepared following biochemical fractionation as described19.

In vitro acetylation assay. We performed HAT assays as reported with minor

modifications54. Briefly, the GST fusion proteins used as substrates were

incubated with HeLa nuclear extracts or recombinant purified GCN5 and

[14C]-acetyl-CoA in HAT buffer (50 mM Tris, pH 7.5, 5% (v/v) glycerol, 0.1 M

EDTA, 50 mM KCl and 2 mM sodium butyrate) in a final volume of 20 ml for

45 min at 30 1C. Acetylated proteins were visualized by phosphorimaging

(Cyclone, Packard) after separation by SDS-PAGE.

GST pull-down assay. We produced [35S]-labeled proteins used for in vitro

binding assays by the TNT Reticulocyte Lysate System (Promega), using the

corresponding pcDNA3 vectors as templates. Preparation of the GST fusion

proteins54 and set up of GST pull-down assays55 were performed as described.

Immunofluorescence. Immunofluorescence analysis was conducted as

described56. The secondary FITC- (sc-2010) and Alexa594-conjugated anti-

bodies (A11032, A11072) were from Santa Cruz Biotechnology and Invitrogen

Molecular Probes, respectively. Confocal fluorescence analysis was performed

on a TCS-SL Leica confocal microscope. Images were acquired using the

Leica software.

RNA interference. Cells were transiently transfected with an RNA against

GCN5 (Dharmacon-SMARTpool selected)29 for 48 h or 72 h at different final

concentrations (75, 100, 150 and 300 nM) by GeneSilencer (Genlantis)

following the manufacturer’s instructions. RNAi control experiments were

performed using a duplex siRNA against Luciferase (Dharmacon).

Statistical analysis. P-values were obtained using the two-tailed Student’s t-test.

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 1 9

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

ACKNOWLEDGMENTSThis work was supported by grants from the FIRB program of the ‘‘Ministerodell’Istruzione, Universita’ e Ricerca,’’ Italy and from the ‘‘Fondazione CRTrieste’’of Trieste, Italy. The authors are indebted to H. Masai (Tokyo MetropolitanInstitute of Medical Science) for helpful discussion and to A. Dutta (University ofVirginia), K. Helin (Biotech Research and Innovation Centre and Centre forEpigenetics), M. Benkirane (Institut de G–enetique Humaine), J. Pines (WellcomeTrust/Cancer Research UK Gurdon Institute) and H. Masai for the gift ofreagents. The authors are grateful to V. Liverani for excellent technicalsupport and to S. Kerbavcic for superb editorial assistance.

AUTHOR CONTRIBUTIONSAll experiments were performed by R.P. and R.M.-M.; A.C. took part in thedesign of the initial CDC6 acetylation experiments; M.G. supervised the work andwrote the manuscript.

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Lei, M. & Tye, B.K. Initiating DNA synthesis: from recruiting to activating the MCMcomplex. J. Cell Sci. 114, 1447–1454 (2001).

2. Bell, S.P. The origin recognition complex: from simple origins to complex functions.Genes Dev. 16, 659–672 (2002).

3. Blow, J.J. & Dutta, A. Preventing re-replication of chromosomal DNA. Nat. Rev. Mol.Cell Biol. 6, 476–486 (2005).

4. Nishitani, H. & Lygerou, Z. Control of DNA replication licensing in a cell cycle. GenesCells 7, 523–534 (2002).

5. Nasmyth, K. Viewpoint: putting the cell cycle in order. Science 274, 1643–1645(1996).

6. Diffley, J.F. Regulation of early events in chromosome replication. Curr. Biol. 14,R778–R786 (2004).

7. Dutta, A. & Bell, S.P. Initiation of DNA replication in eukaryotic cells. Annu. Rev. CellDev. Biol. 13, 293–332 (1997).

8. Stillman, B. Initiation of eukaryotic DNA replication in vitro. Annu. Rev. Cell. Biol.1989, 197–245 (1989).

9. Leatherwood, J. Emerging mechanisms of eukaryotic DNA replication initiation. Curr.Opin. Cell Biol. 10, 742–748 (1998).

10. Coleman, T.R., Carpenter, P.B. & Dunphy, W.G. The Xenopus Cdc6 protein is essentialfor the initiation of a single round of DNA replication in cell-free extracts. Cell 87,53–63 (1996).

11. Tanaka, T., Knapp, D. & Nasmyth, K. Loading of an Mcm protein onto DNA replicationorigins is regulted by Cdc6p and CKDs. Cell 90, 649–660 (1997).

12. Donovan, S., Harwood, J., Drury, L.S. & Diffley, J.F. Cdc6p-dependent loading of Mcmproteins onto pre-replicative chromatin in budding yeast. Proc. Natl. Acad. Sci. USA94, 5611–5616 (1997).

13. Jiang, W., Wells, N.J. & Hunter, T. Multistep regulation of DNA replication by Cdkphosphorylation of HsCdc6. Proc. Natl. Acad. Sci. USA 96, 6193–6198 (1999).

14. Petersen, B.O., Lukas, J., Sorensen, C.S., Bartek, J. & Helin, K. Phosphorylation ofmammalian CDC6 by cyclin A/CDK2 regulates its subcellular localization. EMBO J.18, 396–410 (1999).

15. Petersen, B.O. et al. Cell cycle- and cell growth-regulated proteolysis of mammalianCDC6 is dependent on APC-CDH1. Genes Dev. 14, 2330–2343 (2000).

16. Saha, P. et al. Human CDC6/Cdc18 associates with Orc1 and Cyclin-CDK and isselectively eliminated from the nucleus at the onset of S-phase. Mol. Cell. Biol. 18,2758–2767 (1998).

17. Williams, R.S., Shohet, R.V. & Stillman, B. A human protein related to yeast Cdc6p.Proc. Natl. Acad. Sci. USA 94, 142–147 (1997).

18. Mailand, N. & Diffley, J.F. CDKs promote DNA replication origin licensing in human cellsby protecting Cdc6 from APC/C-dependent proteolysis. Cell 122, 915–926 (2005).

19. Mendez, J. & Stillman, B. Chromatin association of human origin recognition complex,CDC6, and minichromosome maintenance proteins during the cell cycle: assemblyof prereplication complexes in late mitosis. Mol. Cell. Biol. 20, 8602–8612 (2000).

20. Delmolino, L.M., Saha, P. & Dutta, A. Multiple mechanisms regulate subcellularlocalization of human CDC6. J. Biol. Chem. 276, 26947–26954 (2001).

21. Herbig, U., Griffith, J.W. & Fanning, E. Mutation of Cyclin/CDK phosphorylation sites inHsCdc6 disrupts a late step in initiation of DNA replication in human cells. Mol. Biol.Cell 11, 4117–4130 (2000).

22. Alexandrow, M.G. & Hamlin, J.L. Cdc6 chromatin affinity is unaffected by serine-54phosphorylation, S-phase progression, and overexpression of cyclin A. Mol. Cell. Biol.24, 1614–1627 (2004).

23. Timmers, H.T. & Tora, L. SAGA unveiled. Trends Biochem. Sci. 30, 7–10 (2005).24. Burgess, S.M., Ajimura, M. & Kleckner, N. GCN5-dependent histone H3 acetylation

and RPD3-dependent histone H4 deacetylation have distinct, opposing effects onIME2 transcription, during meiosis and during vegetative growth, in budding yeast.Proc. Natl. Acad. Sci. USA 96, 6835–6840 (1999).

25. Howe, L. et al. Histone H3 specific acetyltransferases are essential for cell cycleprogression. Genes Dev. 15, 3144–3154 (2001).

26. Krebs, J.E., Kuo, M.H., Allis, C.D. & Peterson, C.L. Cell cycle-regulated histone acetyl-ation required for expression of the yeast HO gene. Genes Dev. 13, 1412–1421 (1999).

27. Zhang, W., Bone, J.R., Edmondson, D.G., Turner, B.M. & Roth, S.Y. Essential andredundant functions of histone acetylation revealed by mutation of target lysines andloss of the Gcn5p acetyltransferase. EMBO J. 17, 3155–3167 (1998).

28. Paulson, M., Press, C., Smith, E., Tanese, N. & Levy, D.E. IFN-stimulated transcriptionthrough a TBP-free acetyltransferase complex escapes viral shutoff. Nat. Cell Biol. 4,140–147 (2002).

29. Palhan, V.B. et al. Polyglutamine-expanded ataxin-7 inhibits STAGA histone acetyl-transferase activity to produce retinal degeneration. Proc. Natl. Acad. Sci. USA 102,8472–8477 (2005).

30. Takei, Y. et al. MCM3AP, a novel acetyltransferase that acetylates replication proteinMCM3. EMBO Rep. 2, 119–123 (2001).

31. Zarkowska, T. & Mittnacht, S. Differential phosphorylation of the retinoblastomaprotein by G1/S cyclin-dependent kinases. J. Biol. Chem. 272, 12738–12746(1997).

32. Galbiati, L., Mendoza-Maldonado, R., Gutierrez, M.I. & Giacca, M. Regulation of E2F–1 after DNA damage by p300-mediated acetylation and ubiquitination. Cell Cycle 4,930–939 (2005).

33. Takahashi, T., Ohara, E., Nishitani, H. & Masukata, H. Multiple ORC-binding sites arerequired for efficient MCM loading and origin firing in fission yeast. EMBO J. 22,964–974 (2003).

34. Muratoglu, S. et al. Two different Drosophila ADA2 homologues are present in distinctGCN5 histone acetyltransferase-containing complexes. Mol. Cell. Biol. 23, 306–321(2003).

35. Guelman, S. et al. Host cell factor and an uncharacterized SANT domain protein arestable components of ATAC, a novel dAda2A/dGcn5-containing histone acetyltransfer-ase complex in Drosophila. Mol. Cell. Biol. 26, 871–882 (2006).

36. Vaziri, C. et al. A p53-dependent checkpoint pathway prevents rereplication. Mol. Cell11, 997–1008 (2003).

37. Nishitani, H., Lygerou, Z. & Nishimoto, T. Proteolysis of DNA replication licensingfactor Cdt1 in S-phase is performed independently of geminin through its N-terminalregion. J. Biol. Chem. 279, 30807–30816 (2004).

38. Drury, L.S., Perkins, G. & Diffley, J.F. The Cdc4/34/53 pathway targets Cdc6p forproteolysis in budding yeast. EMBO J. 16, 5966–5976 (1997).

39. Duursma, A. & Agami, R. p53-dependent regulation of Cdc6 protein stability controlscellular proliferation. Mol. Cell. Biol. 25, 6937–6947 (2005).

40. Coverley, D., Laman, H. & Laskey, R.A. Distinct roles for cyclins E and A duringDNA replication complex assembly and activation. Nat. Cell Biol. 4, 523–528(2002).

41. Pelizon, C., Madine, M.A., Romanowski, P. & Laskey, R.A. Unphosphorylatablemutants of Cdc6 disrupt its nuclear export but still support DNA replication onceper cell cycle. Genes Dev. 14, 2526–2533 (2000).

42. Oehlmann, M., Score, A.J. & Blow, J.J. The role of Cdc6 in ensuring complete genomelicensing and S phase checkpoint activation. J. Cell Biol. 165, 181–190 (2004).

43. Clay-Farrace, L., Pelizon, C., Santamaria, D., Pines, J. & Laskey, R.A. Humanreplication protein Cdc6 prevents mitosis through a checkpoint mechanism thatimplicates Chk1. EMBO J. 22, 704–712 (2003).

44. Kikuchi, H., Takami, Y. & Nakayama, T. GCN5: a supervisor in all-inclusive control ofvertebrate cell cycle progression through transcription regulation of various cell cycle-related genes. Gene 347, 83–97 (2005).

45. Iizuka, M., Matsui, T., Takisawa, H. & Smith, M.M. Regulation of replication licensingby acetyltransferase Hbo1. Mol. Cell. Biol. 26, 1098–1108 (2006).

46. Burke, T.W., Cook, J.G., Asano, M. & Nevins, J.R. Replication factors MCM2 and ORC1interact with the histone acetyltransferase HBO1. J. Biol. Chem. 276, 15397–15408(2001).

47. Iizuka, M. & Stillman, B. Histone acetyltransferase HBO1 interacts with the ORC1subunit of the human initiator protein. J. Biol. Chem. 274, 23027–23034 (1999).

48. Yang, X.J. Multisite protein modification and intramolecular signaling. Oncogene 24,1653–1662 (2005).

49. Bode, A.M. & Dong, Z. Post-translational modification of p53 in tumorigenesis. Nat.Rev. Cancer 4, 793–805 (2004).

50. Matsuzaki, H. et al. Acetylation of FoxO1 alters its DNA-binding ability and sensitivityto phosphorylation. Proc. Natl. Acad. Sci. USA 102, 11278–11283 (2005).

51. Ozaki, T. et al. Functional implication of p73 protein stability in neuronal cell survivaland death. Cancer Lett. 228, 29–35 (2005).

52. Vervoorts, J., Luscher-Firzlaff, J.M. & Luscher, B. The ins and outs of MYC regulationby posttranslational mechanisms. J. Biol. Chem. 281, 34725–34729 (2006).

53. Chan, H.M., Krstic-Demonacos, M., Smith, L., Demonacos, C. & La Thangue, N.B.Acetylation control of the retinoblastoma tumour-suppressor protein. Nat. Cell Biol. 3,667–674 (2001).

54. Marzio, G. et al. E2F family members are differentially regulated by reversibleacetylation. J. Biol. Chem. 275, 10887–10892 (2000).

55. Marcello, A., Massimi, P., Banks, L. & Giacca, M. Adeno-associated virus type 2 repprotein inhibits human papillomavirus type 16 E2 recruitment of the transcriptionalcoactivator p300. J. Virol. 74, 9090–9098 (2000).

56. Marcello, A. et al. Recruitment of human cyclin T1 to nuclear bodies through directinteraction with the PML protein. EMBO J. 22, 2156–2166 (2003).

ART IC L E S

42 0 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Regulation of active site coupling in glutamine-dependent NAD+ synthetaseNicole LaRonde-LeBlanc1,2, Melissa Resto1,2 & Barbara Gerratana1

NAD+ is an essential metabolite both as a cofactor in energy metabolism and redox homeostasis and as a regulator of cellularprocesses. In contrast to humans, Mycobacterium tuberculosis NAD+ biosynthesis is absolutely dependent on the activity of amultifunctional glutamine-dependent NAD+ synthetase, which catalyzes the ATP-dependent formation of NAD+ at the synthetasedomain using ammonia derived from L-glutamine in the glutaminase domain. Here we report the kinetics and structuralcharacterization of M. tuberculosis NAD+ synthetase. The kinetics data strongly suggest tightly coupled regulation of the catalyticactivities. The structure, the first of a glutamine-dependent NAD+ synthetase, reveals a homooctameric subunit organizationsuggesting a tight dependence of catalysis on the quaternary structure, a 40-A intersubunit ammonia tunnel and structuralelements that may be involved in the transfer of information between catalytic sites.

The importance of NAD+ as the ubiquitous and essential cofactor ofenzymes involved in reduction-oxidation reactions has long beenrecognized1. Recently, nonredox functions of NAD+ have been dis-covered in DNA repair, telomere maintenance, gene silencing, celllongevity, immune response and Ca2+ signaling1. Depletion of cellularNAD+ is deleterious to cellular metabolism2,3. NAD+ metabolismcomprises a network of de novo biosynthetic and recycling pathwaysthat may differ drastically among organisms4–6. The ubiquitous NAD+

synthetase catalyzes the last step in the de novo biosynthesis and, insome recycling pathways, the ATP-dependent transformation of nico-tinic acid adenine dinucleotide (NaAD+) to NAD+ (Fig. 1). Because inhumans some recycling pathways are not dependent on NAD+

synthetase4,7 and M. tuberculosis NAD+ synthetase (NAD+ synthe-taseTB) is essential for NAD+ production in M. tuberculosis8, thisenzyme is an important potential drug target. Inhibitors of NAD+

synthetaseTB have the potential to be used not only in chemo-therapeutic treatments against active tuberculosis but also againstthe tuberculosis reservoir represented by the 2 billion asymptomati-cally infected people8,9.

The two types of NAD+ synthetase are the ammonia-dependentenzyme (NAD+ synthetaseNH3), which is present only in prokar-yotes and uses ammonia as the nitrogen source, and the glutamine-dependent enzyme (NAD+ synthetaseGln), which is present ineukaryotes and prokaryotes including M. tuberculosis5 and usesL-glutamine. The structure of the monofunctional and homodi-meric NAD+ synthetaseNH3 has been solved with differentligands10–14, but no structure of NAD+ synthetaseGln has beenreported. NAD+ synthetaseGln is a multifunctional enzyme: eachsubunit has an N-terminal glutamine amidotransferase (GAT)

domain that hydrolyzes L-glutamine to L-glutamate and NH3 and aC-terminal synthetase domain that carries out the ATP-dependentamidation of NaAD+ (ref. 15).

NAD+ synthetaseGln is the sole member of a new family of theGAT enzymes with a nitrilase-like (C-N bond cleaving) glutaminasedomain16. GAT enzymes catalyze key metabolic reactions in pro-tein, amino acid, cofactor, purine and pyrimidine biosyntheses17,18,characterized by three main events: glutamine hydrolysis, ammoniatransfer from one active site to the other and formation of thesynthase or synthetase product. In some GAT enzymes, the activesites for glutamine hydrolysis and for product formation areconnected by a molecular tunnel implicated in the transfer of thehighly reactive ammonia19,20. Kinetics studies have shown thatthe glutaminase and synthetase activities are in some cases tightlycoupled in spite of their spatial separation18. No common mechan-ism has been identified for the regulation of ammonia transferand catalysis.

The kinetics characterizations reported here establish tightly regu-lated coupling of the catalytic activities of NAD+ synthetaseTB andidentify the formation of the synthetase intermediate (NaAD-AMP) asthe trigger for glutaminase activation. These studies highlight impor-tant differences between the prokaryotic and eukaryotic NAD+

synthetaseGln, with potential implications for drug development. Tounderstand the structural basis for the observed active site coupling ofNAD+ synthetaseGln, we also determined the crystal structure ofNAD+ synthetaseTB. The structure reveals a complex subunit organi-zation and an unprecedented intersubunit ammonia tunnel and, alongwith the results of mutagenesis studies, provides insights on theregulation of active site coupling.

Received 15 September 2008; accepted 27 January 2009; published online 8 March 2009; doi:10.1038/nsmb.1567

1Department of Chemistry and Biochemistry, University of Maryland, College Park, Maryland 20742-2021, USA. 2These authors contributed equally to this work.Correspondence should be addressed to B.G. (bgerrata@umd.edu) or N.L.-L. (nlaronde@umd.edu).

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 2 1

ART IC L E S

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

RESULTSKinetics analysis shows optimal kinetics synergismWe measured the steady-state rate constants for the NAD+ syntheta-seTB–catalyzed reactions using exogenous ammonia or the ammoniaproduced by the glutamine hydrolysis as described in the Supplemen-tary Methods online and reported in Table 1. The turnover numberfor the reaction using ammonia is at least five- to six-fold faster,considering the difference in ionic strength, than for the reaction withL-glutamine. A rate-limiting step involved in L-glutamine hydrolysisand/or ammonia transport could account for this difference (Fig. 1).The steady-state kinetics data support a mechanism of catalysis driven

by kinetic synergism. Decreases in the KM

values of 5- and 12-fold for NaAD+ andATP, respectively, when L-glutamine is thenitrogen source instead of exogenous ammo-nia suggest that L-glutamine binding enhancesthe affinity of the synthetase active site for itssubstrates. A 179-fold increase in the glutami-nase turnover number measured for the reac-tion in the presence of both synthetasesubstrates compared to their absence indicatesa mechanism of activation of the glutaminasedomain triggered by changes occurring at thesynthetase domain. The KM values for L-glu-tamine did not change in the presence orabsence of NaAD+ and ATP. The kcat values

for L-glutamine hydrolysis in the presence of only one of the synthetasesubstrates (kcat ¼ 0.004 ± 0.0003 s�1 and 0.0037 ± 0.0002 s�1 with ATPor NaAD+, respectively) are identical to the value measured in theabsence of synthetase substrates (kcat ¼ 0.0038 ± 0.0002 s�1), indicatingthat both substrates are necessary for glutaminase activation.

Full occupancy of the synthetase domain may be necessary andsufficient for the observed activation of the glutaminase domain or itcould depend on the formation of the NaAD-AMP intermediate. Todistinguish between these two possible mechanisms, we tested theability of AMPCPP, a competitive inhibitor of ATP (Ki ¼ 0.7 ±0.1 mM), to activate the glutaminase domain in the presence of

Synthetase domain

Glutaminase domain

ADP

+

ADP ADP

OH OH OH OHATP/Mg2+

NaAD+ NAD+

AMP/Mg2+ + PPi/Mg2+

NaAD-AMP

O

O-

O

O

O- O

O

O- O

O

O- O

O-

O

O O

O-O P O

O

O

O

NH2

NH3

NH3+

NH3+ NH3

+NH3

S-Cys-E E-Cys

γ-glutamyl thioester intermediate L-Glutamate

NH2E-Cys

L-Glutamine

+

Ade

OHOH

OH OH+

+

N+ N+ N+

Figure 1 The reactions catalyzed at the synthetase and glutaminase domains of NAD+ synthetaseGln.

Table 1 Steady-state kinetic parameters of the wild-type, D656A and C176A NAD+ synthetaseTB–catalyzed reactionsa

Glutamine-dependent wild-type–catalyzed reaction (ionic strength 0.012 M)

Assay Variable substrate Fixed substrates KM (mM)b kcat (s–1) kcat/KM (s�1 mM–1)

Wild-type NAD+ synthetaseTB

NAD+ NaAD+ Gln, ATP 0.13 ± 0.02 0.50 ± 0.02 3.8 ± 0.6

NAD+ ATP Gln, NaAD+ 0.12 ± 0.03 0.60 ± 0.03 5.0 ± 1.0

NAD+ Gln NaAD+, ATP 1.3 ± 0.1 0.55 ± 0.01 0.42 ± 0.03

Glu Gln NaAD+, ATP 1.5 ± 0.3 0.68 ± 0.03 0.4 ± 0.1

Glu Gln None 1.3 ± 0.4 0.0038 ± 0.0002 0.0029 ± 0.0008

D656A NAD+ synthetaseTB

NAD+ Gln NaAD+, ATP 7.8 ± 1.4 0.012 ± 0.0007 0.0015 ± 0.0003

Glu Gln NaAD+, ATP 9.3 ± 1.5 0.048 ± 0.003 0.051 ± 0.0009

Glu Gln None 0.1 ± 0.05 0.0055 ± 0.0003 0.055 ± 0.02

Ammonia-dependent reaction (ionic strength 0.1 M)

Assay Variable substrate Fixed substrates KM (mM)b kcat (s–1) kcat/KM (s–1 mM–1)

Wild-type NAD+ synthetaseTB

NAD+ NaAD+ NH3, ATP 0.7 ± 0.1 2.6 ± 0.1 3.7 ± 0.5

NAD+ ATP NH3, NaAD+ 1.4 ± 0.2 2.9 ± 0.1 2.1 ± 0.3

NAD+ NH3 NaAD+, ATP 20 ± 2 3.0 ± 0.1 0.15 ± 0.02

C176A NAD+ synthetaseTB

NAD+ NaAD+ NH3, ATP 0.65 ± 0.08 3.7 ± 0.1 5.7 ± 0.7

NAD+ ATP NH3, NaAD+ 0.60 ± 0.09 3.0 ± 0.1 5.0 ± 0.8

NAD+ NH3 NaAD+, ATP 15 ± 2 3.3 ± 0.1 0.22 ± 0.03

D656A NAD+ synthetaseTB

NAD+ NH3 NaAD+, ATP 0.089 ± 0.005 0.0045 ± 0.0001 0.051 ± 0.003

aThe enzyme activity was measured in 10 mM MgCl2, 1 mM DTT and 50 mM Tris-HCl, pH 8.3, at 37 1C. bThese values are apparent KM constants at saturating concentrations of the othertwo substrates.

ART IC L E S

42 2 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

NaAD+. We did not detect any activation (kcat ¼ 0.013 ± 0.002 s–1,3.4-fold increase). NAD+ synthetaseTB does not catalyze NAD+ for-mation in the presence of the ATP substrate analog AMPPCP, whichcontains the phosphodiester bond cleaved in the NAD+ synthetase–catalyzed reaction. The lack of AMPPCP activity can be attributed to anonproductive binding mode of AMPPCP. Similarly, the absence ofactivation by AMPCPP could be the result of a binding mode thatinhibits conformational changes required for glutaminase activity.Therefore, we tested the ability of product NAD+ to activate theglutaminase domain in the presence of ATP, but, again, we observedno appreciable increase (kcat ¼ 0.005 ± 0.001 s–1). Taken together,these results indicate that the activation of the glutaminase domain isdependent on the formation of the NaAD-AMP intermediate, con-firming that there is a catalytic coupling of active sites. The stoichio-metry of the products of the NAD+ synthetaseTB reaction reflects theefficiency of active site coupling and ammonia transfer (Table 2).These data show that NAD+ synthetaseTB achieves maximum effi-ciency at or below physiologically relevant concentration of L-gluta-mine (B4.5 mM21) and that the synthetase and glutaminase activitiesare highly synchronized. As the L-glutamine concentration increasedto 20 mM, the activities of the active sites become partially uncoupledand channeling efficiency is reduced to 73%.

The complex quaternary structure of NAD+ synthetaseTB

We determined the crystal structure of the 6-diazo-5-oxo-L-norleucine(DON)–modified NAD+ synthetaseTB in the presence of NaAD+ at2.35-A resolution. The crystals contained four subunits per asym-metric unit, and the expected biological homooctamer is generated byinteraction with an adjacent asymmetric unit related by two-foldcrystallographic symmetry (Fig. 2). We confirmed the 600-kDahomooctameric quaternary structure in solution by gel-filtrationanalysis (data not shown) and found it to be identical to thequaternary structure of the Saccharomyces cerevisiae NAD+ synthetase(NAD+ synthetaseyeast)22. The core of the NAD+ synthetaseGln octamerformed by two stacked central rings of four glutaminase domains isheld together by extensive contacts (1,687 A2) between glutaminasedomains on opposite rings (Fig. 2a). In addition, each glutaminasedomain contacts the synthetase domains of two adjacent subunits(2,002 A2 buried surface area per glutaminase domain) (Fig. 2b). Thecore is decorated by four dimers of synthetase domains at the cornersof a square. Each dimer buries 1,841 A2 of surface area per subunit andis similar to the dimer interface in the NAD+ synthetasesNH3 (ref. 12)(Supplementary Fig. 1 online). The synthetase domains are inclinedtoward the plane between two rings, and dimers are formed fromsynthetase domains of subunits from separate rings (Fig. 2a). Thus,the synthetase domains interlock the rings formed by the glutaminasedomains. As a result of the complex architecture, one subunit contactsfive other subunits in the octamer (Fig. 2).

The glutaminase domain and the glutamine tunnelBy SSM database search23, the N-terminal glutaminase is most similarto the nitrilase PH0642 from Pyrococcus horikoshii (PDB 1J31)24

(Supplementary Fig. 2 online). The glutaminase domain shares thecore structural elements of the nitrilase domain, including a centralb-sandwich surrounded by a-helices. The interface between glutami-nase domains on separated rings is similar to the dimer interface inPH0642 (Supplementary Fig. 2). In an isolated glutaminase domain

Table 2 Stoichiometry analysis for NAD+ synthetaseTB–catalyzed

reactiona

Wild-type NAD+ synthetaseTB

Glutamine concentration AMP/NAD+ Glu/ NAD+ % channel efficiency

1.5 mM n.d. 1.05 ± 0.05 95 ± 5

5.0 mM n.d. 1.12 ± 0.13 89 ± 12

20.0 mM 0.9 ± 0.2 1.36 ± 0.13 73 ± 10

Asp656Ala NAD+ synthetaseTB

Glutamine concentration AMP/NAD+ Glu/ NAD+ % channel efficiency

1.5 mM n.d. 3.1 ± 0.5 32 ± 5

5.0 mM n.d. 3.5 ± 0.4 28 ± 3

20.0 mM n.d. 3.5 ± 0.3 29 ± 3

aAll reactions were carried out in 10 mM MgCl2, 1 mM DTT and 50 mM Tris-HCl, pH 8.3, at37 1C and 0.23 mM NAD+ synthetase. n.d., not determined.

a

b

Figure 2 Oligomeric assembly of NAD+ synthetaseGln from M. tuberculosis.

(a) Quaternary structure of NAD+ synthetaseGln. Glutaminase domains (G1

to G8) line the central cavity, whereas synthetase domains (S1 to S8) are on

the outside. Glutaminase domains in the top ring are illustrated in ribbon

with a transparent surface representation. The subunits are color coded as

indicated in the model shown to the right. NaAD+ bound in the synthetase

active site is shown in CPK. The two opposing tetrameric rings are rotated

by roughly 501 relative to each other around the central four-fold axis. A 901

rotation about the x axis is shown below. Each glutaminase domain in the

ring has limited contacts with two other glutaminase domains located on

the same ring (709 A2 buried surface area per interface per subunit) and

extensive contacts (1,687 A2) with a single glutaminase domain in the

other ring. The glutaminase domains of subunits whose synthetase domains

interact do not interact directly. (b) Surface and ribbon representation of thefour subunits in the asymmetric unit. The synthetase and glutaminase active

sites are indicated by NaAD+ and DON in CPK, respectively. The white

arrows indicate the distances of the active site in G5 to the active sites in

S5 and in S1. There is only limited contact between the glutaminase and

synthetase domains within a monomer (918 A2 buried surface area per

domain per subunit).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 2 3

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

the glutaminase active site would be exposed to solvent, as in all otherGAT enzymes, but in the context of the oligomeric structure of NAD+

synthetaseTB, access to the glutaminase active site is limited. Visualinspection and CAVER analysis25 identified a tunnel of 33 A in length(referred to as the glutamine tunnel) that allows L-glutamine to reachthe glutaminase active site (Fig. 3a). The tunnel is formed by residuesfrom two different subunits. For example, the first two-thirds of theglutamine tunnel to the G5 active site (Fig. 1) is lined with residues ofthe G1 and S1 domains, which are mostly not conserved (Fig. 3a). Theremaining one-third of the glutamine tunnel (average radius 2.7 A)is lined mostly with conserved residues from G5 (SupplementaryFig. 3a online).

Enzymes of the nitrilase superfamily contain a Glu-Lys-Cys catalytictriad with the cysteine as the likely nucleophile for the glutaminasereaction16. In the structure reported here, the glutaminase domain ofNAD+ synthetaseTB was irreversibly covalently modified at the cysteinewith DON in a stable state that mimicks the g-glutamylthioesterintermediate (Figs. 1 and 3b). The glutaminase active site, identifiedby the covalently linked DON to Cys176TB, is located in a loop-richregion (Fig. 3c). The kinetic parameter for the inactivation reaction(kinac / Ki ¼ 0.6 ± 0.1 s–1 mM–1) indicates that NAD+ synthetaseTB

has similar specificity for DON and for L-glutamine (kcat / KM ¼0.4 s–1 mM–1; Table 1). Conformational flexibility in the active sites isindicated by multiple conformations and high B-factors when the fourCys176TB-DON adducts are compared (Supplementary Fig. 4 online).

Interactions occur only between the a-carboxylate and a-aminogroups of this adduct with the side chains of Glu177TB andTyr127TB (Fig. 3c and Supplementary Fig. 4). The positions of theother two residues of the catalytic triad suggest that Lys121TB stabilizesthe negative charge in the transition state for the formation of theglutamylthioester covalent intermediate26, and Glu52TB abstracts aproton from Cys176 and donates it to the ammonia leaving group(Figs. 1 and 3c). An ordered solvent molecule is found in closeproximity to the sulfur of all the Cys176TB-DON adducts and thehydroxyl of Glu52TB. Mutation of either Lys121TB or Glu52TB toalanine abolishes any glutaminase activity27.

To determine whether the glutaminase domain acts reciprocally tocontrol the ammonia-dependent NAD+ synthesis, we measured the

Glu177Tyr127

Phe130

Glu132

Glu52

DON0 mM0.1 mM

Tyr58

DON

Phe180

Lys121

Cys176

O

O O

–O–O–O

OO

250kinac = 0.21 ± 0.02 sec–1

Ki = 0.35 ± 0.06 mM200

150

100

NA

D+ p

rodu

ced

(µM

)

50

0 10 20 30Time (min)

40 50 600

OO-

O

HN + N +

N– N E E

NH3+ NH3

+

NH3+ -N2

52E

176S176S

AH

DON

0.25 mM

0.5 mM1 mM1.5 mM

2 mM4 mM5 mM

a

b d

c Figure 3 DON inhibition of the glutaminase domain. (a) Solid surface

representation of the 33-A glutamine tunnel. Subunits 1 and 5 are in

magenta and green, respectively. DON is shown in CPK. The ammonia

tunnel is shown in mesh, with the first solvent molecule identified in it

shown in cyan sphere. (b) Mechanism of the irreversible inhibition by DON

of the glutaminase domain. (c) Close-up view of the glutaminase active site

(green sticks) with Cys176 covalently modified by DON (white sticks).

In all eight glutaminase domains a Cys176TB-DON adduct was observed,

confirming the assignment of Cys176 as the residue that acts as the

nucleophile in the glutaminase reaction. The first solvent molecule identified

in the ammonia tunnel is shown in cyan sphere. Because the crystals were

grown in 1.8 M ammonium citrate, some of the density modeled as water

could be ammonia. (d) Progress curves for the DON irreversible inhibition of

NAD+ synthetaseTB. kinact, the maximal rate of inactivation, and Kiapp, the

apparent inhibition constant, were obtained by fitting to kobs ¼ kinact [I]/{Ki

app (1 + [S]/Km + [I]/Kiapp} a plot of kobs values versus the concentrations

of DON. kobs values were obtained by a global fit of the progress curves to

[P] ¼ [P]N (1 – e–kobst).

Figure 4 The homooctameric structure of NAD+ synthetaseGln with all eight

intersubunit tunnels. Half of the subunits are shown in transparent surface

and are color coded as in Figure 2a. The remaining subunits (3, 4, 5 and 8)

are not shown for clarity. The tunnels identified by CAVER were not

continuous between the glutaminase and synthetase active sites in all

subunits and the gap is indicated by *. NaAD+ and DON are shown in CPK.

DON (indicated by a wide circle) is buried in the glutaminase domain. Theblack arrows track the CAVER-identified tunnels with glutamine reaching the

glutaminase active site and ammonia departing from this site to the

synthetase active site where NAD+ is formed. G and S indicate the entrance

of glutamine and the exit of NAD+, respectively, in the representation shown

below obtained by a 901 rotation.

ART IC L E S

42 4 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

catalytic properties of the DON-modified enzyme and of the C176Avariant. The C176A variant lacked glutamine-dependent activity butretained ammonia-dependent activity (Table 1). However, the DON-modified enzyme showed only 18% (kcat ¼ 0.48 ± 0.01 s–1) of theammonia-dependent activity of the wild type. This decrease in theactivity could be attributed to the inhibition of conformationalchanges required for the synthetase activity due to the increasedrigidity of the DON-modified enzyme and/or to a limited access tothe synthetase active site through the glutaminase domain arisingfrom the presence of DON (Fig. 3a).

The intersubunit ammonia tunnel and the synthetase domainGlutaminase and synthase active sites in other GAT enzymes areseparated by 18–40 A18–20. In some cases, ammonia transfers through

a molecular tunnel between active sites18–20. In all previous structures ofmultimeric enzymes in which a single polypeptide contains glutami-nase and synthase domains, the tunnel lies within one protomer18–20.The glutaminase and synthetase active sites within a single subunit ofthe NAD+ synthetaseGln are located B55 A from each other (Fig. 2b),B20 A further than the distance between the active sites that areclosest to each other, for example, S1 and G5 (Fig. 2b). In addition tothe glutamine tunnel described above, CAVER25 identified a secondtunnel that connects the closest synthetase and glutaminase activesites, for example, (S1 and G5, Figs. 4 and 5a). The two tunnels jointogether to form a 73-A U-shaped tunnel (Fig. 4). CAVER did notidentify a tunnel between the G1 and S1 active sites. The intersubunitammonia tunnel extends 40 A from G5Cys176TB in the glutaminasedomain to the carboxyl group of NaAD+ in the synthetase domain and

has an average radius of 2.0 A, wide enoughfor an ammonia molecule28. The first thirdfrom the glutaminase active site of this inter-subunit ammonia tunnel is lined with resi-dues from the glutaminase domain (G5), andthe remaining two-thirds is lined with resi-dues of the synthetase domain (S1) (Fig. 5a,band Supplementary Fig. 3b). Ammoniacould be transferred through a hydrogen-bonding network starting at G5Tyr58 andending at G5Asp241TB at the G5-S1 interfaceof the ammonia tunnel (Fig. 5b).

As a result of the tunnel’s convoluted Ushape, one short loop (residues 127–131, theYRE loop) conserved in mycobacterium andeukaryotes (Supplementary Fig. 3b) contactsthe DON molecule (G5Tyr127TB) and threedifferent sections of the tunnel (Fig. 6). Nearthe glutaminase active site, G5Phe130TB formswith two other hydrophobic residues thefirst constriction (0.8 A) of the ammoniatunnel (Figs. 5b and 6). G5Arg128TB interactsat the S1-G5 interface in the glutaminetunnel, whereas G5Glu129TB and G5Tyr131TB

are located at the S1-G5 interface in theammonia tunnel (Figs. 5b and 6). Interac-tions among G5Glu129TB, S1Lys644TB andS1Asp656TB, conserved residues in prokaryo-tic enzymes, mediate the transition from theG5 to the S1 section of the ammonia tunnel(Figs. 5b and 6). Mutation of S1Asp656TB toalanine (S1D656ATB) inhibits activation of theglutaminase domain in the reaction with thesynthetase substrates, as evidence by the only9-fold activation of this variant compared tothe 179-fold activation measured for the wildtype (Table 1). The basal glutaminase turn-over was essentially the same for theS1D656ATB and wild-type reactions. In addi-tion, ammonia transfer from one active siteto the other in the reaction catalyzed byS1D656ATB was significantly compromised,as evidenced by the only B30% maximumchannel efficiency of this variant comparedto the 100% maximum channel efficiencymeasured in the wild-type catalyzed reac-tion (Table 2). Therefore, disruption of the

a

c

b

Figure 5 The ammonia tunnel and the synthetase active site. Subunits 1, 5 and 6 are in magenta, green

and orange, respectively. (a) Solid surface representation of the G5-S1 ammonia tunnel. Large black

arrows indicate the direction of ammonia transfer in this tunnel and the entrance of glutamine in the

glutaminase active site through the glutamine tunnel (mesh). DON and NaAD+ are shown in CPK,

whereas ordered solvent molecules identified are shown in cyan spheres. The two small black arrows

indicate the hydrophobic constrictions identified in the ammonia tunnel. (b) A schematic of the inter-

actions with solvent molecules (cyan spheres) identified in the ammonia tunnel. Residues are coloredaccording to the subunit to which they belong. Residues invariant among the prokaryotic glutamine-

dependent enzymes are indicated by * and those also conserved in the eukaryotic enzymes by **. (c) Fo

– Fc map contoured at 1.0 s of NaAD+ in a close-up view of the synthetase active site. Active site

residues shown in sticks are colored according to the subunit to which they belong. Weak density is

observed for the nicotinic acid ring in all the synthetase domains in our structure, indicating flexibility of

this moiety in the active site. This has been observed in other structures of the NAD+ synthetaseNH3 with

NaAD+ bound as well14, indicating a need for further stabilization, perhaps by binding of the ATP

molecule and/or conformational changes of the flexible loop. The location of the ATP molecule

(transparent sticks) was assigned by superposition with the NAD+ synthetaseNH3 structure with NaAD+

and ATP/Mg2+ bound (PDB 1EE1)14. Mg2+ and an ordered solvent molecule are shown as a transparent

green sphere and a cyan sphere, respectively.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 2 5

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

interactions at the S1-G5 interface and loop YRE yields an enzymethat is markedly wasteful of glutamine and with a substantiallycompromised ability to transfer ammonia between active sites.

A second set of contacts between S6Arg658TB with the side chain ofG5Glu61TB and the backbone carbonyl of S1Arg654TB mediates thetransition from the G5 to the S1 section of the ammonia tunnel(Fig. 5a,b). Thus, a residue from a third subunit, S6, is part of the G5-S1 ammonia tunnel. The S1 section of the tunnel closer to theglutaminase domain is lined with polar residues and contains orderedsolvent molecules. The tunnel narrows to 1.0 A in a second hydro-phobic patch, composed of S1Leu486TB, S1Leu639TB, S1Ala506TB andS1Pro640TB and void of solvent molecules (Fig. 5a,b), to resume itshydrophilic character near the synthetase active site. This combinationof hydrophobic and hydrophilic surfaces of the tunnel differs from theammonia tunnels in all other GAT enzymes, which are either uni-formly hydrophobic18–20,29 or hydrophilic19,28.

At the end of the ammonia tunnel in the synthetase active site theconserved residue S1Asp497TB, close to NaAD+, is positioned to act asa general base for activation of an ammonium ion to ammonia, if anammonium ion is transferred, or as the last site for the hydrogen-bonding interaction if ammonia is transferred (Fig. 5a,b). This rolewas assigned to the Asp173Bac in NAD+ synthetasesNH3 (ref. 13).

The synthetase domain of NAD+ synthetaseTB belongs to the N-type ATP pyrophosphatase family present in two other GAT enzymes,GMP and asparagine synthetases30,31. However, the similarity islimited to the ATP binding site, and the remaining structural elementsamong these synthetase domains are substantially different (Supple-mentary Fig. 5 online). The complete active unit consists of a dimerof two synthetase domains. Whereas the secondary-structural ele-ments in the NAD+ synthetaseNH3, including the dimerization inter-face, are conserved in the synthetase domain, NAD+ synthetaseGln

contains two unique additional helices, a18 and a20, which arerequired for oligomeric assembly (Supplementary Fig. 6 online).Many of the conserved a-helices, most notably a9, are offset and/orlonger in the synthetase domain of NAD+ synthetaseTB compared tothat of the ammonia-dependent enzyme. The extended N-terminalhelical segment of a9 (residues 333–339S1), which is absent in theNAD+ synthetasesNH3 (Fig. 6 and Supplementary Fig. 1), contacts theglutamine tunnel. The C terminus of a9 is located at the binding siteof the S6NaAD+ molecule in the synthetase dimer and contacts theammonia tunnel through S1Arg354TB and S1Gly350TB, which areconserved in the prokaryotic NAD+ synthetasesGln (Fig. 6 and

Supplementary Fig. 3). These residues form hydrogen bonds withthe side chain of S1Asn641TB, whose backbone atoms line theammonia tunnel. Notably, S1Asn641TB is located on a loop liningthe tunnel that contains two of the residues that contribute to theconstriction (constriction 2) in the ammonia tunnel near the synthe-tase active site (Fig. 6). a9 is a connecting structural element that islikely to contribute to regulation of the reactions at the synthetase andglutaminase active sites.

The synthetase domain contains three regions for which weobserved no electron density. Region 1 (residues 402–407, equivalentto loop P1 in the ammonia-dependent enzymes10) acts as a lid whenATP is bound in the ATP binding site and is disordered in the absenceof ATP. A similar role for loop P1 in the synthetase domain of NAD+

synthetasesGln is likely, given the structural similarity of the ATPbinding sites among NAD+ synthetases. Region 2—residues 442–450on the surface of the synthetase domain—is disordered in half of thesynthetase domains in the octamer. Ordering in the other subunits isprobably due to crystal packing. Region 3—residues 542–558, corre-sponding to loop P2 in NAD+ synthetaseNH3—is also always dis-ordered when the ATP binding site is empty. Loop P2 ordering isprobably required to protect the NaAD-AMP intermediate fromhydrolysis14. The N terminus of a17 is connected to loop P2 in theNAD+ synthetasesNH3 and contacts the NaAD+ biding site by hydro-phobic interaction between S1Phe564TB and S1Trp490TB (Fig. 6). Atthe C terminus of a17, the conserved S1Arg576TB interacts withG5Arg128TB from the YRE loop in the glutamine tunnel (Fig. 6).Therefore, a17 is another structural element that is likely to contributeto the regulation of catalysis in NAD+ synthetasesTB.

DISCUSSIONNAD+ synthetase is essential for NAD+ biosynthesis and for thesurvival of both replicating and nonreplicating M. tuberculosis8.Although NAD+ synthetase–independent biosynthetic pathways arepresent in humans4,7, it is unknown whether these pathways can satisfyNAD+ requirements in the absence of NAD+ synthetase. NAD+

synthetaseTB shares only 23% sequence identity with both the yeastand human orthologs, which themselves, in contrast, are 58.4%identical. Kinetic characterizations of NAD+ synthetaseGln have beenreported for only the NAD+ synthetaseyeast (ref. 32). This enzymeshows a modest activation of the glutaminase domain (B15-fold basedon the basal glutaminase kcat from Fig. 4 of ref. 32) that is induced byNaAD+ and not ATP32. NAD+ synthetaseyeast reaches a maximumchanneling efficiency, an indication of active site coupling, of 60%32. Incontrast to NAD+ synthetaseyeast, the activities of the two domains inNAD+ synthetaseTB are more efficiently regulated, leading to negligiblewaste of L-glutamine. In this study we show that the mechanism ofactivation of the glutaminase domain in NAD+ synthetaseTB is different

Figure 6 Connecting elements between glutaminase and synthetase active

sites. Cartoon representation of a-helices (a9, a14 and a17) that connect

the synthetase active site to the glutaminase active site and the tunnel.

Residues involved in interactions between elements are labeled. The

C terminus of a14 is connected to the loop containing S1WTrp490TB, which

is involved in the binding of the nicotinic acid moiety of NaAD+, whereasS1Leu486TB from the center of the helix forms constriction 2. Synthetase

domain residues, which form constriction 2 (indicated by *), and

glutaminase domain residues of the YRE loop, which form constriction 1

(indicated by *), are shown in yellow sticks. Yellow spheres indicate the

positions of key hydrophobic residues that bind to NaAD+ or may sense

interaction at the end of a17. DON and NaAD+ are shown in CPK.

Ammonia and glutamine tunnels are shown in gray solid surface and

mesh, respectively.

ART IC L E S

42 6 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

from that reported for NAD+ synthetaseyeast (ref. 32). The differences inmechanism and sequence between these enzymes encourage thedevelopment of specific NAD+ synthetaseTB inhibitors.

The structure of NAD+ synthetaseTB reported here shows a fasci-nating contrast between extensive intersubunit domain interactionsand relatively limited intrasubunit domain interactions, resulting in anintersubunit ammonia tunnel that is unprecedented among GATenzymes. The ammonia tunnel is unlike any other ammonia tunnelsreported, with its mixed hydrophilic and hydrophobic character20.Constrictions at the hydrophobic patches prevent the transfer ofammonia in the characterized structure, which contains a half-occupied synthetase active site. This is consistent with the hypo-thesis that ammonia transfer occurs only after the formation of theNaAD-AMP intermediate. Notably, Leu486TB and Pro640TB located atconstriction 2 and Phe130TB located at constriction 1 are conservedonly in prokaryotic organisms, which may indicate a difference in theregulation of ammonia transfer between prokaryotic and eukaryoticNAD+ synthetases. Molecular dynamics studies in GAT enzymes andin ammonia-transport proteins have shown that ammonia, not theammonium ion, is transferred, driven by differences in the solvationenergy of ammonia along the tunnel33,34. Whether an ammoniamolecule, rather than an ammonium ion, traverses this ammoniatunnel remains to be seen.

An ammonia tunnel was previously identified in a homology modelof NAD+ synthetaseyeast generated by threading35. The suggestedammonia tunnel is inconsistent with the structure presented here.Using sequence alignment, we identified the corresponding residues inthe three-dimensional structure of NAD+ synthetaseTB to thosereported in NAD+ synthetaseyeast (Supplementary Fig. 3). Most ofthe residues proposed to line the tunnel for the yeast enzyme arescattered in the crystal structure and are located far from active sitesand from the ammonia tunnel identified by CAVER (SupplementaryFig. 7 online). The only exceptions are Tyr532yeast (corresponding toTrp490TB and Phe167Bac), which was shown in this structure and inthe NAD+ synthetaseNH3 (ref. 13) to be involved in NaAD+ binding,Glu177yeast (Asp178TB), which is located on the same loop as Cys176TB

but faces in the opposite direction from the glutaminase active site,and Tyr601yeast (Leu566TB), which lies at the end of a17 helix near thesynthetase active site.

Given the evidence provided here and previously32 that suggestssynergy between the two active sites of the NAD+ synthetasesGln, wehave identified regions of the protein that may act as communicationelements between the two active sites. Two of these elements, the YREloop and a17, may have a role in the communication between thesynthetase active site and the glutaminase active site including con-striction 1. The N terminus of a17 is connected to loop P2 in theNAD+ synthetasesNH3, which should become ordered in the presenceof the NaAD-AMP intermediate, establishing a possible link betweenNaAD-AMP formation and ammonia formation and/or access to thetunnel by contacting the YRE loop. Notably, an alanine variant ofTyr601yeast, which corresponds to Leu566TB near the end of a17(Supplementary Fig. 7), shows poor ammonia-dependent synthe-tase–specific activity32, probably because of an impaired movementof loop P2 in formation of the intermediate. If a17 indeed acts asa communication wire between conformational changes at loop P2and ammonia formation, this variant should have a decreasedactivation of the glutaminase domain in the presence of synthetasesubstrates. Accordingly, the turnover number for the glutaminaseactivity in the presence of the saturating synthetase substrate decreasedby 40-fold for this variant, whereas the decrease in the basal glutami-nase activity was only four-fold32, supporting a compromised

glutaminase-activation mechanism. Two other elements, a14 anda9, are positioned to transmit information about the NaAD+ bindingstate, and a14 may also respond to intermediate formation (Fig. 6). Asa result of its close connection to constriction 2, it induces widening ofthe constriction to allow ammonia to move into the synthetase activesite. The kinetic and structural characterizations reported here lay astrong foundation for further studies aimed at elucidating the regula-tion of the catalytic activities in NAD+ synthetasesGln and for thedesign of antituberculotic drugs.

METHODSCloning and overexpression of native, D656A, C176A and native seleno-

methionine (SeMet) NAD+ synthetase fromM. tuberculosis. We amplified the

gene encoding NAD+ synthetase from M. tuberculosis genomic DNA (donated

by L.-Y. Gao) and cloned it into pSMT3 vector (a gift from C. D. Lima36) at the

BamHI-HindIII sites. We checked the resulting plasmid pMST3/nadE by

sequencing. C176A and D656A mutations were introduced in pMST3/nadE

using The Stratagene QuickChange Site-Directed Mutagenesis kit and verified

by sequencing the entire ORF. Native, C176A and D656 proteins were expressed

in Escherichia coli BL21(DE3) in LB medium containing kanamycin (50 mg

ml�1) by induction with 0.2% (w/v) lactose at 18 1C for 48 h (yield of 8 g of

wet cells per liter). SeMet-derivatized protein was expressed in E. coli

BL21(DE3) in M9 SeMet high-yield media (Medicilon) containing kanamycin

(50 mg ml–1). The cells were induced (at an optical density at 600 nm

(OD600) of 0.5) with 1 mM IPTG at 20 1C, harvested by centrifugation

48 h after induction and frozen as a pellet in liquid nitrogen (yield of 5 g of

wet cells per liter).

Purification and characterization of native, C176A, D656A and native SeMet

NAD+ synthetase from M. tuberculosis. We purified native, C176A, D656A

and native SeMet proteins at 4 1C following the procedure described below.

Table 3 Data collection and refinement statistics for

NAD+ synthetaseTB

SeMet Native

Data collection

Space group P41212 P41212

Cell dimensions

a ¼ b, c (A) 178.21, 214.84 178.13, 215.18

l (A) 0.97930 0.97949

Resolution (A) 30–3.00 30–2.35

Rsym 0.156 (0.320) 0.139 (0.559)

I/sI 9.6 (2.5) 14.4 (2.5)

Completeness (%) 92.9 (66.8) 99.4 (99.0)

Redundancy 7.9 (4.6) 11.4 (5.3)

Refinement

No. reflections 123,320 (8,868) 142,674 (14,038)

Rwork / Rfree (%) 18.4/24.5 (28.7/34.5)

No. atoms 22,082

Protein 20,323

Ligand/ion 226

Solvent 1,443

Mean B-factor (A2) 8.1 (TLS refined)

Waters 1,443

Residues 2,610

DON 4

NaAD+ 4

r.m.s. deviations

Bond lengths (A) 0.015

Bond angles (1) 1.65

*Values in parentheses are for the highest-resolution shell.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 2 7

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Frozen cells were suspended to 0.1 g ml–1 in ice-cold lysis buffer (50 mM Tris,

pH 8.0, 150 mM NaCl, 1 mM benzamidine, 1 mM PMSF, 1 mM DTT, 20%

(v/v) glycerol) and broken by French press at 12,000 psi. After removal of the

cell debris by centrifugation, we purified of the SUMO-tagged NAD+ synthetase

with a linear gradient of 20–120 mM imidazole in 50 mM Tris, pH 8.0, 300 mM

NaCl, 1 mM DTT, 20% (v/v) glycerol using nickel–nitrilotriacetic acid (Ni-

NTA) agarose chromatography (Qiagen). The SUMO tag was cleaved by

dialysis (50 mM Tris, pH 8.0, 1 mM DTT, 350 mM NaCl, 30% (v/v) glycerol)

with the protease Ulp1 at 4 1C for 3 h. We expressed and purified Ulp1 as

previously reported36. Ulp1 cleavage results in NAD+ synthetase with only one

additional amino acid at the N terminus (serine). Pure untagged NAD+

synthetase obtained from the flow through of a Ni-NTA chromatographic step

was dialyzed in 20 mM Tris, pH 7.5, 1 mM DTT, 30% (v/v) gycerol and loaded

onto a Sepharose CL-6B column (Amersham Biosciences). Protein was eluted

with 20 mM Tris, pH 7.5, 1 mM DTT and 15% (v/v) glycerol and concentrated

to B7 mg ml–1 by ultra filtration in an Amicon stirred cell over a YM30

membrane (yield of 5 mg of protein per gram of wet cells). Protein concentra-

tion was measured spectrophotometrically at 280 nm using the extinction

coefficient determined for NAD+ synthetase [e280 ¼ 1.16 (±0.02) ml mg–1

cm–1] by quantitative amino acid analysis (Molecular Structure facility, Uni-

versity of California, Davis). The protein was flash-frozen in liquid nitrogen

and stored at �80 1C. We collected a MALDI-TOF spectrum of the native

enzyme on a Himadzu-Kratos Axima-CFR MALDI-TOF mass spectrometer

(University of Maryland, College Park) in positive-ion mode in the presence of

3,5-dimethoxy-4-hydroxycinnaminic acid, resulting in a mass of 74,786 ±

142 Da (expected 74,787 Da). We estimated the native molecular mass of

purified NAD+ synthetase at 4 1C by gel-filtration chromatography on a

Sepharose CL 6B column prequilibrated in 50 mM Tris, pH 7.5, 0.1 M NaCl,

calibrated with High Molecular Weight Protein Standards (Amersham Bios-

ciences). The successful full incorporation of SeMet in the protein was

determined by MALDI analysis by the School of Pharmacy Mass Spectrometry

Facility, University of Maryland, Baltimore. We acquired UV CD spectra

(300 nm to 185 nm) of C176A, D656A and native proteins at 37 1C on a

Jasco J-810 spectrometer (Supplementary Fig. 8 online). Protein samples were

exchanged into Na2HPO4/NaH2PO4, pH 8.5, using a G-50 spin column.

Kinetics characterizations of the wild-type, of the C176A and the N656A

variants, and of the DON-modified NAD+ synthetaseTB, are described in

Supplementary Methods online.

Protein crystallization and data collection. Before crystallization, protein

samples were thawed and concentrated to 10–15 mg ml–1 using an Amicon

Ultra 30 K centrifugal filter device (Millipore). DON-modified NAD+ synthe-

taseTB was cocrystallyzed in the presence of 2 mM DON and 3 mM NaAD+. We

carried out initial screening for crystallization using a Phoenix Liquid Handling

System for Crystallography (Art Robbins Instruments) with the sitting drop

vapor diffusion method in 96-well microtiter plates. Four screens (Cryo and

PEG Suites (Qiagen), Index (Hampton Research) and Wizard I-II (Emerald

BioSystems)) were tested with two drops and 90 ml of reservoir volume.

One drop contained 0.4 ml of protein solution and 0.4 ml of mother liquor,

whereas the other contained the same volume of protein solution and 0.2 ml of

mother liquor. Crystallization plates were incubated at 20 1C. Several condi-

tions yielded crystals. Among those, we observed square plate crystals in 1.8 M

ammonium citrate tribasic dihydrate. We tested the optimized crystallization

conditions using the hanging drop vapor diffusion method in 24-well plates

with two drops per well and 700 ml of mother liquor. The two drops were

prepared by mixing 1 ml of mother liquor with 1 ml and 2 ml of protein solution,

respectively. We observed plate crystals for native and SeMet proteins in 1.8 M

ammonium citrate tribasic dihydrate, pH 7.0, 5% (v/v) glycerol. Crystals were

flash frozen in mother liquor containing 15% (v/v) glycerol in liquid nitrogen.

Crystals diffracted in the range of 2.3–3.5 A at the synchrotron, indicating

a space group P41212, with unit cell dimensions a ¼ b ¼ 178 A, c ¼ 215 A and

all angles 901. There were four molecules per asymmetric unit, half of the

expected biological octamer. We collected high-resolution data to 2.35 A on a

crystal with dimensions 0.4 � 0.4 � 0.02 mm. Single-wavelength anomalous

data from SeMet-derivatized crystals were collected to 3.0 A. All data were

collected at the NE-CAT beamline (Advanced Photon Source, Argonne

National Laboratories).

Structure determination and refinement. We processed the data using

HKL2000 (ref. 37) and solved the structure using the program HKL2MAP to

identify the heavy atoms and calculate the phases. The resulting map was put

through density modification and automated model building using Resolve38.

NCS restraints were used to improve map quality and removed in later

refinement steps. The resulting model contained roughly 85% of the expected

sequence. Additional model building was performed using COOT39, and

Refmac40 was used for refinement. TLS refinement was used by setting each

monomer as a separate TLS group41. The model contains 2,610 residues of the

expected 2,720 residues, 4 molecules of NaAD+, 4 molecules of DON and 13

glycerol molecules. The four subunits superimpose with an average r.m.s.d. of

0.48 A with 646 residues (performed by Swiss PDB Viewer, http://spdbv.vital-

it.ch/). We prepared molecular graphic figures using PyMol (http://www.pymol.

org/). Detailed data collection and refinement statistics are listed in Table 3.

Accession codes. Protein Data Bank: the coordinates and the structural factors

for M. tuberculosis NAD+ synthetase have been deposited with accession

code 3DLA.

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ACKNOWLEDGMENTSWe thank L.-Y. Gao (University of Maryland) and C.D. Lima (Sloan KetteringInstitute) for M. tuberculosis genomic DNA and for the expression plasmidspSMT3 and Ulp1, respectively. We thank K. Rajashankar for assistance with datacollection and processing. We thank A. Wlodawer and, especially,J.D. Kahn for critiques on the manuscript. This work is based upon researchconducted at the Northeastern Collaborative Access Team beamlines of theAdvanced Photon Source, supported by award RR-15301 from the NationalCenter for Research Resources at the US National Institutes of Health. Use of theAdvanced Photon Source is supported by the US Department of Energy, Office ofBasic Energy Sciences, under Contract No. DE-AC02-06CH11357. This work wassupported by a PRF grant from the American Chemical Society to B.G. and bystart-up funds to B.G. and N.L.-L. from the Department of Chemistry andBiochemistry, University of Maryland, College Park.

AUTHOR CONTRIBUTIONSM.R. cloned, expressed and purified all the proteins and performed all the kineticexperiments; N.L.-L. collected and analyzed crystallographic data and builtand refined the structural model of NAD+ synthetase; B.G. designed all theexperiments, crystallized native and SeMet NAD+ synthetase and contributed tothe building and the refinement of the structural model of NAD+ synthetase;B.G. and N.L.-L. wrote the manuscript.

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Belenky, P., Bogan, K.L. & Brenner, C. NAD+ metabolism in health and disease. TrendsBiochem. Sci. 32, 12–19 (2007).

2. Yang, H. et al. Nutrient-sensitive mitochondrial NAD+ levels dictate cell survival. Cell130, 1095–1107 (2007).

3. Lin, S.J. & Guarente, L. Nicotinamide adenine dinucleotide, a metabolic regulator oftranscription, longevity and disease. Curr. Opin. Cell Biol. 15, 241–246 (2003).

4. Tempel, W. et al. Nicotinamide riboside kinase structures reveal new pathways toNAD+. PLoS Biol. 5, e263 (2007).

5. Magni, G. et al. Enzymology of NAD+ homeostasis in man. Cell. Mol. Life Sci. 61,19–34 (2004).

6. Begley, T.P., Kinsland, C., Mehl, R.A., Osterman, A. & Dorrestein, P. The biosynthesis ofnicotinamide adenine dinucleotides in bacteria. Vitam. Horm. 61, 103–119 (2001).

7. Bieganowski, P. & Brenner, C. Discoveries of nicotinamide riboside as a nutrient andconserved NRK genes establish a Preiss-Handler independent route to NAD+ in fungiand humans. Cell 117, 495–502 (2004).

8. Boshoff, H.I. et al. Biosynthesis and recycling of nicotinamide cofactors in Myco-bacterium tuberculosis. An essential role for NAD in nonreplicating bacilli. J. Biol.Chem. 283, 19329–19341 (2008).

9. Corbett, E.L. et al. The growing burden of tuberculosis: global trends and interactionswith the HIV epidemic. Arch. Intern. Med. 163, 1009–1021 (2003).

10. Jauch, R., Humm, A., Huber, R. & Wahl, M.C. Structures of Escherichia coli NADsynthetase with substrates and products reveal mechanistic rearrangements. J. Biol.Chem. 280, 15131–15140 (2005).

11. Symersky, J., Devedjiev, Y., Moore, K., Brouillette, C. & DeLucas, L. NH3-dependentNAD+ synthetase from Bacillus subtilis at 1 A resolution. Acta Crystallogr. D Biol.Crystallogr. 58, 1138–1146 (2002).

ART IC L E S

42 8 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

12. Rizzi, M. et al. Crystal structure of NH3-dependent NAD+ synthetase from Bacillussubtilis. EMBO J. 15, 5125–5134 (1996).

13. Rizzi, M., Bolognesi, M. & Coda, A. A novel deamido-NAD+-binding site revealed by thetrapped NAD-adenylate intermediate in the NAD+ synthetase structure. Structure 6,1129–1140 (1998).

14. Devedjiev, Y. et al. Stabilization of active-site loops in NH3-dependent NAD+ synthe-tase from Bacillus subtilis. Acta Crystallogr. D Biol. Crystallogr. 57, 806–812 (2001).

15. Bellinzoni, M. et al. Heterologous expression, purification, and enzymatic activity ofMycobacterium tuberculosis NAD+ synthetase. Protein Expr. Purif. 25, 547–557(2002).

16. Brenner, C. Catalysis in the nitrilase superfamily. Curr. Opin. Struct. Biol. 12,775–782 (2002).

17. Raushel, F.M., Thoden, J.B. & Holden, H.M. The amidotransferase family of enzymes:molecular machines for the production and delivery of ammonia. Biochemistry 38,7891–7899 (1999).

18. Huang, X., Holden, H.M. & Raushel, F.M. Channeling of substrates and intermediatesin enzyme-catalyzed reactions. Annu. Rev. Biochem. 70, 149–180 (2001).

19. Nakamura, A., Yao, M., Chimnaronk, S., Sakai, N. & Tanaka, I. Ammonia channelcouples glutaminase with transamidase reactions in GatCAB. Science 312,1954–1958 (2006).

20. Raushel, F.M., Thoden, J.B. & Holden, H.M. Enzymes with molecular tunnels. Acc.Chem. Res. 36, 539–548 (2003).

21. Tullius, M.V., Harth, G. & Horwitz, M.A. Glutamine synthetase GlnA1 is essential forgrowth of Mycobacterium tuberculosis in human THP-1 macrophages and guinea pigs.Infect. Immun. 71, 3927–3936 (2003).

22. Yi, C.K. & Dietrich, L.S. Purification and properties of yeast nicotinamide adeninedinucleotide synthetase. J. Biol. Chem. 247, 4794–4802 (1972).

23. Krissinel, E. & Henrick, K. Secondary-structure matching (SSM), a new tool for fastprotein structure alignment in three dimensions. Acta Crystallogr. D Biol. Crystallogr.60, 2256–2268 (2004).

24. Sakai, N., Tajika, Y., Yao, M., Watanabe, N. & Tanaka, I. Crystal structure ofhypothetical protein PH0642 from Pyrococcus horikoshii at 1.6 A resolution. Proteins57, 869–873 (2004).

25. Petrek, M. et al. CAVER: a new tool to explore routes from protein clefts, pockets andcavities. BMC Bioinformatics 7, 316 (2006).

26. Hung, C.L. et al. Crystal structure of Helicobacter pylori formamidase AmiF reveals acysteine-glutamate-lysine catalytic triad. J. Biol. Chem. 282, 12220–12229 (2007).

27. Bellinzoni, M. et al. Glutamine amidotransferase activity of NAD+ synthetase fromMycobacterium tuberculosis depends on an amino-terminal nitrilase domain. Res.Microbiol. 156, 173–177 (2005).

28. Endrizzi, J.A., Kim, H., Anderson, P.M. & Baldwin, E.P. Crystal structure ofEscherichia coli cytidine triphosphate synthetase, a nucleotide-regulatedglutamine amidotransferase/ATP-dependent amidoligase fusion protein and homologueof anticancer and antiparasitic drug targets. Biochemistry 43, 6447–6463(2004).

29. Mouilleron, S., Badet-Denisot, M.A. & Golinelli-Pimpaneau, B. Glutamine bindingopens the ammonia channel and activates glucosamine-6P synthase. J. Biol. Chem.281, 4404–4412 (2006).

30. Tesmer, J.J., Klem, T.J., Deras, M.L., Davisson, V.J. & Smith, J.L. The crystal structureof GMP synthetase reveals a novel catalytic triad and is a structural paradigm for twoenzyme families. Nat. Struct. Biol. 3, 74–86 (1996).

31. Larsen, T.M. et al. Three-dimensional structure of Escherichia coli asparagine synthetaseB: a short journey from substrate to product. Biochemistry 38, 16146–16157 (1999).

32. Wojcik, M., Seidle, H.F., Bieganowski, P. & Brenner, C. Glutamine-dependent NAD+

synthetase. How a two-domain, three-substrate enzyme avoids waste. J. Biol. Chem.281, 33395–33402 (2006).

33. Fan, Y., Lund, L., Yang, L., Raushel, F.M. & Gao, Y.Q. Mechanism for the transport ofammonia within carbamoyl phosphate synthetase determined by molecular dynamicssimulations. Biochemistry 47, 2935–2944 (2008).

34. Lin, Y., Cao, Z. & Mo, Y. Molecular dynamics simulations on the Escherichia coliammonia channel protein AmtB: mechanism of ammonia/ammonium transport. J. Am.Chem. Soc. 128, 10876–10884 (2006).

35. Bieganowski, P., Pace, H.C. & Brenner, C. Eukaryotic NAD+ synthetase Qns1 containsan essential, obligate intramolecular thiol glutamine amidotransferase domain relatedto nitrilase. J. Biol. Chem. 278, 33049–33055 (2003).

36. Mossessova, E. & Lima, C.D. Ulp1-SUMO crystal structure and genetic analysis revealconserved interactions and a regulatory element essential for cell growth in yeast. Mol.Cell 5, 865–876 (2000).

37. Otwinowski, Z. & Minor, W. Processing of X-ray diffraction data collected in oscillationmode. Methods Enzymol. 276, 307–326 (1997).

38. Wang, J.W. et al. SAD phasing by combination of direct methods with the SOLVE/RESOLVE procedure. Acta Crystallogr. D Biol. Crystallogr. 60, 1244–1253 (2004).

39. Emsley, P. & Cowtan, K. Coot: model-building tools for molecular graphics. ActaCrystallogr. D Biol. Crystallogr. 60, 2126–2132 (2004).

40. Murshudov, G.N., Vagin, A.A. & Dodson, E.J. Refinement of macromolecular structuresby the maximum-likelihood method. Acta Crystallogr. D Biol. Crystallogr. 53, 240–255(1997).

41. Winn, M.D., Isupov, M.N. & Murshudov, G.N. Use of TLS parameters to modelanisotropic displacements in macromolecular refinement. Acta Crystallogr. D Biol.Crystallogr. 57, 122–133 (2001).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 2 9

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Precursor-product discrimination by La protein duringtRNA metabolismMark A Bayfield1,3 & Richard J Maraia1,2

La proteins bind pre-tRNAs at their UUU-3¢OH ends, facilitating their maturation. Although the mechanism by which La bindspre-tRNA 3¢ trailers is known, the function of the RNA binding b-sheet surface of the RNA-recognition motif (RRM1) is unknown.How La dissociates from UUU-3¢OH–containing trailers after 3¢ processing is also unknown. Here we show that La preferentiallybinds pre-tRNAs over processed tRNAs or 3¢ trailer products through coupled use of two sites: one on the La motif and anotheron the RRM1 b-surface that binds elsewhere on tRNA. Two sites provide stable pre-tRNA binding, whereas the processed tRNAand 3¢ trailer are released from their single sites relatively fast. RRM1 loop-3 mutations decrease affinity for pre-tRNA and tRNA,but not for the UUU-3¢OH trailer, and impair tRNA maturation in vivo. We propose that RRM1 functions in activities that aremore complex than UUU-3¢OH binding. Accordingly, the RRM1 mutations also impair an RNA chaperone activity of La. Theresults suggest how La distinguishes precursor from product RNAs, allowing it to recycle onto a new pre-tRNA.

La is the first protein to interact with nascent pre-tRNAs in eukaryotesand remains bound during tRNA processing and modification1,2. Inits best characterized activity, the La domain, comprising a La motif(LM) and RRM1 in a fixed arrangement, protects UUU-3¢OH–containing RNAs from 3¢ exonucleolytic digestion1–4.

Because LM and RRM1 are required for UUU-3¢OH binding5, it wasexpected that this binding would involve the b-sheet surface of LaRRM1 (refs. 6–8). Unexpectedly, a human La crystal structure showsthat, although the LM and RRM1 form a UUU-3¢OH binding cleft,most RNA contacts are to the LM, leaving the b-sheet surface of RRM1unoccupied4. This documented a mode of sequence-specific recogni-tion by an RRM that does not involve its b-sheet surface4. Fouradditional structures of human La with different RNAs confirmedthe LM mode of recognition with the RRM1 b-sheet surface unoccu-pied9. Thus, LM-RRM1 mediates induced-fit UUU-3¢OH binding9

while leaving the RRM1 b- sheet surface unoccupied and its role inRNA binding unclear3,4,9,10.

Some La-related proteins (LARPs) that contain a La domain aretelomerase subunits11,12. LARP7 (also known as PIP7S) is a tumorsuppressor that binds the 7SK small nuclear RNA (snRNA) toregulate the transcription elongation factor P-TEFb13,14. CiliateLARPs recognize UUU-3¢OH–containing telomerase RNA, andLARP7 recognizes UUU-3¢OH of 7SK snRNA, whereas neitherare associated with nascent pre-tRNAs13,15. As the LM-RRMarrangement is conserved in LARPs, they are likely to use RNAbinding modes similar to that of genuine La. However, how the LMand RRM1 b-sheet binding surface work together is not known forany protein.

Some activities of La proteins are more complex than UUU-3¢OHbinding. Yeast La is required for the maturation of structurallyimpaired pre-tRNAs16–18, can promote tRNA folding19 and actsredundantly with tRNA modification enzymes, consistent with arole in RNA structural integrity20. Ciliate LARP p65 induces structuralrearrangement of telomerase RNA12,21, and La shows RNA chaperoneactivity18,22,23. Although these studies document complex functionsrelated to RNA structure and folding, the mechanisms that distinguishthese activities from simple UUU-3¢OH binding are unknown. Tounderstand this, it will be necessary to know how the LM and RRMinteract with RNA dynamically, during simple and complex activitiesand in specific pathways of RNA metabolism.

Modified nucleotides found on La-associated pre-tRNAs indicatethat La is bound to pre-tRNAs during modification24,25. La remainsassociated with pre-tRNA until RNase Z–mediated endonucleolyticcleavage separates the tRNA body from the UUU-3¢OH–containing3¢ trailer, the latter of which varies for different pre-tRNAs16,17,26–29.Whereas pre-tRNA processing occurs in minutes, the half-life of La ishours24,30 and, although abundant, La can be limiting for tRNAmaturation31. To participate in multiple rounds of tRNA maturation,La must dissociate from the UUU-3¢OH–containing trailer productsof processing and recycle onto newly synthesized pre-tRNAs. Ascleaved 3¢ trailers contain UUU-3¢OH, the highest-affinity sequence-specific ligand known for La, it was unclear how efficient dissociationwould occur.

We hypothesized that La might distinguish pre-tRNA from pro-cessed products using two binding sites that together provide higheraffinity than either alone. After processing, the tRNA and 3¢ trailer

Received 24 June 2008; accepted 9 February 2009; published online 15 March 2009; doi:10.1038/nsmb.1573

1Intramural Research Program, Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health, Bethesda, Maryland,USA. 2Commissioned Corps, US Public Health Service, Bethesda, Maryland, USA. 3Present address: Department of Biology, York University, Toronto, ON, Canada.Correspondence should be addressed to R.J.M. (maraiar@mail.nih.gov).

43 0 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

ART IC L E S

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

would readily dissociate from their single binding sites. We show thatthe RRM1 b-sheet surface of human La forms an RNA binding site,distinct from the UUU-3¢OH binding site, that promotes tRNAmaturation in vivo. We also show that this RRM1 binding sitecontributes to the RNA chaperone activity of La.

RESULTSLa binds tRNA using non–UUU-3¢OH–mediated contactsTo test for a tRNA binding site on La distinct from the UUU-3¢OHbinding site, we used the electrophoretic mobility shift assay (EMSA)on a 73-nt tRNAArgACG transcript (which ends with UCG-3¢OH) anda 12-nt 3¢ trailer that ends with UUU-3¢OH, as these representcleavage products of RNase Z26 (indicated as tRNA and UUU-OH3¢ trailer in Fig. 1). La showed a several fold–higher affinity for thistrailer than for the same trailer ending in AAA-3¢OH, consistent withprevious results18 and confirming UUU-3¢OH–specific recognition inour EMSA (not shown). A standard EMSA that contains Mg2+

revealed that La bound avidly to both the 12-nt UUU-3¢OH trailer

and the tRNA (Fig. 1a,b). Although the tRNA does not end withUUU-3¢OH, it may still interact with the UUU-3¢OH binding site,which can accommodate, albeit with lower affinity, bases other thanuracil4. Therefore, we examined binding to the La–Q20A Y23A D33Rmutant (La-QYD in Fig. 1c,d), which is mutated in the LM UUU-3¢OH binding site: Gln20 makes uracil-specific contact; Tyr23 stackswith uracil; and Asp33 makes bidentite contacts to the 2¢ OH and3¢OH of the last nucleotide4,9. La-QYD was debilitated for UUU-3¢OHtrailer binding but supported more binding to the tRNA (Fig. 1c,d).This suggested a UUU-3¢OH–independent mode of tRNA bindingthat is mediated by a site on La other than the RNA 3¢ OH endbinding site in the LM4,9.

Differential sensitivity to Mg2+ reflects distinct binding modesTo further test for differential binding, we compared Kd values forthree relevant RNA species, pre-tRNAArgACG, tRNAArgACG and thefree 12-nt UUU-3¢OH trailer, in two extremes of Mg2+ concentration,0 mM and 10 mM, as well as 1 mM (Fig. 2a–c). Of note is the absenceof divalent cations in the crystal of La bound to UUU-3¢OH4, and therecognized association of divalent cations with tRNA32. The 85-nt pre-tRNAArgACG used in Figure 2 lacks a 5¢ leader but contains the 12-ntUUU-3¢OH–containing trailer covalently linked to tRNAArgACG (thatis, as a contiguous T7 transcript), as this is a substrate for RNase Z26,whereas the 73-nt tRNAArgACG and the 12-nt UUU-3¢OH trailer

a b

c d

0 25 50 100

250

500

La (nM): 0 25 50 100

250

500

F

RNP

F

RNP

UUU-3′OH trailer

UUU-3′OH trailer

tRNA

0 25 50 100

250

500

La-QYD:(nM)

0 25 50 100

250

500

F

RNP

F

RNP

tRNA

0

0.5

1

1.5

2

2.5

3

0 5 10 15 20 25 30 35 40RNA bound (nM)

Bou

nd/fr

ee

0 Mg2+

Pre-tRNA(2.4 nM)

tRNA (7.4 nM)

3′ trailer (19.8 nM)

a

e f

g

b c

d

0

0.5

1

1.5

2

2.5

3

0 10 20 30 6040 50 70RNA bound (nM)

Pre-tRNA(76.9 nM)

tRNA (69 nM)

3′ trailer (13.6 nM)

10 Mg2+

–0.9–0.8–0.7–0.6–0.5–0.4–0.3–0.2–0.1

0

ln B

/Bo

Pre-tRNA

–1.2

–1

–0.8

–0.6

–0.4

–0.2

0

ln B

/Bo

3′ trailer

–2–1.8–1.6–1.4–1.2

–1–0.8–0.6–0.4–0.2

0

ln B

/Bo

tRNA

pre-tRNA

tRNA

3′ trailer

2.4

7.2

19.8

76.9

69.0

13.6

0.031

0.105

1.45

0.113

0.321

0.208

6.2

2.2

3.3

Kd (nM)

0 Mg2+ 10 mM Mg2+ 0 Mg2+/10 Mg2+koff t1/2

11.5

25.2

24.8

1 mM Mg2+

0 20 6040 800

0.51

1.52

2.53

4.5

3.54

5

Pre-tRNA(11.5 nM)

tRNA (25.2 nM)

3′ trailer (24.8 nM)

1 Mg2+

RNA bound (nM)

Time (min)50 150 250

Time (min)50 150 250

Time (min)50 150 250

(min–1) (min)

Figure 2 Two distinct RNA binding sites on La together enhance stable binding to pre-tRNA. (a–c) Scatchard analyses of method 2 EMSAs performed on

pre-tRNAArgACG, tRNAArgACG and the 12-nt UUU-3¢OH trailer at 0, 10 and 1 mM Mg2+. Each titration was done at a constant concentration of La with

concentrations of RNA varying, although the La concentration differed for each individual Scatchard plot; Kd values are provided next to each ligand and in

g. (d–f) Analysis of dissociation of La from pre-tRNAArgACG, tRNAArgACG and the 12-nt UUU-3¢OH trailer from data derived from EMSAs was performed and

analyzed as described33. Timescales 0–350 s. (g) Kd values derived from a–c above and the numerical fraction of the Kd values at 0 and 10 mM Mg2+

(0 Mg/10 Mg), as indicated. Standard errors of the regression coefficient derived from triplicate determinations for pre-tRNAArgACG, tRNAArgACG and the

12-nt UUU-3¢OH trailer were 14.7%, 11.7% and 10.9%, respectively, in 0 mM Mg2+, 5.5%, 7.3% and 8.9%, respectively, in 10 mM Mg2+, and 7.7%,

5.9% and 13.9% in 1 mM Mg2+. koff values were derived from d–f above using standard calculations33. The t1/2 times were then derived from koff. Standard

errors for dissociation of pre-tRNAArgACG, tRNAArgACG and the 12-nt UUU-3¢OH trailer were 8.5%, 7.1% and 10.8%, respectively. The units for koff and

t1/2 are min–1 and min, respectively.

Figure 1 La can bind non–UUU-3¢OH–containing RNA via contacts that are

not mediated by the previously characterized RNA 3¢OH binding site in the

La motif. (a,b) EMSAs reveal human La protein binding to the 12-nt UUU-

3¢OH–containing trailer (a) and a tRNAArgACG transcript that lacks UUU-

3¢OH (b). (c,d) Binding by the mutated protein La–Q20A Y23A D33R

(abbreviated La-QYD), which contains three substitutions in the LM at

residues that are known to be crucial for UUU-3¢OH–specific binding4 (see

text). For each, a constant trace amount of 32P-RNA (B0.1 nM) was

incubated with varying concentrations of La protein, as indicated above.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 3 1

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

represent cleavage products. We used two EMSA methods, each withreactions containing 0 mM Mg2+ or 10 mM Mg2+. In method 1,varying amounts of La were added to trace amounts of 32P-RNA. Thisshowed that binding of the 12-nt trailer was no different in the 0 mMMg2+ and 10 mM Mg2+, whereas tRNA and pre-tRNA binding weremarkedly compromised in 10 mM Mg2+ versus 0 mM Mg2+ (Supple-mentary Fig. 1 online, 0 mM Mg2+ and 10 mM Mg2+). By method 1,the Kd is roughly estimated as the La concentration at which 50% ofthe RNA is shifted. The Kd for the pre-tRNA and tRNA were eachestimated at 5–10 nM in 0 mM Mg2+ and B100 nM in 10 mM Mg2+,whereas the Kd for the 12-nt trailer was B20 nM, with little if anydifference in 0 mM Mg2+ versus 10 mM Mg2+. The large effect ofMg2+ on pre-tRNA and tRNA but not on the 12-nt trailer wasconfirmed by EMSA method 2, as described below (Fig. 2a–c).

For method 2, we focused our analyses on RNA concentrations ofaround the Kd obtained by method 1. For method 2, we added varyingamounts of unlabeled RNA to constant amounts of 32P-RNA and La33.Bound and free 32P-RNA fractions were quantified and Kd determinedby Scatchard analysis. We performed EMSAs in triplicate. Method 2supports a greater amount of data and is independent of the fractionof active protein, and the analysis includes quality-assurance para-meters33. Results for 0 mM Mg2+ and 10 mM Mg2+ are shown inFigure 2a and Figure 2b, respectively and summarized in Figure 2g(see also Supplementary Figs. 2 and 3 online). We obtained a Kd of13.6 nM for the 12-nt UUU-3¢OH trailer in 10 mM Mg2+, similar tothe Kd values of 7.1 nM and 25 nM obtained by others in the presenceof Mg2+ for similar ligands4,34. Our Kd of 2.4 nM for pre-tRNAArg

is similar to the Kd of 7.3 nM for pre-tRNAVal in the absence ofMg2+ (ref. 35).

We asked whether these Kd values were reflective of differentdissociation rates (koff)

33. The data indicated that La binds morestably to pre-tRNA than to the UUU-3¢OH trailer or tRNA (Fig. 2d–f,representative EMSAs shown in Supplementary Fig. 4 online). Thet1/2 times for pre-tRNA, tRNA and the UUU-3¢OH trailer were 6.2min, 2.2 min and 3.3 min, respectively (Fig. 2g).

La binding to pre-tRNA and tRNA wasinhibited by 10 mM Mg2+, whereas bindingto the UUU-3¢OH trailer was slightlyenhanced (Fig. 2g, columns 0 mM Mg2+,10 mM Mg2+ and 0 mM Mg2+/10 mMMg2+). The main conclusion we draw fromthis analysis is that the binding modes usedfor tRNA and the UUU-3¢OH ligands arebiochemically distinct.

Although differential sensitivity to Mg2+

provides biochemical evidence of distinctbinding modes, neither 0 mM Mg2+ nor10 mM Mg2+ represent physiological condi-tions. As summarized in Figure 2g (column1 mM Mg), pre-tRNA had the highestaffinity for La, followed by the 3¢ trailerand tRNA.

A strong difference in binding aviditieswas revealed by competition. In parallel reac-tions containing 1 mM Mg2+, we examined32P–pre-tRNA and 32P-trailer for La bindingin the presence of 100 nM La and increasingamounts of unlabeled competitor RNAs(Fig. 3a,b). Unlabeled 3¢ trailer was largely in-effective for competition with 32P–pre-tRNA,even when present at higher concentration

than La, indicating that La shows a strong preference for pre-tRNAover the UUU-3¢OH–containing trailer (Fig. 3a,c). The same unlabeled3¢ trailer readily competed with 32P-trailer (Fig. 3b,d), indicating itsactivity as a competitor. As expected, unlabeled pre-tRNA competedeffectively with 32P-pre-tRNA and 32P-trailer. The tRNA, whichlacks UUU-3¢OH, competed with 32P–pre-tRNA, but was substantiallyless competitive with the 32P–UUU-3¢OH trailer, consistent withLa accommodating both the tRNA and trailer substrates usingdistinct binding sites. We conclude that La has a strong preferencefor pre-tRNA over the 12-nt UUU-3¢OH trailer, consistent with

trailerpre-tRNA

32P-pre-tRNA 32P-trailer

La RNP-

Free RNA-

-La RNP

-Free RNA

[Competitor] (nM)

Frac

tion

32P

-tra

iler

boun

d

32P-trailer

a b

c d1 2 3 4 5 6 7 8 9 10 11 12

Cold competitor:

[Competitor] (nM): no L

a15 63 25

01,

000

4,00

0no

La

15 63 250

1,00

04,

000

no L

a15 63 25

01,

000

4,00

0

mat-tRNA

13 14 15 16 17 18 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

trailerpre-tRNA

no L

a15 63 25

01,

000

4,00

0no

La

15 63 250

1,00

04,

000

no L

a15 63 25

01,

000

4,00

0

mat-tRNA

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0 1,000 2,000 3,000 4,000

pre-tRNA competitor

Trailer competitor

mat-tRNA competitor

[Competitor] (nM)

Frac

tion

32P

-pre

-tR

NA

boun

d

00.10.20.30.40.50.60.70.80.9

1

0 1,000 2,000 3,000 4,000

32P-pre-tRNA

pre-tRNA competitor

Trailer competitor

mat-tRNA competitor

Figure 3 La shows strong preference for pre-tRNA over a UUU-3¢OH trailer. (a,b) Binding reactions

contained constant amounts of La protein (100 nM) and a trace amount of 32P-RNA (B0.1 nM

pre-tRNA (a), or B0.1 nM UUU-3¢OH trailer (b), and varying amounts of unlabeled (cold) competitor

RNAs as indicated above the lanes and described in the text, in 1 mM Mg2+. (c,d) The data in a and b

were quantified and analyzed in c and d, respectively.

Table 1 Binding properties of mutated La proteins

(a) Effects on RNA binding of different La mutated proteins

La mutant La region tRNA binding UUU-3¢OH trailer

Q20A Y23A D33R LM Slightly reduced Much reduced

R32A K34A K37A LM Unaffected ND

K105A K109A RRM1-a0 Unaffected ND

Y114A F155A RRM1-b2 b3 Unaffected Unaffected

R143A R144A K148A K151A RRM1–loop-3 3–4� reduced Unaffected

K185A K191A K192A RRM1-a3 Unaffected ND

R57A R60A LM NA (other) Unaffected

K16A H19A LM NA (other) Unaffected

R143A K151A RRM1–loop-3 Unaffected ND

R144A K148A RRM1–loop-3 Unaffected ND

(b) Kd values of La-loop protein and the relationship of the Kd values for La-loop and

La (expressed as a fraction: Kd La/Kd La-loop) for three RNA ligands

Ligand Kd La-loop Kd La/La-loop

pre-tRNA 12.9 nM 0.19

tRNA 25.8 nM 0.28

UUU-3¢OH trailer 20.5 nM 0.97

NA, not applicable; ND, not determined.

ART IC L E S

43 2 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

the idea that pre-tRNA uses two binding sites, whereas the 12-ntUUU-3¢OH–containing trailer uses only one.

As shown below, a La RRM1 loop-3 mutant showed decreasedaffinity for tRNA, but not the UUU-3¢OH trailer binding in vitro, anddecreased tRNA maturation in vivo, providing support for thephysiological relevance of the second binding site on RRM1.

La RRM1 functions in pre-tRNA recognitionWe mapped the Mg2+-sensitive UUU-3¢OH–independent bindingactivity to the La domain (not shown). To further localize this activity,we examined surface charge distribution (Supplementary Fig. 5online) and mutated candidate basic surface patches to evaluatetheir effect on RNA binding. We examined several different mutatedproteins for tRNA binding, and in some cases other non–UUU-3¢OH–containing RNAs (not shown, indicated as ‘other’ in Table 1a. RRM1loop-3 (R143A R144A K148A K151A, referred to as ‘La-loop’ mutantbelow), which connects the RRM1 b2 and b3 strands, reduced bindingto pre-tRNA and tRNA four- to five-fold relative to La, with littledecrease in binding to the UUU-3¢OH trailer (Table 1b and Supple-mentary Figs. 6 and 7 online). Moreover, the residual tRNA bindingobserved for La-loop was insensitive to Mg2+ (not shown). DecreasedtRNA binding observed for the La-loop protein is specific, becausemany other mutated proteins did not decrease tRNA binding(Table 1a). The fact that the La-loop mutated protein showed normalUUU-3¢OH–mediated binding, which requires LM-RRM1 intermotifcontacts4,9, indicates that the La domain is not grossly misfolded. Weconclude that loop-3 of RRM1 is an important determinant of aUUU-3¢OH–independent RNA binding site on La. The basic residuesmutated in La-loop-3 are highly conserved in La proteins but notother RRM proteins such as U1A and PABP (Fig. 4a). Loop-3 forms abasic wall on one side of the RRM b-sheet surface (Fig. 4b).

La RRM1 loop-3 functions in tRNA maturation in vivoWe next examined the La-loop mutant for tRNA maturation inS. pombe as monitored by tRNA-mediated suppression (TMS) of apremature stop codon in ade6-704, which alleviates the accumulationof red pigment. By this assay, La can replace the function of S. pombeLa protein (Sla1p) in vivo17,18. We previously characterized suppressortRNA alleles that vary in dependence on La, Rrp6 (3¢ exonucleasecomponent of the exosome), and the RNA polymerase III (Pol III)termination subunit Rpc11p (a Pol III–associated 3¢-5¢ exonuclease),

for efficient maturation18,31. Mutation of the conserved aromaticresidues Tyr114 and Phe155 on the ribonucleoprotein (RNP)-1 and-2 motifs of the RRM1 b-surface compromised maturation of thestructurally defective suppressor pre-tRNA in the yeast strain ySK5,but had less effect in ySH9, which contains a wild-type suppressortRNA18. For the present study, analysis was limited to the wild-typesuppressor tRNA in ySH9 (Fig. 5a,b). Consistent with previous data,La was more active for TMS than the negative controls pRep vectorand the truncated protein La26–408, which lacks residues involved inUUU-3¢OH recognition17,18,36 (Fig. 5a, sectors 1 and 2). The mutantLa–Y114A F155A also served as a control (sector 3). La–loop-3 wasless active than La (Fig. 5a, compare sectors 3 & 5), indicating that oneor more of the basic residues in loop-3 that mediate UUU-3¢OH–independent pre-tRNA binding contribute to functional tRNAmaturation in vivo. We also combined the RRM1 loop-3 and theb-sheet mutations. These mutants, La–Y114A-loop and La–Y114AF155A–loop were increasingly compromised for TMS (Fig. 5a, sectors6 and 7). As shown below, these RRM1 mutants maintain pre-tRNAUUU-OH 3¢ end–protection activity.

The 3¢ end protection function of La protein is standardly mon-itored by northern blotting analysis of endogenous pre-tRNALys

CUU17,18,36–38. An intron probe detects three pre-tRNALysCUU spe-cies, the upper and middle bands, which contain the 3¢ trailer andrequire La for accumulation, and the lower band which accumulatesindependently of La17,18,36–38. The upper band represents a nascent PolIII transcript, whereas the middle species has had its 5¢ leaderremoved, and the lower band reflects the 5¢ and 3¢ end maturedspecies (as indicated to the right of Fig. 5b, above). U5 snRNA wasprobed on the same blot to serve as a control for quantitation (Fig. 5b,below). In cells lacking La or in La26–408, the upper and middle bandsrun as a smear (lanes 1 and 2) due to 3¢ exonucleolytic nibbling17,37.The upper band is stabilized by La17,37, but progressively less so inLa-Y114A, La-F155A and La–Y114A F155A mutants18. The La-loopmutant accumulated less nascent pre-tRNALysCUU (upper band) thandid La (Fig. 5b, compare lanes 3 and 5), consistent with a less stableinteraction of the pre-tRNA with the La-loop protein relative to La as aresult of decreased RRM1-mediated binding.

The compound RRM1 mutant La–Y114A F155A–loop, with muta-tions to the b-sheet surface and loop-3, was most compromised fornascent pre-tRNA accumulation (Fig. 5b, lane 7), consistent with itslow TMS activity (Fig. 5a). Note that, for this and other RRM1mutants, the upper and middle pre-tRNA species appear as distinctbands, a hallmark of 3¢ end protection (Fig. 5b, compare lanes 1 and 2with lanes 4–7). Thus, mutations in two different regions of the RNAbinding surface of RRM1 indicate that RRM1 contributes to tRNAmaturation in vivo. Moreover, the RRM1 effect is distinct frompre-tRNA 3¢ end protection. The cumulative results provide evidencethat La RRM1 contributes to tRNA maturation by mediating UUU-3¢OH–independent binding to pre-tRNA.

2a

b

KGQVLNIQMRR-TLHK-----AFKGSIFVVFDSIKGQILNIQMRR-TLHK-----TFKGSIFAVFDSIKGPIENIQMRR-TLQR-----EFKGSIFIIFNTDYDKVVNLTMRK-HYDKPTKSYKFKGSIFLTFETKFGETENVLMRR-LKPG---DRTFKGSVFITYKTRAGPISAVRMRR-DDDK-----KFKGSVFVEFKEPLGEINQVRLRRDHRNK-----KFNGTVLVEFKTI

FGQILDILVSRS--------LKMRGQAFVIFKEVS--PILSIRVCRDM-----ITRRSLGYAYVNFQQPA

H. sap La M. mus La X. laev La D. mel LaC. eleg La

S. cerev La S. pombe La

H. sap U1A H. sap PABP

--RNP1--

** * *

Loop 3 Tyr114Phe155

Arg143Lys151

Arg144

Lys148

l3 Figure 4 La RRM1 loop-3 mediates UUU-3¢OH–independent tRNA binding.

(a) Sequence alignment of the b2–loop-3–b3 regions of the RRM1s of

the La proteins from seven organisms (Homo sapiens, Mus musculus,

Xenopus laevis, Drosophila melanogaster, Caenorhabditis elegans,

Schizosaccharomyces pombe and Saccharomyces cerevisiae) followed by the

homologous regions of U1A and PABP. Asterisks indicate the basic residues

conserved in loop-3 of La proteins. RNP1 is indicated47; in La, this contains

Phe155, indicated by a vertical line above the sequence. (b) Structure of

the loop-3 side chains relative to the RRM1 b-sheet surface (adopted from

PDB 2VON9). The loop-3 basic residues mutated for this study are shown in

blue. RRM1 is shown in green, with the conserved aromatic residue side

chains (Tyr114 and Phe155) on b-strands 1 and 3, in cyan.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 3 3

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

RRM1 contributes to the RNA chaperone activity of LaThe La protein shows RNA chaperone activity in a cis-splicing assaythat uses a self-splicing intron23. This RNA can misfold in vitro andbecome trapped in an inactive conformation that can be resolved byLa and other RNA chaperone proteins such as StpA, NCp7 and Hfq22.We noted previously that La and a mutated derivative, La–Y114AF155A, were comparably active in this assay, consistent with theirindistinguishable RNA binding activities18. We compared La andLa-loop protein in the cis-splicing assay by monitoring the disappear-ance of full-length intron-containing RNA and the appearance ofspliced product (Fig. 5c, ln U/Uo, y axis). In agreement with previousreports, La activity was reflected by a steep slope in the initial fastphase of the assay23 (Fig. 5c). Whereas the no protein and BSAcontrols showed relatively little activity as expected, La was mostactive, and the La-loop protein was intermediate, even though bothwere present in equal amounts (Fig. 5d). We note that the RNA usedin this assay does not end in UUU-3¢OH. A decrease in RNAchaperone activity for La-loop correlates with a decrease in its non–UUU-3¢OH–mediated RNA binding activity, whereas La–Y114AF155A shows no decrease in RNA chaperone or RNA bindingactivities, as noted previously18.

DISCUSSIONWe report here biochemical and mutational analyses of La RRM1-mediated binding to pre-tRNA. The results indicate that the LM andRRM1 b-sheet surface comprises two binding sites that bind UUU-3¢OH and non–UUU-3¢OH RNAs, respectively. Concurrent use ofboth sites, as occurs for nascent pre-tRNA, provides high-affinitystable binding, whereas use of either site alone provides lower affinity,with less stable binding of processed tRNA and cleaved 3¢ trailers.

Two binding sites support directionality in tRNA maturationConsistent with its presence in a human Pol III holoenzyme andlocalization at Pol III–transcribed genes in yeast and human cells39–41,La is poised to be the first protein to bind newly synthesized pre-tRNAs. How La would dissociate from UUU-3¢OH after separation ofthe trailer and recycle onto new pre-tRNAs was unknown. This was animportant question because the large amount of 3¢ trailers carryingthe sequence-specific ligand UUU-3¢OH that are produced duringtRNA processing might consume La if there was no mechanism fordissociation. The ability to withstand challenge by an excess of 12-ntUUU-3¢OH RNA revealed a strong preference of La for pre-tRNA(Fig. 3). This coupled with the relatively fast dissociation of La from

UUU-3¢OH trailers illustrates a mechanism by which the potentialconsumption of La by UUU-3¢OH trailers can be avoided. The abilityto dissociate after 3¢ trailer cleavage supports directionality of thetRNA pathway (Fig. 6). After dissociation, the end-matured tRNAsbecome substrates of 3¢ end–modifying proteins such as CCA-addingenzyme and tRNA synthetases, and are bound by export factors Los1/Xpo-t and exported from the nucleus27. La can then recycle onto anew pre-tRNA.

Stable pre-tRNA binding contributes to tRNA structural integrityAlthough nascent pre-tRNAs probably have minimal structure, datasuggest that they acquire structure while associated with La. Althoughevidence indicates that La promotes correct folding of pre-tRNAArg

in vivo19, the misfolded pre-tRNAArg was not converted to thecorrectly folded pre-tRNAArg by La in vitro, suggesting involvementof other factors in vivo. Genetic evidence points to tRNA-modifyingenzymes as such factors20,42–44. As a result of high-affinity, stablebinding, pre-tRNAs would have enough time to acquire structure-stabilizing modifications, some of which occur while associated withLa24,25,45,46 (Fig. 6).

For certain functions, it may be beneficial for an RNA bindingprotein to preferentially associate with RNA precursors and dissociatefrom products. Our data indicate this to be true for La. We areunaware of any other noncatalytic RNA binding protein for which thishas been demonstrated.

RRM1 loop-3 is adjacent to and distinct from the LM binding siteIt was thought that the LM and RRM1 b-sheet surface form a singlesite that recognizes UUU-3¢OH6–8. With the appearance of multiplecocrystal structures of La bound to UUU-3¢OH RNAs4,9, this idea hasgiven rise to a two-site model3,4,9,10.

We provided in vitro and in vivo evidence for RRM1 function in themetabolism of a normal pre-tRNA. We used three RNA substrates,differential Mg2+ sensitivity and mutagenesis, which led to theidentification of RRM1 loop-3, adjacent to the b-sheet surface, ashaving an important role for binding of La to pre-tRNA binding, butnot to UUU-3¢OH, in vitro and pre-tRNA maturation in vivo.

For other RRM proteins, loop-3 contacts RNA, in some cases via abasic side chain47 (Fig. 4c). Loop-3 of La forms a basic wall on oneside of the RRM b-sheet surface that projects in the same plane as theconserved aromatic side chains (Fig. 4d). Involvement of La loop-3 inUUU-3¢OH–independent binding is in agreement with previousresults. None of the mutated residues in RRM1 loop-3 was observed

a b c d

M pRep

La 26–4

08

La Y114A

F15

5A

La-lo

op

La–Y

114A

-loop

La-Y

114A

F155A

-loop

1 2 3 4 5 6

1.0

0.41

0.29

0.29

0.04

0.04U5 /

-U5

pRep La26–408

La

La-loop

La–Y114A F155A

La–Y114A-loop

La-Y114AF155A-loop

1 234

567

772-

118-

118-194-

0.20

*

–0.1

0

–0.5

–0.4

–0.3

–0.2

ln U

/Uo

Time (s)

100 150500

1 2 3

BSA

La

La-loop

No protein

*

Figure 5 The La loop mutant is defective in tRNA maturation in vivo. (a) tRNA-mediated suppression (TMS) activity was assayed in S. pombe; pRep is the

empty vector; all other La constructs indicated were cloned in pRep17,18,36,38. (b) Northern blotting analysis of RNAs extracted from cells in a. Above, a blot

probed for intron-containing pre-tRNALysCUU species, which migrate as upper, middle and lower bands (see text). Below, the same blot reprobed for U5

snRNA, which served as a loading control for quantification, normalized to La ¼ 1.0, lane 3. (c) RNA chaperone assays were performed in vitro using the

cis-splicing intron RNA23. Activity is reflected by a decrease in unspliced RNA (ln U/Uo, see Methods) for each of the proteins indicated. Error bars

(standard error of the regression coefficient) reflect triplicate samples for each point. (d) Coomassie blue stained gel of BSA (lane 1), La (lane 2) and La-loop

(lane 3) proteins used in the assay.

ART IC L E S

43 4 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

to contact any other parts of La or UUU-3¢OH–containing RNA in thestructures reported4,6, suggesting that RRM1 is available to bind RNA.NMR revealed that one of the La-loop mutated residues, Lys151,showed a large chemical shift variation in the presence of RNA6. Thenotion that loop-3 residues contribute to an RNA binding platform issupported by the compound mutant, La–Y114A F155A–loop, which ismore severely defective for TMS and in vivo pre-tRNA accumulationthan is either of the La–loop-3 or La–Y114A F155A mutants alone(Fig. 5a,b). The data support the idea that stable RNA binding byRRM1 is mediated by multiple contacts to the b-sheet surface andloop-3, and possibly other residues.

Although the LM and RRM1 form independent structures in theabsence of RNA6,7, their orientations are fixed by contacts to boundRNA4,9. In a previously reported structure, double-stranded RNA4

raised concern about the exit path of the RNA10. Additional cocrystalstructures with all single-stranded RNAs have resolved this concernand added new insight9. In addition to closer LM-RRM1 packing,induced fit and plasticity in the UUU-3¢OH binding cleft, the newstructures show an RNA exit path that indeed differs from the double-stranded RNA–bound structure, as can be appreciated by comparingboth (Supplementary Fig. 8a,b online). These and other considera-tions9 favor two disparate sites on La9 that could bind distant regionsof an RNA separated by intervening RNA structure4, as in Supple-mentary Figure 8c.

As a separate binding platform, RRM1 contributes to versatilityThe basic nature of RRM1 loop-3 suggests RNA backbone contactsthat could support sequence-independent binding to various RNAsthat can derive additional affinity from the UUU-3¢OH binding site.This possibility, and the plasticity of the UUU-3¢OH site9, provideinsight into how La can show versatility for different RNAs.

We note that the affinity gained by two-site binding may be morecomplex than a simple addition of the affinity of each site individually.

We also note that the percentage of La protein that was active for high-affinity pre-tRNA binding was lower than the percentage active fortRNA and 3¢ trailer binding, in all Mg2+ concentrations tested andwhen using the same batch of La protein for the different RNAs. Webelieve that this should not be unexpected, because pre-tRNA bindingrequires both RNA binding sites be active, whereas 3¢ trailer and tRNAbinding each require only one site be active. Thus, the chances thateither of the two sites required for pre-tRNA binding will be inactivewould seem relatively high. We suspect that the orientations of the LMand RRM1 become more fixed relative to each other than when eitheris bound singly to its isolated RNA. Moreover, the manner in whichthe LM and RRM1 are fixed relative to each other may vary fordifferent RNAs.

La RRM1 functions in a complex activityLa proteins also function in the metabolism of non–UUU-3¢OH–containing RNAs. We showed that RRM1 is important for a complex,previously characterized activity of La, RNA chaperone activity.Holding different parts of an RNA by two disparate binding sitesconnected by intervening RNA may promote RNA folding or otherrearrangements by La domain proteins, as has been suggested forother RNA chaperones48,49. Conserved residues specifically on theRRM1 b-sheet of La and Sla1p have been shown to promote thematuration of structurally impaired pre-tRNA18. We suspect that theresults reported here will be useful toward understanding complexactivities of La in its interactions with mRNA and other RNAs, and toLARPs and their RNAs. Coupled use of distinct binding modes may bean important mechanism for RNP dynamics more generally.

METHODSMutagenesis. We carried out mutagenesis by QuikChange XL (Stratagene)

using pREP4-La17 and pET28a-La5 as templates, and verified all constructs

by sequencing.

Protein purification. Plasmids encoding La (pET28a-La) and mutated deriva-

tives were expressed in Escherichia coli BL21 Star(DE3)pLysS (Invitrogen).

Induced proteins were purified by nickel chromatography, concentrated and

desalted into 25 mM Tris, pH 8.0, 1 mM Mg2+, 100 mM KCl, 0.5% (v/v)

NP-40, 10% (v/v) glycerol and 1 mM DTT. Total protein yield was quantified

using the BCA Protein Assay (Pierce) and further assessed by SDS-PAGE and

Coomassie staining to ensure equal quantities of the various proteins.

RNA binding assays. We based the sequences of the 12-nt trailer, pre-tRNA

and mature tRNA substrates on that of the human tRNAArgICG26, with the pre-

tRNA sequence identical to that for the ‘R-11TUUU’ plus one extra terminal

uracil. We prepared and purified radiolabeled RNAs as described18. Briefly,

12-nt trailer (5¢-GUGUAAGCUUUU-3¢, IDT Technologies) was end labeled

using (32P-g)ATP and T4 polynucleotide kinase. Radiolabeled pre-tRNA and

mature tRNA ligands were synthesized by T7 RNA polymerase (Ambion

Megascript) using DNA templates generated by PCR of the human tRNAArg

ICG gene cloned into plasmid pUC19/R-11TUUU (a gift from M. Nashimoto).

*

*

RNase P

RNase Z

*

3′ CCA additionAminoacylationNuclear export

Nascent pre-tRNA

End-matured,modified tRNA

Pol III transcription

La

*

*

* **

***

** *

*

+

LaU

UU

UU

U

U

U U

PPP

UUU3′ trailer

PPP

P P

OH

OH

HO

OH

HO

*Nucleotide or base modification

Figure 6 Model of involvement of La protein in a tRNA maturation pathway.

La seems to be the first protein that binds nascent Pol III transcripts,

including pre-tRNAs (see text), and presumably does so via its UUU-3¢OH

binding cleft and/or RRM1. Different regions of the RNA may become

juxtaposed to RRM1 as the pre-tRNA folds, acquires nucleotide

modifications (indicated by asterisks) and becomes a substrate for

5¢ processing by RNase P. After separation of the tRNA and 3¢ trailer by

cleavage by the endonuclease, RNase Z, La is no longer tethered to two sites

on a single RNA and readily dissociates, free to associate with new nascent

pre-tRNA. The released, end-matured, modified tRNA becomes the substrate

of 3¢ end–modifying proteins such as the CCA-adding enzyme and tRNA

synthetases, and is exported from the nucleus.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 3 5

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Labeled RNAs were then PAGE purified. For method 1, approximately 15,000

c.p.m. of 5¢ 32P end–labeled 12-nt trailer or 3,000 c.p.m. of pre-tRNA or

mature tRNA ligand, (B0.1 nM each) were incubated with various amounts of

La proteins in 20 ml containing 20 mM Tris, pH 8.0, varying concentrations

of Mg2+, 100 mM KCl and 5 mM b-mercaptoethanol. Complexes were

incubated at 37 1C for 30 min, cooled on ice for 20 min, and then resolved

on 8% (w/v) polyacrylamide nondenaturing gels at 4 1C and 150V. Kd values

were approximated as the concentration of protein at which half of the RNA

substrate was bound33.

Method 2 was performed as described33. Briefly, after Kd values were

determined by method 1, an appropriate concentration of protein was

incubated with varying RNA concentrations estimated to give between 20%

and 80% protein occupancy, calculated using scatplan.xls (available at the

Setzer laboratory homepage http://www.biosci.missouri.edu/setzer/setzlab-

mu.htm#Spreadsheets). The concentration of protein active for each ligand

under different conditions can be derived from Scatchard analysis of method 2

(see the Setzer laboratory homepage). The active La concentration (and the

mass concentration (nM) used for each titration in Figure 2a–c is listed here in

parentheses) determined for pre-tRNAArgACG, tRNAArgACG and 12-nt UUU-

3¢OH trailer, to be 5.5 (10), 21 (20) and 51.5 (45) nM, respectively, in 0 Mg2+,

84 (250), 94 (250), and 46 (60) nM, respectively, in 10 mM Mg2+, and 40 (80),

86 (80) and 129 (160) nM, respectively, in 1 mM Mg2+. Complexes were

formed using constant RNA concentrations of 32P-RNA (B0.1 nM)

supplemented with cold RNA to yield the final RNA concentrations noted in

the figures. These RNA-protein complexes were formed and resolved as for

method 1. Bound and free fractions were quantitated using a Fuji-FLA 3000

phosphorimager; in most cases, the bound fraction included c.p.m. with

mobility above the free band. Using the bound and free quantities and the

known total concentrations of RNA added, we determined Kd values

using a nonlinear curve-fitting algorithm (the scatchd2.xls spreadsheet

found at the Setzer laboratory homepage) that fit the data to a standard

Scatchard analysis. Our Scatchard plots of method 2 yield straight lines

indicative of 1:1 stoichiometry. We note that our and others’ analysis of La

RNA binding by method 1 do indicate 1:2 stoichiometry; this occurs

only at high concentrations of La (for example, see Supplementary

Fig. 1, La concentration Z40 nM, reflected by a second RNP complex higher

in the gel). Consistent with a 1:1 stoichiometry, all of our method 2 analyses

were performed with relatively low La concentrations, for which only one

RNP is seen.

Dissociation rate constant assays. We performed dissociation rate constant

assays in the absence of Mg2+, essentially as described, in triplicate33 using the

ratern0.xls spreadsheet (Setzer laboratory homepage). Typical EMSA gels used

for dissociation rate are shown in Supplementary Figure 4. Briefly, La-RNA

complexes were formed in EMSA buffer using a concentration of La (200 nM)

determined to give approximately 95% 32P-RNA binding. We then determined

the concentrations of unlabeled competitor RNA that, when mixed with the32P-RNA before La addition, yielded 10–20% La-bound 32P-RNA (2,000 nM;

Supplementary Fig. 4, lanes 1). Dissociation rate constants were determined by

first forming complexes using only 32P-RNA (as in lane 2), then at time ¼ 0,

adding the predetermined amount of competitor cold RNA (as per lane 1), and

loading aliquots of the dissociation reaction at the noted time points. By

measuring the dissociation of the 32P-RNA from the La-bound form to the

unbound form over time, we calculated the dissociation rates of La for different

RNA substrates.

tRNA-mediated suppression experiments. We performed tRNA-mediated

suppression18 and northern blotting17 as described. The probe for U5 RNA

was 5¢-CTGGTAAAAGGCAAGAAACAGATACG-5¢.

RNA chaperone activity. RNA chaperone activity was monitored as

described23, except that RNA was radiolabeled with 32P instead of 35S, and

proteins assayed were added simultaneously with GTP. Intron self-splicing

was measured by quantification of the exponential decay of the unspliced

precursor (Unspliced/Spliced ¼ U) from the starting value of unspliced

precursor (Uo/So ¼ Uo), expressed as ln U/Uo over time. Unspliced and

spliced products were measured using phosphorimager analysis. The DNA

template for the precursor RNA was made by PCR amplification of a minimal

T4 phage self-splicing intron (gift from R. Schroeder, University of Vienna,

using an oligonucleotide containing a T7 promoter), followed by T7 transcrip-

tion and PAGE purification.

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ACKNOWLEDGMENTSWe thank D. Setzer for advice, protocols and RNA binding data analysis tools,M. Nashimoto (Niigata University of Pharmacy and Applied Life Sciences) forthe human tRNAArgACG gene and R. Schroeder (University of Vienna) forcis-splicing intron DNA. We thank D. Setzer, D. Engelke and M. Teplova forcomments. This work was supported by the Intramural Research Program of theUS National Institute of Child Health and Human Development, NationalInstitutes of Health.

AUTHOR CONTRIBUTIONSM.A.B. performed all experiments; R.J.M. and M.A.B. designed the study andanalyzed the data; R.J.M. wrote the paper with editing by M.A.B.

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Maraia, R.J. & Intine, R.V. Recognition of nascent RNA by the human La antigen:conserved and diverged features of structure and function. Mol. Cell. Biol. 21,367–379 (2001).

2. Wolin, S.L. & Cedervall, T. The La protein. Annu. Rev. Biochem. 71, 375–403(2002).

3. Maraia, R.J. & Bayfield, M.A. The La protein-RNA complex surfaces. Mol. Cell 21,149–152 (2006).

4. Teplova, M. et al. Structural basis for recognition and sequestration of UUU-OH3¢-termini of nascent RNA pol III transcripts by La, a rheumatic disease autoantigen.Mol. Cell 21, 75–85 (2006).

5. Goodier, J.L., Fan, H. & Maraia, R.J. A carboxy-terminal basic region controls RNApolymerase III transcription factor activity of human La protein. Mol. Cell. Biol. 17,5823–5832 (1997).

6. Alfano, C. et al. Structural analysis of cooperative RNA binding by the La motif andcentral RRM domain of human La protein. Nat. Struct. Mol. Biol. 11, 323–329(2004).

7. Dong, G., Chakshusmathi, G., Wolin, S.L. & Reinisch, K.M. Structure of the La motif: awinged helix domain mediates RNA binding via a conserved aromatic patch. EMBO J.23, 1000–1007 (2004).

8. Kenan, D.J. & Keene, J.D. La gets its wings. Nat. Struct. Mol. Biol. 11, 303–305(2004).

9. Kotik-Kogan, O., Valentine, E.R., Sanfelice, D., Conte, M.R. & Curry, S. Structuralanalysis reveals conformational plasticity in the recognition of RNA 3¢ ends by thehuman La protein. Structure 16, 852–862 (2008).

10. Curry, S. & Conte, M.R. A terminal affair: 3¢-end recognition by the human La protein.Trends Biochem. Sci. 31, 303–305 (2006).

11. Aigner, S. et al. Euplotes telomerase contains an La motif protein produced byapparent translational frameshifting. EMBO J. 19, 6230–6239 (2000).

12. Stone, M.D. et al. Stepwise protein-mediated RNA folding directs assembly oftelomerase ribonucleoprotein. Nature 446, 458–461 (2007).

13. He, N. et al. A La-related protein modulates 7SK snRNP integrity to suppress P-TEFb-dependent transcriptional elongation and tumorigenesis. Mol. Cell 29, 588–599(2008).

14. Krueger, B.J. et al. LARP7 is a stable component of the 7SK snRNP while P-TEFb,HEXIM1 and hnRNP A1 are reversibly associated. Nucleic Acids Res. 36, 2219–2229(2008).

15. Aigner, S., Postberg, J., Lipps, H.J. & Cech, T.R. The Euplotes La motif protein p43 hasproperties of a telomerase-specific subunit. Biochemistry 42, 5736–5747 (2003).

16. Yoo, C.J. & Wolin, S.L. The yeast La protein is required for the 3¢ endonucleolyticcleavage that matures tRNA precursors. Cell 89, 393–402 (1997).

17. Intine, R.V.A. et al. Transfer RNA maturation is controlled by phosphorylation of thehuman La antigen on serine 366. Mol. Cell 6, 339–348 (2000).

18. Huang, Y., Bayfield, M.A., Intine, R.V. & Maraia, R.J. Separate RNA-binding surfaceson the multifunctional La protein mediate distinguishable activities in tRNA matura-tion. Nat. Struct. Mol. Biol. 13, 611–618 (2006).

19. Chakshusmathi, G., Kim, S.D., Rubinson, D.A. & Wolin, S.L.A. La protein requirementfor efficient pre-tRNA folding. EMBO J. 22, 6562–6572 (2003).

20. Copela, L.A., Chakshusmathi, G., Sherrer, R.L. & Wolin, S.L. The La protein functionsredundantly with tRNA modification enzymes to ensure tRNA structural stability. RNA12, 644–654 (2006).

21. Blackburn, E.H. The end of the (DNA) line. Nat. Struct. Biol. 7, 847–850 (2000).22. Schroeder, R., Barta, A. & Semrad, K. Strategies for RNA folding and assembly.

Nat. Rev. Mol. Cell Biol. 5, 908–919 (2004).23. Belisova, A. et al. RNA chaperone activity of protein components of human Ro RNPs.

RNA 11, 1084–1094 (2005).

ART IC L E S

43 6 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

24. Hendrick, J.P., Wolin, S.L., Rinke, J., Lerner, M.R. & Steitz, J.A. Ro small cytoplasmicribonucleoproteins are a subclass of La ribonucleoproteins: further characterization ofthe Ro and La small ribonucleoproteins from uninfected mammalian cells. Mol. Cell.Biol. 1, 1138–1149 (1981).

25. Maraia, R.J. & Intine, R.V. La protein and its associated small nuclear and nucleolarprecursor RNAs. Gene Expr. [review] 10, 41–57 (2002).

26. Nashimoto, M., Nashimoto, C., Tamura, M., Kaspar, R.L. & Ochi, K. The inhibitoryeffect of the autoantigen La on in vitro 3¢ processing of mammalian precursor tRNAs.J. Mol. Biol. 312, 975–984 (2001).

27. Hopper, A.K. & Phizicky, E.M. tRNA transfers to the limelight. Genes Dev. 17,162–180 (2003).

28. Phizicky, E.M. Have tRNA, will travel. Proc. Natl. Acad. Sci. USA 102, 11127–11128(2005).

29. Haeusler, R.A. & Engelke, D.R. Spatial organization of transcription by RNA polymer-ase III. Nucleic Acids Res. 34, 4826–4836 (2006).

30. Pfeifle, J., Anderer, F.A. & Franke, M. Multiple phosphorylation of human SS-B/Laautoantigen and its effect on poly(U) and autoantibody binding. Biochim. Biophys.Acta 928, 217–226 (1987).

31. Huang, Y., Intine, R.V., Mozlin, A., Hasson, S. & Maraia, R.J. Mutations in the RNApolymerase III subunit Rpc11p that decrease RNA 3¢ cleavage activity increase3¢-terminal oligo(U) length and La-dependent tRNA processing. Mol. Cell. Biol. 25,621–636 (2005).

32. Draper, D.E., Grilley, D. & Soto, A.M. Ions and RNA folding. Annu. Rev. Biophys.Biomol. Struct. 34, 221–243 (2005).

33. Setzer, D.R. Measuring equilibrium and kinetic constants using gel retardation assays.Methods Mol. Biol. 118, 115–128 (1999).

34. Ohndorf, U.M., Steegborn, C., Knijff, R. & Sondermann, P. Contributions of theindividual domains in human La protein to its RNA 3¢-end binding activity. J. Biol.Chem. 276, 27188–27196 (2001).

35. Horke, S., Reumann, K., Schweizer, M., Will, H. & Heise, T. Nuclear trafficking of Laprotein depends on a newly identified NoLS and the ability to bind RNA. J. Biol. Chem.279, 26563–26570 (2004).

36. Intine, R.V., Dundr, M., Misteli, T. & Maraia, R.J. Aberrant nuclear trafficking of Laprotein leads to disordered processing of associated precursor tRNAs. Mol. Cell 9,1113–1123 (2002).

37. Van Horn, D.J., Yoo, C.J., Xue, D., Shi, H. & Wolin, S.L. The La protein inSchizosaccharomyces pombe: a conserved yet dispensable phosphoprotein that func-tions in tRNA maturation. RNA 3, 1434–1443 (1997).

38. Bayfield, M.A., Kaiser, T.E., Intine, R.V. & Maraia, R.J. Conservation of a maskednuclear export activity of La proteins and its effects on tRNA maturation. Mol. Cell.Biol. 27, 3303–3312 (2007).

39. Wang, Z., Luo, T. & Roeder, R.G. Identification of an autonomously initi-ating RNA polymerase III holoenzyme containing a novel factor that is selectivelyinactivated during protein synthesis inhibition. Genes Dev. 11, 2371–2382(1997).

40. Fairley, J.A. et al. Human La is found at RNA polymerase III-transcribed genes in vivo.Proc. Natl. Acad. Sci. USA 102, 18350–18355 (2005).

41. French, S.L. et al. Visual analysis of the yeast 5S rRNA gene transcriptome: regulationand role of La protein. Mol. Cell. Biol. 28, 4576–4587 (2008).

42. Anderson, J. et al. The essential Gcd10p-Gcd14p nuclear complex is required for1-methyladenosine modification and maturation of initiator methionyl-tRNA. GenesDev. 12, 3650–3662 (1998).

43. Calvo, O. et al. GCD14p, a repressor of GCN4 translation, cooperates with Gcd10p andLhp1p in the maturation of initiator methionyl-tRNA in Saccharomyces cerevisiae.Mol. Cell. Biol. 19, 4167–4181 (1999).

44. Johansson, M.J. & Bystrom, A.S. Dual function of the tRNA(m5U54)methyltransferasein tRNA maturation. RNA 8, 324–335 (2002).

45. Nishikura, K. & De Robertis, E.M. RNA processing in microinjected Xenopus oocytes.Sequential addition of base modifications in the spliced transfer RNA. J. Mol. Biol.145, 405–420 (1981).

46. Maglott, E.J., Deo, S.S., Przykorska, A. & Glick, G.D. Conformational transitions of anunmodified tRNA: implications for RNA folding. Biochemistry 37, 16349–16359(1998).

47. Maris, C., Dominguez, C. & Allain, F.H. The RNA recognition motif, a plasticRNA-binding platform to regulate post-transcriptional gene expression. FEBS J.272, 2118–2131 (2005).

48. Mikulecky, P.J. et al. Escherichia coli Hfq has distinct interaction surfaces for DsrA,rpoS and poly(A) RNAs. Nat. Struct. Mol. Biol. 11, 1206–1214 (2004).

49. Rajkowitsch, L. & Schroeder, R. Dissecting RNA chaperone activity. RNA 13,2053–2060 (2007).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 3 7

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

S16 throws a conformational switch during assembly of30S 5¢ domainPriya Ramaswamy1,3 & Sarah AWoodson2

Rapid and accurate assembly of new ribosomal subunits is essential for cell growth. Here we show that the ribosomal proteinsmake assembly more cooperative by discriminating against non-native conformations of the Escherichia coli 16S ribosomal RNA.We used hydroxyl radical footprinting to measure how much the proteins stabilize individual ribosomal RNA tertiary interactions,revealing the free-energy landscape for assembly of the 16S 5¢ domain. When ribosomal proteins S4, S17 and S20 bind the5¢ domain RNA, a native and a non-native assembly intermediate are equally populated. The secondary assembly protein S16suppresses the non-native intermediate, smoothing the path to the native complex. In the final step of 5¢ domain assembly, S16drives a conformational switch at helix 3 that stabilizes pseudoknots in the 30S decoding center. Long-range communicationbetween the S16 binding site and the decoding center helps to explain the crucial role of S16 in 30S assembly.

Rapidly dividing cells must produce hundreds of new ribosomes eachminute1,2. Consequently, the process of ribosome assembly must beaccurate, so that each subunit is active, and stringently controlled, sothat the capacity for protein synthesis matches the rate of growth3,4.Large ribosomal RNAs (rRNAs) form metastable structures that canlead to errors in assembly5. How the ribosomal proteins and externalassembly factors remodel these intermediates is important to thefidelity of ribosome assembly.

Reconstitution studies on the E. coli 30S ribosomal subunitshowed that the ribosomal proteins induce large changes in thestructure of the 16S rRNA that underlie the cooperativity andhierarchy of the 30S assembly map6–8 (Fig. 1a,b). Whereas themechanisms by which the central and 3¢ domains of the 16S rRNAare assembled have been addressed9–12, assembly of the 16S5¢ domain, which makes up the body of the 30S subunit13,14

(Fig. 1c), is poorly understood. A 16S fragment containing the5¢ domain forms a stable ribonucleoprotein (RNP) with ribosomalproteins S4, S17, S20 and S16 (ref. 15). Primary assembly proteinsS4, S17 and S20 bind the naked rRNA, whereas binding of S16requires S4 and S20 (ref. 16; Fig. 1b). As the 5¢ domain is the first tobe transcribed in vivo and its proteins make interdomain contacts,rapid formation of its stable rRNA and rRNA-protein interactionsnucleates 30S assembly17,18.

Using hydroxyl radical footprinting, we previously showed that theE. coli 16S 5¢ domain RNA can form all of the backbone interactionspredicted by the structure of the 30S subunit in the absence ofproteins19. However, interactions between helices 15 and 17 requiredmore than 5 mM MgCl2, and some helices were protected lessstrongly in the naked RNA than in native 30S ribosomes. Thus,

the 5¢ domain proteins are needed to stabilize the rRNA tertiarystructure in physiological Mg2+ concentrations.

Moreover, time-resolved footprinting showed that half of the5¢ domain RNA became kinetically trapped in non-native foldingintermediates when refolded in vitro in 20 mM MgCl2

19. Tertiaryinteractions between helix 15 and helix 17 required the longest time toform (B1 min), probably owing to misfolding of the central junctionbetween helices 5, 6 and 6a. These results raised the question ofwhether the proteins also change the pathway of assembly, avoidingunproductive conformations.

To determine whether ribosomal proteins redirect the foldingpathway of the rRNA, we probed the assembly landscape of theE. coli 16S 5¢ domain RNP using quantitative hydroxyl radicalfootprinting. This method detects the solvent accessibility of indivi-dual residues along the RNA backbone, providing a detailed picture ofthe RNA tertiary interactions20. Information about the thermo-dynamic stability of each contact in the presence and absence of theproteins was obtained by probing the complexes over a wide range ofMg2+ concentrations.

The results show that binding of S16 to helices 15 and 17 results in aconformational switch at helix 3, 30 A away, which stabilizes tertiaryinteractions in the 30S decoding site. We also find that S16 increasesthe cooperativity of RNP assembly by preferentially stabilizing thenative configuration of helices in the lower half of the 5¢ domain, whiledisfavoring non-native assembly intermediates. Together, these resultshelp explain the crucial role of S16 in 30S assembly. They alsodemonstrate that discrimination against non-native structures isanother way in which RNA-protein interactions increase the selectivityof molecular self-assembly.

Received 29 September 2008; accepted 9 March 2009; published online 3 April 2009; doi:10.1038/nsmb.1585

1Program in Cell, Molecular and Developmental Biology and Biophysics and 2T. C. Jenkins Department of Biophysics, Johns Hopkins University, Baltimore, Maryland,USA. 3Present address: Department of Biochemistry and Biophysics, University of California San Francisco, San Francisco, California, USA. Correspondence should beaddressed to S.A.W. (swoodson@jhu.edu).

43 8 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

ART IC L E S

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

RESULTSStability of the naked 16S 5¢ domain RNATo determine how ribosomal proteins stabilize the folded 16S5¢ domain RNA, we compared the naked rRNA with RNPs containingthe primary binding proteins S4, S17 and S20, or S4, S17 and S20 plusprotein S16. The structures of the complexes were probed by hydroxylradical in 330 mM KCl and 0–30 mM MgCl2 (see Methods andSupplementary Figs. 1–3 online). The extent of cleavage was quanti-fied at more than 65 independent segments of the rRNA back-bone. In general, the stability of RNA tertiary structure is inverselyrelated to the Mg2+ dependence of the foldingtransitions21–23. Thus, we expect the RNAinteractions to form at lower Mg2+ concentra-tions when more proteins join the complex.The extent of cleavage in hydroxyl radical

correlates with the solvent accessibility of each ribose20, which reflectsthe sum of all folding equilibria that lead to exposure or protection ofthat residue. Therefore, the Mg2+ required to protect each segment ofthe RNA reflects the free energy of specific assembly intermediates.

In the absence of proteins (Fig. 2, Supplementary Table 1 andSupplementary Fig. 3b online), we found that tertiary interactions inthe RNA were heterogeneous and fell into three general categories.Some interactions required little Mg2+ to be stable (pink, Fig. 2b), butthey were not protected to the same extent as were control reactionsrun in parallel on native 30S subunits. Others were not protected,even up to 30 mM Mg2+ (green, Fig. 2b). Still others folded in twodistinct phases, suggesting the presence of folding intermediates(black, Fig. 2b).

Residues that were protected in o2 mM MgCl2 included nucleo-tides in helices 17 and 18 adjacent to the ‘upper’ five helix junctionthat binds with protein S4. A stable core of tertiary structure was alsovisible around helix 6a and the central junction, which aligns theinterface between helix 6/6a and helix 7 (red, Fig. 3a). The lowerjunction between helices 7–10 folded in 2–13 mM MgCl2 (orange andgreen, Fig. 3a).

Many other regions of the 5¢ domain remained exposed to hydroxylradical in 20 mM Mg2+ (blue, Fig. 3a), including helices that form thebinding sites for the primary assembly proteins S4, S17 and S20. Inour previous studies, the naked 5¢ domain RNA was almost completelyfolded in 20 mM MgCl2 and 120 mM NH4Cl19. The lesser stability ofthe rRNA tertiary structure reported here reflects the competitionbetween Mg2+ and 330 mM K+ for access to the RNA and the largersize of the K+ ion relative to NH4

+ (ref. 24). K+ is often used toreconstitute 30S subunits and more closely mimics the intracellularenvironment. Thus, under ‘physiological’ conditions, the ribosomalproteins are needed to fully stabilize the rRNA.

Stable 5¢ domain RNPTo determine whether the 5¢ domain RNA can assemble completelywith the four 5¢ domain proteins, we incubated the rRNA withproteins S4, S16, S17 and S20 in 0–30 mM MgCl2 before hydroxyl

17

16

18

a b

c

S17

S16 S6 S9 S19

S13

S14S10

S11

S5

S21 S3 S2

S12

S18

S20 S4 S8 S15 S7

3

5′

3′

S4

S16

S17

S20

5

15

1413

12

7

810

9

11

66a

4

2

5′ domain 3′ domainCentral

1

Figure 1 Structure of the E. coli 5¢ domain. (a) Secondary structure of the

16S rRNA55 with 5¢ domain nucleotides 21–562 in blue. Helices are

numbered as in refs. 13,56. (b) 30S assembly map16 with four 5¢ domain

proteins used in this study in color. (c) Structure of the 5¢ domain in the

E. coli 30S ribosome (PDB 2AVY)28, which forms the body of the small

subunit. S4, pink; S16, blue; S17, green; S20, yellow.

1

0.8

MgCl2 (mM)a b

c

NT

NT

30S

30S

0.01

0.05

0.1

1 2 3 71.5

12 15 20 25 350.2

0.3

0.5

G A H K K RNA only

388–390469

250–254

388–390

469

250–254

0.6

0.4

0.2427–429

451–452

481–483487–488

495–496

499–500504–505

516–518

521–522

525–527

529–531

533–536

513

498

469,471

0

0.01 0.1

RNA + 4 proteins

[MgCl2] (mM)1 10 100

Y

1

0.8

0.6

0.4

0.2

0.01 0.1

[MgCl2] (mM)

1 10 100

0

Y

Figure 2 Hydroxyl radical footprinting of the

5¢ domain in the presence and absence of

proteins. (a) The 16S 5¢ domain (nucleotides

21–562) was folded for 30–40 min at 37 1C in

0–35 mM MgCl2 before Fe(II)-EDTA footprinting

and primer extension (see Methods). Lanes: NT,

no treatment; H, RNA only in 80 mM K+-HEPES;

K, RNA only in 80 mM K+-HEPES plus 330 mM

KCl; G A, sequence ladder; 30S, native 30S

ribosomes. Protections due to RNA-RNA contacts

predicted by the structure of the 30S ribosome

are indicated on the right. (b) Folding transitions

for individual RNA-RNA contacts in the absence

of proteins. The relative extent of protectionY was normalized to controls on native 30S

ribosomes (squares) and fit to two- or four-state

models (see Methods). Green, nucleotides

250–254; black, nucleotides 388–390; pink,

nucleotide 469. (c) Fits for the same nucleotides

as in b, but in the presence of proteins S4, S16,

S17 and S20. Symbols as in b.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 3 9

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

radical footprinting with Fe(II)-EDTA. As expected, binding of thefour 5¢ domain proteins strongly stabilized the tertiary interactions inthe 16S 5¢ domain (Fig. 2c and Supplementary Fig. 3a). In 5 mMMgCl2, many tertiary interactions that were undetectable in the RNAalone were formed in the 5¢ domain RNP to the same degree as in thenative 30S subunit (Fig. 3). Moreover, the pattern of hydroxyl radicalprotection and the changes in chemical base modification wereconsistent with previous footprinting studies of 5¢ domain pro-teins25–27 (Supplementary Fig. 1 and 2) and with the backbonecontacts predicted by crystal structures of the 30S ribosome13,28 (seeMethods). Thus, the 5¢ domain RNP assembles completely underphysiological conditions.

Primary binding proteinsWe next asked to what extent the three primary assembly proteinswithout S16 could stabilize the three-dimensional structure of therRNA. When we incubated the 5¢ domain RNA with a mixture of S4,S17 and S20, most of the expected rRNA tertiary contacts formed in2.3 mM MgCl2, (Fig. 3b and Supplementary Fig. 3c). Only a fewinteractions between helices 8 and 6, 10 and 17, at the tip of helix 12and in helix 18, required more than 2.3 mM MgCl2 to formcompletely. Thus, S4, S17 and S20 together stabilize nearly all thenative tertiary contacts in the 5¢ domain, except those near the helix 18pseudoknot and a few positions in the core of the domain.

The primary binding proteins also perturb the ensemble of initialRNA structures. When the naked 5¢ domain RNA was incubated in330 mM KCl without Mg2+, most of the RNA backbone wasmoderately cleaved (Fig. 2a), suggesting that the entire domain isdynamic or disordered, adopting many conformations. By contrast,certain nucleotides were cleaved much more strongly in low Mg2+

concentrations in the presence of S4, S17 and S20 than in the nakedRNA (Fig. 4a,b). The exposed nucleotides are located in helix 5(nucleotides 50–60, 352, 355, 365), helix 7 (118), helix 8 (176–177),helix 12 (314), helix 13 (328) and helix 15 (370, 372, 392, 396), where

they participate in tertiary interactions with adjacent helices in themature 30S subunit13,14 (Fig. 4c,d).

Notably, the residues exposed at low Mg2+ concentration by bindingof S4, S17 and S20 overlap the binding site for protein S16(refs. 25,27; Fig. 4d). As the S4–S17–S20 complexes were titratedwith Mg2+, the exposed residues became protected from hydroxylradical cleavage, consistent with formation of additional tertiarystructure (Fig. 4a,b and Supplementary Fig. 4 online). Thus, theprimary assembly proteins not only stabilize the overall tertiarystructure of the 5¢ domain RNA, they also specifically pre-organizethe S16 binding site. Moreover, S4, S17 and S20 in combination narrowthe ensemble of assembly intermediates, favoring rRNA conformationsthat are prepared to make the desired tertiary interactions. These resultshelp explain the cooperative interactions between S16 and S4 or S20 inthe 30S assembly map16.

RNAa b c

0

S4–S17–S20

≤20 mM Mg2+

S4–S17–S20–S16

Figure 3 Global stabilization of rRNA tertiary structure by ribosomal proteins. Individual residues in the 5¢ domain RNA were clustered according to the

[Mg2+]1/2 of the folding transition: red, 0–2.3 mM; orange, 2.3–4.9 mM; green, 4.9–13.4 mM; blue, 413 mM. Residues protected in two transitions were

ranked according to the midpoint of the second transition (further details in Supplementary Table 1). Two-dimensional schematics and three-dimensional

ribbons prepared as in Figure 1. (a) 5¢ domain RNA only; (b) RNA plus S4, S17 and S20; (c) RNA plus S4, S17, S20 and S16.

GANNHKK

RNA onlya c

db

35 mM

30 mM

314

H17

H4

H16

H15

H13

H17

H4

H15

30S

H5

H13

H8

H10 H8

H5

328

352355

365370372

314

328

352355

365370372

392396

392396

S4–S17–S20

0.01

GANNHKKM 0.01

Figure 4 Primary assembly proteins pre-organize the S16 binding site.

Enhanced hydroxyl radical cleavage of specific nucleotides in low Mg2+

concentrations when proteins S4, S17 and S20 are bound, relative to the

naked RNA. These residues become protected in 20 mM MgCl2. Lanes arelabeled as in Figure 2. (a) RNA only. (b) RNA plus S4, S17 and S20. See

Supplementary Figure 4 for additional data. (c,d) Exposed residues in low

Mg2+ concentrations (red) overlap the S16 binding site (nucleotides 51 and

362–364 on both sides of helix 5, and nucleotides 120, 315, 324 and

390–393)25,27,38.

ART IC L E S

44 0 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

S16 binds the native 5¢ domain coreProtein S16 is essential for viability29, and its absence strongly affectsthe kinetics of 30S reconstitution in vitro30. When bound to the 16SrRNA, S16 induces large changes in the conformation of 5¢ and centraldomains25,27, consistent with its important role in 30S assembly. Tounderstand the function of S16 in assembly, we deduced the specificeffects of S16 on the stability of the 5¢ domain by comparing Mg2+-titrations of S4–S17–S20 complexes with titrations of complexescontaining four proteins (Fig. 3b,c).

In the 30S subunit, S16 straddles a C-loop motif in helix 15 thatstacks against bases in a sharp kink in helix 17, and Tyr17 in S16donates a hydrogen bond to the 2¢ OH of A374 in the C-loop13.Addition of S16 stabilized this important tertiary contact in the rRNA,reducing the midpoint for protection of nucleotides 481–483 at thekink in helix 17 from 4.9 mM to 2.4 mM MgCl2 (SupplementaryTable 1). More unexpectedly, we found that S16 also stabilized manybackbone contacts with helix 6/6a in the core of the 5¢ domain,including contacts with helix 8 (nucleotides 151–153), helix 13 andS20 (nucleotides 107–108), helix 10 and the tip of helix 17 (nucleo-tides 203–204). Helix 6 forms a spur that juts into the solvent,emerging from a bundle of helices at the base of the 30S subunit13,14.Thus, S16 binding directly stabilizes the interaction between helices 15and 17 and indirectly improves helix packing around the 30S spur.

S16 changes the structure of helix 12S16 was previously shown to induce a shift in the secondary structureof helix 12 and to decrease the accessibility of G31 in helix 3 and basesin the central pseudoknot (helix 2)27. We observed that S16 stabilizedtertiary interactions between helix 3, 12 and 18, which form part of theinterface between the body of the 30S subunit and the platform. Theseinclude a contact between the tip of helix 12 (nucleotides 295–297)and the minor groove of helix 3, as well as an interaction betweennucleotide 301 and a flexible loop in S17 (Fig. 3c).

S16 stabilizes the decoding site pseudoknotIn addition to its interactions with helix 15 and the domain core, theMg2+ titrations showed that protein S16 stabilizes the pseudoknotformed by base pairs 505–507 and 524–526 in helix 18 at physiologicalMg2+ concentrations (Fig. 3c and Supplementary Table 1). The helix18 pseudoknot positions the universally conserved G530 in the 30Sdecoding site and is essential for protein synthesis31. In the 30Ssubunit, nucleotides 505–507 are buried by a kink in the RNAbackbone created by the pseudoknot and by contact with theN-terminal domain of protein S4. These residues were only partiallyprotected in complexes with S4 alone (P.R. and S.A.W., unpublisheddata), indicating that interactions elsewhere in the rRNA are impor-tant for the stability of the helix 18 pseudoknot. Addition of S16lowered the [Mg2+]1/2 for protection of nucleotides 505–506 to2.3 mM (versus 5.6 mM with S4–S17–S20) and increased the maxi-mum protection to 80% of the value obtained for 30S controls. Thus,in our experiments, the helix 18 pseudoknot forms in physiologicalMg2+ only when S16 is added to the complex.

Reorganization of intermediate complexesAlthough many 5¢ domain interactions become more stable asproteins join the RNP, segments of the rRNA backbone showed amore complex change in solvent accessibility that revealed the reorga-nization of intermediate RNPs during the assembly process (Fig. 5).When the 5¢ domain RNA was incubated with the three primarybinding proteins, certain residues were protected in two transitions,indicating the presence of one or more intermediates in the pathway.For example, in the S4–S17–S20 complex, nucleotides 379–380 inhelix 15 fold in two stages. These nucleotides are partially protected in0 mM Mg2+ and fully protected in 10 mM Mg2+ (Fig. 5a). A secondgroup of residues, such as those in helix 12, become lightly protectedin 0.1 mM Mg2+ and fully protected in 10 mM Mg2+ (Fig. 5b).

Notably, a third group of residues in helix 18 were partiallyprotected in 0–1 mM Mg2+, exposed to hydroxyl radical cleavagebetween 1 mM Mg2+ and 5 mM Mg2+ and reprotected in 10–20 mMMg2+ (Fig. 5c). This oscillation in the solvent accessibility of the RNAbackbone is best explained by the remodeling of tertiary interactionsduring assembly. We observed the same behaviors for other residues inthese helices, consistent with a structural change affecting the entirehelix (Supplementary Figs. 5 and 6 online). Single-protein complexesshow similar multistage folding (P.R. and S.A.W., unpublisheddata), suggesting that these structural changes are intrinsic to the 5¢domain RNA.

S16 suppresses non-native assembly intermediatesAddition of S16 eliminated this multistage folding, causing tertiaryinteractions with helices 12, 15 and 18 to form cooperatively over anarrow range of Mg2+ concentrations (Fig. 5 and SupplementaryFig. 6). Not only did the final folding transition occur at a lower Mg2+

concentration when S16 was added, but the first folding transition

1

0.8

a

b

c

Helix 15

+1°

+1°

+1°

1° +S16

1° +S16

RNA

RNA

1° +S16

Helix 12

Helix 18

RNA

0.6

0.4

0.2

0

0.8

0.6

0.4

0.2

0

0.8

0.6

0.4

0.2

0

0.01 0.1 1 10 100

[MgCl2] (mM)

Y

Y

Y

Figure 5 S16 discriminates against non-native assembly intermediates.

(a,b) Formation of tertiary interactions in helice 15, 12 and 18. The extent

of protection (Y ) relative to native 30S ribosomes was as above (see

Methods). Representative titrations are shown for (a) nucleotides 379–380

(helix 15); (b) nucleotide 315 (helix 12); (c) nucleotide 501–502 (helix 18).

Open circles, RNA only; filled squares, S4–S17–S20 (11); filled circles,

S4–S16–S17–S20. The lower baseline varies for residues in helix 18 in

titrations with four proteins, owing to heavy cleavage of unfolded RNA

controls in these particular experiments. Further data are shown in

Supplementary Figure 5.

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 4 1

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

moved to a higher Mg2+ concentration or disappeared. Thus, bindingof S16 stabilizes certain assembly intermediates while antagonizingothers. We observed the conversion of multistage transitions in theabsence of S16 to two-state transitions in the presence of S16 for manyresidues in the 5¢ domain, suggesting that fewer assembly intermedi-ates are populated when S16 binds the rRNA.

Conformational switch at helix 3The structures of candidate assembly intermediates were deduced byclustering residues that followed similar folding transitions whenbound by S4–S17–S20. Residues folding in two stages were locatedon both faces of helix 15, the inward face of helix 17 and the tips ofhelices 11 and 18 (blue, Fig. 6). The second cluster of residues map tothe base of helix 12, the interface between helices 6a and 10, and helix16 where it contacts helix 18 (orange, Fig. 6). In the 30S subunit, helix15 lies between helix 17 and helix 6a, whereas the other side of helix 6apacks against helices 8 and 10 (refs. 13,14). Thus, these two clustersrepresent a set of helix packing interactions in the lower half ofthe domain.

Because these residues are protected in two stages, we deducethat the core of the 5¢ domain adopts at least two folded con-formations during assembly: a native-like conformation (IN) that issimilar to the 30S structure, and a non-native conformation (InC)that leaves certain residues exposed to hydroxyl radical. In themature subunit, helices 6, 6a and 12 are arranged co-axially and arelinked to each other and to helix 5 through the central junction.Residues in these helices fold together (orange), suggesting that thedifference between the native and non-native intermediates may bedue to alternative base-pairing of the central junction, which wehave observed previously19.

Residues that are exposed in moderate Mg2+ concentrations andreprotected in high Mg2+ concentrations cluster in helix 18 and in thetip of helix 12 (nucleotides 295–297) (pink, Fig. 6, and Supplemen-tary Fig. 3c). The backbone of these residues is protected by helix 3,which lies between helix 18 and the tip of helix 12 in the maturesubunit. Consequently, the transient exposure of helices 12 and 18 tosolvent at moderate Mg2+ concentration can be explained by a switchin the relative orientation of helix 3.

When we compared folding transitions in different parts of the5¢ domain (Fig. 6a), it was apparent that helix 18 (pink) becomesexposed over the same Mg2+ concentration range in which helix 15and the base of helix 12 (blue and orange) become fully protected

(Fig. 6a). Therefore, we infer that interactions between helices 3 and18 break when the native-like IN intermediate is populated. Thecorrelation between the orientation of helix 3 and the IN state iscorroborated by changes in the solvent accessibility of all the residuesin these helices and in the core of the 5¢ domain (SupplementaryFig. 6). Core residues that are protected in a single transition centeredat 1–2 mM Mg2+ (nucleotides 151–153, 469, 370–374) probably reflectinteractions that exist in the native-like core but not in the non-nativecore. Other residues (nucleotides 203–204, 301, 481–483, nucleotides505–506) require 5 mM or 10 mM Mg2+ to become protected in theabsence of S16, representing contacts that appear in the native RNP. Asdiscussed below, the self-consistency of the footprinting results allowsus to propose a structural model for assembly of the 16S 5¢ domain.

DISCUSSIONMany RNA binding proteins preferentially bind and stabilize thenative three-dimensional structure of their RNA target. In multi-protein RNPs such as the signal recognition particle (SRP) and the 16Scentral domain, the resulting protein-induced changes in the structureof the RNA favor the addition of subsequent proteins, drivingcooperative assembly of the entire complex32–34. Our studies on the16S 5¢ domain show that S16 not only stabilizes the native conforma-tion of the rRNA, it also destabilizes certain rRNA interactions at earlystages of assembly. These results reveal a second important role ofRNA binding proteins in RNP assembly, which is to suppressunproductive intermediates. We propose that the suppression ofmetastable intermediates is an important factor in the cooperativityof RNP assembly.

Assembly energy landscapeThe actions of the 5¢ domain proteins can be understood in terms ofan energy landscape for assembly (Fig. 7), as previously proposedbased on the kinetics of 30S assembly35. As each protein binds therRNA, the relative free energies of the rRNA structures are changed,resulting in a new distribution of assembly intermediates. Together,the three primary assembly proteins S4, S17 and S20 stabilize nearly allthe RNA tertiary interactions in the 16S 5¢ domain. However, theseform at different Mg2+ concentrations, suggesting that assembly passesthrough at least two intermediates, IN and InC (see below). S16 lowersthe free energy of IN but not InC, preventing the accumulation of InC

complexes. Consequently, when S16 is present, assembly proceedsdirectly from IN to N under physiological conditions (2–5 mM Mg2+).

Evidence for competing I states comes from the multistage changesin backbone accessibility as the complexes are stabilized by Mg2+. Onepotential explanation for the plateau in the protection of helix 15 isthat assembly passes through an intermediate in which helix 15 is onlypartially buried (scheme I). However, if ribosomal proteins such asS16 bind the native rRNA more strongly than non-native or unfoldedrRNA, then, according to the thermodynamic cycle, they must

1

0.8

a b

c

H15

H12 H180.6

0.4

0.2

0.01

H16

H12

H12

H11 H7H17

H17

H16

H15

H4

H3

H18

H15

H6

H6

H10

H10

H18

H3

0.1[MgCl2] (mM)

10 10010

Y

Figure 6 RNA conformational changes during assembly. Residues protected

in two or more steps during assembly in the presence of S4, S17 and S20

form three clusters: blue, 20–60% protected in KCl and fully protected in

1 mM MgCl2; orange, 20–30% protected in 0.1 mM MgCl2 and fully

protected in 1 mM MgCl2; pink, partially protected in KCl, exposed in

1 mM MgCl2 and reprotected in 10 mM MgCl2. (a) Sample titration curves

for H15 nucleotides 379–380 (blue), H12 nucleotide 312 (orange) and

H18 nucleotide 495–496 (pink). Curves for other residues are shown in

Supplementary Figure 6. (b,c) Residues in each cluster are projected on the

two-dimensional and three-dimensional structures of the 30S 5¢ domain.

Stereo ribbon of the three-dimensional structure as in Figure 1c, with proteins

shown as semitransparent surfaces: S4, pink; S17, green; S20, yellow.

ART IC L E S

44 2 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

stabilize native RNA interactions more than they do non-nativeinteractions. If the partially protected intermediate in scheme Icontains only native RNA interactions, then both the intermediateand the native state should be stabilized when S16 binds, and bothtransitions should occur at lower Mg2+ concentrations. However,we observe that the initial folding transition is disfavored whenS16 is added to the reaction (Fig. 5a and Supplementary Fig. 6).Moreover, the transient exposure of helix 18 during assembly in theabsence of S16 strongly argues for a second intermediate with adifferent conformation.

Scheme I:

U

Cleaved 50% protected Protected

NIN

An alternative model that is consistent with all of our data is thatassembly involves at least two intermediates with different structures(scheme II): IN in which helix 12 is buried and helix 18 is exposed, andInC in which helix 12 is exposed and helix 18 is buried. We proposethat some residues in helix 15 are protected in both IN and InC. If InC

forms at low Mg2+ concentrations, this would explain why suppres-sion of InC by S16 increases cleavage of the helix 15 backbone. In this‘energy landscape’ model, the extent of protection plateaus when IN

and InC have similar free energies, and thus neither intermediatedominates the population (Fig. 7a).

Scheme II:

UCleaved

Cleaved

Protected

ProtectedN

InC

IN

Protein S16 smooths the pathway of assembly considerably bystabilizing IN more than InC, such that only IN is populated(Fig. 7a). This could occur by selective binding to IN. However, S16could also raise the free energy of InC. In either case, once there is asubstantial energy gap between the two intermediate states, only themost stable state will be populated at equilibrium.

The four-state model in scheme II can account for all of themultistage changes in backbone accessibility. By contrast, the oscillat-ing protection of helix 18 cannot be explained by the three-statemodel in scheme I (Supplementary Fig. 7 online). Thus, at least twointermediates are needed to explain assembly of the 5¢ domain.Additional intermediates containing only one or two proteins arealso likely. Therefore, the free-energy landscape for assembly is almostcertainly more complex than depicted in Figure 7.

Mechanism of 5¢ domain assemblyFrom the footprinting results presented here, we propose the followingminimal mechanism for 5¢ domain assembly (Fig. 7b,c): at low Mg2+

concentrations, the 5¢ domain RNA forms an ensemble of partly foldedstates with a minimal set of stable tertiary interactions (IC). Thisensemble contains both native-like and non-native configurations ofcore helices 6, 6a, 7 and 11 that are bound and further stabilized byprimary assembly proteins S4, S17 and S20. In IN, we propose thathelix 3 is oriented away from the upper five-helix junction, leaving thetip of helix 12 and the stem of helix 18 exposed to solvent (Fig. 7b).Although we do not know the exact orientation of helix 3 in IN, aswitch in the topology of the right angle junction between helices 3 and4 could swing helix 3 away from the rest of the 5¢ domain36. In InC,helix 3 is aligned correctly with helix 18, but the base of helix 12 isunprotected owing to the perturbed conformation of the core helicesand the central junction between helices 5, 6a and 12.

When S16 binds (or when the Mg2+ concentration reaches 3 mM),IN becomes more stable than InC, resulting in full protection of helix 15and exposure of helix 18 and the tip of helix 12 (Fig. 6a). Helix 18 and12 are finally reprotected as helix 3 swings back into its native orien-tation (N). Under equilibrium conditions, we cannot distinguishwhether InC proceeds to N directly in the absence of S16, or whethernon-native complexes must first transform to IN before reaching thenative RNP (dashed arrows, Fig. 7c). In the presence of S16, we pro-pose that most of the complexes assemble along the path from IN to N.

The fact that ribosome assembly can proceed by more than onepathway was previously shown by the kinetics of 30S ribosomeassembly, as measured by time-resolved footprinting of the RNAbackbone37 and the rates of protein addition measured by MS35. Theequilibrium and kinetic assembly intermediates of the 5¢ domain arequalitatively similar, even though the time-resolved experiments werecarried out on the entire 16S rRNA and total 30S proteins in 20 mMMgCl2. The most stable interactions in the core of the 5¢ domainformed within 50 ms. By contrast, the folding kinetics of the upperjunction were slower (2 s) and multiphasic37, consistent with slowerconformational changes in this region of the rRNA. One discrepancy isthat the tip of helix 12 (nucleotides 295–297) is rapidly protected in the

RNA

a

c

b

∆G

+S16

Nativecore

H1,H2

N

H18

H18

H18

S16

S16

H4 H12

H12

H12

H17

H15

H3

H3 in

H3 out

Icore

IC

IN

InC

InCNon-nativecore

IN

N

+1°

Figure 7 Model for assembly of the 30S 5¢ domain. (a) Free-energy diagram

illustrating how selective stabilization of a native-like intermediate by S16

depopulates competing Is and results in more cooperative assembly; 11,

S4–S17–S20. (b) Helices that switch conformation when bound by S16

(blue surface) are highlighted in a structure of the 30S ribosome (PDB

2AVY). (c) An ensemble of partly folded RNA (IC) containing different

configurations of core helices are bound and further stabilized by primary

assembly proteins S4, S17 and S20 (pale pink, green and yellow). In the

absence of S16, a native-like (IN) and a non-native intermediate (InC) have

similar free energies and are both formed. S16 (light blue) preferentially

stabilizes IN, resulting in depopulation of InC and a smoother transition to

the native RNP (N). Long-distance communication between the binding site

of S16 in H17 (cyan) and H15 (blue) and helix 3 (purple) stabilizes the

helix 18 pseudoknot (pink) in the 30S decoding center. Equilibrium data

do not distinguish paths from InC to N (dashed arrows).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 4 3

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

16S rRNA, possibly because of the presence of S16 and the high Mg2+

concentration, or the presence of other 16S sequences.

Protein S16 and 30S assemblyThe crucial role of S16 in the assembly of 30S subunits25,30 can beexplained by the conformational switch in helix 3 that comes about bythe preferential stabilization of IN. S16 was previously shown tochange the secondary structure of helix 12 and stabilize the centralpseudoknot27. We hypothesize that tight binding of S16 to helices 15and 17 is communicated via helix 4 to helix 3, and via helix 17 to thestem of helix 18. We note that the kink in helix 17 and the helix 18pseudoknot both become protected in 2–5 mM MgCl2, during thetransition from IN to the native RNP (Fig. 7). Tertiary interactionsbetween helices 3, 18 and the tip of helix 12 are expected to stabilizethe helix 18 pseudoknot. In the 30S subunit, helix 18 is presumablyfurther stabilized by its interactions with protein S12, which is the nextprotein to join the complex in the 30S assembly map38.

By changing the orientation of helix 3, binding of S16 influencesthe connections between the head, body and platform of the smallsubunit that are crucial for forming the decoding site and for finalmaturation of the 30S subunit39,40. First, helix 18 is itself anessential component of the decoding site31,41. Second, in the 30Sribosome, helix 3 stacks with helix 1, which in turn contributes halfof the central pseudoknot (helix 2). Reorientation of helix 3 duringthe transition from IN to the native RNP may help to ensure thathelix 3 is not locked in place until helix 12 is correctly aligned withhelix 6a in the core of the 5¢ domain.

In the cell, ribosome assembly is coupled to transcription42,allowing secondary and tertiary structures to form at the 5¢ end ofthe 16S rRNA before the 3¢ end has been synthesized. When foldingis sequential, local structure is favored over long-distance interac-tions such as helix 3 (ref. 43). Wagner and co-workers havesuggested that the rRNA leader serves as a scaffold for assemblyof the 5¢ domain44,45, by forming metastable interactions that holdthe place of downstream partners until the rest of the 16S rRNAis transcribed. They found that the leader interacts with helix 6 and is cross-linked to nucleotides in helices 3, 4, 5, 7 and 11a,precisely those regions of the 5¢ domain that can form alternativestructures19. The cold-sensitive mutation C23U in helix 1 inhibits5¢ processing of the 16S rRNA46 and produces 30S particlesresembling reconstitution intermediates (RI) formed in vitro atlow temperature6. Thus, an interesting possibility is that theconformational switch we have identified in helix 3 is coupled toprocessing of the pre-rRNA and later steps of 30S subunit assembly.

METHODSPurification of recombinant ribosomal proteins. We overexpressed E. coli

ribosomal proteins S16, S17 and S20 and purified them by ion-exchange

chromatography as described47. Plasmids for overexpression of S16, S17 and

S20 were a gift from G. Culver. We overexpressed S4 as described48 (gift from

D. Draper) and purified it as above. Protein footprints were verified using

DMS footprinting49,50.

Purification of 5¢ domain. We transcribed the 542-nt 5¢ domain RNA in vitro

with T7 RNA polymerase from pRNA1 (ref. 19) using standard methods and

purified it by denaturing 4% PAGE. The RNA concentration was determined by

UV absorption at 260 nm and e260 ¼ 5.4 � 106 M–1 cm–1.

Assembly of ribosomal protein complexes. For hydroxyl radical footprinting

reactions, 12 pmol 5¢ domain RNA was folded 30–40 min at 37 1C in 42 ml

reconstitution buffer (80 mM K+-HEPES, pH 7.6, 330 mM KCl, 20 mM MgCl2,

0.01% (v/v) Nikkol detergent, 6 mM b-mercaptoethanol) before treatment

with Fe(II)-EDTA. Where stated, we varied the MgCl2 concentration from

0 mM to 30 mM. The 5¢ domain RNA was previously shown to fold in less than

5 min19. We prepared protein-RNA complexes by pre-incubating 12 pmol

RNA in reconstitution buffer containing the desired MgCl2 concentration for

15–20 min at 37 1C, before addition of 5-molar equivalents of S16, S17 or S20

and 4-molar equivalents of S4 in a total volume of 56 ml. The RNA was

incubated with the proteins for an additional 45 min at 37 1C.

We determined the optimum protein:RNA ratios by titrating 50 pmol

5¢ domain RNA with individual ribosomal proteins in reconstitution buffer

at RNA: protein ratios ranging from 1:1 to 1:10 in a 42 ml volume. Protein

contacts were saturated with 5-molar equivalents of S16, S17 or S20 and

4-molar equivalents of S4. The protein-RNA co-incubation time required to

achieve saturation was determined by verifying the integrity of RNA-protein

complexes on a native 8% polyacrylamide gel between 0–1.5 h; 40 min was

sufficient for maximal complex formation, and longer incubation produced no

change in the extent of hydroxyl radical protection.

DMS modification of protein complexes. We folded 50 pmol of 5¢ domain

RNA alone, with the individual proteins or all four proteins in 50 ml of

reconstitution buffer as described above, with RNA:protein ratios of 1:0.25,

1:0.5, 1:1, 1:2, 1:3 and 1:5 at 42 1C for 1 h. Samples were cooled on ice for

10 min and then treated with DMS as described50. The RNA was extracted once

with equal volumes of phenol and 1:25 isoamyl:chloroform and precipitated

before primer extension.

Fe(II)-EDTA reactions. Following folding of the 5¢ domain RNA or assembly

of 5¢ domain RNPs, we removed 14 ml containing 3 pmol of 5¢ domain RNA

from each reaction and treated it with Fenton reagents as described51. We

carried out control reactions in parallel with each titration on native E. coli 30S

ribosomal subunits, 5¢ domain RNA in 80 mM K+-HEPES, 80 mM K+-HEPES

plus 330 mM KCl, or reconstitution buffer. 30S subunits (10 pmol) in 14 ml

reconstitution buffer were heat activated for 30 min at 42 1C, placed on ice

for 10 min and treated on ice with 2 ml 10 mM Fe(II)-EDTA for 1 min as

described above. Samples were analyzed by primer extension and AMV reverse

transcriptase as described19.

Data analysis. We compared protected regions of the RNA backbone with the

solvent-accessible surface area for each C4¢ atom in the 5¢ domain of the 16S

rRNA using coordinates from the structure of the E. coli 30S ribosome28 and

the program Calc-surf52. The extent of cleavage at each position in the

5¢ domain RNA was quantified with a Molecular Dynamics Phosphorimager

and normalized to a reference nucleotide whose intensity did not change over

the course of the experiment53. We obtained the fractional saturation (Y) by

normalizing the relative cleavage to the amount of cleavage in 80 mM HEPES

(Y ¼ 0) and buffer plus 20 mM MgCl2 or native 30S subunits (Y ¼ 1),

whichever was greater.

Assuming that the extent of hydroxyl radical cleavage reflects the equili-

brium between an ‘open’ and ‘closed’ state at each nucleotide, we fit the

fractional saturation (Y) of each backbone protection versus Mg2+ concentra-

tion (C) to an isotherm for a cooperative two-state equilibrium,�Y ¼ �Y0 + ðC=CmÞn=½1 + ðC=CmÞn�, in which n is the Hill coefficient and Y0 is

the extent of protection in 330 mM KCl and no MgCl2. Multiphasic transitions

were fit to a four-state model in which the statistical weight of each term was

taken from the equilibrium constant for the open and closed state,

Ki ¼ ðC=Cm;iÞn;i (refs. 22,54):

�Y ¼ ðC=Cm;2Þn;2 + ðC=Cm;N Þn;N

1 + ðC=Cm;1Þn;1 + ðC=Cm;2Þn;2 + ðC=Cm;NÞn;Nð1Þ

The initial state (Icore) and one intermediate are assumed to be completely

exposed, whereas the second intermediate and the final state (N) are fully

protected. In many cases, the data could be described equally well by two

sequential transitions,

�Y ¼ A1ðC=Cm;1Þn;1

1 + ðC=Cm;1Þn;1+

A2ðC=Cm;2Þn;2

1 + ðC=Cm;2Þn;2ð2Þ

in which A1 and A2 are the magnitude of change in protection with each step.

Parameters from equation (2) were used for the purposes of clustering the

ART IC L E S

44 4 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

data. Reproducibility in the fit parameters between experimental trials was

typically ±20% for Cm or ±50% for n, but no greater than than ±50% (Cm)

or ±100% (n).

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ACKNOWLEDGMENTSThe authors thank G. Culver (Univ. Rochester) and D. Draper, R. Moss andT. Adilakshmi (Johns Hopkins Univ. (JHU)) for gifts of plasmids andT. Adilakshmi, A. Cukras, J. Brunelle (JHU) and R. Green (JHU andHoward Hughes Medical Institute) for their help and advice. This work wassupported by a grant from the US National Institutes of Health (GM60819).

AUTHOR CONTRIBUTIONSP.R. performed experiments, analyzed and interpreted data and wrote the paper;S.A.W. conceived the project, interpreted the data and wrote the paper.

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Nierhaus, K.H. The assembly of prokaryotic ribosomes. Biochimie 73, 739–755(1991).

2. Warner, J.R., Vilardell, J. & Sohn, J.H. Economics of ribosome biosynthesis. ColdSpring Harb. Symp. Quant. Biol. 66, 567–574 (2001).

3. Wilson, D.N. & Nierhaus, K.H. The weird and wonderful world of bacterial ribosomeregulation. Crit. Rev. Biochem. Mol. Biol. 42, 187–219 (2007).

4. Kaczanowska, M. & Ryden-Aulin, M. Ribosome biogenesis and the translation processin Escherichia coli. Microbiol. Mol. Biol. Rev. 71, 477–494 (2007).

5. Thirumalai, D. & Woodson, S.A. Kinetics of folding of protein and RNA. Acc. Chem.Res. 29, 433–439 (1996).

6. Traub, P. & Nomura, M. Structure and function of Escherichia coli ribosomes. VI.Mechanism of assembly of 30 s ribosomes studied in vitro. J. Mol. Biol. 40, 391–413(1969).

7. Held, W.A. & Nomura, M. Rate determining step in the reconstitution of Escherichiacoli 30S ribosomal subunits. Biochemistry 12, 3273–3281 (1973).

8. Stern, S., Powers, T., Changchien, L.M. & Noller, H.F. RNA-protein interactions in 30Sribosomal subunits: folding and function of 16S rRNA. Science 244, 783–790(1989).

9. Powers, T., Stern, S., Changchien, L.M. & Noller, H.F. Probing the assembly of the 3¢major domain of 16 S rRNA. Interactions involving ribosomal proteins S2, S3, S10,S13 and S14. J. Mol. Biol. 201, 697–716 (1988).

10. Agalarov, S.C., Sridhar Prasad, G., Funke, P.M., Stout, C.D. & Williamson, J.R.Structure of the S15,S6,S18-rRNA complex: assembly of the 30S ribosome centraldomain. Science 288, 107–113 (2000).

11. Recht, M.I. & Williamson, J.R. RNA tertiary structure and cooperative assembly of alarge ribonucleoprotein complex. J. Mol. Biol. 344, 395–407 (2004).

12. Grondek, J.F. & Culver, G.M. Assembly of the 30S ribosomal subunit: positioningribosomal protein S13 in the S7 assembly branch. RNA 10, 1861–1866 (2004).

13. Wimberly, B.T. et al. Structure of the 30S ribosomal subunit. Nature 407, 327–339(2000).

14. Schluenzen, F. et al. Structure of functionally activated small ribosomal subunit at 3.3A resolution. Cell 102, 615–623 (2000).

15. Weitzmann, C.J., Cunningham, P.R., Nurse, K. & Ofengand, J. Chemical evidence fordomain assembly of the Escherichia coli 30S ribosome. FASEB J. 7, 177–180 (1993).

16. Held, W.A., Ballou, B., Mizushima, S. & Nomura, M. Assembly mapping of 30 Sribosomal proteins from Escherichia coli. Further studies. J. Biol. Chem. 249,3103–3111 (1974).

17. Nowotny, V. & Nierhaus, K.H. Assembly of the 30S subunit from Escherichia coliribosomes occurs via two assembly domains which are initiated by S4 and S7.Biochemistry 27, 7051–7055 (1988).

18. Powers, T., Daubresse, G. & Noller, H.F. Dynamics of in vitro assembly of 16 S rRNAinto 30 S ribosomal subunits. J. Mol. Biol. 232, 362–374 (1993).

19. Adilakshmi, T., Ramaswamy, P. & Woodson, S.A. Protein-independent folding pathwayof the 16S rRNA 5 ¢ domain. J. Mol. Biol. 351, 508–519 (2005).

20. Tullius, T.D. & Greenbaum, J.A. Mapping nucleic acid structure by hydroxyl radicalcleavage. Curr. Opin. Chem. Biol. 9, 127–134 (2005).

21. Rook, M.S., Treiber, D.K. & Williamson, J.R. An optimal Mg2+ concentration for kineticfolding of the tetrahymena ribozyme. Proc. Natl. Acad. Sci. USA 96, 12471–12476(1999).

22. Pan, J., Thirumalai, D. & Woodson, S.A. Magnesium-dependent folding of self-splicingRNA: exploring the link between cooperativity, thermodynamics, and kinetics. Proc.Natl. Acad. Sci. USA 96, 6149–6154 (1999).

23. Uchida, T., He, Q., Ralston, C.Y., Brenowitz, M. & Chance, M.R. Linkage of monovalentand divalent ion binding in the folding of the P4–P6 domain of the Tetrahymenaribozyme. Biochemistry 41, 5799–5806 (2002).

24. Heilman-Miller, S.L., Thirumalai, D. & Woodson, S.A. Role of counterion condensationin folding of the Tetrahymena ribozyme. I. Equilibrium stabilization by cations. J. Mol.Biol. 306, 1157–1166 (2001).

25. Powers, T. & Noller, H.F. Hydroxyl radical footprinting of ribosomal proteins on 16SrRNA. RNA 1, 194–209 (1995).

26. Stern, S., Wilson, R.C. & Noller, H.F. Localization of the binding site for protein S4 on16 S ribosomal RNA by chemical and enzymatic probing and primer extension. J. Mol.Biol. 192, 101–110 (1986).

27. Stern, S., Changchien, L.M., Craven, G.R. & Noller, H.F. Interaction of proteins S16,S17 and S20 with 16 S ribosomal RNA. J. Mol. Biol. 200, 291–299 (1988).

28. Schuwirth, B.S. et al. Structures of the bacterial ribosome at 3.5 A resolution. Science310, 827–834 (2005).

29. Persson, B.C., Bylund, G.O., Berg, D.E. & Wikstrom, P.M. Functional analysis of theffh-trmD region of the Escherichia coli chromosome by using reverse genetics.J. Bacteriol. 177, 5554–5560 (1995).

30. Held, W.A. & Nomura, M. Escherichia coli 30 S ribosomal proteins uniquely requiredfor assembly. J. Biol. Chem. 250, 3179–3184 (1975).

31. Powers, T. & Noller, H.F. A functional pseudoknot in 16S ribosomal RNA. EMBO J 10,2203–2214 (1991).

32. Nagai, K. et al. Structure, function and evolution of the signal recognition particle.EMBO J. 22, 3479–3485 (2003).

33. Doudna, J.A. & Batey, R.T. Structural insights into the signal recognition particle.Annu. Rev. Biochem. 73, 539–557 (2004).

34. Williamson, J.R. Assembly of the 30S ribosomal subunit. Q. Rev. Biophys. 38,397–403 (2005).

35. Talkington, M.W., Siuzdak, G. & Williamson, J.R. An assembly landscape for the 30Sribosomal subunit. Nature 438, 628–632 (2005).

36. Chworos, A. et al. Building programmable jigsaw puzzles with RNA. Science 306,2068–2072 (2004).

37. Adilakshmi, T., Bellur, D.L. & Woodson, S.A. Concurrent nucleation of 16S folding andinduced fit in 30S ribosome assembly. Nature 455, 1268–1272 (2008).

38. Brodersen, D.E., Clemons, W.M. Jr, Carter, A.P., Wimberly, B.T. & Ramakrishnan, V.Crystal structure of the 30 S ribosomal subunit from Thermus thermophilus: structureof the proteins and their interactions with 16 S RNA.. J. Mol. Biol 316, 725–768(2002.).

39. Brink, M.F., Verbeet, M.P. & de Boer, H.A. Formation of the central pseudoknot in16S rRNA is essential for initiation of translation. EMBO J. 12, 3987–3996(1993).

40. Poot, R.A., van den Worm, S.H., Pleij, C.W. & van Duin, J. Base complementarity inhelix 2 of the central pseudoknot in 16S rRNA is essential for ribosome functioning.Nucleic Acids Res. 26, 549–553 (1998).

41. Ogle, J.M. et al. Recognition of cognate transfer RNA by the 30S ribosomal subunit.Science 292, 897–902 (2001).

42. Lewicki, B.T., Margus, T., Remme, J. & Nierhaus, K.H. Coupling of rRNA transcriptionand ribosomal assembly in vivo. Formation of active ribosomal subunits in Escherichiacoli requires transcription of rRNA genes by host RNA polymerase which cannot bereplaced by bacteriophage T7 RNA polymerase. J. Mol. Biol. 231, 581–593(1993).

43. Heilman-Miller, S.L. & Woodson, S.A. Effect of transcription on folding of theTetrahymena ribozyme. RNA 9, 722–733 (2003).

44. Pardon, B. & Wagner, R. The Escherichia coli ribosomal RNA leader nut regioninteracts specifically with mature 16S RNA. Nucleic Acids Res. 23, 932–941(1995).

45. Besancon, W. & Wagner, R. Characterization of transient RNA-RNA interactionsimportant for the facilitated structure formation of bacterial ribosomal 16S RNA.Nucleic Acids Res. 27, 4353–4362 (1999).

46. Dammel, C.S. & Noller, H.F. A cold-sensitive mutation in 16S rRNA provides evidencefor helical switching in ribosome assembly. Genes Dev. 7, 660–670 (1993).

47. Culver, G.M. & Noller, H.F. Efficient reconstitution of functional Escherichia coli 30Sribosomal subunits from a complete set of recombinant small subunit ribosomalproteins. RNA 5, 832–843 (1999).

48. Baker, A.M. & Draper, D.E. Messenger RNA recognition by fragments of ribosomalprotein S4. J. Biol. Chem. 270, 22939–22945 (1995).

49. Moazed, D., Stern, S. & Noller, H.F. Rapid chemical probing of conformation in 16 Sribosomal RNA and 30 S ribosomal subunits using primer extension. J. Mol. Biol. 187,399–416 (1986).

50. Stern, S., Moazed, D. & Noller, H.F. Structural analysis of RNA using chemical andenzymatic probing monitored by primer extension. Methods Enzymol. 164, 481–489(1988).

51. Latham, J.A. & Cech, T.R. Defining the inside and outside of a catalytic RNA molecule.Science 245, 276–282 (1989).

52. Gerstein, M. A resolution-sensitive procedure for comparing protein surfaces and itsapplication to the comparison of antigen-combining sites. Acta Crystallogr. A 48,271–276 (1992).

53. Hsieh, M. & Brenowitz, M. Quantitative kinetics footprinting of protein-DNA associa-tion reactions. Methods Enzymol. 274, 478–492 (1996).

54. Fang, X., Pan, T. & Sosnick, T.R. A thermodynamic framework and cooperativity in thetertiary folding of a Mg2+-dependent ribozyme. Biochemistry 38, 16840–16846(1999).

55. Gutell, R.R. Comparative sequence analysis and the structure of 16S and 23S rRNA.in. Ribosomal RNA: Structure, Evolution, Processing, and Function in Protein Bio-synthesis. (eds. Zimmerman, R.A. & Dahlberg, A.E.) 111–128 (CRC, Boca Raton,Florida, USA, 1996).

56. Glotz, C. & Brimacombe, R. An experimentally-derived model for the secondarystructure of the 16S ribosomal RNA from Escherichia coli. Nucleic Acids Res. 8,2377–2395 (1980).

ART IC L E S

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 4 5

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

CK2a phosphorylates BMAL1 toregulate the mammalian clockTeruya Tamaru1, Jun Hirayama2,3, Yasushi Isojima4,Katsuya Nagai4, Shigemi Norioka5, Ken Takamatsu1 &Paolo Sassone-Corsi2

Clock proteins govern circadian physiology and their functionis regulated by various mechanisms. Here we demonstratethat Casein kinase (CK)-2a phosphorylates the core circadianregulator BMAL1. Gene silencing of CK2a or mutation of thehighly conserved CK2-phosphorylation site in BMAL1, Ser90,result in impaired nuclear BMAL1 accumulation and disruptionof clock function. Notably, phosphorylation at Ser90 follows arhythmic pattern. These findings reveal that CK2 is an essentialregulator of the mammalian circadian system.

The circadian clock orchestrates intrinsic timing in most organismsand controls a large variety of physiological and metabolic pro-grams1. The molecular core of the circadian clock is constituted bymultiple gene products that operate in transcriptional-translationalfeedback loops1,2. The BMAL1–CLOCK heterodimer is central tothe clock mechanism as it drives and maintains circadian oscilla-tions. Several studies indicate that clock proteins are importanttargets of post-translational modifications3–7, and their rhythmicphosphorylation seems to be a crucial step for regulated func-tion2,8,9. Indeed, BMAL1 phosphorylation is rhythmic in serumshock–synchronized fibroblasts10. Furthermore, drastic circadianphenotypes in the fly are caused by mutations in kinases such asDoubletime (DBT), the fly ortholog of CKIe, Shaggy (SGG), thefly ortholog of Glycogen synthase kinase 3 (GSK3), and CK2.Drosophila melanogaster CK2 specifically regulates the nuclearentry of Period (PER), thereby controlling the fly clock11–13. Inmammals, the tau mutant phenotype, characterized by impaired

circadian rhythmicity, is due to a mutation in the gene encodingthe clock-regulating kinase CKIe14.

Microsequencing of purified p45PFK, a kinase previously implicatedin circadian control in mammals15, demonstrates that it correspondsto CK2a (Supplementary Fig. 1 online). Notably, purified p45PFK

phosphorylates BMAL1 and CLOCK in vitro15,16 (SupplementaryFig. 2 online).

The CK2 holoenzyme is constituted by two copies of the catalytic(CK2a) and regulatory (CK2b) subunits (a2b2). The CK2amonomer also exists as an active form in vivo. To assess the role ofCK2b in modulating CK2a-mediated BMAL1 phosphorylation, weco-incubated the subunits in kinase assays. CK2a alone phosphory-lated GST-BMAL1, whereas CK2b inhibited BMAL1 phosphorylationin a dose-dependent manner (Fig. 1; see Supplementary Methodsonline). These results demonstrate that CK2a monomer phosphor-ylates BMAL1.

To establish the physiological role of CK2a-mediated BMAL1phosphorylation, we used microRNA interference (miR)-mediatedsilencing in dexamethasone (Dex)-synchronized NIH-3T3 mousefibroblasts. A miR-CK2a silencing vector specifically reducedCK2a levels to less than the half that of the control (P ¼ 0.002)(Fig. 2a), resulting in markedly impaired circadian Per2 oscilla-tion, as monitored by real-time Per2 promoter–driven luciferasebioluminescence4 (Fig. 2b). In fact, after a first peak with sig-nificantly reduced amplitude, Per2 oscillation was almost abolishedin miR-CK2a–transfected cells (Fig. 2b). These results wereconfirmed by using an alternative target sequence (Supplemen-tary Fig. 3 online). Thus, CK2a is essential for mammaliancircadian rhythmicity.

Notably, silencing of CK2a leads to substantial cytoplasmic BMAL1retention 24 h after Dex treatment (Fig. 2c), a time when BMAL1 isnormally mostly nuclear10. As the function of the CLOCK–BMAL1complex is intimately dependent on the nuclear translocationof BMAL1 (ref. 17), we speculate that CK2a exerts its control onthe mammalian circadian clock by dictating BMAL1 intracellular

CBBa bCK2αβ CK2αβ

(–)

(kDa)GST-BMAL1

1169770

45

31

GST

CK2α

pCK2β

1.25

15′60′

1

0.75

Inco

rpor

ated

Pi/B

MA

L1(p

mol

/pm

ol)

0.5

0.25

00 0.2 0.5 1 2

+ [γ-32P]-ATPCK2αβCK2α

CK2β/CK2α (pmol/pmol)

pGST-BMAL1

Autoradiography

1 2 1 2 1 2 1 2

Figure 1 CK2a phosphorylates BMAL1 in vitro. (a) Mixtures of CK2a(50 ng; visible in Coomassie Brilliant Blue staining) and CK2b (10 ng;visible in autoradiography after phosphorylation) subunits were used in an

in vitro kinase assay, with or without ATP, using glutathione S-transferase

(GST) (lane 1) or GST-BMAL1 (lane 2) as substrates. (b) CK2a (4 pmol)

and CK2b (0–8 pmol) were used in kinase assays with GST-BMAL1 as the

substrate (500 ng; optimal ratio to CK2a). GST-BMAL1 kinase activities

(average values) are plotted at each CK2b/CK2a ratio and normalized

against activity in the absence of CK2b with error bars (± s.d.). Photographs

are representative of duplicate experiments.

Received 27 November 2008; accepted 18 February 2009; published online 29 March 2009; doi:10.1038/nsmb.1578

1Department of Physiology, Toho University School of Medicine, Tokyo, Japan. 2Department of Pharmacology, School of Medicine, University of California, Irvine,California, USA. 3Medical Top Track Program, Medical Research Institute, Tokyo Medical and Dental University, Bunkyo-ku, Tokyo, Japan. 4Division of ProteinMetabolism, Institute for Protein Research, Osaka University, Suita, Osaka, Japan. 5Division of Protein Chemistry, Institute for Protein Research, Osaka University,Suita, Osaka, Japan. Correspondence should be addressed to P.S.-C. (psc@uci.edu).

44 6 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

BR I E F COMMUNICAT IONS

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

localization. This is reminiscent of the role described for theD. melanogaster CK2 in controlling PER nuclear entry13.

We searched the BMAL1 primary sequence for CK2 consensusphosphoacceptor sites (S/T-X-X-D/E). The serine residue at position90 is highly conserved among all vertebrate BMAL1s (SupplementaryFig. 4b online), and a MALDI-TOF/MS analysis indicated thatit is indeed a potential CK2-phosphorylation site (SupplementaryFig. 4a). Notably, Ser90 is not conserved in D. melanogaster CYCLE,the fly counterpart of BMAL1, consistent with the notion that CK2acannot phosphorylate CYCLE13.

To establish the role of BMAL1 Ser90 phosphorylation in circadianfunction, we performed rescue experiments by expressing wild-typeBMAL1 (BMAL1-WT), BMAL1-S90A mutant or GFP in mouseembryonic fibroblasts (MEFs) derived from Bmal1-null mice, whichhave a dysfunctional circadian clock4,6,7 (Fig. 2d). Per2 expression wasmonitored in Dex-synchronized MEFs by a real-time reporter assay.BMAL1-WT rescued circadian Per2 expression in Dex-synchronizedMEFs (Supplementary Fig. 5 online), whereas the BMAL1-S90Amutant was unable to do so (Fig. 2f), despite its having normalDNA binding activity (data not shown). These results indicatethat BMAL1 phosphorylation at Ser90 is essential for circadiangene expression.

Ectopically expressed BMAL1-WT accumulated in the nucleus 24 hafter Dex treatment (Fig. 2f), paralleling the behavior of nativeBMAL1 in fibroblasts10. In contrast, BMAL1-S90A remained mostlycytoplasmic (Fig. 2f). Consistent with previous reports indicating amatching localization pattern between BMAL1 and CLOCK17,BMAL1-S90A suppressed nuclear CLOCK accumulation, possiblybecause of its decreased interaction potential with CLOCK (Supple-mentary Fig. 6 online). These data are consistent with the changein BMAL1 localization upon CK2a silencing (Fig. 2c) and stress

the notion that circadian function requires phosphorylation ofBMAL1 and its nuclear accumulation.

To assess the temporal pattern of Ser90 phosphorylation in vivo,we raised an antibody specific for Ser90-phosphorylated BMAL1.Specificity was validated because this antibody (P-BMAL1-S90)detects Myc-BMAL1 but not the Myc-BMAL1-S90A mutant orMyc-GFP (Supplementary Fig. 7 online). Using the P-BMAL1-S90 antibody, we examined whether silencing of CK2a indeedimpaired BMAL1 phosphorylation at Ser90. The P-BMAL1-S90antibody detected a band corresponding to the highly phosphory-lated protein (apparent migration pattern corresponding to about85–90 kDa) in the concentrated BMAL1 immunoprecipitated fromthe control vector–transfected NIH-3T3 cells at 24 h after Dextreatment (Fig. 3a). The targeting vectors for miR-CK2a substan-tially decreased the P-BMAL1-S90 signal (Fig. 3a). Notably, knock-down of CK2a strongly affects the total phosphorylation state ofBMAL1, suggesting that CK2a may influence or modulate multiplephosphorylation events on BMAL1 by other kinases, includingERKs and CKIe18,19.

Next, we examined whether P-BMAL1-S90 levels oscillated in vivo.In Dex-synchronized NIH-3T3 cells, Ser90 phosphorylation showed arhythmic pattern with peaks at 20–24 h and 44–48h, paralleling PER1level oscillation (Fig. 3b). At peaking times, the P-BMAL1-S90 signalgradually shifted to a higher molecular weight (B90 kDa), corre-sponding to the hyperphosphorylated forms. The timing of BMAL1-S90 phosphorylation fits with the cyclic nuclear entry of BMAL1 inDex-synchronized cells (data not shown) and serum shock–synchro-nized cells10. The levels of CK2 subunits (a¢ is a subtype of CK2catalytic subunit) are basically constant during the circadian cycle(Fig. 3b); thus, they are unlikely to determine P-BMAL1-S90 oscilla-tion. Finally, on the basis of the BMAL1 phosphorylation pattern

Figure 2 CK2a and BMAL1-Ser90 regulate

nuclear accumulation and clock function.

(a–c) CK2a silencing or depletion affects BMAL1

nuclear localization and clock function. NIH-3T3

cells were transfected with miR targeting vectors

for mouse CK2a and the validated negative

control (control) vector, and were then Dex

synchronized. (a) Immunoblot analysis for CK2aand actin in cell lysates, with one representative

experiment shown. Normalized average CK2alevels from triplicate experiments are shown

below with error bars (± s.d.). (b) Circadian Per2

expression was monitored using a real-time

bioluminescence assay. Normalized average

values from multiple experiments (n ¼ 5) areplotted with error bars (± s.d. of the first and

second peaks). (c) Localization of BMAL1,

visualized by immunofluorescence (red) 24 h

after Dex treatment. Nuclei were stained with

DAPI (blue), and miRNAi-transfected, GFP-

positive cells appear green. Photographs

are representative of triplicate experiments.

Percentages of cells whose BMAL1 is

predominantly nuclear among the GFP-positive

cells from triplicate experiments are plotted as

means ± s.e.m. (***, P o 0.001). (d–f) Mutation

of BMAL1-Ser90 affects nuclear localization and

clock function. (d) Lysates from Bmal1�/� MEFs stably expressing Myc-BMAL1-WT, Myc-BMAL1-S90A and Myc-GFP were analyzed with anti-Myc antibody.

(e) Circadian Per2 expression was not restored by BMAL1-S90A, as monitored a real-time bioluminescence assay. Normalized average values from multiple

experiments (n ¼ 5) are plotted with error bars (± s.d. of the first and second peaks). (f) MEFs 24 h after Dex treatment were stained with anti-BMAL1

(green) antibody. Nuclei were stained with DAPI (blue). Photographs are representative of mutiple (n ¼ 4) experiments. Values of nuclear BMAL1-dominant

cells (%) in MEFs from multiple (n ¼ 4) experiments are plotted as means ± s.e.m. (***, P o 0.001).

TF (h) 24a

d e f

b c24 48 48 1.2

Control

miR-CK2α

Control

BMAL1

MergeDAPIGFP

WT

80***

***

604020

100

75

50

25

0

Nuc

lear

BM

AL1

-do

min

ant c

ells

(%

)N

ucle

ar B

MA

L1-

dom

inan

t cel

ls (

%)

S90AGFP

miR-CK2α

Contro

l

Myc

-GFP

Myc

-BM

AL1-W

T

Myc

-BM

AL1-S

90A

miR

-CK2α

WT

S90A

1.0

0.8

0.6

Rel

ativ

e P

er2-

Luc

activ

ity

0.4

0.2

0

1.2

1.0

0.8

0.6

Rel

ativ

e P

er2-

Luc

activ

ity

0.4

0.2

0

0 12 24 36 48Time post Dex treatment (h)

60 72

0 12 24 36 48Time post Dex treatment (h)

60 72

BMAL1

Myc-BMAL1-WT

Myc-BMAL1-S90A

Merge (DAPI)

84 96

0+ +

+ +

0 24 24Dex (h)Control

miR-CK2α

CK2α

Actin

1.0

0.5

Nor

mal

ized

CK

2α c

onte

nt(m

iR-C

Ka/

cont

rol)

0

BR I E F COMMUNICAT IONS

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 4 7

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

(Fig. 3a,b), it would seem that CK2a could have a triggering role inmultiple phosphorylation events.

Our findings demonstrate that CK2a is a BMAL1 kinase that has anessential role in the mammalian clock by regulating BMAL1 nuclearentry. The single Ser90 phosphorylation site is essential forcircadian rhythmicity and its mutation causes a phenotype similarto that caused by depletion of CK2a (compare Fig. 2b,c and 2e,f).These findings suggest that CK2a has a pivotal role in the mammaliancircadian clock, by controlling BMAL1 intracellular distribution.Although it is likely that CK2 influences circadian physiology atmultiple levels because it phosphorylates a large array of cellularproteins20, the remarkable specificity observed here on BMAL1-Ser90 provides a lead to further investigations. Finally, as BMAL1 isSUMOylated and acetylated4,7, the regulatory pathway presented heremay participate in the interplay between phosphorylation and otherpost-translational modifications.

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ACKNOWLEDGMENTSWe thank M. Okada and T. Takao for MS analysis, and E.G. Krebs, L. Dongxia,J. S. Takahashi, C.A. Bradfield, S.M. Reppert, D.R. Weaver, M. Ikeda andC. Nishio for reagents, discussions and help. This work was supported by theHuman Frontiers in Science Program Organization and the Japanese Ministry ofEducation, Culture, Sports, Science and Technology (MEXT; T.T. and K.T.) and

by the Cancer Research Coordinating Committee of the University of Californiaand from the US National Institutes of Health (P.S.-C.).

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Reppert, S.M. & Weaver, D.R. Nature 418, 935–941 (2002).2. Harms, E., Kivimae, S., Young, M.W. & Saez, L. J. Biol. Rhythms 19, 361–373

(2004).3. Doi, M., Hirayama, J. & Sassone-Corsi, P. Cell 125, 497–508 (2006).4. Hirayama, J. et al. Nature 450, 1086–1090 (2007).5. Gekakis, N. et al. Science 280, 1564–1569 (1998).6. Bunger, M.K. et al. Cell 103, 1009–1017 (2000).7. Cardone, L. et al. Science 309, 1390–1394 (2005).8. Lee, C., Etchegaray, J.P., Cagampang, F.R.A., Loudon, S.I. & Reppert, S.M. Cell 107,

855–867 (2001).9. Tomita, J., Nakajima, M., Kondo, T. & Iwasaki, H. Science 307, 251–254 (2005).10. Tamaru, T. et al. Genes Cells 8, 973–983 (2003).11. Price, J.L. et al. Cell 94, 83–95 (1998).12. Martinek, S., Inonog, S., Manoukian, A.S. & Young, M.W. Cell 105, 769–779 (2001).13. Lin, J.M. et al. Nature. 420, 816–20 (2002).14. Lowrey, P.L. et al. Science 288, 483–492 (2000).15. Tamaru, T., Okada, M., Nagai, K., Nakagawa, H. & Takamatsu, K. J. Neurochem. 72,

2191–2197 (1999).16. Tamaru, T. & Okada, M. Eur. J. Biochem. 238, 152–159 (1996).17. Kondratov, R.V. et al. Genes Dev. 17, 1921–1932 (2003).18. Sanada, K., Okano, T. & Fukada, Y. J. Biol. Chem. 277, 267–271 (2002).19. Eide, E.J., Vielhaber, E.L., Hinz, W.A. & Virshup, D.M. J. Biol. Chem. 277,

17248–17254 (2002).20. Meggio, F. & Pinna, L.A. FASEB J. 17, 349–368 (2003).

a b

CK2α

Actin

WB:

WB:

miR

BMAL1

P-BMAL1-S90

Time after Dex treatment (h)

Actin

Lysate

P-BMAL1-S90

BMAL1

CK2α

CK2α′

CK2β

PER1

Time after Dex treatment (h)

P-BMAL1-S90

Nor

mal

ized

con

tent

0

1.01.2

0.80.60.40.2

0 2 4 8 12 16 20 24 28 32 36 40 44 48 52

0 4 8 12 16 20 24 28 32 36 40 44 4852

PER1

IP: BMAL1

IP: BMAL1

Lysate

C CK2α

Figure 3 Circadian phosphorylation of BMAL1-Ser90 by CK2a in vivo. (a) NIH-3T3 cells expressing miR-CK2a and a control vector 24 h after Dex treatment

were subjected to BMAL1-immunoprecipitation (IP) followed by immunoblot analysis (WB) for BMAL1, P-BMAL1-S90, CK2a and actin (Lysate). Arrows and

bars designate BMAL1 and P-BMAL1-S90. (b) BMAL1 immunoprecipitates and lysates of NIH-3T3 cells at each time point after Dex treatment were used in

immunoblot analysis for P-BMAL1-S90, BMAL1, PER1, CK2 subunits and actin. Arrows and bars designate BMAL1 and P-BMAL1-S90. Normalized average

values for P-BMAL1-S90 and PER1 from triplicate experiments are plotted in the graph with error bars (± s.d.). All the photographs are representative of

triplicate experiments.

BR I E F COMMUNICAT IONS

44 8 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

Distinct transcriptional outputsassociated with mono- anddimethylated histone H3arginine 2Antonis Kirmizis1, Helena Santos-Rosa1, Christopher J Penkett2,Michael A Singer3, Roland D Green3 & Tony Kouzarides1

Dimethylation of histone H3 Arg2 (H3R2me2) maintainstranscriptional silencing by inhibiting Set1 mediatedtrimethylation of H3K4. Here we demonstrate that Arg2 is alsomonomethylated (H3R2me1) in yeast but that its functionalcharacteristics are distinct from H3R2me2: (i) H3R2me1 doesnot inhibit histone H3 Lys4 (H3K4) methylation; (ii) it ispresent throughout the coding region of genes; and (iii) itcorrelates with active transcription. Collectively, these resultsindicate that different H3R2 methylation states have definedroles in gene expression.

Covalent post-translational modifications of histones have a funda-mental role in chromatin structure and function1. Histone argininemethylation is one such modification that has been linked totranscriptional regulation2. Arginine residues are methylated onthe terminal guanidino nitrogens and can exist in three differentmethylation states: monomethylated (me1),symmetrically dimethylated (me2s) or asym-metrically dimethylated (me2a)3. Studies inmammalian and yeast cells have demon-strated that histone arginine methylationcan influence both gene activation andrepression4–7. For example, we have recentlyshown that asymmetric dimethylation of his-tone H3 Arg2 (H3R2me2a) in yeast contri-butes to transcriptional repression byinhibiting trimethylation of H3K4. Specifi-cally, H3R2me2a inhibits H3K4me3 by block-ing the PHD domain of the Set1 complexcomponent Spp1 from binding to methylatedH3K4 and, therefore, abrogating H3K4 tri-methylation by the Set1 methyltransferase8.Similarly, in mammals, H3R2me2a, which iscatalyzed by PRMT6, blocks binding of theWD40 domain of WDR5 to histone H3, thusinhibitng H3K4 trimethylation mediated by

MLL1 (refs. 9–11). Overall, these studies described a function fordimethylarginines in transcriptional regulation. However, the role ofmonomethylarginine still remains unexplored, even though thismodification is detected in vivo on mammalian histones6.

Using an antibody against the monomethylated form of H3R2, weshow that this modification (H3R2me1) occurs on yeast histone H3(Supplementary Fig. 1a online). The antibody recognizes only Arg2on histone H3, as mutation of arginine to alanine (H3R2A) orglutamine (H3R2Q) abolishes the signal (Supplementary Fig. 1b).Furthermore, we demonstrate the specificity of this antibody towardthe monomethylated version of H3R2 by dot blot and peptide-competition analyses (Supplementary Fig. 1c,d).

To determine whether H3R2me1 functions in a similar manner toH3R2me2a, we first examined the effect of this methyl H3R2 state onthe recruitment of Spp1 to H3K4 in vitro and in vivo. In pull-downassays, H3R2me1 did not block the interaction of the Spp1 PHDfinger with dimethylated H3K4 peptides (Fig. 1a, lane 4). Consistently,chromatin immunoprecipitation (ChIP) assays showed that in vivoSpp1 occupancy on the 5¢ end of genes can coincide with H3R2me1enrichment but not with the presence of H3R2me2a (Fig. 1b). Wenext sought to determine the effect of H3R2me1 on the ability of Set1to trimethylate H3K4. In agreement with the above results, thepurified Set1 complex was able to methylate an H3R2me1 peptidealmost as well as an unmodified peptide (Fig. 1c). The Set1 activitywas shown to be specific, because an H3K4me3 peptide was notmethylated, and the presence of H3R2me2a reduced this activity(Fig. 1c, lanes 3 and 4). These results show that H3R2me1 does not

Spp1 binding on peptides Spp1 binding in cells

Set1p activity on peptides

Autoradiogram

Coomassie

MYC-Spp1H3R2me1H3R2me2a

YAL0

35W

YLR34

2W

YLR35

5C

Gene name

YAL0

18C

Fol

d en

richm

ent (

Log2

)

1 2 3 4

3

2

1

0

GST-Spp1PHD

GST

Coomassie

Inpu

tH3K

4me2

+ R

2me1

H3K4m

e3

H3R2m

e2a

H3R2m

e1

H3 1–21H3K

4me2

+ R

2me2

a

H3K4m

e2

H3 1–21

1 2 3 4

a b c

Figure 1 H3R2me1 does not block activity of the Set1 complex toward H3K4. (a) Pull-down assays

using synthesized peptides and recombinant glutathione S-transferase (GST)-tagged Spp1PHD or

GST only as a negative control. Equal loading of peptides is monitored by Coomassie staining. (b) ChIP

analysis of logarithmically growing yeast cells using antibodies toward Myc-tagged Spp1, H3R2me1

and H3R2m2a. Error bars represent s.e.m. for duplicate experiments. (c) In vitro methyltransferase

assays using purified Set1 complex and synthesized peptides. Equal amounts of peptides were

used in the methyltransferase reactions, as shown by Coomassie staining.

Received 18 March 2008; accepted 29 January 2009; published online 8 March 2009; doi:10.1038/nsmb.1569

1Gurdon Institute and Department of Pathology, Cambridge CB2 1QN, UK. 2EMBL-European Bioinformatics Institute, Wellcome Trust Genome Campus, Hinxton,Cambridge CB10 1SD, UK. 3NimbleGen Systems Inc., Madison, WI 53711, USA. Correspondence should be addressed to T.K. (t.kouzarides@gurdon.cam.ac.uk).

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 4 9

BR I E F COMMUNICAT IONS

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

abrogate H3K4 trimethylation (unlike H3R2me2a) and suggestthat H3R2 monomethylation has a distinct function from asym-metric dimethylation.

Having determined that H3R2me1 has functional characteristicsdistinct from H3R2me2a, we attempted to determine the role ofH3R2me1 in transcription. We used a high-resolution, genome-wideChIP-chip (ChIP combined with microarray) analysis to determinethe location of H3R2me1 occupancy and its relationship to geneexpression (Supplementary Methods online). We found thatH3R2me1 was localized mainly within tran-scriptional units and was present in 85% ofyeast genes (Supplementary Fig. 2 online).To determine the relationship of H3R2me1 togene expression, we first divided 5,065 genesinto five groups according to their transcrip-tional rate, as previously determined12, andthen examined the average enrichment ofH3R2me1 for each gene group (Fig. 2a).Average gene profiles of H3R2me1 indicatedthat this modification is present evenlythroughout the entire coding region of tran-scriptionally active genes. The enrichment ofH3R2me1 correlated with levels of transcrip-tional activity, because the most active geneswere the most enriched in this modification(Fig. 2a, left). In contrast, analysis of previ-ously reported data of H3R2me2a showedthat this mark is present on the 3¢ end ofgenes and correlates with transcriptional

repression8 (Fig. 2a, right). Additionally,H3R2me2a covers inactive genes entirely(Fig. 2b, left).

We next divided all genes into threedifferentially expressed categories (inactive,moderately active and active) to investigatethe relationship of H3R2me1 with otherhistone methyl marks. H3R2me1 overlaps tosome extend with the active lysine methylmarks H3K4me3, H3K36me3 and H3K79me3on representative moderately transcribed(YAL023C) and active (YLR390W) genes(Fig. 2b). Unlike H3R2me2a, the mono-methylated form of this residue overlapswith H3K4me3 at the 5¢end of moderatelyactive genes (Fig. 2b, middle). Thus, takentogether, these results indicate that H3R2me1coincides with all other active modificationson yeast genes, and its correlation to tran-scription is opposite to that of H3R2me2a.

The above analyses suggested thatH3R2me1 and H3R2me2a might have oppo-site roles in transcription. Therefore, we nextasked whether H3R2me1 and H3R2me2a aredynamically exchanged on nucleosomes uponinduction of gene expression. To test this, weused the sporulation pathway as a modelsystem of inducible gene activity. ChIP ana-lysis of cells grown in rich media (repressedcondition) showed high enrichment ofH3R2me2a at sporulation genes (Hop1 andSpr3), as expected, whereas H3R2me1 was not

detected at all on the same nucleosomes (Fig. 3a, time 0 h). Mostnotably, shifting the cells in sporulation media, which inducedactivation of these genes (Fig. 3a, left), completely reversed the levelsof the two modifications at those same nucleosomes. H3R2me1 wasrobustly enriched, but there were no traces of H3R2me2a (Fig. 3a,time 18 h). These results confirm that the monomethylated state ofH3R2 is associated with transcriptional activation.

As methylation at H3R2 was associated with sporulation genes, wenext sought to determine whether H3R2 is necessary for sporulation

H3R2me1 H3R2me2a

mRNA per hour Least active

Most active

1–44–88–16>16

Fol

d en

richm

ent

1.8

1.6

1.4

1.2

1.0

0.8

0.6

Fol

d en

richm

ent

1.8

1.6

1.4

1.2

1.0

0.8

0.6

Average gene

Inactive

Rel

ativ

e en

richm

ent t

o H

3 (L

og2)

YAL018C YAL023C YLR390W

4

0

4

04

04

0

4

0

4

0

4

04

04

0

4

0

4

0

4

04

04

0

4

0

Moderately active Active

H3R2me1

H3R2me2a

H3K4me3

H3K36me3

H3K79me3

Average gene

mRNA per hourmRNA per hourmRNA per hourmRNA per hour

<1a

b

Figure 2 H3R2me1 associates with transcriptional activation. (a) Composite profiles of ChIP-chip

experiments. (b) ChIP-chip analysis compares the distribution of various histone methylation marks

across three differentially expressed genes. The H3R2me2a and H3K4me3 data have been described8.

12 0.05

0.04

0.03

0.02

0.01

0

0.05

0.04

0.03

0.02

0.01

00 18Time (h)

0 18Time (h)

0.05

0.04

0.03

0.02

0.01

00 18Time (h)

0.05

0.04

0.03

0.02

0.01

00 18Time (h)

0 18Time (h)

H3WT = 11%H3R2A = 1%

H3WT = 9%H3R2A = 0%

Meiosis II

HOP1

SPR3

Fol

d in

duct

ion

Rel

ativ

e en

richm

ent t

o H

3

Gene expression H3R2me1 ChlP H3R2me2a ChlP Meiosis l

9

6

3

0

12

0 18Time (h)

Fol

d in

duct

ion

9

6

3

0

a b

Figure 3 H3R2 is necessary for sporulation. (a) Gene expression analysis and ChIP experiments during

induction of sporulation. Gene expression levels were normalized to a gene, HSD1, whose transcription

remains unchanged before and after sporulation. Error bars represent s.e.m. for duplicate experiments.

(b) Sporulation assays of H3WT and H3R2A strains.

BR I E F COMMUNICAT IONS

45 0 VOLUME 16 NUMBER 4 APRIL 2009 NATURE STRUCTURAL & MOLECULAR BIOLOGY

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

in yeast. We grew cells that expressed either wild-type histone H3(H3WT) or the mutant H3R2A in sporulation media for 7 d and then,using microscopy, we counted the number of cells that had undergonemeiosis I or meiosis II. Mutation of H3R2 to alanine resulted in asevere defect of sporulation, as only 1% of cells managed to undergomeiotic nuclear divisions, as opposed to 20% of H3WT cells (Fig. 3b).This result is consistent with the dynamic regulation of H3R2methylation on sporulation genes and suggests that H3R2 has animportant role in the early stages of the sporulation process in yeast.

In summary, this report unveils for the first time a role for amonomethylarginine state in transcription. The results presented hereprovide evidence that H3R2me1 is a methylation state that occursin vivo on yeast nucleosomes. The presence of H3R2me1 correlateswith transcriptional activity, opposite to the relationship ofH3R2me2a with gene expression. Although both H3R2 methylationstates are enriched within the coding region of genes, their distributionis also different: monomethylation is enriched throughout thecoding region of active genes, whereas dimethylation is enrichedthroughout inactive genes and toward the 3¢ end of active genes.The distinct distribution patterns of mono- and dimethylation marksare a strong indicator that these two modifications are associated withdifferent functions.

The functional difference between these H3R2 methylation states isprobably conserved in higher eukaryotes. Recent findings show thatPRMT6, the predominant H3R2 methyltransferase, catalyzes prefer-entially asymmetric dimethylation, implying the existence of a distinctenzyme that carries out monomethylation10,13. The need for twoseparate enzymes to catalyze these modifications suggests that,possibly, these two methylation states function differently. Theidentity of an enzyme that would catalyze exclusively H3R2me1 inmammals or yeast remains elusive. Combinatorial deletions of thethree putative yeast arginine methyltransferases (Rmt1, Rmt2 andHsl7) and individual deletions of 35 other yeast methyltransferases donot affect the levels of this modification (data not shown andSupplementary Table 1 online).

The precise function of H3R2me1 in the process of transcriptionalactivation remains to be resolved. There are at least two distinctmechanisms by which H3R2me1 can function in gene expression. Inone model, the monomethyled state is a ‘passive’ mark on chromatinthat is used to identify actively transcribed regions that need to be

silenced. In this model, the monomethylation mark is deposited onactive genes to allow subsequent dimethylation and consequentrepression of transcription. In a second model, monomethylationdictates a function (such as the recruitment of a protein) that isnecessary for genes to become or remain active. Such a model wouldbe analogous to the recruitment of chromatin effectors by specificlysine methylation states14,15. Future studies will aim to decipher themolecular mechanism used by H3R2me1 during gene expression.

Accession codes. Gene Expression Omnibus: Microarray data setshave been deposited under accession number GSE14453.

Note: Supplementary information is available on the Nature Structural & MolecularBiology website.

ACKNOWLEDGMENTSWe thank members of the T.K. laboratory for helpful discussions and M. Gilchristfor help with depositing genomic data. This work was supported by postdoctoralfellowship grants to A.K. from the European Molecular Biology Organization(EMBO) and Marie Curie. The T.K. laboratory is funded by grants from CancerResearch UK (CRUK) and the 6th Research Framework Program of the EuropeanUnion (Epitron and Heroic).

COMPETING INTERESTS STATEMENTThe authors declare competing financial interests: details accompany the full-textHTML version of the paper at http://www.nature.com/nsmb/

Published online at http://www.nature.com/nsmb/

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions/

1. Kouzarides, T. Cell 128, 693–705 (2007).2. Wysocka, J., Allis, C.D. & Coonrod, S. Front. Biosci. 11, 344–355 (2006).3. Turner, B.M. Nat. Struct. Mol. Biol. 12, 110–112 (2005).4. Bauer, U.M., Daujat, S., Nielsen, S.J., Nightingale, K. & Kouzarides, T. EMBO Rep. 3,

39–44 (2002).5. Ma, H. et al. Curr. Biol. 11, 1981–1985 (2001).6. Strahl, B.D. et al. Curr. Biol. 11, 996–1000 (2001).7. Wang, H. et al. Science 293, 853–857 (2001).8. Kirmizis, A. et al. Nature 449, 928–932 (2007).9. Guccione, E. et al. Nature 449, 933–937 (2007).10. Hyllus, D. et al. Genes Dev. 21, 3369–3380 (2007).11. Iberg, A.N. et al. J. Biol. Chem. 283, 3006–3010 (2007).12. Holstege, F.C. et al. Cell 95, 717–728 (1998).13. Lakowski, T.M. & Frankel, A. J. Biol. Chem. 283, 10015–10025 (2008).14. Ruthenburg, A.J., Allis, C.D. & Wysocka, J. Mol. Cell 25, 15–30 (2007).15. Sims, R.J. III & Reinberg, D. Genes Dev. 20, 2779–2786 (2006).

BR I E F COMMUNICAT IONS

NATURE STRUCTURAL & MOLECULAR BIOLOGY VOLUME 16 NUMBER 4 APRIL 2009 4 5 1

©20

09 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.