Nature Structural Molecular Biology April

www.nature.com/nsmb EDITORIAL OFFICE 75 Varick Street, Fl 9, New York, NY 10013-1917 Tel: (212) 726 9331, Fax: (212)...

264 downloads 1640 Views 5MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

www.nature.com/nsmb

EDITORIAL OFFICE 75 Varick Street, Fl 9, New York, NY 10013-1917 Tel: (212) 726 9331, Fax: (212) 679 0735 Editor: Boyana Konforti Senior Editor: Michelle Montoya Associate Editors: Inês Chen, Sabbi Lall Assistant Editor: Maria Hodges Copy Editor: Carrie Patis Production Editors: Sabina Eberle, Jamel Wooten Senior Illustrator: Katie Vicari Illustrator: Kimberly Caesar Cover Design: Erin Boyle Editorial Assistant: Christina Polizoto

[email protected]

© 2009 Nature America, Inc. All rights reserved.

MANAGEMENT OFFICES NPG New York 75 Varick Street, Fl 9, New York, NY 10013-1917 Tel: (212) 726 9200, Fax: (212) 696 9006 Publisher: Stephanie Diment Executive Editor: Linda Miller Chief Technology Officer: Howard Ratner Head of Nature Research & Reviews Marketing: Sara Girard Marketing Manager: Leah Rodriguez Assistant Production Coordinator: Karen Wilson Head of Web Services: Anthony Barrera Web Production Manager: Susan Kline NPG London The Macmillan Building, 4 Crinan Street, London N1 9XW Tel: 44 207 833 4000, Fax: 44 207 843 4996 Managing Director: Steven Inchcoombe Publishing Director: Alison Mitchell Editor-in-Chief, Nature Publications: Philip Campbell Marketing Director: Della Sar Director of Web Publishing: Timo Hannay NPG Nature Asia-Pacific Chiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843 Tel: 81 3 3267 8751, Fax: 81 3 3267 8746 Publishing Director — Asia Pacific: David Swinbanks Associate Director: Antoine E. Bocquet Manager: Koichi Nakamura Operations Director: Hiroshi Minemura Marketing Manager: Masahiro Yamashita Asia-Pacific Sales Director: Kate Yoneyama Asia-Pacific Sales Manager: Ken Mikami DISPLAY ADVERTISING [email protected] (US/Canada) [email protected] (Europe) [email protected] (Asia) Global Head of Advertising and Sponsorship: Dean Sanderson, Tel: (212) 726 9350, Fax: (212) 696 9482 Global Head of Display Advertising: Andrew Douglas, Tel: 44 207 843 4975, Fax: 44 207 843 4996 Asia-Pacific Sales Manager: Ken Mikami, Tel: 81 3 3267 8765, Fax: 81 3 3267 8746 Display Account Managers: New England: Sheila Reardon, Tel: (617) 399 4098, Fax: (617) 426 3717 New York/Mid-Atlantic/Southeast: Jim Breault, Tel: (212) 726 9334, Fax: (212) 696 9481 Midwest: Mike Rossi, Tel: (212) 726 9255, Fax: (212) 696 9481 West Coast South: George Lui, Tel: (415) 781 3804, Fax: (415) 781 3805 West Coast North: Bruce Shaver, Tel: (415) 781 6422, Fax: (415) 781 3805 Germany/Switzerland/Austria: Sabine Hugi-Fürst, Tel: 41 52761 3386, Fax: 41 52761 3419 United Kingdom/Ireland: Jeremy Betts, Tel: 44 207 843 4968, Fax: 44 207 843 4749 Scandinavia/Iceland/Spain/Portugal: Evelina Rubio-Hakansson, Tel: 44 207 843 4079, Fax: 44 207 843 4749 France/Belgium/The Netherlands/Italy/Israel/Eastern Europe: Nicola Wright, Tel: 44 207 843 4959, Fax: 44 207 843 4749 Greater China/Singapore: Gloria To, Tel: 852 2811 7191, Fax: 852 2811 0743 NATUREJOBS [email protected] (US/Canada) [email protected] (Europe) [email protected] (Asia) US Sales Manager: Ken Finnegan, Tel: (212) 726 9248, Fax: (212) 696 9482 European Sales Manager: Dan Churchward, Tel: 44 207 843 4966, Fax: 44 207 843 4596 Asia-Pacific Sales Manager: Ayako Watanabe, Tel: 81 3 3267 8765, Fax: 81 3 3267 8746 SITE LICENSE BUSINESS UNIT Americas: Tel: (888) 331 6288 Asia/Pacific: Tel: 81 3 3267 8751 Australia/New Zealand: Tel: 61 3 9825 1160 India: Tel: 91 124 2881054/55 ROW: Tel: 44 207 843 4759

[email protected] [email protected] [email protected] [email protected] [email protected]

CUSTOMER SERVICE www.nature.com/help Senior Global Customer Service Manager: Gerald Coppin For all print and online assistance, please visit www.nature.com/help Purchase subscriptions: Americas: Nature Structural & Molecular Biology, Subscription Dept., 342 Broadway, PMB 301, New York, NY 10013-3910. Tel: (866) 363 7860, Fax: (212) 689 9108 Europe/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 812 358 Japan: 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 8746 India: 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 [email protected] Nature 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 321 531 US Reprints: Tel: (212) 726 9278, Fax: (212) 679 0843

volume 16 number 4 APRIL 2009

E d i t o r ia l 345

Pressing information


news and views 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

346

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

348

A tipping point for mistranslation and disease Paul Schimmel & Min Guo  see also p 353 and p 359

350

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

351

Yeast as budding stem cells? Inês Chen

352

research highlights

articles

Structural and kinetic analysis of a M. tuberculosis NAD+ synthetase gives insight into the coordination of catalysis. (p 421)

Ala

353

Bases in the anticodon loop of tRNA GGC prevent misreading Hiroshi Murakami, Atsushi Ohta & Hiroaki Suga  see also p 348

359

A sequence element that tunes Escherichia coli tRNA GGC to ensure accurate decoding Sarah 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 determination Chunnian 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 motifs Hyeung Kim, Ok-Hee Lee, Huawei Xin, Liuh-Yow Chen, Jun Qin, Heekyung Kate Chae, Shiaw-Yih Lin, Amin Safari, Dan Liu & Zhou Songyang

Ala

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.

i


volume 16 number 4 APRIL 2009

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)

380

Polyglutamine disruption of the huntingtin exon 1 N terminus triggers a complex aggregation mechanism Ashwani 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 IscS Salvatore 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 ribosome Christopher S Fraser, John W B Hershey & Jennifer A Doudna

405

Tertiary interactions within the ribosomal exit tunnel Andrey Kosolapov & Carol Deutsch

412

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

421

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

430

Precursor-product discrimination by La protein during tRNA metabolism Mark A Bayfield & Richard J Maraia

438

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

b r i e f c o m m u n i c at i o n s 446

CK2α phosphorylates BMAL1 to regulate the mammalian clock Teruya 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 2 Antonis Kirmizis, Helena Santos-Rosa, Christopher J Penkett, Michael A Singer, Roland D Green & Tony Kouzarides

nature structural & molecular biology classified

See back pages.

Precursor-product discrimination by the La protein is examined. (p 430)

nature structural & molecular biology

iii

E d i to r i a l

Pressing information


One 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.

A

s 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-soughtafter 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

a dequate 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

nature structural & molecular biology volume 16 number 4 APRIL 2009

345

news and views

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

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-

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: [email protected]

346

miR-9

miR-A?

miR-B? An

TLX mRNA miR-9? HDAC5 mRNA

An miR-9?

X mRNA

? An

miR-9

pre-miR-9-1 TLX

Cytoplasm Nucleus pre-miR-9-1

pri-miR-9-1 An

X? HDAC5 TLX

mir-9-1 locus

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.

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

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

volume 16 number 4 APRIL 2009 nature structural & molecular biology

Katie Vicari


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.


news and views 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

verexpression. Although these data imply o 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 TLXknockdown 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 crossrepression 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.


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).

347

news and views

A tipping point for mistranslation and disease Paul 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. 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 singlestranded 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.

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: [email protected]

348

a GTP

+

50S 30S

GTPase activation

Accommodation/ peptidyl transfer

Binding Rejection

Rejection

mRNA

b

+

GDP

T Ψ C loop

Minihelix domain

60 55

C Ψ T D

50

76

C C A

65 7

+

+

c

1

D loop

20

32

15

38 25 10

5′ 32

38

U 34 36 A 35 G G C

Anticodon loop

mRNA

3′

Ala

Figure 1 A 32-38 pair in tRNAGGC 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 Ala tRNAGGC. (c) The 32-38 pair, residing close to the anticodon triplet, affects codon-anticodon decoding.

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 codonanticodon 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

a nticodon 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.


news and views Table 1 Human mitochondrial tRNAs with mutations at 32-38 and their clinical phenotypes tRNA gene

Mutation site

Nucleotide change

Clinical 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.

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 wildtype 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

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 sugarphosphate 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


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. ACKNOWLEDGMENTS We 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).

349

news and views

At the (3′) end, you’ll turn to meiosis Alberto 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. 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,

Alberto Moldón and José Ayté are in the Oxidative Stress and Cell Cycle Group, Universitat Pompeu Fabra, Barcelona, Spain. e-mail: [email protected]

350

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.

Furthermore, in elegantly described e xperiments, 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 meiosisspecific 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.


news and views


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.

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 r egulation during meiosis in fission yeast. But we can be assured that other processes will participate in this complex network.

ACKNOWLEDGMENTS Work 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).

Yeast as budding stem cells? 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 m in the next round of replication and passed along to one of the b 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


351

research highlights


Orchestrating mitosis The 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

NMR in living cells Atomic-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

Don’t you know you’re toxic?

Digesting a switch When 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 doublestranded 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,

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

352

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

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


ARTICLES

Bases in the anticodon loop of tRNAAla GGC prevent misreading


Hiroshi 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 the anticodon triplets. Here we report that evolutionarily conserved pairs of bases at positions 32 and 38 in tRNAAla GGC prevent misreading of a near-cognate valine codon, GUC. The tRNAAla GGC molecules with the conserved A32-U38 and C32-G38 pairs do not read GUC, whereas those with three representative nonconserved pairs, U32-U38, U32-A38 and C32-A38, direct the misincorporation of alanine at this valine codon into the peptide chain. Overexpression of the nonconserved tRNAAla GGC in Escherichia 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 an aminoacyl-tRNA (aa-tRNA) whose anticodon base-pairs with the cognate codon encoded in the mRNA. It is well known that the third base pair in the codon-anticodon interaction tolerates a wobble pair represented by a U G interaction, in addition to the canonical Watson-Click base pair, however, this decoding does not generally alter the identity of the amino acid in the peptide, and thus its fidelity is maintained1. On the other hand, if a U G mispair at the first or second base pair occurs, such a codon-anticodon interaction does accompany an amino acid alteration, and therefore this kind of miscoding is generally prohibited. Still, some examples of misreading of the first codon have been reported in literature. For instance, in Saccharomyces cerevisiae, amber (UAG) mutations generated by UV irradiation are read by Gln-tRNAGln CUG, causing termination to be suppressed by glutamine incorporation2–6. In E. coli, mutation of the AGC (serine) codon to GGC (glycine) codon at the catalytic Ser68 residue in b-lactamase is suppressed by the endogenous Ser-tRNASer GCU, although the probability of misreading resulting in glycine incorporation was estimated to be less than 1 in 1,000 (ref. 7). It is also known that a base mutation or mutations near the junction of arms in the tRNA cloverleaf structure diminish decoding fidelity. One of the well-known cases is the G24A mutation in the D-stem of tRNATrp CCA, the so-called Hirsh suppressor tRNA, which misreads CGG (arginine) and UGA (see Fig. 1a for a reference of the 8,9 base position in a tRNA structure, tRNAAla GGC) . It was recently shown Trp that the Hirsh suppressor tRNACCA elevates the rates of both GTP hydrolysis and accommodation independently from the codonanticodon interaction, and thus the misreading described above occurs9. 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 WatsonCrick base pair in the anticodon stem of tRNATrp CUG increased the frequency of misreading of the first position wobble10,11. For instance, tRNATrp CUG bearing the G27-A43 mispair misread the UAG amber codon 40 times more frequently than the wild-type pair. Taken together with other biochemical data, it was postulated that such mutations possibly alter the angle of the junction of the anticodon stem and the central tRNA L-shaped structure, increasing the frequency of wobble reading10. Some bases in tRNA anticodon loop are also known to contribute to the maintenance of decoding fidelity. Although a typical example is base modifications in the anticodon loop that disrupt codon recognition12,13, here we focus on sequence variations in the anticodon loop. For instance, E. coli has tRNAGly with three isoacceptors for GGN (N can be any base) codons, whereas Mycoplasma mycoides has only tRNAGly UCC for reading these codons. It turns out that the difference in the sequence of the anticodon Gly loop between E. coli tRNAGly UCC and M. mycoides tRNAUCC is a base at position 32, in which the former has U32 whereas the latter has C32, both pairing with A38. Notably, the U32C mutation introduced into E. coli tRNAGly UCC made it capable of reading all four glycine codons14,15. This suggests that the base at position 32 in the anticodon loop influences the tolerance of the U34 U and U34 C mispairs in codon-anticodon recognition. As described earlier, however, this misreading does not accompany an amino acid alteration. Hence, the study described above does not explain the importance of these bases at positions 32 and 38 in decoding fidelity. Nevertheless, this work prompted us to investigate whether the conservation of positions 32 and 38 contributes to the ability of tRNAs to correctly decode cognate codons in E. coli.

1Research Center for Advanced Science and Technology, University of Tokyo, Meguro-ku, Tokyo, Japan. 2Department of Chemistry and Biotechnology, Graduate School of Engineering, University of Tokyo, Bunkyo-ku, Tokyo, Japan. Correspondence should be addressed to H.M. ([email protected]) or H.S. ([email protected]).

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

NATURE STRUCTURAL & MOLECULAR BIOLOGY

VOLUME 16

NUMBER 4

APRIL 2009

353

ARTICLES

a

b

76

Frequency (%)

E. coli tRNA

73

1

Conserved

19

5 24 27

43

0 32

Nonconserved

38 0


Ala

Figure 1 Structure of wild-type tRNAGGC and its variants (a) wild-type E. coli tRNAAla GGC. (b) Frequency in the occurrence of the 32-38 pair in Ala 84 nonredundant sequences of bacterial tRNAAla GGC. The tRNAGGC with the A32-U38 and C32-G38 pairs are referred to as conserved tRNAAla GGC, whereas those with the U32-U38, U32-A38 and C32-A38 pairs are referred to as nonconserved tRNAAla GGC.

RESULTS An evolutionary bias of the 32-38 pair in tRNAAla GGC First, we used the tDNA database to look for evolutionary bias in the 32-38 pair. Prior to our study, a 1982 report on the comparison of 42 kinds of E. coli and bacterial phage tRNA sequences focusing on the anticodon stem-loop region, proposed that some of the base pairs in the anticodon stem and the bases at positions 37 and 38 might show a preference for certain nucleotides depending upon the base at position 36, which forms the first base pair in the codon-anticodon interaction16. This finding has led to ‘the extended anticodon hypothesis’, which posits that these bases evolved to optimize translation efficiency and, possibly, decoding fidelity. Furthermore, this hypothesis was experimentally verified by suppression of the amber codon by mutant Glu tRNATrp CUA, and of the ochre codon by mutant tRNAUUA, demonstrating that tRNAs with an extended anticodon sequence showed the highest suppression efficiency17–20. More recently, 5,601 bacterial tRNA sequences were extracted from the tDNA database and used to analyze the statistical conservation of bases at the 32 and 38 positions21. Certain tRNAs have a specific subset of combinations that differ from those of other tRNAs. For instance, 99% of tRNAAla GGC contain either A32-U38 (77%) or C32-G38 (22%), whereas the bases contained in bacterial tRNAs in general have frequencies of 52% for C32-A38, 17% for U32-A38, 11% for U32-U38 and 8% for C32-C38. Notably, tRNAAla GGC derivatives with nonconserved pairs such as U32-U38 and U32-A38 dissociate from the

Figure 2 Decoding efficiency of the GCC codon by tRNAAla GGC 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 tRNAAla GGC in the wPURE system. The tRNA mix Tyr Asp Lys consists of in vitro transcripts of tRNAfMet CAU, tRNAGUA, tRNAGUC and tRNACUU. The peptide was expressed at 37 1C for 15 min in the presence of 0.2 mM proteinogenic 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 tRNAAla GGC variant in the wPURE system. (d) Tricine SDS-PAGE analysis of the competitive decoding of the GCC codon by Ala-tRNAAla GGC and Leu-tRNALeu GGC in the wPURE system. The competition contained 3 mM Leu Ala tRNAGGC and each tRNAGGC variant to a concentration of 3 mM. Arrows indicate Ala-peptide (A) and Leu-peptide (C).

354

VOLUME 16

A site of the E. coli ribosome four to ten times more slowly than those containing A32-U38 (ref. 22). Moreover, the U32C mutation of tRNAGly CCC, which is 98% conserved with the U32-A38 pair, increases the affinity of the tRNA not only to the cognate codon but also to the near-cognate codons involving third position mismatches21. These results imply that the 32-38 pair influences the affinity of tRNAs in the A site; however, again, these third position mismatches in the codonanticodon interaction do not alter the amino acid, so it is unclear whether the evolutionary force driving this bias in the 32-38 pair, which depends on the anticodon triplet, arises from the need to control efficiency in translation or decoding fidelity. We also independently searched the bacterial tDNA database23 to assess the sequence bias in the 32-38 pair in 84 nonredundant sequences of bacterial tRNAAla GGC. We found a trend similar to that previously described21 (Fig. 1). Note that no bacterial tRNAAla GGC contains U32-A38 and C32-A38 pairs, whereas archeal tRNAAla GGC has the U32-A38 pair (18% out of 17 nonredundant sequences) but, again, no C32-A38 pair. Because the above sequence bias in the 32-38 pair possibly determines the translation efficiency of Ala-tRNAAla GGC, we analyzed the difference in translation efficiency of in vitro transcripts between E. coli tRNAAla GGC (Fig. 1a) with the conserved A32-U38 pair (wild type) or the C32-G38 pair, as well as the nonconserved U32-U38, U32-A38 and C32-A38 pairs (Fig. 1b). For simplicity, we refer the former and latter sets of tRNAAla GGC as conserved and nonconserved tRNAAla GGC, respectively. No change in the decoding efficiency of the GCC cognate codon To assess the translation efficiency of each tRNAAla GGC variant, we used an E. coli cell-free translation system that was specially reconstituted for this experiment. In this system, the native tRNAs were entirely substituted with in vitro transcripts of four tRNAs (tRNAfMet CAU, Asp Lys tRNATyr GUA, tRNAGUC and tRNACUU; we refer to the mixture of these tRNAs as ‘tRNA mix’) along with a tRNAAla GGC, referred to as the wPURE system (w stands for ‘withdrawn’). To validate whether this

a mRNA 5′

3′

Peptide

b

c Natural tRNAs

+

–

–

tRNA mix

–

–

+

+

tRNAGGC (32-38)

–

–

–

A-U

Lanes

1

2

3

4

Ala

d

(32-38) Lanes

(A)

(B)

(B) Ala

Leu tRNAGGC

Lanes

38

Codon

(A)

tRNAGGC (32-38)

32

Anticodon loop

–

–

– 1

A-U C-G U-U U-A C-A 2 3 4 5 6

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

+

–

+

+

+

+

+

1

2

3

4

5

6

7

(A) (C)

NUMBER 4

APRIL 2009


ARTICLES 3′

mRNA 5′ Peptide

c Natural tRNAs tRNA mix tRNAVal GAC Lanes

+ – – 1

– – – 2

– + – 3

– + + 4

32

e Calculated mass: Val-peptide m/z 1655.795 [M+H]+ Ala-peptide m/z 1627.763 [M+H]+

38

Anticodon loop Codon Ala (32-38) tRNAGGC

Lanes

– A-U C-G U-U U-A C-A 1

2

3

4

5

6

Intensity (a.u.)

b

d

200

Intensity (a.u.)

a

200

3 µM tRNALeu GAC

Obs.=1656.419

Val-peptide Codon:GUC

150

(D) 1600

1680

0

(A)

(C)


(B)

0.3 µM

Obs.=1628.287

Ala-peptide Codon:GUC

Ala (32-38) tRNAGGC

Lanes

150

tRNALeu GAC

A-U C-G U-U U-A 6 7 8 9

C-A 10

100 50 0

(B)

C-A 5

(C)

100 50

Ala tRNAGGC (32-38) A-U C-G U-U U-A Lanes 1 2 3 4

1600

1,200

(C)

1680

(D) 1,400

1,600 m/z

1,800

2,000

Figure 3 Influence of the sequence variation of the 32-38 pair in tRNAAla GGC 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 14 in vitro transcript of tRNAVal GAC in the wPURE system. Other conditions were the same as Figure 2b. Arrows indicate Val-peptide (A) and [ C]aspartate (B). (c) Tricine SDS-PAGE analysis of the peptide in the presence of tRNA mix and each tRNAAla GGC 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. Leu Leu Ala (e) Competitive decoding of the GUC codon by tRNAAla GGC variants and tRNAGAC. In lanes 1–5, 3 mM tRNAGAC and 3 mM each tRNAGGC variant were used; Ala variant were used. Arrows indicate Ala-peptide (C), Leu-peptide (D). in lanes 6–10, 0.3 mM tRNALeu and 3 mM each tRNA GAC GGC

wPURE system was able to function like the ordinary PURE system24 for the expression of a model peptide consisting of amino acids assigned by the above tRNAs, a 13-mer peptide, MKKKADYKDDDDK (italicized residues indicate a Flag peptide sequence), was expressed from the corresponding mRNA (Fig. 2a) in the presence of wild-type 14 tRNAAla GGC and [ C]Asp in both systems. We determined the expression level of the peptide by the intensity of the radioactive band following tricine SDS-PAGE, showing that the wPURE system functioned like the ordinary PURE system for the expression of this peptide (Fig. 2b, lane 1 versus lane 4). Most importantly, the expression was tRNAAla GGC dependent (lanes 3 and 4). MALDI-TOF analysis of the peptide expressed in the wPURE system also confirmed the accuracy of expression (data not shown), indicating that correct reading of the GCC codon could be achieved by tRNAAla GGC. We then tested the tRNAAla GGC variants (Fig. 1) in the wPURE system for the decoding ability of the respective tRNAs to the GCC cognate codon. 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 efficiency is likely to reflect the intrinsic decoding ability of each tRNAAla GGC to the GCC codon. Unexpectedly, we observed no difference in incorporation efficiency (Fig. 2c). To avoid exhausting the energy source of translation, we terminated the reaction described above after 15 min (Supplementary Fig. 2 online); however, it was still possible that the difference in the decoding ability of each tRNAAla GGC was so small that the apparent translation efficiency was not sensitive enough to reflect to the actual value under such conditions. We therefore performed an additional experiment to 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 carrying the anticodon loop sequence of E. coli wild-type tRNAAla GGC (Supplementary Figs. 1 and 3 online). In fact, when we added tRNALeu GGC to


VOLUME 16

the wPURE system instead of tRNAAla GGC, translation of the same mRNA took place smoothly (Fig. 2d, lane 1). Notably, this leucinecontaining peptide (Leu-peptide) appeared as a faster-migrating band than the alanine-containing peptide (Ala-peptide) band in tricine SDS-PAGE (Fig. 2d, lanes 1 and 2). MALDI-TOF analysis also revealed a molecular mass consistent with the Leu-peptide (data not shown), indicating that the single substitution of alanine to leucine in this peptide altered its migration properties. Thus, this feature allowed us to use tricine SDS-PAGE to conveniently assess the expression level of the individual peptides in competition assays between tRNAAla GGC and tRNALeu GGC. We observed no appreciable difference in the intensities between the Ala- and Leu-peptides generated by any of tRNAAla GGC variants competing with tRNALeu GGC (lanes 3–7). These experiments clearly showed that the conserved and nonconserved tRNAAla GGC variants were able to decode the GCC cognate codon with similar efficiencies. We thus suspected that the evolutionary conservation of the 32-38 pair in tRNAAla GGC arose for a different reason(s). The 32-38 pair controls misreading of GUC near-cognate codon As sequence variation in the 32-38 pair did not affect decoding efficiency, we turned our investigation toward its decoding fidelity. The wobble pairing at the second G35 in tRNAAla GGC to a near-cognate valine codon, GUC, would be expected to alter the amino acid incorporation from valine to alanine. We therefore prepared another mRNA template based on the previously used mRNA in which the GCC codon was substituted with a GUC codon, and tested whether misreading 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 the expression of peptide (Fig. 3b, lane 3). Even though the isolated background-level peptide was present only in trace amounts, MALDITOF analysis revealed that it was consistent with the molecular mass

NUMBER 4

APRIL 2009

355


ARTICLES of the valine-containing peptide (Val-peptide) as a major peak (data not shown). This suggests that the background expression can be attributed to a trace amount of tRNAVal GAC contaminating the wPURE system. On the other hand, addition of the in vitro transcript of tRNAVal GAC to the wPURE system markedly elevated the expression level of peptide (Fig. 3b, lane 4). We then tested whether alanine misincorporation at the GUC codon could be induced by addition of tRNAAla GGC variants to the wPURE system. The presence of wild-type or C32-G38 tRNAAla GGC slightly increased the background expression, presumably owing to misreading of the GUC codon resulting in alanine incorporation into the peptide chain (Fig. 3c, lanes 1–3). Unexpectedly, the presence of nonconserved tRNAAla GGC (U32-U38, U32-A38 and C32-A38) substantially increased the expression level (Fig. 3c, lanes 4–6, respectively). MALDI-TOF analysis of the isolated peptide showed a single major peak of molecular mass corresponding to the Ala-peptide (Fig. 3d). This result clearly shows that the background incorporation at the GUC codon by the contaminated tRNAVal GAC was completely competed out by the nonconserved tRNAAla GGC. Even though the nonconserved tRNAAla GGC misreads GUC effectively in the wPURE system, in E. coli the cognate tRNAVal GAC coexists endogenously and thus competes out such a misreading event. Therefore, it was necessary to assess how effectively misreading occurred under the competitive conditions. Because the Val-peptide and the Ala-peptide had nearly the same migration pattern in tricine SDS-PAGE (Fig. 3b,c), it was difficult to quantitatively assess the competition. Instead, we engineered a tRNALeu containing the native anticodon loop sequence of E. coli tRNAVal GAC (Supplementary Fig. 3c) and used it as a competitor against each tRNAAla GGC variant. As expected on the basis of previous experiments28–30, LeuRS charged leucine onto the engineered tRNALeu GAC (Supplementary Fig. 1) and the resulting Leu-tRNALeu GAC decoded the mRNA GUC codon, yielding the Leu-peptide. Because the Leu-peptide migrated faster than the Ala-peptide in tricine-SDS-PAGE, we could readily visualize the degree of competition (Fig. 3e). When we added an equal amount of each tRNAAla GGC variant and tRNALeu GAC to the wPURE system, only the Leu-peptide band was observed in all cases, suggesting that each Ala-tRNAAla GGC variant was completely competed out by Leu-tRNALeu GAC (Fig. 3e, lanes 1–5). However, when we reduced the concentration of the tRNALeu GAC to one-tenth that of tRNAAla GGC, a faint but clearly visible Ala-peptide band appeared in the presence of the nonconserved tRNAAla GGC (Fig. 3e, lanes 6–10). Particularly, the frequency of misreading of GUC by Ala-tRNAAla GGC containing the C32-A38 pair reached approximately 30% (Fig. 3e, lane 10). This result clearly indicates that the 32-38 pair in tRNAAla GGC controls misreading of the near-cognate GUC codon. Overexpression of the nonconserved tRNAAla GGC is toxic in E. coli The above in vitro experiments clearly demonstrated that the nonconserved tRNAAla GGC misreads the near-cognate GUC codon involving the G35U wobble pair. We wondered whether this misreading event could occur in vivo, so that the nonconserved tRNAAla GGC acts as a toxigenic tRNA. We transformed E. coli BL21 cells with a vector that could overexpress each conserved or nonconserved tRNAAla GGC variant under the control of an arabinose promoter (Supplementary Fig. 4 online). The tranformed cells were grown individually on either 0.2% (w/v) glucose (negative control) or 0.2% (w/v) arabinose on LB agar plates at 42 1C. Before induction of tRNA expression, all cells appeared as healthy as the untransformed control cells (Fig. 4a). Upon induction, cells expressing the conserved tRNAAla GGC showed no change in growth, whereas those expressing the nonconserved tRNAAla GGC became

356

VOLUME 16

a

b 0.2 % Glucose

No tRNA

0.2 % Arabinose

A-U

No tRNA

C-G

C-A

U-A

U-U

A-U

C-G

C-A

U-A

U-U

Figure 4 Overexpression of the conserved or nonconserved tRNAAla GGC in E. coli (BL21). Each tRNAAla GGC 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.

unhealthy (Fig. 4b). Particularly, those expressing the nonconserved tRNAAla GGC with U32-A38 or C32-A38 were unable to grow. These U32-A38 and C32-A38 pairs were never found in the tRNAAla GGC sequence database, indicating that the sequence bias of the 32-38 pair in tRNAAla GGC probably appeared to avoid formation of toxigenic tRNAs in vivo. It should be noted that at 37 1C most cells appeared to be healthy, with the exception of those cells expressing the nonconserved tRNAAla GGC with C32-A38, which grew slightly more slowly (data not shown). This temperature sensitivity may suggest that the frequency of the misreading of the GUC codon by tRNAAla GGC with the nonconserved C32-A38 pair is not marked because the codon is predominantly read correctly by the cognate tRNAVal GAC. However, in some proteins the resulting valine to alanine substitution would cause them to be less stable, resulting in loss of function at 42 1C. This probably led to the observed temperature-dependent cell growth. Nonetheless, our demonstration clearly shows that the nonconserved tRNAAla GGC is toxic in vivo and is therefore not conserved in the repertoire of functional tRNAs. DISCUSSION Here we provide in vitro evidence that the nonconserved tRNAAla GGC (Fig. 1) misreads its near-cognate valine codon, GUC, resulting in misincorporation of alanine into the valine site of the peptide chain (Fig. 3). In contrast, misreading of this codon by the conserved tRNAAla GGC (Fig. 1) is minimal and thus is readily competed out by the cognate tRNALeu GAC (Fig. 3). This observation is also valid in vivo, where overexpression of the nonconserved tRNAAla GGC is toxic, whereas that of the conserved tRNAAla GGC is not (Fig. 4). These results imply that the reason for the evolutionary force selecting the 32-38 pair in tRNAAla 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, which occur before and after GTP hydrolysis, respectively31,32. In the initial selection step, incorrect tRNA is rejected by rapid dissociation of the ternary complex of aa-tRNA–EF-Tu–GTP from the A site and the sluggish rate of GTP hydrolysis33,34. Even though GTP hydrolysis occasionally occurs for the incorrect aa-tRNA, in the next proofreading step the slow accommodation rate of the incorrect aa-tRNA to the peptidyl-transferase center results in its rejection, and therefore incorrect reading of the noncognate codon is avoided34. It is likely that the sequence variation of 32-38 pair in tRNAAla GGC also influences either or both steps of aa-tRNA selection. It was reported that the nonconserved tRNAAla GGC(U32-U38 or U32-A38) binding to the cognate GCC codon has a slower dissociation rate from the A site than 22 the conserved tRNAAla GGC(A32-U38 or C32-G38) . Therefore, an

NUMBER 4

APRIL 2009



ARTICLES explanation for the increase in the frequency of misreading of GUC by such nonconserved tRNAAla GGC is also due to their slow dissociation rate from the ribosome. Recently, various kinetics measurements were performed for misreading of near-cognate codons, including GUC, by the conserved tRNAAla GGC(A32-U38, C32-G38) and non35 conserved tRNAAla GGC(U32-A38 or C32-A38) . The apparent rate of peptide bond formation in misreading of the GUC codon by the two nonconserved tRNAAla GGC(U32-A38 or C32-A38) is elevated to the level of that which occurs during reading of the cognate GCC codon. Clearly, this result is consistent with our finding that the nonconserved tRNAAla 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 U32A38 and C32-A38 pairs, belonging to the largest structural family I, form noncanonical structures involving bifurcated hydrogen bonds. In contrast, the U32-U38 pair, categorized in family II, forms a single, noncanonical hydrogen bond. Structures for the A32-U38 and C32G38 pairs, in family III, cannot yet be predicted because of insufficient available structural information. It should be noted that families I and II combined constitute about 93% of bacterial tRNAs36, implying that these base pairs evolved to maximize the decoding ability of tRNAs on the ribosome. In the present study, we have shown that, paradoxically, the family I tRNAAla GGC with U32-A38 or C32-A38 and the family II tRNAAla with U32-U38 misread GUC codon. Consequently, the rare GGC family III pairs, A32-U38 and C32-G38, are found in the naturally occurring tRNAAla GGC. This suggests that the decoding fidelity of tRNAAla GGC is tuned by selecting uncommon 32-38 pairs during the evolution. Presumably, similar unique sequence biases that tune decoding fidelity can be found in many regions of the tRNA body sequence37. More extensive sequence analyses of tRNAs and biochemical studies on such evolutionarily biased variants will be important to reveal the mechanism of decoding fidelity in translation. METHODS Materials. 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 Val other all experiments. The concentrations of tRNAAla GGC variants, tRNAGAC, the Leu engineered tRNALeu GGC and tRNAGAC 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,


VOLUME 16

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 measurements 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 tRNAAla GGC 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 tRNAAla GGC 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 & Molecular Biology website. ACKNOWLEDGMENTS We thank O.C. Uhlenbeck and S. Ledoux for their invaluable discussion. This work was supported by grants from the Japan Society for the Promotion of 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 Technology Program in the New Energy and Industrial Technology Development Organization (NEDO) to H.S., the Industrial Technology Research Grant Program in NEDO (05A02513a) to H.M., and the Takeda Science Foundation. AUTHOR CONTRIBUTIONS This study was designed by H.M., A.O. and H.S.; all of the experiments were performed 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 is expressed as an amber suppressor when present in high copy number in yeast. Gene 29, 69–76 (1984). 3. Pure, G.A., Robinson, G.W., Naumovski, L. & Friedberg, E.C. Partial suppression of an ochre mutation in Saccharomyces cerevisiae by multicopy plasmids containing a normal 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 codonanticodon pairing: amber suppression by a yeast glutamine tRNA. Gene 49, 383–388 (1986).

NUMBER 4

APRIL 2009

357


ARTICLES Gln can suppress amber codons and is 5. Weiss, W.A. & Friedberg, E.C. Normal yeast tRNACAG 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 Gln can effect partial suppression of amber mutations in the yeast normal tRNACAG 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 codonanticodon 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. Interaction between 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 the recognition 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 determines ability 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 a weak, 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 anticodon arm 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 tRNAGluSuoc 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 loop are independent by genetic criteria. Nucleic Acids Res. 15, 4669–4686 (1987). 21. Olejniczak, M. & Uhlenbeck, O.C. tRNA residues that have coevolved with their anticodon 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 of tRNAs 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 tRNA sequences and sequences of tRNA genes. Nucleic Acids Res. 26, 148–153 (1998).

358

VOLUME 16

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 the identity of a transfer RNA. Nature 333, 140–145 (1988). 26. Francklyn, C. & Schimmel, P. Aminoacylation of RNA minihelices with alanine. Nature 337, 478–481 (1989). 27. Tamura, K., Asahara, H., Himeno, H., Hasegawa, T. & Shimizu, M. Identity elements of Escherichia coli tRNAAla. J. Mol. Recognit. 4, 129–132 (1991). 28. Normanly, J., Ogden, R.C., Horvath, S.J. & Abelson, J. Changing the identity of a transfer RNA. Nature 321, 213–219 (1986). 29. Asahara, H. et al. Recognition nucleotides of Escherichia coli tRNALeu and its elements facilitating discrimination from tRNASer and tRNATyr. J. Mol. Biol. 231, 219–229 (1993). 30. Tukalo, M., Yaremchuk, A., Fukunaga, R., Yokoyama, S. & Cusack, S. The crystal structure of leucyl-tRNA synthetase complexed with tRNALeu in the post-transferediting conformation. Nat. Struct. Mol. Biol. 12, 923–930 (2005). 31. Rodnina, M.V. & Wintermeyer, W. Fidelity of aminoacyl-tRNA selection on the ribosome: kinetic and structural mechanisms. Annu. Rev. Biochem. 70, 415–435 (2001). 32. Rodnina, M.V. & Wintermeyer, W. Ribosome fidelity: tRNA discrimination, proofreading and induced fit. Trends Biochem. Sci. 26, 124–130 (2001). 33. Pape, T., Wintermeyer, W. & Rodnina, M.V. Complete kinetic mechanism of elongation factor 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 and proofreading of aminoacyl-tRNA on the ribosome. EMBO J. 18, 3800–3807 (1999). 35. Ledoux, S., Olejniczak, M. & Uhlenbeck, O.C. A sequence element that tunes Escherichia coli tRNAAla GGC to ensure accurate decoding. Nat. Struct. Mol. Biol. advance online publication, doi:10.1038/nsmb.1581 (22 March 2009). 36. Auffinger, P. & Westhof, E. Singly and bifurcated hydrogen-bonded base-pairs in tRNA anticodon hairpins and ribozymes. J. Mol. Biol. 292, 467–483 (1999). 37. Saks, M.E. & Conery, J.S. Anticodon-dependent conservation of bacterial tRNA gene sequences. RNA 13, 651–660 (2007). 38. Milligan, J.F., Groebe, D.R., Witherell, G.W. & Uhlenbeck, O.C. Oligoribonucleotide synthesis 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 physiologically stable cyclic peptides. ACS Chem. Biol. 3, 120–129 (2008). 41. Kawakami, T., Murakami, H. & Suga, H. Messenger RNA-programmed incorporation of multiple 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-resistant peptides closed by a nonreducible inter-side-chain bond. ACS Chem. Biol. 3, 241–249 (2008).

NUMBER 4

APRIL 2009


ARTICLES

A sequence element that tunes Escherichia coli tRNAAla GGC to ensure accurate decoding


Sarah 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 tRNAAla GGC to a more common U32-A38 pair results in a tRNA that performs almost normally on cognate codons but is unusually efficient in reading nearcognate codons. Pre–steady state kinetic measurements on E. coli ribosomes show that, unlike the wild-type tRNAAla GGC, the misreading mutant tRNAAla shows rapid GTP hydrolysis and no detectable proofreading on near-cognate codons. Similarly, GGC Ala tRNAAla GGC mutated to contain C32-G38, a pair that is found in some bacterial tRNAGGC sequences, was able to decode only the cognate codons, whereas tRNAAla GGC 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 evolved to suppress translation of near-cognate codons.

Numerous biochemical experiments suggest that the 45 elongator aminoacyl-tRNAs (aa-tRNAs) in E. coli act as equivalent substrates of the translational machinery. More than 20 different E. coli aa-tRNAs were found to bind elongation factor Tu (EF-Tu) with similar affinities1, and 8 show nearly identical rates of dissociation from the A site and the P site of encoded E. coli ribosomes2. Recent experiments have shown that ten different E. coli aa-tRNAs have nearly identical ternary complex binding affinities to the ribosomal entry site and similar rates of GTP hydrolysis and peptide bond formation during decoding3. Despite their uniform functional properties, aa-tRNAs are quite different from one another chemically. Phylogenetic analysis of tRNA sequences from 145 bacteria with fully sequenced genomes indicates that each tRNA isoacceptor has a unique set of consensus residues distributed throughout the molecule4. In addition, each tRNA species contains different types and numbers of post-transcriptional modifications in the anticodon loop and the tertiary core5. Several experiments have shown that when these consensus residues are mutated or when one or more of the modifications are removed, the uniform functional properties of the aa-tRNA are lost. For example, removing all the post-transcriptional modifications from aa-tRNAs weakens the binding to the ribosomal A or P sites of several aa-tRNAs2. Base pair changes in the T-stem of individual tRNAs can either weaken or strengthen their binding affinity to EF-Tu6. Removal of selected modifications or changes in the sequence within the body of a suppressor tRNA can also either increase or decrease its ability to decode a termination codon in vivo7–9. These experiments suggest that the overall chemical composition of every aa-tRNA has been ‘tuned’ by evolution such that each aa-tRNA functions equivalently in the decoding process.

Although the emerging data support the view that tRNA sequences are idiosyncratically tuned for uniform behavior during decoding, it does not explain the underlying reason why this has occurred. It is unclear what the evolutionary disadvantage would be if the different aa-tRNAs showed a range of affinities for the ribosome or proceeded through the decoding pathway at different rates. One possibility is that the uniform behavior is related to the need for aa-tRNAs to undergo accurate decoding. Each aa-tRNA must read its cognate codons, but it must not efficiently read the structurally similar near-cognate codons containing a single-nucleotide mismatch. Numerous experiments have shown that the introduction of certain tRNA mutations or the removal of an individual post-transcriptional modification can lead to misreading10–14 or translational frameshifting15–18 in vivo. However, a mechanistic understanding of this phenomenon is limited to the G24A mutation of E. coli tRNATrp, which substantially promotes misreading of several near-cognate codons19. Here we evaluated mutations in the anticodon loop of E. coli tRNAAla GGC that are known to stabilize binding to ribosomes for their ability to read near-cognate codons. To achieve this, we used kinetic and thermodynamic assays that measure different steps in the decoding process. RESULTS Mutating A32-U38 has little effect on cognate decoding tRNAAla GGC is the minor alanine isoacceptor in E. coli that selectively reads its complementary GCC and wobble GCU codons20. The major isoacceptor tRNAAla UGC is capable of reading all four alanine codons, so tRNAAla GGC is not essential for growth, although its deletion causes a slow-growth phenotype in minimal media21. One of the distinctive structural features of tRNAAla GGC is the A-U pair at positions 32 and 38 at the top of the anticodon loop (Fig. 1). This combination of residues

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. ([email protected]).



VOLUME 16

NUMBER 4

APRIL 2009

359


ARTICLES

Figure 1 Secondary structure of E. coli tRNAAla GGC. 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 tRNAAla GGC but are not conserved among all bacterial tRNAAla GGC. Positions 32 and 38 in the anticodon loop are numbered.

is rare in bacterial tRNAs, being present only in tRNAAla GGC and (ref. 5). Recent experiments measuring the binding of tRNAPro GGG E. coli tRNAAla GGC to the A site of ribosomes encoded with a complementary GCC codon showed that the identity of the nucleotide pair at positions 32 and 38 modulates the tRNA binding affinity22. Whereas the wild-type tRNAAla GGC demonstrated an A site binding affinity similar to other deacylated elongator tRNAs2, replacement of the A32-U38 pair by U-A, U-U or A-A pairs caused the binding affinity 22 of tRNAAla GGC to be four- to ten-fold stronger . Introduction of the C32-G38 pair, which is present in tRNAAla GGC sequences of some other bacteria, had no effect on A site binding. This suggested that the rare A32-U38 pair and its phylogenetic alternative, C32-G38, have evolved to weaken the tRNAAla GGC binding affinity for ribosomes to ensure that its affinity is similar to that of other tRNAs. Because the binding affinity of aa-tRNAs to the ribosomal A site does not measure a step in the normal decoding process, discerning the relevance of the 32-38 pair for tRNAAla GGC function requires assays that measure decoding directly. As mutated tRNAAla GGC sequences are most easily tested using unmodified tRNA transcripts, we first compared the decoding properties of unmodified tRNAAla GGC to previous data obtained for its fully modified counterpart. We assayed unmodified Ala-tRNAAla GGC on E. coli ribosomes programmed with a 27-nucleotide derivative of the initiation region of T4 gp32 mRNA displaying the cognate GCC codon in the A site and an AUG codon in the P site. As previously described in greater detail3, we used three different assays to evaluate the ability of Ala-tRNAAla GGC to undergo decoding. First, we determined the Kd of a catalytically inactive ternary complex bound to the entry site of E. coli ribosomes containing tRNAfMet in the P site23 (Fig. 2a). Second, we determined the rate of GTP hydrolysis by the ternary complex at several encoded ribosome concentrations to deduce kGTPmax, the GTPase rate at saturating ribosome concentrations (Fig. 2b,c). Finally, we measured kpep, the observed rate of peptide bond formation between fMet-tRNAfMet and Ala Ala-tRNAAla GGC (Fig. 2d). The unmodified tRNAGGC and the previously Ala assayed modified tRNAGGC showed similar Kd values (2.3 nM

360

VOLUME 16

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 determined under identical reaction conditions3. Thus, unlike with several other tRNAs2,10,24, the post-transcriptional modifications have only a small effect on the decoding process of tRNAAla GGC in the conditions used in these in vitro experiments. This is likely to reflect the fact that native tRNAAla GGC has no modifications in the anticodon loop and only five modifications in the tertiary core, which do not directly contact the ribosome20,25,26 (Fig. 1). Although removal of the modifications in the tertiary core of tRNA can destabilize tRNA structure, these effects are minimal in the buffer containing 10 mM MgCl2. However, in buffers containing lower MgCl2 concentrations, such as the highfidelity buffers often used in translation studies, transcripts are not fully folded27–30. We next compared the ability of the unmodified wild-type tRNA, tRNAAla GGC (wt), to decode its cognate codons with a tightly binding double mutant where the wild-type A32-U38 pair was changed to a 22 U32-A38 pair, tRNAAla GGC (UA) . Both of these tRNAs were effective in decoding the GCC and GCU codons, but the ternary complex containing tRNAAla GGC (UA) bound to the perfectly complementary GCC codon approximately two-fold more tightly and the wobble GCU codon about four-fold more tightly than tRNAAla GGC (wt) (Table 1). However, tRNAAla GGC (UA) showed kGTPmax and kpep values that were indistinguishable from tRNAAla GGC (wt) on both cognate codons (Table 1). Thus, it seems that, although mutating the A32U38 pair to U32-A38 in tRNAAla GGC slightly increases the affinity of the ternary complex for the ribosome, it does not affect the subsequent kinetic steps of decoding under the conditions used here. The A32-U38 pair in tRNAAla GGC prevents misreading We assayed the binding affinities of the wild-type and mutant tRNAAla GGC ternary complexes to the ribosomal entry site using the two near-cognate alanine codons GCA and GCG, which introduce an A-G or G-G mismatch into the third position of the codon-anticodon helix. As would be expected on the basis of previous studies comparing wild-type tRNA binding to near-cognate codons19,31, both ternary complexes bound the mismatched codons much more weakly than the cognate codons (Fig. 2a and Table 1). However, the ternary complex containing tRNAAla GGC (UA) bound the near-cognate codons at least five-fold more tightly than complexes containing tRNAAla GGC (wt). Hence, the stabilizing effect of mutating the A32-U38 base pair to U-A is similar or even slightly greater on the near-cognate codons than was observed with the cognate codons. The rate of ternary complex association to ribosomes is the same for both cognate and nearcognate codons32, so it is likely that the stabilizing effect of the mutation is due to a slower dissociation rate of the ternary complex off the ribosome. This would result in tRNAAla GGC (UA) being less accurate in the initial selection step of decoding33. This suggests that one reason the A32-U38 pair in tRNAAla GGC has evolved to be so well conserved is to destabilize ternary complex binding to ribosomes and thereby improve the accuracy of the initial selection steps of decoding. To assess whether the 32-38 pair also influences translation accuracy in the subsequent steps of decoding, it was first necessary to determine how well the tRNAAla GGC (wt) transcript can decode a near-cognate codon. It was possible to obtain values of kGTP at several ribosome concentrations and estimate a kGTPmax using the near-cognate GCA codon, despite the fact that the ternary complex binds more weakly to ribosomes containing mismatched codons (Fig. 2b,c). The value of kGTPmax was 2.4 s–1, 13-fold slower than the value obtained for the cognate GCC codon. The formation of dipeptide bond on the nearcognate GCA codon occurs much more slowly than kpep ¼ 1.7 s–1

NUMBER 4

APRIL 2009


ARTICLES The tighter binding tRNAAla GGC (UA) showed markedly different behavior from a 1 b 1 tRNAAla GGC (wt) in decoding the near-cognate GCA codon. tRNAAla GGC (UA) showed a value of 27 s–1, substantially faster k GTPmax 0.8 0.8 than the value of 2.4 s–1 observed with tRNAAla GGC (wt) and essentially the same rate 0.6 0.6 as observed on its cognate codon (Fig. 2b,c and Table 1). Similarly, the kpep value of 2.1 –1 on the GCA near-cognate codon was also s 0.4 0.4 significantly accelerated, such that it is nearly equal to the value determined using the 0.2 0.2 cognate GCC codon (Fig. 2d and Table 1). The fact that the fraction of dipeptide formed reached the same level with near-cognate 0 0 0.001 0.01 0.1 1 10 100 0.1 1 10 100 1,000 104 codons as with cognate codons indicates Time (s) Ribosome concentration (nM) that the mutant tRNA is not considerably 1.5 rejected off the ribosome in the presence of 20 0.35 c d the near-cognate codon. In other words, 1 0.3 tRNAAla GGC (UA) seems to evade the proof0.5 reading process by stimulating the forward15 0 0 2 4 0.25 Ribosome concentration (µM) reaction rates so that it efficiently makes dipeptide on the near-cognate codon. 0.2 The difference in initial selection rates for 10 tRNAs on cognate versus near-cognate codons 0.15 is increased in buffers containing low concentrations of MgCl2 and with polyamines33,36, so 0.1 5 we asked whether tRNAAla GGC (UA) can also 0.05 decode a near-cognate codon in such a highfidelity buffer (Supplementary results online). 0 0 Although the Kd of the different ternary com0 1 2 3 4 5 0.001 0.01 0.1 1 10 Ribosome concentration (µM) Time (s) plexes could not be determined even on the cognate codons in this buffer, owing to poor Ala Figure 2 Comparison of tRNAAla GGC (wt) to tRNAGGC (UA) on the GCC cognate and GCA near-cognate stability of the ribosome/ternary complex codons. (a) Equilibrium dissociation curves of catalytically inactive ternary complexes binding to the adduct in the filter-retention assay, we were ribosomal entry site. (b) Time course of GTP hydrolysis at a ribosome concentration of 1.7 mM. able to measure the rates of GTP hydrolysis and (c) Ribosome saturation curve of GTP hydrolysis. (d) Time course of dipeptide formation between Ala (wt) on the GCA codon could peptide bond formation. The apparent rate of fMet-tRNAfMet and Ala-tRNAAla . Dipeptide formation for tRNA GGC GGC GTP hydrolysis determined in high-fidelity not be fit to a simple exponential, so no line was drawn. buffer with 2 mM ribosomes showed that, as in the 10 mM MgCl2 buffer, tRNAAla GGC obtained with the cognate codon, but we could only estimate kpep to be (wt) has a fast rate of hydrolysis on the cognate codon and a much less than 0.05 s–1 (Fig. 2d; see Methods). The slower values of kGTPmax slower rate on the near-cognate GCA codon, whereas tRNAAla GGC (UA) has and kpep have been explained by an induced-fit mechanism in which a similar, fast rate of GTP hydrolysis on both codons (Supplementary the mismatched codon-anticodon interaction causes incorrect adapta- Fig. 1 online). However, unlike in the 10 mM MgCl2 buffer (Fig. 2b), tion of the tRNA to the ribosome and thereby prevents the ribosomal the extent of GTP hydrolysis achieved at long incubation times was conformational changes needed to promote rapid catalysis33–35. only 20%, reflecting the fact that a substantial fraction of the EF-Tu/GTP Fraction GTP hydrolyzed Fraction fMet-Ala dipeptide

kGTP apparent (s–1)

Fraction bound

kGTP apparent (s–1)


Ala tRNAGGC (UA) GCA

Ala tRNAGGC (wt) GCA

Ala tRNAGGC (UA) GCC

Ala (wt) GCC tRNAGGC

Table 1 Thermodynamic and kinetic parameters for different tRNAAla GGC on cognate and near-cognate codons Kd (nM)a Codon

kGTPmax (s–1)b

kpep (s–1)a

A-U

U-A

A-U

U-A

A-U

GGC GCU

2.3 ± 0.40 5.9 ± 0.93

1.0 ± 0.23 1.3 ± 0.24

31 ± 13 23 ± 3.3

27 ± 11 25 ± 3.1

1.7 ± 0.21 1.4 ± 0.17

GCG GCA

B1,000c B1,000c

2.4 ± 0.23

26 ± 18

175 ± 30 210 ± 52

GUC ACC

C-G

C-A

1.8 ± 0.69 1.4 ± 0.16

1.5 ± 0.70 1.5 ± 0.36

1.4 ± 0.27 1.8 ± 0.15

o0.05d

2.1 ± 0.34

o0.05d

3.4 ± 0.64

o0.05d

1.4 ± 0.36 0.49 ± 0.16

o0.05d o0.05d

1.6 ± 0.35 2.8 ± 0.79

o0.05d

U-A

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. value as precise Kd determination exceeded the limits of accurate measurement. dEstimated limit (see Methods).

cEstimated


VOLUME 16

NUMBER 4

APRIL 2009

361

ARTICLES 0.35

Fraction fMet-Ala dipeptide

0.30 0.25 Ala tRNAGGC (wt) GCC

0.20

Ala tRNAGGC (UA) GCC Ala tRNAGGC (wt) GUC

0.15

Ala tRNAGGC (UA) GUC Ala tRNAGGC (wt) ACC

0.10

Ala tRNAGGC (UA) ACC

0.05 0 0.001

0.01

0.1

1

10


Time (s)

Figure 3 Time course of peptide bond formation for tRNAAla GGC (wt) and tRNAAla GGC (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).

did not have an aa-tRNA stably bound. This presumably arises from the poor folding of the transcript tRNA in this buffer. Experiments measuring the rate of peptide bond formation in the high-fidelity buffer showed that tRNAAla GGC (wt) rapidly formed dipeptide in the presence of the cognate codon and not the GCA near-cognate codon, whereas tRNAAla GGC (UA) could efficiently form dipeptide on both codons (Supplementary Fig. 2 online). Because the data collected in the high-fidelity buffer resembled the data collected in the 10 mM MgCl2 buffer in which tRNA folding was not compromised, we used the 10 mM MgCl2 buffer for the remainder of the experiments. To determine whether tRNAAla GGC (UA) is also capable of misreading other near-cognate codons, we used the kpep assay to monitor the rate of 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 nearcognate codons with rates and extents of reaction similar to the cognate GCC codon, whereas tRNAAla GGC (wt) showed slow rates (Fig. 3 and Table 1). This indicates that tRNAAla GGC (UA) has lost its ability to perform accurate decoding on any near-cognate codon. To determine whether misreading was a phenomenon specific to the U32-A38 pair, we tested two other mutant tRNAAla GGC molecules using the kpep assay with the two cognate codons and the three nearcognate codons (Table 1). We mutated tRNAAla GGC to contain two other 32–38-nucleotide pairs: one (C32-A38) is commonly found in bacterial tRNAs other than tRNAAla GGC, and another (C32-G38) is conserved in 22% of known bacterial tRNAAla GGC sequences but is present in only 1.2% of all bacterial tRNAs5. tRNAAla GGC (CA), representing a 32-38 pair present in 52% of all bacterial tRNAs5, was able to read cognate codons normally, but it also rapidly and efficiently misread all three near-cognate codons, similarly to tRNAAla GGC (UA). Ala In contrast, tRNAAla GGC (CG) behaved similarly to tRNAGGC (wt) in effectively reading the cognate codons, but showed slow rates of kpep on near-cognate codons. These results are consistent with the fact that tRNAAla GGC in some bacteria contains the rare C32-G38 pair in place of A32-U38, but none contains the C32-A38 or U32-A38 pairs5. DISCUSSION The A32-U38 pair was originally identified as a sequence element that destabilized binding of tRNAs to the ribosomal A site22. It was

362

VOLUME 16

hypothesized that the purpose of this pair in bacterial tRNAAla GGC and tRNAPro GGG was to off-set the stabilizing effect of their GC-rich codonanticodon pairs so that they would act similarly to other tRNAs when decoding their cognate codons. Although this view may be correct, experiments presented here measuring decoding on near-cognate codons using pre–steady state kinetics make it clear that a crucial role of this base pair is to prevent misreading. When this A32-U38 pair is mutated to a more common 32-38 pair, the resulting ternary complex can not only bind ribosomes somewhat more tightly than the wild-type tRNA but can also stimulate GTP hydrolysis and peptide bond formation equally well on both cognate and near-cognate codons. As these effects would reduce the ability of tRNAAla GGC to distinguish cognate from near-cognate codons in both the initial selection and proofreading steps of decoding, it is likely that the A32-U38 pair was selected to maintain translational accuracy. Once bound to the ribosome, it is astonishing how well tRNAAla GGC (UA) can function on near-cognate codons. Both the maximal rate of GTP hydrolysis and the rate and extent of peptide bond formation are indistinguishable from what is observed for tRNAAla GGC (wt) with its cognate codon. In other words, in these assays the ribosome does not detect that a mismatched codonanticodon complex has formed, and it allows peptide bond formation to occur normally without any proofreading. This ability of tRNAAla GGC (UA) to efficiently read near-cognate codons far exceeds the in vitro effects of error-inducing antibiotics37,38. Although the G24A mutation of E. coli tRNATrp also shows substantial misreading in vitro, it shows no difference in binding to near-cognate codons on the ribosome as a ternary complex and still shows substantially reduced levels of peptide bond formation on mismatched codons, indicating that some proofreading occurs19. tRNAAla GGC (UA) can rapidly undergo GTP hydrolysis and peptide bond formation on the near-cognate GCA codon, even in the low-magnesium, high-fidelity buffer in which the unmodified transcript is not as well folded as in the standard 10 mM MgCl2 buffer. Several of the results presented here have been confirmed using an in vitro translation assay with purified components to prepare oligopeptides from a defined mRNA39. Transcripts of tRNAAla GGC containing one of the more common 32-38 pairs (C-A, U-A or U-U) were effective at incorporating alanine at a GUC (valine) codon, whereas Ala tRNAAla GGC (wt) and tRNAGGC (CG) were not. It is notable that, when a competitor tRNA with an anticodon cognate to the GUC codon was added to the reaction, misincorporation by the tRNAAla GGC mutants was strongly suppressed, presumably because the competitor ternary complex can bind its cognate codon much more tightly than the tRNAAla GGC mutants. Effective competition by correctly matched tRNAs probably also explains why expression of the misreading tRNAAla GGC (CA) in E. coli has only modest effects on bacterial growth39. The 32-38 pair modulates the binding of the Ala-tRNAAla GGC ternary complex to the ribosomal entry site with a trend similar to how it modulates binding of the deacylated tRNAAla GGC to the ribosomal A site22. It is likely that in both cases the explanation of the sequence specificity lies in the structure of the anticodon loop, because the 32-38 pair of tRNAPhe present in high-resolution crystal structures does not seem to interact directly with the 30S ribosome in either complex35,40. As discussed previously41, the A32-U38 pair may form a stable Watson-Crick pair that in turn allows U33 and A37 to form a base pair, resulting in a 3-nucleotide anticodon loop. The observed weaker binding of the wild-type A32-U38 pair would then be due to the energy required to break these base pairs to rearrange the loop into a more open conformation upon codon binding. An

NUMBER 4

APRIL 2009



ARTICLES alternative, less specific explanation for the destabilizing effect of the A32-U38 pair may be that, compared to other nonconserved nucleotide pairs, this particular pair is in some way less able to stabilize the codon-anticodon helix through stacking interactions. A different explanation is required to account for how tRNAAla GGC (UA) can efficiently stimulate its rapid forward rates on near-cognate codons once it is bound to ribosomes. Although no high-resolution X-ray structure of the ternary complex bound to ribosomes is available, medium-resolution cryo-EM structures suggest that the structures of tRNA and possibly EF-Tu are distorted when the ternary complex binds to a cognate codon in the entry site26,42,43. It has been proposed that, when a mismatched codon is present, the altered structure of the codon-anticodon helix prohibits this distortion in the ternary complex, leading to weaker binding and rejection after GTP hydrolysis34. Presumably, tRNA mutations that promote misreading have altered distortability or dynamics that allow them to fit into the ribosome correctly despite the mismatched codon, as described in the original ‘waggle’ theory44. For example, the misreading G24A mutation of tRNATrp lies close to a major site of distortion in the ribosome-bound ternary complexes19,26,43. Although no obvious distortion of this complex is observed in the region of the 32-38 pair, the resolutions of the structures are low. In addition, it is unclear whether each ternary complex will distort identically as a result of their differing tRNA sequences. However, the fact that tRNATrp and tRNAAla GGC use different positions to avoid the same inaccurate decoding phenotype highlights the idea that each tRNA is tuned by different elements. Although it seems that one important selective pressure on tRNA sequences seems to be to perform equivalently in translating their cognate codons, experiments presented here highlight the fact that tRNA 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 decoding accuracy, it is uncertain whether this will always be the case. Mutating other tRNAAla GGC consensus elements does not seem to greatly affect ribosome binding22, but these mutants’ ability to misread remains to be tested. It is possible that the extensive and complex tRNA sequence requirements associated with each anticodon reflect the apparent need for tRNA to show a characteristic deformability to ensure accurate decoding. Other elements, such as post-transcriptional modifications and the identity of the amino acid, are likely to be important for how the aa-tRNA functions on the ribosome, similarly to how the nature of the 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 a slower rate of dipeptide formation if esterified to tRNAPhe rather than tRNAPro (ref. 46). If this is the case, mutations of tRNA consensus elements may not always directly affect aa-tRNA function on cognate codons but may instead affect their ability to avoid translating nearcognate codons. METHODS Materials. 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¢-GGCAAGGAGGUAAAAAUGXXXGCACGU3¢, 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 tRNAAla GGC was transcribed from templates generated by primer extension of overlapping DNA


VOLUME 16

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 determined 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 singleexponential 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 Ala tRNAAla GGC (wt) or tRNAGGC (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, tRNAAla GGC shows a rate of rejection that is unusually Ala (refs. 19,31,33). slow compared those of to tRNAPhe, tRNATrp and tRNAUGC 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 experiments 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 & Molecular Biology website.

NUMBER 4

APRIL 2009

363

ARTICLES ACKNOWLEDGMENTS We thank H. Suga and H. Murakami for discussions and for sharing their unpublished data. M.O. is supported by the ‘‘Homing’’ grant from the Foundation for Polish Science. This work was funded by the US National Institutes of Health grant GM037552 (to O.C.U.) AUTHOR CONTRIBUTIONS S.L. and M.O. performed the experiments; S.L., M.O. and O.C.U. contributed to the 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 Escherichia coli 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 transfer RNAs 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 the ribosome. Mol. Cell 31, 114–123 (2008). 4. Saks, M.E. & Conery, J.S. Anticodon-dependent conservation of bacterial tRNA gene sequences. RNA 13, 651–660 (2007). 5. Sprinzl, M., Horn, C., Brown, M., Ioudovitch, A. & Steinberg, S. Compilation of tRNA sequences 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 for binding by EF-Tu. RNA 13, 835–840 (2007). 7. Yarus, M., Cline, S., Raftery, L., Wier, P. & Bradley, D. The translational efficiency of tRNA 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 tRNA backbone-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 of unnatural amino acids into proteins by nonsense suppression. J. Biol. Chem. 271, 23169–23175 (1996). 10. Kru¨ger, M.K., Pedersen, S., Hagervall, T.G. & Sørensen, M.A. The modification of the wobble 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-5oxyacetic acid in tRNAPro cmo5UGG promotes reading of all four proline codons in vivo. RNA 10, 1662–1673 (2004). 12. Claesson, C. et al. Glycine codon discrimination and the nucleotide in position 32 of the anticodon loop. J. Mol. Biol. 247, 191–196 (1995). 13. Tsai, F. & Curran, J.F. tRNAGln 2 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 tRNAPro GGG of Salmonella 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 of reading 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 of tRNAGly reduce T4 gene 60 translational bypassing efficiency. EMBO J. 18, 2 2886–2896 (1999). 19. Cochella, L. & Green, R. An active role for tRNA in decoding beyond codon:anticodon pairing. Science 308, 1178–1180 (2005). 20. Mims, B.H., Prather, N.E. & Murgola, E.J. Isolation and nucleotide sequence analysis of 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 GU in tRNAAla by alanyl-tRNA synthetase. Science 271, 195–197 (1996).

364

VOLUME 16

22. Olejniczak, M., Dale, T., Fahlman, R.P. & Uhlenbeck, O.C. Idiosyncratic tuning of tRNAs 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 elongation factor 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 tRNA modification 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. Science 292, 883–896 (2001). 26. Valle, M. et al. Cryo-EM reveals an active role for aminoacyl-tRNA in the accommodation process. EMBO J. 21, 3557–3567 (2002). 27. Sampson, J.R. & Uhlenbeck, O.C. Biochemical and physical characterization of an unmodified 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 conserved modified 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 an unmodified tRNA: implications for RNA folding. Biochemistry 37, 16349–16359 (1998). 30. Serebrov, V., Vassilenko, K., Kholod, N., Gross, H.J. & Kisselev, L. Mg2+ binding and structural stability of mature and in vitro synthesized unmodified Escherichia coli tRNAPhe. Nucleic Acids Res. 26, 2723–2728 (1998). 31. Kothe, U. & Rodnina, M.V. Codon reading by tRNAAla with modified uridine in the wobble position. Mol. Cell 25, 167–174 (2007). 32. Gromadski, K.B., Daviter, T. & Rodnina, M.V. A uniform response to mismatches in codon-anticodon complexes ensures ribosomal fidelity. Mol. Cell 21, 369–377 (2006). 33. Pape, T., Wintermeyer, W. & Rodnina, M. Induced fit in initial selection and proofreading of aminoacyl-tRNA on the ribosome. EMBO J. 18, 3800–3807 (1999). 34. Sanbonmatsu, K.Y. Alignment/misalignment hypothesis for tRNA selection by the ribosome. Biochimie 88, 1075–1089 (2006). 35. Ogle, J.M., Murphy, F.V., Tarry, M.J. & Ramakrishnan, V. Selection of tRNA by the ribosome 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 discrimination on the ribosome. Mol. Cell 13, 191–200 (2004). 37. Pape, T., Wintermeyer, W. & Rodnina, M.V. Complete kinetic mechanism of elongation factor 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 decoding region 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 tRNAAla GGC 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 their anticodon 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 in the codon-recognition complex. Nat. Struct. Biol. 9, 849–854 (2002). 43. Li, W. et al. Recognition of aminoacyl-tRNA: a common molecular mechanism revealed by cryo-EM. EMBO J. 27, 3322–3331 (2008). 44. So¨ll, 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-tRNAs to 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-alkylamino acids 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 for assaying aminoacylation and peptide bond formation. Methods 44, 74–80 (2008).

NUMBER 4

APRIL 2009


ARTICLES

A feedback regulatory loop involving microRNA-9 and nuclear receptor TLX in neural stem cell fate determination © 2009 Nature America, Inc. All rights reserved.

Chunnian 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 specifically in neurogenic areas of the brain and may be involved in neural stem cell self-renewal and differentiation. We showed previously that the nuclear receptor TLX is an essential regulator of neural stem cell self-renewal. Here we show that miR-9 suppresses TLX expression to negatively regulate neural stem cell proliferation and accelerate neural differentiation. Introducing a TLX expression vector that is not prone to miR-9 regulation rescued miR-9–induced proliferation deficiency and inhibited precocious differentiation. In utero electroporation of miR-9 in embryonic brains led to premature differentiation and outward migration of the transfected neural stem cells. Moreover, TLX represses expression of the miR-9 pri-miRNA. By forming a negative regulatory loop 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 understand the molecular mechanisms underlying stem cell self-renewal and differentiation. Neural stem cells are a subset of undifferentiated precursors that retain the ability to proliferate and self-renew and have the capacity to give rise to both neuronal and glial lineages1–4. Although the functional properties of neural stem cells have been studied extensively, how their self-renewal and differentiation is regulated is not completely understood. Accumulating evidence indicates that both transcriptional and post-transcriptional regulation mechanisms are important for regulating genes essential for neural stem cell self-renewal and neurogenesis. miRNAs are a recently identified large family of 20–22-nucleotide noncoding RNAs involved in numerous cellular processes, including development, proliferation and differentiation5,6. Thus, miRNAs are potentially key post-transcriptional regulators in stem cell self-renewal and differentiation. Distinct sets of miRNAs have been shown to be specifically expressed in embryonic stem cells7,8. Loss of Dicer1 causes embryonic lethality and loss of stem cell populations9,10. Argonaute family members, key components of the RNA-induced silencing complex (RISC), are required for maintaining germline stem cells in various species11. These observations together support a role for miRNAs in stem cell biology. Several brain-specific miRNAs have recently been identified. Among these miRNAs, miR-9 is expressed specifically in neurogenic regions of the brain during neural development and in adulthood12–15. Whether miR-9 has a role in neural stem cell self-renewal and differentiation remains to be determined.

We have shown that TLX is an essential regulator of neural stem cell self-renewal16. TLX maintains adult neural stem cells in an undifferentiated and self-renewable state, in part through transcriptional repression of its downstream target genes, encoding p21 and phosphatase and tensin homolog (Pten), by complexing with histone deacetylases17. Recently, TLX-positive neural stem cells have been shown to have a role in spatial learning and memory18. In addition to its function in adult brains, TLX is also involved in neural development by regulating cell-cycle progression in neural stem cells of the developing brain19–21. TLX is therefore a key regulator that acts to establish the undifferentiated and self-renewable state of neural stem cells, although aspects of its regulation are enigmatic. Here we demonstrate that miR-9 suppresses TLX expression through the 3¢ untranslated region (UTR) of TLX mRNA, which, in turn, regulates neural stem cell proliferation and differentiation. Increased expression of miR-9 led to reduced mouse neural stem cell proliferation and accelerated neural differentiation, whereas antisense knockdown of miR-9 led to increased neural stem cell proliferation. Introducing a TLX expression vector lacking the endogenous TLX 3¢ UTR rescued proliferation deficiency induced by miR-9 overexpression and reversed miR-9–promoted precocious differentiation. These results suggest that miR-9 regulates neural stem cell proliferation and differentiation, at least in part, through targeting TLX mRNA via its 3¢ UTR. In utero electroporation of miR-9 into ventricular zone neural stem cells in embryonic mouse brains triggered premature differentiation and outward migration of the transfected cells, similar to that induced by electroporation of the TLX

Department of Neurosciences, Center for Gene Expression and Drug Discovery, Beckman Research Institute of City of Hope, Duarte, California, USA. Correspondence should be addressed to Y.S. ([email protected]). Received 1 April 2008; accepted 17 February 2009; published online 29 March 2009; doi:10.1038/nsmb.1576


VOLUME 16

NUMBER 4

APRIL 2009

365

ARTICLES

1

c Relative luciferase activity

100

d5

d3

4.9

1.7 0 50

b

d7

5.7

d5

GAPDH

d

80

e 100

100

80

80

60

40

20

60

40

20

0

miR-9 f mutant miR-9

WT 3′ UTR Mut 3′ UTR

60

40

20

0 1

miR-9

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.

d7

U6 relative miR-9 levels

25 100 (nM)

C

d3

TLX

0


d0

miR-9

Relative luciferase activity

d0

Relative luciferase activity

a

g

2

C

miR-9 miR-9 mutant

miR-9 miR-9 mutant

TLX

TLX

GAPDH

GAPDH

small interfering RNA (siRNA)21. Furthermore, TLX binds to the 3¢ genomic sequences of miR-9-1 to inhibit its expression. MiR-9 and TLX thus form a feedback loop to regulate the switch between neural stem cell proliferation and differentiation. RESULTS MiR-9 represses TLX expression by targeting its 3¢ UTR We hypothesized that TLX is targeted by miRNAs to regulate its expression. Using the miRanda (http://www.microrna.org)22 and TargetScan (http://genes.mit.edu/targetscan)23 algorithms, miR-9 was predicted to have a target site in the TLX 3¢ UTR. This target site is conserved in human, mouse, dog and chicken TLX (Supplementary Fig. 1a online). As TLX is specifically expressed in vertebrate forebrains and is an essential regulator of neural stem cell self-renewal, we first asked whether the candidate TLX-targeting miR-9 is expressed in the brain and, specifically, whether this miRNA is expressed in neural stem cells or in their differentiated progeny. Northern blotting revealed that miR-9 is expressed specifically in the brain (Supplementary Fig. 1b), consistent with previous reports12,13,24. The size of the miRNA was as expected for the mature miR-9 (22 bp). MiR-9 is also expressed in neural stem cells (d0, Fig. 1a). Notably, the expression of miR-9 is upregulated during neural differentiation (Fig. 1a), in contrast to the reduced expression of TLX (Fig. 1b). To validate whether miR-9 targets TLX, we made a luciferase reporter construct with the mouse 1.4-kb TLX 3¢ UTR containing the predicted miR-9 target site and flanking sequences inserted into the 3¢ UTR of a Renilla luciferase reporter gene in a siCHECK vector. Increasing amounts of RNA duplexes of mature miR-9 were transfected into human embryonic kidney HEK 293 cells along with the corresponding reporter gene. We observed dose-dependent repression of the reporter gene in miR-9–treated cells (Fig. 1c). In addition to the synthetic RNA duplexes, miR-9 was also expressed using a microRNA expression vector, MDH1-PGK-GFP2 (ref. 25), into which we had cloned a 489-nucleotide (nt) miR-9-1 genomic sequence including the 89-nt miR-9 hairpin precursor and the 200-nt genomic sequence flanking each side of the precursor. The miR-9 expression vector repressed luciferase reporter gene activity (Fig. 1d), similarly to the

366

VOLUME 16

miR-9 RNA duplexes (Fig. 1c), suggesting that miR-9 suppresses TLX expression through its 3¢ UTR. Base-pairing between the miRNA seed sequence and its target gene is needed for miRNA-mediated repression of the target mRNA26,27. To test whether the predicted miR-9 target site in TLX 3¢ UTR is crucial for repression of TLX expression by miR-9, we introduced point mutations disrupting complementarity in the predicted miR-9 target site into TLX 3¢ UTR in the luciferase reporter construct. Mutation of the miR-9 target site abolished the repression by miR-9 (Fig. 1e). A miR-9 mutant with compensatory mutations repressed a luciferase reporter gene with a mutated TLX 3¢ UTR (Fig. 1e). These results strongly suggest that miR-9 represses TLX expression through the predicted target site in TLX 3¢ UTR. Next we tested whether miR-9 targets TLX in neural stem cells. We transfected mature miR-9 RNA duplexes into neural stem cells and 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 downregulates TLX expression. MiR-9 regulates neural stem cell fate determination To examine whether miR-9 regulates neural stem cell proliferation, we transfected neural stem cells with increasing concentrations of miR-9 RNA duplexes and measured cell proliferation by 5-bromodeoxy uridine (BrdU) labeling of dividing cells. Transfection of miR-9 led to dose-dependent inhibition of cell proliferation (Fig. 2a,b) with a minimal effect on cell death (Supplementary Fig. 1c,d). Accordingly, reduced expression of TLX and increased expression of p21, encoded by a target gene that is repressed by TLX17, was observed in miR-9– transfected cells in a dose-dependent manner (Fig. 2c). This gene expression profile is consistent with miR-9–induced inhibition of neural stem cell proliferation (Fig. 2a,b), suggesting that miR-9 negatively regulates neural stem cell proliferation, presumably through downregulation of TLX signaling. To determine whether overexpression of miR-9 regulates neural stem cell differentiation, we transfected neural stem cells with miR-9 RNA duplexes and cultured them under varying conditions. As complete withdrawal of epidermal growth factor (EGF) and fibroblast growth factor (FGF) from the medium led to considerable cell death (data not shown), we cultured neural stem cells in N2-supplemented media with low EGF and FGF concentrations (1 ng ml–1), which allowed cell viability with minimal cell proliferation. Over a 7-d time course, no difference in neuronal and glial differentiation could be

NUMBER 4

APRIL 2009


ARTICLES

b

Merged

BrdU+ cells (%)

50 nM

40 35 30 25 20 15 10 5 0

c

*

*

0 50 100 200 miR-9 (nM) miR-9 (nM)

0

100 nM

*

50 100 200 TLX GAPDH miR-9 (nM) 50 100 200

200 nM

0

TLX p21

30 Control

25 20 15 10

miR-9

GFAP+ cells (%)

Control

e

5 0

1

2

25

*

20 15 10 5 0

1

2

a

TLX∆3′ UTR – +

c

C –

+

miR-9 TLX∆3′ UTR

35

TLX total

30

Actin

b

BrdU

Merged


VOLUME 16

NUMBER 4

25 20 15 10

+miR-9

0

d 25 GFAP+ cells (%)

–miR-9

20

TLX∆3′ UTR

*

5

+miR-9

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¢s t-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.

–miR-9 +miR-9

**

C

detected between control RNA and miR-9–transfected cells (data not shown), suggesting that overexpression of miR-9 alone is not sufficient to trigger neuronal or glial differentiation. However, when neural stem cells were induced for differentiation using forskolin or retinoic acid, transfection of miR-9 promoted both astroglial and neuronal differentiation, leading to an increase in the percentage of glial fibrillary acidic protein (GFAP)-positive astrocytes and Tuj1-positive neurons at day 3 of differentiation (Fig. 2d,e), mimicking day 5 of differentiation in control cells (data not shown). These results indicate that miR-9 accelerates differentiation of neural stem cells that have been primed for differentiation. To determine whether the effect of miR-9 transfection on neural stem cell proliferation and differentiation is mediated through TLX, we stably transduced neural stem cells with a vector expressing a version of TLX lacking its 3¢ UTR (TLXD3¢ UTR). Transfection of miR-9 had no effect on the expression of TLXD3¢ UTR, although miR-9 downregulated endogenous TLX expression levels (Fig. 3a). Expression of

TLXD3¢ UTR led to a 1.24-fold increase in neural stem cell proliferation. Co-transfection of TLXD3¢ UTR and miR-9 substantially reversed the proliferative deficiency induced by miR-9 (Fig. 3b,c). Furthermore, whereas transfection of miR-9 increased astroglial differentiation in control neural stem cells, we detected no appreciable increase in astrocyte differentiation in TLXD3¢ UTR-transduced cells upon miR-9 treatment (Fig. 3d). These results strongly suggest that miR-9 regulates neural stem cell proliferation and differentiation, at least in part, by inhibiting TLX expression through its 3¢ UTR. We further investigated the role of miR-9 in neural stem cell proliferation using 2¢-O-methyl antisense RNA oligonucleotides as small RNA inhibitors28,29. We synthesized 2¢-O-methyl antisense oligonucleotide against miR-9 and transfected it into neural stem cells, with 2¢-O-methyl antisense oligonucleotide against green fluorescent protein (GFP) included as a negative control. Treatment of

BrdU+ cells (%)

*

30

Tuj1+ cells (%)

d

miR-9


Actin

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 BrdUpositive (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 neuronal differentiation. 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.

–miR-9

BrdU

0 nM :miR-9

a

12 34 C TLX∆3′ UTR –miR-9 +miR-9

*

15

10

5

0 C

APRIL 2009

TLX∆3′ UTR

367

ARTICLES

c miR-9

BrdU

GAPDH

miR-9 antisense

miR-9 antisense TLX p21

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.

d 70

U6

bC

Merged

*

60 BrdU+ cells (%)

C miR-9 antisense

Control

a

50 40 30 20 10 0

antisense oligonucleotides against miR-9 led to substantial knockdown of miR-9 mature forms (Fig. 4a). The expression of TLX was upregulated in miR-9 antisense RNA-treated neural stem cells, along with decreased expression of p21 (Fig. 4b). BrdU-labeling analysis revealed that knockdown of miR-9 led to an increase in cell proliferation (1.37-fold, Fig. 4c,d), consistent with enhanced cell proliferation in TLXD3¢ UTR–transduced neural stem cells (Fig. 3b,c).

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

miR-9 stimulates neural differentiation in the brain Regulation of miR-9 gene expression by TLX During development, neural stem cells reside in the ventricular zone Three genes, miR-9-1, miR-9-2 and miR-9-3, encode miR-9 in and migrate out into the cortical plate upon differentiation. To the mouse genome. Both miR-9-1 and miR-9-2 are expressed in determine whether miR-9 influences neural stem cell proliferation mammalian brains14, whereas miR-9-3 expression has not been and differentiation in vivo, we introduced miR-9 RNA duplexes detected in vertebrate brains31. To determine whether miR-9 gene into neural stem cells in the ventricular zone of mouse brains at expression is affected by TLX expression, we performed reverseembryonic day E13.5 by in utero electroporation. Electroporated transcription PCR (RT-PCR) to assess the expression levels of miRbrains were analyzed at E15.5. Cells that had taken up miR-9 were 9-1 and miR-9-2 pri-miRNAs in brains of wild-type and TLX-null labeled green by coexpression of GFP. Transfection of miR-9 led to a mice. Notably, expression of both miR-9-1 and miR-9-2 pri-miRNAs marked decrease of cells that were positively labeled for Ki67, a proliferative marker * a 1.2 b c 20 (Fig. 5a), and a substantial increase in the number of cells that migrated from the ven1 15 tricular zone to the cortical plate (Fig. 5b,c), 0.8 suggesting that miR-9 negatively regulates * neural stem cell proliferation and accelerates 10 0.6 neural differentiation. Immunostaining revealed that transfection 0.4 of miR-9 led to decreased TLX expression 5 (Fig. 5d and Supplementary Fig. 2 online). 0.2 The miR-9–transfected cells that migrated to 0 0 the cortical plate lost the neural progenitor C miR-9 C miR-9 marker nestin (Fig. 5d). Instead, these cells d GFP TLX Merge GFP Nestin Merge GFP DCX Merge expressed the neuronal marker double cortin (DCX, Fig. 5d), indicating neuronal differControl -GFP entiation. In contrast, the control small RNA–transfected cells that remained in the ventricular zone were nestin-positive and miR-9 -GFP lacked DCX expression (Fig. 5d). These results are similar to those obtained from in utero electroporation of a TLX siRNA21. Figure 5 In utero electroporation of miR-9 in embryonic neural stem cells. (a) In utero electroporation Furthermore, in utero electroporation of of miR-9 decreased cell proliferation in the ventricular zone (VZ) of embryonic brains. Proliferating cells TLXD3¢ UTR along with miR-9 rescued the were labeled by Ki67. Percentage of Ki67-positive cells relative to GFP-positive cells (Ki67+GFP+/ precocious migration induced by miR-9 GFP+) in miR-9–electroporated brains was calculated and normalized with the percentage of Ki67transfection (Supplementary Fig. 3 online). positive cells in control RNA (C)–electroporated brains. Error bars indicate s.d. * indicates P ¼ 0.002 These results strongly suggest that miR-9 by the Student’s t-test. (b) Electroporation of miR-9 led to precocious outward cell migration. regulates neural stem cell differentiation 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 through targeting TLX expression in vivo. right images show 20 magnification. (c) Quantification of control RNA (C) and miR-9–electroporated In addition to neuronal differentiation, we cells (GFP+ cells) that migrated to the cortical plate (CP). Error bars indicate s.d. of the mean. also examined whether miR-9 has a role in * indicates P ¼ 0.02 by the Student’s t-test. (d) Immunostaining of cells from control-GFP or glial differentiation in vivo. Gliogenesis occurs miR-9-GFP–electroporated brains.

368

VOLUME 16

GFP+ cells in CP (%)

Fold difference in % of Ki67+GFP+/GFP+ cells in VZ


C miR-9 antisense

NUMBER 4

APRIL 2009


ARTICLES

a

d

–/– WT TLX

Input IgG1 IgG2 AcH3 H3K4me3 HDAC5 H3K9me3

miR-9-1

TLX-1/2

miR-9-2

b

c

1340 RE miR-9-1 TLX-1/2 1365/1799 1840 5338 2211 RE TLX-3 miR-9-1 5338 2419 2545 5411 RE miR-9-1 TLX-4 5338 5511 5750

Input IgG anti-TLX TLX-1/2 TLX-3


TLX-4

TLX-3

e 3,500 Relative luciferase activity

GAPDH

3,000

*

2,500 2,000 1,500 1,000 500

0 TLX – + miR-9-1-luc WT

– + MT

was upregulated in brains of TLX-null mice (Fig. 6a), suggesting that expression of miR-9 precursors can be repressed by TLX. Sequence analysis revealed several consensus TLX binding sites in the flanking regions of the miR-9-1 (Fig. 6b) and miR-9-2 genes (data not shown). We used chromatin immunoprecipitation (ChIP) to further explore whether TLX regulates miR-9 expression by direct binding to miR-9 genomic sequences. MiR-9-1 was chosen for this analysis because it is induced more substantially in TLX-null brains (Fig. 6a). We designed three pairs of primers for ChIP analysis, with one 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 sites 3 (TLX-3) and site 4 (TLX-4) individually. ChIP assays revealed that TLX 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 previously characterized REST binding site32,33, whereas no binding was detected on TLX-4, which is downstream of the REST binding site (Fig. 6c). Consistent with TLX binding to miR-9-1 genomic sequences, we also detected the TLX-interacting transcriptional co-repressor HDAC5 on the TLX binding sites, TLX-1/2 and TLX-3, in the miR-9-1 genomic locus (Fig. 6d). Furthermore, these sites are associated with the repressive chromatin marker, trimethylated histone H3 lysine 9 (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, we cloned the 1.2-kb miR-9-1 downstream genomic sequence that contains the consensus TLX binding sites, TLX-1/2 and TLX-3, and inserted it downstream of a Renilla luciferase reporter gene in the siCHECK vector. Co-transfection of TLX with the reporter gene in neural stem cells led to a 2.2-fold repression of the reporter activity (Fig. 6e). Mutation of the TLX binding sites substantially relieved the repression mediated by TLX (Fig. 6e). These results together suggest that TLX represses miR-9 expression by binding to the consensus TLX binding sites in the 3¢ genomic sequence of miR-9-1. DISCUSSION We show here that microRNA miR-9 and nuclear receptor TLX form a feedback regulatory loop to regulate neural stem cell proliferation and differentiation. TLX is highly expressed in neural stem cells but is repressed upon differentiation16; in contrast, the level of the miR-9 mature form is increased upon differentiation13,24,36. The temporal relationship between miR-9 and TLX expression resembles that between miR-124 and its target genes lamc1, itgb1 and REST 32,37,38. In both instances, when miRNA expression is low, the targets tend to be expressed at high levels. Conversely, the expression of these targets is downregulated as the miRNAs accumulate. These data support the


VOLUME 16

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.

hypothesis that miRNAs induced during differentiation function to ensure proper cell fate transitions by suppressing leftover stem cell maintenance transcripts in stem cells39. This study demonstrates that miR-9 has an important role in neural stem cell proliferation and differentiation and that TLX is a key target of miR-9 in neural stem cells. Every miRNA could have multiple target genes23,26 and, indeed, several target genes have been predicted and some tested for miR-9, including those encoding the transcription factors REST, FoxG1, Senseless and Hairy/E(spl), and components of the FGF signaling pathway32,33,40–42. One of the questions addressed here is whether the cell proliferation and differentiation effect mediated by miR-9 in neural stem cells is directly related to repression of TLX expression. Transfection of miR-9 into neural stem cells that are stably transduced with a TLX-expressing vector lacking the 3¢ UTR showed that such ectopically expressed TLX rescued the proliferative deficiency induced by overexpression of miR-9 and compromised miR-9–induced precocious differentiation. The result of this rescue experiment suggests that miR-9 regulates neural stem cell proliferation and differentiation through repression of TLX expression. Although TLX is an important target gene of miR-9, other targets may also have a role in miR-9 function in neural stem cells. Recent evidence suggests that miRNAs often act as fine-tuning devices rather than as primary gene regulators43. Consistent with this concept, we failed to induce neural differentiation by overexpressing miR-9 alone in cultured neural stem cells. Instead, we detected an accelerated differentiation program when differentiation of neural stem cells was induced in culture or in E13.5 brains, where active endogenous neurogenesis occurs. Furthermore, it has been suggested that inhibiting a miRNA may not generate a strong or even detectable phenotype, as expression of its target genes are already repressed at the transcriptional level, whereas overexpressing a miRNA in cells where its target genes are highly expressed may render the action of the miRNA more detectable37. In accordance with this theory, we failed to detect a change in cell differentiation in miR-9 antisense RNA–treated neural stem cells (data not shown). However, we were able to detect a precocious differentiation program upon miR-9 overexpression in neural stem cells that were primed for differentiation. In addition to being a direct target of miR-9, TLX also transcriptionally inhibits miR-9 genes, suggesting a negative feedback loop between TLX and miR-9 that, perhaps, allow rapid transition from neural stem cells to differentiated cells. In neural stem cells, TLX is expressed at relatively high levels. During differentiation, as TLX levels decrease, miR-9 expression accumulates. In turn, miR-9 suppresses TLX expression post-transcriptionally to further promote neural differentiation. This regulatory loop may represent a key mechanism to sense the intricate balance between cell proliferation and

NUMBER 4

APRIL 2009

369

ARTICLES differentiation and to confer cell fate determination in a timely manner. Overall, our study suggests that the brain-specific miRNA miR-9 has a key role in vertebrate brain development. miR-9 provides a novel strategy to control neural stem cell fate determination by forming a feedback regulatory loop with TLX. METHODS


Neural 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 sitedirected 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 antibody16. 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 northern 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¢.

370

VOLUME 16

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 Signaling 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 & Molecular Biology website. ACKNOWLEDGMENTS We 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 (Massachusetts Institute of Technology) for providing the microRNA expression vector MDH1PGK-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, National Institute of Neurological Disorders and Stroke grant R01 NS059546 (to Y.S.). AUTHOR CONTRIBUTIONS Y.S., C.Z. and G.S. designed the project; C.Z., G.S. and S.L. performed the experiments; 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. Multipotent progenitor 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).

NUMBER 4

APRIL 2009



ARTICLES 10. Wienholds, E., Koudijs, M.J., van Eeden, F.J., Cuppen, E. & Plasterk, R.H. The microRNA-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: tentacles that reach into RNAi, developmental control, stem cell maintenance, and tumorigenesis. 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 microRNA array 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 mammalian microRNA 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 the developing 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 neural stem cells. Nature 427, 78–83 (2004). 17. Sun, G., Yu, R.T., Evans, R.M. & Shi, Y. Orphan nuclear receptor TLX recruits histone deacetylases 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-positive neural 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 of lateral telencephalic progenitor domains. J. Neurosci. 23, 10568–10576 (2003). 21. Li, W. et al. Nuclear receptor TLX regulates cell cycle progression in neural stem cells of 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 neuronal differentiation. Genome Biol. 5, R13 (2004). 25. Chen, C.Z., Li, L., Lodish, H.F. & Bartel, D.P. MicroRNAs modulate hematopoietic lineage differentiation. Science 303, 83–86 (2004). 26. Lewis, B.P., Burge, C.B. & Bartel, D.P. Conserved seed pairing, often flanked by adenosines, indicates that thousands of human genes are microRNA targets. Cell 120, 15–20 (2005).


VOLUME 16

27. Brennecke, J., Stark, A., Russell, R.B. & Cohen, S.M. Principles of microRNA-target recognition. PLoS Biol. 3, e85 (2005). 28. Meister, G., Landthaler, M., Dorsett, Y. & Tuschl, T. Sequence-specific inhibition of microRNA- and siRNA-induced RNA silencing. RNA 10, 544–550 (2004). 29. Hutvagner, G., Simard, M.J., Mello, C.C. & Zamore, P.D. Sequence-specific inhibition of small RNA function. PLoS Biol. 2, e98 (2004). 30. Qian, X. et al. Timing of CNS cell generation: a programmed sequence of neuron and glial 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. Science 317, 1220–1224 (2007). 32. Conaco, C., Otto, S., Han, J.J. & Mandel, G. Reciprocal actions of REST and a microRNA promote neuronal identity. Proc. Natl. Acad. Sci. USA 103, 2422–2427 (2006). 33. Wu, J. & Xie, X. Comparative sequence analysis reveals an intricate network among REST, 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 microRNAs modulate 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 developing neural tube. Genes Dev. 21, 531–536 (2007). 38. Visvanathan, J., Lee, S., Lee, B., Lee, J.W. & Lee, S.K. The microRNA miR-124 antagonizes 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 mRNA repression and evolution. Science 310, 1817–1821 (2005). 40. Bredenkamp, N., Seoighe, C. & Illing, N. Comparative evolutionary analysis of the FoxG1 transcription factor from diverse vertebrates identifies conserved recognition sites 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 of sensory organ precursors in Drosophila. Genes Dev. 20, 2793–2805 (2006). 42. Leucht, C. et al. MicroRNA-9 directs late organizer activity of the midbrain-hindbrain boundary. Nat. Neurosci. 11, 641–648 (2008). 43. Hornstein, E. & Shomron, N. Canalization of development by microRNAs. Nat. Genet. 38, S20–S24 (2006).

NUMBER 4

APRIL 2009

371

ARTICLES

TRF2 functions as a protein hub and regulates telomere maintenance by recognizing specific peptide motifs


Hyeung 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 TRF1 and TRF2 associate with a collection of molecules necessary for telomere maintenance and cell-cycle progression. However, the specificity and the mechanisms by which TRF2 communicates with different signaling pathways remain largely unknown. Using oriented peptide libraries, we demonstrate that the TRFH domain of human TRF2 recognizes [Y/F]XL peptides with the consensus motif YYHKYRLSPL. Disrupting the interactions between the TRF2 TRFH domain and its targets resulted in telomeric DNAdamage responses. Furthermore, our genome-wide target analysis revealed phosphatase nuclear targeting subunit (PNUTS) and microcephalin 1 (MCPH1) as previously unreported telomere-associated proteins that directly interact with TRF2 via the [Y/F]XL motif. PNUTS and MCPH1 can regulate telomere length and the telomeric DNA-damage response, respectively. Our findings indicate that an array of TRF2 molecules functions as a protein hub and regulates telomeres by recruiting different signaling molecules via a linear sequence code.

Telomere dysfunction has been implicated in cancer and aging1–9. Mammalian chromosomal ends contain long tracts of duplex telomere repeats with 3¢ single-stranded G overhangs10. The telosome/shelterin complex, which includes TRF1, TRF2, TRF1-interacting nuclear factor 2 (TIN2), RAP1 (also known as TERF2IP), TPP1 (for TINT1/PIP1/PTOP) and protection of telomeres 1 (POT1), helps to maintain telomere integrity by protecting the telomeres from chromosomal abnormalities and DNA-damage responses due to telomere replication, recombination and erosion11,12. Both TRF1 and TRF2 contain a TRFH domain, which mediates homodimerization, and a myb domain, which directly binds the telomeric double-stranded DNA13,14. In addition to telomeric DNA, TRF1 and TRF2 also associate with various proteins involved in telosome assembly, telomere-length regulation, DNA replication, repair, end joining, recombination and cell-cycle control11,12,15. Consistent with the essential roles of TRF1 and TRF2, homozygous inactivation of either gene resulted in early embryonic lethality in mice16,17. In cultured cells, impairment of TRF2 function (for example, dominant negative expression of TRF2DBDM, which lacks the basic and myb domains) led to DNA-damage responses18,19, telomere loop deletion20 or anaphase bridging21. However, the mechanisms of TRF2-mediated interaction and the direct targets of TRF2 remain elusive. One known target of TRF2 in telomere maintenance is the exonuclease Apollo22,23. Indeed, biochemical and structural analyses revealed 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 to harbor distinct binding specificities, suggesting differential recruitment of distinct proteins by different TRFH domains. However, the specific determinant for TRF2TRFH recognition and the identities of other TRF2 targets remain unknown. Here we investigated the specificity of TRF2TRFH and demonstrated that TRF2TRFH is a protein domain that recognizes specific peptides with the [Y/F]XL motif. Through proteomic analyses, we identified several [Y/F]XL motif–containing proteins that can directly interact with TRF2 and mediate telomere-length control and end protection. Our results indicate that an array of TRF2 molecules at the telomeres serves as a protein hub for telomeric signaling. RESULTS Determining the binding specificity of TRF2TRFH We reasoned that TRF2TRFH might represent a modular proteinprotein interaction domain whose specificity could be studied using the oriented peptide library technique25. We therefore synthesized an oriented 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 its interaction with TRFH24, we partially fixed the corresponding positions (P0, P+2 and P+4) in the library (as indicated by square brackets) (Fig. 1a). Peptide mixtures that specifically associated with glutathione S-transferase (GST)-TRF2TRFH fusion proteins were isolated and sequenced. Among the four aromatic residues partially fixed

1Verna

and Marrs McLean Department of Biochemistry and Molecular Biology, Baylor College of Medicine, Houston, Texas 77030, USA. 2Department of System Biology, The University of Texas M.D. Anderson Cancer Center, Houston, Texas 77054, USA. 3These authors contributed equally to this work. Correspondence should be addressed to Z.S. ([email protected]). Received 10 September 2008; accepted 10 February 2009; published online 15 March 2009; doi:10.1038/nsmb.1575

372

VOLUME 16

NUMBER 4

APRIL 2009


ARTICLES

a

Oriented peptide library

–4 –3 –2 –1 0 +1 +2 +3 +4 +5

Apollo

R G L A L K Y L

L

Library

K G X X X X F X I X P X N Y L H V W

+

GST-TRFH domain beads

c 160

T P V N FP value

Position

Bound peptides

Peptide sequencing and decoding

120 TRF2-YRL Kd = 190nM

80 40

Selectivity

3 2.5 2 1.5 1 0.5 0

Selectivity

3 2.5 2 1.5 1 0.5 0

Selectivity


b

3 2.5 2 1.5 1 0.5 0

P–4

KRHDENQS TGP A I L VMF YW

P–1


P+2


3 2.5 2 1.5 1 0.5 0 3 2.5 2 1.5 1 0.5 0 3 2.5 2 1.5 1 0.5 0

P–3


P0


P+3


3 2.5 2 1.5 1 0.5 0 3.5 3 2.5 2 1.5 1 0.5 0 3 2.5 2 1.5 1 0.5 0

0 P–2

KRHDENQS TGPA I L VMF YW

P+1


P+5


0

5 10 15 20 25 30 35 40 TRF2 concentration (µM)

d Peptide

Protein

TRF2

YRL:

AKGYYHKYRLSPLNA

0.19

P–4, P–3:

AKGAAHKYRLSPLNA

0.63

P–2:

AKGYYAKYRLSPLNA

0.24

P0:

AKGYYHKARLSPLNA

>15

P+1:

A KGYYHKYALSPLNA

2.7

P+2:

A KGYYHKYRASPLNA

NB

P+3:

A KGYYHKYRLAPLNA

0.20

P+4:

A KGYYHKYRLSALNA

3.5

P+5:

A KGYYHKYRLSPANA

0.65

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.

at position P0, TRF2TRFH preferred tyrosine and phenylalanine to a lesser extent) for binding (Fig. 1b). At the P+2 position, leucine (but not isoleucine or valine) was selected. Additional selections at other positions were also evident, including tyrosine at P3, lysine at P1 and arginine at P+1. Indeed, a synthesized consensus peptide YRL (KGYYHKYRLSPLN) bound the TRFH domain of TRF2 with high affinity (190 nM) (Fig. 1c). Furthermore, alanine substitution of the YXL motif in the YRL peptide resulted in its loss of TRF2 interaction, whereas alanine substitution of the P+1 residue arginine or the P+4 residue proline reduced its affinity by more than ten-fold (Fig. 1d). These results indicate that TRF2TRFH recognizes specific peptide sequences with the core motif of [Y/F]XL. TRF2–[Y/F]XL interaction is crucial for telomere maintenance These findings suggest that TRFH domains may recruit different signaling molecules via a linear sequence code, in a manner similar to other protein-protein interaction modules such as the SH3 and WW domains26,27. In addition, disruption of TRF2TRFH interaction with its cellular targets may trigger telomere dysfunction. To further investigate the biological importance of the TRFH–[Y/F]XL motif interaction, we expressed the TRF2 consensus peptide in tandem repeats (2YRL) in human HTC75 cells. We reasoned that this tandem repeat peptide should occupy the two [Y/F]XL binding sites on a TRFH dimer and act as dominant negatives to inhibit endogenous TRF2 activity by competing for TRF2TRFH binding. Indeed, whereas the chromatin association of either TRF1 or TRF2 remained intact (Supplementary Fig. 1 online), expression of this peptide did elicit 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 (2YRA)


VOLUME 16

that does not bind TRF2 had no effect, underlining the important role of 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 TRF2 Arg109, is essential for the TRF2TRFH–YRL interaction24. We therefore generated the F120A mutant form of TRF2 (TRF2 FA, Fig. 2c) and found that it expressed at a level comparable to wild-type TRF2 expression (Fig. 2d). Consistent with our biochemical and structural analyses, the F120A mutation led to a dramatic reduction in YRL peptide binding (Fig. 2d). Furthermore, although TRF2 FA still retained its ability to interact with wild-type TRF2 (Fig. 2e), its expression in HTC75 cells increased the percentage of TIF-containing cells compared to that of cells expressing wild-type TRF2 (8% versus 2%; Fig. 2f–h). Notably, the percentage of TRF2 FA–expressing cells that contained TIF was similar to that of TRF2DBDM-expressing cells (Fig. 2g). The mechanism of how TRF2DBDM expression triggers telomeric DNA-damage responses has remained poorly understood18,29. Vastly overexpressed TRF2DBDM can lead to the displacement of endogenous TRF2 from telomeres21. In our experiments, modestly overexpressed TRF2DBDM could associate with telomeres and did not drastically alter the chromatin association of endogenous TRF2 (Supplementary Fig. 2a,b online). We reasoned that the intact TRFH domain on TRF2DBDM allowed it to act in a dominant negative manner, preventing multiple [Y/F]XL motif–containing proteins from binding to endogenous TRF2. Indeed, alanine mutation of Phe120 on TRF2DBDM abolished the effect of TRF2DBDM expression in TIF assays (Fig. 2f,g). Collectively, these experiments indicate that the association of TRFH with different [Y/F]XL motif–containing targets is crucial for TRF2-mediated telomere protection in human cells.

NUMBER 4

APRIL 2009

373

ARTICLES

P < 0.005 25

GST fusion

SFB2×YRL

TIF (%)

20

SFB-2×YRL

SFB2×YRA

TRF1

Merged

TRF2

Flag WB Input (4%)

TRF2FA

Streptavidin pull down

5

F2

TRF2 ∆B∆M FA

TRF2 TRF2 F120A (FA) TRF2 ∆B∆M

*

GST WB

g

10

TRF2 ∆B∆M FA TRF2 TRFH

8 Input (4%)

TIF (%)

*

TRF2 ∆B∆M

F2

TR

TR

GST

TR FA F2 TR F2 FA

GST-TRF2

FA

FA

Flag WB

F2

Myb

TR

449 495

Flag tagged

GST

F2 TR F2

A

L

YR 2×

YR

GST-TRF2 F2

244 TRFH

e

TR

1 45

2×

Ve

ct or

0

B

Pull down

6 4

Identification of TIN2, PNUTS and MCPH1 as TRF2TRFH targets The identification and analysis of specific TRFH binding partners should greatly facilitate the understanding of TRF2 function and telomere maintenance. To this end, we performed a genome-wide search of potential TRF2-interacting partners based on our peptide library data of TRF2TRFH (Supplementary Fig. 3 online). From this list, we selected several human proteins that have been implicated in signal transduction and RNA or DNA regulation for further analysis (Fig. 3a and Supplementary Fig. 4 online). On the basis of their sequences, we synthesized [Y/F]XL peptides and measured their affinities to TRF2TRFH. Three of the peptides (from TIN2, PNUTS and MCPH1, respectively) bound TRF2TRFH with affinities in the low micromolar range (Supplementary Fig. 4), and we followed up further on these interactions. Notably, TIN2 and PNUTS were also identified in our large-scale immunoprecipitation and MS analysis of TRF2 (Fig. 3b) and RAP1 protein complexes30,31. As in Apollo, TIN2 (known to associate with TRF2) also harbors a TRFH domain binding motif (FNL) (Supplementary Fig. 4). Indeed, TRF2TRFH can interact with the TIN2 FNL peptide with a modest affinity that is dependent on the FXL motif (Supplementary Fig. 5a–c online). It has been suggested that TIN2 may bind TRF2 through two distinct regions, the high-affinity TIN2 N-terminal region and the low-affinity region containing the TRFH binding motif24. The TRFH– TIN2 interaction was not stable enough to survive co-immunoprecipitation24. To further explore the association of the TIN2 FNL motif with TRF2 in vivo, we studied the TRF2–TIN2 interaction in live cells

374

VOLUME 16

F ∆B 2 F ∆ A ∆B M ∆M FA

TR

h

F2

Figure 2 The TRF2-[YF]XL interaction is important for telomere maintenance. (a) Telomere dysfunction 2 induced foci (TIF) analysis of HTC75 cells expressing the consensus YRL peptide. SFB-2YRL– or 0 SFB -2YRA–expressing cells were immunostained with anti-53BP1 (green) and TRF2 antibodies (red). TRF2 TRF2 TRF2 TRF2 (b) Quantification of TIF-positive cells. Only cells with Z7 53BP1 foci that colocalized with TRF2 foci FA ∆B∆M ∆B∆M FA 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-2YRL 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 TRF2-YFPn ∆B∆M-YFPn using streptavidin beads. The amount of co-precipitated GST-tagged protein was detected by anti-GST TRF2 WB Endo. TRF2 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 Actin Actin WB 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 YFPntagged 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. TR


53BP1

GST WB

15 10

c

f

Mock SFB-2×YRL Mock

F2 TR F2 TR FA F2 TR F2 FA TR F2 TR F2 TR FA F2 TR F2 FA

d

b 30

Merged

TR

TRF2

Vector

53BP1

TR

a

through the bimolecular fluorescence complementation (BiFC) assay32,33. Here we tagged TRF2 and TIN2, respectively, with the N-terminal half of Venus yellow fluorescent protein (YFP) (YFPnTRF2) and the C-terminal domain of YFP (YC-TIN2). These proteins were stably expressed in HTC75 cells for fluorescence complementation analysis. Consistent with the notion of multiple domains mediating the TRF2–TIN2 interaction, the FNL motif mutant TIN2AA showed reduced association with TRF2 (Supplementary Fig. 5d). The TRF2– TIN2 interaction is unlikely to be mediated through TRF1, because the TRF1 binding mutant TIN2AA can still interact with TRF2 and TRF1 does not interact directly with TRF2. In addition, interaction of TRF2 FA to endogenous TIN2 was decreased (Fig. 4), indicating that the FNL motif contributes to the TRF2–TIN2 interaction in vivo. Other TRF2-associated proteins identified include PNUTS, a nuclear targeting subunit of the protein phosphatase PP1 (ref. 34), and the microcephaly syndrome protein MCPH1 (also known as BRIT1), a BRCT domain–containing protein that functions in DNA-damage responses35–38. Neither protein has been shown previously to interact with telomere proteins. To confirm the TRF2–PNUTS and TRF2– MCPH1 interactions, we first carried out co-immunoprecipitation experiments using antibodies against the endogenous TRF2 and RAP1 complexes. Anti-RAP1 immunoprecipitation brought down endogenous 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 TRF2 foci, whereas Flag-MCHP1 co-localized with endogenous TRF2.

NUMBER 4

APRIL 2009


ARTICLES

PNUTS

230 VVVSDKYNLKPIPLKRQ

83 1 MCPH1 BRCT

Figure 3 TRF2 specifically interacts with YXLcontaining 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 Cterminal 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.

In p

907 940 ZnF

R

357 433 PP1D

In

73 147 TF2S

1 PNUTS

d

pu t( 2 AP .5% ) 1 Ig IP G C tr

c

ut Ig (1% G ) C R tr AP 1 IP

a

MCPH1

TRF2 648 835 BRCT BRCT

TRF2

RAP1

RAP1 324 ETFEEKYRLSPTLSSTK

e

b 220 160 120 100 80 70 60 50 40 30

Rad50 PNUTS NBS1 Mre11 TRF2

YXL motif protein

TRF2

Merged

FlagPNUTS

RAP1

FlagMCPH1

MW Anti(kDa) Flag IP

confirmed the importance of the YXLXP motif in mediating TRF2 binding (Fig. 4a). To determine whether TRF2–PNUTS or TRF2–MCPH1 association is dependent on the YXL motif in vivo, we generated alanine substitution mutants of PNUTS (PNUTS-AA) and MCPH1 (MCPH1-AA). Flag-tagged wild-type and mutant PNUTS or MCPH1 was then coexpressed with TRF2 in 293T cells for co-precipitation experiments. GST-TRF2 was able to specifically pull down wild-type

YXL-dependent interaction between PNUTS/MCPH1 and TRF2 Next, we synthesized YXL motif–containing peptides based on PNUTS 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

a

b TRF1

c GST

TRF2

NB

1.1

VSDKANLKPIPLKRQ

NB

>30

VSDKYNAKPIPLKRQ

NB

NB

VSDKYNLKAIPLKRQ

NB

4.8

G

VSDKYNLKPIPLKRQ

ETFEEKYRLSPTLSST

NB

0.42

ETFEEKYRASPTLSST

NB

NB

Flag WB

TRF2

GST

TRF2

Ve c M tor C P M H1 C P Ve H1A ct A M or C P M H1 C PH 1A A Ve ct o M r C P M H1 C Ve PH ct 1A M or A C M PH C 1 PH 1A A

Protein

PNUTS PNUTS AA PNUTS AA PNUTS

ST TR F TR 2 TRF2 F FA TR 2 F2 F TR A TRF2 F TR 2 F TRF2 A F G 2 ST FA

Peptide

PNUTS

MCPH1-SFB

Flag WB

GSTTRF2

GST-TRF2 GST WB GST WB

GST GST Input (2%)

MCPH1 WB TIN2 WB POT1 WB RAP1 WB Flag WB Flag-TRF2

Flag-TRF2 FA

% YFP Positives

PNUTS WB

Pull down

Input (20%)

e 25

Pull down

f PNUTS

20

TRF2

YRL [Peptide](µM)

15 Yc

10

YRA

- 1 10 100 1 10 100 PNUTS WB

YFPn

MCPH1 WB

5

TIN2 WB PNUTST RF2

0 Y YC C-P YC -P NU N TS U TS YC AA -T IN 2

d

In pu t( Fl 8% ag ) I Ig P G C In tr pu t( Fl 10 ag % IP ) Ig G C tr


TIN2

TRF2 WB Input (10%)

YFP

RAP1 IP

IgG IP

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 Flagtagged 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.


VOLUME 16

NUMBER 4

APRIL 2009

375

ARTICLES

b

c

6 5.5 5

+ – – – +

+ – – + –

4

+ + – – +

+ + – + –

60 50

21 35 46 55 68 Population doubling

MCPH1 sh Control sh P < 0.001 P = 0.005

40 30

Flag WB

20

TRF2 WB

10

RAP1 WB

0

Actin WB

M C

7

+ – + + –

MCPH1 WB

Vector PNUTS PNUTS∆C

4.5

+ – + – +

M C PH 1 PH 1A A

6.5

– – – + –

M oc k

7

– – – – +

%TIF

Flag-TPP1∆C22: YFPn-MCPH1AA: YFPn-MCPH1: MCPH1 sh: Control sh: YFPn-MCPH1 Endo MCPH1

M oc k

Average telomere length (kb)

a

TPP1∆C22

d

YFPn-MCPH1 Control shRNA

MCPH1 shRNA

Control sh TPP1∆C22

MCPH1 sh TPP1∆C22


YFPn-MCPH1AA

MCPH1 sh TPP1∆C22


MCPH1 sh TPP1∆C22

53BP1


TRF2

Merged

e

DNA-damage repair

Telomere length

Telosome and telomerase Telomerase

Telomere DNA

Peptide motif binding modules

PNUTS but not PNUTS-AA (Fig. 4b). Similarly, TRF2 interaction with MCPH1-AA was considerably impaired compared to its interaction with wild-type MCPH1 (Fig. 4c), strongly supporting the notion that both PNUTS and MCPH1 bind TRF2 through the YXL motif. To confirm whether the TRF2–PNUTS interaction is mediated through TRF2TRFH, we coexpressed Flag-tagged PNUTS with GST-tagged wild-type TRF2 or TRF2 FA. In these cells, TRF2, but not TRF2 FA, could bring down PNUTS, even though the F120A mutation did not affect TRF2 dimerization (Figs. 2e and 4b). Moreover, the ability of Flag–TRF2 FA to precipitate endogenous PNUTS and MCPH1 was greatly reduced (Fig. 4d). These observations indicate that the TRF2–PNUTS and the TRF2–MCPH1 interactions are 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, using the YFPn-TRF2 constrcut described above and PNUTS or MCPH1 tagged with with the C-terminal domain of YFP (YC-PNUTS and YC-MCPH1, respectively). Whereas YFP-positive cells were virtually absent in control cells, about 14% of the cells coexpressing YFPn-TRF2 and YC-PNUTS were YFP positive (Fig. 4e), indicating an interaction between TRF2 and PNUTS in vivo. This number is comparable to the percentage of YFP-positive cells in cells coexpressing YFPnTRF2 and YC-TIN2 (20%), which served as a positive control33. However, coexpression of YFPn-TRF2 and the YC-PNUTS-AA mutant did not result in increased fluorescence complementation over background (Fig. 4e). Similarly, MCPH1, but not MCPH1-AA, complemented YFPn-TRF2 in BiFC assays (Supplementary Fig. 6 online). These data provide further support that the YXL motif is crucial for the TRF2–PNUTS and TRF2–MCPH1 interactions in cells.

376

VOLUME 16

Figure 5 MCPH1 and PNUTS regulate DNAdamage 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. FlagTPP1DC22–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 MCPH1AA 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.

The identification of multiple TRF2TRFH targets raises the possibility that these proteins may compete for TRF2 binding in cells. To test this, we investigated whether the YRL peptide could cause differential displacement of endogenous PNUTS and MCPH1 from TRF2. Indeed, increasing the concentration of the YRL peptide reduced the association of both PNUTS and MCPH1 with TRF2 in the anti-RAP1 immunoprecipitates (Fig. 4f). Notably, the PNUTS–TRF2 interaction seemed to be more sensitive than the MCPH1–TRF2 interaction in this peptidetitration experiment. This difference in sensitivity can be correlated with the affinities of their corresponding YXL peptides for the TRFH domain (that is, the PNUTS peptide binds more weakly than the MCPH1 peptide). These observations open up the possibility that the affinity of the YXL motif may affect the outcome of competition between different TRFH binding proteins. PNUTS and MCPH1 regulate telomere length and end protection We 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 telomerase-positive HTC75 cells resulted in modest telomere elongation (Fig. 5a and Supplementary Fig. 7 online) but had little effect on TIF formation (data not shown), indicating a role for PNUTS in telomerelength maintenance but not DNA-damage responses. It should be noted that PNUTSDC may not act as an ideal dominant negative protein, so that the telomeric activity of PNUTS may have been underestimated in these assays. Because MCPH1 has been implicated in DNA damage–response pathways35–38, we hypothesized that the TRF2–MCPH1 interaction might regulate DNA-damage responses at the telomeres. To test this, we used a mutant form of TPP1 (TPP1DC22), whose expression results 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

NUMBER 4

APRIL 2009


ARTICLES


and Supplementary Fig. 8 online), indicating a requirement of MCPH1 for foci formation in response to DNA damage at the telomeres. Furthermore, an RNA interference (RNAi)-resistant form of wild-type MCPH1, but not MCPH1-AA, rescued the effect of MCPH1 knockdown (Fig. 5b–d), suggesting that the TRF2–MCPH1 interaction modulates the function of MCPH1 at the telomeres. These data collectively demonstrate that PNUTS and MCPH1 are physiological targets of TRF2 and are likely to function in distinct pathways. DISCUSSION The ability of TRF1 and TRF2 to bind telomeric sequences and thereby help to organize telomere chromatin structure and recruit other proteins to the telomeres has long been appreciated. Recent studies have further hypothesized that TRF1 and TRF2 may serve as molecular platforms for the recruitment and assembly of the telomere interaction network (‘telomere interactome’)12. However, the mechanisms by which this ever-expanding list of TRF1- and TRF2-interacting proteins contribute to TRF protein function remain unclear. The data presented here support the model that TRFH domains represent telomere-specific domains that recognize linear peptide sequence motifs, in a manner similar to that of many known protein modules such as the SH3 and WW domains. These sequences would effectively serve as molecular glue, allowing for the telomeric association of various signaling molecules and enabling TRF1 and TRF2 to function as hubs at the telomeres. The long, repetitive DNA sequences at the telomere end enable its association with arrays of TRF1 and TRF2 molecules to accommodate temporal, combinatorial and perhaps developmental regulation of diverse signaling cascades (Fig. 5e). For example, our data indicate that TRF2 arrays can function as a telomeric hub via TRF2TRFH and recruit at least four [Y/F]XL motif proteins: TIN2, Apollo, MCPH1 and PNUTS. The TRF2–TIN2 interaction regulates telosome formation and telomerase recruitment39,40, whereas the TRF2–Apollo and TRF2–MCPH1 interactions regulate DNA damage–repair responses22,23 (Fig. 5e). In addition, the TRF2–PNUTS association modulates telomere length. Because each TRF2 homodimer contains two [Y/F]XL motif binding sites, it is possible that two different TRFH targets can be recruited to the same TRF2 homodimer, promoting communication between two distinct signaling branches. In this sense, TRF2 arrays serve not merely as a hub but as a structural platform. It will be important to identify other proteins that directly interact with TRF2 and to dissect the function of and cross-talk between the TRF2 targets. To this end, the specificity of TRFH domains for particular linear sequences as determined by our peptide library experiments should prove highly valuable in predicting possible targets. In this study, we have successfully identified MCPH1 and PNUTS as new targets of the TRF2 TRFH domain. It has been proposed that telomeres are protected by a single large 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 a complex. However, TIN2 contains a [Y/F]XL motif, and the TIN2 binding pockets on TRF1 and TRF2 overlap with the binding pockets of other [Y/F]XL targets of TRF1 and TRF2, such as PINX1, Apollo and PNUTS (data not shown). As a result, we suggest that the telosome/shelterin complex may be one of several complexes (possibly competing with each other) at the telomeres at any given time. Consistent with this notion, we found that TRF2 complexes are heterogeneous30. A large fraction of TRF2 and RAP1 was detected in distinct peaks from the telosome. In addition, a distinct TRF2–RAP1 complex has been implicated in telomere


VOLUME 16

nonhomologous end joining41,42. Furthermore, interactions of the TRFH mutant TRF2 FA with [Y/F]XL proteins MCPH1, PNUTS and TIN2 were compromised, but its interaction with RAP1 was not (Fig. 4d). The ability of a TRFH domain to recruit different targets indicates a much more ‘proactive’ role for TRF2 in determining the assortment of complexes at the telomeres. Our findings point to new avenues into which the function of TRFH-containing proteins can be probed and offer new clues regarding the mechanisms of telomere dysfunction relevant to cancer and aging. METHODS Protein expression and purification. We expressed human TRF2TRFH (residues 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 YFPfragment 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 (2YRL) or YRA (2YRA) 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 (CalBiochem), 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 9Hflouren-9-ylmethoxycarbonyl chemistry. The synthesized peptides were incubated overnight with 2 equivalent of fluorescein isothiocyanate (FITC) in pyridine/dimethylformamide/dichloromethane (50:29:21, v/v). The FITClabeled 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 FITClabeled peptides (50 nM) at room temperature for 5–30 min. Fluorescence

NUMBER 4

APRIL 2009

377

ARTICLES polarization was subsequently measured in a 384-well plate using a Victor V plate reader (Perkin Elmer).


Immunoprecipitation, western blotting and immunofluorescence. For largescale 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 lentiviral 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 RNAiresistant 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 & Molecular Biology website. ACKNOWLEDGMENTS We thank S.Y. Jung and Q. He for technical help and M. Lei (University of Michigan) for the GST-TRF2TRFH fusion proteins. We thank J. Pennington and T. Palzkill for peptide synthesis. Work in the laboratories of Z.S. and D.L. is supported by awards from the US National Institutes of Health, the US Department of Defense, American Heart Association, the Welch foundation and the American Cancer Society. Z.S. is funded by the Leukemia and Lymphoma Society.

378

VOLUME 16

AUTHOR CONTRIBUTIONS H.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. and H.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-guanine base pairs. Cell 51, 899–908 (1987). 11. de Lange, T. Shelterin: the protein complex that shapes and safeguards human telomeres. Genes Dev. 19, 2100–2110 (2005). 12. Songyang, Z. & Liu, D. Inside the mammalian telomere interactome: regulation and regulatory activities of telomeres. Crit. Rev. Eukaryot. Gene Expr. 16, 103–118 (2006). 13. Broccoli, D., Smogorzewska, A., Chong, L. & de Lange, T. Human telomeres contain two 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 functional telomere structure. J. Biol. Chem. 279, 1442–1448 (2004). 17. Celli, G.B. & de Lange, T. DNA processing is not required for ATM-mediated telomere damage response after TRF2 deletion. Nat. Cell Biol. 7, 712–718 (2005). 18. Takai, H., Smogorzewska, A. & de Lange, T. DNA damage foci at dysfunctional telomeres. Curr. Biol. 13, 1549–1556 (2003). 19. Denchi, E.L. & de Lange, T. Protection of telomeres through independent control of ATM and ATR by TRF2 and POT1. Nature 448, 1068–1071 (2007). 20. Wang, R.C., Smogorzewska, A. & de Lange, T. Homologous recombination generates T-loop-sized deletions at human telomeres. Cell 119, 355–368 (2004). 21. van Steensel, B., Smogorzewska, A. & de Lange, T. TRF2 protects human telomeres from end-to-end fusions. Cell 92, 401–413 (1998). 22. van Overbeek, M. & de Lange, T. Apollo, an Artemis-related nuclease, interacts with TRF2 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 protect telomeres from DNA repair. Curr. Biol. 16, 1303–1310 (2006). 24. Chen, Y. et al. A shared docking motif in TRF1 and TRF2 used for differential recruitment 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 signaling modular domains. Prog. Biophys. Mol. Biol. 71, 359–372 (1999). 27. Yaffe, M.B. et al. A motif-based profile scanning approach for genome-wide prediction of signaling pathways. Nat. Biotechnol. 19, 348–353 (2001). 28. d’Adda di Fagagna, F. et al. A DNA damage checkpoint response in telomere-initiated senescence. Nature 426, 194–198 (2003). 29. Karlseder, J., Broccoli, D., Dai, Y., Hardy, S. & de Lange, T. p53- and ATM-dependent apoptosis induced by telomeres lacking TRF2. Science 283, 1321–1325 (1999). 30. Liu, D., O’Connor, M.S., Qin, J. & Songyang, Z. Telosome, a mammalian telomereassociated 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 protein complex and modulation of telomere length. J. Biol. Chem. 279, 28585–28591 (2004). 32. Hu, C.D. & Kerppola, T.K. Simultaneous visualization of multiple protein interactions in living cells using multicolor fluorescence complementation analysis. Nat. Biotechnol. 21, 539–545 (2003). 33. Chen, L.Y., Liu, D. & Songyang, Z. Telomere maintenance through spatial control of telomeric proteins. Mol. Cell. Biol. 27, 5898–5909 (2007).

NUMBER 4

APRIL 2009


ARTICLES 40. Xin, H. et al. TPP1 is a homologue of ciliate TEBP-b and interacts with POT1 to recruit telomerase. Nature 445, 559–562 (2007). 41. Bae, N.S. & Baumann, P.A. RAP1/TRF2 complex inhibits nonhomologous end-joining at 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 BRCA1dependent DNA damage response. Science 316, 1202–1205 (2007). 44. Songyang, Z. & Cantley, L.C. The use of peptide library for the determination of kinase peptide 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: assembly of prereplication complexes in late mitosis. Mol. Cell. Biol. 20, 8602–8612 (2000).


34. Allen, P.B., Kwon, Y.G., Nairn, A.C. & Greengard, P. Isolation and characterization of PNUTS, 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 damage responsive protein that regulates the Brca1-Chk1 pathway, implicating checkpoint dysfunction in microcephaly. Proc. Natl. Acad. Sci. USA 102, 15105–15109 (2005). 36. Rai, R. et al. BRIT1 regulates early DNA damage response, chromosomal integrity, and cancer. Cancer Cell 10, 145–157 (2006). 37. Alderton, G.K. et al. Regulation of mitotic entry by microcephalin and its overlap with ATR signalling. Nat. Cell Biol. 8, 725–733 (2006). 38. Wood, J.L., Singh, N., Mer, G. & Chen, J. MCPH1 functions in an H2AX-dependent but MDC1-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 and TIN2 interaction in high-order telomeric complex assembly. Proc. Natl. Acad. Sci. USA 103, 11874–11879 (2006).


VOLUME 16

NUMBER 4

APRIL 2009

379

ARTICLES

Polyglutamine disruption of the huntingtin exon 1 N terminus triggers a complex aggregation mechanism


Ashwani 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-like aggregates. We show here that the 17-amino-acid flanking sequence (HTTNT) N-terminal to the polyQ in the toxic huntingtin exon 1 fragment imparts onto this peptide a complex alternative aggregation mechanism. In isolation, the HTTNT peptide is a compact coil that resists aggregation. When polyQ is fused to this sequence, it induces in HTTNT, in a repeat-length dependent fashion, a more extended conformation that greatly enhances its aggregation into globular oligomers with HTTNT cores and exposed 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 disordered peptide and have implications for the molecular mechanism of Huntington’s disease.

There are nine known expanded CAG repeat diseases, in which expansion of a disease protein’s polyQ sequence beyond a threshold repeat length causes progressive neurodegeneration through a predominantly gain-of-function mechanism1. In Huntington’s disease, the repeat-length threshold is about 37 glutamines2. A major challenge to understanding disease mechanisms has been to discover physical properties of polyQ proteins that show repeat-length dependence in this threshold regime and that therefore might serve as a link in the progression from genetics to disease. PolyQcontaining aggregates are ubiquitously observed in these diseases1, and aggregation rates of polyQ sequences increase as repeat length increases3, mirroring correlations between the repeat length and the risk of disease and age of onset1. These observations led to the hypothesis that repeat length–dependent aggregation of polyQ is the triggering event in the mechanism of expanded CAG repeat diseases. Not all data support this hypothesis, however. In particular, in cell and animal models, disease progression is not always correlated with aggregate burden, as measured by inclusions revealed by light microscopy4. There are also inconsistent reports of the nature of polyQ aggregates. Thus, whereas simple polyQ peptides follow a nucleated growth polymerization mechanism with direct formation of amyloidlike aggregates3,5–8, aggregation products of the polyQ-containing disease protein huntingtin (HTT) exon 1 include, in addition to amyloid fibrils9, oligomeric and protofibrillar structures10,11 that many feel are more relevant to disease pathology12.

Although the human huntingtin (HTT) gene encodes a protein of more than 3,500 amino acids, expression of the first exon of the gene in cell and animal models is sufficient to replicate much of the pathology of Huntington’s disease1, and there is growing evidence that proteolytic release of a fragment containing exon 1 is required for toxicity13. The amino acid sequence of the translation product of human HTT exon 1, which includes the polyQ sequence, is shown in Table 1. As polyQ repeats are the only apparent common feature of the nine expanded polyQ repeat disease proteins1, we have extensively studied a series of simple polyQ peptides that contain flanking lysine residues added for solubility5–8,14. We found that these peptides aggregate via a nucleated growth polymerization mechanism in which the critical nucleus is a rarely populated form of the monomer5–8. In these peptides, increases in aggregation rates for peptides with longer polyQ repeat lengths are associated with more favorable equilibrium constants for nucleus formation5. We also showed previously that the proline-rich flanking sequence (oligoPro) on the Cterminal side of the polyQ in exon 1 reduces aggregation kinetics and aggregate stability but does not fundamentally change the aggregation mechanism14. Its effect is also directional; oligoPro added to the N terminus of polyQ has no impact on aggregation14. Intrigued by these oligoPro effects, we turned our attention to the 17-amino-acid sequence at the N terminus of the HTT protein, just upstream from the polyQ segment. In this paper, we describe detailed analysis of the in vitro aggregation mechanism of chemically synthesized peptide models for human HTT exon 1 that include this

1Department of Structural Biology and 2Pittsburgh Institute for Neurodegenerative Diseases, University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania 15260, USA. 3Department of Molecular and Cellular Biochemistry, University of Kentucky, Lexington, Kentucky 40536, USA. 4These authors contributed equally to this work. Correspondence should be addressed to R.W. ([email protected])

Received 5 May 2008; accepted 30 January 2009; published online 8 March 2009; doi:10.1038/nsmb.1570

380

VOLUME 16

NUMBER 4

APRIL 2009


ARTICLES


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 is grounded in a kind of reciprocal cross-talk between the polyQ and HTTNT segments, both of which are intrinsically unfolded protein sequences15. This work reveals a complex pathway featuring various aggregate structures and suggests an unanticipated degree of conformational communication between adjacent disordered elements in intrinsically unfolded protein sequences. These data are consistent with previous reports showing modulation of polyQ aggregation by flanking sequences in general16–21 and the HTT N terminus in particular22,23. RESULTS The role of the HTT N terminus in exon 1 aggregation To explore possible effects of the HTT exon 1 N-terminal 17 amino acids on polyQ aggregation, we generated the peptide HTTNTQ35 (Table 1), subjected it to a disaggregation procedure required to ensure the absence of pre-existing aggregates24 and studied its aggregation 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) more rapidly than 3 mM Q35 (Fig. 1a). In addition, the peptide HTTNTQ36P10 aggregates somewhat less rapidly than HTTNTQ35 but

much more rapidly than Q35 (Fig. 1a); this shows that, although the aggregation-suppressing ability of oligoPro also operates within the exon 1 context, the enhancing effect of HTTNT is dominant over the suppressing effect of oligoPro. As discussed above, when oligoPro was placed C-terminal to polyQ, the aggregation rate was diminished compared with that of the polyQ sequence alone (compare 25 mM Q35P10 with 25 mM Q35 in Figure 1a), but oligoPro placed N-terminal to polyQ had no effect14 (data not shown). In contrast, the HTTNT effect did not seem to depend on where the HTTNT was placed. Thus, Q35HTTNT shows an aggregation rate that is much faster than that of Q35 and not significantly (P 4 0.01 at t ¼ 0.75 h) different to that of HTTNTQ35 (Fig. 1a). Despite the apparent dominant effect by the HTTNT sequence, the kinetics of a series of HTTNTQN peptides continued to show a strong polyQ repeat–length dependence, as shown previously for both simple polyQ peptides25 and for recombinantly produced exon 1 peptides9. Thus, whereas HTTNT, HTTNTQ3 and HTTNTQ15 all aggregated sluggishly, HTTNTQ25 aggregated over a period of 1–2 d and, as discussed above, HTTNTQ35 aggregated within a few hours (Fig. 1b). Although some of the peptides examined in Figure 1b contain a F17W mutation to allow certain fluorescence experiments (described below), this mutation did not appreciably affect 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 KK QQQQQQQQQQ QQQQQ----- ---------- ---------- KK

Q15 Q20 Q29 Q30 Q35 Q35P10

KK QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- KK KK QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQ- ---------- KK KK QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ ---------- KK KK QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQ----- KK MATLEKLMKA FESLKSF--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQ----- KK

NT

MATLEKLMKA FESLKSF--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQ----- KK

NT

MATLEKLMKA FESLKSF-amide

HTT Q35 Q35HTTNT

KK QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQ----- MATLEKLMKA FESLKSF

HTT HTTNTQ3(F17W) NT

FRET-HTT Q3 HTTNTQ15(F17W) HTTNTQ25(F17W) HTTNTQ30P6 HTTNTQ30P6(F17W) HTTNTQ20P10 HTTNTQ20P10(F11A F17A) HTTNTQ20P10(M1X M8X) NT

HTT Q20P10(M1X M8X F11A F17A) HTTNTQ20P10(F17W) HTTNTQ20P10(F11W) FRET-HTTNTQ20P10 HTTNTQ37P10(F17W) FRET-HTTNTQ37P10 HTTNTK2Q36

MATLEKLMKA FESLKSW--- QQQ ¥ATLEKLMKA FESLKSW--- QQQ MATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQ----- ---------- ---------- KK MATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQQQQQQ QQQQQ----- ---------- KK MATLEKLMKA FESLKSF--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ ---------- PPPPPP---- KK MATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ ---------- PPPPPP---- KK MATLEKLMKA FESLKSF--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK MATLEKLMKA AESLKSA--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK XATLEKLXKA FESLKSF--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK XATLEKLXKA AESLKSA--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK

MATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK MATLEKLMKA WESLKSF--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK ¥ATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQQQQQQ ---------- ---------- PPPPPPPPPP KK MATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQQQ--- PPPPPPPPPP KK ¥ATLEKLMKA FESLKSW--- QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQQQ--- PPPPPPPPPP KK MATLEKLMKA FESLKSF-KK QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQQ QQQQQQ---- KK

BKK QQQQQQQQQQ QQQQQQQQQQ QQQQQQQQQ- ---------- KK Biotinyl-Q29 HTT-C, PQLPQPPPQA QPLLPQPQPP PPPPPPPPGP AVAEEPPLHR P; ¥, nitrotyrosine; X, methionine sulfoxide; B, biotin.

a


VOLUME 16

NUMBER 4

APRIL 2009

381

ARTICLES

a120

b

100 100 HTTNT, 5 µM Q35, 3 µM Q35, 25 µM HTTNTQ35, 3 µM HTTNTK2Q36, 3 µM Q35P10, 25 µM Q35HTTNT, 3 µM HTTNTQ36P10, 3 µM

60

40

20

80

% monomer

% monomer

80 HTTNT HTTNTQ25 HTTNTQ15 HTTNTQ3 HTTNTQ35

60 40 20 0

0

10

20

0 0

20

40

60 Time (h)

80

100

c 100

d –12

50

60

HTTNTQ30P6 Q30

–14

60 40

log slope

% monomer

40

120

80


30 Time (h)

F17W F11W WT F11A F17A M1X M8X M1X M8X F11A F17A

20 0 0

200

400 Time (h)

–16

–18 600

800

–6.5

–6.0

–5.5 –5.0 –4.5 log [peptide]

–4.0

–3.5

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 ¼ 0.688, s.d. ¼ ±1.8), Q35HTTNT (3 mM, R2 ¼ 0.996, s.d. ¼ ±3.0), HTTNTQ36P10 (3 mM, R2 ¼ 0.992, s.d. ¼ ±4.3), Q35P10 (25 mM, R2 ¼ 0.973, s.d. ¼ ±0.8), HTTNTK2Q36 (3 mM, R2 ¼ 0.981, s.d. ¼ ±1.0). (b) Role of polyQ repeat length on 5 mM peptides. HPLC sedimentation assay following aggregation of HTTNT (R2 ¼ 0.746, s.d. ¼ ±2.3), HTTNTQ3 (F17W) (R2 ¼ 0.966, s.d. ¼ ±2.1), HTTNTQ15 (F17W) (R2 ¼ 0.971, s.d. ¼ ±3.2), HTTNTQ25 (F17W) (R2 ¼ 0.992, 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 ¼ 0.987, s.d. ¼ ±2.5); M1X M8X (R2 ¼ 0.998, s.d. ¼ ±1.4); M1X M8X F11A F17A (R2 ¼ 0.628, 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.

HTTNTQ20P10 background show that replacement of either Phe11 peptide aggregates, electron micrographs showed that the initial or Phe17 with tryptophan in the HTTNT segment yields aggrega- products were oligomeric and protofibrillar (Fig. 2). Later in the tion kinetics similar to those of the corresponding wild-type time course, various other aggregated structures appeared that were more fibril-like (Fig. 2). A relationship between the formation of such peptide (Fig. 1c). Previously, we found that early stages of the aggregation of simple oligomeric products and non-nucleated, downhill aggregation kinetics polyQ peptides show a modest concentration dependence consistent (Fig. 1d) fitting a classical colloidal coagulation model has been with a nucleated growth polymerization mechanism and a critical suggested previously for other protein-aggregation reactions27,28. nucleus of one5, as shown for Q30 in Figure 1d. In contrast, the initial stages of aggregation of the exon 1–related sequence HTTNTQ30P6 HttNT conformation and exon 1 aggregation kinetics (previously, we found no appreciable difference between a Pro6 and a There are several possible explanations for the aggregation-enhancing Pro10 sequence14) yields a log-log plot of initial kinetics versus ability of the HTTNT sequence. (i) As addition of lysine residues to a concentration5 (Fig. 1d) with a slope of a b c f approximately 1, corresponding to a calculated critical nucleus (n*) of about 1. The n* ¼ 1 value indicates a rate/concentration d relationship for the early aggregation kinetics g of HTTNTQ30P6 that is consistent with a none nucleated, ‘downhill’ aggregation mechanism26 for oligomer formation, without a kinetic barrier to spontaneous aggregation. h j k l This analysis suggests that polyQ peptides NT containing the HTT sequence spontaneously aggregate by an entirely different mechanism than that of simple polyQ pepi tides, a conclusion that is further supported by additional data presented below. Previously, we found that when simple polyQ monomers undergo spontaneous Figure 2 Electron micrographs of various HTTNT-related aggregates. HTTNTQ30P6 was incubated in aggregation in aqueous solution, the earliest PBS 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 observable aggregates have fibril-associated 100 h (k). HTTNTQ3 (F17W) was incubated in PBS at 37 1C for 800 h (l). All samples were transferred properties similar to those of the final pro- directly from reaction mixture to freshly glow-discharged carbon-coated grids and stained with 1% (v/v) duct5. In contrast, when the HTTNTQ30P6 uranyl acetate. Scale bar ¼ 50 nm.

382

VOLUME 16

NUMBER 4

APRIL 2009


ARTICLES

a

Figure 3 State of expansion of the HTTNT peptide in solution. (a) Fractional migration (Kav) versus log molecular weight (MW) of various peptides in sizeexclusion 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-toend separation calculated from FRET measurements for mutants FRETHTTNTQ3, 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).

0.4 HTTNT

0.35

Kav

0.3 Bal 0.25

Pro14

Q15

Apr

Q20

0.2

Ins Q29

0.15

Q35

0.1 3.1

b

3.3

3.5 log MW

FRET distance (Å)

3.9

36

**

34

*

32 30 28 26 24 22 20 Q3

Q20P10

Q37P10

Q3 / urea

HTTNT context

sequence sometimes discourages aggregation29, HTTNT might enhance exon 1 aggregation compared to our simple polyQ peptides because it replaces the Lys-Lys pair at the N terminus of the model polyQ peptides (see introduction and Table 1). However, the aggregation kinetics of 3 mM HTTNTK2Q35, an exon 1 analog containing a Lys-Lys pair inserted between HTTNT and Q35, were not significantly (P 4 0.01 at t ¼ 0.3 h) different from the kinetics of 3 mM HTTNTQ35 but at the same time were significantly (P o 0.001 at t ¼ 10.5 h) faster than the kinetics of 3 mM Q35 (Fig. 1a). In addition, mutations that reduce the hydrophobic character of HTTNT without altering charged residues abrogated the rapid aggregation kinetics in an exon 1 peptide (Fig. 1c), confirming that the HTTNT segment provides some positive, sequencespecific contribution to aggregation rather than simply replacing a Lys-Lys aggregation suppressor. (ii) An obvious alternative possibility is that HTTNT itself might be highly aggregation prone. Unexpectedly, however, we found that a peptide consisting solely of the HTTNT sequence aggregated sluggishly under our standard conditions (Fig. 1a,b). (iii) Another possible explanation is that the combination of all or part of HTTNT with the first portion of the polyQ sequence might create a new aggregation sequence motif with much more robust

a

10

[θ] millidegrees.cm2.dmol–1

0

b

aggregation kinetics than either polyQ or HTTNT alone. However, HTTNT peptides containing short to intermediate lengths of polyQ also aggregated slowly (Fig. 1b), and robust aggregation kinetics ensued for longer polyQ constructs even when HTTNT was attached to the polyQ C terminus (Fig. 1a), which creates an entirely different sequence at the Qn-HTTNT junction; these data suggest that polyQ abutted to HTTNT does not simply create a powerful, novel, linear aggregation motif. The HTTNTK2Q35 data (Fig. 1a) also argues against a novel aggregation motif because, if there were such a motif at the junction of the HTTNT and polyQ sequences, the highly charged Lys-Lys insertion would be expected to disrupt it. (iv) It is also possible that the HTTNT sequence might normally exist as a well-behaved oligomeric species, such as a dimer or trimer, that can accelerate aggregation by concentrating and/ or orientating the polyQ elements. However, size-exclusion chromatography (SEC) mobility showed that the HTTNT peptide migrates as a monomer (Fig. 3), and SEC elution profiles (data not shown) showed no evidence of higher assembly states. Likewise, CD spectra of HTTNT did not change with respect to concentration (Fig. 4a), consistent with a non-associating system. (v) Another possibility is that, in analogy to a recent report of the ability of expanded polyQ to destabilize a folded protein domain20, the HTTNT segment may exist in a folded and/or compact state that resists aggregation but unfolds or extends when attached to expanded polyQ, enhancing its aggregation; disruption of the 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 mechanism for the HTTNT effect is the fact that most peptides the size of HTTNT do not fold into stable, globular structures but, rather, are disordered. Notably, however, analytical SEC suggests that the HTTNT sequence is actually relatively compact in solution. A series of simple polyQ peptides yielded migration rates in SEC that fit a straight line

80 60

[θ] millidegrees.cm2.dmol–1


3.7

–10

–20

–30

40 20 0 –20 –40

–40 190

200

210 220 230 240 Wavelength (nm)

250

260


–60 190

200

VOLUME 16

210 220 230 240 Wavelength (nm)

NUMBER 4

250

APRIL 2009

260

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% polyproline 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.

383

ARTICLES

a

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) crosspeaks 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.

1 5 10 15 M A T L E K L M K A F ES L KS F dαN(i,i) dαN(i,i+1) dNN(i,i+1) dαN(i,i+2) +0.2 ∆δ Hα 0.0 (p.p.m.) –0.2

b 4.10 4.15

Glu5 Lys6 Thr3 Hα-Glu5 HN

Ala10 Glu12 Lys15

4.20

1H

(p.p.m.)

4.25

Thr3

4.30 4.35

Lys9

Leu4

Leu14

Leu7

Ser13

4.40

Ser16

4.45 M8

4.50

Phe11

4.60

Phe17

(p.p.m.)

8.25

1H

8.45

Leu7-Met8

8.40

8.35

8.30

8.25

8.20

8.40 8.35 8.30 1HN (p.p.m.)

8.25

8.20

Leu7-Lys6

8.30 8.35

Glu5-Leu4

8.40 8.45 8.50 8.50

8.45

(Fig. 3a). As expected, peptides that are predicted to be extended and thus have larger hydrodynamic radii, such as an alanine-rich peptide strongly favoring an a-helical structure31 and a Pro14 peptide favoring polyproline type II helix32, migrated faster than polyQ peptides of equivalent length. Insulin and aprotinin, which show smaller hydrodynamic radii caused by their compact structures, migrated—as expected—more slowly than the polyQ peptides of equivalent length. Within this data set, the HTTNT peptide also migrated substantially more slowly than expected for a peptide of its molecular weight, suggesting that—like insulin and aprotinin—HTTNT has a relatively small hydrodynamic radius. To confirm the compactness of the HTTNT peptide, we conducted a fluorescence-based resonance energy transfer (FRET) experiment in which we replaced Met1 of HTTNT with the resonance energy acceptor nitrotyrosine, and Phe17 with the resonance energy donor tryptophan33 in the context of an HTTNTQ3 sequence (‘FRET-HTTNTQ3’). Compared with HTTNTQ3 containing only the F17W replacement, the tryptophan fluorescence of the FRET-HTTNTQ3 peptide decreased by about 50% (data not shown), corresponding to a calculated average separation between donor and acceptor groups of 24 ± 0.5 A˚ (Fig. 3b). This is substantially shorter than the average end-to-end distance of 34.5 ± 4 A˚ calculated34 for a 17-residue peptide in an extended statistical coil conformation. This theoretical value is supported by a FRET analysis of the HTTNTQ3 peptide in 6 M urea, which yielded a separation of 33.9 ± 0.5 A˚ (Fig. 3b). Thus, in agreement with the SEC data, HTTNT in the HTTNTQ3 context in native buffer seems to be markedly collapsed. Given this evidence for a collapsed structure in an isolated HTTNT peptide, we investigated whether it was possible for expanded polyQ sequences 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 tryptophan in the HTTNT sequence did not significantly change (P 4 0.01).

384

VOLUME 16

HTTNT has no stable secondary structure Although several features of the above results are consistent with HTTNT (in the absence of a connected expanded polyQ) being a compact domain, the short length of HTTNT makes it unlikely (but not impossible) that it possesses a unique, folded structure. To probe the secondary and tertiary structure of HTTNT, we applied two solution methods. The first, CD spectral analysis, proved equivocal. The CD spectrum of HTTNT at 35 1C (Fig. 4a) and the difference spectrum resulting from subtraction of a 35 1C spectrum from a 5 1C spectrum (data not shown) lack strong secondary-structure features, suggesting the absence of a stable structure. At the same time, deconvolution analysis of the 35 1C spectrum predicts substantial a-helical structure (Fig. 4, see legend), consistent with projections based on amino acid sequence23,35. This interpretation of the CD spectrum is problematic, however, because the protein data sets used in programs such as ContinLL are thought to be unsuitable for deconvoluting CD spectra of short peptides36. Exon 1 1.0 0.9 0.8 PONDR score


8.50 8.20

However, the corresponding separation for the Q37 repeat peptide expanded to 32 ± 1 A˚ (P o 0.01), a value approaching the 34-A˚ range obtained both by measurement of HTTNTQ3 in urea and by calculation based on an assumed statistical coil configuration (Fig. 3b). The data are thus consistent with a mechanism in which the HTTNT segment of exon 1 is normally in a collapsed state that is resistant to aggregation but that, when connected to an expanded polyQ sequence, becomes extended and labile to aggregation—an effect analogous to what has previously been observed for a globular protein20.

Ala2 Hα-Thr3 HN

4.55

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0

50

100 150 200 250 300 350 400 450 500 550 600 Residue number

Figure 6 PONDR analysis of the first 600 amino acids of the human huntingtin 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/).

NUMBER 4

APRIL 2009


ARTICLES 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 ¼ 0.994, 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, nonincubated monomer). Below, equivalent masses of isolated aggregates (no material in the ‘M’ column in this row).

Emission max. (nm)

a 356 354 352 350 348 346

b 100 80

60

% reaction

% reaction

50

40

40 30 20 10

20

0

0

2

4 6 Time (h)

8

10

0

25

50

75

100

125

Time (h)

c

M

1

4

8 15 25 40 70

Reaction mix: Aggregates:

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 show that HTTNT adopts predominantly unfolded, random-coil conformations characterized by small spectral dispersion, small secondary chemical shifts and strong sequential Ha(i) HN(i + 1) NOEs with few sequential HN(i) HN(i + 1) or medium-range NOEs. A single, weak, medium-range NOE cross-peak connects the Thr3 Ha and Glu5 HN protons (Fig. 5b). The region around these residues shows the largest negative Ha secondary chemical shifts and the most HN(i) HN(i + 1) NOE cross-peaks (Fig. 5a), indicating transient existence of a few residues in the a-helix quadrant of F,C space. Despite this slight propensity, there is clearly no stable a-helix in this peptide in solution under physiological conditions.

a

+ –

A multistep aggregation mechanism When the assembly of simple polyQ peptides into amyloid-like aggregates is monitored by multiple analytical techniques, the data from all measures track closely, suggesting the absence of reaction intermediates5. In contrast, HTTNT-containing exon 1 peptides show different 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 structure38, generated reaction curves that were delayed compared to the

c 100

–

d

90

355

0.8 0.6

350

0.4

345

0.2

340 0

0.0 100 200 300 400 500 600 700 800 900 Time (h)

% aggregration

1.0

Elongation rate (fmol h–1)

80

b 360 Emission max. (nm)


0

Thus, in the absence of an expanded polyQ, HTTNT adopts a conformation that lacks appreciable secondary and tertiary structural features but at the same time exists in a collapsed state. Results from the 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 of engaging in coupled folding and binding interactions15 (Fig. 6). Such sequences tend to be collapsed coils in native buffer15, often taking up a-helical conformations when complexed with their binding partners. Consistent with this, HTTNT has essentially no a-helical structure in isolation (see above) but takes on a substantial amount of a-helical structure in the presence of low concentrations of trifluoroethanol, as revealed by CD (Fig. 4b).

M

14

24

40

50

60

80 100 200 300 400

70

e

60 50

Turns, α-helix 310-helix turns, β-sheet β-aggregates unordered Gln NH2 def. β-sheet

40

Gln C=O stretch

6

30 5

20

4

10

3

0 0

100 200 300 400 500 600 700 800 900 Time (h)

2

1 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 1,550 1,600 1,650 1,700 residues in the mutant peptides F11W (————, R2 ¼ 0.994, s.d. ¼ ±0.5) and F17W Amide I frequency (cm–1) (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.


VOLUME 16

NUMBER 4

APRIL 2009

385


ARTICLES all aggregates isolated after Trp17 burial feature a single b-sheet band at B1,626 cm1 and are indistinguishable from each other and from aggregates of simple polyQ (Fig. 8e). Notably, at a time when the major changes in aggregate structure were occurring, as eviNT NT Figure 9 Mechanism of HTT -mediated exon 1 aggregation. The HTT domain (green) unfolds in a denced by ThT binding, tryptophan fluorespolyQ 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 cence shift, FTIR, EM and polyQ antibody involve both HTTNT and polyQ in amyloid-like structure; oligoPro (black) is not incorporated into the binding, the vast bulk of the exon 1 peptide core. This drawing is schematic and is not meant to imply any details of aggregate structure, except (480%) did not pellet after centrifugation that final aggregates are rich in b-sheet, are fibrillar and involve both HTTNT and polyQ. Although the (Figs. 7 and 8) and migrated as a single, initial formation of oligomers shows non-nucleated, downhill kinetics, it is likely that a nucleation event monomeric (Fig. 3a) species in SEC (M.J. takes place stochastically within the prefibrillar aggregate population, as shown in brackets, to trigger and R.W., unpublished data). Coincident rapid amyloid growth. with the timing of the changes in aggregate structure, there was a marked increase in the curves from the HPLC sedimentation (~) and light-scattering (J) rate of monomer loss, suggesting that a nucleation event had occurred. assays (Fig. 7b and 8c). Thus, for the HTTNTQ30P6 peptide (Fig. 7) at 6.25 h, the difference between the ThT value (D) and the HPLC value DISCUSSION (~) is statistically significant (P o 0.01); at later time points (19 h, The results presented here are consistent with the model shown in 120 h) the difference is insignificant (P 4 0.01). The observation of Figure 9. In this model, the HTTNT peptide has an intrinsic tendency the ThT lag followed by a burst suggests that the initial aggregates are to collapse into an aggregation-resistant compact coil state, but an not amyloid-like, whereas later aggregates are. FTIR analysis also attached, expanded polyQ sequence induces in this segment a more suggests that the initial aggregation product is qualitatively differ- extended state. When the HTTNT sequence is extended, it becomes ent—showing more coil and less b-structure—compared to aggregates susceptible to formation of a metastable, micelle-like aggregate, with isolated later in the time course (Fig. 8e). This is also consistent with the HTTNT segment making up the loosely packed core and the the EM data (Fig. 2). flanking polyQ sequence excluded from the core. Although this initial Details of the structures of these intermediates are revealed by aggregation reaction is non-nucleated, amyloid nuclei seem to rise further analysis of HTTNTQ30P6 (Fig. 7) and HTTNTQ20P10 (Fig. 8) stochastically from among these initial, prefibrillar aggregates. With aggregates isolated from ongoing reactions. In particular, dot blot the emergence of amyloid fibril structure, the remaining monomeric analysis of these aggregates showed that polyQ in the initial aggregates fraction—which represents the vast majority of the exon 1 molecules is readily accessible to an antibody against a linear polyQ epitope but at the time when these postulated nucleation events are taking place— is masked in the later aggregates (Figs. 7c and 8d). In addition, the fuels the ongoing fibril-elongation reaction. Although many details of initially formed aggregates are poor templates for recruiting polyQ- this proposed mechanism remain to be worked out, it seems that containing peptides, whereas the later, fibrillar aggregates are capable longer polyQ sequences favor aggregation in two ways: first, by of seeding polyQ elongation (Fig. 8b). Seeding activity is a feature disrupting the HTTNT compact coil and thereby facilitating initial particularly associated with amyloid-like aggregates39. Furthermore, HTTNT aggregation; and, second, by improving the efficiency whereas a tryptophan residue inserted in place of phenylalanine at of the nucleation events proposed to occur within the prefibrillar position 11 or 17 was solvent exposed both in the monomeric peptide aggregate population. and in the initial aggregates, it became less solvent accessible (based on In many respects, the mechanism shown in Figure 9 resembles the its fluorescence maximum) as the aggregation reaction proceeded, ‘conformational conversion’ model for spontaneous amyloid growth with a time course that parallels the disappearance of polyQ antibody proposed for yeast prion amyloid formation40. In that model, relabinding epitopes (Figs. 7a (’); 8B (,m)). In contrast, when the tively loosely structured spherical oligomers form first, then convert isolated HTTNT peptide itself aggregated slowly, the tryptophan into a more structured oligomer that can grow by recruiting other residue of the F17W mutant remained fully solvent exposed in the oligomers or monomers. In Figure 9, we envision that the nucleation aggregate fraction, even after 800 h (Fig. 7a (D)); this is consistent process for exon 1 aggregation consists of the stochastic rearrangement with the apparently inability of aggregates of HTTNT lacking a long of some oligomers or protofibrils into amyloid-like structures capable polyQ track to progress beyond the oligomer stage, in agreement with of rapidly propagating via monomer addition. That is, although some EM analysis (Fig. 2l). Although the initial aggregates of HTTNTQN prefibrillar aggregates readily dissociate back to monomers (M.J. and peptides seem to be less tightly packed than the later aggregates, they R.W., unpublished data), those aggregates that successfully undergo are sufficiently structured that they are protected from proteolyis by the nucleation process seem to be required intermediates, under these experimental conditions and, therefore, to be on-pathway41 to amytrypsin, in contrast to soluble, monomeric HTTNT (Fig. 8a). NT Thus, the initial aggregates in this HTT -mediated mechanism are loid formation. The kinetics of the only known alternative route to nonfibrillar oligomers, with their HTTNT segments composing the core amyloid formation—the previously described5 nucleated growth but their polyQ segments remaining unstructured and available for mechanism—are too slow to provide the observed rapid fibril formaantibody binding. Subsequently, the polyQ elements also become tion (A.K.T. and R.W., unpublished data). Providing further support integrated into the aggregate core structure, which takes on a more for the scheme shown in Figure 9, and extracting details on the fibrillar character both in the polyQ and HTTNT elements. Growth into mechanism and kinetics of nucleation process and other assembly larger fibrillar assemblies, as seen in EM, take place at later incubation steps are the focus of current work. Amino acid sequence has been viewed as having two major roles in times. FTIR analysis (Fig. 8e) showed that the initial aggregates (before burial of Trp17) contain a substantial amount of coil and turn influencing protein aggregation and amyloid formation30. First, conformations in addition to b-structure. Notably, FTIR spectra of globular proteins become more amyloidogenic when their folded

386

VOLUME 16

NUMBER 4

APRIL 2009



ARTICLES structures are destabilized, for example, by point mutations. Second, amyloidogenic proteins possess specific primary-sequence elements that constitute the core regions of the amyloid structure, and mutations within these regions can abrogate or enhance amyloid formation. Our studies on HTT exon 1 aggregation, however, reveal other means by which sequence can influence aggregation rate. Recently, we reported that an oligoPro sequence on the C-terminal side of a polyQ tends to decrease aggregation rate and aggregate stability, apparently by favoring more aggregation-resistant conformations14. Here we show that adjacent elements of intrinsically unfolded polypeptides can engage in an ‘aggregation two-step’ in which one element supplies rapid initial aggregation kinetics (but is insufficient to form amyloid or provide stability) whereas a second, connected element (which in isolation aggregates relatively slowly) provides the means to a highly stable amyloid structure. In a further twist, we show that a degree of compactness in the unstructured HTTNT sequence provides protection against aggregation and that the previously described destabilizing ability of a connected polyQ20,42 disrupts this protective structure and opens the door to robust aggregation. Together, these results suggest that the rules by which primary sequences inform the direction and kinetics of protein aggregation may be much more complex than previously believed29. Perhaps 25% of the proteome consists of sequences that do not exist as ordered globular or fibrous structures but rather are intrinsically unfolded43. The results in this paper, together with our previous studies on the proline-rich sequence of HTT exon 114, illustrate the dramatic effects on peptide conformation and aggregation imparted through subtle sequence effects within a single disordered polypeptide. In particular, the ability of disordered polyQ sequences to influence the folding stability of adjacent domains20,42 is an unprecedented phenomenon in protein science whose limits, and still obscure mechanism, are yet to be elucidated. What is particularly unexpected about our data on the effect of expanded polyQ on HTTNT is the implied ability of the compact coil state of HTTNT to resist aggregation; classically, compact but disordered ‘molten globule’ states of globular proteins have been considered to be aggregation-prone44. PolyQ destabilization of adjacent domains may account for the observation that longer polyQ tracts are often associated with reductions in protein activity7, a trend suggesting the possibility that some loss of function45 might accompany the toxic gain of function that is generally thought to dominate polyQ diseases1. We believe HTTNT derives its unusual impact on aggregation mechanism 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 the cell15. Although this designation for HTTNT is based solely on PONDR analysis (Fig. 6), the biophysical properties of HTTNT are entirely consistent with this notion, and HTT is well known to possess many interaction partners46. Our results add to a growing literature on interactions between polyQ and its flanking sequences in aggregation reactions. After initial reports on the ability of flanking sequences to modulate aggregation of polyQ disease proteins in cells47, several papers16–19,21 appeared describing aggregation by a flanking sequence that facilitates aggregation of the polyQ portion, sometimes as a clearly defined initial step19,21 and sometimes initiated by the destabilizing influence of an adjacent expanded polyQ20. Our results put into a molecular biophysics context several recent cell-based studies of the role of the HTTNT sequence in exon 1 aggregation and toxicity. Using models in which exon 1 fragments are expressed in mammalian or yeast cells, several groups have shown


VOLUME 16

that the presence or absence of HTTNT, as well as mutations within or adjacent to HTTNT, introduce complex alterations in subcellular localization, aggregate formation and/or cytotoxicity or growth retardation22,23,35,48. Studies showing that binding of HTTNT by a specific immunoglobulin fragment inhibits exon 1 aggregation and toxicity in several cell models49 are consistent with our data showing that HTTNT aggregation is the first step in exon 1 aggregation and that this depends on destabilization of a compact state in HTTNT. Despite its lack of a-helical structure, the ability of HTTNT to form a-helix structure in response to its environment (Fig. 4b) is consistent both with suggestions that HTTNT may mediate exon 1 targeting to membrane fractions23,35 and with our hypothesis that HTTNT is a MoRF sequence that mediates coupled folding and binding to protein targets. The ability of HTTNT to markedly alter the polyQ aggregation mechanism and the ability of HTTNT mutations (Fig. 1c) and binding factors49 to abrogate or modulate this effect are likely to contribute to some observed cellular effects. In fact, our recognition that polyQ sequences may aggregate by entirely different mechanisms, depending on flanking sequences, suggests that at least some of the diversity of aggregate morphologies observed in exon 1 cell and animal models11,50 may ultimately be explained by the biophysical character of the exon 1 sequence itself. METHODS Materials. 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, hexafluoroisopropanol (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 spectrapolarimeter 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 experiments 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 WATERGATE W5 pulse sequence52. We acquired two dimensional homonuclear NOESY and TOCSY53 data at 5 1C, using a 1.5-s recycle delay and mixing

NUMBER 4

APRIL 2009

387

ARTICLES 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 min1, 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 aggregatefree at the end of the FRET measurements by EM, HPLC, ThT and 901 lightscattering measurements. Fluorescence spectra were recorded at room temperature (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 M1 cm1 and nitrotyrosine e381 nm ¼ 2,200 M1 cm1, 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-cm1 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

388

VOLUME 16

obtained between 1,700–1,800 cm1. 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 necessitated 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. ACKNOWLEDGMENTS The authors acknowledge J. Ko and P. Patterson (California Institute of Technology) for the gift of the MW1 antibody, and T. Fullam (Allegheny College) for providing a set of aggregation kinetics data. We also acknowledge the following funding sources that contributed to the work described here: NIH R01 AG019322 (R.W.); Huntington’s Disease Society of America postdoctoral fellowship (V.M.C.); NSF MCB-0444049 (T.P.C.); Petroleum Research Fund/ American Chemical Society 43138-AC4 (T.P.C.); grant #4100026429 from the Commonwealth 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 onset differs 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 expanded polyglutamine repeat diseases. in Protein Misfolding Diseases: Current and Emerging Principles 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 body formation reduces levels of mutant huntingtin and the risk of neuronal death. Nature 431, 805–810 (2004).

NUMBER 4

APRIL 2009



ARTICLES 5. Chen, S., Ferrone, F. & Wetzel, R. Huntington’s disease age-of-onset linked to 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. in Genetic 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 aggregation nucleation 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-like protein aggregates in vitro and in vivo. Cell 90, 549–558 (1997). 10. Poirier, M.A. et al. Huntingtin spheroids and protofibrils as precursors in polyglutamine fibrilization. J. Biol. Chem. 277, 41032–41037 (2002). 11. Wacker, J.L., Zareie, M.H., Fong, H., Sarikaya, M. & Muchowski, P.J. Hsp70 and Hsp40 attenuate formation of spherical and annular polyglutamine oligomers by partitioning 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 dysfunction and degeneration due to mutant huntingtin. Cell 125, 1179–1191 (2006). 14. Bhattacharyya, A. et al. Oligoproline effects on polyglutamine conformation and aggregation. 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 of the 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 not all: 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 innterplay between PolyQ and protein context delays aggregation by forming a reservoir of protofibrils. PLoS ONE 1, e111 (2006). 19. Ellisdon, A.M., Thomas, B. & Bottomley, S.P. The two-stage pathway of ataxin-3 fibrillogenesis involves a polyglutamine-independent step. J. Biol. Chem. 281, 16888–16896 (2006). 20. Ignatova, Z. & Gierasch, L.M. Extended polyglutamine tracts cause aggregation and structural 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 a polyglutamine-containing chimera is a multistep process initiated by the flanking sequence. J. Biol. Chem. 282, 36736–36743 (2007). 22. Duennwald, M.L., Jagadish, S., Muchowski, P.J. & Lindquist, S. Flanking sequences profoundly 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-cellular localization, 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 highperformance liquid chromatography-based sedimentation assay. Methods Enzymol. 413, 34–74 (2006). 25. Chen, S., Berthelier, V., Yang, W. & Wetzel, R. Polyglutamine aggregation behavior in vitro 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 common critical 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-associated protein 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 short alanine-based peptides. Proc. Natl. Acad. Sci. USA 86, 5286–5290 (1989).


VOLUME 16

32. Schuler, B., Lipman, E.A. & Eaton, W.A. Probing the free-energy surface for protein folding 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 modulate huntingtin aggregation, nuclear entry and toxicity. Hum. Mol. Genet. 16, 2600–2615 (2007). 36. Whitmore, L. & Wallace, B.A. DICHROWEB, an online server for protein secondary structure 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. Methods Enzymol. 309, 274–284 (1999). 39. O’Nuallain, B., Williams, A.D., Westermark, P. & Wetzel, R. Seeding specificity in amyloid 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 conformational information by a prion determinant. Science 289, 1317–1321 (2000). 41. Kodali, R. & Wetzel, R. Polymorphism in the intermediates and products of amyloid assembly. Curr. Opin. Struct. Biol. 17, 48–57 (2007). 42. Bevivino, A.E. & Loll, P.J. An expanded glutamine repeat destabilizes native ataxin-3 structure 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 a molten globule intermediate. J. Biol. Chem. 274, 32897–32903 (1999). 45. Cattaneo, E. et al. Loss of normal huntingtin function: new developments in Huntington’s disease research. Trends Neurosci. 24, 182–188 (2001). 46. Kaltenbach, L.S. et al. Huntingtin interacting proteins are genetic modifiers of neurodegeneration. PLoS Genet. 3, e82 (2007). 47. Nozaki, K., Onodera, O., Takano, H. & Tsuji, S. Amino acid sequences flanking polyglutamine stretches influence their potential for aggregate formation. Neuroreport 12, 3357–3364 (2001). 48. Steffan, J.S. et al. SUMO modification of Huntingtin and Huntington’s disease pathology. Science 304, 100–104 (2004). 49. Colby, D.W. et al. Potent inhibition of huntingtin aggregation and cytotoxicity by a disulfide bond-free single-domain intracellular antibody. Proc. Natl. Acad. Sci. USA 101, 17616–17621 (2004). 50. Wanderer, J. & Morton, A.J. Differential morphology and composition of inclusions in the 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-like features 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 NMR spectroscopy. J. Magn. Reson. 132, 125–129 (1998). 53. Bax, A. & Davis, D.G. MLEV-17-based two-dimensional homonuclear magnetization transfer spectroscopy. J. Magn. Reson. 65, 355–360 (1985). 54. Schwarzinger, S. et al. Sequence-dependent correction of random coil NMR chemical shifts. 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. Protein Eng. 12, 485–490 (1999). 57. Ko, J., Ou, S. & Patterson, P.H. New anti-huntingtin monoclonal antibodies: implications 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 circular dichroism spectra: comparison of CONTIN, SELCON, and CDSSTR methods with an expanded reference set. Anal. Biochem. 287, 252–260 (2000). 59. Jackson, M. & Mantsch, H.H. The use and misuse of FTIR spectroscopy in the determination of protein structure. Crit. Rev. Biochem. Mol. Biol. 30, 95–120 (1995). 60. Venyaminov, S. & Kalnin, N.N. Quantitative IR spectrophotometry of peptide compounds in water (H2O) solutions. I. Spectral parameters of amino acid residue absorption bands. Biopolymers 30, 1243–1257 (1990).

NUMBER 4

APRIL 2009

389

ARTICLES

Bacterial frataxin CyaY is the gatekeeper of iron-sulfur cluster formation catalyzed by IscS


Salvatore 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 lethal neurodegenerative disease. It is believed that frataxin is an iron chaperone that participates in iron metabolism. We have tested this hypothesis using the bacterial frataxin ortholog, CyaY, and different biochemical and biophysical techniques. We observe that CyaY participates in iron-sulfur (Fe-S) cluster assembly as an iron-dependent inhibitor of cluster formation, through binding to the desulfurase IscS. The interaction with IscS involves the iron binding surface of CyaY, which is conserved throughout the frataxin family. We propose that frataxins are iron sensors that act as regulators of Fe-S cluster formation to fine-tune the quantity of Fe-S cluster formed to the concentration of the available acceptors. Our observations provide new perspectives for understanding FRDA and a mechanistic model that rationalizes the available knowledge on frataxin.

Friedreich’s ataxia (FRDA) is a relentlessly progressive neurodegenerative disease which leads to the death of the affected individuals. Although classified as rare, this recessive pathology has B1 carrier in every 120 individuals1. Discovery of the responsible gene in 1996 established that FRDA is caused by deficiency of frataxin, a small essential protein highly conserved from bacteria to humans2. In eukaryotes, frataxin is nuclearly encoded, translated in the cytoplasm and then imported into mitochondria, where it is finally matured3. The cellular function of frataxin remains controversial. It is commonly accepted that frataxin is involved in iron metabolism: partial depletion of frataxin has been shown to increase mitochondrial iron levels and to decrease the activity of Fe-S cluster proteins (reviewed in ref. 1). In vitro, frataxins from different species bind both Fe2+ and Fe3+ with a defined stoichiometry, although with relatively low affinity and specificity4–7. Bioinformatics, genetic and biochemical evidence has shown that frataxin binds to ferrochelatase5,8–10, and to essential components of the Fe-S cluster machinery11–13, thus implicating the protein in heme metabolism and Fe-S cluster formation. Two hypotheses have been suggested to explain the exact role of frataxin in these processes. According to one, frataxin and its orthologs are iron chaperones whose role is to provide the iron necessary for Fe-S and heme assembly14. The second, based on the capacity of the protein to form iron-loaded multimers in vitro, proposes an iron-storage function to scavenge the toxic iron in a sheltered, but readily available, form15. We have challenged these hypotheses and studied the effects of frataxin 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 in which the genes encoding the protein components of the Fe-S cluster machinery (but not frataxin) are grouped in the isc operon. Additionally, bacterial frataxin orthologs (the CyaY proteins) consist of only a conserved globular domain, thus avoiding possible complications caused by the presence of the mitochondrial import signal. Fe-S cluster formation is a complex enzymatic reaction, still not entirely understood, that requires several steps: conversion of cysteine into alanine with formation of a persulfide, its transfer to an acceptor and the formation of the cluster. We followed the effect of CyaY on the kinetics of the reaction in vitro using purified proteins. In this assay, the cluster is formed under strict anaerobic conditions on the scaffold protein IscU, either chemically, using sulfide and ferric ammonium citrate as sources of sulfur and iron, respectively, or enzymatically13,16,17. In the latter case, the desulfurase IscS transfers sulfur from cysteine onto IscU, with the concomitant uptake of iron to form a Fe-S cluster. IscU further transfers the Fe-S cluster to its final acceptors. Using this assay, we show that Escherichia coli frataxin is not merely an iron chaperone with a neutral involvement in the enzymatic process 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 of 2Fe-2S clusters. This function does not involve large aggregates of frataxin, which could store iron in a bioavailable form. Our work provides a new perspective on the cellular role of frataxin and suggests a molecular mechanism to explain FRDA.

1National

Institute for Medical Research, The Ridgeway, London, UK. 2Dipartimento di Biochimica e Biofisica, II Universita’ degli Studi di Napoli, Napoli, Italy. 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 Moe´culaire, Intitut National de la Recherche´ Agronomique, Jouy-en-Josas, France. Correspondence should be address to A.P. ([email protected]). 3Dipartimento


390

VOLUME 16

NUMBER 4

APRIL 2009


ARTICLES

c 10

0.35 CyaY No CyaY

0.30

6

0.20

4

mdeg

0.15

2

0.10 0 300 –2

0.05 0.00

400

–4 300

350

b 0.09

400 450 500 Wavelength (nm)

d

0.08

CyaY CyaY added at 180 s No CyaY

mdeg

4

0.05 0.04

3 2 1

0.02

Chemical reaction

0.01

0 0

0

60

0

50

0

40

0

0 10

0

0

0

80

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 by adding 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.

4,

0

20 4,

0

60

00 3,

3,

0

0 40

2,

0

80

20

1,

1,

0

–1 60

0

700

5

2+

Fe preloaded CyaY

0.03


6

Enzymatic reactions CyaY

0.06

7

600

Wavelength (nm)

600

No CyaY

0.07 Absorbance

550

500

30

Absorbance

CyaY No CyaY

8

0.25

20

a

Time (s)

Time (s)

RESULTS CyaY is an inhibitor of enzymatic Fe-S cluster formation We implemented an in vitro reconstitution assay by adapting published protocols18–21. Specific care was paid to choosing the appropriate concentrations of this complex, multicomponent system and to ascertain the state of folding of each component (see the Results section of Supplementary Methods online). Fe-S reconstitution on IscU was first monitored by absorbance spectroscopy: 2Fe-2S clusters are characterized by maxima at 400 nm and 456 nm and a shoulder at 510 nm, whereas 4Fe-4S clusters have a broad absorption centered at B390 nm18 (Fig. 1a). Formation of Fe-S clusters was detected as an increase of absorbance in the range of 400–550 nm and by the appearance of a brown-red color, as previously described. As expected, the IscS-mediated enzymatic reaction was both faster and more efficient than the non-enzymatic reaction (Fig. 1b). When CyaY was added to the enzymatic reaction, we observed two main effects: the absorption spectrum at the end of the reaction changed shape, with a marked attenuation of the band at 456 nm and almost complete disappearance of the shoulder at 510 nm (Fig. 1a). The absorbance ratio between the bands at 400 nm and 456 nm increased progressively as the reaction proceeded. In the presence of CyaY, the kinetics of the process were also distinctly slower (Fig. 1b and Supplementary Fig. 1 online). The brown-red color was observed also in this case, although it was less intense than in the absence of CyaY. To check whether the presence of CyaY could facilitate iron delivery, as would be expected for an iron chaperone, we compared the kinetics when iron was added to the reaction solution to those obtained when CyaY was preloaded with iron. We incubated a CyaY solution with an iron excess, either anaerobically (for Fe2+) or aerobically (for Fe3+), and then added the CyaY solution anaerobically to the enzymatic mixture. Apart from a small lag phase that may reflect iron availability, the kinetics with iron-preloaded CyaY were practically indistinguishable from those obtained by adding iron independently (Fig. 1b). This suggests that, if the effect of CyaY is linked to its iron binding properties, it does not depend on whether the protein takes iron from the solution or is the carrier, as expected from the weak affinities of CyaY both for Fe2+ and Fe3+ (ref. 22).


VOLUME 16

To check whether the observed effect was due to the presence of any agent able to bind iron and interfere with the enzymatic activity in a nonspecific way, we repeated the experiments replacing CyaY with an excess of either citrate or calmodulin. The first is a well-known iron chelator; the second has been seen to form iron-loaded aggregates in the presence of excess iron (S.A. and A.P., unpublished data). In neither case did we observe interfering effects on the kinetics of the enzymatic reaction (data not shown). Taken together, this evidence suggests that the role of CyaY is not simply that of an iron chaperone but that of an inhibitor. CyaY specifically inhibits formation of 2Fe-2S clusters We also used CD spectroscopy to follow the effect of CyaY on Fe-S cluster formation. This technique is complementary to absorbance spectroscopy as it avoids complications due to the overlapping absorption spectra of other iron-bound components. It is also sensitive to the cluster nuclearity: 2Fe-2S and 4Fe-4S clusters can clearly be distinguished by CD. 2Fe-2S gives intense contributions in the range 300–700 nm. 4Fe-4S clusters, with the exception of 2[4Fe-4S] bacterial ferredoxins23, have weak bands that can barely be distinguished from the baseline at the concentrations used here. Enzymatic cluster formation on IscU showed clear evidence for formation of a 2Fe-2S cluster with maxima at 340 nm, 430 nm and 510 nm, and minima at 370 nm and 560 nm, in agreement with the literature20,21,24 (Fig. 1c). In the presence of CyaY, the rate of 2Fe-2S cluster formation was much slower and the reaction efficiency greatly reduced, as shown by the strong drop (480%) in the plateau intensity (Fig. 1d). Minor differences in the kinetics of the different experiments were due to several factors. Different enzyme preparations had slightly different activities. CD measurements were overall slightly faster because, by this technique, we followed only one pathway and used larger IscU–IscS ratios. This was possible because, for technical reasons, the reaction could be initiated directly in the spectrophotometer, thus the early events of the reaction were not lost. This could not be done for the absorbance measurements, for which we had to transfer the cuvette from the chamber to the spectrophotometer. The inhibitory effect of CyaY seemed to be stronger when monitored by CD than by absorbance spectroscopy. This is probably because CD monitors only formation of 2Fe-2S clusters on IscU, whereas absorbance also detects other iron-containing species.

NUMBER 4

APRIL 2009

391

ARTICLES

Non-enzymatic IscU reconstitution

6

CyaY No CyaY

5

5

mdeg

3 2

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.

Cluster transfer from chemically loaded IscU to Fdx CyaY No CyaY

6

4

mdeg

c7

4 3

1 2

0

b

100

0

7

500

200 300 400 Time (s)

4

mdeg

200

0.04

CyaY No CyaY

5

0

d 0.045

Enzymatic Fdx reconstitution

6

1

600

Absorbance

–1

3 2 1

800

400 600 Time (s)

Dosage of Ala produced No CyaY [Ala] CyaY

0.035 [Ala] 0.03 0.025

0

0 60 0 1, 20 0 1, 80 0 2, 40 0 3, 00 0 3, 60 0

60 0

50 0

40 0

30 0

20 0

10 0

0

Time (s)

Time (s)

To confirm that inhibition is due to CyaY and is independent of the order in which the components are added, we recorded kinetics in which CyaY was injected into the mixture only after starting the reaction (Fig. 1d). Again, we observed no appreciable differences in the presence of CyaY preloaded with iron or when Fe2+ was used instead of Fe3+ (data not shown). These results indicate that CyaY has a strong and specific inhibitory effect on the formation of 2Fe-2S clusters. Dissecting Fe-S cluster assembly into its components During Fe-S cluster assembly, IscS-mediated formation of a cluster on IscU is followed by cluster transfer from IscU to the final acceptor. The second process is known to be slower than the first24. To assess which step is affected by CyaY, we dissected the pathway as follows. First, we carried out non-enzymatic (chemical) cluster formation on IscU in the presence and in the absence of CyaY and followed the process by CD (Fig. 2a). The absence of any observable effect indicates that CyaY intervenes 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

50

N (p.p.m.)

GST

40 35

21 14

2.5 5 7.5 10 12.5 [IscS] (µM)

0.12 0.10

46

115

Trp61

21

28

23

120

Glu18

Glu19

29

130

42 22 30 35 34 43 33

135

39

44

125

D22 0

e

His7

Asp76

110

CyaY

45

d 105

No CyaY

15

31

IscS

c 55

Fluorescence

66 45

M

MW (kDa)

b ar ke r ST s G co S n pu T-C trol ll- ya do Y w n

a G


protein can form Fe-S clusters in the presence of IscS or other desulfurases, also in the absence of IscU23. (Fig. 2b). We observed a strong decrease in the kinetics in the presence of CyaY. The effect was comparable with that observed in the presence of IscU as a scaffold (Fig. 1d), indicating that the inhibitory effect of CyaY involves IscS and not IscU. To confirm these results and to assess further the role of IscS, we analyzed cluster transfer from chemically reconstituted holo IscU (that is, Fe-S cluster–loaded IscU) to Fdx in the presence and in the absence of CyaY (Fig. 2c). Transfer is known to occur when the two proteins are mixed because of the lower affinity of IscU for the cluster21. Because the CD signal of cluster-loaded Fdx is more intense than that of cluster-loaded IscU24 (Supplementary Fig. 2 online), the reaction led 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 tested whether CyaY acts on the IscS desulfurase activity by assessing the efficiency of enzymatic cysteine-to-alanine conversion. We started the reaction in the presence of CyaY (5 mM) using low cysteine concentrations (10 mM). When the reaction reached a plateau, we injected additional cysteine to reach a total added concentration at 20 mM, causing the reaction to start again and to proceed with the same rate observed after the first addition (Fig. 2d). We collected aliquots of the mixture at several time points. The alanine concentration, as estimated by amino acid analysis, increased steadily and at the two plateaus was

0.02

–1

10

9

8 H (p.p.m.)

Absorbance

a7

0.08

0.04 0.02

Asp31 0.00

7

1

No CyaY CyaY_D31K CyaY_E19KD22K CyaY_W61R CyaY_H7KD76K CyaY

0.06

0

1,200

2,400 3,600 Time (s)

4,800

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 of an 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.

392

VOLUME 16

NUMBER 4

APRIL 2009


ARTICLES

Absorbance

0.25

[Fe2+] 100 µM

0.08 0.07

0.2 0.15 0.1

25 µM

0.05

10 µM 5 µM

0.06 0.05 0.04 0.03 0.02 0.01

1, 20 0 1, 80 0 2, 40 0 3, 00 0 3, 60 0 4, 20 0 4, 80 0

60

0

0 0

Leu21, Asn35, Leu39, Thr42, Gly46, Lys48, Thr64 and Gln98. These residues all map onto the protein surface that contains the iron binding sites6 (Fig. 3d). The effect of CyaY on IscS is therefore the consequence of a direct interaction between the two molecules that does not compete with IscU binding.

b 0.09

0

2

4 6 8 10 12 14 [CyaY] (µM)

Time (s)


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 using 50 mM IscU, 250 mM cysteine, 3 mM DTT and 1 mM IscS.

10.0 ± 0.8 mM and 18.9 ± 0.4 mM, respectively, showing that all the cysteine initially added has been converted into alanine. These results indicate that inhibition of 2Fe-2S cluster formation results from a direct effect of CyaY on IscS and is independent of the nature of the acceptor. CyaY does not inhibit the desulfurase activity of IscS because the enzyme continues converting its substrate into alanine until it is completely consumed. CyaY forms a specific complex with IscS Our 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 overexpressed isc operon proteins13. To strengthen these results, we tested whether a pull-down also ‘fished out’ the endogenous proteins: we bound GST-tagged CyaY to a glutathione Sepharose column and then passed an E. coli extract through the column. Under these conditions, we detected the presence of endogenous IscS but not IscU (Fig. 3a). Because IscU also interacts with IscS, we tested whether CyaY competes with IscU for the same site on IscS using fluorescence spectroscopy. As all three proteins contain tryptophan residues, we attached a fluorophore, AlexaFluor532, to the more reactive sulfhydryl group on IscU. The monophasic curve obtained by titrating labeled IscU (IscU*) with IscS confirmed the presence of a direct interaction between IscU and IscS with a dissociation constant consistent with the value previously reported (1 mM)20 (Fig. 3b). The fluorescence intensities obtained by individually titrating IscU* with IscS in the presence of an excess of CyaY (apo or Fe3+ loaded) were superimposable with those obtained in the absence of CyaY, suggesting an absence of competition. We next used NMR to map the surface of interaction on CyaY. When iron-free and iron-preloaded 15N labeled CyaY was titrated aerobically or anaerobically with unlabeled IscU, no spectral perturbations were observed (data not shown). In contrast, titration of CyaY with IscS produced two effects: a progressive disappearance of the whole CyaY spectrum, as expected from formation of the large molecular complex of CyaY with the IscS dimer (90 kDa), and a chemical shift perturbation that specifically affected a limited number of peaks (Fig. 3c). The residues strongly affected already at a 1:0.5 CyaY:IscS molar ratio were Asp22, Asp23, Ser28, Asp29, Glu33, Ile34, Phe43 and Glu44. Other residues affected at higher CyaY:IscS molar ratios (1:0.75) were


VOLUME 16

CyaY mutations affect cluster reconstitution kinetics To explore the role of CyaY residues in or around the surface of IscS binding, we produced different mutants, chosen among those already well characterized and known to have an effect in vivo or in vitro. CyaY_D22K, CyaY_D31K, CyaY_E19K D22K and CyaY_E18K E19K D22K affect residues that are completely or partially conserved and are known to be involved in the main iron binding site6,25,26. The equivalent mutations in yeast frataxin led to progressively severe phenotypes in vivo27. We also tested a CyaY_W61R mutant because the equivalent position in human frataxin is associated with a severe FRDA case28. Although not immediately perturbed in our titration with IscS, this residue is in a region contiguous to the iron binding interface. A mutant of residues not perturbed by IscS titration and far away from the iron binding surface (CyaY_H7K D76K)4 was used as a control. We had previously confirmed that these mutants retain their fold4,29. With the exception of the negative control CyaY_H7K D76K, which behaved like the wild type, the mutants showed progressively faster kinetics of Fe-S formation on IscU, with initial rates similar to that observed in the absence of CyaY (Fig. 3e). CyaY_W61K has a comparatively smaller effect. These data validate the surface for CyaY–IscS interaction and confirm a role of CyaY in Fe-S cluster reconstitution. The effect of CyaY is iron-concentration dependent As the surface of CyaY interacting with IscS is also involved in iron binding, we tested whether CyaY’s inhibitory effect could be sensitive to variations in iron concentration. The initial reaction rates depend on the iron concentration, both in the presence and in the absence of CyaY, which is expected because iron is a substrate in the overall cluster-formation reaction (Fig. 4a). However, the difference in rates at the same Fe2+ concentration in the absence (closed symbols) and in the presence (open symbols) of CyaY becomes much more marked at higher iron concentrations. CyaY in 20 mM Tris-HCl, 100 mM NaCl 2+ CyaY+Fe (1:20) in 20 mM Tris-HCl, 100 mM NaCl 2+ CyaY+Fe (1:20) in 20 mM HEPES, no salt

600 500

Monomer

400

mAU

Closed symbols: CyaY 0.3 Open symbols: no CyaY

Initial rate

a

Iron-induced aggregates

300 200 100 0 4

6

8

10

12 14 Volume (mI)

16

18

20

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.

NUMBER 4

APRIL 2009

393

ARTICLES


As an independent control, we measured the rate of cluster formation as a function of CyaY concentration, while keeping the iron and enzyme concentrations constant (Fig. 4b). We observed a decrease in the rate of cluster formation with increasing CyaY concentrations. The effect saturated at 5 mM CyaY, much lower than the IscU concentration used in our assays but close to the concentration of IscS, confirming that CyaY affects IscS rather than IscU. These results indicate that the CyaY effect depends on the amount of iron present in the environment. CyaY inhibition does not require formation of multimer Frataxin has been suggested to act as a ferritin-like iron scavenger by forming large, spherical polymers of distinct stoichiometry15. We used gel filtration to test whether oligomerization of CyaY could be important for the observed effects (Fig. 5). Freshly prepared ironfree CyaY eluted as a monomer, as previously described4. Under the conditions of the kinetic measurements, CyaY eluted again as a monomer. We did not observe species of large molecular weight, even when the protein was treated with an Fe2+ excess to create conditions favorable for oligomers formation4. Detectable amounts of oligomers could be observed only when the experiment was carried out at low ionic strength, in the absence of DTT and using HEPES rather than Tris-HCl. These results indicate that the effect of CyaY is not linked to formation of oligomeric species. It must be completely attributed to the monomeric form of CyaY and its stoichiometric interaction with IscS. DISCUSSION We have investigated the effect of frataxin in the enzymology of Fe-S cluster biogenesis using purified proteins from E. coli. In using the bacterial system, we postulated that, given the high homology and structural conservation within the frataxin family, the main features of the mechanism(s) by which frataxins functions must be the same, albeit with species-specific adaptations. This assumption is supported by complementation 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 presence of CyaY should either enhance the enzymatic rates (if the rate-limiting step of the reaction depended on iron delivery) or otherwise have no effect. We observed instead that, far from facilitating iron delivery, the presence of CyaY inhibits the reaction. By dissecting the complex pathway of Fe-S cluster formation, we have shown conclusively that IscS is indispensable for the CyaY action and that the effect of CyaY is mediated through a direct interaction with IscS, in agreement with a previous report13. Although we cannot in principle exclude an interaction of CyaY also with IscU in the context of the tertiary complex (IscU–IscS–CyaY), we see that CyaY has little or no effect on the kinetics of the processes involving IscU or other scaffold proteins. CyaY does not alter the IscS desulfurase activity but is an inhibitor of cluster formation. Finally, we have shown 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-S formation was not observed earlier. In vivo experiments, although crucial for the identification of the hallmarks of the disease, could not provide details on the mechanism of the process. Most of the previous in vitro work has, on the other hand, compared the effect of frataxin on the chemical reaction, thus missing the most important component, the desulfurase enzyme4,32. The only report that compares the effect of CyaY on the enzymatic reaction was carried out by incubating

394

VOLUME 16

IscS with the substrate (cyteine) for 2 h before adding Fe3+-preloaded CyaY13. Under these conditions, cysteine is almost completely converted into alanine, as we have observed (Fig. 2d). The mapped IscS interface includes CyaY residues that are highly conserved and that have been implicated in iron binding, indicating that our overall conclusions can be generalized, although there might be some degree of species variability. Accordingly, we observed an iron-dependent effect of CyaY on Fe-S cluster formation, in agreement with the observation that, in yeast, binding of the frataxin ortholog Yfh1 with the Isu1–Nfs1 complex (equivalent to IscU–IscS) is iron dependent11. Notably, an iron-dependent effect of the activity of ferrochelatase, the enzyme involved in another frataxin pathway, has also been reported10. We propose that frataxins function as iron sensors and suggest a mechanism for their action (Fig. 6). By a negative-regulation mechanism, frataxins would act as gate keepers of Fe-S assembly by fine tuning the quantity of Fe-S clusters formed to match the concentration of the apo acceptors. Frataxins would have low affinity for the IscS–IscU system at normal iron levels. At any even small iron imbalance (that is, an excess of iron as compared to final acceptors), the affinity of the protein for IscS would increase. One may wonder why such a regulation mechanism is needed, considering that the isc operon is under the control of the transcription factor IscR33. The necessity of regulation both at the transcriptional and posttranslational levels can be explained by considering that IscR regulates the whole operon, and its action will require more time than the immediate response that a component that interacts directly on the central component of the machinery, the enzyme, could have by sensing the iron concentrations. In FRDA, and even more so in knockout models, where the phenotype is exacerbated, such a regulatory mechanism would be absent. As a regulator that tunes the quantity of Fe-S cluster formed to the availability of the apo acceptors, any reduction or depletion of frataxin levels would upset this equilibrium and lead to an imbalance in the amount of the Fe-S clusters produced with respect to the apo acceptors. Even a small iron accumulation could result in the formation of Fe-S clusters at a rate incompatible with the concentration of final acceptors. Fe-S clusters are labile species that cannot exist without an acceptor or carrier. They would therefore fall apart, producing free iron, which in turn would give rise to Fenton chemistry. The FRDA phenotype, including the early damage of Fe-S cluster proteins observed both in mammalian and yeast knockouts, could be a direct, early consequence of this process8,34. The resulting free Fe3+, which is highly insoluble, would first form small and amorphous nanoparticles and eventually precipitate, generating the large deposits detected in vivo8,9,34. Our model provides an answer to several observations that are currently unexplained. Neither the molecular chaperone nor the ferritin-like hypotheses14,15 fully account for the sequence conservation26 and the essential nature of frataxins34. Our CyaY mutations of conserved or semiconserved residues explain the severe phenotype of equivalent mutants in yeast27,35. One of the main difficulties with the ferritin-like hypothesis is the presence of a mitochondrial ferritin36. Iron transport in mitochondria, where large amounts of citrate and other iron transporters are present, is not a satisfactory answer. An iron-sensor role instead explains frataxin’s low affinity for iron and the long-term accumulation of the FRDA symptoms: a sensor requires weak affinities. Even small frataxin concentrations would be sufficient for cell viability, but frataxin deficiency would eventually trigger long-term catastrophic effects. It also becomes clear why frataxin is associated with oxidative stress and why time-dependent

NUMBER 4

APRIL 2009


ARTICLES

a

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.

In the presence of normal iron concentrations Frataxin Final acceptor

IscU IscS dimer

Cluster-loaded final acceptor

+ Cys Ala

b

In the presence of any iron surplus

+

Slower rates Cys Ala

Iron-loaded frataxin

c

In conditions of frataxin deficiency

+

No rate control

Fenton chemistry


Cys Ala Surplus of Fe-S clusters

Insoluble iron precipitates

intramitochondrial iron accumulation in frataxin-deficient organisms is observed after the onset of the pathology and after inactivation of the Fe-S–dependent enzymes. Finally, a role for the CyaY monomer rather than an aggregate is in agreement with previous work showing that an oligomerization-deficient mutant of Yfh1 can still participate in Fe-S cluster biogenesis or heme assembly37. We believe that, although more work is needed to establish the details of the molecular mechanism of IscS inhibition exerted by CyaY and provide a direct description of the eukaryotic system, our work opens an entirely new perspective to understanding the role of frataxins. This will hopefully promote new studies to clarify its links with the FRDA pathology. METHODS Protein 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 concentrations 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.


VOLUME 16

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 monitored 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 reconstitution 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 reconstituted 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

NUMBER 4

APRIL 2009

395

ARTICLES 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 & Molecular Biology website. ACKNOWLEDGMENTS We 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 the Mill Hill NMR Centre for technical support. The project was supported by PUR funds (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 an intronic GAA triplet repeat expansion. Science 271, 1423–1427 (1996). 3. Koutnikova, H. et al. Studies of human, mouse and yeast homologues indicate a mitochondrial function for frataxin. Nat. Genet. 16, 345–351 (1997). 4. Adinolfi, S., Trifuoggi, M., Politou, A.S., Martin, S. & Pastore, A. A structural approach to 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 the iron 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 associated with 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 frataxin homologue (Yfh1). Hum. Mol. Genet. 12, 879–889 (2003). 10. Yoon, T. & Cowan, J.P. Frataxin-mediated iron delivery to ferrochelatase in the final step of heme biosynthesis. J. Biol. Chem. 279, 25943–25946 (2004). 11. Gerber, J., Muhlenhoff, U. & Lill, R. An interaction between frataxin and Isu1/Nfs1 that is crucial for Fe/S cluster synthesis on Isu1. EMBO Rep. 4, 906–911 (2003). 12. Ramazzotti, A., Vanmansart, V. & Foury, F. Mitochondrial functional interactions between frataxin and Isu1p, the iron-sulfur cluster scaffold protein, in Saccharomyces cerevisiae. FEBS Lett. 557, 215–220 (2004).

396

VOLUME 16

13. Layer, G., Ollagnier-de Choudens, S., Sanakis, Y. & Fontecave, M. Iron-sulfur cluster biosynthesis: characterization of Escherichia coli CYaY as an iron donor for the assembly 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 an iron 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: implications 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, and formation 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 of iron 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 for iron-sulfur cluster biosynthesis: sequential assembly of [2Fe-2S] and [4Fe-4S] clusters in IscU. Biochemistry 39, 7856–7862 (2000). 19. Yuvaniyama, P., Agar, J.N., Cash, V.L., Johnson, M.K. & Dean, D.R. NifS-directed assembly 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 from IscS 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 of Clostridium 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 solution structure of frataxin. Structure 8, 695–707 (2000). 26. Pastore, C., Franzese, M., Sica, F., Temussi, P. & Pastore, A. Understanding the binding 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 essential role in Fe-S cluster assembly. EMBO Rep. 8, 194–199 (2007). 28. Labuda, M., Poirier, J. & Pandolfo, M. A missense mutation (W155R) in an American patient with Friedreich Ataxia. Hum. Mutat. 13, 506–509 (1999). 29. Correia, A.R., Adinolfi, S., Pastore, A. & Gomes, C. Conformational stability of human frataxin 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 functions between eukaryotic frataxin and the Escherichia coli frataxin homolog CyaY. FEMS Yeast Res. 7, 1276–1284 (2007). 31. Cavadini, P., Gellera, C., Patel, P.I. & Isaya, G. Human frataxin maintains mitochondrial 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, represses expression 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 early embryonic lethality without iron accumulation. Hum. Mol. Genet. 9, 1219–1226 (2000). 35. Leidgens, S. The function of yeast frataxin in iron-sulfur cluster biogenesis: a systematic 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 yeast frataxin homologue Yfh1 is dispensable in vivo. EMBO Rep. 5, 1096–1101 (2004).

NUMBER 4

APRIL 2009


ARTICLES

The pathway of hepatitis C virus mRNA recruitment to the human ribosome


Christopher 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 controlled by translation-initiation factors. Some viruses use an internal ribosome entry site (IRES) in their mRNA to harness ribosomes independently of initiation factors. We show here that a ribosome conformational change that is induced upon hepatitis C viral IRES binding is necessary but not sufficient for correct mRNA positioning. Using directed hydroxyl radical probing to monitor the assembly of IRES-containing translation-initiation complexes, we have defined a crucial step in which mRNA is stabilized upon initiator tRNA binding. Unexpectedly, however, this stabilization occurs independently of the AUG codon, underscoring the importance of initiation factor–mediated interactions that influence the configuration of the decoding channel. These results reveal how an IRES RNA supplants some, but not all, of the functions normally carried out by protein factors during initiation of protein synthesis.

Protein synthesis in all cells begins with binding and positioning of an mRNA on the small ribosomal subunit. In eukaryotes, the 5¢ 7-methyl guanosine cap structure on most mRNAs triggers an assembly of translation factors that recruit the 40S ribosomal subunit. This highly regulated process, leading to 60S subunit joining and formation of an active 80S ribosome, requires at least 12 initiation factors composed of roughly 28 polypeptides1,2. Some viral mRNAs lack a 5¢ cap and instead include a structured RNA sequence, the IRES, in the 5¢ untranslated region (UTR), which functions in place of some or all of the canonical initiation factors3,4. Some of the most detailed information to date for this type of mechanism has come from the study of the IRES found in the hepatitis C virus (HCV) mRNA5,6. An efficient 40S subunit initiation complex on the HCV IRES requires only two initiation factors, eIF2 and eIF3 (ref. 7). The eIF2 complex recruits the initiator tRNA (Met-tRNAi), whereas the much larger eIF3 complex enhances formation of the 40S subunit initiation complex on the 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 the head, body and platform of the 40S subunit. This binding region comprises the channel through which the mRNA enters the 40S subunit, the ribosome decoding sites (aminoacyl (A) site, peptidyl (P) site and exit (E) site) and the exit channel through which the mRNA leaves the 40S subunit11 (Fig. 1a). The entry channel is occluded 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 a structural change in 40S subunits bound to the wild-type IRES whereby the mRNA entry channel seemed to be more open relative to 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 of domain II for mRNA entry into the binding channel in the absence of initiation factors15,16. Subsequent cryo-EM reconstructions of the 40S subunit bound to the cricket paralysis viral IRES or to initiation factors eIF1 and eIF1A showed similar conformational changes in the 40S subunit head, hinting at a common mechanism of mRNA loading by viral IRESs and cap-dependent initiation factors14,17. Another clue to the mechanism of mRNA loading came from the discovery that a subunit of eIF3, eIF3j, binds stably to 40S subunits only in the absence of mRNA9,18. Directed hydroxyl radical probing showed that the C terminus of eIF3j lies in the 40S mRNA entry channel, where it presumably disfavors mRNA binding in the absence of other initiation factors19. Upon initiator tRNA recruitment, as part of the ternary complex, eIF3j is displaced and mRNA binding is enhanced. How viral IRESs trigger this crucial switch, leading to proper positioning of the viral mRNA on the 40S subunit, is unknown. To address this question, we used directed hydroxyl radical probing and ribosome toeprinting of reconstituted translation-initiation complexes to determine the steps required for HCV IRES–mediated

1Howard Hughes Medical Institute, University of California, Berkeley, California 94720, USA. 2Department of Biochemistry and Molecular Medicine, School of Medicine, 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: Section of Molecular and Cellular Biology, College of Biological Sciences, University of California, Davis, California 95616, USA. Correspondence should be addressed to J.A.D. ([email protected]).

Received 29 September 2008; accepted 3 February 2009; published online 15 March 2009; doi:10.1038/nsmb.1572


VOLUME 16

NUMBER 4

APRIL 2009

397

40S EDTA/Fe –cys+EDTA/Fe –cys+HCV+EDTA/Fe –cys+HCV∆II+EDTA/Fe S152C S152C+HCV S152C+HCV∆II S217C S217C+HCV S217C+HCV∆II T235C T235C+HCV T235C+HCV∆II A241C A241C+HCV A241C+HCV∆II

ARTICLES Figure 1 Directed hydroxyl radical probing of IIIb 18S rRNA from BABE-Fe–eIF3j–40S–HCV IRES complexes. (a) The 40S subunit structure based mRNA exit channel on a cryo-EM reconstruction13 viewed from the h bk subunit interface with landmarks indicated: IIIa E pt IIIc P A, A-site; P, P-site; E, E-site; bk, beak; b, body; C T AG mRNA A entry channel Lane pt, platform; h, head. (b) The 5¢ UTR of the HCV II b mRNA consists of four domains (I–IV); the IRES IIId domains (II–IV) with subdomains (a–f) of domain I III are indicated. (c) Representation of eIF3j, IIIe 40S 1486–1491 5′ indicating the positions of cysteine mutations IIIf used for BABE-Fe attachment (above). Modeled 3′ 1502–1504 IV positions of eIF3j amino acids in the Thermus 1508 thermophilus 30S crystal structure, adapted from eIF3j a previous publication19 (below). The boxed area 152 217 235 241 258 1 provides a detailed view of the mRNA entry channel and A-site with helices 18, 32 and 34 1523–1526 indicated. Nucleotides cleaved in these helices for each experiment are shown in Supplementary h32 Fig. 1a. (d) Primer-extension analysis of 18S h34 rRNA cleaved by BABE-Fe–modified eIF3j. h18 Sequencing lanes are indicated (C, T, A and G). + OR eIF3j Other control lanes include 40S subunits in the absence or presence of EDTA/Fe, mock30S derivatized eIF3j ( cys+EDTA/Fe), in the HCV∆II 40S HCV 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.

b

d

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

a


c

mRNA positioning in the 40S decoding channel. We established that mRNA and the C terminus of eIF3j do not bind simultaneously in the ribosome entry channel and that eIF3j displacement signals mRNA entry and binding in the channel. Using this as an assay for initiation complex formation, we show that the 40S conformational change induced by the IRES domain II is necessary but not sufficient for mRNA entry into the decoding channel. In addition, both the mRNA strand downstream of the AUG codon and the ternary complex are required for the mRNA to displace eIF3j. Notably, this effect does not require AUG recognition by the initiator tRNA, implying that correct mRNA positioning is a function of ribosome conformation rather than mRNA-tRNA base pairing. Our results support an ordered pathway for HCV IRES–mediated translation initiation in which an eIF3- and IRES-stabilized conformational state of the 40S subunit favors viral mRNA entry into the decoding channel but only after ternary complex binding. RESULTS IRES domain II promotes mRNA entry into the binding channel Previous studies showed that initiation factors eIF1, eIF1A, eIF2 and eIF3 are necessary for stable mRNA positioning in the ribosomal decoding channel, as monitored by toeprinting and reduced eIF3j affinity to the 40S subunit19,20. To test whether the HCV IRES similarly favors mRNA binding and eIF3j displacement from the mRNA binding channel, we monitored 18S ribosomal RNA (rRNA) cleavages induced by bromoacetamidobenzyl-EDTA/Fe (BABE-Fe)modified eIF3j proteins19 (Fig. 1c). As shown previously, eIF3j proteins containing a single BABE-Fe moiety at several positions near the C terminus lead to site-specific rRNA cleavages in the mRNA entry channel (helix 34) in complexes containing only the 40S subunit and eIF3j. Addition of the wild-type HCV IRES mRNA to this complex largely prevented these cleavages (Fig. 1d and

398

VOLUME 16

Supplementary Fig. 1b online). This observation could indicate that the C terminus of eIF3j dissociates from the 40S mRNA binding cleft upon HCV mRNA association. Consistent with this idea, we observed no cleavage of the mRNA segment of the HCV IRES RNA in IRES–40S–eIF3j complexes, suggesting that eIF3j is no longer in the channel (data not shown). To distinguish between competitive binding of eIF3j and HCV mRNA and HCV mRNA–induced protection of 18S rRNA cleavage sites, we mapped the 40S subunit position on the IRES mRNA by toeprinting21–23. Consistent with previous results7,15, the association of the HCV mRNA with the 40S subunit induces a toeprint at nucleotides +20 and +21 downstream of the AUG codon, reflecting the leading edge of the 40S subunit on the mRNA (Fig. 2a). This toeprint was inhibited upon addition of increasing concentrations of eIF3j, indicating that the association of mRNA with the entry channel requires eIF3j displacement. Notably, the toeprint located at nucleotides +3 and +4 is still evident, even at a large molar excess of eIF3j (5–40 mM), suggesting that eIF3j influences only mRNA association with the entry channel and not HCV mRNA association with the rest of the 40S subunit. This effect probably explains our previous data indicating that a short mRNA can bind concurrently with eIF3j19. As the conformational changes induced by domain II of the HCV IRES have been suggested to facilitate mRNA entry into the binding channel, we tested an HCV mRNA lacking this domain (denoted as HCVDII in figures) in our cleavage assay. Previous studies showed that this truncated form of the HCV IRES binds with similar affinity to the 40S subunit but fails to induce efficient translation initiation7,24. We found that the cleavages induced at nucleotides 1486–1491 in helix 34 by BABE-Fe–modified C-terminal positions on eIF3j were restored (Fig. 1d), suggesting that domain II is necessary for eIF3j displacement from, and mRNA binding to, the decoding channel. However, it is not possible from these data to determine the order in which these events

NUMBER 4

APRIL 2009


ARTICLES

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

HCV HCV+40S

b

40S EDTA/Fe –cys+EDTA/Fe –cys+HCV∆ORF+EDTA/Fe S152C S152C+HCV∆ORF S217C S217C+HCV∆ORF T235C T235C+HCV∆ORF A241C A241C+HCV∆ORF

channel with mRNA. To test this possibility, we analyzed BABE-Fe–modified eIF3j– induced cleavage of 18S rRNA in reconstituted 40S–eIF3–HCV IRES mRNA complexes. This was achieved by the addition of BABE-Fe–modified eIF3j together with C T AG an eIF3 complex purified without endogenLane +elF3j CT AG ous eIF3j attached (eIF3Dj). In contrast to 1 2 3 4 5 6 7 8 9 10 Lane complexes containing the eIF3j subunit 3′ alone, those containing intact eIF3 yielded +3/4 similar cleavage patterns to those observed 1486–1491 40S in the absence of the HCV IRES mRNA 5′ (HCV) A P E (Fig. 3a). Similar experiments using 1502–1504 Entry channel Decoding site Exit channel HCVDII produced small, but reproducibly 1508 enhanced, cleavage intensities in rRNA helix ATG +3/4 34 relative to wild-type HCV IRES mRNA (Fig. 3a). These results suggest that the eIF3 +20/21 complex enables eIF3j to compete more 1523–1526 +20/21 40S effectively with HCV IRES mRNA for bindGUA 5′ (HCV) 3′ ing to the mRNA entry channel, particularly A P E 40S Entry channel Decoding site Exit channel in the absence of the IRES domain II– induced 40S conformational change. These + data raise the question of how the HCV elF3j mRNA can efficiently associate with the mRNA entry channel in the presence of HCV∆ORF eIF3j and the eIF3 complex during the process of translation initiation. Figure 2 Toeprinting analysis of the 40S–HCV–eIF3j complexes. (a) Lanes C, T, A and G depict Previously, we showed that ternary comsequencing lanes corresponding to HCV mRNA, with the AUG codon indicated. Toeprinting reactions of plex association with the 40S subunit HCV mRNA in the absence (lane 5) or presence (lane 6) of 40S subunits is shown. Additional reactions enhances the affinity of a short, unstructured including 40S subunits in the presence of 5 mM (lane7), 10 mM (lane 8), 20 mM (lane 9), or 40 mM mRNA for the 40S subunit in the presence of (lane 10) eIF3j are indicated. The positions of toeprints that correspond to 40S–HCV complexes are eIF3j (ref. 19), hinting at a role for the ternary indicated (+3/4 and +20/21). Numbering is from the A (+1) of the AUG codon. Cartoons depicting the complex in displacing eIF3j during mRNA 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 loading. To test this possibility, we used sitecodon (HCVDORF). Sequencing and control lanes are indicated, as described in Figure 1d. The lanes directed hydroxyl radical probing to detercorresponding to eIF3j in the absence (lanes 9, 11, 13 and 15) or presence (lanes 10, 12, 14 mine the effect of the ternary complex on and 16) of HCVDORF are indicated. As described in Figure 1d, the colored circles correspond to HCV IRES mRNA association in the 40S the components added in each reaction. subunit mRNA entry channel in the presence of eIF3j and intact eIF3Dj. Unfortunately, it was not possible to use the cleavage site that occur. We next tested an otherwise wild-type version of the HCV IRES we have described in helix 34 for this purpose because association of mRNA truncated after the AUG codon to determine whether the the ternary complex alone protects helix 34 from eIF3j-induced rRNA cleavage sites induced by BABE-Fe–modified eIF3j are generated cleavage (Supplementary Fig. 2a online). This occurs even though when the mRNA strand does not extend into the eIF3j binding site on eIF3j is still present in the mRNA entry channel, as indicated by the the 40S subunit. Although this shortened mutant associates with the unchanged cleavage intensity of helix 18 by BABE-Fe–modified eIF3j 40S subunit (data not shown), the rRNA cleavage sites were essentially in the presence of ternary complex (Supplementary Fig. 2b). Cleavage unchanged relative to those observed for 40S–eIF3j complexes of helix 18 by BABE-Fe–modified eIF3j was unchanged by the (Fig. 2b). This suggests that, in addition to the conformational association of eIF3Dj with the 40S subunit in the absence (Supplechanges induced by domain II, the presence of mRNA in the entry mentary Fig. 2c), or presence (Supplementary Fig. 2d), of the HCV channel and A-site of the 40S subunit is required to promote eIF3j mRNA, and this allowed visualization of this region to determine displacement. It should be noted that domain II does not stably whether any ternary complex–specific changes occur during HCV associate with the 40S subunit in the absence of other domains of the mRNA association. Upon recruitment of the ternary complex to the 40S subunit in the presence of HCV mRNA and eIF3Dj, cleavage HCV IRES. intensities at all nucleotide positions in helix 18 in the mRNA entry channel were diminished (Fig. 3b). This observation implies that, eIF3 regulates HCV mRNA entry into the binding channel Whereas the C-terminal half of eIF3j is located in the mRNA once the ternary complex associates, even in the presence of intact binding channel of the 40S subunit, its N terminus interacts with eIF3, HCV IRES mRNA enters the entry channel and eIF3j is the eIF3b subunit of eIF325. If maintained in the presence of the displaced. Notably, this effect was not seen in complexes containing 40S–eIF3–HCV IRES mRNA complex, this interaction should HCVDII (Fig. 3b). One explanation for these results is that AUG recognition by the increase the local concentration of eIF3j on the 40S subunit and thus enhance eIF3j’s ability to compete for binding to the entry initiator tRNA stabilizes mRNA on the 40S subunit. To test this, we

a

G


UA


VOLUME 16

NUMBER 4

APRIL 2009

399

b

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

C T AG

40S EDTA/Fe –cys+EDTA/Fe –cys+HCV+elF3∆j+EDTA/Fe –cys+HCV∆II+elF3∆j+EDTA/Fe S217C S217C+HCV+elF3∆j S217C+HCV∆II+elF3∆j T235C T235C+HCV+elF3∆j T235C+HCV∆II+elF3∆j A241C A241C+HCV+elF3∆j A241C+HCV∆II+elF3∆j

a

C T A G

40S EDTA/Fe –cys+EDTA/Fe –cys+HCV+elF3∆j+TC+EDTA/Fe –cys+HCV∆AUG+elF3∆j+TC+EDTA/Fe –cys+HCV∆II+elF3∆j+TC+EDTA/Fe S217C S217C+HCV+elF3∆j+TC S217C+HCV∆AUG+elF3∆j+TC S217C+HCV∆II+elF3∆j+TC T235C T235C+HCV+elF3∆j+TC T235C+HCV∆AUG+elF3∆j+TC T235C+HCV∆II+elF3∆j+TC A241C A241C+HCV+elF3∆j+TC A241C+HCV∆AUG+elF3∆j+TC A241C+HCV∆II+elF3∆j+TC

ARTICLES

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 Lane

Lane

601–609

1486–1491

616

1502–1504 1508

626–628


1523–1526

40S

elF3j

636–639

40S

+

TC

+

OR elF3j

+

+

OR

OR

elF3∆j HCV

HCV∆II

elF3∆j HCV

HCV∆AUG

mutated the AUG codon and surrounding nucleotides (ATCATG to AAAAAA) in domain IV of the HCV IRES and analyzed complexes containing this construct by site-directed hydroxyl radical probing. Unexpectedly, in the absence of the AUG codon, ternary complex recruitment still promotes HCV mRNA binding in the 40S subunit entry 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 HCV IRES mRNA in the entry channel in a process that is independent of AUG codon recognition by the initiator tRNA. eIF3 binding triggers structural changes in the 40S subunit How do eIF3 and the ternary complex stabilize mRNA in the 40S subunit decoding channel irrespective of AUG codon recognition? An important clue to a possible mechanism came from observations of distinct conformational states of the 40S subunit by using cryo-EM to compare empty 40S subunits to 40S subunits complexed with viral IRES mRNA or initiation factors eIF1 and eIF1A13,14,17. These structures showed similar motions of the head relative to the body of the 40S subunit, altering the architecture of the mRNA binding channel, which could thereby affect access to the mRNA binding channel. To test whether these characterized structural changes are detectable using our site-directed hydroxyl radical probing assay, we formed each complex and determined the cleavage pattern from hydroxyl radicals generated from eIF3j. To visualize its association with the 40S subunit, the HCV mRNA was truncated at the AUG codon to avoid eIF3j dissociation, as discussed above. We used 18S rRNA cleavages resulting from BABE-Fe–generated hydroxyl radicals from different sites on eIF3j to estimate changes in the three-dimensional structure of the 40S subunit, as previously validated for the bacterial ribosome26. Distance changes between the location of the BABE-Fe moiety and a specific nucleotide in the rRNA result in altered

400

VOLUME 16

HCV∆II

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.

cleavage intensities, with more intense cleavages representing reduced distances between BABE-Fe and the affected sites. We compared the rRNA cleavage patterns generated in parallel experiments using two different BABE-Fe–modified eIF3j variants (Fig. 1c). Cleavage sites to be monitored were selected in the region of the 40S subunit beak (Fig. 1a) because the conformation of this structure changes substantially upon IRES or initiation factor binding13,14,17. Consistent with the cryo-EM structures, 18S rRNA cleavage patterns generated from BABE-Fe positions in eIF3j changed upon the association of saturating amounts of eIF1 and eIF1A or the HCV mRNA with the 40S subunit (Fig. 4a,b). Specifically, for both complexes, we observed reduced cleavage intensities at nucleotides 1268– 1270, 1295–1298 and 1305–1307 in helices 32 and 33 generated from position 152 in eIF3j (Fig. 4a,b, compare lanes 9 and 10). However, in contrast to the recruitment of HCV mRNA, eIF1 and eIF1A association with the 40S subunit resulted in increased cleavage intensities of nucleotides 1279–1284 generated from BABE-Fe tethered to amino acid positions 152 and 217 on eIF3j (Fig. 4a,b, compare lanes 9 and 10). Therefore, although eIF1 and eIF1A, or the HCV mRNA, induce similar structural changes in the 40S subunit, these data obtained in solution imply some differences in the way these components alter the 40S subunit structure, particularly surrounding the mRNA entry channel and A-site. It should be noted that the observed cleavage differences could also result partly or entirely from direct interactions between eIF3j with eIF1 and eIF1A. We have previously shown that the interaction between eIF3j and eIF1A is anticooperative19, which probably indicates altered conformations of eIF3j and eIF1A on the 40S subunit. We next tested whether the eIF3 complex alters the conformation of the 40S subunit upon binding. To this end, we compared 18S rRNA cleavages generated from BABE-Fe tethered to positions in eIF3j in the absence or presence of the eIF3 complex. Association of the eIF3 complex seems to alter the cleavage pattern in a similar, but distinct,

NUMBER 4

APRIL 2009


also observed differences in cleavage-intensity changes between complexes containing eIF1 and eIF1A, or eIF3j in helix 18 (Supplementary Fig. 2b), confirming the similar, but unique, conformations that these initiation factors promote in the 40S subunit.

1 2 3

4 5 6

7 8

9 10

1 2 3

4 5 6

7 8

1268–1270

1268–1270

1279–1284

1279–1284

1295–1298

1295–1298

1305–1307

1305–1307

1318–1321

1318–1321

1 2 3 4 5 6

S217C

S217C+eIF3∆j

S152C

7 8

S152C+eIF3∆j

C T A G

9 10 11 12 Lane

40S

S217C+HCV∆ORF

S217C

S152C+HCV∆ORF

S152C

c

EDTA/Fe –cys+EDTA/Fe –cys+eIF3∆j+EDTA/Fe

Ternary complex binding induces a 40S structural change A reasonable explanation for ternary complex–induced stability of mRNA, irrespective of AUG recognition, may be that there are additional conformational changes in the 40S subunit induced by the ternary complex itself. To test this, we again compared rRNA cleavage patterns generated from BABE-Fe–modified eIF3j in the location of the 40S subunit beak upon recruitment of the ternary complex to the 40S subunit in the absence of other initiation factors. Changes in the cleavage intensities from hydroxyl radicals generated from BABE-Fe tethered to position 152 in eIF3j were apparent (Fig. 5, compare lanes 9 and 10). Cleavage intensities increased markedly at nucleotide positions 1268–1270 (helix 32) and 1279–1284 (helix 33), and increased slightly at nucleotides 1295–1298 (helix 33). These data suggest a conformational change upon ternary complex recruitment in

–cys+HCV+EDTA/Fe

C T A G

11 12 Lane

40S EDTA/Fe –cys+EDTA/Fe

S217C+eIF1/1A

S217C

S152C+eIF1/1A

C T A G

b

S152C

a

–cys+eIF1/1A+EDTA/Fe

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 and 217 were enhanced by the addition of eIF3 or eIF1 and eIF1A to the 40S subunit (Fig. 4a,c, compare lanes 9 and 10). However, in contrast to the cleavage patterns observed upon addition of the HCV IRES or eIF1 and eIF1A, the cleavage intensities of nucleotides 1295–1298 and 1305–1307 by hydroxyl radicals generated from position 152 in eIF3j remained unchanged upon addition of eIF3 (Fig. 4a,c, compare lanes 11 and 12). Notably, the eIF3 complex, eIF1 and eIF1A, or the HCV IRES mRNA all reduced the cleavage intensities of nucleotides 1268– 1270 in helix 32 that are generated from eIF3j modified with BABE-Fe at position 152 (Fig. 4a–c, compare lanes 9 and 10). Thus, the eIF3 complex stabilizes a specific 40S subunit conformation that has some similarities to the structures previously reported for HCV mRNA or eIF1 and eIF1A bound to the 40S subunit. However, these conformational changes are not sufficient to favor mRNA binding to the decoding channel, as detected by eIF3j displacement or 40S toeprinting20, in the absence of additional initiation factors. Furthermore, we

40S EDTA/Fe –cys+EDTA/Fe


ARTICLES

9 10 11 12 Lane 1268–1270 1279–1284

1295–1298

40S

1305–1307

1318–1321

40S

40S eIF1A

eIF3j

eIF3j

+

eIF3j eIF1

+

+ eIF3∆j

HCV∆ORF

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.


VOLUME 16

NUMBER 4

APRIL 2009

401

C T A G

40S EDTA/Fe –cys+EDTA/Fe –cys+TC+EDTA/Fe S152C S152C+TC S217C S217C+TC T235C T235C+TC A241C A241C+TC

ARTICLES

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

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.

Lane 1268–1270 1279–1284

1295–1298


1305–1307

1318–1321

40S

elF3j

+ TC

which nucleotides in the 40S subunit beak move closer to the location of amino acid 152 in eIF3j (or into a more favorable environment or position for attack). Notably, the cleavage intensities generated from eIF3j position 152 at nucleotides 1305–1307, or cleavage of nucleotides 1279–1284 generated 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 inside the mRNA entry channel indicated that this region of the 40S subunit also does not change in relation to the location of eIF3j upon ternary complex recruitment (Supplementary Fig. 2b). These data reveal that the ternary complex promotes a structural rearrangement of the beak of the 40S subunit that would be likely to alter the conformation of the mRNA binding channel, especially in the A-site of the 40S subunit. This change may contribute to the association of the mRNA with the 40S subunit in a manner that is independent of AUG codon recognition by the initiator tRNA. Notably, although a thermodynamically stable complex between the ternary complex and 40S subunit persists in the absence of any other initiation factors, it is expected that other initiation factors are required for accelerating the rate of complex formation. DISCUSSION In this study, we investigated the streamlined initiation mechanism used by the HCV IRES mRNA to examine the mechanism of mRNA positioning in the 40S mRNA binding channel. On the basis of site-directed hydroxyl radical probing and toeprinting, we established that eIF3j and HCV IRES mRNA binding to the 40S subunit mRNA entry channel is mutually exclusive. This enabled us to use the dissociation of eIF3j from the mRNA entry channel as an indicator

402

VOLUME 16

of mRNA binding to the 40S subunit entry channel. A proposed model for the pathway of HCV mRNA association with the 40S subunit is presented in Figure 6. In agreement with previous data7,13, the conformational change in the 40S subunit induced by domain II of the HCV IRES is required for mRNA positioning in the binding channel. This conformational change alone is not sufficient to allow the HCV mRNA to associate with the entry channel when both eIF3j and eIF3 are present. Instead, eIF3j displacement and consequent association of the HCV mRNA with the entry channel require ternary complex recruitment to the 40S subunit. Of note, the toeprinting competition experiment shown in Figure 2a requires a large molar excess of eIF3j to compete with HCV mRNA binding to the 40S subunit. This is probably required because the HCV mRNA is tethered to the 40S subunit through domains II and III, increasing its local concentration, whereas the absence of the interaction with the eIF3 complex reduces the local concentration of eIF3j. Although it is possible that the large excess of eIF3j used for toeprinting may bind nonspecifically to the 40S subunit, this seems unlikely because the physiological local concentration of eIF3j on the 40S subunit would probably be high owing to its interaction with the eIF3 complex. Unexpectedly, our data show that the stabilizing effect of the ternary complex on HCV mRNA association does not require base-pairing between the AUG codon and the anticodon of the initiator tRNA. This finding contradicts toeprinting data that indicated a strong requirement of the AUG codon for HCV mRNA association with the entry channel7. However, toeprinting enables detection of only the most thermodynamically stable complexes, such as those enhanced by the codon-anticodon interaction. In addition, mutation of the authentic AUG codon reduces protein synthesis by only 50–60% in vitro27. The remaining activity may be due to the use of the ACG codon situated two codons downstream of the AUG codon. Because toeprinting is not possible with the mutant AUG construct7, we are not able to determine whether our construct actually allows a codon-anticodon interaction to occur at the ACG codon. Therefore, although our data

A GU 3′

5′

A GU

3′ Domain II– induced conformation (mRNA entry channel opening)

5′ 3′

GUA

5′

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 complex places 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.

NUMBER 4

APRIL 2009



ARTICLES suggest that AUG recognition by the initiator tRNA is not required for eIF3j dissociation in our assay, we cannot rule out the weaker interaction between the initiator tRNA and the ACG codon, which may be essential for mRNA stabilization and eIF3j dissociation. Notably, recent data indicate that eIF3j can dissociate mRNA from the 40S subunit during ribosome recycling but only in the presence of eIF1 (ref. 20). As eIF1 does not affect the affinity of eIF3j binding to the 40S subunit19, this is probably due to the effect of eIF1 in dissociating the P-site–bound tRNA, which would subsequently shift the equilibrium toward eIF3j binding to the entry channel over that of mRNA. However, a detailed kinetics analysis is required before the role of eIF3j in mRNA dissociation can be made, as it is plausible that eIF3j influences only mRNA association. In this study, we also investigated the mRNA binding propensities of different 40S subunit conformations induced upon association of the HCV IRES or initiation factors. In agreement with cryo-EM studies13,14, our data provide evidence that the HCV IRES or eIF1 and eIF1A promote similar but distinct conformational changes in the 40S subunit. Notably, we show that the eIF3 complex also induces a conformational change in the 40S subunit that has some similarities to those induced by the HCV mRNA or eIF1 and eIF1A. Such eIF3mediated effects on the 40S subunit may provide an explanation for its involement in mRNA recruitment to the 40S subunit in addition to its role in recruiting the cap binding complex in vitro and in vivo8,9,28. A recent report also suggested a possible conformational change in the mRNA entry channel when eIF3 associates with the 40S subunit29. However, those experiments were completed in the absence of eIF3j but instead contained poly(U) to provide affinity for the eIF3 complex to the 40S subunit. Therefore, it is not possible to determine from those experiments what relative effects eIF3 and poly(U) have on the structure of the 40S subunit. Ternary complex association with the 40S subunit without other components induces a conformational change in the head of the 40S subunit, which may have a role in promoting eIF3j dissociation and/or mRNA stability with the entry channel. Notably, this conformational change is not sufficient to promote HCV mRNA association with the entry channel because the ternary complex cannot stabilize HCV mRNA in the entry channel in the absence of domain II. Therefore, because eIF1, eIF1A, eIF3 and the ternary complex enable AUG recognition on an unstructured mRNA30, it is tempting to speculate that eIF1 and eIF1A probably provide the necessary conformational change in the 40S subunit to allow mRNA positioning in the binding channel, as proposed14. On the basis of these findings, we propose that either the HCV IRES, or eIF1 and eIF1A, can induce a similar conformational state of the 40S subunit that favors mRNA loading into the decoding channel with consequential displacement of eIF3j upon ternary complex binding. This conclusion explains how the HCV IRES functionally replaces eIF1 and eIF1A during translation initiation on the human ribosome.

METHODS Sample 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 bromoacetamidobenzylEDTA/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


VOLUME 16

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 mutagenesis 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 electrophoresis 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 & Molecular Biology website. ACKNOWLEDGMENTS We thank D. King at the University of California, Berkeley, for expert MS analysis of modified proteins. We gratefully acknowledge members of the Doudna laboratory for discussions and comments on the manuscript. In particular, we would like to thank R. Spanggord for advice on hydroxyl radical probing and F. Siu for advice on RNA transcription protocols. This work was supported in part by a grant from the US National Institutes of Health to J.A.D. and J.W.B.H. AUTHOR CONTRIBUTIONS C.S.F. performed the experiments; C.S.F., J.W.B.H. and J.A.D. designed experiments 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 in eukaryotes. in Translational Control in Biology and Medicine (eds. Mathews, M.B., Sonenberg, N. & Hershey, J.W.B.) 87–128 (Cold Spring Harbor Laboratory Press, Cold Spring 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. in Translational 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).

NUMBER 4

APRIL 2009

403


ARTICLES 4. Elroy-Stein, O. & Merrick, W.C. translation initiation via cellular internal ribosome entry sites. in Translational Control in Biology and Medicine (eds. Mathews, M.B., Sonenberg, N. & Hershey, J.W.B.) 155–172 (Cold Spring Harbor Laboratory Press, Cold Spring Harbor, NY, 2007). 5. Pisarev, A.V., Shirokikh, N.E. & Hellen, C.U. Translation initiation by factor-independent 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 viral translation 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 prokaryoticlike mode of cytoplasmic eukaryotic ribosome binding to the initiation codon during internal translation initiation of hepatitis C and classical swine fever virus RNAs. Genes Dev. 12, 67–83 (1998). 8. Trachsel, H., Erni, B., Schreier, M.H. & Staehelin, T. Initiation of mammalian protein synthesis. 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 initiation factors 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 human translation initiation complexes by the hepatitis C virus internal ribosome entry site RNA. 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 ribosome and elongation fractors. in Translational Control in Biology and Medicine (eds. Mathews, M.B., Sonenberg, N. & Hershey, J.W.B.) pp. 59–85 (Cold Spring Harbor Laboratory Press, Cold Spring Harbor, NY, 2007). 13. Spahn, C.M. et al. Hepatitis C virus IRES RNA-induced changes in the conformation of the 40S ribosomal subunit. Science 291, 1959–1962 (2001). 14. Passmore, L.A. et al. The eukaryotic translation initiation factors eIF1 and eIF1A induce 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 the interaction of 40S ribosomal subunits with the internal ribosomal entry site of hepatitis C 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 to human 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 factors from 48S complexes during ribosomal subunit joining and the link between establishment 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 decoding center of the human 40S ribosomal subunit. Mol. Cell 26, 811–819 (2007).

404

VOLUME 16

20. Pisarev, A.V., Hellen, C.U. & Pestova, T.V. Recycling of eukaryotic posttermination ribosomal complexes. Cell 131, 286–299 (2007). 21. Hartz, D., McPheeters, D.S., Traut, R. & Gold, L. Extension inhibition analysis of translation 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 as measured 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. Nucleic Acids Res. 26, 4853–4859 (1998). 24. Kieft, J.S., Zhou, K., Jubin, R. & Doudna, J.A. Mechanism of ribosome recruitment by hepatitis C IRES RNA. RNA 7, 194–206 (2001). 25. ElAntak, L., Tzakos, A.G., Locker, N. & Lukavsky, P.J. Structure of eIF3b RNA recognition motif and its interaction with eIF3j: structural insights into the recruitment of 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 to RNA. Methods Enzymol. 318, 175–190 (2000). 27. Reynolds, J.E. et al. Unique features of internal initiation of hepatitis C virus RNA translation. EMBO J. 14, 6010–6020 (1995). 28. Jivotovskaya, A.V., Valasek, L., Hinnebusch, A.G. & Nielsen, K.H. Eukaryotic translation initiation factor 3 (eIF3) and eIF2 can promote mRNA binding to 40S subunits independently 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. Ribosomal position 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 initiation factors 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 required for 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 mammalian 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 the peptide-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 iondependent tertiary fold. J. Mol. Biol. 292, 513–529 (1999). 35. Lomakin, I.B., Kolupaeva, V.G., Marintchev, A., Wagner, G. & Pestova, T.V. Position of eukaryotic initiation factor eIF1 on the 40S ribosomal subunit determined by directed hydroxyl 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 of eukaryotic translation initiation complexes. Methods Enzymol. 430, 147–177 (2007).

NUMBER 4

APRIL 2009


ARTICLES

Tertiary interactions within the ribosomal exit tunnel


Andrey Kosolapov & Carol Deutsch Although tertiary folding of whole protein domains is prohibited by the cramped dimensions of the ribosomal tunnel, dynamic tertiary interactions may permit folding of small elementary units within the tunnel. To probe this possibility, we used a b-hairpin and an a-helical hairpin from the cytosolic N terminus of a voltage-gated potassium channel and determined a probability of folding for each at defined locations inside and outside the tunnel. Minimalist tertiary structures can form near the exit port of the tunnel, a region that provides an entropic window for initial exploration of local peptide conformations. Tertiary subdomains of the nascent peptide fold sequentially, but not independently, during translation. These studies offer an approach for diagnosing the molecular basis for folding defects that lead to protein malfunction and provide insight into the role of the ribosome during early potassium channel biogenesis.

Protein folding begins with the birth of the peptide within the ribosomal tunnel1–8, but the dimensions of this tunnel, 100 A˚ in length and 10–20 A˚ in width9–12, limit the dimensions of a folded peptide that can fit inside the tunnel. For example, the complete folding of a tertiary monomeric functional T1 domain of the voltage-gated potassium channel Kv1.3, which is approximately 27 27 25 A˚13,14 (Fig. 1a), is precluded1. However, we may ask whether 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 the tunnel has not been investigated explicitly. The elementary tertiary folding unit has not been established. The T1 domain is in the cytoplasmic N terminus of voltage-gated potassium (Kv) channels. It is a recognition domain with a crucial role in the folding and oligomerization of Kv channel proteins, enabling formation of the tetrameric channel15–17, its axonal targeting18 and possibly the opening and closing of the permeation pathway for potassium ions13,19–21. We have previously shown that T1 domains tetramerize while still attached to ribosomes22 and that the whole T1 domain folds into its tertiary structure only after the primary sequence emerges from the ribosomal tunnel and the linker between T1 and the first 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 fold inside the ribosomal tunnel? In this paper, we assess the extent and relative ease of T1 subdomain folding and the compartment in which this folding occurs. We also address a key mechanistic issue in protein folding in general, namely, whether subdomains of nascent peptides fold sequentially in the tunnel or in a concerted fashion upon emergence, and whether the folding of one nascent subdomain depends on the synthesis and/or folding of another subdomain.

Here we establish that tertiary intrapeptide interactions occur within the tunnel and that stable hairpins form upon emergence from the tunnel. In addition, folding of a C-terminal hairpin depends on the presence and status of the N-terminal b-hairpin. Moreover, the nascent peptide–tunnel complex may be more dynamic than previously thought, and a distal portion of the tunnel (near the exit port) provides an entropic window for exploration of conformational space by the nascent peptide. RESULTS Subdomain folding To study elementary tertiary folding of T1 subdomains at different stages of biogenesis, we chose two subdomains: a pair of antiparallel b-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 of cysteines (Cys53 and Cys66, denoted 53C66C) into the b-hairpin and a different pair (Cys125 and Cys149, denoted 125C149C) into the ahelical hairpin. In each case, the cysteines are exposed in the folded monomer and are thus available to cross-linking reagents under the conditions of our assay1,23. Moreover, these cysteines are predicted to be within 4–6 A˚ of each other if the subdomain is folded similarly to that 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 intermaleimide distance of B6 A˚. If the engineered cysteines in each of the T1 subdomains shown in Figure 1b come within 4–7 A˚ of each other, then they can be cross-linked with PDM. To distinguish cross-linked from non–cross-linked products, we use methoxypolyethylene glycol maleimide (PEG-MAL) and methoxypolyethylene glycol

Department of Physiology, University of Pennsylvania, Philadelphia, Pennsylvania 19104-6085, USA. Correspondence should be addressed to C.D. ([email protected]). Received 26 June 2008; accepted 30 January 2009; published online 8 March 2009; doi:10.1038/nsmb.1571


VOLUME 16

NUMBER 4

APRIL 2009

405

ARTICLES

a

to the ribosome) at various distances from the peptidyl transferase center (PTC; Fig. 1b) using different enzyme restriction sites (Supplementary Table 1 online) that eliminate the stop codon. This produces different chain lengths between the PTC and the cysteine at the C terminus of each subdomain. The cross-linking assay was performed on each of these double-cysteine constructs. In addition, similar assays were carried out on constructs containing only a single cysteine of each cysteine pair to calculate the probability of crosslinking, Pxlink1 (see Methods). Each mRNA for the engineered nascent peptide was translated in a cell-free, membrane-free rabbit reticulocyte system with added 35S-methionine and then subjected to the crosslinking assay1 (see Methods). After labeling with PDM, samples were first treated with SDS and then pegylated to assay for either free available 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) from constructs with different numbers of amino acids between the PTC and the C-terminal cysteine, DPTC (calculated as the number of the last residue included by the restriction site minus the residue number of the engineered cysteine; for example, 91 – 66 ¼ 25 for the first construct shown in Fig. 2a). Regardless of the DPTC and number of cysteines in the construct, both N- and C-terminal cysteines were available for pegylation (lanes 1) and were completely labeled by PDM (lanes 2). Moreover, all PDM-modified single-cysteine constructs show a similar efficiency of labeling with PEG-SH, regardless of DPTC (lane 3 for all single-cysteine constructs). For a DPTC of 52 (outside tunnel), both the b-hairpin and the a-hairpin doublecysteine nascent peptides that were modified with PDM produced negligible gel shifts with PEG-SH (lanes 3, second rows, columns 1 and 4, respectively). This indicates that no free peptidylmaleimides were present—that is, the cysteines are cross-linked and subdomains are mostly folded. In contrast, double-cysteine peptides with DPTC of 19 residues or 25 residues (deeper in the tunnel) were not cross-linked

b lle125

Glu149

β1

α4

α5

β2

53

66

α4 β2 Gln66

β1

∆ PTC

NH2 α5

∆ PTC

NH2 125

149


Arg53

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 an a-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.

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 labeled with PEG-MAL24, and a free peptidyl-maleimide can be labeled with PEG-SH23. A cross-linked subdomain will fail to undergo a gel shift with either pegylation reagent, PEG-MAL or PEG-SH. A probability of cross-linking, Pxlink, can be calculated, as described1. We consider that the extent of cross-linking reflects tertiary interactions and, depending on the location of the engineered cysteine pairs, the probability of tertiary folding. To determine the cellular compartment (for example, ribosomal tunnel versus cytosol) in which subdomain folding occurs, we generated double-cysteine constructs as nascent peptides (still attached

a PEG-MAL PEG-SH PDM

53C66C + – –

53C

+ – – + + +

∆PTC 25

+ – –

+ – +

66C – + +

2

+ – +

125C149C PEG-MAL PEG-SH PDM

– + +

+ – –

+ – +

+ – – 2

1

1

3

1

1

0 1

2

3

∆PTC 19

2

+ – +

1

3

2

3

b

149C – + +

+ – –

+ – +

– + +

1

1

0

0

0 1

1

2

3

2

α-hairpin

66C

149C

66

1

45

0

0 1

β-hairpin

1

1

1

0

0 2 3

1

45

0

∆PTC 322

∆PTC 239

3

1

1

∆PTC 52

1

31 22

1

66

2

2 ∆PTC 52

125C

– + +

2

0 1

+ – –

1

2

3

2

0

0

0 1

1

3

2

3

1

2

3

0 1

2

3

∆PTC 25

1

31

1

0

22

0 ∆PTC 19

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.

406

VOLUME 16

NUMBER 4

APRIL 2009


ARTICLES 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 crosslinking 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 be cross-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.

a 1.0

P xlink

0.8 0.6

α

β

0.4 0.2 0.0 0.0

b P xlink

0.4

0.6 F acc

0.8

1.0

0.4

0.6 F acc

0.8

1.0

62C75C 101C125C

1.0 0.8

0.2

∆PTC > 285

0.6

62C75C 101C125C

0.4

59C72C 86C99C


0.2 0.0 0.0

0.2

and therefore not folded. In this case, the PDM-treated peptides contained free peptidylmaleimides that were pegylated with PEG-SH to produce gel shifts (bands 1 and 2 in lanes 3, first row, columns 1 and 4). Single-cysteine constructs were not cross-linked, yet were modified by PDM to contain a free maleimide that was pegylated by PEG-SH (band 1 in lanes 3, first row, columns 2, 3, 5 and 6). Crosslinking data for additional constructs with DPTC values of 27 and 39 for the b-hairpin and 24 and 29 for the a-hairpin are shown in Supplementary Figure 1 online. Location of the subdomain in the tunnel The interpretation of these cross-linking data requires that we know the location of the subdomain, inside or outside of the tunnel, which depends on both DPTC and the secondary structure of the peptide between the C-terminal cysteine of the subdomain and the PTC. To determine the subdomain location, we used an accessibility assay (Fig. 2b and Supplementary Fig. 2 online). This assay also relies on a mass-tagging strategy using PEG-MAL2,24, but without the SDS pretreatment used in the cross-linking assay (see Methods)1. The final extent of PEG-MAL labeling of a cysteine in the last 20 A˚ of the tunnel is monotonically dependent on the distance of the modifiable cysteine from the PTC2. Accessibility is therefore an accurate estimator of distance from the PTC2 (see Discussion). For each subdomain, we engineered a single cysteine, corresponding to the C-terminal cysteine of each cysteine pair, into the nascent peptide, which was treated with PEG-MAL, and determined the final extent of labeling. For b- and a-hairpins located outside the ribosomal tunnel—that is, DPTC is 322 and 239, respectively (top gels)—the fraction of pegylated protein is Z0.88 (Fig. 2b). In contrast, the fraction of pegylated protein is r0.14 for constructs where the C-terminal cysteines of the b- and a-subdomains were placed at 25 residues and 19 residues, respectively, from the PTC (bottom gels). These cysteines, located inside the tunnel, are 80 A˚ or less from the PTC2. Relationship between subdomain location and folding To determine the tunnel location where hairpin folding occurs, we combined the results from the cross-linking and accessibility assays for


VOLUME 16

each hairpin. In Figure 3a, we plot the probability of cross-linking (Pxlink) against the fraction of accessible peptide (Facc). The midpoint of each PF curve, F50, is the fraction accessible when Pxlink is 50% of the maximum (dotted lines). Thus, we can compare Pxlink at different distances from the PTC and the relative ease of folding for each subdomain at the same location. Both subdomains apparently begin to fold before they emerge completely from the ribosomal tunnel, and maximum folding occurs after the subdomains emerge from the tunnel. Pxlink saturates below 100% completion owing to rapid binding of PDM molecules to each cysteine23. The F50 values are 0.57 ± 0.10 for the b-hairpin and 0.37 ± 0.06 for the a-hairpin, values that are not significantly different (P ¼ 0.075, Z-test), suggesting that the relative ease of folding of the two different hairpins is similar. Is subdomain folding affected by the interaction of the ribosomal tunnel with the subdomain, or is it due solely to the intrinsic properties of the peptide? To address this question, we first released the nascent peptides from the ribosome using RNase, isolated the released peptide and assayed it for cross-linking. A nascent peptide in which the C terminus of the b-hairpin sequence was located 25 residues from the PTC was released and gave nearly maximal crosslinking for the hairpin (Supplementary Fig. 3 online), suggesting that it folded completely. In contrast, a released peptide in which the C terminus of the a-hairpin sequence was 19 residues from the PTC cross-linked only 55% of the maximum value attained while attached to the ribosome. Notably, when this sequence was first positioned more distally in the tunnel, for example, 24 residues and 29 residues from the PTC, and then released, the a-hairpin apparently folded more completely, at 71% and 76% of the maximum, respectively. This behavior suggests that several factors may determine the probability of folding of the a-hairpin: the location of the a-hairpin sequence in the tunnel before release (so that the tunnel promotes a folding-competent state of the peptide), the total length of the nascent peptide, and/or the number and nature of the residues between the hairpin C terminus and the PTC. In addition, the maximally folded a-hairpin present in the crystal structure may represent only one of several hairpin structures. This a-hairpin behavior contrasts with the

NUMBER 4

APRIL 2009

407

ARTICLES

a

β1

β2

α4

α5

α4

α5

NH2

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.

PTC Full length

Truncated N122M β1

β2

α4

α5

NH2

Mutant T65D

b

1.0

Pxlink

0.8 0.6 0.4 0.2

12 12 5C 5C 14 14 9C 9C 65 D

53 53 C C6 66 6 C C 65 D

1 12 25 5C C1 14 49 9C C ∆N


0.0

folding requirements for the b-hairpin, which is constrained by the tunnel but apparently folds completely once it is released. Two additional conclusions may be drawn from the PF curves shown in Figure 3a. First, folding of the upstream b-hairpin does not require the presence and/or folding of the a-hairpin, as evidenced by maximum folding for the DPTC 52 peptide, which lacks the a-hairpin. Second, tertiary folding of both hairpins is prohibited in short nascent intermediates with relatively inaccessible C-terminal cysteines (DPTC o 21). However, the a-hairpin shows a significant amount of folding (P ¼ 0.04, Pxlink ¼ 0.26, 34% of its maximum) when DPTC ¼ 24, that is, when the C-terminal cysteine is inside the tunnel (Facc ¼ 0.25), suggesting a dynamic equilibrium between folded and unfolded species (Supplementary Results online). Evidence for an entropic window The PF curves shown in Figure 3a make two predictions. First, at locations far outside the tunnel (DPTC 4 239), the hairpin subdomains, as well as the whole T1 domain1, are folded. We can therefore predict from the crystal structure of the T1 domain which side chains in the folded T1 are too distant from one another to be intramolecularly cross-linked (414 A˚1) and should therefore show a low Pxlink. Second, at tunnel locations within 80 A˚ of the PTC, no folding occurs. Thus, pairs of distant cysteines engineered anywhere along the nascent peptide, including nonhairpin sequences, should also show a low Pxlink within these 80 A˚. Each prediction derives from a different constraint on tertiary folding. In the first case, the whole stably folded T1 domain has a network of intramolecular interactions that restricts its mobility, decreasing the degrees of freedom allowed for any given stretch of primary sequence within the folded T1. In the second case, the tunnel itself is too small to allow intramolecular tertiary interactions in the nascent peptide. To test these predictions, we designed two separate double-cysteine constructs, 62C75C and 101C125C, in the region adjacent to the b-hairpin and a-hairpin, respectively. The cysteines in the 62C75C and 101C125C pairs are separated by the same number of amino acids, respectively, as in the 53C66C (b-hairpin) and 125C149C (a-hairpin) pairs. In contrast to intrapair proximities (6–7 A˚) for 53C66C and 125C149C, both 62C75C and 101C125C are farther apart in the folded T1 monomer, B25 A˚ and B17 A˚, respectively.

408

VOLUME 16

As predicted, cross-linking efficiencies for 62C75C (Fig. 3b, red squares) and 101C125C (blue triangles) are low when these constructs were placed either deep inside or completely outside of the tunnel (Facc r 0.5 or Z 0.9, respectively), consistent with the lack of crosslinking reported for two cysteines located 14 A˚ apart in the whole folded T1 domain1. However, between these two regions of restricted conformational entropy there is a region where some cross-linking is permitted (Fig. 3b, shaded area), even though in each case the introduced cysteines within the pair are distant from one another in the mature T1 structure. We used chain lengths that position 62C75C and 101C125C in the more distal regions of the tunnel, at DPTC values of 30, 35, 37, 43, 67, 313 and 27, 38, 41, 43, 48, 263, respectively. Short constructs (Fig. 3b, open symbols) with Facc of 0.6–0.95 cross-linked to give a Pxlink of Z 0.6, suggesting that this region of the tunnel is dynamic and allows the peptide to explore conformational space, therefore constituting an entropic window (shaded region) for sampling potential folding partners. For these constructs, both accessibility measurements and calculations based on secondary structure between the C-terminal cysteine and the PTC indicate that these C-terminal cysteines are located either just within the tunnel at the exit port or outside in the immediate vicinity of the tunnel. To confirm findings in the restricted region, we engineered two constructs with DPTC 4 285. We engineered 86C99C between the b- and a-hairpin sequences. These cysteines are 14 A˚ apart in the mature folded structure. We also engineered 59C72C in the b-hairpin sequence. These cysteines are 25 A˚ apart in the mature folded structure. Pxlink for both 86C99C (Fig. 3b, filled green circle) and 59C72C (filled inverted black triangle) is B0.3, in agreement with 62C75C (filled red square) at this same location. All filled symbols represent cysteine pairs far outside the tunnel on a long tether, which have Pxlink of 0.2–0.3. To confirm findings in the interaction-permissive region, we engineered a construct containing 59C72C that poises this cysteine pair in the vicinity of the exit port but outside the tunnel. The Pxlink for this construct was 40.6 (data not shown). These findings permit a more precise interpretation of the cross-linking results for hairpin residues 53C66C and 125C149C, which are in close proximity in the T1 domain (Fig. 3a). Although these residues in each hairpin initially interact as part of the exploration of conformational space allotted by the tunnel’s entropic window, the hairpins fold into stable structures only in the context of the longer, folded T1 domain. Cooperative subdomain folding The 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.

NUMBER 4

APRIL 2009



ARTICLES This hypothesis predicts that folding of the a-hairpin sequence should be sensitive to deletion of its preceding T1 residues. To explore this possibility, we truncated the T1 domain at position 122 (Fig. 4), which yields an N-terminally deleted nascent peptide (DN) containing the a-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-hairpin folding depends on the presence of the N-terminal T1 domain. In addition to these long-tethered DN constructs with a folding domain outside of the tunnel, we investigated shorter DN constructs, DN-DPTC24 and DN-DPTC29, in which the folding domains reside near the end of the tunnel. The truncated DN-DPTC24 folded less efficiently than the nontruncated peptide (Pxlink is 0.06 ± 0.12, n ¼ 2 versus 0.26 ± 0.03, n ¼ 3, respectively, (Supplementary Fig. 4 online)). Truncated DN-DPTC29, which is also near the end of the tunnel and provides a larger dynamic range, has a Pxlink of 0.30 ± 0.02 (n ¼ 2) versus 0.46 ± 0.03 (n ¼ 5) for the nontruncated construct. For both DN-DPTC24 and DN-DPTC29 constructs lacking the N-terminal b-hairpin, folding of the a-hairpin was less efficient, again consistent with folding of a C-terminal hairpin depending on the presence and status of the N-terminal b-hairpin sequence. In contrast, the b-hairpin apparently still folded in the absence of the C-terminal region of T1, including the a-hairpin sequence. A 118-residue nascent peptide that includes the b-hairpin sequence (DPTC ¼ 86), but not the 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-hairpin formation 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 T1 tetramer. Kv1.3 with a T65D mutation does not tetramerize and only achieves one-half as much tertiary folding as wild-type T1 (refs. 23,25). We therefore used the T65D mutation to investigate two issues. Does the T65D mutation impair tertiary folding of the b-hairpin? Does the T65D mutation affect a-hairpin formation? We introduced the T65D mutation into both a 53C66C background (DPTC 322) and a 125C149C background (DPTC 239) that places the b- and a-hairpins far outside the tunnel. Pxlink in the T65D mutant is 0.46 for the b-hairpin and 0.26 for the a-hairpin, each significantly lower 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 with accumulated disruption for the more C-terminal location of the a-hairpin. We suggest that propagated disruption from the site of mutation all along the folded T1 interface is responsible for defective quaternary structure formation of the T65D mutant T1 domain. Folding of the a-hairpin depends on the presence and status of the b-hairpin. DISCUSSION Subdomain folding Although it is too cramped inside the tunnel to accommodate the entire T1 domain of Kv channels in its fully folded state1, consistent with previous investigations of other proteins6,26,27 and the geometric analysis of Moore and co-workers28, tertiary interactions do occur in the tunnel, particularly in its last 20 A˚. This region supports helix formation1–4,29, specifically, folding of the a5 helix of the T1 domain studied herein3. Why is this region permissive for tertiary interactions? First, the dimensions of the tunnel in this region may be wider30 than the 20 A˚ estimated from the crystal structures of the ribosome, which lack both nascent peptides and attendant chaperones. Second, the ribosomal tunnel may be more dynamic than previously thought (whereas Steitz’s estimate of the diameter of the largest sphere that can


VOLUME 16

fit inside a peptideless tunnel, 13.7 A˚31, can accommodate the bhairpin, an increase of less than two-fold is required to accommodate the a-hairpin). Moreover, the entrance of the ribosomal tunnel undergoes conformational changes during peptide elongation32 and the tunnel is believed to have a gate33–35. Third, tertiary- and secondary-structure formation may be coupled36 and thus potentiated in this distal portion of the tunnel that favors secondary folding3,29. An a-helix requires less room than a hairpin of similar length and amino acid composition, and thus an a-helix may form along the first 80 A˚ of the tunnel, independent of tertiary folding, but the converse may not hold. For the a4-a5 hairpin, tertiary structures in the last 20 A˚ of the tunnel may be favored by coupled secondary-structure formation. Moreover, chaperone proteins hovering at the exit port of the tunnel may facilitate tertiary and/or coupled folding. Although these two subdomains differ in size and length of the hairpin, and the number of hydrophobic interactions along their respective hairpin interfaces, our cross-linking assay suggests that the ease of folding of the b- and the a-hairpin are not significantly different. The b-hairpin evidently folds despite the absence of the a-hairpin. Similarly, the a-hairpin apparently folds in the absence of the 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 the full-length peptide, which contains both b- and a-subdomains. Moreover, mutation of b-hairpin residue 65 (T65D) causes a decrease in folding of both the b- and a-hairpins. In the full-length mature Kv1.3, T65D prevents both quaternary and complete tertiary folding of the T1 domain and, consequently, function of the Kv1.3 channel23,25. This is expected because tertiary folding and tetramerization of T1 are coupled25. We now understand that the basis for this defect in tertiary folding involves the propagated disruption of subdomain formation. Our findings underscore the usefulness of this approach in diagnosing the molecular basis for folding defects. Information about the extent of folding at discrete locations along the tunnel is embedded in the cross-linking-accessibility (PF) curves in Figure 3a. The locations are deduced from a cysteine-accessibility assay. This assay is based on the monotonic dependence of the fraction pegylated for a given cysteine in an all-extended tape measure on the distance of that cysteine from the PTC2. We have compared the cysteine-accessibility results obtained for a-hairpin constructs with those obtained for this all-extended tape measure. The accessibility of the C-terminal cysteine in the tunnel, in a-hairpin constructs, is similar to the accessibility of cysteines in the tape measure with the same DPTC, demonstrating that these cysteines reside at the same location in the tunnel. This similarity is expected, given that the segment of the peptide between the C-terminal cysteine of the a-hairpin (149C) and the PTC is completely extended3. Thus, the measured accessibility of 149C is not an artifact of secondary-structure formation in the region downstream from 149C, nor is it an artifact of blocked access to 149C by the upstream a4 and a5 helices in the tunnel. Dynamics of the nascent peptide–tunnel complex Our results suggest that the nascent peptide–tunnel complex may be more dynamic than originally believed. At one location (DPTC 24), the level of cross-linking suggests that the a-hairpin exists in a time-averaged equilibrium between an extended a4-a5 species and a tertiary hairpin. Dynamics are also manifest in the distal portion of the tunnel that provides an entropic window for initial exploration of local peptide regions for folding. This represents a restricted time and space in which the nascent peptide visits potential tertiary

NUMBER 4

APRIL 2009

409


ARTICLES conformational states. It is possible that this region, the last 20 A˚ near the exit port, is lined by chaperone proteins, which operationally extend the length of an 80 A˚ tunnel. Estimates of the length of the tunnel vary from 80 A˚ to 112 A˚2,9,10,12,28. Regardless, this region shows a monotonically increasing accessibility with increasing distance from the PTC2,30. If this increased accessibility represents a widening of the tunnel, this will allow an increased number and mobility of water molecules and ions, and therefore an increased driving force for burying hydrophobic residues at a tertiary folded interface (hydrophobic effect37). Both a4-a5 and b1-b2 hairpins contain hydrophobic residues that are in close contact all along their folded interfaces13. The final tertiary hairpin fold is stabilized by constraints of the additional secondary and tertiary interactions of the whole folded T1 domain outside the tunnel. Our results highlight three principles that are likely to have critical roles in protein folding during biogenesis. The first is that intramolecular tertiary interactions occur before the nascent peptide has fully emerged from the ribosomal tunnel, allowing for a regulatory role of folding by the ribosome and its attendant chaperones. The second is that the nascent peptide in the exit port of the tunnel is structurally dynamic and can therefore explore a myriad of conformations before fully emerging from the ribosome. This may enable the peptide to fold more efficiently than it would in a less confined space. The third is that tertiary interactions are influenced by subdomain folding elsewhere in the nascent peptide. We propose that these three principles constrain and shape the earliest stages of protein folding and may be used to help us to understand the molecular basis for folding defects that underlie protein malfunction. METHODS Constructs 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 ml1 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. Orthophenyldimaleimide (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

410

VOLUME 16

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 pegylation. 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 containing 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 ml1) 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 & Molecular Biology website. ACKNOWLEDGMENTS We thank R. Horn, J. Frank and U. Hartl for careful reading of the manuscript and R. Horn for helpful discussion. This research was funded by the US National Institutes of Health grant GM 52302 to C.D. AUTHOR CONTRIBUTIONS A.K. performed the experiments; A.K. and C.D. designed the research, interpreted the results and wrote the manuscript.

NUMBER 4

APRIL 2009


ARTICLES


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 of Kv1.3 during biogenesis. Neuron 44, 295–307 (2004). 2. Lu, J. & Deutsch, C. Secondary structure formation of a transmembrane segment in Kv channels. 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 secretory proteins differ in FRET-detected folding far inside the ribosome and in their exposure to ribosomal proteins. Cell 116, 725–736 (2004). 5. Mingarro, I., Nilsson, I., Whitley, P. & von Heijne, G. Different conformations of nascent polypeptides during translocation across the ER membrane. BMC Cell Biol. 1, 3 (2000). 6. Kowarik, M., Kung, S., Martoglio, B. & Helenius, A. Protein folding during cotranslational 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. Nucleic Acid Res. Mol. Biol. 66, 41–66 (2001). 8. Matlack, K.E. & Walter, P. The 70 carboxyl-terminal amino acids of nascent secretory proteins are protected from proteolysis by the ribosome and the protein translocation apparatus 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 atomic structure 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 of ribosome activity in peptide bond synthesis. Science 289, 920–930 (2000). 11. Menetret, J.F. et al. The structure of ribosome-channel complexes engaged in protein translocation. Mol. Cell 6, 1219–1232 (2000). 12. Beckmann, R. et al. Architecture of the protein-conducting channel associated with the translating 80S ribosome. Cell 107, 361–372 (2001). 13. Minor, D.L. et al. The polar T1 interface is linked to conformational changes that open the voltage-gated potassium channel. Cell 102, 657–670 (2000). 14. Kreusch, A., Pfaffinger, P.J., Stevens, C.F. & Choe, S. Crystal structure of the tetramerization 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 hydrophilic amino-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+ channel assembly. Neuron 11, 67–76 (1993).


VOLUME 16

17. Xu, J., Yu, W., Jan, J.N., Jan, L. & Li, M. Assembly of voltage-gated potassium channels. Conserved hydrophilic motifs determine subfamily-specific interactions between 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 to T1 domain structure. Nat. Struct. Biol. 7, 403–407 (2000). 20. Kurata, H.T. et al. Amino-terminal determinants of U-type inactivation of voltage-gated K+ channels. J. Biol. Chem. 277, 29045–29053 (2002). 21. Wang, G. & Covarrubias, M. Voltage-dependent gating rearrangements in the intracellular 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 ER membranes while nascent Kv peptides are still attached to ribosomes. Biochemistry 40, 10934–10946 (2001). 23. Kosolapov, A. & Deutsch, C. Folding of the voltage-gated K+ channel T1 recognition domain. J. Biol. Chem. 278, 4305–4313 (2003). 24. Lu, J. & Deutsch, C. Pegylation: a method for assessing topological accessibilities in Kv1.3. Biochemistry 40, 13288–13301 (2001). 25. Robinson, J.M. & Deutsch, C. Coupled tertiary folding and oligomerization of the T1 domain of Kv channels. Neuron 45, 223–232 (2005). 26. Kolb, V.A., Makeyev, E.V. & Spirin, A.S. Folding of firefly luciferase during translation in a cell-free system. EMBO J. 13, 3631–3637 (1994). 27. Makeyev, E.V., Kolb, V.A. & Spirin, A.S. Enzymatic activity of the ribosome-bound nascent polypeptide. FEBS Lett. 378, 166–170 (1996). 28. Voss, N.R., Gerstein, M., Steitz, T.A. & Moore, P.B. The geometry of the ribosomal polypeptide 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 the ribosomal 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 in erythromycin 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 cellular regulation. Nat. Struct. Biol. 10, 366–370 (2003). 35. Tu, D., Blaha, G., Moore, P.B. & Steitz, T.A. Structures of MLSBK antibiotics bound to mutated 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? Trends Biochem. Sci. 28, 18–25 (2003). 37. Chandler, D. Interfaces and the driving force of hydrophobic assembly. Nature 437, 640–647 (2005).

NUMBER 4

APRIL 2009

411

ARTICLES

Acetylation by GCN5 regulates CDC6 phosphorylation in the S phase of the cell cycle


Roberta 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 the early G1 phase of the cell cycle, a process requiring tight regulation to ensure that proper origin licensing occurs once per cell cycle. Here we show that, in late G1 and early S phase, CDC6 is found in a complex also containing Cyclin A, cyclin-dependent kinase (CDK)-2 and the acetyltransferase general control nonderepressible 5 (GCN5). GCN5 specifically acetylates CDC6 at three lysine residues flanking its cyclin-docking motif, and this modification is crucial for the subsequent phosphorylation of the protein by Cyclin A–CDKs at a specific residue close to the acetylation site. GCN5-mediated acetylation and site-specific phosphorylation of CDC6 are both necessary for the relocalization of the protein to the cell cytoplasm in the S phase, as well as to regulate its stability. This two-step, intramolecular regulatory program by sequential modification of CDC6 seems to be essential for proper S-phase progression.

The exact duplication of the genome once per cell division is a prerequisite for allowing proper cell proliferation. In eukaryotic cells, this is primarily achieved by exercising tight control over the process of initiation of DNA replication. In particular, regulation is achieved by temporally separating the assembly of the pre-replicative complex (pre-RC) at origins of DNA replication from the actual start of DNA synthesis1. Origin ‘licensing’ is sequential with the binding of the six-subunit origin recognition complex (ORC) to origin DNA; this determines the recruitment of the CDC6 and CDT1 factors, which in turn promote the loading of the putative DNA replicative helicase minichromosome maintenance protein (MCM) complex2,3. The licensed origins are only subsequently triggered to initiate DNA replication by the concerted actions of the S phase–promoting kinases (CDC7 and CDKs4). To prevent re-replication, origins are thus regulated in such a way that once they have fired they are brought back to an unlicensed state, and re-licensing is inhibited until completion of mitosis5. Multiple lines of evidence accumulated over the last few years have indicated that the CDC6 protein is essential for both the initiation of DNA replication and the regulation of licensing in Saccharomyces cerevisiae, Schizosaccharomyces pombe, Xenopus laevis and mammals (reviewed in refs. 6–8). This protein is indispensable for the formation and maintenance of pre-RCs and for the loading of MCM onto origins of replication9–12. Both the S. cerevisiae Cdc6p and S. pombe Cdc18 homologs are CDK substrates, and their phosphorylation at the G1-S boundary leads to ubiquitin-mediated proteolysis of the two proteins. In S. pombe, the degradation of Cdc18p seems to be a straightforward mechanism to restrict the frequency of origin usage to once per cell

cycle, because overexpression of Cdc18p induces re-replication and mutants lacking CDK consensus sites are more efficient than wild-type Cdc18p at this process (reviewed in ref. 6). Human CDC6 is also phosphorylated by CDKs at three specific serine residues in its N-terminal domain, at positions 54, 74 and 106; these seem to be the only functional CDK sites in CDC6 (refs. 13–15). The protein, however, is not destroyed during S phase, but remains present throughout S, G2 and M phase, and it is eventually degraded, by the nonmitotic form of the anaphase-promoting complex/cyclosome (APC/C) E3 ubiquitin ligase, only after entering the G1 phase of the subsequent cell cycle; the same event also occurs in cells that have exited the cell cycle13–17. Recent work has indicated that, in late G1 phase for cells exiting quiescence or at the end of mitosis in cycling cells, CDKs containing Cyclin E phosphorylate CDC6 and that this modification prevents APC/C-mediated proteolysis of the protein18. Prevention of CDC6 degradation would be a mechanism to ensure pre-RC assembly and origin licensing during a ‘window of opportunity’ when other APC/C substrates that are inhibitory for these processes, such as geminin and Cyclin A, are instead degraded. CDC6 phosphorylation by Cyclin E–CDKs provides a molecular explanation for pre-RC assembly and origin licensing in the G1 phase. At this time point, CDC6 is exclusively found in the nucleus and binds to chromatin19. Once the cells enter the S phase, however, at least part of the endogenously expressed protein is relocalized to the cytoplasm13,14,16,20,21, an occurrence that is particularly evident when the protein is overexpressed22. The relocalization of CDC6 is temporally coincident with the appearance of Cyclin A, and it seems to be specifically dependent on its phosphorylation by Cyclin A–containing

1Molecular Biology Laboratory, Scuola Normale Superiore, AREA della Ricerca del CNR, Pisa, Italy. 2Molecular Medicine Laboratory, International Centre for Genetic Engineering and Biotechnology (ICGEB), Trieste, Italy. 3Department of Biomedicine, Faculty of Medicine, University of Trieste, Italy. 4Present address: IFOM-Institute of Molecular Oncology-Foundation, Milan, Italy. 5These authors contributed equally to this work. Correspondence should be addressed to M.G. ([email protected]).

Received 12 May 2008; accepted 4 March 2009; published online 3 April 2009; doi:10.1038/nsmb.1583

412

VOLUME 16

NUMBER 4

APRIL 2009


ARTICLES

Ac-CDC6 (endog.) *

*

WB anti-CDC6

R

KS K

K

RS K

K

KS R

K92R K105R

R

RS K

K105R K109R K92R K105R K109R (K3R)

K

RS R

R

RS R

g Flag-CDC6 Vector + HA-GCN5 – HA-GCN5mut – Ac-CDC6 (transf.) GCN5/GCN5mut CDC6/CDC6(K3R) (transf.) CDC6/CDC6(K3R) (transf.) GCN5/GCN5mut Tubulin

f

– + –

Flag-CDC6(K3R) – – +

+ – –

– + – WB anti-Ac-Lys WB anti-HA IP anti-Flag WB anti-Flag WB anti-Flag WB anti-HA WB anti-tubulin

WCL

CDC6 proteins GCN5

* * *


K92R K105R K109R

Ac-GCN5

*

P Ac Ac

KS K

C D C 1– 6 90 91 –5 11 61 1– 18 56 6– 1 56 1

G ST

C L IP C o IP ntr an ol tiAc -L ys

*

W

b

d

Cy dock motif

K

B

1–90 Ac-GST-CDC6 91–561 111–561 186–561

Ac-Histones

Ac

CDC6

105 106 109

A NLS KEN box

e

NES

92

CDC6

ATPase Destr Cy dock ORC box motif hom Leu zip

G S C T D C K9 6 K12R 0 K1 5R 0 K9 9R 2 K1 R K 05 10 R 5R K K K1 92R 10 09 K 9R R 10 (K 5R 3R )

c

G ST G ST -C D C 6

a NE

GST

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 GCN5dependent 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 all mutants (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.

CDKs, because it is induced by overexpression of Cyclin A but not of Cyclin E (ref. 14). These observations point to the existence of subtle differences in the phosphorylation pattern that the different CDKs impinge on CDC6 in subsequent phases of the cell cycle, suggesting the existence of additional mechanisms that regulate the intracellular fate and stability of CDC6. Resulting from a search for human pre-RC components that might interact with cellular histone acetyltransferases (HATs), here we describe our serendipitous discovery that human CDC6 associates with, and is acetylated by, GCN5. In yeast, this protein represents the enzymatic subunit of the transcriptional regulatory complex SAGA (Spt-Ada-Gcn5 acetyltransferase)23 and seems essential for cell-cycle progression24–27. We show that, in human cells, CDC6 takes part in the formation of a complex with Cyclin A–CDK2 that also includes GCN5, that this HAT acetylates CDC6 during the early S phase of the cell cycle and, finally, that this event is crucial for proper cellcycle progression. RESULTS GCN5 acetyltransferase binds and acetylates CDC6 While searching for cellular HATs that interact with the pre-RC members, 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 whether endogenous CDC6 was also acetylated, HeLa cell extracts were immunoprecipitated using an anti–acetyllysine antibody and immunoblotted using an anti-CDC6 antibody. A 63-kDa band corresponding to


VOLUME 16

endogenous CDC6 protein was readily detected (Fig. 1b); analogous findings 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 with trichostatin A (TSA), an inhibitor of cellular deacetylases, carried out immunoprecipitation of cell lysates with an anti-Flag antibody and immunoblotted with an anti–acetyllysine antibody. We detected acetylated Flag-CDC6 in the anti-Flag immunoprecipitates (Supplementary Fig. 1a online). Next we observed that GST-CDC6 is a specific substrate of recombinant GCN5 in an in vitro HAT assay (Supplementary Fig. 1b). By using a series of GST-CDC6 fusion proteins (Fig. 1c), we concluded that the target region for GCN5 acetylation lies between amino acids 91 and 110 of CDC6 (Fig. 1d) and that the three lysines present in this region (at positions 92, 105 and 109; Fig. 1e) were all targets for GCN5-mediated acetylation (Fig. 1f). Factor acetylation by HATs is often concomitant with the specific binding of the enzyme to its substrate. Indeed, this was also found to be the case for GCN5 and CDC6; the interaction between the two proteins requires the integrity of 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 cell lysates, we estimated that B18% of total CDC6 was acetylated inside the cells (Supplementary Fig. 2a–d online). This level was clearly enhanced by the overexpression of enzymatically active GCN5 but not of its catalytically inactive mutant GCN5mut28; this enhancement was higher than that obtained by cell treatment with trichostatin A (TSA) (Supplementary Fig. 2e). Similar to endogenous CDC6, acetylation of

NUMBER 4

APRIL 2009

413

ARTICLES levels, but it led to a substantial accumulation of the protein inside the cells while decreasing the levels of Cyclin E and Cyclin A, suggesting perturbed cell-cycle progression (Fig. 2b). These results initially disclosed an unexpected link between GCN5mediated CDC6 acetylation and the specific phosphorylation of the protein on Ser106. To further explore this issue, we analyzed the levels of phosphorylation of transfected wild-type CDC6, of the K3R mutant and of an additional mutant we constructed bearing a serineto-alanine substitution at position 106 (S106A; Fig. 2c). Notably, the K3R mutant was not phosphorylated on Ser106, similarly to the CDC6 acetylation by GCN5 affects Ser106 phosphorylation The acetylated lysine residues of CDC6 frame the cyclin-docking motif S106A mutant and unlike wild-type CDC6 (Fig. 2d). Both mutants, of the protein20. In particular, Lys105 is adjacent to Ser106, one of the however, were still phosphorylated on Ser54. The anti–pSer106-CDC6 three CDC6 serine residues (54, 74 and 106) that are specifically antibody was still able to recognize the K3R mutant when phosphoryphosphorylated by the CDKs13,14. We therefore questioned whether lated in vitro by Cyclin A–CDK, thus indicating that the K3R GCN5-dependent acetylation might affect CDC6 phosphorylation. We mutation per se did not impair epitope recognition (Supplementary initially observed that the overexpression of GCN5 increased the levels Fig. 3 online). To further define the role of GCN5 in mediating endogenous CDC6 of CDC6 phosphorylated at Ser106 (pSer106), as detected either after immunoprecipitation (Fig. 2a) or by straight western blotting Ser106 phosphorylation, we knocked down GCN5 by RNA inter(Fig. 2b). In the same lysates, no modification was observed for ference29. GCN5 depletion markedly increased the levels of total CDC6 phosphorylated at Ser54 (pSer54). In contrast to wild-type CDC6 (Fig. 2e) and selectively inhibited CDC6 Ser106 phosphorylaGCN5, the expression of GCN5mut did not increase CDC6 pSer106 tion, while leaving Ser54 phosphorylation unaltered (Fig. 2f). In the same siRNA-treated cell lysates, the level of MCM3 protein, a factor also modified by + HA-GCN5 – HA-GCN5 + + + a d – Vector + Flag-CDC6 + – – acetylation30, was unaltered. In addition, IP anti-CDC6 pS106 Flag-CDC6(K3R) – + – CDC6 pS106 WB anti-CDC6 silencing of GCN5 did not decrease the levels Flag-CDC6(S106A) – – + CDC6 CDC6 pS106 WB anti-CDC6 pS106 of CDK2–Cyclin A nor it impaired phosphorWB anti-CDC6 CDC6 pS54 WB anti-CDC6 pS54 ylation of other CDK2–Cyclin A targets, such WCL GCN5 WB anti-HA GCN5 WB anti-HA as retinoblastoma-1 (Rb)31 (Supplementary Tubulin WB anti-tubulin CDC6 & mutants WB anti-Flag Fig. 4a online). Tubulin WB anti-tubulin Collectively, these data are consistent with HA-GCN5 – + – b HA-GCN5mut the possibility that GCN5-mediated acetyla– – + e siGCN5 Mock Vector + – – tion perturbs the levels of CDC6 protein WB anti-GCN5 GCN5 CDC6 WB anti-CDC6 by specifically affecting its phosphorylation WB anti-CDC6 CDC6 CDC6 pS106 WB anti-CDC6 pS106 on Ser106. WB anti-tubulin Tubulin CDC6 pS54 WB anti-CDC6 pS54

Tubulin

WB anti-tubulin

5 N C

uc

-G

f

-L

WB anti-HA

si

WB anti-Cyclin A

oc k

Cyclin A GCN5/GCN5mut

48h 72h 48h 72h 150 nM 300 nM

WCL

si

WB anti-Cyclin E

M

Cyclin E

WB anti-CDC6

CDC6 pS54 WCL IP anti-CDC6 pS54

K3R R S106A

R

R A

5

KS K

WB anti-CDC6 WCLIP anti-CDC6 pS106 uc si -G C N

CDC6 K

CDC6 pS106

si -L

P Ac Ac

oc k

Cy dock motif

105 106 109

Ac

M

c 92


transfected wild-type CDC6 was also markedly increased in response to wild-type GCN5 but not to GCN5mut (Fig. 1g). In contrast, the CDC6 K3R mutant, which was not acetylated by GCN5 in vitro, was not sensitive to GCN5 overexpression in vivo, although it still coimmunoprecipitated with the enzyme. Collectively, these results clearly indicate that CDC6 lysines 92, 105 and 109 are specific targets for GCN5 acetylation both in vitro and inside the cells.

GCN5

WB anti-GCN5

MCM3

WB anti-MCM3

Tubulin

WB anti-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 with CDC6 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 antiGCN5 siRNA, or with an irrelevant siRNA anti-luciferase. The lower immunoblots show additional results from whole-cell lysates using the indicated antibodies.

414

VOLUME 16

NUMBER 4

APRIL 2009

GCN5 complexes with Cyclin A–CDK2 and acetylates CDC6 in S phase We analyzed the levels of CDC6 acetylation in human glioblastoma T98G cells, which accumulate in G0 upon serum starvation and then synchronously enter G1 after readdition of serum32,33 (Fig. 3a). The levels of GCN5 were reduced in cell lysates from serum-starved cells as compared to asynchronous cells; however, they started to increase 6 h after serum addition (early G1), peaked at 20 h (early S) and decreased at 24 h (mid S-phase) (Fig. 3b). The levels of CDC6 were almost undetectable under serum starvation and started to progressively increase from 16 h after serum re-addition onward, in accordance with earlier observations18. Of interest, acetylated CDC6 showed a clear enrichment at 20 h (early S), followed by a return to basal levels at 24 h (mid S). Thus, the amounts of GCN5 and acetylated CDC6 both peak in early S phase. Similar findings were also obtained in synchronized HeLa cells (data not shown). Next we monitored the levels of the two CDC6 phospho-isoforms during cell-cycle


ARTICLES Figure 3 CDC6 acetylation is cell cycle AS Time after dependent. (a) T98G cells synchronization. Flow T98G 72 h Add serum serum addition (h): 6 16 20 24 28 cytometry profiles of asynchronous cells (AS), cells Serum WB anti-CDC6 CDC6 starvation cells blocked in G0 by serum starvation (0 h) or CDC6 pS54 WB anti-CDC6 pS54 16 h 0h 6h cells at different times after serum stimulation, (late G1-S) IP anti-CDC6 pS106 (early G1) CDC6 pS106 WB anti-CDC6 with the deduced cell-cycle phase indicated. CDK2 WB anti-CDK2 (b) Acetylated CDC6 peaks in early S phase. The Cyclin E WB anti-Cyclin E 28 h 24 h 20 h upper four immunoblots show the levels of the (early S) (mid S) Cyclin A WB anti-Cyclin A indicated proteins in whole-cell extracts from Tubulin WB anti-tubulin cells synchronized at different time points. The lower blot shows the levels of acetylated CDC6 Time after Time after after immunoprecipitation with an anti-CDC6 serum serum 20 24 addition (h): AS 6 antibody followed by western blotting with an addition (h): AS 0 6 16 20 24 28 IP anti-CDC6 CDC6 WB anti-CDC6 anti-Ac-Lys antibody. (c) Phosphorylation of CDC6 WB anti-GCN5 GCN5 Cyclin A WB anti-Cyclin A on Ser106 is concomitant with the appearance of CDC6 WB anti-CDC6 IP anti-GCN5 GCN5 WB anti-CDC6 WB anti-GCN5 CDC6 Cyclin A, at the same time point at which CDC6 Tubulin WB anti-tubulin is acetylated. The immunoblots show the levels WB anti-CDC6 IP antiCDC6 pS106 Ac-CDC6 GCN5 WB anti-Ac-Lys IP antiWB anti-GCN5 CDC6 pS106 of the indicated proteins in whole-cell lysates WB anti-CDC6 CDC6 CDC6 of synchronized cells, with the exception of CDC6 pS54 WB anti-CDC6 IP antiCDC6 pSer106, which was detected after WB anti-GCN5 CDC6 pS54 GCN5 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.

a

c


b

d

progression. After 6 h (late G1) phosphorylation of Ser54 was low, whereas from 16 h onward it markedly increased (Fig. 3c). Phosphorylation of Ser106 was less pronounced at 16 h as compared to phosphorylationof Ser54, and pSer106 levels peaked at 20–24 h. Notably, CDC6 is known to be phosphorylated in the S phase in a Cyclin A–CDK2–dependent manner13,14,16,20,21. Indeed, in the T98G synchronization, Cyclin E levels peaked at 6 h (early G1) and 16 h (late G1), whereas Cyclin A started to appear at 16 h and its levels increased at 20 h (early S) and 24 h (mid S) (Fig. 3c). These findings are thus consistent with the possibility that CDC6 Ser106 phosphorylation might be attributable to Cyclin A–CDK2. Indeed, we observed that Ser106 phosphorylation substantially increased upon Cyclin A (but not Cyclin E) overexpression, whereas the levels of pSer54 remained unaltered (Supplementary Fig. 4b). We proceeded to investigate whether the interaction between CDC6 and GCN5 might vary during cell-cycle progression. As concluded from co-immunoprecipitation experiments using extracts from synchronized T98G cells, binding between CDC6 and GCN5 was maximal at 20 h after serum addition, the same time point at which both GCN5 levels and CDC6 acetylation peaked (Fig. 3d). Both pSer54 and pSer106 CDC6 bound GCN5 at this time point. However, the amount of co-immunoprecipitated GCN5 was higher in the anti-pSer106 immunoprecipitates. This was even more clear at 24 h after serum addition (mid S), when co-immunoprecipitation was detected using the only the anti-pSer106 but not the anti-pSer54 antibody. Taken together, the cell-cycle experiments showed that: (i) CDC6 associated with GCN5 in early S phase; (ii) at the same time point, the protein became acetylated; (iii) binding between GCN5 and CDC6 was preferential for the CDC6 pSer106 phospho-isoform; (iv) acetylation, phosphorylation and GCN5 binding were concomitant with the expression of Cyclin A; (v) the overexpression of Cyclin A selectively increased phosphorylation of CDC6 on Ser106. As GCN5 often participates in the formation of multicomponent protein complexes in different species23,34,35, we tested whether it might also associate with Cyclin A–CDK2. Indeed, both GCN5 and CDC6 were coimmunoprecipitated together with CDK2 by an antibody against Cyclin A, as was CDK2 by an antibody against GCN5 (Supplementary Fig. 5a and 5b online, respectively).


VOLUME 16

The observations above are consistent with the possibility that CDC6 phosphorylation on Ser106 might preferentially occur on acetylated CDC6. Accordingly, we found that higher levels of acetylated CDC6 were associated with pSer106 CDC6 rather than with pSer54 CDC6 or total CDC6 (Supplementary Fig. 6a,b online). CDC6 modifications regulate its subcellular localization In human cells, CDC6 is phosphorylated in a Cyclin A–CDK2– dependent manner during the S phase and then translocated to the cytosol and subsequently degraded6,13,14,20,21. Confocal immunofluorescence analysis performed in nonsynchronized HeLa cells using a monoclonal anti-CDC6 antibody revealed that in about 55% of the cells endogenous CDC6 was nuclear, whereas in about 45% of cells the protein had a cytoplasmic localization (Fig. 4a). However, most (490%) of the asynchronous cells that expressed Cyclin A, an S-phase marker, belonged to the subset of cells with exclusive cytoplasmic localization of CDC6 (Fig. 4b). In keeping with these findings, CDC6 was localized in the cytoplasm in more than 90% of the cells in which Cyclin A had been overexpressed (Fig. 4c). We examined whether the acetylation of the factor might influence its subcellular localization. Indeed, we found that cell treatment with TSA increased the number of cells with exclusive cytoplasmic localization of CDC6 (68% versus 45%, P o 0.01; Fig. 4c). Analogous results were obtained by overexpression of active GCN5 (71% of cells with cytoplasmic localization among those positive for the expression of the transfected protein, P o 0.01), but not of GCN5mut (Fig. 4d). Flow cytometry experiments on the cells transfected with wild-type GCN5 showed an appreciable increase in the number of S-phase cells, similar to that obtained by overexpression of Cyclin A (Supplementary Fig. 7 online). Finally, we knocked down GCN5 expression by RNA interference; this treatment increased the number of cells in the G1 phase (Fig. 4e) and determined a prevalently nuclear CDC6 localization (Fig. 4f). Next we explored the subcellular localization of transfected CDC6 S106A, 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 to endogenous CDC6 (B55% of the cells with nuclear localization;

NUMBER 4

APRIL 2009

415

mutant was detected in the cytoplasm. Of interest, however, the subnuclear distribution of the two mutants was markedly different, in that the S106A protein was found in both the nucleosolic and chromatin fractions (30% and 63% respectively), whereas the K3R mutant was detectable only in the chromatin pellet (92%). These findings are thus consistent with the conclusion that CDC6 acetylation is essential for the detachment of CDC6 from chromatin, whereas Ser106 phosphorylation later promotes relocalization of the protein to the cytoplasm. In keeping with this conclusion, when lysates from nontransfected cells were partitioned in the same manner, endogenous CDC6 pSer54 was found distributed in both the cytoplasmic and nuclear insoluble compartments, similarly to the distribution of total CDC6. In contrast, CDC6 pSer106 was present exclusively in the cytoplasmic fraction (Supplementary Fig. 9 online).

75

50

50

25

25

0

0

Merge

N

si-GCN5

HeLa cells

C

100 80 60 G2-M S G0/G1

40 20 0 si -L -G uc C N 5

WB anti-tubulin

si

75

IF anti-CDC6 IF anti-HA (endogenous CDC6) (GCN5/GCN5mut)

+HA-GCN5mut

WB anti-GCN5

f

100

IF anti-CDC6 (endogenous CDC6)

75 50 25 0

Mock N

C

100 75 50

si-Luc

25 0

N

C

100 75 50

si-GCN5

25 0

N

C

IF anti-GCN5 (endogenous GCN5)

Merge

Number of CDC6+ cells (%)

100

N C N C IF anti-CDC6 IF anti-Cyclin A (endogenous CDC6) (endogenous Cyclin A) Merge

+HA-GCN5

GCN5 Tubulin

100


0

– + Cell-cycle distribution (%)

25

Number of CDC6 cells (%)

50

+ –

100


d

75

si-Luc si-GCN5

si-Luc 100

Number of Cyclin A– positive cells (%)

IF anti-CDC6 (endogenous CDC6)

b

100

Effects of CDC6 modifications on cell-cycle progression Next we examined whether the change in chromatin association and subcellular localization of the two CDC6 mutants perturbed normal cell-cycle progression. Overexpression of wild-type CDC6 or some of its other mutants does not exert a strong effect in primary cells, because multiple mechanisms (including CDT1 inhibition by geminin

e

Number of Number of HA-positive cells (%) HA-positive cells (%)

c

+

a IF anti-CDC6 (endogenous CDC6)

Fig. 5a). Similarly to what was observed for endogenous CDC6, this distribution was modified by cell treatment with TSA, at which CDC6 became cytoplasmic in 465% of the cells (P o 0.01). Notably, both the S106A and the K3R mutants were strictly nuclear, and their localization was altered neither by TSA treatment (Fig. 5a) nor by GCN5 overexpression (Supplementary Fig. 8 online). Both mutants, however, were still able to co-immunoprecipitate endogenous Cyclin A and hemagglutin-tagged GCN5, with similar efficiency to the wildtype protein (data not shown). Taken together, these observations clearly indicate that both acetylation and Ser106 phosphorylation of CDC6 are required for its relocation to the cytoplasm. Next we checked whether the exclusive nuclear localization of the CDC6 Ser106 and K3R mutants might reflect the association of these proteins with chromatin. Whole-cell lysates from cells transfected with wild-type CDC6 or with the two mutants were fractionated to separate the cytosolic, nuclear soluble and nuclear insoluble compartments19. The wild-type protein was found to localize in the insoluble nuclear compartment (B50% of total) and in the nucleosolic and cytoplasmic fractions (15% and 35% respectively; see Fig. 5b for western blots and Fig. 5c for quantification). In sharp contrast, and in agreement with the immunofluorescence data, less than 10% of either



ARTICLES

100

75 50 25 0

N

C

N

C

N

C

75 50 25 0

75 50 25 0

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) and endogenous 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.

416

VOLUME 16

NUMBER 4

APRIL 2009


ARTICLES (from 16% to 3%). This result is consistent with the conclusion that the overexpression of the K3R mutant, which is tightly chromatinbound, specifically impairs S-phase progression without affecting entry into the S phase. We observed no detectable re-replication with any of the mutants. Cytoplasmic transport of human CDC6 correlates with degradation of the protein6,13,14,20,21. We therefore wanted to assess the stability of the constitutively nuclear CDC6 S106A and K3R mutants relative to wild-type CDC6. After treatment of asynchronous cells with cycloheximide (CHX) to block protein synthesis, the half-life of transfected wild-type CDC6 was less than 1 h, similar to endogenous CDC6 and analogous with published findings39. In contrast, both the S106A and K3R mutants were remarkably more stable; in particular, after 4 h of CHX treatment, more than 75% of K3R was still present inside

and p53-dependent DNA-damage response) prevent cellular DNA re-replication36,37. However, overexpression of CDC6 in p53 / cells has been shown to escape checkpoint inhibition and to induce, to some extent, DNA re-replication38. We therefore analyzed the effect of our 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, we visualized BrdU incorporation and DNA content by flow cytometry. As shown in Figure 5d and quantified in Figure 5e, the overexpression of both wild-type CDC6 and the S106A mutant determined a reduction in the number of cells incorporating BrdU; in sharp contrast, the K3R mutant markedly increased the number of cells involved in DNA synthesis (from 43% to 67%), while reducing the number of cells in G1 (from 44% to 30%) and, most notably, in G2-M Flag-CDC6(K3R)

b

Flag-CDC6(S106A)

WCL Cyt Sol Ins Flag-CDC6 Flag-CDC6(S106A)

100

100

100

100

100

75

75

75

75

75

75

50

50

50

50

50

50

25

25

25

25

25

25

0

0

N C

d Ins Sol Cyt

60 40 20

0

N C

-BrdU

104

+BrdU BrdU

0

N C

N C

Mock

BrdU

10

102 101 G1

G2-M

G1

G2-M

10 C CD 6( C C S1 6 D 06 C 6( A) K3 R )

0

C

D

f

Time after CHX addition (h): 0 CDC6 (endog.)

e Mock CDC6 CDC6(S106A) CDC6(K3R)

80 60

2

20

Tubulin

0

G0/G1

BrdU

G2/M

CDC6

CDC6(S106A)

BrdU

BrdU

CDC6(K3R) BrdU

102 101 G1

0

G2-M

G1

G2-M

G1 G2-M

200 400 600 FL2A (PI)

4 WB anti-CDC6

WB anti-Flag

Tubulin Flag-CDC6(S106A)

WB anti-ORC2

100

WB anti-tubulin

Flag-CDC6

40

ORC2

g 1

Tubulin 100

WB anti-tubulin

103

100

200 400 600 FL2A (PI)

Tubulin

N C

104

BrdU

3

0

0

0

WB anti-tubulin WB anti-Flag

% CDC6 protein

N C

FL1H (BrdU)

0

80 % of total

WB anti-Flag

Flag-CDC6(K3R) 100

FL1H (BrdU)

Number of cells (%)

IF anti-Flag (transfected CDC6)

Flag-CDC6

c 100

Cell cycle distribution (%)


a

75

Flag-CDC6(K3R) Flag-CDC6(S106A)

50 Flag-CDC6 25

CDC6(endog.)

WB anti-tubulin 0 WB anti-Flag

Flag-CDC6(K3R) Tubulin

WB anti-tubulin

0

1

2

3

4

CHX (h)

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) from untransfected 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.


VOLUME 16

NUMBER 4

APRIL 2009

417

ARTICLES CDC6 NUCLEAR

CDC6 CYTOPLASMIC Relocalization to the cytoplasm

Degradation Ser54, Ser74 phosphorylation by Cyclin E–CDK2 APC/C

Lys92, Lys105, Lys109 acetylation by GCN5

CDK2

GCN5 P P

P P

Ac Ac

Cyclin E

Ser106 phosphorylation by Cyclin A–CDK2

Release from chromatin P P

Ac

CDK2

GCN5 Ac Ac

P Ac

Cyclin A CDC6

CDC6

G0

G1

CDC6

CDC6

G1-S

S

Cyclin E 6h

16 h

20 h

24 h

28 h


Cyclin A

the cells (Fig. 5f,g). Thus, by promoting CDC6 detachment from chromatin and relocalization to the cytoplasm, acetylation facilitates CDC6 degradation. This conclusion fully concurs with the observation that GCN5 depletion and GCN5mut overexpression both markedly increased the total amount of CDC6 (Fig. 2e and 2b, respectively). DISCUSSION Our data show that CDC6 is acetylated in vivo by the GCN5 HAT at three lysines that frame the Cyclin-docking motif in the N terminus of the protein, a region that is not directly involved in pre-RC formation38. Acetylation regulates the levels of CDC6 pSer106, because the overexpression of GCN5 increases phosphorylation at this residue, whereas the triple mutant in the three acetylated lysines is no longer phosphorylated. During the cell cycle, acetylation of CDC6 occurs in the early S phase, when the levels of both Cyclin A and GCN5 peak. At this time point, both GCN5 and CDC6 are found in a complex with Cyclin A and CDK2. Cyclin A–mediated phosphorylation of CDC6 on Ser106 requires prior acetylation of CDC6. Finally, cell treatment with a deacetylase inhibitor or overexpression of GCN5 force cytoplasmic relocalization of both endogenous and transfected CDC6, an effect that is also obtained by transfection of Cyclin A. Consistently, CDC6 proteins bearing mutations at either Ser106 or at the three lysines that are acetylated are exclusively nuclear and their stability is increased. Work originally performed in the Xenopus in vitro replication system indicated that Cyclins E and A have specialized roles during the transition from G0 to S phase40. Whereas Cyclin E stimulates preRC assembly, Cyclin A activates DNA synthesis by replication complexes that are already assembled, while also inhibiting the assembly of new complexes. Thus, Cyclin E opens a ‘window of opportunity’ for pre-RC assembly that is closed by Cyclin A40. Recent work indicates that an essential mechanism that permits the opening of this window is the specific phosphorylation of CDC6 by Cyclin E, which prevents degradation of the protein by the APC/C and thus permits pre-RC assembly18. This conclusion has been reached mainly by using an antibody against CDC6 phosphorylated on Ser54. Our work extends these findings further by showing that, when the cells enter the S phase, CDC6 pSer54 is found in a complex that also includes GCN5, Cyclin A and CDK2. At this moment, GCN5 specifically acetylates CDC6. This modification determines its release from chromatin and allows its further phosphorylation on Ser106, followed by relocalization of the protein to the cytoplasm and its eventual degradation. The finding that the K3R mutant, which is not acetylated, is still normally phosphorylated on Ser54 but not at all on

418

VOLUME 16

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 interact in early S-phase cells. Phosphorylation of CDC6 on Ser106 determines the relocalization of the protein to the cytoplasm, followed by its degradation.

Ser106 and is specifically chromatin bound is fully consistent with this conclusion. Equally consistent are the observations that overexpression of the K3R mutant leads to a marked increase in the number of BrdU-positive cells and a substantial decrease of both G1 and, most remarkably, G2-M cells and that both the K3R and S106A mutants are considerably more stable than the wild-type protein. A model summarizing these findings is drawn schematically in Figure 6. Our results uncover the events that occur at a step that follows preRC assembly and origin firing, namely the steps that coincide with Cyclin A’s appearance. Work performed in X. laevis using an XCdc6 protein mutated in the three CDK phosphorylation sites has indeed suggested that XCdc6 phosphorylation by CDKs is not essential for either regulated binding of XCdc6 to chromatin nor for the subsequent loading of the MCMs, thus suggesting that Cdc6 phosphorylation might be required at later stages of the replication process41. Indeed, our work indicates that acetylation and subsequent specific phosphorylation of CDC6 on Ser106 are essential to allow detachment of the protein from chromatin, relocalization to the cytoplasm and degradation, and that these events are essential to ensure proper cell-cycle progression. Consistent with this conclusion, experimental evidence obtained in both X. laevis and human cells has shown that overexpression of CDC6 in G2 cells inhibits mitosis by inducing a checkpoint pathway involving Chk1 (refs. 42,43), and that this property is modulated by the phosphorylation of CDC6 at particular residues43. The role of CDC6 subcellular localization has roused much controversy in recent years. Several authors have reported convincingly that in the G1 phase the protein is exclusively nuclear, whereas during the S phase most of it relocalizes to the cytoplasm13,14,16,20,21; however other data have suggested that these effects might be a peculiarity of an exogenously overexpressed protein, and that most of the endogenously expressed protein would remain essentially nuclear throughout the cell cycle19,22. Our results in fact challenge this latter conclusion, and clearly indicate that human endogenous CDC6 (stained by an antiCDC6 monoclonal antibody) is found in both the nucleus and the cytoplasm in nonsynchronous cells. Consistent with the reported binding and kinase activity of Cyclin A–CDK2 (refs. 14,20,41), in most of the cells that express endogenous Cyclin A, CDC6 is found only in the cytoplasm (Fig. 4b), and Cyclin A overexpression forces the cytoplasmic relocalization of endogenous CDC6 (Fig. 4c), as also reported by other laboratories14,18. CDC6 acetylation has an essential role in determining the cytoplasmic relocalization of the protein. This conclusion is supported by the observation that the overexpression of GCN5 (but not that of its catalytically inactive mutant) forces the

NUMBER 4

APRIL 2009



ARTICLES cytoplasmic relocation of both endogenous and transfected CDC6, whereas the GCN5 knockdown has the opposite effect. In a consistent manner, both the K3R mutant (which is not acetylated) and the S106A mutant (which is not phosphorylated on Ser106) have a strictly nuclear localization, irrespective of GCN5 overexpression (Fig. 5). These data reinforce the conclusion that the cytoplasmic relocalization of CDC6 is strictly dependent on the consequential acetylation and phosphorylation of CDC6 on Ser106. The results presented in this manuscript also underscore the role of GCN5 as a general cell-cycle regulator44. Indeed, this HAT seems to have a key role in regulating the expression of several cell cycle–related genes, such as Cyclin A, Cyclin D3, PCNA and CDC25B. The observation that GCN5 also regulates the function of one key regulator of pre-RC formation and controls licensing extends this concept further. The finding that CDC6 is specifically acetylated by GCN5 and that this HAT finely tunes its function does not exclude the possibility that the protein might also be the substrate of other HATs. Indeed, recent work performed in Xenopus has shown that CDC6 might 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 protein acetylation and phosphorylation might occur in different proteins as two closely interconnected modifications that are part of a multistep regulatory program48. Examples of other factors in which these modifications are strictly related and, in some instances, sequential, include p53 and p73, Rb, FOXO1 and MYC49–53. It will be interesting to understand the exact changes, resulting from CDC6 acetylation, in the molecular structure of the protein and how these changes might impinge on the subsequent site-specific phosphorylation. METHODS Plasmids. We constructed the expression vectors pFlag-CDC6 and pGEX20TCDC6 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, respectively. 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 pcDNA3HA-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 synchronized 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 cytometry analysis on a FACSCalibur (Becton Dickinson) to simultaneously determine


VOLUME 16

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 sc20698), 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 immunoprecipitation (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) fractions 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 antibodies (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 & Molecular Biology website.

NUMBER 4

APRIL 2009

419

ARTICLES ACKNOWLEDGMENTS This work was supported by grants from the FIRB program of the ‘‘Ministero dell’Istruzione, Universita’ e Ricerca,’’ Italy and from the ‘‘Fondazione CRTrieste’’ of Trieste, Italy. The authors are indebted to H. Masai (Tokyo Metropolitan Institute of Medical Science) for helpful discussion and to A. Dutta (University of Virginia), K. Helin (Biotech Research and Innovation Centre and Centre for Epigenetics), M. Benkirane (Institut de G–e´ne´tique Humaine), J. Pines (Wellcome Trust/Cancer Research UK Gurdon Institute) and H. Masai for the gift of reagents. The authors are grateful to V. Liverani for excellent technical support and to S. Kerbavcic for superb editorial assistance. AUTHOR CONTRIBUTIONS All experiments were performed by R.P. and R.M.-M.; A.C. took part in the design of the initial CDC6 acetylation experiments; M.G. supervised the work 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. Lei, M. & Tye, B.K. Initiating DNA synthesis: from recruiting to activating the MCM complex. 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. Genes Cells 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. Cell Dev. 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 essential for 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 replication origins 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 Mcm proteins onto pre-replicative chromatin in budding yeast. Proc. Natl. Acad. Sci. USA 94, 5611–5616 (1997). 13. Jiang, W., Wells, N.J. & Hunter, T. Multistep regulation of DNA replication by Cdk phosphorylation 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 of mammalian 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 mammalian CDC6 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 is selectively 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 cells by 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: assembly of prereplication complexes in late mitosis. Mol. Cell. Biol. 20, 8602–8612 (2000). 20. Delmolino, L.M., Saha, P. & Dutta, A. Multiple mechanisms regulate subcellular localization of human CDC6. J. Biol. Chem. 276, 26947–26954 (2001). 21. Herbig, U., Griffith, J.W. & Fanning, E. Mutation of Cyclin/CDK phosphorylation sites in HsCdc6 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-54 phosphorylation, 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 on IME2 transcription, during meiosis and during vegetative growth, in budding yeast. Proc. Natl. Acad. Sci. USA 96, 6835–6840 (1999).

420

VOLUME 16

25. Howe, L. et al. Histone H3 specific acetyltransferases are essential for cell cycle progression. Genes Dev. 15, 3144–3154 (2001). 26. Krebs, J.E., Kuo, M.H., Allis, C.D. & Peterson, C.L. Cell cycle-regulated histone acetylation 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 and redundant functions of histone acetylation revealed by mutation of target lysines and loss of the Gcn5p acetyltransferase. EMBO J. 17, 3155–3167 (1998). 28. Paulson, M., Press, C., Smith, E., Tanese, N. & Levy, D.E. IFN-stimulated transcription through 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 acetyltransferase 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 protein MCM3. EMBO Rep. 2, 119–123 (2001). 31. Zarkowska, T. & Mittnacht, S. Differential phosphorylation of the retinoblastoma protein 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 are required 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 distinct GCN5 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 are stable components of ATAC, a novel dAda2A/dGcn5-containing histone acetyltransferase complex in Drosophila. Mol. Cell. Biol. 26, 871–882 (2006). 36. Vaziri, C. et al. A p53-dependent checkpoint pathway prevents rereplication. Mol. Cell 11, 997–1008 (2003). 37. Nishitani, H., Lygerou, Z. & Nishimoto, T. Proteolysis of DNA replication licensing factor Cdt1 in S-phase is performed independently of geminin through its N-terminal region. J. Biol. Chem. 279, 30807–30816 (2004). 38. Drury, L.S., Perkins, G. & Diffley, J.F. The Cdc4/34/53 pathway targets Cdc6p for proteolysis in budding yeast. EMBO J. 16, 5966–5976 (1997). 39. Duursma, A. & Agami, R. p53-dependent regulation of Cdc6 protein stability controls cellular proliferation. Mol. Cell. Biol. 25, 6937–6947 (2005). 40. Coverley, D., Laman, H. & Laskey, R.A. Distinct roles for cyclins E and A during DNA replication complex assembly and activation. Nat. Cell Biol. 4, 523–528 (2002). 41. Pelizon, C., Madine, M.A., Romanowski, P. & Laskey, R.A. Unphosphorylatable mutants of Cdc6 disrupt its nuclear export but still support DNA replication once per cell cycle. Genes Dev. 14, 2526–2533 (2000). 42. Oehlmann, M., Score, A.J. & Blow, J.J. The role of Cdc6 in ensuring complete genome licensing 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. Human replication protein Cdc6 prevents mitosis through a checkpoint mechanism that implicates Chk1. EMBO J. 22, 704–712 (2003). 44. Kikuchi, H., Takami, Y. & Nakayama, T. GCN5: a supervisor in all-inclusive control of vertebrate cell cycle progression through transcription regulation of various cell cyclerelated genes. Gene 347, 83–97 (2005). 45. Iizuka, M., Matsui, T., Takisawa, H. & Smith, M.M. Regulation of replication licensing by 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 ORC1 interact with the histone acetyltransferase HBO1. J. Biol. Chem. 276, 15397–15408 (2001). 47. Iizuka, M. & Stillman, B. Histone acetyltransferase HBO1 interacts with the ORC1 subunit 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 sensitivity to phosphorylation. Proc. Natl. Acad. Sci. USA 102, 11278–11283 (2005). 51. Ozaki, T. et al. Functional implication of p73 protein stability in neuronal cell survival and death. Cancer Lett. 228, 29–35 (2005). 52. Vervoorts, J., Luscher-Firzlaff, J.M. & Luscher, B. The ins and outs of MYC regulation by 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 reversible acetylation. J. Biol. Chem. 275, 10887–10892 (2000). 55. Marcello, A., Massimi, P., Banks, L. & Giacca, M. Adeno-associated virus type 2 rep protein inhibits human papillomavirus type 16 E2 recruitment of the transcriptional coactivator p300. J. Virol. 74, 9090–9098 (2000). 56. Marcello, A. et al. Recruitment of human cyclin T1 to nuclear bodies through direct interaction with the PML protein. EMBO J. 22, 2156–2166 (2003).

NUMBER 4

APRIL 2009


ARTICLES

Regulation of active site coupling in glutaminedependent NAD+ synthetase


Nicole 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 cellular processes. In contrast to humans, Mycobacterium tuberculosis NAD+ biosynthesis is absolutely dependent on the activity of a multifunctional glutamine-dependent NAD+ synthetase, which catalyzes the ATP-dependent formation of NAD+ at the synthetase domain using ammonia derived from L-glutamine in the glutaminase domain. Here we report the kinetics and structural characterization of M. tuberculosis NAD+ synthetase. The kinetics data strongly suggest tightly coupled regulation of the catalytic activities. The structure, the first of a glutamine-dependent NAD+ synthetase, reveals a homooctameric subunit organization suggesting a tight dependence of catalysis on the quaternary structure, a 40-A˚ intersubunit ammonia tunnel and structural elements that may be involved in the transfer of information between catalytic sites.

The importance of NAD+ as the ubiquitous and essential cofactor of enzymes involved in reduction-oxidation reactions has long been recognized1. Recently, nonredox functions of NAD+ have been discovered in DNA repair, telomere maintenance, gene silencing, cell longevity, immune response and Ca2+ signaling1. Depletion of cellular NAD+ is deleterious to cellular metabolism2,3. NAD+ metabolism comprises a network of de novo biosynthetic and recycling pathways that may differ drastically among organisms4–6. The ubiquitous NAD+ synthetase catalyzes the last step in the de novo biosynthesis and, in some recycling pathways, the ATP-dependent transformation of nicotinic acid adenine dinucleotide (NaAD+) to NAD+ (Fig. 1). Because in humans some recycling pathways are not dependent on NAD+ synthetase4,7 and M. tuberculosis NAD+ synthetase (NAD+ synthetaseTB) is essential for NAD+ production in M. tuberculosis8, this enzyme is an important potential drug target. Inhibitors of NAD+ synthetaseTB have the potential to be used not only in chemotherapeutic treatments against active tuberculosis but also against the tuberculosis reservoir represented by the 2 billion asymptomatically infected people8,9. The two types of NAD+ synthetase are the ammonia-dependent enzyme (NAD+ synthetaseNH3), which is present only in prokaryotes and uses ammonia as the nitrogen source, and the glutaminedependent enzyme (NAD+ synthetaseGln), which is present in eukaryotes and prokaryotes including M. tuberculosis5 and uses L-glutamine. The structure of the monofunctional and homodimeric NAD+ synthetaseNH3 has been solved with different ligands10–14, but no structure of NAD+ synthetaseGln has been reported. NAD+ synthetaseGln is a multifunctional enzyme: each subunit has an N-terminal glutamine amidotransferase (GAT)

domain that hydrolyzes L-glutamine to L-glutamate and NH3 and a C-terminal synthetase domain that carries out the ATP-dependent amidation of NaAD+ (ref. 15). NAD+ synthetaseGln is the sole member of a new family of the GAT enzymes with a nitrilase-like (C-N bond cleaving) glutaminase domain16. GAT enzymes catalyze key metabolic reactions in protein, amino acid, cofactor, purine and pyrimidine biosyntheses17,18, characterized by three main events: glutamine hydrolysis, ammonia transfer from one active site to the other and formation of the synthase or synthetase product. In some GAT enzymes, the active sites for glutamine hydrolysis and for product formation are connected by a molecular tunnel implicated in the transfer of the highly reactive ammonia19,20. Kinetics studies have shown that the glutaminase and synthetase activities are in some cases tightly coupled in spite of their spatial separation18. No common mechanism has been identified for the regulation of ammonia transfer and catalysis. The kinetics characterizations reported here establish tightly regulated coupling of the catalytic activities of NAD+ synthetaseTB and identify the formation of the synthetase intermediate (NaAD-AMP) as the trigger for glutaminase activation. These studies highlight important differences between the prokaryotic and eukaryotic NAD+ synthetaseGln, with potential implications for drug development. To understand the structural basis for the observed active site coupling of NAD+ synthetaseGln, we also determined the crystal structure of NAD+ synthetaseTB. The structure reveals a complex subunit organization and an unprecedented intersubunit ammonia tunnel and, along with the results of mutagenesis studies, provides insights on the regulation of active site coupling.

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. ([email protected]) or N.L.-L. ([email protected]).

Received 15 September 2008; accepted 27 January 2009; published online 8 March 2009; doi:10.1038/nsmb.1567


VOLUME 16

NUMBER 4

APRIL 2009

421

ARTICLES by kinetic synergism. Decreases in the KM values of 5- and 12-fold for NaAD+ and O N+ N+ N+ ADP ADP ADP ATP, respectively, when L-glutamine is the O O O 2+ 2+ OH OH ATP/Mg2+ + nitrogen source instead of exogenous ammo+ AMP/Mg + PPi/Mg OH OH OH OH OH OH nia suggest that L-glutamine binding enhances NH3 + NAD NaAD-AMP NaAD+ the affinity of the synthetase active site for its Glutaminase domain substrates. A 179-fold increase in the glutamiNH3+ NH3+ NH3+ nase turnover number measured for the reacNH3 O O NH2 O O S-Cys-E tion in the presence of both synthetase + E-Cys + E-Cys OO O O substrates compared to their absence indicates O O γ-glutamyl thioester intermediate L-Glutamate L-Glutamine a mechanism of activation of the glutaminase domain triggered by changes occurring at the + Gln Figure 1 The reactions catalyzed at the synthetase and glutaminase domains of NAD synthetase . synthetase domain. The KM values for L-glutamine did not change in the presence or absence of NaAD+ and ATP. The kcat values for L-glutamine hydrolysis in the presence of only one of the synthetase RESULTS Kinetics analysis shows optimal kinetics synergism substrates (kcat ¼ 0.004 ± 0.0003 s1 and 0.0037 ± 0.0002 s1 with ATP + We measured the steady-state rate constants for the NAD syntheta- or NaAD+, respectively) are identical to the value measured in the seTB–catalyzed reactions using exogenous ammonia or the ammonia absence of synthetase substrates (kcat ¼ 0.0038 ± 0.0002 s1), indicating produced by the glutamine hydrolysis as described in the Supplemen- that both substrates are necessary for glutaminase activation. tary Methods online and reported in Table 1. The turnover number Full occupancy of the synthetase domain may be necessary and for the reaction using ammonia is at least five- to six-fold faster, sufficient for the observed activation of the glutaminase domain or it considering the difference in ionic strength, than for the reaction with could depend on the formation of the NaAD-AMP intermediate. To distinguish between these two possible mechanisms, we tested the L-glutamine. A rate-limiting step involved in L-glutamine hydrolysis and/or ammonia transport could account for this difference (Fig. 1). ability of AMPCPP, a competitive inhibitor of ATP (Ki ¼ 0.7 ± The steady-state kinetics data support a mechanism of catalysis driven 0.1 mM), to activate the glutaminase domain in the presence of Synthetase domain

O

O


O-

O

O O P O O-

NH2

Ade

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) kcat (s–1)

kcat/KM (s1 mM–1)

0.50 ± 0.02 0.60 ± 0.03

3.8 ± 0.6 5.0 ± 1.0

1.3 ± 0.1 1.5 ± 0.3

0.55 ± 0.01 0.68 ± 0.03

0.42 ± 0.03 0.4 ± 0.1

1.3 ± 0.4

0.0038 ± 0.0002

0.0029 ± 0.0008

0.012 ± 0.0007 0.048 ± 0.003

0.0015 ± 0.0003 0.051 ± 0.0009

KM (mM)b

Assay

Variable substrate

Fixed substrates

NAD+

NaAD+

NAD+

ATP

Gln, ATP Gln, NaAD+

NAD+ Glu

Gln Gln

NaAD+, ATP NaAD+, ATP

Glu

Gln

None

NAD+ Glu

Gln Gln

NaAD+, ATP NaAD+, ATP

7.8 ± 1.4 9.3 ± 1.5

Glu

Gln

None

0.1 ± 0.05

Wild-type NAD+ synthetaseTB 0.13 ± 0.02 0.12 ± 0.03

D656A NAD+ synthetaseTB

0.0055 ± 0.0003

0.055 ± 0.02

kcat (s–1)

kcat/KM (s–1 mM–1)

Ammonia-dependent reaction (ionic strength 0.1 M) KM (mM)b

Assay

Variable substrate

Fixed substrates

NAD+ NAD+

NaAD+ ATP

NH3, ATP NH3, NaAD+

0.7 ± 0.1 1.4 ± 0.2

2.6 ± 0.1 2.9 ± 0.1

3.7 ± 0.5 2.1 ± 0.3

NAD+

NH3

NaAD+, ATP

20 ± 2

3.0 ± 0.1

0.15 ± 0.02

NAD+

NaAD+

NH3, ATP

0.65 ± 0.08

3.7 ± 0.1

5.7 ± 0.7

NAD+ NAD+

ATP NH3

NH3, NaAD+ NaAD+, ATP

0.60 ± 0.09 15 ± 2

3.0 ± 0.1 3.3 ± 0.1

5.0 ± 0.8 0.22 ± 0.03

NAD+

NH3

NaAD+, ATP

0.0045 ± 0.0001

0.051 ± 0.003

Wild-type NAD+ synthetaseTB

C176A NAD+ synthetaseTB

D656A NAD+ synthetaseTB 0.089 ± 0.005

aThe

enzyme activity was measured in 10 mM MgCl2, 1 mM DTT and 50 mM Tris-HCl, pH 8.3, at 37 1C. two substrates.

422

VOLUME 16

bThese

NUMBER 4

values are apparent KM constants at saturating concentrations of the other

APRIL 2009


ARTICLES Table 2 Stoichiometry analysis for NAD+ synthetaseTB–catalyzed reactiona Wild-type NAD+ synthetaseTB Glutamine concentration 1.5 mM 5.0 mM 20.0 mM

AMP/NAD+

Glu/ NAD+

n.d.

1.05 ± 0.05

95 ± 5

n.d. 0.9 ± 0.2

1.12 ± 0.13 1.36 ± 0.13

89 ± 12 73 ± 10

% channel efficiency

Asp656Ala NAD+ synthetaseTB AMP/NAD+

Glu/ NAD+

1.5 mM 5.0 mM

n.d. n.d.

3.1 ± 0.5 3.5 ± 0.4

32 ± 5 28 ± 3

20.0 mM

n.d.

3.5 ± 0.3

29 ± 3

Glutamine concentration

% channel efficiency

aAll


reactions were carried out in 10 mM MgCl2, 1 mM DTT and 50 mM Tris-HCl, pH 8.3, at 37 1C and 0.23 mM NAD+ synthetase. n.d., not determined.

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+ formation in the presence of the ATP substrate analog AMPPCP, which contains the phosphodiester bond cleaved in the NAD+ synthetase– catalyzed reaction. The lack of AMPPCP activity can be attributed to a nonproductive binding mode of AMPPCP. Similarly, the absence of activation by AMPCPP could be the result of a binding mode that inhibits conformational changes required for glutaminase activity. Therefore, we tested the ability of product NAD+ to activate the glutaminase domain in the presence of ATP, but, again, we observed no appreciable increase (kcat ¼ 0.005 ± 0.001 s–1). Taken together, these results indicate that the activation of the glutaminase domain is dependent on the formation of the NaAD-AMP intermediate, confirming that there is a catalytic coupling of active sites. The stoichiometry of the products of the NAD+ synthetaseTB reaction reflects the efficiency of active site coupling and ammonia transfer (Table 2). These data show that NAD+ synthetaseTB achieves maximum efficiency at or below physiologically relevant concentration of L-glutamine (B4.5 mM21) and that the synthetase and glutaminase activities are highly synchronized. As the L-glutamine concentration increased to 20 mM, the activities of the active sites become partially uncoupled and 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+ at 2.35-A˚ resolution. The crystals contained four subunits per asymmetric unit, and the expected biological homooctamer is generated by interaction with an adjacent asymmetric unit related by two-fold crystallographic symmetry (Fig. 2). We confirmed the 600-kDa homooctameric quaternary structure in solution by gel-filtration analysis (data not shown) and found it to be identical to the quaternary structure of the Saccharomyces cerevisiae NAD+ synthetase (NAD+ synthetaseyeast)22. The core of the NAD+ synthetaseGln octamer formed by two stacked central rings of four glutaminase domains is held together by extensive contacts (1,687 A˚2) between glutaminase domains on opposite rings (Fig. 2a). In addition, each glutaminase domain contacts the synthetase domains of two adjacent subunits (2,002 A˚2 buried surface area per glutaminase domain) (Fig. 2b). The core is decorated by four dimers of synthetase domains at the corners of a square. Each dimer buries 1,841 A˚2 of surface area per subunit and is similar to the dimer interface in the NAD+ synthetasesNH3 (ref. 12) (Supplementary Fig. 1 online). The synthetase domains are inclined toward the plane between two rings, and dimers are formed from synthetase domains of subunits from separate rings (Fig. 2a). Thus, the synthetase domains interlock the rings formed by the glutaminase domains. As a result of the complex architecture, one subunit contacts five other subunits in the octamer (Fig. 2). The glutaminase domain and the glutamine tunnel By SSM database search23, the N-terminal glutaminase is most similar to the nitrilase PH0642 from Pyrococcus horikoshii (PDB 1J31)24 (Supplementary Fig. 2 online). The glutaminase domain shares the core structural elements of the nitrilase domain, including a central b-sandwich surrounded by a-helices. The interface between glutaminase domains on separated rings is similar to the dimer interface in PH0642 (Supplementary Fig. 2). In an isolated glutaminase domain

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 A˚2 buried surface area per interface per subunit) and extensive contacts (1,687 A˚2) 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 the four 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 A˚2 buried surface area per domain per subunit).


VOLUME 16

NUMBER 4

a

b

APRIL 2009

423

ARTICLES

a

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]/ {Kiapp (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).

c Tyr127 Glu177 Phe180 Phe130

DON

Glu132 Lys121 Cys176 Tyr58

Glu52

d

b

DON 0 mM 0.1 mM 0.25 mM

+

O +

+

NH3


AH

O

N–

DON

E

O 176

N+

NH3

-N2

NH3

52

O

O

–

O

S H

O-

O 176

S

N+ N

E

E

2 mM 4 mM 5 mM

kinac = 0.21 ± 0.02 sec–1 200 Ki = 0.35 ± 0.06 mM 150 100

+

O –

O

NAD produced (µM)

250 O –

0.5 mM 1 mM 1.5 mM

50 0 0

10

20 30 40 Time (min)

50

60

the glutaminase active site would be exposed to solvent, as in all other GAT enzymes, but in the context of the oligomeric structure of NAD+ synthetaseTB, access to the glutaminase active site is limited. Visual inspection and CAVER analysis25 identified a tunnel of 33 A˚ in length (referred to as the glutamine tunnel) that allows L-glutamine to reach the glutaminase active site (Fig. 3a). The tunnel is formed by residues from two different subunits. For example, the first two-thirds of the glutamine tunnel to the G5 active site (Fig. 1) is lined with residues of the G1 and S1 domains, which are mostly not conserved (Fig. 3a). The remaining one-third of the glutamine tunnel (average radius 2.7 A˚) is lined mostly with conserved residues from G5 (Supplementary Fig. 3a online). Enzymes of the nitrilase superfamily contain a Glu-Lys-Cys catalytic triad with the cysteine as the likely nucleophile for the glutaminase reaction16. In the structure reported here, the glutaminase domain of NAD+ synthetaseTB was irreversibly covalently modified at the cysteine with DON in a stable state that mimicks the g-glutamylthioester intermediate (Figs. 1 and 3b). The glutaminase active site, identified by the covalently linked DON to Cys176TB, is located in a loop-rich region (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 ¼ 0.4 s–1 mM–1; Table 1). Conformational flexibility in the active sites is indicated by multiple conformations and high B-factors when the four Cys176TB-DON adducts are compared (Supplementary Fig. 4 online).

Interactions occur only between the a-carboxylate and a-amino groups of this adduct with the side chains of Glu177TB and Tyr127TB (Fig. 3c and Supplementary Fig. 4). The positions of the other two residues of the catalytic triad suggest that Lys121TB stabilizes the negative charge in the transition state for the formation of the glutamylthioester covalent intermediate26, and Glu52TB abstracts a proton from Cys176 and donates it to the ammonia leaving group (Figs. 1 and 3c). An ordered solvent molecule is found in close proximity to the sulfur of all the Cys176TB-DON adducts and the hydroxyl of Glu52TB. Mutation of either Lys121TB or Glu52TB to alanine abolishes any glutaminase activity27. To determine whether the glutaminase domain acts reciprocally to control the ammonia-dependent NAD+ synthesis, we measured the

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. The black 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.

424

VOLUME 16

NUMBER 4

APRIL 2009


ARTICLES a molecular tunnel between active sites18–20. In all previous structures of multimeric enzymes in which a single polypeptide contains glutaminase and synthase domains, the tunnel lies within one protomer18–20. The glutaminase and synthetase active sites within a single subunit of the NAD+ synthetaseGln are located B55 A˚ from each other (Fig. 2b), B20 A˚ further than the distance between the active sites that are closest to each other, for example, S1 and G5 (Fig. 2b). In addition to the glutamine tunnel described above, CAVER25 identified a second tunnel that connects the closest synthetase and glutaminase active sites, for example, (S1 and G5, Figs. 4 and 5a). The two tunnels join together to form a 73-A˚ U-shaped tunnel (Fig. 4). CAVER did not The intersubunit ammonia tunnel and the synthetase domain identify a tunnel between the G1 and S1 active sites. The intersubunit Glutaminase and synthase active sites in other GAT enzymes are ammonia tunnel extends 40 A˚ from G5Cys176TB in the glutaminase separated by 18–40 A˚18–20. In some cases, ammonia transfers through domain to the carboxyl group of NaAD+ in the synthetase domain and has an average radius of 2.0 A˚, wide enough for an ammonia molecule28. The first third a b from the glutaminase active site of this intersubunit ammonia tunnel is lined with residues from the glutaminase domain (G5), and the remaining two-thirds is lined with residues of the synthetase domain (S1) (Fig. 5a,b and Supplementary Fig. 3b). Ammonia could be transferred through a hydrogenbonding network starting at G5Tyr58 and ending at G5Asp241TB at the G5-S1 interface of the ammonia tunnel (Fig. 5b). As a result of the tunnel’s convoluted U shape, one short loop (residues 127–131, the YRE loop) conserved in mycobacterium and eukaryotes (Supplementary Fig. 3b) contacts the DON molecule (G5Tyr127TB) and three different sections of the tunnel (Fig. 6). Near the glutaminase active site, G5Phe130TB forms with two other hydrophobic residues the first constriction (0.8 A˚) of the ammonia c tunnel (Figs. 5b and 6). G5Arg128TB interacts at the S1-G5 interface in the glutamine tunnel, whereas G5Glu129TB and G5Tyr131TB are located at the S1-G5 interface in the ammonia tunnel (Figs. 5b and 6). Interactions among G5Glu129TB, S1Lys644TB and S1Asp656TB, conserved residues in prokaryotic enzymes, mediate the transition from the G5 to the S1 section of the ammonia tunnel (Figs. 5b and 6). Mutation of S1Asp656TB to alanine (S1D656ATB) inhibits activation of the Figure 5 The ammonia tunnel and the synthetase active site. Subunits 1, 5 and 6 are in magenta, green glutaminase domain in the reaction with the and orange, respectively. (a) Solid surface representation of the G5-S1 ammonia tunnel. Large black synthetase substrates, as evidence by the only 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, 9-fold activation of this variant compared to whereas ordered solvent molecules identified are shown in cyan spheres. The two small black arrows the 179-fold activation measured for the wild indicate the hydrophobic constrictions identified in the ammonia tunnel. (b) A schematic of the intertype (Table 1). The basal glutaminase turnactions with solvent molecules (cyan spheres) identified in the ammonia tunnel. Residues are colored over was essentially the same for the according to the subunit to which they belong. Residues invariant among the prokaryotic glutamineS1D656ATB and wild-type reactions. In addidependent enzymes are indicated by * and those also conserved in the eukaryotic enzymes by **. (c) Fo tion, ammonia transfer from one active site – 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 to the other in the reaction catalyzed by S1D656ATB was significantly compromised, 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 as evidenced by the only B30% maximum NaAD+ bound as well14, indicating a need for further stabilization, perhaps by binding of the ATP channel efficiency of this variant compared molecule and/or conformational changes of the flexible loop. The location of the ATP molecule to the 100% maximum channel efficiency + NH3 + (transparent sticks) was assigned by superposition with the NAD synthetase structure with NaAD measured in the wild-type catalyzed reacand ATP/Mg2+ bound (PDB 1EE1)14. Mg2+ and an ordered solvent molecule are shown as a transparent tion (Table 2). Therefore, disruption of the green sphere and a cyan sphere, respectively.


catalytic properties of the DON-modified enzyme and of the C176A variant. The C176A variant lacked glutamine-dependent activity but retained ammonia-dependent activity (Table 1). However, the DONmodified enzyme showed only 18% (kcat ¼ 0.48 ± 0.01 s–1) of the ammonia-dependent activity of the wild type. This decrease in the activity could be attributed to the inhibition of conformational changes required for the synthetase activity due to the increased rigidity of the DON-modified enzyme and/or to a limited access to the synthetase active site through the glutaminase domain arising from the presence of DON (Fig. 3a).


VOLUME 16

NUMBER 4

APRIL 2009

425

ARTICLES


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+, whereas S1Leu486TB 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.

interactions at the S1-G5 interface and loop YRE yields an enzyme that is markedly wasteful of glutamine and with a substantially compromised ability to transfer ammonia between active sites. A second set of contacts between S6Arg658TB with the side chain of G5Glu61TB and the backbone carbonyl of S1Arg654TB mediates the transition 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 G5S1 ammonia tunnel. The S1 section of the tunnel closer to the glutaminase domain is lined with polar residues and contains ordered solvent molecules. The tunnel narrows to 1.0 A˚ in a second hydrophobic patch, composed of S1Leu486TB, S1Leu639TB, S1Ala506TB and S1Pro640TB and void of solvent molecules (Fig. 5a,b), to resume its hydrophilic character near the synthetase active site. This combination of hydrophobic and hydrophilic surfaces of the tunnel differs from the ammonia tunnels in all other GAT enzymes, which are either uniformly hydrophobic18–20,29 or hydrophilic19,28. At the end of the ammonia tunnel in the synthetase active site the conserved residue S1Asp497TB, close to NaAD+, is positioned to act as a general base for activation of an ammonium ion to ammonia, if an ammonium ion is transferred, or as the last site for the hydrogenbonding interaction if ammonia is transferred (Fig. 5a,b). This role was assigned to the Asp173Bac in NAD+ synthetasesNH3 (ref. 13). The synthetase domain of NAD+ synthetaseTB belongs to the Ntype ATP pyrophosphatase family present in two other GAT enzymes, GMP and asparagine synthetases30,31. However, the similarity is limited to the ATP binding site, and the remaining structural elements among these synthetase domains are substantially different (Supplementary Fig. 5 online). The complete active unit consists of a dimer of two synthetase domains. Whereas the secondary-structural elements in the NAD+ synthetaseNH3, including the dimerization interface, are conserved in the synthetase domain, NAD+ synthetaseGln contains two unique additional helices, a18 and a20, which are required for oligomeric assembly (Supplementary Fig. 6 online). Many of the conserved a-helices, most notably a9, are offset and/or longer in the synthetase domain of NAD+ synthetaseTB compared to that of the ammonia-dependent enzyme. The extended N-terminal helical segment of a9 (residues 333–339S1), which is absent in the NAD+ synthetasesNH3 (Fig. 6 and Supplementary Fig. 1), contacts the glutamine tunnel. The C terminus of a9 is located at the binding site of the S6NaAD+ molecule in the synthetase dimer and contacts the ammonia tunnel through S1Arg354TB and S1Gly350TB, which are conserved in the prokaryotic NAD+ synthetasesGln (Fig. 6 and

426

VOLUME 16

Supplementary Fig. 3). These residues form hydrogen bonds with the side chain of S1Asn641TB, whose backbone atoms line the ammonia tunnel. Notably, S1Asn641TB is located on a loop lining the tunnel that contains two of the residues that contribute to the constriction (constriction 2) in the ammonia tunnel near the synthetase active site (Fig. 6). a9 is a connecting structural element that is likely to contribute to regulation of the reactions at the synthetase and glutaminase active sites. The synthetase domain contains three regions for which we observed no electron density. Region 1 (residues 402–407, equivalent to loop P1 in the ammonia-dependent enzymes10) acts as a lid when ATP is bound in the ATP binding site and is disordered in the absence of ATP. A similar role for loop P1 in the synthetase domain of NAD+ synthetasesGln is likely, given the structural similarity of the ATP binding sites among NAD+ synthetases. Region 2—residues 442–450 on the surface of the synthetase domain—is disordered in half of the synthetase domains in the octamer. Ordering in the other subunits is probably due to crystal packing. Region 3—residues 542–558, corresponding to loop P2 in NAD+ synthetaseNH3—is also always disordered when the ATP binding site is empty. Loop P2 ordering is probably required to protect the NaAD-AMP intermediate from hydrolysis14. The N terminus of a17 is connected to loop P2 in the NAD+ synthetasesNH3 and contacts the NaAD+ biding site by hydrophobic interaction between S1Phe564TB and S1Trp490TB (Fig. 6). At the C terminus of a17, the conserved S1Arg576TB interacts with G5Arg128TB from the YRE loop in the glutamine tunnel (Fig. 6). Therefore, a17 is another structural element that is likely to contribute to the regulation of catalysis in NAD+ synthetasesTB. DISCUSSION NAD+ synthetase is essential for NAD+ biosynthesis and for the survival of both replicating and nonreplicating M. tuberculosis8. Although NAD+ synthetase–independent biosynthetic pathways are present in humans4,7, it is unknown whether these pathways can satisfy NAD+ requirements in the absence of NAD+ synthetase. NAD+ synthetaseTB shares only 23% sequence identity with both the yeast and human orthologs, which themselves, in contrast, are 58.4% identical. Kinetic characterizations of NAD+ synthetaseGln have been reported for only the NAD+ synthetaseyeast (ref. 32). This enzyme shows a modest activation of the glutaminase domain (B15-fold based on the basal glutaminase kcat from Fig. 4 of ref. 32) that is induced by NaAD+ and not ATP32. NAD+ synthetaseyeast reaches a maximum channeling efficiency, an indication of active site coupling, of 60%32. In contrast to NAD+ synthetaseyeast, the activities of the two domains in NAD+ synthetaseTB are more efficiently regulated, leading to negligible waste of L-glutamine. In this study we show that the mechanism of activation of the glutaminase domain in NAD+ synthetaseTB is different

NUMBER 4

APRIL 2009



ARTICLES from that reported for NAD+ synthetaseyeast (ref. 32). The differences in mechanism and sequence between these enzymes encourage the development of specific NAD+ synthetaseTB inhibitors. The structure of NAD+ synthetaseTB reported here shows a fascinating contrast between extensive intersubunit domain interactions and relatively limited intrasubunit domain interactions, resulting in an intersubunit ammonia tunnel that is unprecedented among GAT enzymes. The ammonia tunnel is unlike any other ammonia tunnels reported, with its mixed hydrophilic and hydrophobic character20. Constrictions at the hydrophobic patches prevent the transfer of ammonia in the characterized structure, which contains a halfoccupied synthetase active site. This is consistent with the hypothesis that ammonia transfer occurs only after the formation of the NaAD-AMP intermediate. Notably, Leu486TB and Pro640TB located at constriction 2 and Phe130TB located at constriction 1 are conserved only in prokaryotic organisms, which may indicate a difference in the regulation of ammonia transfer between prokaryotic and eukaryotic NAD+ synthetases. Molecular dynamics studies in GAT enzymes and in ammonia-transport proteins have shown that ammonia, not the ammonium ion, is transferred, driven by differences in the solvation energy of ammonia along the tunnel33,34. Whether an ammonia molecule, rather than an ammonium ion, traverses this ammonia tunnel remains to be seen. An ammonia tunnel was previously identified in a homology model of NAD+ synthetaseyeast generated by threading35. The suggested ammonia tunnel is inconsistent with the structure presented here. Using sequence alignment, we identified the corresponding residues in the three-dimensional structure of NAD+ synthetaseTB to those reported in NAD+ synthetaseyeast (Supplementary Fig. 3). Most of the residues proposed to line the tunnel for the yeast enzyme are scattered in the crystal structure and are located far from active sites and from the ammonia tunnel identified by CAVER (Supplementary Fig. 7 online). The only exceptions are Tyr532yeast (corresponding to Trp490TB and Phe167Bac), which was shown in this structure and in the 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 the synthetase active site. Given the evidence provided here and previously32 that suggests synergy between the two active sites of the NAD+ synthetasesGln, we have identified regions of the protein that may act as communication elements between the two active sites. Two of these elements, the YRE loop and a17, may have a role in the communication between the synthetase active site and the glutaminase active site including constriction 1. The N terminus of a17 is connected to loop P2 in the NAD+ synthetasesNH3, which should become ordered in the presence of the NaAD-AMP intermediate, establishing a possible link between NaAD-AMP formation and ammonia formation and/or access to the tunnel by contacting the YRE loop. Notably, an alanine variant of Tyr601yeast, which corresponds to Leu566TB near the end of a17 (Supplementary Fig. 7), shows poor ammonia-dependent synthetase–specific activity32, probably because of an impaired movement of loop P2 in formation of the intermediate. If a17 indeed acts as a communication wire between conformational changes at loop P2 and ammonia formation, this variant should have a decreased activation of the glutaminase domain in the presence of synthetase substrates. Accordingly, the turnover number for the glutaminase activity in the presence of the saturating synthetase substrate decreased by 40-fold for this variant, whereas the decrease in the basal glutaminase activity was only four-fold32, supporting a compromised


VOLUME 16

glutaminase-activation mechanism. Two other elements, a14 and a9, are positioned to transmit information about the NaAD+ binding state, and a14 may also respond to intermediate formation (Fig. 6). As a result of its close connection to constriction 2, it induces widening of the constriction to allow ammonia to move into the synthetase active site. The kinetic and structural characterizations reported here lay a strong foundation for further studies aimed at elucidating the regulation of the catalytic activities in NAD+ synthetasesGln and for the design of antituberculotic drugs. METHODS Cloning and overexpression of native, D656A, C176A and native selenomethionine (SeMet) NAD+ synthetase from M. 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 ml1) 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

P41212

P41212

178.21, 214.84 0.97930

178.13, 215.18 0.97949

Data collection Space group Cell dimensions a ¼ b, c (A˚) l (A˚) Resolution (A˚) Rsym I/sI Completeness (%) Redundancy Refinement No. reflections

30–3.00 0.156 (0.320)

30–2.35 0.139 (0.559)

9.6 (2.5) 92.9 (66.8)

14.4 (2.5) 99.4 (99.0)

7.9 (4.6)

123,320 (8,868)

11.4 (5.3)

142,674 (14,038)

Rwork / Rfree (%) No. atoms

18.4/24.5 (28.7/34.5) 22,082

Protein Ligand/ion

20,323 226

Solvent Mean B-factor (A˚2)

1,443 8.1 (TLS refined)

Waters

1,443

Residues DON

2,610 4

NaAD+ r.m.s. deviations Bond lengths (A˚) Bond angles (1)

4 0.015 1.65

*Values in parentheses are for the highest-resolution shell.

NUMBER 4

APRIL 2009

427


ARTICLES 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 (NiNTA) 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 concentration 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, University 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 Biosciences). 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+ synthetaseTB 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 conditions 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).

428

VOLUME 16

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.vitalit.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 & Molecular Biology website. ACKNOWLEDGMENTS We thank L.-Y. Gao (University of Maryland) and C.D. Lima (Sloan Kettering Institute) for M. tuberculosis genomic DNA and for the expression plasmids pSMT3 and Ulp1, respectively. We thank K. Rajashankar for assistance with data collection and processing. We thank A. Wlodawer and, especially, J.D. Kahn for critiques on the manuscript. This work is based upon research conducted at the Northeastern Collaborative Access Team beamlines of the Advanced Photon Source, supported by award RR-15301 from the National Center for Research Resources at the US National Institutes of Health. Use of the Advanced Photon Source is supported by the US Department of Energy, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357. This work was supported by a PRF grant from the American Chemical Society to B.G. and by start-up funds to B.G. and N.L.-L. from the Department of Chemistry and Biochemistry, University of Maryland, College Park. AUTHOR CONTRIBUTIONS M.R. cloned, expressed and purified all the proteins and performed all the kinetic experiments; N.L.-L. collected and analyzed crystallographic data and built and refined the structural model of NAD+ synthetase; B.G. designed all the experiments, crystallized native and SeMet NAD+ synthetase and contributed to the 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. Trends Biochem. Sci. 32, 12–19 (2007). 2. Yang, H. et al. Nutrient-sensitive mitochondrial NAD+ levels dictate cell survival. Cell 130, 1095–1107 (2007). 3. Lin, S.J. & Guarente, L. Nicotinamide adenine dinucleotide, a metabolic regulator of transcription, longevity and disease. Curr. Opin. Cell Biol. 15, 241–246 (2003). 4. Tempel, W. et al. Nicotinamide riboside kinase structures reveal new pathways to NAD+. 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 of nicotinamide adenine dinucleotides in bacteria. Vitam. Horm. 61, 103–119 (2001). 7. Bieganowski, P. & Brenner, C. Discoveries of nicotinamide riboside as a nutrient and conserved NRK genes establish a Preiss-Handler independent route to NAD+ in fungi and humans. Cell 117, 495–502 (2004). 8. Boshoff, H.I. et al. Biosynthesis and recycling of nicotinamide cofactors in Mycobacterium 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 interactions with the HIV epidemic. Arch. Intern. Med. 163, 1009–1021 (2003). 10. Jauch, R., Humm, A., Huber, R. & Wahl, M.C. Structures of Escherichia coli NAD synthetase 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-dependent NAD+ synthetase from Bacillus subtilis at 1 A˚ resolution. Acta Crystallogr. D Biol. Crystallogr. 58, 1138–1146 (2002).

NUMBER 4

APRIL 2009



ARTICLES 12. Rizzi, M. et al. Crystal structure of NH3-dependent NAD+ synthetase from Bacillus subtilis. EMBO J. 15, 5125–5134 (1996). 13. Rizzi, M., Bolognesi, M. & Coda, A. A novel deamido-NAD+-binding site revealed by the trapped 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+ synthetase from Bacillus subtilis. Acta Crystallogr. D Biol. Crystallogr. 57, 806–812 (2001). 15. Bellinzoni, M. et al. Heterologous expression, purification, and enzymatic activity of Mycobacterium 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 intermediates in enzyme-catalyzed reactions. Annu. Rev. Biochem. 70, 149–180 (2001). 19. Nakamura, A., Yao, M., Chimnaronk, S., Sakai, N. & Tanaka, I. Ammonia channel couples 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 for growth 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 adenine dinucleotide synthetase. J. Biol. Chem. 247, 4794–4802 (1972). 23. Krissinel, E. & Henrick, K. Secondary-structure matching (SSM), a new tool for fast protein 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 of hypothetical protein PH0642 from Pyrococcus horikoshii at 1.6 A˚ resolution. Proteins 57, 869–873 (2004). 25. Petrek, M. et al. CAVER: a new tool to explore routes from protein clefts, pockets and cavities. BMC Bioinformatics 7, 316 (2006). 26. Hung, C.L. et al. Crystal structure of Helicobacter pylori formamidase AmiF reveals a cysteine-glutamate-lysine catalytic triad. J. Biol. Chem. 282, 12220–12229 (2007). 27. Bellinzoni, M. et al. Glutamine amidotransferase activity of NAD+ synthetase from Mycobacterium tuberculosis depends on an amino-terminal nitrilase domain. Res. Microbiol. 156, 173–177 (2005).


VOLUME 16

28. Endrizzi, J.A., Kim, H., Anderson, P.M. & Baldwin, E.P. Crystal structure of Escherichia coli cytidine triphosphate synthetase, a nucleotide-regulated glutamine amidotransferase/ATP-dependent amidoligase fusion protein and homologue of anticancer and antiparasitic drug targets. Biochemistry 43, 6447–6463 (2004). 29. Mouilleron, S., Badet-Denisot, M.A. & Golinelli-Pimpaneau, B. Glutamine binding opens 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 structure of GMP synthetase reveals a novel catalytic triad and is a structural paradigm for two enzyme families. Nat. Struct. Biol. 3, 74–86 (1996). 31. Larsen, T.M. et al. Three-dimensional structure of Escherichia coli asparagine synthetase B: 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 of ammonia within carbamoyl phosphate synthetase determined by molecular dynamics simulations. Biochemistry 47, 2935–2944 (2008). 34. Lin, Y., Cao, Z. & Mo, Y. Molecular dynamics simulations on the Escherichia coli ammonia 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 contains an essential, obligate intramolecular thiol glutamine amidotransferase domain related to nitrilase. J. Biol. Chem. 278, 33049–33055 (2003). 36. Mossessova, E. & Lima, C.D. Ulp1-SUMO crystal structure and genetic analysis reveal conserved 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 oscillation mode. 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. Acta Crystallogr. D Biol. Crystallogr. 60, 2126–2132 (2004). 40. Murshudov, G.N., Vagin, A.A. & Dodson, E.J. Refinement of macromolecular structures by 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 model anisotropic displacements in macromolecular refinement. Acta Crystallogr. D Biol. Crystallogr. 57, 122–133 (2001).

NUMBER 4

APRIL 2009

429

ARTICLES

Precursor-product discrimination by La protein during tRNA metabolism


Mark 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 binds pre-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 preferentially binds pre-tRNAs over processed tRNAs or 3¢ trailer products through coupled use of two sites: one on the La motif and another on the RRM1 b-surface that binds elsewhere on tRNA. Two sites provide stable pre-tRNA binding, whereas the processed tRNA and 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 are more complex than UUU-3¢OH binding. Accordingly, the RRM1 mutations also impair an RNA chaperone activity of La. The results 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 eukaryotes and remains bound during tRNA processing and modification1,2. In its 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 was expected that this binding would involve the b-sheet surface of La RRM1 (refs. 6–8). Unexpectedly, a human La crystal structure shows that, 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 RRM1 unoccupied 4. This documented a mode of sequence-specific recognition by an RRM that does not involve its b-sheet surface4. Four additional structures of human La with different RNAs confirmed the LM mode of recognition with the RRM1 b-sheet surface unoccupied9. Thus, LM-RRM1 mediates induced-fit UUU-3¢OH binding9 while leaving the RRM1 b- sheet surface unoccupied and its role in RNA binding unclear3,4,9,10. Some La-related proteins (LARPs) that contain a La domain are telomerase subunits11,12. LARP7 (also known as PIP7S) is a tumor suppressor that binds the 7SK small nuclear RNA (snRNA) to regulate the transcription elongation factor P-TEFb13,14. Ciliate LARPs recognize UUU-3¢OH–containing telomerase RNA, and LARP7 recognizes UUU-3¢OH of 7SK snRNA, whereas neither are associated with nascent pre-tRNAs13,15. As the LM-RRM arrangement is conserved in LARPs, they are likely to use RNA binding modes similar to that of genuine La. However, how the LM and RRM1 b-sheet binding surface work together is not known for any protein.

Some activities of La proteins are more complex than UUU-3¢OH binding. Yeast La is required for the maturation of structurally impaired pre-tRNAs16–18, can promote tRNA folding19 and acts redundantly with tRNA modification enzymes, consistent with a role in RNA structural integrity20. Ciliate LARP p65 induces structural rearrangement of telomerase RNA12,21, and La shows RNA chaperone activity18,22,23. Although these studies document complex functions related to RNA structure and folding, the mechanisms that distinguish these activities from simple UUU-3¢OH binding are unknown. To understand this, it will be necessary to know how the LM and RRM interact with RNA dynamically, during simple and complex activities and in specific pathways of RNA metabolism. Modified nucleotides found on La-associated pre-tRNAs indicate that La is bound to pre-tRNAs during modification24,25. La remains associated with pre-tRNA until RNase Z–mediated endonucleolytic cleavage separates the tRNA body from the UUU-3¢OH–containing 3¢ 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 is hours24,30 and, although abundant, La can be limiting for tRNA maturation31. To participate in multiple rounds of tRNA maturation, La must dissociate from the UUU-3¢OH–containing trailer products of processing and recycle onto newly synthesized pre-tRNAs. As cleaved 3¢ trailers contain UUU-3¢OH, the highest-affinity sequencespecific ligand known for La, it was unclear how efficient dissociation would occur. We hypothesized that La might distinguish pre-tRNA from processed products using two binding sites that together provide higher affinity than either alone. After processing, the tRNA and 3¢ trailer

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. ([email protected]).

Received 24 June 2008; accepted 9 February 2009; published online 15 March 2009; doi:10.1038/nsmb.1573

430

VOLUME 16

NUMBER 4

APRIL 2009


ARTICLES

b

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 UUU3¢OH–containing trailer (a) and a tRNAArgACG transcript that lacks UUU3¢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.

500

250

100

tRNA 50

0

500

250

100

50

25

La (nM):

0

UUU-3′OH trailer

25

a RNP

RNP F

F

d

500

250

100

tRNA 50

0

100 250 500

50

25

0

UUU-3′OH trailer

La-QYD: (nM)

25

c

RNP

RNP

F

would readily dissociate from their single binding sites. We show that the RRM1 b-sheet surface of human La forms an RNA binding site, distinct from the UUU-3¢OH binding site, that promotes tRNA maturation in vivo. We also show that this RRM1 binding site contributes to the RNA chaperone activity of La. RESULTS La binds tRNA using non–UUU-3¢OH–mediated contacts To test for a tRNA binding site on La distinct from the UUU-3¢OH binding site, we used the electrophoretic mobility shift assay (EMSA) on a 73-nt tRNAArgACG transcript (which ends with UCG-3¢OH) and a 12-nt 3¢ trailer that ends with UUU-3¢OH, as these represent cleavage products of RNase Z26 (indicated as tRNA and UUU-OH 3¢ trailer in Fig. 1). La showed a several fold–higher affinity for this trailer than for the same trailer ending in AAA-3¢OH, consistent with previous results18 and confirming UUU-3¢OH–specific recognition in our EMSA (not shown). A standard EMSA that contains Mg2+ revealed that La bound avidly to both the 12-nt UUU-3¢OH trailer

b

3

0 Mg

2

2+

3′ trailer (13.6 nM)

Pre-tRNA (2.4 nM)

0 0

tRNA (7.4 nM)

1

5 10 15 20 25 30 35 40 RNA bound (nM)

Pre-tRNA

tRNA (69 nM)

0.5 0

e

Pre-tRNA (76.9 nM)

0

10 20 30 40 50 60 70 RNA bound (nM)

f

3′ trailer 0 –0.2 –0.4 –0.6 –0.8 –1

50

150 250 Time (min)

Differential sensitivity to Mg2+ reflects distinct binding modes To further test for differential binding, we compared Kd values for three relevant RNA species, pre-tRNAArgACG, tRNAArgACG and the free 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 absence of divalent cations in the crystal of La bound to UUU-3¢OH4, and the recognized association of divalent cations with tRNA32. The 85-nt pretRNAArgACG used in Figure 2 lacks a 5¢ leader but contains the 12-nt UUU-3¢OH–containing trailer covalently linked to tRNAArgACG (that is, 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

5 2+ 1 Mg 4.5 4 3′ trailer (24.8 nM) 3.5 3 2.5 2 tRNA (25.2 nM) 1.5 1 Pre-tRNA 0.5 (11.5 nM) 0 0 20 40 60 80 pre-tRNA RNA bound (nM) tRNA tRNA 3′ trailer 0 –0.2 –0.4 –0.6 –0.8 –1 –1.2 –1.4 –1.6 –1.8 –2 50 150 250 Time (min)

g

1.5

1

0 –0.1 –0.2 –0.3 –0.4 –0.5 –0.6 –0.7 –0.8 –0.9

10 Mg

2

1.5 0.5

d

c

3 2.5

3′ trailer (19.8 nM)

ln B/Bo

Bound/free

2.5

2+

ln B/Bo

a

ln B/Bo


F

and the tRNA (Fig. 1a,b). Although the tRNA does not end with UUU-3¢OH, it may still interact with the UUU-3¢OH binding site, which can accommodate, albeit with lower affinity, bases other than uracil4. Therefore, we examined binding to the La–Q20A Y23A D33R mutant (La-QYD in Fig. 1c,d), which is mutated in the LM UUU3¢OH binding site: Gln20 makes uracil-specific contact; Tyr23 stacks with uracil; and Asp33 makes bidentite contacts to the 2¢ OH and 3¢OH of the last nucleotide4,9. La-QYD was debilitated for UUU-3¢OH trailer binding but supported more binding to the tRNA (Fig. 1c,d). This suggested a UUU-3¢OH–independent mode of tRNA binding that is mediated by a site on La other than the RNA 3¢ OH end binding site in the LM4,9.

–1.2

50

150 250 Time (min)

K (nM)

d t1/2 koff 2+ 0 Mg2+ 10 mM Mg 1 mM Mg2+ 0 Mg2+/10 Mg2+ (min–1) (min)

2.4

76.9

11.5

0.031

0.113 6.2

7.2

69.0

25.2

0.105

0.321 2.2

19.8

13.6

24.8

1.45

0.208 3.3

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.


VOLUME 16

NUMBER 4

APRIL 2009

431

ARTICLES La binding to pre-tRNA and tRNA was inhibited by 10 mM Mg2+, whereas binding pre-tRNA mat-tRNA trailer mat-tRNA pre-tRNA trailer Cold competitor: to the UUU-3¢OH trailer was slightly enhanced (Fig. 2g, columns 0 mM Mg2+, [Competitor] (nM): 10 mM Mg2+ and 0 mM Mg2+/10 mM La RNP-La RNP Mg2+). The main conclusion we draw from this analysis is that the binding modes used Free RNAfor tRNA and the UUU-3¢OH ligands are -Free RNA biochemically distinct. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Although differential sensitivity to Mg2+ 32P-trailer 32P-pre-tRNA c d provides biochemical evidence of distinct 0.8 1 binding modes, neither 0 mM Mg2+ nor 0.9 0.7 Trailer competitor 0.8 10 mM Mg2+ represent physiological condi0.6 0.7 mat-tRNA competitor 0.5 tions. As summarized in Figure 2g (column 0.6 0.5 0.4 1 mM Mg), pre-tRNA had the highest pre-tRNA competitor 0.4 0.3 affinity for La, followed by the 3¢ trailer 0.3 0.2 mat-tRNA competitor 0.2 Trailer competitor and tRNA. 0.1 0.1 pre-tRNA competitor A strong difference in binding avidities 0 0 0 1,000 2,000 3,000 4,000 0 1,000 2,000 3,000 4,000 was revealed by competition. In parallel reac[Competitor] (nM) [Competitor] (nM) tions containing 1 mM Mg2+, we examined 32P–pre-tRNA and 32P-trailer for La binding Figure 3 La shows strong preference for pre-tRNA over a UUU-3¢OH trailer. (a,b) Binding reactions 32 in the presence of 100 nM La and increasing contained constant amounts of La protein (100 nM) and a trace amount of P-RNA (B0.1 nM pre-tRNA (a), or B0.1 nM UUU-3¢OH trailer (b), and varying amounts of unlabeled (cold) competitor amounts of unlabeled competitor RNAs RNAs as indicated above the lanes and described in the text, in 1 mM Mg2+. (c,d) The data in a and b (Fig. 3a,b). Unlabeled 3¢ trailer was largely inwere quantified and analyzed in c and d, respectively. effective for competition with 32P–pre-tRNA, even when present at higher concentration represent cleavage products. We used two EMSA methods, each with than La, indicating that La shows a strong preference for pre-tRNA reactions containing 0 mM Mg2+ or 10 mM Mg2+. In method 1, over the UUU-3¢OH–containing trailer (Fig. 3a,c). The same unlabeled varying amounts of La were added to trace amounts of 32P-RNA. This 3¢ trailer readily competed with 32P-trailer (Fig. 3b,d), indicating its showed that binding of the 12-nt trailer was no different in the 0 mM activity as a competitor. As expected, unlabeled pre-tRNA competed Mg2+ and 10 mM Mg2+, whereas tRNA and pre-tRNA binding were effectively with 32P-pre-tRNA and 32P-trailer. The tRNA, which markedly compromised in 10 mM Mg2+ versus 0 mM Mg2+ (Supple- lacks UUU-3¢OH, competed with 32P–pre-tRNA, but was substantially mentary Fig. 1 online, 0 mM Mg2+ and 10 mM Mg2+). By method 1, less competitive with the 32P–UUU-3¢OH trailer, consistent with the Kd is roughly estimated as the La concentration at which 50% of La accommodating both the tRNA and trailer substrates using the RNA is shifted. The Kd for the pre-tRNA and tRNA were each distinct binding sites. We conclude that La has a strong preference estimated at 5–10 nM in 0 mM Mg2+ and B100 nM in 10 mM Mg2+, for pre-tRNA over the 12-nt UUU-3¢OH trailer, consistent with whereas the Kd for the 12-nt trailer was B20 nM, with little if any difference in 0 mM Mg2+ versus 10 mM Mg2+. The large effect of Table 1 Binding properties of mutated La proteins Mg2+ on pre-tRNA and tRNA but not on the 12-nt trailer was confirmed by EMSA method 2, as described below (Fig. 2a–c). (a) Effects on RNA binding of different La mutated proteins For method 2, we focused our analyses on RNA concentrations of La mutant La region tRNA binding UUU-3¢OH trailer around the Kd obtained by method 1. For method 2, we added varying LM Slightly reduced Much reduced amounts of unlabeled RNA to constant amounts of 32P-RNA and La33. Q20A Y23A D33R LM Unaffected ND Bound and free 32P-RNA fractions were quantified and Kd determined R32A K34A K37A RRM1-a0 Unaffected ND by Scatchard analysis. We performed EMSAs in triplicate. Method 2 K105A K109A RRM1-b2 b3 Unaffected Unaffected supports a greater amount of data and is independent of the fraction Y114A F155A of active protein, and the analysis includes quality-assurance para- R143A R144A K148A K151A RRM1–loop-3 3–4 reduced Unaffected RRM1-a3 Unaffected ND meters33. Results for 0 mM Mg2+ and 10 mM Mg2+ are shown in K185A K191A K192A R57A R60A LM NA (other) Unaffected Figure 2a and Figure 2b, respectively and summarized in Figure 2g LM NA (other) Unaffected (see also Supplementary Figs. 2 and 3 online). We obtained a Kd of K16A H19A R143A K151A RRM1–loop-3 Unaffected ND 2+ 13.6 nM for the 12-nt UUU-3¢OH trailer in 10 mM Mg , similar to R144A K148A RRM1–loop-3 Unaffected ND the Kd values of 7.1 nM and 25 nM obtained by others in the presence 2+ Arg 4,34 of Mg for similar ligands . Our Kd of 2.4 nM for pre-tRNA is similar to the Kd of 7.3 nM for pre-tRNAVal in the absence of (b) Kd values of La-loop protein and the relationship of the Kd values for La-loop and Mg2+ (ref. 35). La (expressed as a fraction: Kd La/Kd La-loop) for three RNA ligands We asked whether these Kd values were reflective of different Ligand Kd La/La-loop Kd La-loop dissociation rates (koff )33. The data indicated that La binds more 12.9 nM 0.19 stably to pre-tRNA than to the UUU-3¢OH trailer or tRNA (Fig. 2d–f, pre-tRNA 25.8 nM 0.28 representative EMSAs shown in Supplementary Fig. 4 online). The tRNA 20.5 nM 0.97 t1/2 times for pre-tRNA, tRNA and the UUU-3¢OH trailer were 6.2 UUU-3¢OH trailer NA, not applicable; ND, not determined. min, 2.2 min and 3.3 min, respectively (Fig. 2g). no La 15 63 250 1,000 4,000

32P-trailer

no La 15 63 250 1,000 4,000 no La 15 63 250 1,000 4,000

no La 15 63 250 1,000 4,000

no La 15 63 250 1,000 4,000 no La 15 63 250 1,000 4,000


Fraction 32P-pre-tRNA bound

432

b

32P-pre-tRNA

Fraction 32P-trailer bound

a

VOLUME 16

NUMBER 4

APRIL 2009


ARTICLES

a

2

l

H. sap La M. mus La X. laev La D. mel La C. eleg La S. pombe La S. cerev La H. sap U1A H. sap PABP

b

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.

3

Loop 3

** * * KGQVLNIQMRR-TLHK-----AFKGSIFVVFDSI KGQILNIQMRR-TLHK-----TFKGSIFAVFDSI KGPIENIQMRR-TLQR-----EFKGSIFIIFNTD YDKVVNLTMRK-HYDKPTKSYKFKGSIFLTFETK FGETENVLMRR-LKPG---DRTFKGSVFITYKTR AGPISAVRMRR-DDDK-----KFKGSVFVEFKEP LGEINQVRLRRDHRNK-----KFNGTVLVEFKTI --RNP1-FGQILDILVSRS--------LKMRGQAFVIFKEVS --PILSIRVCRDM-----ITRRSLGYAYVNFQQPA

Lys151

Arg143 Tyr114

Phe155

Lys148


Arg144

the idea that pre-tRNA uses two binding sites, whereas the 12-nt UUU-3¢OH–containing trailer uses only one. As shown below, a La RRM1 loop-3 mutant showed decreased affinity for tRNA, but not the UUU-3¢OH trailer binding in vitro, and decreased tRNA maturation in vivo, providing support for the physiological relevance of the second binding site on RRM1. La RRM1 functions in pre-tRNA recognition We mapped the Mg2+-sensitive UUU-3¢OH–independent binding activity to the La domain (not shown). To further localize this activity, we examined surface charge distribution (Supplementary Fig. 5 online) and mutated candidate basic surface patches to evaluate their effect on RNA binding. We examined several different mutated proteins for tRNA binding, and in some cases other non–UUU-3¢OH– containing RNAs (not shown, indicated as ‘other’ in Table 1a. RRM1 loop-3 (R143A R144A K148A K151A, referred to as ‘La-loop’ mutant below), which connects the RRM1 b2 and b3 strands, reduced binding to pre-tRNA and tRNA four- to five-fold relative to La, with little decrease in binding to the UUU-3¢OH trailer (Table 1b and Supplementary Figs. 6 and 7 online). Moreover, the residual tRNA binding observed for La-loop was insensitive to Mg2+ (not shown). Decreased tRNA binding observed for the La-loop protein is specific, because many other mutated proteins did not decrease tRNA binding (Table 1a). The fact that the La-loop mutated protein showed normal UUU-3¢OH–mediated binding, which requires LM-RRM1 intermotif contacts4,9, indicates that the La domain is not grossly misfolded. We conclude that loop-3 of RRM1 is an important determinant of a UUU-3¢OH–independent RNA binding site on La. The basic residues mutated in La-loop-3 are highly conserved in La proteins but not other RRM proteins such as U1A and PABP (Fig. 4a). Loop-3 forms a basic wall on one side of the RRM b-sheet surface (Fig. 4b). La RRM1 loop-3 functions in tRNA maturation in vivo We next examined the La-loop mutant for tRNA maturation in S. pombe as monitored by tRNA-mediated suppression (TMS) of a premature stop codon in ade6-704, which alleviates the accumulation of red pigment. By this assay, La can replace the function of S. pombe La protein (Sla1p) in vivo17,18. We previously characterized suppressor tRNA alleles that vary in dependence on La, Rrp6 (3¢ exonuclease component of the exosome), and the RNA polymerase III (Pol III) termination subunit Rpc11p (a Pol III–associated 3¢-5¢ exonuclease),


VOLUME 16

for efficient maturation18,31. Mutation of the conserved aromatic residues Tyr114 and Phe155 on the ribonucleoprotein (RNP)-1 and -2 motifs of the RRM1 b-surface compromised maturation of the structurally defective suppressor pre-tRNA in the yeast strain ySK5, but had less effect in ySH9, which contains a wild-type suppressor tRNA18. For the present study, analysis was limited to the wild-type suppressor tRNA in ySH9 (Fig. 5a,b). Consistent with previous data, La was more active for TMS than the negative controls pRep vector and the truncated protein La26–408, which lacks residues involved in UUU-3¢OH recognition17,18,36 (Fig. 5a, sectors 1 and 2). The mutant La–Y114A F155A also served as a control (sector 3). La–loop-3 was less active than La (Fig. 5a, compare sectors 3 & 5), indicating that one or more of the basic residues in loop-3 that mediate UUU-3¢OH– independent pre-tRNA binding contribute to functional tRNA maturation in vivo. We also combined the RRM1 loop-3 and the b-sheet mutations. These mutants, La–Y114A-loop and La–Y114A F155A–loop were increasingly compromised for TMS (Fig. 5a, sectors 6 and 7). As shown below, these RRM1 mutants maintain pre-tRNA UUU-OH 3¢ end–protection activity. The 3¢ end protection function of La protein is standardly monitored by northern blotting analysis of endogenous pre-tRNALys CUU17,18,36–38. An intron probe detects three pre-tRNALysCUU species, the upper and middle bands, which contain the 3¢ trailer and require La for accumulation, and the lower band which accumulates independently of La17,18,36–38. The upper band represents a nascent Pol III transcript, whereas the middle species has had its 5¢ leader removed, and the lower band reflects the 5¢ and 3¢ end matured species (as indicated to the right of Fig. 5b, above). U5 snRNA was probed 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 bands run 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 in La-Y114A, La-F155A and La–Y114A F155A mutants18. The La-loop mutant accumulated less nascent pre-tRNALysCUU (upper band) than did La (Fig. 5b, compare lanes 3 and 5), consistent with a less stable interaction of the pre-tRNA with the La-loop protein relative to La as a result of decreased RRM1-mediated binding. The compound RRM1 mutant La–Y114A F155A–loop, with mutations to the b-sheet surface and loop-3, was most compromised for nascent pre-tRNA accumulation (Fig. 5b, lane 7), consistent with its low TMS activity (Fig. 5a). Note that, for this and other RRM1 mutants, the upper and middle pre-tRNA species appear as distinct bands, a hallmark of 3¢ end protection (Fig. 5b, compare lanes 1 and 2 with lanes 4–7). Thus, mutations in two different regions of the RNA binding surface of RRM1 indicate that RRM1 contributes to tRNA maturation in vivo. Moreover, the RRM1 effect is distinct from pre-tRNA 3¢ end protection. The cumulative results provide evidence that La RRM1 contributes to tRNA maturation by mediating UUU3¢OH–independent binding to pre-tRNA.

NUMBER 4

APRIL 2009

433

ARTICLES

La–Y114A F155A 7

6 5 1 2

4 3 La

pRep

*

11872194118U5 / *

c

1

2

3

4

5

6

7 -U5

d

0 No protein

–0.1 ln U/Uo

M

La-Y114A F155A-loop

08 Y1 14 La A F -lo 1 La op 55A – La Y11 - 4 F1 Y1 A55 14 loo A- A p lo op

b

La-loop

pR ep La 26 La –4

La–Y114A-loop

0. 04 0. 04 1. 0 0. 41 0. 29 0. 29 0. 20

a

–0.2

La-loop

BSA

50

100 Time (s)

1 2 3

–0.3 La –0.4 –0.5

0

La26–408

150


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.

RRM1 contributes to the RNA chaperone activity of La The La protein shows RNA chaperone activity in a cis-splicing assay that uses a self-splicing intron23. This RNA can misfold in vitro and become trapped in an inactive conformation that can be resolved by La and other RNA chaperone proteins such as StpA, NCp7 and Hfq22. We noted previously that La and a mutated derivative, La–Y114A F155A, were comparably active in this assay, consistent with their indistinguishable RNA binding activities18. We compared La and La-loop protein in the cis-splicing assay by monitoring the disappearance of full-length intron-containing RNA and the appearance of spliced product (Fig. 5c, ln U/Uo, y axis). In agreement with previous reports, La activity was reflected by a steep slope in the initial fast phase of the assay23 (Fig. 5c). Whereas the no protein and BSA controls showed relatively little activity as expected, La was most active, and the La-loop protein was intermediate, even though both were present in equal amounts (Fig. 5d). We note that the RNA used in this assay does not end in UUU-3¢OH. A decrease in RNA chaperone activity for La-loop correlates with a decrease in its non– UUU-3¢OH–mediated RNA binding activity, whereas La–Y114A F155A shows no decrease in RNA chaperone or RNA binding activities, as noted previously18. DISCUSSION We report here biochemical and mutational analyses of La RRM1mediated binding to pre-tRNA. The results indicate that the LM and RRM1 b-sheet surface comprises two binding sites that bind UUU3¢OH and non–UUU-3¢OH RNAs, respectively. Concurrent use of both sites, as occurs for nascent pre-tRNA, provides high-affinity stable 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 maturation Consistent with its presence in a human Pol III holoenzyme and localization at Pol III–transcribed genes in yeast and human cells39–41, La is poised to be the first protein to bind newly synthesized pretRNAs. How La would dissociate from UUU-3¢OH after separation of the trailer and recycle onto new pre-tRNAs was unknown. This was an important question because the large amount of 3¢ trailers carrying the sequence-specific ligand UUU-3¢OH that are produced during tRNA processing might consume La if there was no mechanism for dissociation. The ability to withstand challenge by an excess of 12-nt UUU-3¢OH RNA revealed a strong preference of La for pre-tRNA (Fig. 3). This coupled with the relatively fast dissociation of La from

434

VOLUME 16

UUU-3¢OH trailers illustrates a mechanism by which the potential consumption of La by UUU-3¢OH trailers can be avoided. The ability to dissociate after 3¢ trailer cleavage supports directionality of the tRNA pathway (Fig. 6). After dissociation, the end-matured tRNAs become substrates of 3¢ end–modifying proteins such as CCA-adding enzyme and tRNA synthetases, and are bound by export factors Los1/ Xpo-t and exported from the nucleus27. La can then recycle onto a new pre-tRNA. Stable pre-tRNA binding contributes to tRNA structural integrity Although nascent pre-tRNAs probably have minimal structure, data suggest that they acquire structure while associated with La. Although evidence indicates that La promotes correct folding of pre-tRNAArg in vivo19, the misfolded pre-tRNAArg was not converted to the correctly folded pre-tRNAArg by La in vitro, suggesting involvement of other factors in vivo. Genetic evidence points to tRNA-modifying enzymes as such factors20,42–44. As a result of high-affinity, stable binding, pre-tRNAs would have enough time to acquire structurestabilizing modifications, some of which occur while associated with La24,25,45,46 (Fig. 6). For certain functions, it may be beneficial for an RNA binding protein to preferentially associate with RNA precursors and dissociate from products. Our data indicate this to be true for La. We are unaware of any other noncatalytic RNA binding protein for which this has been demonstrated. RRM1 loop-3 is adjacent to and distinct from the LM binding site It was thought that the LM and RRM1 b-sheet surface form a single site that recognizes UUU-3¢OH6–8. With the appearance of multiple cocrystal structures of La bound to UUU-3¢OH RNAs4,9, this idea has given rise to a two-site model3,4,9,10. We provided in vitro and in vivo evidence for RRM1 function in the metabolism of a normal pre-tRNA. We used three RNA substrates, differential Mg2+ sensitivity and mutagenesis, which led to the identification of RRM1 loop-3, adjacent to the b-sheet surface, as having an important role for binding of La to pre-tRNA binding, but not to UUU-3¢OH, in vitro and pre-tRNA maturation in vivo. For other RRM proteins, loop-3 contacts RNA, in some cases via a basic side chain47 (Fig. 4c). Loop-3 of La forms a basic wall on one side of the RRM b-sheet surface that projects in the same plane as the conserved aromatic side chains (Fig. 4d). Involvement of La loop-3 in UUU-3¢OH–independent binding is in agreement with previous results. None of the mutated residues in RRM1 loop-3 was observed

NUMBER 4

APRIL 2009


ARTICLES Pol III transcription

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.

Nascent pre-tRNA PPP

La U OH U U

P

PP

RNase P

* *

We also note that the percentage of La protein that was active for highaffinity pre-tRNA binding was lower than the percentage active for tRNA and 3¢ trailer binding, in all Mg2+ concentrations tested and when using the same batch of La protein for the different RNAs. We believe that this should not be unexpected, because pre-tRNA binding requires both RNA binding sites be active, whereas 3¢ trailer and tRNA binding each require only one site be active. Thus, the chances that either of the two sites required for pre-tRNA binding will be inactive would seem relatively high. We suspect that the orientations of the LM and RRM1 become more fixed relative to each other than when either is bound singly to its isolated RNA. Moreover, the manner in which the LM and RRM1 are fixed relative to each other may vary for different RNAs.

U

*

HO U

U La 3′ trailer UUU OH

+


RNase Z P

* *

* * *

U U HO U

*Nucleotide or base modification

P

* * * * * * *

OH

3′ CCA addition Aminoacylation Nuclear export

End-matured, modified tRNA

to contact any other parts of La or UUU-3¢OH–containing RNA in the structures 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. The notion that loop-3 residues contribute to an RNA binding platform is supported by the compound mutant, La–Y114A F155A–loop, which is more severely defective for TMS and in vivo pre-tRNA accumulation than 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 by RRM1 is mediated by multiple contacts to the b-sheet surface and loop-3, and possibly other residues. Although the LM and RRM1 form independent structures in the absence of RNA6,7, their orientations are fixed by contacts to bound RNA4,9. In a previously reported structure, double-stranded RNA4 raised concern about the exit path of the RNA10. Additional cocrystal structures with all single-stranded RNAs have resolved this concern and added new insight9. In addition to closer LM-RRM1 packing, induced fit and plasticity in the UUU-3¢OH binding cleft, the new structures show an RNA exit path that indeed differs from the doublestranded RNA–bound structure, as can be appreciated by comparing both (Supplementary Fig. 8a,b online). These and other considerations9 favor two disparate sites on La9 that could bind distant regions of an RNA separated by intervening RNA structure4, as in Supplementary Figure 8c. As a separate binding platform, RRM1 contributes to versatility The basic nature of RRM1 loop-3 suggests RNA backbone contacts that could support sequence-independent binding to various RNAs that can derive additional affinity from the UUU-3¢OH binding site. This possibility, and the plasticity of the UUU-3¢OH site9, provide insight into how La can show versatility for different RNAs. We note that the affinity gained by two-site binding may be more complex than a simple addition of the affinity of each site individually.


VOLUME 16

La RRM1 functions in a complex activity La 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 sites connected by intervening RNA may promote RNA folding or other rearrangements by La domain proteins, as has been suggested for other RNA chaperones48,49. Conserved residues specifically on the RRM1 b-sheet of La and Sla1p have been shown to promote the maturation of structurally impaired pre-tRNA18. We suspect that the results reported here will be useful toward understanding complex activities of La in its interactions with mRNA and other RNAs, and to LARPs and their RNAs. Coupled use of distinct binding modes may be an important mechanism for RNP dynamics more generally. METHODS Mutagenesis. 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 derivatives 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 pretRNA 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).

NUMBER 4

APRIL 2009

435


ARTICLES 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/setzlabmu.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 UUU3¢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 the 32P-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

436

VOLUME 16

T4 phage self-splicing intron (gift from R. Schroeder, University of Vienna, using an oligonucleotide containing a T7 promoter), followed by T7 transcription and PAGE purification. Note: Supplementary information is available on the Nature Structural & Molecular Biology website. ACKNOWLEDGMENTS We thank D. Setzer for advice, protocols and RNA binding data analysis tools, M. Nashimoto (Niigata University of Pharmacy and Applied Life Sciences) for the human tRNAArgACG gene and R. Schroeder (University of Vienna) for cis-splicing intron DNA. We thank D. Setzer, D. Engelke and M. Teplova for comments. This work was supported by the Intramural Research Program of the US National Institute of Child Health and Human Development, National Institutes of Health. AUTHOR CONTRIBUTIONS M.A.B. performed all experiments; R.J.M. and M.A.B. designed the study and analyzed 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-OH 3¢-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 RNA polymerase 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 and central 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: a winged 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. Structural analysis reveals conformational plasticity in the recognition of RNA 3¢ ends by the human 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 by apparent translational frameshifting. EMBO J. 19, 6230–6239 (2000). 12. Stone, M.D. et al. Stepwise protein-mediated RNA folding directs assembly of telomerase ribonucleoprotein. Nature 446, 458–461 (2007). 13. He, N. et al. A La-related protein modulates 7SK snRNP integrity to suppress P-TEFbdependent 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 has properties 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¢ endonucleolytic cleavage that matures tRNA precursors. Cell 89, 393–402 (1997). 17. Intine, R.V.A. et al. Transfer RNA maturation is controlled by phosphorylation of the human 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 surfaces on the multifunctional La protein mediate distinguishable activities in tRNA maturation. Nat. Struct. Mol. Biol. 13, 611–618 (2006). 19. Chakshusmathi, G., Kim, S.D., Rubinson, D.A. & Wolin, S.L.A. La protein requirement for 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 functions redundantly with tRNA modification enzymes to ensure tRNA structural stability. RNA 12, 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).

NUMBER 4

APRIL 2009



ARTICLES 24. Hendrick, J.P., Wolin, S.L., Rinke, J., Lerner, M.R. & Steitz, J.A. Ro small cytoplasmic ribonucleoproteins are a subclass of La ribonucleoproteins: further characterization of the 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 nucleolar precursor RNAs. Gene Expr. [review] 10, 41–57 (2002). 26. Nashimoto, M., Nashimoto, C., Tamura, M., Kaspar, R.L. & Ochi, K. The inhibitory effect 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 polymerase III. Nucleic Acids Res. 34, 4826–4836 (2006). 30. Pfeifle, J., Anderer, F.A. & Franke, M. Multiple phosphorylation of human SS-B/La autoantigen 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 RNA polymerase III subunit Rpc11p that decrease RNA 3¢ cleavage activity increase 3¢-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 the individual 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 La protein 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 La protein leads to disordered processing of associated precursor tRNAs. Mol. Cell 9, 1113–1123 (2002).


VOLUME 16

37. Van Horn, D.J., Yoo, C.J., Xue, D., Shi, H. & Wolin, S.L. The La protein in Schizosaccharomyces pombe: a conserved yet dispensable phosphoprotein that functions in tRNA maturation. RNA 3, 1434–1443 (1997). 38. Bayfield, M.A., Kaiser, T.E., Intine, R.V. & Maraia, R.J. Conservation of a masked nuclear 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 initiating RNA polymerase III holoenzyme containing a novel factor that is selectively inactivated 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: regulation and role of La protein. Mol. Cell. Biol. 28, 4576–4587 (2008). 42. Anderson, J. et al. The essential Gcd10p-Gcd14p nuclear complex is required for 1-methyladenosine modification and maturation of initiator methionyl-tRNA. Genes Dev. 12, 3650–3662 (1998). 43. Calvo, O. et al. GCD14p, a repressor of GCN4 translation, cooperates with Gcd10p and Lhp1p 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)methyltransferase in 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 an unmodified tRNA: implications for RNA folding. Biochemistry 37, 16349–16359 (1998). 47. Maris, C., Dominguez, C. & Allain, F.H. The RNA recognition motif, a plastic RNA-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).

NUMBER 4

APRIL 2009

437

ARTICLES

S16 throws a conformational switch during assembly of 30S 5¢ domain


Priya Ramaswamy1,3 & Sarah A Woodson2 Rapid and accurate assembly of new ribosomal subunits is essential for cell growth. Here we show that the ribosomal proteins make 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 the 5¢ domain RNA, a native and a non-native assembly intermediate are equally populated. The secondary assembly protein S16 suppresses the non-native intermediate, smoothing the path to the native complex. In the final step of 5¢ domain assembly, S16 drives a conformational switch at helix 3 that stabilizes pseudoknots in the 30S decoding center. Long-range communication between 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 each minute1,2. Consequently, the process of ribosome assembly must be accurate, so that each subunit is active, and stringently controlled, so that the capacity for protein synthesis matches the rate of growth3,4. Large ribosomal RNAs (rRNAs) form metastable structures that can lead to errors in assembly5. How the ribosomal proteins and external assembly factors remodel these intermediates is important to the fidelity of ribosome assembly. Reconstitution studies on the E. coli 30S ribosomal subunit showed that the ribosomal proteins induce large changes in the structure of the 16S rRNA that underlie the cooperativity and hierarchy of the 30S assembly map6–8 (Fig. 1a,b). Whereas the mechanisms by which the central and 3¢ domains of the 16S rRNA are assembled have been addressed9–12, assembly of the 16S 5¢ domain, which makes up the body of the 30S subunit13,14 (Fig. 1c), is poorly understood. A 16S fragment containing the 5¢ domain forms a stable ribonucleoprotein (RNP) with ribosomal proteins S4, S17, S20 and S16 (ref. 15). Primary assembly proteins S4, S17 and S20 bind the naked rRNA, whereas binding of S16 requires S4 and S20 (ref. 16; Fig. 1b). As the 5¢ domain is the first to be transcribed in vivo and its proteins make interdomain contacts, rapid formation of its stable rRNA and rRNA-protein interactions nucleates 30S assembly17,18. Using hydroxyl radical footprinting, we previously showed that the E. coli 16S 5¢ domain RNA can form all of the backbone interactions predicted by the structure of the 30S subunit in the absence of proteins19. However, interactions between helices 15 and 17 required more than 5 mM MgCl2, and some helices were protected less strongly in the naked RNA than in native 30S ribosomes. Thus,

the 5¢ domain proteins are needed to stabilize the rRNA tertiary structure in physiological Mg2+ concentrations. Moreover, time-resolved footprinting showed that half of the 5¢ domain RNA became kinetically trapped in non-native folding intermediates when refolded in vitro in 20 mM MgCl219. Tertiary interactions between helix 15 and helix 17 required the longest time to form (B1 min), probably owing to misfolding of the central junction between helices 5, 6 and 6a. These results raised the question of whether the proteins also change the pathway of assembly, avoiding unproductive conformations. To determine whether ribosomal proteins redirect the folding pathway of the rRNA, we probed the assembly landscape of the E. coli 16S 5¢ domain RNP using quantitative hydroxyl radical footprinting. This method detects the solvent accessibility of individual residues along the RNA backbone, providing a detailed picture of the RNA tertiary interactions20. Information about the thermodynamic stability of each contact in the presence and absence of the proteins was obtained by probing the complexes over a wide range of Mg2+ concentrations. The results show that binding of S16 to helices 15 and 17 results in a conformational switch at helix 3, 30 A˚ away, which stabilizes tertiary interactions in the 30S decoding site. We also find that S16 increases the cooperativity of RNP assembly by preferentially stabilizing the native configuration of helices in the lower half of the 5¢ domain, while disfavoring non-native assembly intermediates. Together, these results help explain the crucial role of S16 in 30S assembly. They also demonstrate that discrimination against non-native structures is another way in which RNA-protein interactions increase the selectivity of molecular self-assembly.

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 be addressed to S.A.W. ([email protected]). Received 29 September 2008; accepted 9 March 2009; published online 3 April 2009; doi:10.1038/nsmb.1585

438

VOLUME 16

NUMBER 4

APRIL 2009


ARTICLES

a

b

5′ domain

S17 S20

Central 3′ domain

S4

S8

S16

S15

S6

S18 S9

S11

18

17

S10

S5

2 16

3 4

14

S21

S4

5′ 3′

S17

6 S16

7 10

S2

S3

5 6a

11

S14

c

1

13 12

S19 S13

S12

15

S7

8 S20

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.

RESULTS Stability of the naked 16S 5¢ domain RNA To determine how ribosomal proteins stabilize the folded 16S 5¢ domain RNA, we compared the naked rRNA with RNPs containing the primary binding proteins S4, S17 and S20, or S4, S17 and S20 plus protein S16. The structures of the complexes were probed by hydroxyl radical in 330 mM KCl and 0–30 mM MgCl2 (see Methods and Supplementary Figs. 1–3 online). The extent of cleavage was quantified at more than 65 independent segments of the rRNA backbone. In general, the stability of RNA tertiary structure is inversely related to the Mg2+ dependence of the folding transitions21–23. Thus, we expect the RNA a interactions to form at lower Mg2+ concentrations when more proteins join the complex. The extent of cleavage in hydroxyl radical

Stable 5¢ domain RNP To determine whether the 5¢ domain RNA can assemble completely with the four 5¢ domain proteins, we incubated the rRNA with proteins S4, S16, S17 and S20 in 0–30 mM MgCl2 before hydroxyl

b

MgCl2 (mM)

NT G A NT H K K 30S 30S 0.01 0.05 0.1 0.2 0.3 0.5 1 1.5 2 3 7 12 15 20 25 35


9

correlates with the solvent accessibility of each ribose20, which reflects the sum of all folding equilibria that lead to exposure or protection of that residue. Therefore, the Mg2+ required to protect each segment of the RNA reflects the free energy of specific assembly intermediates. In the absence of proteins (Fig. 2, Supplementary Table 1 and Supplementary Fig. 3b online), we found that tertiary interactions in the RNA were heterogeneous and fell into three general categories. Some interactions required little Mg2+ to be stable (pink, Fig. 2b), but they were not protected to the same extent as were control reactions run in parallel on native 30S subunits. Others were not protected, even up to 30 mM Mg2+ (green, Fig. 2b). Still others folded in two distinct phases, suggesting the presence of folding intermediates (black, Fig. 2b). Residues that were protected in o2 mM MgCl2 included nucleotides in helices 17 and 18 adjacent to the ‘upper’ five helix junction that binds with protein S4. A stable core of tertiary structure was also visible around helix 6a and the central junction, which aligns the interface between helix 6/6a and helix 7 (red, Fig. 3a). The lower junction between helices 7–10 folded in 2–13 mM MgCl2 (orange and green, Fig. 3a). Many other regions of the 5¢ domain remained exposed to hydroxyl radical in 20 mM Mg2+ (blue, Fig. 3a), including helices that form the binding sites for the primary assembly proteins S4, S17 and S20. In our previous studies, the naked 5¢ domain RNA was almost completely folded in 20 mM MgCl2 and 120 mM NH4Cl19. The lesser stability of the rRNA tertiary structure reported here reflects the competition between Mg2+ and 330 mM K+ for access to the RNA and the larger size of the K+ ion relative to NH+4 (ref. 24). K+ is often used to reconstitute 30S subunits and more closely mimics the intracellular environment. Thus, under ‘physiological’ conditions, the ribosomal proteins are needed to fully stabilize the rRNA.

1 0.8 0.6

RNA only 250–254 388–390 469

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 protection Y¯ 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.


0.4 427–429

0.2

451–452

0 0.01

469,471 481–483 487–488 495–496 498 499–500 504–505 513 516–518 521–522

c

1 0.8 0.6

0.1

1 [MgCl2] (mM)

10

100

10

100

RNA + 4 proteins 250–254 388–390 469

Y 0.4

525–527 529–531

0.2

533–536

0 0.01

0.1

1 [MgCl2] (mM)

VOLUME 16

NUMBER 4

APRIL 2009

439

ARTICLES

a

0

b

RNA

c

S4–S17–S20

S4–S17–S20–S16

≤20 mM Mg2+

Primary binding proteins We next asked to what extent the three primary assembly proteins without S16 could stabilize the three-dimensional structure of the rRNA. When we incubated the 5¢ domain RNA with a mixture of S4, S17 and S20, most of the expected rRNA tertiary contacts formed in 2.3 mM MgCl2, (Fig. 3b and Supplementary Fig. 3c). Only a few interactions between helices 8 and 6, 10 and 17, at the tip of helix 12 and in helix 18, required more than 2.3 mM MgCl2 to form completely. Thus, S4, S17 and S20 together stabilize nearly all the native tertiary contacts in the 5¢ domain, except those near the helix 18 pseudoknot and a few positions in the core of the domain. The primary binding proteins also perturb the ensemble of initial RNA structures. When the naked 5¢ domain RNA was incubated in 330 mM KCl without Mg2+, most of the RNA backbone was moderately cleaved (Fig. 2a), suggesting that the entire domain is dynamic 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 naked RNA (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 the mature 30S subunit13,14 (Fig. 4c,d). Notably, the residues exposed at low Mg2+ concentration by binding of S4, S17 and S20 overlap the binding site for protein S16 (refs. 25,27; Fig. 4d). As the S4–S17–S20 complexes were titrated with Mg2+, the exposed residues became protected from hydroxyl radical cleavage, consistent with formation of additional tertiary structure (Fig. 4a,b and Supplementary Fig. 4 online). Thus, the primary assembly proteins not only stabilize the overall tertiary structure of the 5¢ domain RNA, they also specifically pre-organize the S16 binding site. Moreover, S4, S17 and S20 in combination narrow the ensemble of assembly intermediates, favoring rRNA conformations that are prepared to make the desired tertiary interactions. These results help explain the cooperative interactions between S16 and S4 or S20 in the 30S assembly map16.

a

c

RNA only GANNHKK

0.01

radical footprinting with Fe(II)-EDTA. As expected, binding of the four 5¢ domain proteins strongly stabilized the tertiary interactions in the 16S 5¢ domain (Fig. 2c and Supplementary Fig. 3a). In 5 mM MgCl2, many tertiary interactions that were undetectable in the RNA alone were formed in the 5¢ domain RNP to the same degree as in the native 30S subunit (Fig. 3). Moreover, the pattern of hydroxyl radical protection and the changes in chemical base modification were consistent with previous footprinting studies of 5¢ domain proteins25–27 (Supplementary Fig. 1 and 2) and with the backbone contacts predicted by crystal structures of the 30S ribosome13,28 (see Methods). Thus, the 5¢ domain RNP assembles completely under physiological conditions.

H17

35 mM 314 H16 328 352 355

H15

H4

H13

H5

365 370 372

392 396 H10

b GANNHKKM

H8

d

S4–S17–S20 30S 0.01


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.

30 mM 314 328

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 are labeled 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.

440

VOLUME 16

H17 352 355 365 370 372

H5

H4 H15

H13 392 396

NUMBER 4

APRIL 2009

H8


ARTICLES

a

1 0.8

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.

Helix 15 +1°

0.6 Y

1° +S16

0.4 0.2

b

RNA

0 Helix 12 0.8

1° +S16

0.6 Y +1°

0.4

RNA

0.2

c

0


0.8

Helix 18 1° +S16

0.6 RNA

Y 0.4

0.2

+1°

0 0.01

0.1

1

10

100

[MgCl2] (mM)

S16 binds the native 5¢ domain core Protein S16 is essential for viability29, and its absence strongly affects the kinetics of 30S reconstitution in vitro30. When bound to the 16S rRNA, S16 induces large changes in the conformation of 5¢ and central domains25,27, consistent with its important role in 30S assembly. To understand the function of S16 in assembly, we deduced the specific effects of S16 on the stability of the 5¢ domain by comparing Mg2+titrations of S4–S17–S20 complexes with titrations of complexes containing four proteins (Fig. 3b,c). In the 30S subunit, S16 straddles a C-loop motif in helix 15 that stacks against bases in a sharp kink in helix 17, and Tyr17 in S16 donates 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 the kink in helix 17 from 4.9 mM to 2.4 mM MgCl2 (Supplementary Table 1). More unexpectedly, we found that S16 also stabilized many backbone contacts with helix 6/6a in the core of the 5¢ domain, including contacts with helix 8 (nucleotides 151–153), helix 13 and S20 (nucleotides 107–108), helix 10 and the tip of helix 17 (nucleotides 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 15 and 17 and indirectly improves helix packing around the 30S spur. S16 changes the structure of helix 12 S16 was previously shown to induce a shift in the secondary structure of helix 12 and to decrease the accessibility of G31 in helix 3 and bases in the central pseudoknot (helix 2)27. We observed that S16 stabilized tertiary interactions between helix 3, 12 and 18, which form part of the interface between the body of the 30S subunit and the platform. These include a contact between the tip of helix 12 (nucleotides 295–297) and the minor groove of helix 3, as well as an interaction between nucleotide 301 and a flexible loop in S17 (Fig. 3c).


VOLUME 16

S16 stabilizes the decoding site pseudoknot In addition to its interactions with helix 15 and the domain core, the Mg2+ titrations showed that protein S16 stabilizes the pseudoknot formed by base pairs 505–507 and 524–526 in helix 18 at physiological Mg2+ concentrations (Fig. 3c and Supplementary Table 1). The helix 18 pseudoknot positions the universally conserved G530 in the 30S decoding site and is essential for protein synthesis31. In the 30S subunit, nucleotides 505–507 are buried by a kink in the RNA backbone created by the pseudoknot and by contact with the N-terminal domain of protein S4. These residues were only partially protected in complexes with S4 alone (P.R. and S.A.W., unpublished data), indicating that interactions elsewhere in the rRNA are important for the stability of the helix 18 pseudoknot. Addition of S16 lowered the [Mg2+]1/2 for protection of nucleotides 505–506 to 2.3 mM (versus 5.6 mM with S4–S17–S20) and increased the maximum protection to 80% of the value obtained for 30S controls. Thus, in our experiments, the helix 18 pseudoknot forms in physiological Mg2+ only when S16 is added to the complex. Reorganization of intermediate complexes Although many 5¢ domain interactions become more stable as proteins join the RNP, segments of the rRNA backbone showed a more complex change in solvent accessibility that revealed the reorganization of intermediate RNPs during the assembly process (Fig. 5). When the 5¢ domain RNA was incubated with the three primary binding 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 in helix 15 fold in two stages. These nucleotides are partially protected in 0 mM Mg2+ and fully protected in 10 mM Mg2+ (Fig. 5a). A second group of residues, such as those in helix 12, become lightly protected in 0.1 mM Mg2+ and fully protected in 10 mM Mg2+ (Fig. 5b). Notably, a third group of residues in helix 18 were partially protected in 0–1 mM Mg2+, exposed to hydroxyl radical cleavage between 1 mM Mg2+ and 5 mM Mg2+ and reprotected in 10–20 mM Mg2+ (Fig. 5c). This oscillation in the solvent accessibility of the RNA backbone is best explained by the remodeling of tertiary interactions during assembly. We observed the same behaviors for other residues in these helices, consistent with a structural change affecting the entire helix (Supplementary Figs. 5 and 6 online). Single-protein complexes show similar multistage folding (P.R. and S.A.W., unpublished data), suggesting that these structural changes are intrinsic to the 5¢ domain RNA. S16 suppresses non-native assembly intermediates Addition of S16 eliminated this multistage folding, causing tertiary interactions with helices 12, 15 and 18 to form cooperatively over a narrow range of Mg2+ concentrations (Fig. 5 and Supplementary Fig. 6). Not only did the final folding transition occur at a lower Mg2+ concentration when S16 was added, but the first folding transition

NUMBER 4

APRIL 2009

441

ARTICLES

a

0.8 Y

H17

H18

H15

0.6

H12

H18 H16

0.4

H3

0.2 0 0.01

c H16

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.

b

1

H4 0.1

1 10 [MgCl2] (mM)

H18

H15

100

H12

H6

H3 H12

H11

H7

H17 H15

H10

H6


H10

moved to a higher Mg2+ concentration or disappeared. Thus, binding of S16 stabilizes certain assembly intermediates while antagonizing others. We observed the conversion of multistage transitions in the absence of S16 to two-state transitions in the presence of S16 for many residues in the 5¢ domain, suggesting that fewer assembly intermediates are populated when S16 binds the rRNA. Conformational switch at helix 3 The structures of candidate assembly intermediates were deduced by clustering residues that followed similar folding transitions when bound by S4–S17–S20. Residues folding in two stages were located on both faces of helix 15, the inward face of helix 17 and the tips of helices 11 and 18 (blue, Fig. 6). The second cluster of residues map to the base of helix 12, the interface between helices 6a and 10, and helix 16 where it contacts helix 18 (orange, Fig. 6). In the 30S subunit, helix 15 lies between helix 17 and helix 6a, whereas the other side of helix 6a packs against helices 8 and 10 (refs. 13,14). Thus, these two clusters represent a set of helix packing interactions in the lower half of the domain. Because these residues are protected in two stages, we deduce that the core of the 5¢ domain adopts at least two folded conformations during assembly: a native-like conformation (IN) that is similar to the 30S structure, and a non-native conformation (InC) that leaves certain residues exposed to hydroxyl radical. In the mature subunit, helices 6, 6a and 12 are arranged co-axially and are linked to each other and to helix 5 through the central junction. Residues in these helices fold together (orange), suggesting that the difference between the native and non-native intermediates may be due to alternative base-pairing of the central junction, which we have observed previously19. Residues that are exposed in moderate Mg2+ concentrations and reprotected in high Mg2+ concentrations cluster in helix 18 and in the tip of helix 12 (nucleotides 295–297) (pink, Fig. 6, and Supplementary 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 mature subunit. Consequently, the transient exposure of helices 12 and 18 to solvent at moderate Mg2+ concentration can be explained by a switch in the relative orientation of helix 3. When we compared folding transitions in different parts of the 5¢ domain (Fig. 6a), it was apparent that helix 18 (pink) becomes exposed over the same Mg2+ concentration range in which helix 15 and the base of helix 12 (blue and orange) become fully protected

442

VOLUME 16

(Fig. 6a). Therefore, we infer that interactions between helices 3 and 18 break when the native-like IN intermediate is populated. The correlation between the orientation of helix 3 and the IN state is corroborated by changes in the solvent accessibility of all the residues in these helices and in the core of the 5¢ domain (Supplementary Fig. 6). Core residues that are protected in a single transition centered at 1–2 mM Mg2+ (nucleotides 151–153, 469, 370–374) probably reflect interactions that exist in the native-like core but not in the non-native core. Other residues (nucleotides 203–204, 301, 481–483, nucleotides 505–506) require 5 mM or 10 mM Mg2+ to become protected in the absence of S16, representing contacts that appear in the native RNP. As discussed below, the self-consistency of the footprinting results allows us to propose a structural model for assembly of the 16S 5¢ domain. DISCUSSION Many RNA binding proteins preferentially bind and stabilize the native three-dimensional structure of their RNA target. In multiprotein RNPs such as the signal recognition particle (SRP) and the 16S central domain, the resulting protein-induced changes in the structure of the RNA favor the addition of subsequent proteins, driving cooperative assembly of the entire complex32–34. Our studies on the 16S 5¢ domain show that S16 not only stabilizes the native conformation of the rRNA, it also destabilizes certain rRNA interactions at early stages of assembly. These results reveal a second important role of RNA binding proteins in RNP assembly, which is to suppress unproductive intermediates. We propose that the suppression of metastable intermediates is an important factor in the cooperativity of RNP assembly. Assembly energy landscape The actions of the 5¢ domain proteins can be understood in terms of an energy landscape for assembly (Fig. 7), as previously proposed based on the kinetics of 30S assembly35. As each protein binds the rRNA, 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 all the RNA tertiary interactions in the 16S 5¢ domain. However, these form at different Mg2+ concentrations, suggesting that assembly passes through at least two intermediates, IN and InC (see below). S16 lowers the free energy of IN but not InC, preventing the accumulation of InC complexes. Consequently, when S16 is present, assembly proceeds directly from IN to N under physiological conditions (2–5 mM Mg2+). Evidence for competing I states comes from the multistage changes in backbone accessibility as the complexes are stabilized by Mg2+. One potential explanation for the plateau in the protection of helix 15 is that assembly passes through an intermediate in which helix 15 is only partially buried (scheme I). However, if ribosomal proteins such as S16 bind the native rRNA more strongly than non-native or unfolded rRNA, then, according to the thermodynamic cycle, they must

NUMBER 4

APRIL 2009


ARTICLES

a

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).

b RNA

+1°

H18

+S16 Icore

∆G

H3 H12

H4 H17 H15

InC IN N

c

H18 H3 out H12 S16 H1,H2 IN

Native core

The four-state model in scheme II can account for all of the multistage changes in backbone accessibility. By contrast, the oscillating protection of helix 18 cannot be explained by the three-state model in scheme I (Supplementary Fig. 7 online). Thus, at least two intermediates are needed to explain assembly of the 5¢ domain. Additional intermediates containing only one or two proteins are also likely. Therefore, the free-energy landscape for assembly is almost certainly more complex than depicted in Figure 7.

S16


H18 H3 in

IC

N

H12

Non-native core

InC

stabilize native RNA interactions more than they do non-native interactions. If the partially protected intermediate in scheme I contains only native RNA interactions, then both the intermediate and the native state should be stabilized when S16 binds, and both transitions should occur at lower Mg2+ concentrations. However, we observe that the initial folding transition is disfavored when S16 is added to the reaction (Fig. 5a and Supplementary Fig. 6). Moreover, the transient exposure of helix 18 during assembly in the absence of S16 strongly argues for a second intermediate with a different conformation. Scheme I: U

IN

Cleaved 50% protected

N Protected

An alternative model that is consistent with all of our data is that assembly involves at least two intermediates with different structures (scheme II): IN in which helix 12 is buried and helix 18 is exposed, and InC in which helix 12 is exposed and helix 18 is buried. We propose that some residues in helix 15 are protected in both IN and InC. If InC forms at low Mg2+ concentrations, this would explain why suppression 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 intermediate dominates the population (Fig. 7a). Scheme II: IN U Cleaved

Cleaved InC

N Protected

Protected

Protein S16 smooths the pathway of assembly considerably by stabilizing IN more than InC, such that only IN is populated (Fig. 7a). This could occur by selective binding to IN. However, S16 could also raise the free energy of InC. In either case, once there is a substantial energy gap between the two intermediate states, only the most stable state will be populated at equilibrium.


VOLUME 16

Mechanism of 5¢ domain assembly From the footprinting results presented here, we propose the following minimal mechanism for 5¢ domain assembly (Fig. 7b,c): at low Mg2+ concentrations, the 5¢ domain RNA forms an ensemble of partly folded states with a minimal set of stable tertiary interactions (IC). This ensemble contains both native-like and non-native configurations of core helices 6, 6a, 7 and 11 that are bound and further stabilized by primary assembly proteins S4, S17 and S20. In IN, we propose that helix 3 is oriented away from the upper five-helix junction, leaving the tip 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, a switch in the topology of the right angle junction between helices 3 and 4 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 is unprotected owing to the perturbed conformation of the core helices and 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 15 and exposure of helix 18 and the tip of helix 12 (Fig. 6a). Helix 18 and 12 are finally reprotected as helix 3 swings back into its native orientation (N). Under equilibrium conditions, we cannot distinguish whether InC proceeds to N directly in the absence of S16, or whether non-native complexes must first transform to IN before reaching the native RNP (dashed arrows, Fig. 7c). In the presence of S16, we propose that most of the complexes assemble along the path from IN to N. The fact that ribosome assembly can proceed by more than one pathway was previously shown by the kinetics of 30S ribosome assembly, as measured by time-resolved footprinting of the RNA backbone37 and the rates of protein addition measured by MS35. The equilibrium and kinetic assembly intermediates of the 5¢ domain are qualitatively similar, even though the time-resolved experiments were carried out on the entire 16S rRNA and total 30S proteins in 20 mM MgCl2. The most stable interactions in the core of the 5¢ domain formed within 50 ms. By contrast, the folding kinetics of the upper junction were slower (2 s) and multiphasic37, consistent with slower conformational changes in this region of the rRNA. One discrepancy is that the tip of helix 12 (nucleotides 295–297) is rapidly protected in the

NUMBER 4

APRIL 2009

443

ARTICLES


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 assembly The crucial role of S16 in the assembly of 30S subunits25,30 can be explained by the conformational switch in helix 3 that comes about by the preferential stabilization of IN. S16 was previously shown to change the secondary structure of helix 12 and stabilize the central pseudoknot27. We hypothesize that tight binding of S16 to helices 15 and 17 is communicated via helix 4 to helix 3, and via helix 17 to the stem of helix 18. We note that the kink in helix 17 and the helix 18 pseudoknot both become protected in 2–5 mM MgCl2, during the transition from IN to the native RNP (Fig. 7). Tertiary interactions between helices 3, 18 and the tip of helix 12 are expected to stabilize the helix 18 pseudoknot. In the 30S subunit, helix 18 is presumably further stabilized by its interactions with protein S12, which is the next protein to join the complex in the 30S assembly map38. By changing the orientation of helix 3, binding of S16 influences the connections between the head, body and platform of the small subunit that are crucial for forming the decoding site and for final maturation of the 30S subunit39,40. First, helix 18 is itself an essential component of the decoding site31,41. Second, in the 30S ribosome, helix 3 stacks with helix 1, which in turn contributes half of the central pseudoknot (helix 2). Reorientation of helix 3 during the transition from IN to the native RNP may help to ensure that helix 3 is not locked in place until helix 12 is correctly aligned with helix 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 of the 16S rRNA before the 3¢ end has been synthesized. When folding is sequential, local structure is favored over long-distance interactions such as helix 3 (ref. 43). Wagner and co-workers have suggested that the rRNA leader serves as a scaffold for assembly of the 5¢ domain44,45, by forming metastable interactions that hold the place of downstream partners until the rest of the 16S rRNA is transcribed. They found that the leader interacts with helix 6 a nd is cross-linked to nucleotides in helices 3, 4, 5, 7 and 11a, precisely those regions of the 5¢ domain that can form alternative structures19. The cold-sensitive mutation C23U in helix 1 inhibits 5¢ processing of the 16S rRNA46 and produces 30S particles resembling reconstitution intermediates (RI) formed in vitro at low temperature6. Thus, an interesting possibility is that the conformational switch we have identified in helix 3 is coupled to processing of the pre-rRNA and later steps of 30S subunit assembly. METHODS Purification 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

444

VOLUME 16

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 equilibrium between an ‘open’ and ‘closed’ state at each nucleotide, we fit the fractional saturation (Y¯) of each backbone protection versus Mg2+ concentration (C) to an isotherm for a cooperative two-state equilibrium, Y ¼ Y0 + ðC=Cm Þn =½1 + ðC=Cm Þn , in which n is the Hill coefficient and Y¯0 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;i Þn;i (refs. 22,54): Y ¼

ðC=Cm;2 Þn;2 + ðC=Cm;N Þn;N 1 + ðC=Cm;1 Þn;1 + ðC=Cm;2 Þn;2 + ðC=Cm;N Þn;N

ð1Þ

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, A1 ðC=Cm;1 Þn;1 A2 ðC=Cm;2 Þn;2 Y ¼ n;1 + 1 + ðC=Cm;1 Þ 1 + ðC=Cm;2 Þn;2

ð2Þ

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

NUMBER 4

APRIL 2009


ARTICLES 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 & Molecular Biology website. ACKNOWLEDGMENTS The authors thank G. Culver (Univ. Rochester) and D. Draper, R. Moss and T. Adilakshmi (Johns Hopkins Univ. (JHU)) for gifts of plasmids and T. Adilakshmi, A. Cukras, J. Brunelle (JHU) and R. Green (JHU and Howard Hughes Medical Institute) for their help and advice. This work was supported by a grant from the US National Institutes of Health (GM60819). AUTHOR CONTRIBUTIONS P.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. Cold Spring Harb. Symp. Quant. Biol. 66, 567–574 (2001). 3. Wilson, D.N. & Nierhaus, K.H. The weird and wonderful world of bacterial ribosome regulation. Crit. Rev. Biochem. Mol. Biol. 42, 187–219 (2007). 4. Kaczanowska, M. & Ryden-Aulin, M. Ribosome biogenesis and the translation process in 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 Escherichia coli 30S ribosomal subunits. Biochemistry 12, 3273–3281 (1973). 8. Stern, S., Powers, T., Changchien, L.M. & Noller, H.F. RNA-protein interactions in 30S ribosomal 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 central domain. Science 288, 107–113 (2000). 11. Recht, M.I. & Williamson, J.R. RNA tertiary structure and cooperative assembly of a large ribonucleoprotein complex. J. Mol. Biol. 344, 395–407 (2004). 12. Grondek, J.F. & Culver, G.M. Assembly of the 30S ribosomal subunit: positioning ribosomal 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 3.3 A˚ resolution. Cell 102, 615–623 (2000). 15. Weitzmann, C.J., Cunningham, P.R., Nurse, K. & Ofengand, J. Chemical evidence for domain 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 S ribosomal 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 coli ribosomes 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 rRNA into 30 S ribosomal subunits. J. Mol. Biol. 232, 362–374 (1993). 19. Adilakshmi, T., Ramaswamy, P. & Woodson, S.A. Protein-independent folding pathway of 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 radical cleavage. Curr. Opin. Chem. Biol. 9, 127–134 (2005). 21. Rook, M.S., Treiber, D.K. & Williamson, J.R. An optimal Mg2+ concentration for kinetic folding 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-splicing RNA: 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 monovalent and divalent ion binding in the folding of the P4–P6 domain of the Tetrahymena ribozyme. Biochemistry 41, 5799–5806 (2002). 24. Heilman-Miller, S.L., Thirumalai, D. & Woodson, S.A. Role of counterion condensation in folding of the Tetrahymena ribozyme. I. Equilibrium stabilization by cations. J. Mol. Biol. 306, 1157–1166 (2001).


VOLUME 16

25. Powers, T. & Noller, H.F. Hydroxyl radical footprinting of ribosomal proteins on 16S rRNA. RNA 1, 194–209 (1995). 26. Stern, S., Wilson, R.C. & Noller, H.F. Localization of the binding site for protein S4 on 16 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. Science 310, 827–834 (2005). 29. Persson, B.C., Bylund, G.O., Berg, D.E. & Wikstrom, P.M. Functional analysis of the ffh-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 required for 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 30S ribosomal 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 and induced 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: structure of 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 in 16S 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 in helix 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 transcription and ribosomal assembly in vivo. Formation of active ribosomal subunits in Escherichia coli requires transcription of rRNA genes by host RNA polymerase which cannot be replaced 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 the Tetrahymena ribozyme. RNA 9, 722–733 (2003). 44. Pardon, B. & Wagner, R. The Escherichia coli ribosomal RNA leader nut region interacts specifically with mature 16S RNA. Nucleic Acids Res. 23, 932–941 (1995). 45. Besancon, W. & Wagner, R. Characterization of transient RNA-RNA interactions important 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 evidence for helical switching in ribosome assembly. Genes Dev. 7, 660–670 (1993). 47. Culver, G.M. & Noller, H.F. Efficient reconstitution of functional Escherichia coli 30S ribosomal subunits from a complete set of recombinant small subunit ribosomal proteins. RNA 5, 832–843 (1999). 48. Baker, A.M. & Draper, D.E. Messenger RNA recognition by fragments of ribosomal protein S4. J. Biol. Chem. 270, 22939–22945 (1995). 49. Moazed, D., Stern, S. & Noller, H.F. Rapid chemical probing of conformation in 16 S ribosomal 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 and enzymatic 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 its application 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 association reactions. Methods Enzymol. 274, 478–492 (1996). 54. Fang, X., Pan, T. & Sosnick, T.R. A thermodynamic framework and cooperativity in the tertiary 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 Biosynthesis. (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 secondary structure of the 16S ribosomal RNA from Escherichia coli. Nucleic Acids Res. 8, 2377–2395 (1980).

NUMBER 4

APRIL 2009

445

B R I E F C O M M U N I C AT I O N S


Teruya Tamaru1, Jun Hirayama2,3, Yasushi Isojima4, Katsuya Nagai4, Shigemi Norioka5, Ken Takamatsu1 & Paolo Sassone-Corsi2 Clock proteins govern circadian physiology and their function is regulated by various mechanisms. Here we demonstrate that Casein kinase (CK)-2a phosphorylates the core circadian regulator BMAL1. Gene silencing of CK2a or mutation of the highly conserved CK2-phosphorylation site in BMAL1, Ser90, result in impaired nuclear BMAL1 accumulation and disruption of clock function. Notably, phosphorylation at Ser90 follows a rhythmic pattern. These findings reveal that CK2 is an essential regulator of the mammalian circadian system. The circadian clock orchestrates intrinsic timing in most organisms and controls a large variety of physiological and metabolic programs1. The molecular core of the circadian clock is constituted by multiple gene products that operate in transcriptional-translational feedback loops1,2. The BMAL1–CLOCK heterodimer is central to the clock mechanism as it drives and maintains circadian oscillations. Several studies indicate that clock proteins are important targets of post-translational modifications3–7, and their rhythmic phosphorylation seems to be a crucial step for regulated function2,8,9. Indeed, BMAL1 phosphorylation is rhythmic in serum shock–synchronized fibroblasts10. Furthermore, drastic circadian phenotypes in the fly are caused by mutations in kinases such as Doubletime (DBT), the fly ortholog of CKIe, Shaggy (SGG), the fly ortholog of Glycogen synthase kinase 3 (GSK3), and CK2. Drosophila melanogaster CK2 specifically regulates the nuclear entry of Period (PER), thereby controlling the fly clock11–13. In mammals, the tau mutant phenotype, characterized by impaired 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.

circadian rhythmicity, is due to a mutation in the gene encoding the clock-regulating kinase CKIe14. Microsequencing of purified p45PFK, a kinase previously implicated in circadian control in mammals15, demonstrates that it corresponds to CK2a (Supplementary Fig. 1 online). Notably, purified p45PFK phosphorylates BMAL1 and CLOCK in vitro15,16 (Supplementary Fig. 2 online). The CK2 holoenzyme is constituted by two copies of the catalytic (CK2a) and regulatory (CK2b) subunits (a2b2). The CK2a monomer also exists as an active form in vivo. To assess the role of CK2b in modulating CK2a-mediated BMAL1 phosphorylation, we co-incubated the subunits in kinase assays. CK2a alone phosphorylated GST-BMAL1, whereas CK2b inhibited BMAL1 phosphorylation in a dose-dependent manner (Fig. 1; see Supplementary Methods online). These results demonstrate that CK2a monomer phosphorylates BMAL1. To establish the physiological role of CK2a-mediated BMAL1 phosphorylation, we used microRNA interference (miR)-mediated silencing in dexamethasone (Dex)-synchronized NIH-3T3 mouse fibroblasts. A miR-CK2a silencing vector specifically reduced CK2a levels to less than the half that of the control (P ¼ 0.002) (Fig. 2a), resulting in markedly impaired circadian Per2 oscillation, as monitored by real-time Per2 promoter–driven luciferase bioluminescence4 (Fig. 2b). In fact, after a first peak with significantly reduced amplitude, Per2 oscillation was almost abolished in miR-CK2a–transfected cells (Fig. 2b). These results were confirmed by using an alternative target sequence (Supplementary Fig. 3 online). Thus, CK2a is essential for mammalian circadian rhythmicity. Notably, silencing of CK2a leads to substantial cytoplasmic BMAL1 retention 24 h after Dex treatment (Fig. 2c), a time when BMAL1 is normally mostly nuclear10. As the function of the CLOCK–BMAL1 complex is intimately dependent on the nuclear translocation of BMAL1 (ref. 17), we speculate that CK2a exerts its control on the mammalian circadian clock by dictating BMAL1 intracellular

a

CBB

b

Autoradiography

CK2αβ CK2αβ CK2α CK2αβ 32 + [γ- P]-ATP (–) (kDa) 116 97 70

1

2

1

2

1

2

1

2 GSTBMAL1

45 CK2α

Incorporated Pi/BMAL1 (pmol/pmol)

CK2a phosphorylates BMAL1 to regulate the mammalian clock

1.25 1 0.75

15′ 60′

0.5 0.25 0 0 0.2 0.5 1 2 CK2β/CK2α (pmol/pmol)

pGST-BMAL1

31

pCK2β

GST

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 Protein Metabolism, 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. ([email protected]).

Received 27 November 2008; accepted 18 February 2009; published online 29 March 2009; doi:10.1038/nsmb.1578

446

VOLUME 16

NUMBER 4

APRIL 2009


B R I E F C O M M U N I C AT I O N S Figure 2 CK2a and BMAL1-Ser90 regulate Control miR-CK2α 1.2 TF (h) 24 24 48 48 nuclear accumulation and clock function. 0 Dex (h) 0 24 24 Control (a–c) CK2a silencing or depletion affects BMAL1 1.0 BMAL1 + + Control miR-CK2α nuclear localization and clock function. NIH-3T3 miR-CK2α + + 0.8 cells were transfected with miR targeting vectors CK2α Merge for mouse CK2a and the validated negative 0.6 DAPI control (control) vector, and were then Dex Actin GFP 0.4 synchronized. (a) Immunoblot analysis for CK2a and actin in cell lysates, with one representative *** 0.2 1.0 80 experiment shown. Normalized average CK2a 60 0 levels from triplicate experiments are shown 0.5 40 0 12 24 36 48 60 72 84 96 20 below with error bars (± s.d.). (b) Circadian Per2 Time post Dex treatment (h) 0 expression was monitored using a real-time bioluminescence assay. Normalized average values from multiple experiments (n ¼ 5) are BMAL1 Merge (DAPI) plotted with error bars (± s.d. of the first and 1.2 Myc-BMAL1-WT second peaks). (c) Localization of BMAL1, 1.0 visualized by immunofluorescence (red) 24 h *** after Dex treatment. Nuclei were stained with 100 0.8 WT DAPI (blue), and miRNAi-transfected, GFP75 0.6 S90A positive cells appear green. Photographs GFP Myc-BMAL1-S90A 50 0.4 are representative of triplicate experiments. 25 Percentages of cells whose BMAL1 is 0.2 predominantly nuclear among the GFP-positive 0 0 cells from triplicate experiments are plotted as 0 12 24 36 48 60 72 means ± s.e.m. (***, P o 0.001). (d–f) Mutation Time post Dex treatment (h) 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).

f

W T S9 0A

Nuclear BMAL1dominant cells (%)

Relative Per2-Luc activity

e

Nuclear BMAL1dominant cells (%)

Normalized CK2α content (miR-CKa/control)

M yc M -GF yc P M -BM yc A -B L1 M -W AL T 1S9 0A

d


c

C m ont iR ro -C l K2 α

b Relative Per2-Luc activity

a

localization. This is reminiscent of the role described for the D. melanogaster CK2 in controlling PER nuclear entry13. We searched the BMAL1 primary sequence for CK2 consensus phosphoacceptor sites (S/T-X-X-D/E). The serine residue at position 90 is highly conserved among all vertebrate BMAL1s (Supplementary Fig. 4b online), and a MALDI-TOF/MS analysis indicated that it is indeed a potential CK2-phosphorylation site (Supplementary Fig. 4a). Notably, Ser90 is not conserved in D. melanogaster CYCLE, the fly counterpart of BMAL1, consistent with the notion that CK2a cannot phosphorylate CYCLE13. To establish the role of BMAL1 Ser90 phosphorylation in circadian function, we performed rescue experiments by expressing wild-type BMAL1 (BMAL1-WT), BMAL1-S90A mutant or GFP in mouse embryonic fibroblasts (MEFs) derived from Bmal1-null mice, which have a dysfunctional circadian clock4,6,7 (Fig. 2d). Per2 expression was monitored in Dex-synchronized MEFs by a real-time reporter assay. BMAL1-WT rescued circadian Per2 expression in Dex-synchronized MEFs (Supplementary Fig. 5 online), whereas the BMAL1-S90A mutant was unable to do so (Fig. 2f), despite its having normal DNA binding activity (data not shown). These results indicate that BMAL1 phosphorylation at Ser90 is essential for circadian gene expression. Ectopically expressed BMAL1-WT accumulated in the nucleus 24 h after Dex treatment (Fig. 2f), paralleling the behavior of native BMAL1 in fibroblasts10. In contrast, BMAL1-S90A remained mostly cytoplasmic (Fig. 2f). Consistent with previous reports indicating a matching localization pattern between BMAL1 and CLOCK17, BMAL1-S90A suppressed nuclear CLOCK accumulation, possibly because of its decreased interaction potential with CLOCK (Supplementary Fig. 6 online). These data are consistent with the change in BMAL1 localization upon CK2a silencing (Fig. 2c) and stress


VOLUME 16

the notion that circadian function requires phosphorylation of BMAL1 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 or Myc-GFP (Supplementary Fig. 7 online). Using the P-BMAL1S90 antibody, we examined whether silencing of CK2a indeed impaired BMAL1 phosphorylation at Ser90. The P-BMAL1-S90 antibody detected a band corresponding to the highly phosphorylated protein (apparent migration pattern corresponding to about 85–90 kDa) in the concentrated BMAL1 immunoprecipitated from the control vector–transfected NIH-3T3 cells at 24 h after Dex treatment (Fig. 3a). The targeting vectors for miR-CK2a substantially decreased the P-BMAL1-S90 signal (Fig. 3a). Notably, knockdown of CK2a strongly affects the total phosphorylation state of BMAL1, suggesting that CK2a may influence or modulate multiple phosphorylation events on BMAL1 by other kinases, including ERKs and CKIe18,19. Next, we examined whether P-BMAL1-S90 levels oscillated in vivo. In Dex-synchronized NIH-3T3 cells, Ser90 phosphorylation showed a rhythmic pattern with peaks at 20–24 h and 44–48h, paralleling PER1 level oscillation (Fig. 3b). At peaking times, the P-BMAL1-S90 signal gradually shifted to a higher molecular weight (B90 kDa), corresponding to the hyperphosphorylated forms. The timing of BMAL1S90 phosphorylation fits with the cyclic nuclear entry of BMAL1 in Dex-synchronized cells (data not shown) and serum shock–synchronized cells10. The levels of CK2 subunits (a¢ is a subtype of CK2 catalytic subunit) are basically constant during the circadian cycle (Fig. 3b); thus, they are unlikely to determine P-BMAL1-S90 oscillation. Finally, on the basis of the BMAL1 phosphorylation pattern

NUMBER 4

APRIL 2009

447

B R I E F C O M M U N I C AT I O N S IP: BMAL1

52

48

40 44

28 32 36

24

20

8

12 16

Time after Dex treatment (h)

2

WB:

4

b

0

a IP: BMAL1

Actin Lysate

48

40 44

0

52

CK2α′

PER1

32 36

CK2α

CK2α

28

PER1

BMAL1

20 24

BMAL1

P-BMAL1-S90

P-BMAL1-S90

1.2 1.0 0.8 0.6 0.4 0.2 0

8 12 16

WB:

P-BMAL1S90

4

miR CK2α

Normalized content

C

Time after Dex treatment (h)

CK2β Actin


Lysate

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.

(Fig. 3a,b), it would seem that CK2a could have a triggering role in multiple phosphorylation events. Our findings demonstrate that CK2a is a BMAL1 kinase that has an essential role in the mammalian clock by regulating BMAL1 nuclear entry. The single Ser90 phosphorylation site is essential for circadian rhythmicity and its mutation causes a phenotype similar to that caused by depletion of CK2a (compare Fig. 2b,c and 2e,f). These findings suggest that CK2a has a pivotal role in the mammalian circadian clock, by controlling BMAL1 intracellular distribution. Although it is likely that CK2 influences circadian physiology at multiple levels because it phosphorylates a large array of cellular proteins20, the remarkable specificity observed here on BMAL1Ser90 provides a lead to further investigations. Finally, as BMAL1 is SUMOylated and acetylated4,7, the regulatory pathway presented here may participate in the interplay between phosphorylation and other post-translational modifications. Note: Supplementary information is available on the Nature Structural & Molecular Biology website. ACKNOWLEDGMENTS We 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 and C. Nishio for reagents, discussions and help. This work was supported by the Human Frontiers in Science Program Organization and the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT; T.T. and K.T.) and

448

VOLUME 16

by the Cancer Research Coordinating Committee of the University of California and 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).

NUMBER 4

APRIL 2009


B R I E F C O M M U N I C AT I O N S

Distinct transcriptional outputs associated with mono- and dimethylated histone H3 arginine 2 Dimethylation of histone H3 Arg2 (H3R2me2) maintains transcriptional silencing by inhibiting Set1 mediated trimethylation of H3K4. Here we demonstrate that Arg2 is also monomethylated (H3R2me1) in yeast but that its functional characteristics are distinct from H3R2me2: (i) H3R2me1 does not inhibit histone H3 Lys4 (H3K4) methylation; (ii) it is present throughout the coding region of genes; and (iii) it correlates with active transcription. Collectively, these results indicate that different H3R2 methylation states have defined roles in gene expression.

2m 3R e1 2m H 3K e2a 4m e3 H

1–

3

H

H

8C

YA L0 1

2W

35 5C

R

YL

34

R

YL

YA L0 35 W

3R

21

Fold enrichment (Log2)

1–

2

3K 1 4 H me 3K 2 4 H me2 3K 4m + R e2 2m e + R 2a 2m e1

3

H

H

pu

t

Covalent post-translational modifications of histones have a fundamental role in chromatin structure and function1. Histone arginine methylation is one such modification that has been linked to transcriptional regulation2. Arginine residues are methylated on the terminal guanidino nitrogens and can exist in three different methylation states: monomethylated (me1), symmetrically dimethylated (me2s) or asymSpp1 binding in cells metrically dimethylated (me2a)3. Studies in a Spp1 binding on peptides b c MYC-Spp1 3 mammalian and yeast cells have demonSet1p activity on peptides H3R2me1 strated that histone arginine methylation H3R2me2a 2 can influence both gene activation and repression4–7. For example, we have recently GST-Spp1PHD 1 shown that asymmetric dimethylation of histone H3 Arg2 (H3R2me2a) in yeast contriAutoradiogram GST butes to transcriptional repression by 0 inhibiting trimethylation of H3K4. SpecifiCoomassie Coomassie cally, H3R2me2a inhibits H3K4me3 by block1 2 3 4 1 2 3 4 ing the PHD domain of the Set1 complex Gene name component Spp1 from binding to methylated H3K4 and, therefore, abrogating H3K4 tri- Figure 1 H3R2me1 does not block activity of the Set1 complex toward H3K4. (a) Pull-down assays methylation by the Set1 methyltransferase8. 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 Similarly, in mammals, H3R2me2a, which is analysis of logarithmically growing yeast cells using antibodies toward Myc-tagged Spp1, H3R2me1 catalyzed by PRMT6, blocks binding of the and H3R2m2a. Error bars represent s.e.m. for duplicate experiments. (c) In vitro methyltransferase WD40 domain of WDR5 to histone H3, thus assays using purified Set1 complex and synthesized peptides. Equal amounts of peptides were inhibitng H3K4 trimethylation mediated by used in the methyltransferase reactions, as shown by Coomassie staining. In


Antonis Kirmizis1, Helena Santos-Rosa1, Christopher J Penkett2, Michael A Singer3, Roland D Green3 & Tony Kouzarides1

MLL1 (refs. 9–11). Overall, these studies described a function for dimethylarginines in transcriptional regulation. However, the role of monomethylarginine still remains unexplored, even though this modification is detected in vivo on mammalian histones6. Using an antibody against the monomethylated form of H3R2, we show that this modification (H3R2me1) occurs on yeast histone H3 (Supplementary Fig. 1a online). The antibody recognizes only Arg2 on histone H3, as mutation of arginine to alanine (H3R2A) or glutamine (H3R2Q) abolishes the signal (Supplementary Fig. 1b). Furthermore, we demonstrate the specificity of this antibody toward the monomethylated version of H3R2 by dot blot and peptidecompetition analyses (Supplementary Fig. 1c,d). To determine whether H3R2me1 functions in a similar manner to H3R2me2a, we first examined the effect of this methyl H3R2 state on the recruitment of Spp1 to H3K4 in vitro and in vivo. In pull-down assays, H3R2me1 did not block the interaction of the Spp1 PHD finger with dimethylated H3K4 peptides (Fig. 1a, lane 4). Consistently, chromatin immunoprecipitation (ChIP) assays showed that in vivo Spp1 occupancy on the 5¢ end of genes can coincide with H3R2me1 enrichment but not with the presence of H3R2me2a (Fig. 1b). We next sought to determine the effect of H3R2me1 on the ability of Set1 to trimethylate H3K4. In agreement with the above results, the purified Set1 complex was able to methylate an H3R2me1 peptide almost as well as an unmodified peptide (Fig. 1c). The Set1 activity was shown to be specific, because an H3K4me3 peptide was not methylated, and the presence of H3R2me2a reduced this activity (Fig. 1c, lanes 3 and 4). These results show that H3R2me1 does not

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. ([email protected]).

Received 18 March 2008; accepted 29 January 2009; published online 8 March 2009; doi:10.1038/nsmb.1569


VOLUME 16

NUMBER 4

APRIL 2009

449

B R I E F C O M M U N I C AT I O N S repression8 (Fig. 2a, right). Additionally, H3R2me2a covers inactive genes entirely (Fig. 2b, left). Most active We next divided all genes into three 1.8 1.8 H3R2me1 H3R2me2a differentially expressed categories (inactive, 1.6 1.6 moderately active and active) to investigate the relationship of H3R2me1 with other 1.4 1.4 histone methyl marks. H3R2me1 overlaps to 1.2 1.2 some extend with the active lysine methyl 1.0 1.0 marks H3K4me3, H3K36me3 and H3K79me3 0.8 0.8 on representative moderately transcribed 0.6 0.6 (YAL023C) and active (YLR390W) genes Average gene (Fig. 2b). Unlike H3R2me2a, the monoAverage gene methylated form of this residue overlaps with H3K4me3 at the 5¢end of moderately Active Inactive Moderately active b 4 4 4 active genes (Fig. 2b, middle). Thus, taken H3R2me1 0 0 0 together, these results indicate that H3R2me1 4 4 4 coincides with all other active modifications H3R2me2a 0 0 0 on yeast genes, and its correlation to tran4 4 4 scription is opposite to that of H3R2me2a. H3K4me3 0 0 0 The above analyses suggested that 4 4 4 H3R2me1 and H3R2me2a might have oppoH3K36me3 0 0 0 site roles in transcription. Therefore, we next 4 4 4 asked whether H3R2me1 and H3R2me2a are H3K79me3 0 0 0 dynamically exchanged on nucleosomes upon YAL023C YLR390W YAL018C induction of gene expression. To test this, we used the sporulation pathway as a model Figure 2 H3R2me1 associates with transcriptional activation. (a) Composite profiles of ChIP-chip system of inducible gene activity. ChIP anaexperiments. (b) ChIP-chip analysis compares the distribution of various histone methylation marks lysis of cells grown in rich media (repressed across three differentially expressed genes. The H3R2me2a and H3K4me3 data have been described8. condition) showed high enrichment of H3R2me2a at sporulation genes (Hop1 and Spr3), as expected, whereas H3R2me1 was not abrogate H3K4 trimethylation (unlike H3R2me2a) and suggest detected at all on the same nucleosomes (Fig. 3a, time 0 h). Most that H3R2 monomethylation has a distinct function from asym- notably, shifting the cells in sporulation media, which induced activation of these genes (Fig. 3a, left), completely reversed the levels metric dimethylation. Having determined that H3R2me1 has functional characteristics of the two modifications at those same nucleosomes. H3R2me1 was distinct from H3R2me2a, we attempted to determine the role of robustly enriched, but there were no traces of H3R2me2a (Fig. 3a, H3R2me1 in transcription. We used a high-resolution, genome-wide time 18 h). These results confirm that the monomethylated state of ChIP-chip (ChIP combined with microarray) analysis to determine H3R2 is associated with transcriptional activation. As methylation at H3R2 was associated with sporulation genes, we the location of H3R2me1 occupancy and its relationship to gene expression (Supplementary Methods online). We found that next sought to determine whether H3R2 is necessary for sporulation H3R2me1 was localized mainly within transcriptional units and was present in 85% of Meiosis l Gene expression H3R2me1 ChlP H3R2me2a ChlP a b yeast genes (Supplementary Fig. 2 online). 0.05 0.05 12 0.04 0.04 To determine the relationship of H3R2me1 to 9 0.03 0.03 gene expression, we first divided 5,065 genes 6 HOP1 0.02 0.02 into five groups according to their transcrip3 0.01 0.01 tional rate, as previously determined12, and H3WT = 11% 0 0 0 then examined the average enrichment of 0 18 0 18 0 18 H3R2A = 1% Time (h) Time (h) Time (h) H3R2me1 for each gene group (Fig. 2a). Meiosis II 12 0.05 0.05 Average gene profiles of H3R2me1 indicated 0.04 0.04 9 that this modification is present evenly 0.03 0.03 throughout the entire coding region of tran6 SPR3 0.02 0.02 scriptionally active genes. The enrichment of 3 0.01 0.01 H3R2me1 correlated with levels of transcrip0 0 0 H3WT = 9% 0 18 0 18 0 18 tional activity, because the most active genes Time (h) Time (h) Time (h) H3R2A = 0% were the most enriched in this modification (Fig. 2a, left). In contrast, analysis of previFigure 3 H3R2 is necessary for sporulation. (a) Gene expression analysis and ChIP experiments during ously reported data of H3R2me2a showed induction of sporulation. Gene expression levels were normalized to a gene, HSD1, whose transcription that this mark is present on the 3¢ end of remains unchanged before and after sporulation. Error bars represent s.e.m. for duplicate experiments. genes and correlates with transcriptional (b) Sporulation assays of H3WT and H3R2A strains. mRNA per hour mRNA per hour mRNA per hour mRNA per hour mRNA per hour

Least active

Fold enrichment

16

450

VOLUME 16

Relative enrichment to H3

Fold induction

Fold induction

Relative enrichment to H3 (Log2)


Fold enrichment

a

NUMBER 4

APRIL 2009



B R I E F C O M M U N I C AT I O N S 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 undergone meiosis I or meiosis II. Mutation of H3R2 to alanine resulted in a severe defect of sporulation, as only 1% of cells managed to undergo meiotic nuclear divisions, as opposed to 20% of H3WT cells (Fig. 3b). This result is consistent with the dynamic regulation of H3R2 methylation on sporulation genes and suggests that H3R2 has an important role in the early stages of the sporulation process in yeast. In summary, this report unveils for the first time a role for a monomethylarginine state in transcription. The results presented here provide evidence that H3R2me1 is a methylation state that occurs in vivo on yeast nucleosomes. The presence of H3R2me1 correlates with transcriptional activity, opposite to the relationship of H3R2me2a with gene expression. Although both H3R2 methylation states are enriched within the coding region of genes, their distribution is also different: monomethylation is enriched throughout the coding region of active genes, whereas dimethylation is enriched throughout inactive genes and toward the 3¢ end of active genes. The distinct distribution patterns of mono- and dimethylation marks are a strong indicator that these two modifications are associated with different functions. The functional difference between these H3R2 methylation states is probably conserved in higher eukaryotes. Recent findings show that PRMT6, the predominant H3R2 methyltransferase, catalyzes preferentially asymmetric dimethylation, implying the existence of a distinct enzyme that carries out monomethylation10,13. The need for two separate enzymes to catalyze these modifications suggests that, possibly, these two methylation states function differently. The identity of an enzyme that would catalyze exclusively H3R2me1 in mammals or yeast remains elusive. Combinatorial deletions of the three putative yeast arginine methyltransferases (Rmt1, Rmt2 and Hsl7) and individual deletions of 35 other yeast methyltransferases do not affect the levels of this modification (data not shown and Supplementary Table 1 online). The precise function of H3R2me1 in the process of transcriptional activation remains to be resolved. There are at least two distinct mechanisms by which H3R2me1 can function in gene expression. In one model, the monomethyled state is a ‘passive’ mark on chromatin that is used to identify actively transcribed regions that need to be


VOLUME 16

silenced. In this model, the monomethylation mark is deposited on active genes to allow subsequent dimethylation and consequent repression of transcription. In a second model, monomethylation dictates a function (such as the recruitment of a protein) that is necessary for genes to become or remain active. Such a model would be analogous to the recruitment of chromatin effectors by specific lysine methylation states14,15. Future studies will aim to decipher the molecular mechanism used by H3R2me1 during gene expression. Accession codes. Gene Expression Omnibus: Microarray data sets have been deposited under accession number GSE14453. Note: Supplementary information is available on the Nature Structural & Molecular Biology website. ACKNOWLEDGMENTS We thank members of the T.K. laboratory for helpful discussions and M. Gilchrist for help with depositing genomic data. This work was supported by postdoctoral fellowship grants to A.K. from the European Molecular Biology Organization (EMBO) and Marie Curie. The T.K. laboratory is funded by grants from Cancer Research UK (CRUK) and the 6th Research Framework Program of the European Union (Epitron and Heroic). COMPETING INTERESTS STATEMENT The authors declare competing financial interests: details accompany the full-text HTML 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. 2. 3. 4.

Kouzarides, T. Cell 128, 693–705 (2007). Wysocka, J., Allis, C.D. & Coonrod, S. Front. Biosci. 11, 344–355 (2006). Turner, B.M. Nat. Struct. Mol. Biol. 12, 110–112 (2005). 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).

NUMBER 4

APRIL 2009

451

Nature Structural Molecular Biology April

Molecular Modelling In Structural Biology

Nature Reviews Molecular Cell Biology

Molecular Biology

Molecular biology

Molecular Biology

Molecular Biology

Molecular Biology

Molecular Biology

Foundations of Structural Biology

Advances in Structural Biology, Volume 6 (Advances in Structural Biology)

Structural Genomics and High Throughput Structural Biology

Structural Proteomics: High-Throughput Methods (Methods in Molecular Biology)

Genomic Structural Variants (Methods in Molecular Biology, v838)

Analytical Molecular Biology

Genetics and Molecular Biology

Encyclopedia of Molecular Biology

Molecular Biology Problem Solver

Receptor Molecular Biology

Essentials of Molecular Biology

Encyclopedia of Molecular Biology

Topology in molecular biology

Molecular Biology and Biotechnology

Enzymes of Molecular Biology

Molecular Biology and Biotechnology

Analytical Molecular Biology

Molecular Biology Problem Solver

Computational molecular biology

Fundamental Molecular Biology

Molecular Cell Biology

Molecular Biology of Spirochetes

Nature Structural Molecular Biology April

Molecular Modelling In Structural Biology

Nature Reviews Molecular Cell Biology

Molecular Biology

Molecular biology

Molecular Biology

Molecular Biology

Molecular Biology

Molecular Biology

Foundations of Structural Biology

Advances in Structural Biology, Volume 6 (Advances in Structural Biology)

Structural Genomics and High Throughput Structural Biology

Structural Proteomics: High-Throughput Methods (Methods in Molecular Biology)

Genomic Structural Variants (Methods in Molecular Biology, v838)

Analytical Molecular Biology

Genetics and Molecular Biology

Encyclopedia of Molecular Biology

Molecular Biology Problem Solver

Receptor Molecular Biology

Essentials of Molecular Biology

Encyclopedia of Molecular Biology

Topology in molecular biology

Molecular Biology and Biotechnology

Enzymes of Molecular Biology

Molecular Biology and Biotechnology

Analytical Molecular Biology

Molecular Biology Problem Solver

Computational molecular biology

Fundamental Molecular Biology

Molecular Cell Biology

Molecular Biology of Spirochetes

Recommend Documents