Cell Journal, Volume 144, Number 1, January 7, 2011

Volume 144 Number 1 January 7, 2011

www.cell.com

Presenilin Carves Neural Circuits Chromatin and Alternative Splicing

Your bacteria's favorite! Eppendorf Eporator ® The Eppendorf Eporator: the solution for efficiently transforming bacteria and yeast! The Eppendorf Eporator is the perfect fit for you and your lab. Ask the specialists—your bacteria and yeast—they will confirm that the product features and performance of the Eppendorf Eporator make it very easy for you to carry out your experiments and obtain excellent results.

Unique features of the Eppendorf Eporator Easy one-button start l Intuitive use guided by color display l Fast and freely programmable function keys l Integrated electroporation chamber for safety l Compact, space-saving design l USB port for data export l

For more information visit www.eppendorfna.com/eporator

www.eppendorf.com Þ Email: [email protected] In the U.S.: Eppendorf North America, Inc. 800-645-3050 Þ In Canada: Eppendorf Canada Ltd. 800-263-8715

015.A1.0103.A © 2011 Eppendorf AG

NEW!

. W E N

Intelligent design

inCuSaFe™ copper enriched stainless steel interior

+

+

Single-beam, dual capture infrared CO2 sensor

+

SafeCell UV protection in situ

+

1

Hydrogen peroxide vapor sterilization in situ

Good laboratory technique

Sterisonic™ GxP Performance and Productivity Delivers Best Efficiency Value: 2 Hours SANYO Sterisonic™

14 Hours Brand X

H2O2 sterilization vs. high heat sterilization = Uptime (Hours) = Downtime (Hours)

7KHUDSLGin situ+2VHTXHQFHUHWXUQVWKH IXOO\VWHULOL]HG6WHULVRQLFŒ*[3WRQRUPDOXVHTXLFNHU WKDQDQ\FRPSHWLWLYHLQFXEDWRULQWKHZRUOG

The industry’s first in situ H2O2 sterilization with the fastest turnaround. For maximum productivity in clinical, general purpose or the most highly compliant GMP applications, the new SANYO Sterisonic™ GxP CO2 incubator offers an impressive return on investment. With multiple contamination control safeguards, exclusive on-board H2O2 sterilization, FDA-21CFR data capture and graphical LCD display, the Sterisonic™ GxP rewards good laboratory technique with performance ww.steriso onic.com or call 800-858-8442. you can trust. Learn more, visit ww pictured above: Sterisonic™ GxP MCO-19AICUVH with rapid H2O2 vapor sterilization system.

FREEER OFF [

D E TA IL

!

E S O N L IN

]

FREE! FREE!

H2O2 sterilization system acessory, plus BD labware consumables kit. Limited time upgrade offer with purchase of the Sterisonic™ GxP MCO 19AICUVH. ($900 Value) BD stem cell starter kit with Sterisonic™ GxP quote. No purchase necessary! Act now. Supplies are limited. ($150 Value)

www.sterisonic.com ©2009 Sanyo Biomedical OWS 1015 05/09

Supplying Reagent, Pre-clinical and cGMP Grade Proteins to the Biopharmaceutical and Vaccine Development Community

Recombinant Vaccine Components !   ! P. aeruginosa rEPA ! 

 ! E. coli  • C. difficile 
 COMING SOON!

• Circumsporozoite Protein Reagent Proteins is now your single source for reagent, pre-clinical and cGMP grade proteins. See our new website for a complete listing of products and special discounts of up to 30% on recombinant human proteins.

Win an iPad ! Make a purchase by Feb 28, 2011 for a chance to win an iPad!

Editor Emilie Marcus Senior Deputy Editor Elena Porro Deputy Editors Robert Kruger Connie M. Lee Scientific Editors Karen Carniol Kara Cerveny Michaeleen Doucleff Fabiola Rivas Niki Scaplehorn Lara Szewczak Senior Managing Editor Meredith Adinolfi Deputy Managing Editor Andy Smith Lead Illustrator Andrew A. Tang Illustrators Yvonne Blanco Kate Mahan Production Staff Reyna Clancy Editorial Assistant Mary Beth O’Leary

Editorial Board Abul Abbas C. David Allis Genevie`ve Almouzni Uri Alon Angelika Amon Johan Auwerx Richard Axel Cori Bargmann Konrad Basler Bonnie Bassler David Baulcombe Jeffrey Benovic Carolyn Bertozzi Wendy Bickmore Elizabeth Blackburn Joan Brugge Lewis Cantley Joanne Chory David Clapham Andrew Clark Hans Clevers Stephen Cohen Pascale Cossart Jeff Dangl Ted Dawson Pier Paolo di Fiore Marileen Dogterom Julian Downward Bruce Edgar Steve Elledge Anne Ephrussi Ronald Evans Witold Filipowicz Marco Foiani Elaine Fuchs Yukiko Goda Stephen Goff Joe Goldstein

Douglas Green Leonard Guarente Taekjip Ha Daniel Haber Ulrike Heberlein Nobutaka Hirokawa Mark Hochstrasser Arthur Horwich Tony Hunter James Hurley Richard Hynes Thomas Jessell Narry Kim Mary-Claire King David Kingsley Frank Kirchhoff Richard Kolodner John Kuriyan Robert Lamb Mark Lemmon Beth Levine Wendell Lim Jennifer Lippincott-Schwartz Dan Littman Richard Losick Scott Lowe Tom Maniatis Matthias Mann Kelsey Martin Joan Massague´ Iain Mattaj Satyajit Mayor Ruslan Medzhitov Craig Mello Tom Misteli Tim Mitchison Alex Mogilner Paul Nurse Roy Parker

Dana Pe’er Kathrin Plath Carol Prives Klaus Rajewsky Venki Ramakrishnan Rama Ranganathan Anne Ridley James Roberts Alexander Rudensky Helen Saibil Joshua Sanes Randy Schekman Ueli Schibler Joseph Schlessinger Hans Scho¨ler Trina Schroer Geraldine Seydoux Kevan Shokat Pamela Sklar Nahum Sonenberg James Spudich Paul Sternberg Bruce Stillman Azim Surani Keiji Tanaka Craig Thompson Robert Tjian Ju¨rg Tschopp Ulrich von Andrian Gerhard Wagner Detlef Weigel Alan Weiner Jonathan Weissman Matthew Welch Tian Xu Shinya Yamanaka Marino Zerial Xiaowei Zhuang Huda Zoghbi

Cell Office Cell, Cell Press, 600 Technology Square, 5th Floor, Cambridge, Massachusetts 02139 Phone: (+1) 617 661 7057, Fax: (+1) 617 661 7061, E-mail: [email protected] Online Publication: http://www.cell.com Cell (ISSN 0092-8674) is published biweekly by Cell Press, 600 Technology Square, 5th Floor, Cambridge, Massachusetts 02139. The institutional subscription rate for 2011 is $1,605 (US and Canada) or $1,847 (elsewhere). The individual subscription rate is $320 (US and Canada) or $363 (elsewhere). The individual copy price is $50. Periodicals postage paid at Boston, Massachusetts and additional mailing offices. Postmaster: send address changes to Elsevier Customer Service Americas, Cell Press Journals, 11830 Westline Industrial Drive, St. Louis, MO 63146, USA. The paper used in this publication meets the requirments of ANSI/NISO Z39.48-1992 (Permanence of Paper). Printed by Dartmouth Printing Company, Hanover, NH.

¬,OOK¬!GAIN

$ISCOVER¬-ORE s¬!CCESS¬TO¬THE¬¬#ELL¬0RESS¬PRIMARY¬RESEARCH¬JOURNALS¬AND¬ ¬4RENDS¬REVIEWS¬TITLES ¬ALL¬ON¬THE¬SAME¬PLATFORM s¬)MPROVED ¬MORE¬ROBUST¬ARTICLE¬AND¬AUTHOR¬SEARCH s¬6IDEO ¬ANIMATIONS¬AND¬SOUND¬lLES s¬%ASY¬TO¬NAVIGATE¬HOME¬PAGE ¬ARTICLES¬PAGES¬AND¬ARCHIVE

WWWCELLCOM

Cell Press President & CEO Lynne Herndon Editor in Chief, Vice President of Content Development Emilie Marcus Vice President of Business Development Joanne Tracy Vice President of Web Development and Operations Keith Wollman Senior Product Manager Mark Van Hussen Director of Marketing Jonathan Atkinson Production Manager Meredith Adinolfi

Display Advertising Northeast/Mid-Atlantic: Victoria Macomber, ph: 508 928 1255; fax: 508 928 1256; e-mail: [email protected] Midwest/Southeast/Eastern Canada: Inez Herrero-Redman, ph: 585 678 4395; fax: 585 678 4722; e-mail: i.herrero@elsevier. com Northwest/Southwest/Western Canada: Lori Young, ph: 646 370 6312; fax: 212 462 1915; e-mail: [email protected] California: Elizabeth Loennborn, ph: 714 655 1877; fax: 214 452 9627; e-mail: [email protected] UK/Europe: James Kenney, ph: +44 20 7424 4216; fax: +44 18 6585 3136; e-mail: [email protected] Asia: Wendy Xie, ph: +86 10 8520 8827; e-mail: w.xie@ elsevier.com Classified Advertising United States and Canada: Gordon Sheffield, Key Account Manager, ph: 617 386 2189; fax: 617 397 2805; e-mail: g.sheffi[email protected]

Press Officer Cathleen Genova

UK, Europe, and Asia: Sabrina Dodge, Key Account Manager, ph: +44 20 7424 4997; fax: +44 18 6585 3136; e-mail: [email protected]

ª2011 Elsevier Inc. All rights reserved. This journal and the individual contributions contained in it are protected under copyright by Elsevier Inc., and the following terms and conditions apply to their use:

advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made. Although all advertising material is expected to conform to ethical (medical) standards, inclusion in this publication does not constitute a guarantee or endorsement of the quality or value of such product or of the claims made of it by its manufacturer.

Photocopying: Single photocopies of single articles may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee are required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for nonprofit educational classroom use. For information on how to seek permission, visit www.elsevier. com/permissions or call (+44) 1865 843830 (UK) / (+1) 215 239 3804 (US). Permissions: For information on how to seek permission, visit www.elsevier.com/ permissions or call (+44) 1865 843830 (UK) / (+1) 215 239 3804 (US). Derivative Works: Subscribers may reproduce tables of contents or prepare lists of articles including summaries for internal circulation within their institutions. Permission of the Publisher is required for resale or distribution outside the institution. Permission of the Publisher is required for all other derivative works, including compilations and translations (please consult www.elsevier.com/permissions). Electronic Storage or Usage: Permission of the Publisher is required to store or use electronically any material contained in this journal, including any article or part of an article (please consult www.elsevier.com/permissions). Except as outlined above, no part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without prior written permission of the Publisher. Notice: No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence, or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein. Because of rapid

Reprints: Article reprints are available through Cell’s reprint service; for information, contact Nicholas Pavlow (e-mail: [email protected]; ph: (+1) 212 633 3960). Subscription Orders and Inquiries: Mail, fax, or e-mail address changes to Elsevier Customer Service Americas, allowing 4–6 weeks for processing. Lost or damaged issues will be replaced, subject to availability, if Cell Press is notified within the claim period (US and airmail delivery: 3 months from issue date; surface delivery: 4 months from issue date). Periodical delivery in the US can take up to 3 weeks. Airmail delivery can take 2–4 weeks. The price of a single copy of Cell is $50 (excluding special issues). All orders must be prepaid and in writing. Please include the volume and issue number, payment (check or credit card, MasterCard, Visa, or American Express only), and a delivery address. Allow 4–6 weeks for delivery. Mailing address: Elsevier Customer Service Americas, Cell Press Journals, 11830 Westline Industrial Drive, St. Louis, MO 63146, USA. Toll-free phone within USA/Canada: 866 314 2355; phone for outside US/Canada: (+1) 314 453 7038; fax: (+1) 314 523 5170; e-mail: [email protected]; internet: www.cellpress.com or <www.cell.com>. Funding Body Agreements and Policies: Elsevier has established agreements and developed policies to allow authors whose articles appear in journals published by Elsevier to comply with potential manuscript archiving requirements as specified as conditions of their grant awards. To learn more about existing agreements and policies, visit http://www.cell.com/cellpress/FundingBodyAgreements. Guide for Authors: For a full and complete guide for authors, please go to www.cell.com/ authors.

MASTERING CHANGE

Breakthrough in 5-hmC quantitation for epigenetics Interested in simplifying the study of DNA methylation, particularly 5-hydroxymethylcytosine? Try the EpiMark™ 5-hmC and 5-mC Analysis Kit, a robust enzymatic method for the locus-specific detection of methylated (5-mC) and hydroxymethylated (5-hmC) cytosine. As the first commercially available PCR-based kit to reproducibly identify and quantitate the presence of 5-hmC, this simple 3-step protocol will expand your potential for epigenetics research and biomarker discovery. Identify and quantitate methylation states with the EpiMark™ 5-hmC and 5-mC Analysis Kit

Advantages: s Reproducible quantitation of 5-hmC and 5-mC

% 5-hmC

s Easy-to-use protocols

% 5-mC % Unmethylated C

s Compatible with existing techniques (PCR)

100

s Amenable to high throughput

80 60

Visit neb.com/epigenetics to learn more, and explore the complete listing of EpiMark validated products from NEB.

40 20 0 Brain

Liver

Heart

Spleen

Analysis of the different methylation states in Balb/C mouse tissue samples shows a variation in the amount of 5-hmC present at locus 12.

CLONING & MAPPING

DNA AMPLIFICATION & PCR

RNA ANALYSIS

PROTEIN EXPRESSION & ANALYSIS

GENE EXPRESSION & CELLULAR ANALYSIS

www.neb.com

Leading Edge Cell Volume 144 Number 1, January 7, 2011 IN THIS ISSUE CELL CULTURE 5

New Year’s Diets

PREVIEWS 9

Cancer Genomes Evolve by Pulverizing Single Chromosomes

M. Meyerson and D. Pellman

11

Mitochondrial Stress Signals Revise an Old Aging Theory

D.K. Woo and G.S. Shadel

13

Fishing for Biomarkers with Antigen Mimics

T.M. Lindstrom and W.H. Robinson

REVIEW 16

Epigenetics in Alternative Pre-mRNA Splicing

R.F. Luco, M. Allo, I.E. Schor, A.R. Kornblihtt, and T. Misteli

SNAPSHOT 158

Chromatin Remodeling: INO80/SWR1

Y. Bao and X. Shen

Antibodies and Related Reagents for Signal Transduction Research

XP

™

Announcing

Monoclonal Antibodies,

eXceptional Monoclonal Technology™

exclusively from Cell Signaling Technology ®

development and production of antibodies with eXceptional Performance™ (XP™). This revolutionary and proprietary technology provides us with access to a broad range of antibody-producing B cells unattainable with traditional monoclonal technologies, allowing more comprehensive screening and the identification of XP monoclonal antibodies.

eXceptional specificity™ + eXceptional sensitivity™ + eXceptional stability and reproducibility™ = eXceptional Performance™ XMT coupled with our extensive antibody validation and stringent quality control delivers XP monoclonal antibodies with eXceptional Performance in the widest range of applications. untreated

EGF-treated (100 ng/ml, 15 min.) kDa

A43 1 Bx PC 3 He La

XMT™, an eXceptional Monoclonal Technology™ from Cell Signaling Technology®, enables the

Events

200

EGFR Receptor

140 100 80 60

EGF Receptor

human lung carcinoma

Jurkat cells (red) and Kyse70 cells (blue)

A-549 cells

EGF Receptor (D38B1) XP™ Rabbit mAb #4267 demonstrates eXceptional Performance in a wide range of applications. See www.cellsignal.com for full validation details.

for more information on XMT™ visit...

www.cellsignal.com China www.cst-c.com.cn

| Europe www.cellsignal.eu | Japan www.cstj.co.jp

Tel: 978-867-2300

| E-mail: [email protected]

© 2010 Cell Signaling Technology, Inc. XMT™, XP™ , eXceptional Monoclonal Technology™, eXceptional specificity™, eXceptional sensitivity™, eXceptional stability and reproducibility™, eXceptional Performance™ and Cell Signaling Technology® are trademarks of Cell Signaling Technology, Inc. / DRAQ5® is a registered trademark of Biostatus Limited

Above: Confocal IF analysis of rat colon using Phospho-β-Catenin (Ser675) (D2F1) XP™ Rabbit mAb #4176. Red = DY-554 phalloidin. Blue = DRAQ5® #4084.

Articles Cell Volume 144 Number 1, January 7, 2011 27

Massive Genomic Rearrangement Acquired in a Single Catastrophic Event during Cancer Development

P.J. Stephens, C.D. Greenman, B. Fu, F. Yang, G.R. Bignell, L.J. Mudie, E.D. Pleasance, K.W. Lau, D. Beare, L.A. Stebbings, S. McLaren, M.-L. Lin, D.J. McBride, I. Varela, S. Nik-Zainal, C. Leroy, M. Jia, A. Menzies, A.P. Butler, J.W. Teague, M.A. Quail, J. Burton, H. Swerdlow, N.P. Carter, L.A. Morsberger, C. Iacobuzio-Donahue, G.A. Follows, A.R. Green, A.M. Flanagan, M.R. Stratton, P.A. Futreal, and P.J. Campbell

41

The Cul4-Ddb1Cdt2 Ubiquitin Ligase Inhibits Invasion of a Boundary-Associated Antisilencing Factor into Heterochromatin

S. Braun, J.F. Garcia, M. Rowley, M. Rougemaille, S. Shankar, and H.D. Madhani

55

Crystal Structure and Allosteric Activation of Protein Kinase C bII

T.A. Leonard, B. Ro´z_ ycki, L.F. Saidi, G. Hummer, and J.H. Hurley

67

Amyloid-like Aggregates Sequester Numerous Metastable Proteins with Essential Cellular Functions

H. Olzscha, S.M. Schermann, A.C. Woerner, S. Pinkert, M.H. Hecht, G.G. Tartaglia, M. Vendruscolo, M. Hayer-Hartl, F.U. Hartl, and R.M. Vabulas

79

The Cell-Non-Autonomous Nature of Electron Transport Chain-Mediated Longevity

J. Durieux, S. Wolff, and A. Dillin

92

Dynamics between Stem Cells, Niche, and Progeny in the Hair Follicle

Y.-C. Hsu, H.A. Pasolli, and E. Fuchs

106

Presenilin-Dependent Receptor Processing Is Required for Axon Guidance

G. Bai, O. Chivatakarn, D. Bonanomi, K. Lettieri, L. Franco, C. Xia, E. Stein, L. Ma, J.W. Lewcock, and S.L. Pfaff

119

Tunable Signal Processing in Synthetic MAP Kinase Cascades

E.C. O’Shaughnessy, S. Palani, J.J. Collins, and C.A. Sarkar

(continued)

R&D Systems Tools for Cell Biology Research™

Proteins Antibodies ELISAs Assay Services MultiAnalyte Profiling Activity Assays

R&D Systems Quantikine® ELISAs

The Most Referenced Immunoassays A direct measure of product quality is the frequency of citations in the scientific literature. R&D Systems has more than 20 years of experience designing, testing, and optimizing the most cited ELISA kits in the world. Find out why scientists trust R&D Systems ELISAs more than any other brand. R&D Systems is the Most Referenced ELISA Manufacturer

NEW Quantikine ELISA Kits

Stem Cells ELISpot Kits 17.3%

Flow Cytometry

R&D Systems 41.8%

Cell Selection

7.9% 3.7% 2.5% 2.1% 1.6% 1.4%

Approximately 42% of Referenced Immunoassays are Developed and Manufactured by R&D Systems. A survey of 860 manuscripts from 44 journals was conducted to compare the number of citations specifying the use of R&D Systems ELISAs to the number citing ELISAs from other commercial sources. A total of 433 ELISA citations referencing immunoassays from 66 different vendors were identified in the survey.

y_1-Acid Glycoprotein yAngiopoietin-like 3 yCathepsin V yClusterin yDkk-1 yEGF R/ErbB1 yEG-VEGF/PK1 yFetuin A yFGF-21 yGalectin-3 yGas 6 yGDF-15 yIL-17A/F Heterodimer yIL-19 yLipocalin-2/NGAL yMBL yProprotein Convertase 9/PCSK9 yPeriostin/OSF-2 yProgranulin yST2/IL-1 R4 yThrombomodulin/CD141 yTie-1 yTIM-1/KIM-1

For more information visit our website at www.RnDSystems.com/go/ELISA For research use only. Not for use in diagnostic procedures.

R&D Systems, Inc. www.RnDSystems.com R&D Systems Europe, Ltd. www.RnDSystems.co.uk R&D Systems China Co., Ltd. www.RnDSystemsChina.com.cn

RESOURCES 132

Identification of Candidate IgG Biomarkers for Alzheimer’s Disease via Combinatorial Library Screening

M.M. Reddy, R. Wilson, J. Wilson, S. Connell, A. Gocke, L. Hynan, D. German, and T. Kodadek

143

Phenotypic Landscape of a Bacterial Cell

R.J. Nichols, S. Sen, Y.J. Choo, P. Beltrao, M. Zietek, R. Chaba, S. Lee, K.M. Kazmierczak, K.J. Lee, A. Wong, M. Shales, S. Lovett, M.E. Winkler, N.J. Krogan, A. Typas, and C.A. Gross

POSITIONS AVAILABLE

On the cover: Presenilin-1 has been linked to Alzheimer’s disease (AD) because it is required for cleavage of amyloid-b precursor protein (APP), contributing to the loss of neuronal connections in AD pathogenesis. In this issue, Bai et al. (pp. 106–118) demonstrate that protease cleavage activity of Presenilin-1 is required for establishing neuronal connections by regulating axon guidance signaling during neural development. The cover image design is inspired from the art of paper cutting. The scissor represents Presenilin-dependent protease activity, which has functional roles in shaping neural circuitry during development. Artwork by Jamie Simon, Salk Institute.

Announcing an innovative new textbook from Academic Cell Primer to The Immune Response, Academic Cell Update Edition By Tak W. Mak and Mary Saunders

Primer to The Immune Response, Academic Cell Update Edition, is an invaluable resource for students who need a concise but complete and understandable introduction to immunology. Academic Cell textbooks contain premium journal content from Cell Press and are part of a new cutting-edge textbook/journal collaboration designed to help today’s instructors teach students to “think like a scientist.”

academiccell.com

Primer to The Immune Response Academic Cell Update Edition Tak W. Mak The Campbell Family Institute for Breast Cancer Research, Ontario, Canada Mary Saunders The Campbell Family Institute for Breast Cancer Research, Ontario, Canada Paperback/456 pages ISBN: 9780123847430 $79.95/£54.99/€64.95

Academic Cell is a dynamic textbook publishing partnership between Academic Press and Cell Press, two market-leading publishers bringing scientific advances from the world of life science research into the classroom. Order online now from: elsevierdirect.com/9780123847430 Request and examination copy from textbooks.elsevier.com

Twitter.com/academiccell

Facebook.com/academiccell

Leading Edge

In This Issue Shattering Chromosomes in Cancer PAGE 27

Somatic mutations driving cancer are thought to accumulate gradually over time. But could cancer-relevant mutations also arise in a single catastrophic event of chromosome shattering? Stephens et al. provide evidence for such a phenomenon, which they term chromothripsis, whereby tens to hundreds of genomic rearrangements occur in a single cellular crisis. The authors document chromothripsis in a number of human cancers and show that several oncogenic lesions can emerge from one genomic crisis. The findings have important implications for the temporal emergence of some cancers.

Ubiquitination Sculpts Chromatin Boundaries PAGE 41

Epe1, an S. pombe JmjC family protein, accumulates at boundaries between heterochromatin and euchromatin. This localization is puzzling, as Epe1 is recruited by HP1 silencing factors that are distributed throughout heterochromatin. Braun et al. demonstrate that the Cul4-Ddb1Cdt2 ubiquitin ligase fosters degradation of Epe1 within the body of heterochromatin, leading to accumulation at the boundaries. Ubiquitin-dependent sculpting of chromosomal protein distribution may be a general mechanism for proper assembly of chromatin domains.

Activation Is a Two-Step for PKCs PAGE 55

Protein kinase C isozymes (PCKs) are signal transducers that translocate to cell membranes, where they are activated by lipid dyacylglycerol binding to their C1A and C1B domains. Now, Leonard et al. solve the crystal structure of fulllength PKC bII, revealing an unexpected intermediate in the activation pathway where the active site is accessible to substrate but kept inactive by the C1B domain clamping down on a key phenylalanine side chain. This is reversed upon membrane binding, activating the enzyme. These findings define a novel regulatory mechanism for protein kinases.

Making Sense of the Aggregate Proteome PAGE 67

Protein aggregation is linked to neurodegeneration and other diseases, but how aggregation perturbs a cell is unclear. Using quantitiative proteomics, Olzscha et al. demonstrate that cytotoxicity correlates with the capacity of aggregates to sequester multiple cellular proteins with distinct properties, such as large size and an enrichment in unstructured regions. Many of the interacting proteins also occupy essential hub positions in cellular protein networks. Thus, by targeting a metastable subproteome, amyloid-like aggregation may cause multifactorial cell toxicity.

Presenilin Functions Pre-AD PAGE 106

Mutations in the intramembrane protease Presenilin-1 are the most common cause of familial Alzheimer’s disease. Bai et al. now show that Presenilin-1 has a role in establishing neuronal connections during development of the mouse spinal cord. The absence of Presenilin-1 disrupts the sequential cleavage of DCC, a receptor for the axon guidance cue Netrin, resulting in axon guidance defects. These findings provide a molecular link between neural circuit formation and neurodegeneration. Cell 144, January 7, 2011 ª2011 Elsevier Inc. 1

.%7


Reach Your Ideal Candidate! så0OSTåONEåORåMULTIPLEåå JOBåOPENINGSå så0ROMOTEåANåUPCOMINGåå CONFERENCEåORåEVENTå så!NNOUNCEåAåGRANTAWARDå så%LEVATEåYOURåå ORGANIZATIONSåPROlLE

careers.cell.com

Mitochondria Tweet about Aging PAGE 79

Linking mitochondria to aging has focused on reactive oxygen species as the main culprit. Durieux et al. now demonstrate that alteration of mitochondrial function in key tissues is essential for establishing and maintaining a prolongevity cue in worms. Interestingly, mitochondrial perturbation in one tissue is perceived and acted upon by the mitochondrial stress response pathway in a distal tissue. That mitochondrial function can be communicated cell non-autonomously to set the rate of aging suggests the existence of an unknown secreted signal termed a mitokine.

Synthetic Signaling Cascades Deconstructed PAGE 119

O’Shaughnessy et al. construct a minimal mammalian MAP kinase cascade in yeast and mathematically dissect the contributions of intrinsic and extrinsic perturbations to the systems-level properties of the insulated cascade. Their findings demonstrate the critical importance of relative kinase concentration in dictating the activation profile of MAPKs, identify cascading itself as a mechanism for generating ultrasensitivity, and detail design rules for the construction of synthetic cascades with distinct activation characteristics.

Good Antibody Hunting PAGE 132

Disease-specific antibodies have potential to serve as biomarkers in a simple blood test, but identifying such antibodies is challenging. In this issue, Reddy et al. present a general approach, based on screening synthetic peptide mimetics, for identifying disease-specific antibodies. They apply this method to identify two antibodies that may represent candidate biomarkers for Alzheimer’s disease.

Getting at the Root of the Hair Cycle PAGE 92

Hair follicles undergo cyclical bouts of hair growth, a process requiring activation of otherwise quiescent stem cells (SCs). Hsu et al. exploit this feature to define the point when activated SCs become irreversibly committed along their way to making hair. Early SC descendents retain stemness and return to the niche when hair growth stops, becoming the main source of SCs for the next hair cycle. Surprisingly, some proliferating descendents that have irreversibly lost stemness also home back to the niche, where they function to provide quiescent signaling cues that control the hair cycle.

The Book of E. coli PAGE 143

High-throughput genome sequencing has created a need for large-scale approaches for elucidating gene function. Nichols et al. combine chemical genomics with quantitative fitness measurements to attribute functional roles to hundreds of E. coli genes. The approach and the data set provide a rich community resource and yield insights into several aspects of bacterial physiology, including genome organization and antibiotic mechanisms.

Cell 144, January 7, 2011 ª2011 Elsevier Inc. 3

3TRUGGLINGåTOåKEEPåUPåWITHåå THEåLATESTåLIFEåSCIENCEåNEWS

#ELLå$AILY¬.EWS¬!GGREGATORåHASåTHEåSOLUTION 3UBSCRIBEåTOå&2%%å$AILYå.EWSå!LERTSåATåNEWSCELLCOMåANDå GETåTHEåLATESTåLIFEåSCIENCEåHEADLINESåDELIVEREDåTOåYOURåINåBOX

NEWSCELLCOM

Leading Edge

Cell Culture: New Year’s Diets If 2011 shapes up like the previous year, then 30% of Americans will go on a diet. In fact, many of us have already jumped on a weight loss regime to counteract overindulgences during the holidays. In this Cell Culture, we take a peek at the biology behind popular diets, exploring the metabolic and neurological tricks that ramp up lipid metabolism and curb appetite before and during meals.

Atkin’s Attacks Adipose Many dieters looking to drop a quick 5 lbs will turn to the tried-and-true Aktin’s Diet. First described almost 50 years ago, the Aktin’s Diet purportedly ‘‘turns your body into a fat-burning machine’’ by severely restricting the intake of sugars and starches. Indeed, numerous animal studies indicate that low-carbohydrate, high-fat diets, such as Aktin’s, induce weight loss by triggering a shift from carbohydrate metabolism to fat oxidation. What orchestrates this metabolic switch? The mammalian brain can’t metabolize fat. Thus, as blood sugar and insulin levels plummet during an Aktin’s diet, the liver starts to generate an alternative fuel— ketones. Fatty acids are released from adipose tissue and oxidized in the liver to the ketones acetoacetate and hydroxybutrate. Using microarray analysis, Badman et al. found that a strict Aktin’s-like diet increases expression of the fibroblast growth factor 21 (FGF21) by >25-fold in livers of mice. Disrupting this hormone-like signal blocks ketogenesis and dramatically raises levels of lipid and cholesterol in the ‘‘Ketogenic diets’’ boost FGF21 secreblood. Simultaneously, Inagaki et al. demonstrated that overexpressing FGF21 tion, which triggers a switch from carboboosts ketogenesis by 5-fold, decreases insulin levels, and stimulates lipid break- hydrate to fat metabolism. Image by K. down in adipose tissue. Both studies identified the nuclear receptor PPARa as the Mahan. activator of FGF21 transcription in the liver during fasting or low-carbohydrate, high-fat diets. Together, these findings suggest that Aktin’s-like diets work by mimicking a state of starvation. Remarkably, ‘‘ketogenic diets,’’ such as the Aktin’s, are effective at treating drug-resistant epilepsy, and 30% of patients become seizure free on this regime. Although the precise mechanisms underlying this effect are still unknown, a study by Ma et al. (2007) found that ketones reduce the firing of GABAergic neurons by opening ATP-sensitive potassium channels at the cell surface. The authors propose that ketones decrease ATP levels at the plasma membrane because they are oxidized in mitochondria but not in the cytosol as glucose is. Inagaki, T., et al. (2007). Cell Metab. 5, 415–425. Badman, M.K., et al. (2007). Cell Metab. 5, 426–437. Ma, W., et al. (2007). J. Neurosci. 27, 3618–3625.

Flattening the Belly Launched in 2009 by a New York Times best seller, the Flat Belly Diet is a relative newcomer on the weight loss block. Its premise is that adding monounsaturated fats to each meal redistributes body fat away from the abdomen, resulting in a ‘‘flat belly.’’ Although these waist-slimming claims are still controversial, a study by Schwartz et al. (2008) indicates that monounsaturated fats may contribute to weight loss by serving as precursors for a satiation signal generated in the gut during meals. Food in the small intestine stimulates the production of oleoylethanolamide (OEA), a ‘‘lipid messenger’’ that, when administrated pharmacologically, reduces calorie intake in mice by increasing the time between meals. Searching for the dietary source of OEA, Schwartz et al. injected a variety of compounds into the gut of mice. Oleic Olives contain the monounsaturated fat oleic acid, which serves as a precursor acid, a monounsaturated fat in olive oil, was the only compound tested that increased to a satiation signal during meals. This levels of OEA in the small intestine while also regulating enzymes that synthesize and photo is licensed from Flickr user Steve degrade OEA. Next, Schwartz et al. used a combination of biochemistry and mass Jurvetson (http://www.flickr.com/photos/ spectrometry to show that the lipid transporter CD36 facilitates the movement of oleic jurvetson/) under a Creative Commons acid into the small intestine’s epithelial cells, where the lipid serves as a direct Attribution license. precursor for OEA synthesis. Finally, they demonstrated that oleic acid’s effect on appetite depends both on the CD36 transporter and the nuclear receptor PPARa, a master regulator of fat storage and utilization. It will be interesting to determine whether other monounsaturated fats in the Flat Belly Diet, such as linoleic acid in almonds, also contribute to appetite suppression and reduced caloric intake in a similar fashion as oleic acid. Schwartz, G.J., et al. (2008). Cell Metab. 8, 281–288. Cell 144, January 7, 2011 ª2011 Elsevier Inc. 5

60

years of leadership in human genetics research, education and service. 1948–2008 www.ashg.org

Sprinkle Sensations In addition to metabolic signals from the gut, such as OEA, sensory signals also contribute to satiation and meal termination. In fact, one of the latest diet fads, the Sensa Diet, banks on the hypothesis that enhancing the odor of foods—by sprinkling fragrant crystals of flavors and scents atop meals—decreases calorie intake by boosting satiation signals in the central nervous system (CNS). When you eat, food aromas take one of two pathways to reach the olfactory epithelium: the orthonasal route through the nose and the retronasal route through the mouth. Although both pathways lead to the same olfactory receptors, Small et al. (2005) found that aromas traveling each route activate distinct regions in the human CNS. First, the authors inserted one tube under the nostrils (orthonasal route) and one tube at the nasopharynx (retronasal route) of volunteers. They then delivered a chocolate aroma through each tube and examined brain activity by functional magnetic resonance imaging (fMRI). Orthonasal delivery activated a region involved in reward anticipation Odors reach olfactory receptors by two (i.e., amygdala), whereas retronasal delivery triggered a network implicated in the plea- routes, through the nose (yellow dots) sure experienced when the reward is actually received (e.g., medial orbitofrontal and through the mouth (blue dots). Image cortex), suggesting that odors in the retronasal route probably contribute more to sati- courtesy of Neuron, 10.1016/j.neuron. ation signals than those in the orthonasal route. So does enhanced activity in these 2005.07.022. brain centers equate to smaller meals? To begin testing this hypothesis, Ruijschop et al. (2009) used a special mass spectrometry technique that quantifies in real-time aroma compounds generated in an individual’s retronasal passage. When volunteers were told to eat until they were ‘‘full and satisfied,’’ individuals that released more retronasal odorants tended to consume fewer total calories. However, the authors also found that the total concentration of aromas released from a given food varies significantly among individuals, suggesting that the complexity of a food’s odor, with or without Sensa sprinkles, is only one component contributing to satiation signals in the brain. Small, D.M., et al. (2005). Neuron 47, 593–605. Ruijschop, R.M., et al. (2009). J. Agric. Food Chem. 57, 9888–9894.

The South Beach Many weight loss regimes start with a short-term period of highly restricted eating intended to kick-start your metabolism. For example, the South Beach Diet begins with a two-week ‘‘Phase I’’ that removes all sugars and carbohydrates from meals. Such severe dietary restrictions obviously have positive effects on insulin signaling. But could this carbohydrate hiatus also reduce cravings by resetting dopamine signaling in the brain’s reward centers? Food gives humans pleasure by releasing dopamine in the dorsal striatum, and the amount of pleasure directly correlates with the amount of dopamine released. A recent study by Johnson and Kenny (2010) found that unlimited access to highly pleasurable food—cheesecake, bacon, frosting, and chocolate—dampens dopamine signaling in the dorsal striatum of rats and induces behavioral changes similar to those seen in animals addicted to narcotics. First, the authors train the rats that turning a wheel triggers a pleasurable stimulation from an electrode implanted in their hypothalamus. After 40-day access to the decadent foods, the rats not only gain weight, but their sensitivity to pleasurable feelings drops; the electrical pulse required to keep the rodents turning the wheel increases by 40%. Moreover, these ‘‘reward deficits’’ persist two weeks after withdrawal from the palatable foods—a significantly longer time than similar effects caused by narcotic addiction. Finally, Kenny and Could a boost in dopamine signaling Johnson show that decreased expression of dopamine D2 receptors in the dorsal curb cravings in ‘‘Phase I’’ of the South striatum accelerates the addiction-like behavior. Beach Diet? This photo is licensed from Interestingly, fMRI studies indicate that dopamine D2 receptors in the dorsal striaFlickr user miamism (http://www.flickr. tum also play a prominent role in appetite regulation in humans. Stice et al. (2008) com/photos/miamism/) under a Creative Commons Attribution license. found that dopamine signaling in response to food tends to scale inversely with body mass and future weight gain. Furthermore, this relationship is stronger in individuals with an allele of the dopamine D2 receptor associated with decreased expression, suggesting that the more pleasure we feel from food, the less we eat. It will be interesting to determine whether temporary carbohydrate ‘‘withdrawal’’ during ‘‘Phase I’’ of the South Beach Diet boosts dopamine signaling and pleasure sensitivity in dieters. Johnson, P.M., and Kenny, P.J. (2010). Nat. Neurosci. 13, 635–641. Stice, E., et al. (2008). Science 322, 449–452. Michaeleen Doucleff Cell 144, January 7, 2011 ª2011 Elsevier Inc. 7

BE THE FIRST

to read the latest issue of any Cell Press journal.

Register for Cell Press Email Alerts and get the complete table of contents as soon as the issue publishes online — FREE! Cell Press Email Alerts deliver the news, research, and commentaries featured in each journal’s latest issue, including the full title of every article, direct links to the articles, and the complete author list. Plus, to save you time, each research article has a brief summary highlighting its significant findings. You don’t have to be a subscriber to sign up for Cell Press Email Alerts. While subscribers have instant access to the full text of all articles listed in the Email Alerts, non-subscribers can read the abstracts of all articles as well as the full text of the issue’s Featured Article.

www.cellpress.com

Leading Edge

Previews Cancer Genomes Evolve by Pulverizing Single Chromosomes Matthew Meyerson1,2,4,* and David Pellman3,5,6 1Department

of Medical Oncology, Dana-Farber Cancer Institute, Boston, MA 02115 of Pathology 3Department of Cell Biology Harvard Medical School, Boston, MA 02115 4Broad Institute, Cambridge, MA 02142 5Howard Hughes Medical Institute, Chevy Chase, MD 20815-6789 6Department of Pediatric Hematology/Oncology, Dana-Farber Cancer Institute and Children’s Hospital, Boston, MA 02115 *Correspondence: [email protected] DOI 10.1016/j.cell.2010.12.025 2Department

A report in this issue describes ‘‘chromothripsis,’’ a new mechanism for genetic instability in cancer cells. Chromothripsis appears to be a cataclysmic event in which a single chromosome is fragmented and then reassembled. The phenomenon raises important questions of how chromosome rearrangements can be confined to defined genome segments. We tend to think of tumor evolution as the gradual acquisition of mutations that can occur with a uniform chance across the whole genome: a series of genetic changes that stimulate growth, attenuate cell death, destroy checkpoint controls, promote further genetic instability, and enable metastasis (Stratton et al., 2009; Nowell, 1976). For many tumors, this idea of gradual alteration of the genome matches the appearance of tumors under the microscope, where malignant lesions can develop from benign lesions. However, this is not always the case. Cancer genomes can also evolve by ‘‘punctuated equilibrium’’-like mechanisms in which one-off cataclysmic events generate the potential for multiple concurrent mutations. For example, critical shortening of telomeres triggers breakage-fusionbridge cycles that result in gene amplification and other chromosome rearrangements (McClintock, 1941; Sahin and Depinho, 2010). Developing tumor cells can also make a single large evolutionary step by failing cytokinesis, whereby the doubling of the centrosome number produces a storm of aneuploidy (Fujiwara et al., 2005). In this issue of Cell, Stephens et al. (2011) describe a new type of cataclysmic event that they call chromothripsis (Greek; chromos for chromosome, thripsis for shattered into pieces) in which chromosomes are broken into many pieces and then stitched back together (Figure 1).

These findings come amidst a flood of information from the large-scale resequencing of cancer genomes, which is providing important insights into the evolutionary paths available to developing cancers (Stratton et al., 2009). Such efforts help to identify changes that contribute to tumorigenesis, but also may reveal ‘‘passenger’’ alterations that create potential burdens on tumor cells that could be exploited for therapeutics. Thus, understanding the ways that cancer genomes can evolve is important; the underlying evolutionary mechanisms should constrain the composition of the chromosomes in the mature tumor cell. In their current work, Stephens et al. use paired-end next-generation sequencing across multiple cancer samples to determine chromosomal structure and, in particular, the breakpoints of copy number alterations. With this approach, they have identified a new type of chromosomal disruption in cancer whereby there are repeated switches in copy number state along the length of a chromosome or other genomic segment, often with hundreds of breakpoints within a chromosome arm. The chromosomal segments vary in copy number primarily by single segment changes: for example, a region with two copies would be followed by a single copy, followed by two, followed by three (Figure 1). Strikingly, these alterations are primarily limited to a single chromosome or, in some cases,

a few chromosomes that appear to be co-coordinately altered. As the chromosomes appear to be shattered and then stitched back together, they have coined the term chromothripsis. A combination of genome resequencing and analysis of single-nucleotide polymorphism arrays in cell lines and primary tumors suggests that chromothripsis occurs in 2%–3% of cancers, spanning a wide variety of tumor types. In certain tumors, such as osteosarcomas and chordomas, chromothripsis is observed in up to 25% of samples. Chromothripsis may lead to the generation of amplifications of one or more oncogenes or to the deletion of one or more tumor suppressor genes. For example, one small cell lung cancer cell line contains a normal copy of chromosome 8 and a massively rearranged derivative chromosome 8 with all the hallmarks of chromothripsis. This cell line also contains large numbers of double minute chromosomes comprised of 15 distinct segments of chromosome 8, all rearranged to one another and leading to amplification of the MYC oncogene. Most strikingly, fluorescence in situ hybridization (FISH) experiments demonstrate that the amplified sequences on the double minute chromosome are absent from the derivative chromosome 8. This strongly suggests that a single copy of chromosome 8 shattered and that most fragments were stitched together to generate the derivative chromosome, but other

Cell 144, January 7, 2011 ª2011 Elsevier Inc. 9

to one or two chromosomes fragments, including the MYC or to a single chromosome gene, were stitched together arm. The underlying mechainto a circular double minute nism is unclear; however, the chromosome whose amplifiauthors speculate that this cation conferred a growth could possibly be linked to critadvantage—all occurring at ical telomere shortening (Pamthe same time. In another palona et al., 2010). Short teloexample, a chordoma DNA meres can cause chromatid sample exhibits a complex fusion and the bridging of rearrangement that simuldicentric chromosomes across taneously disrupts the the cytokinetic furrow. The CDKN2A, WRN, and FBXW7 resolution of bridging chromotumor suppressor genes, somes is known to produce each present at different Figure 1. Stitching Together Shattered Chromosomes by Chromonuclear protrusions and fraglocations in the genome. In thripsis ments that, in principle, could principle, chromothripsis may Chromothripsis is proposed to involve the shattering of a single chromosome, a small group of chromosomes, or a single chromosome arm. The fragments, spatially isolate a chromosome. also promote cancer by genor a subset of the fragments, are then stitched together by nonhomologous Altogether, the discovery of eration of new fusion genes end-joining. The mechanism by which these alterations are confined to a small chromothripsis by Stephens as well. Given the complexity segment of the genome is not defined. et al. reveals a new way that of chromothriptic alterations, it will be a challenge to find a statistical (PCC), a phenomenon that was first ob- cancer genomes can evolve. In what approach to determine the functional served in cell fusion experiments (Rao appears to be a single step, numerous and Johnson, 1970; Sperling and Rao, genes can be mutated, amplified, and retargets of these alterations. Because chromothripsis The authors argue that the chromo- 1974). When chromosomes from an S arranged. thriptic events are likely to occur in a single phase nucleus are induced to undergo occurs in such a wide variety of tumors, catastrophic event rather than a series of chromosome condensation by signals the underlying mechanism is likely to subsequent and random alterations. from chromosomes derived from a cell reflect as yet undefined general properThree pieces of evidence suggest the in mitosis, the incompletely replicated ties of human cancer. possibility that chromothriptic changes chromosomes from the S phase nucleus have occurred in a single event. First, shatter. It is therefore tempting to specu- ACKNOWLEDGMENTS the number of copy number states found late that chromothripsis could initiate on the altered chromosome is restricted during mitosis by a PCC-like mechanism. M.M. is founder, shareowner, and consulting The next question is how the fragments advisor of Foundation Medicine. to two; under a model of progressive alterations, many copy number states might be stitched together. In principal, would be expected. Second, in the higher some information about the initial shatterREFERENCES copy number states, heterozygosity is ing as well as the stitching together might preserved; if there were progressive be gleaned from the sequence of the junc- Fujiwara, T., Bandi, M., Nitta, M., Ivanova, E.V., alterations, any early occurring deletion tions of the fragments on the derivative Bronson, R.T., and Pellman, D. (2005). Nature would eliminate heterozygosity. Finally, chromosome. For example, telomere 437, 1043–1047. the alterations cluster to a greater degree fusions between sister chromatids are McClintock, B. (1941). Genetics 26, 234–282. than would be expected from sequential expected to produce a large number of Murnane, J.P. (2006). DNA Repair (Amst.) 5, 1082– alterations in the chromosome. A statis- head-to-head duplications (Murnane, 1092. tical analysis based on Monte Carlo simu- 2006). However, chromothripsis pro- Nowell, P.C. (1976). Science 194, 23–28. lations of the progressive model is also duces highly complex derivative chromo- Pampalona, J., Soler, D., Genesca, A., and Tusell, consistent with the view that the limited somes that lack an identifiable signa- L. (2010). Mutat. Res. 683, 16–22. number of copy number states is very ture—the segments on the derivative Rao, P.N., and Johnson, R.T. (1970). Nature 225, unlikely to have occurred by chance chromosome have been joined by a seem- 159–164. through sequential alteration, again ingly random mechanism. The sequence Sahin, E., and Depinho, R.A. (2010). Nature 464, arguing for a catastrophic or ‘‘punctuated at the junction of each segment shows 520–528. either a lack of homology or microhomolequilibrium’’ model. Sperling, K., and Rao, P.N. (1974). Chromosoma What mechanisms could produce such ogy between the joined segments. Thus, 45, 121–131. massive but highly localized changes in the main conclusion we can draw from Stephens, P.J., Greenman, C.D., Beiyuan, F., the genome? The first interesting question the sequence analysis is that the ends Yang, F., Bignell, G.R., Mudie, L.J., Pleasance, is how does the chromosome get shat- are likely joined by the nonhomologous E.D., Lau, K.W., Beare, D., Stebbings, L.A., et al. (2011). Cell 144, this issue, 27–40. tered? One well-known mechanism by end-joining DNA repair system. Finally, we are left with the fascinating Stratton, M.R., Campbell, P.J., and Futreal, P.A. which chromosomes can be ‘‘pulverized’’ is premature chromosome compaction puzzle of how the pulverization is confined (2009). Nature 458, 719–724.

10 Cell 144, January 7, 2011 ª2011 Elsevier Inc.

Leading Edge

Previews Mitochondrial Stress Signals Revise an Old Aging Theory Dong Kyun Woo1 and Gerald S. Shadel1,2,* 1Department

of Pathology of Genetics Yale University School of Medicine, New Haven, CT 06520-8023, USA *Correspondence: [email protected] DOI 10.1016/j.cell.2010.12.023 2Department

In this issue, Durieux et al. (2011) describe a tissue-specific signal, originating from mitochondria, that acts cell non-autonomously to regulate life span in the nematode, C. elegans. This new finding provides a first step toward resolving the relative contributions of mitochondrial free radical damage and signaling mechanisms in aging. Mitochondrial dysfunction is central to theories of aging, with the production of high levels of reactive oxygen species (ROS) receiving the lion’s share of attention. Durieux et al. (2011) now suggest that mitochondria influence longevity in C. elegans not only through the production of ROS but also via a stress-evoked signal that acts in a cell-non-autonomous manner to regulate mitochondrial protein homeostasis. This exciting development highlights the need to rethink theories of aging to incorporate mitochondrial signaling events. As our understanding of mitochondria expands, so too does our appreciation for their diverse contributions to pathologies associated with aging and the regulation of life span. Mitochondria do much more than produce ATP via oxidative phosphorylation. These dynamic organelles participate in myriad cellular processes including apoptosis, ion homeostasis, and oxygen sensing; and their integrity is safeguarded by multiple quality control mechanisms, including the mitochondrial unfolded protein response (UPR) (Figure 1). Mitochondria also produce ROS, which were once thought of only as agents of oxidative damage but are now known to have important signaling functions (Hamanaka and Chandel, 2010). Although changes in mitochondrial respiration and ROS are implicated in aging, their relationship with longevity is complicated (Barja, 2007). Depending on the circumstance, alterations in mitochondrial respiration can either increase or decrease ROS

production and life span (for instance, see Bonawitz et al., 2007). Mild inhibition of mitochondrial activity extends life span in C. elegans. This effect is not simply a result of decreased oxidative damage-induced aging by ROS because electron transport chain (ETC) inhibition counteracts aging only when it occurs during a defined developmental period, the L3/L4 larval stages. One explanation for the time dependence of this phenomenon is that mitochondrial status is sensed during development, somehow programming rates of aging in adulthood. This type of mitochondrial adaptive response is similar to the concept of hormesis, which in the context of aging describes the heightened ability of an organism to respond to a certain condition (such as oxidative stress) following prior exposure to that condition (Gems and Partridge, 2008), and implies an underlying epigenetic basis. Support for a mitochondria-mediated hormetic effect in aging is provided by experiments in which C. elegans were fed a glucose-restricted diet, which caused an increase in mitochondrial respiration and ROS that, in turn, enhanced stress-resistance pathways and life span (Schulz et al., 2007). The recent discovery of a longevity-promoting, ROS-dependent gene expression program that occurs upon ETC inhibition (Lee et al., 2010) may begin to define a mitochondrial retrograde signaling pathway that could, in principle, promote a hormetic effect, supporting the idea that early exposure to mitochondrial

ROS leads to enhanced oxidative stress resistance later in life (Schulz et al., 2007). The current study by Durieux et al. contributes novel insight into how early exposure to mitochondrial stress promotes life-span extension while also raising new questions about the identity and transduction of mitochondrial adaptive signals. Previous studies have shown that global ETC inhibition can promote longevity, but with a cost to overall fitness, causing decreased fecundity and reduced motility (Dillin et al., 2002). Durieux et al. now show that tissue-specific knockdown of mitochondrial ETC components in either neurons or intestinal cells is sufficient for life-span extension without any of the deleterious effects associated with global downregulation of ETC. This perhaps unexpected effect does not appear to involve enhanced overall stress resistance but relies on activation of the mitochondrial UPR pathway (Haynes and Ron, 2010). Durieux et al. provide direct support for this idea by showing that the key transcriptional coactivator of mitochondrial UPR, the ubiquitin-like protein UBL-5, is indispensible for the increase in life span that results from mild mitochondrial ETC inhibition. Further linking the mitochondrial UPR and the ETC longevity pathway, the authors demonstrate that UBL-5 expression is required during the same developmental window as decreased ETC activity to favor extended life span. Importantly, both UBL-5 and ETC inhibition are dispensable for life-span extension via the

Cell 144, January 7, 2011 ª2011 Elsevier Inc. 11

and deciphering the relationwell-characterized insulin/ ships between these proIGF-1 signaling pathway, providing more evidence cesses in various tissues that at least two separate will allow for a more comprepathways regulate longevity. hensive mitochondrial theory Perhaps the most striking of aging. A deeper underfinding from this new study standing of the mechanisms is that the ETC-mediated that activate mitochondrial longevity pathway operates UPR and the identification at some level through a of life-span-promoting mitocell-non-autonomous mechkine(s) may yield new theraanism. The authors show pies for age-related diseases that neuron-specific ETC and contribute to longer, impairment induces a mitohealthier lives for humans. chondrial UPR response in the intestine and extends life ACKNOWLEDGMENTS span, a result most easily explained by an endocrine This work was supported by signal that relays changes Program Project Grant ES011163 in mitochondrial function/ from the NIH. stress in neurons to intestinal cells. Although the chemical REFERENCES nature of this signal remains unknown, the authors sugBarja, G. (2007). Rejuv. Res. 10, Figure 1. Longevity Is Regulated by Cell-Autonomous and -Non215–224. gest that a diffusible moleAutonomous Mitochondrial Stress Pathways cule, a ‘‘mitokine,’’ is released Severe mitochondrial dysfunction, which can include increased production of Bonawitz, N.D., Chatenay-Lapointe, reactive oxygen species (ROS), promotes aging. However, mild mitochondrial from certain tissues, broadM., Pan, Y., and Shadel, G.S. (2007). impairment and stress stimulate retrograde signaling and extend life span in casting a mitochondrial proCell Metab. 5, 265–277. C. elegans. Adaptive mitochondrial retrograde pathways relay mitochondrial longevity signal to target stress signals to the nucleus, activating mitochondrial quality control genes. Dillin, A., Hsu, A.-L., Arantes-Olitissues, such as the intestine Beyond maintaining mitochondrial integrity and function, both of which are viera, N., Lehrer-Graiwer, J., Hsin, necessary for maximal life span, Durieux et al. (2011) uncover another arm of (Figure 1). This development H., Fraser, A.G., Kamath, R.S., this adaptive response, showing that cell-non-autonomous signaling of the adds a completely new layer Ahringer, J., and Kenyon, C. mitochondrial unfolded protein response (UPR) between tissues is involved in (2002). Science 298, 2398–2401. of regulation to the concept life-span extension. They propose that a diffusible factor, a mitokine, is released from one tissue in response to mitochondrial stress and relays of adaptive mitochondrial Durieux, J., Wolff, S., and Dillin, A. longevity cues to other tissues. Together, cell-autonomous and -non-autonsignaling (hormesis) that may (2011). Cell 144, this issue, 79–91. omous mitochondrial stress signals likely cooperate to extend life span in explain other observations response to mild ETC inhibition. Gems, D., and Partridge, L. (2008). that link the nervous and Cell Metab. 7, 200–203. endocrine systems to regulaHamanaka, R.B., and Chandel, N.S. (2010). Trends tion of aging and life span in higher origin matters. How neuronal versus muscular mitokine signaling differs and Biochem. Sci. 35, 505–513. organisms (Russell and Kahn, 2007). The study by Durieux et al. is ground- whether a single systemic mitokine Haynes, C.M., and Ron, D. (2010). J. Cell Sci. 123, breaking in many respects, but questions promotes life-span extension remain 3849–3855. remain. For example, ETC knockdown in unclear. Another possibility may be the body-wall muscle upregulates mitochon- need for synergy between cell-autono- Lee, S.J., Hwang, A.B., and Kenyon, C. (2010). Curr. Biol. 20, 2131–2136. drial UPR in the intestine, but, unlike mous and cell-non-autonomous mitoneuronal ETC inhibition, it does not lead chondrial signals (Figure 1) in specific Russell, S.J., and Kahn, C.R. (2007). Nat. Rev. Mol. to life-span extension. Thus, although tissues for aging to be delayed. Therefore, Cell Biol. 8, 681–691. a mitochondrial UPR-mitokine signal can elucidating the precise connections Schulz, T.J., Zarse, K., Voigt, A., Urban, N., be transmitted to the intestine from at between mitochondrial ROS and mito- Birringer, M., and Ristow, M. (2007). Cell Metab. least two different tissues, the tissue of chondrial UPR in the context of longevity 6, 280–293.

12 Cell 144, January 7, 2011 ª2011 Elsevier Inc.

Leading Edge

Previews Fishing for Biomarkers with Antigen Mimics Tamsin M. Lindstrom1,2,* and William H. Robinson1,2,* 1Division

of Immunology and Rheumatology, Department of Medicine, Stanford University School of Medicine, Stanford, CA 94305, USA Affairs Palo Alto Health Care System, Palo Alto, CA 94304, USA *Correspondence: [email protected] (T.M.L.), [email protected] (W.H.R.) DOI 10.1016/j.cell.2010.12.022 2Veterans

Current efforts to identify antibodies that are biomarkers of disease rely on knowing the antigens they target. In many diseases, however, the relevant antigens are unknown. Reddy et al. (2010) now present an approach for discovering antibody biomarkers that avoids the need for antigen identification. Biomarkers, objective indicators of a specific biological state, have the potential to illuminate the pathogenesis of disease and to transform its management. Biomarkers may aid in diagnosing disease, predicting disease onset, and selecting appropriate therapy. However, with a few exceptions, this promise has yet to be fulfilled (Rifai et al., 2006)—partly because, for many diseases, molecular biomarkers have yet to be identified. Antibodies are one type of molecular biomarker. Because antibodies function by binding specific antigens, attempts to identify antibody biomarkers have so far involved using antigens to capture antibodies that are overproduced in disease. The problem with this approach is that in many diseases, particularly autoimmune diseases, the antigen that triggers the immune response is unknown. In this issue of Cell, Reddy et al. (2010) report a new approach for the discovery of antibody biomarkers, one that requires no knowledge of the specificity of the immune response. Instead of putative antigens, the authors use an array of random synthetic molecules to pinpoint diseaseassociated antibodies (see Figure 1). Identifying relevant antigens is at the heart of current approaches for discovering antibody biomarkers. Array-based approaches depend on exposing serum samples from patients to an ordered array of putative antigens, capturing those antibodies that bind antigens on the arrays, and measuring their levels (Robinson et al., 2002). Antibodies that are present at significantly higher levels in the serum of patients with the disease of interest

(compared to control serum from either healthy patients or patients with an unrelated disease) are candidate biomarkers. One major drawback of these antigen arrays is that they are biased, given that antigens are selected based on the likelihood that they play a role in the disease. Antigen arrays are thus ill-suited to de novo discovery. Less biased are the high-density antigen arrays, which comprise clones of a human cDNA library expressed in bacterial or insect cells (Auger et al., 2009; Horn et al., 2006). However, because they comprise either recombinant proteins or biomolecules isolated from tissues irrelevant to the disease, neither high-density nor conventional antigen arrays recapitulate the spectrum of posttranslational modifications that can occur in humans. This omission is another major drawback because many of the antigens that play a role in autoimmune disease are molecules with posttranslational modifications that elicit disease only in their modified form (Doyle and Mamula, 2001). Unlike arraybased approaches, mass-spectrometric approaches start with the extraction of biomolecules directly from diseased tissues and thus do take into account relevant posttranslational modifications. Immunoblotting with antibodies from patients’ sera can then pinpoint specific disease antigens, which can be identified using mass spectrometry (Wu and Mohan, 2009). Nonetheless, mass-spectrometric approaches suffer from their own set of drawbacks; for instance, the selection of the diseased tissue to be analyzed, as well as imperfections in

antigen isolation and sample preparation, introduce some bias. Reddy et al. tackle the problem from a new angle, devising an unbiased, highthroughput approach that is predicated on posttranslational modification. They keep the array format and readout the same as that of current antigen arrays but change the content of the arrays— the antibody bait. They reason that the primary antigens (those that trigger the initial immune response) are most likely to be biomolecules that are not only modified but modified in an abnormal way, owing to a pathological process characteristic of the disease. This concept resonates with current thinking, for instance about rheumatoid arthritis, an autoimmune disease affecting the joints. A key target of the aberrant antibody response in rheumatoid arthritis is a protein that has undergone citrullination (Whiting et al., 2010), a posttranslational modification that occurs during inflammation and cell death. So rather than use unmodified biomolecules, the authors use a combinatorial library of unnatural, synthetic molecules that might by chance mimic the antibody-binding site of the primary antigen. The premise is that these synthetic molecules, termed peptoids, can form shapes that cannot be formed by unmodified biomolecules. Through mimicry, then, peptoids might be able to pinpoint antibodies that are important to the disease process and thus aid in the discovery of biomarkers. To test their hypothesis, Reddy et al. initially use arrays of 4608 different peptoids to fish for antibodies associated

Cell 144, January 7, 2011 ª2011 Elsevier Inc. 13

Figure 1. Using Peptoids to Discover Antibody Biomarkers of Disease Reddy et al. (2010) describe a new approach for identifying antibody biomarkers of disease. They use arrays of synthetic molecules termed peptoids to capture antibodies from patients’ serum. To measure the levels of the IgG antibodies they capture, the authors use a fluorescently labeled anti-IgG antibody. A peptoid (Peptoid X) that by chance mimics the antibody-binding site of a key disease antigen will retain much more antibody from patients’ serum than from normal (nondiseased) serum. Antibody binding to Peptoid X could then serve as a marker of disease. In separate experiments, it may be possible to identify the endogenous target of the captured antibody by using the antibody to fish its antigen out of serum.

with experimental autoimmune encephalomyelitis (EAE), a mouse model of multiple sclerosis (an autoimmune disease targeting myelin sheaths). They identify three peptoids (named AMogP1–3, after the Mog peptide used to induce EAE in mice) that bind much more antibody in serum from mice with EAE than in serum from healthy mice, control mice immunized with ovalbumin, or mice with systemic lupus erythematosus, another autoimmune disease. The authors go on to show that antibody binding to AMogP1–3 can differentiate between healthy mice and mice with EAE. These results define antibody binding to AMogP1–3 as a biomarker of EAE. Although one particular antigen (the primary antigen) triggers the initial antibody response in autoimmune disease,

additional antigens (secondary antigens) form as the disease progresses, leading to the production of additional antibodies. Compared to antibodies against secondary antigens, antibodies against the primary antigen are more likely to be specific to the disease and to therefore serve as biomarkers. If the antibodies that bind AMogP1–3 recognize the primary antigen in EAE, then Mog should be their endogenous target. The authors garner two pieces of evidence showing that this is the case. First, they show that antibody reactivity to AMogP1–3 arises at the same time as the reactivity to Mog itself. Second, and more importantly, serum from EAE mice (which have been immunized with Mog) no longer reacts with AMogP1–3 once it has been depleted of anti-Mog antibodies. These findings

14 Cell 144, January 7, 2011 ª2011 Elsevier Inc.

provide proof of concept that an unnatural molecule can uncover the antibody that recognizes a disease-triggering antigen. But can such a peptoid-based approach be applied to human disease, which is much more varied than the carefully controlled disease induced in genetically identical laboratory mice? To address this question, the authors turn to Alzheimer’s disease. Although not classically considered an autoimmune disease, Alzheimer’s disease involves aberrations in levels of antibodies against b-amyloid (Britschgi et al., 2009) and ATP synthase (Vacirca et al., 2010). By screening serum samples against 15,000 peptoids, Reddy et al. identify three peptoids that can distinguish patients with Alzheimer’s disease from age-matched healthy individuals. If further studies validate antibody binding to these peptoids as a biomarker of Alzheimer’s (through evaluation in independent patient cohorts and in a larger number of samples), this new assay could greatly improve the management of Alzheimer’s disease, a disease for which there is currently no objective diagnostic. Given that antibodies play not only pathogenic but also protective roles in Alzheimer’s disease (Britschgi et al., 2009; Vacirca et al., 2010), searching for antibodies whose levels are abnormally low in Alzheimer’s disease patients may lead to the discovery of additional biomarkers. Although the authors’ approach is compelling, key questions remain about its clinical usefulness and its ability to uncover antibodies that are meaningful in terms of pathogenesis. Are the Alzheimer’s antibodies identified in this study true biomarkers of Alzheimer’s disease? Do the findings in the mouse model of multiple sclerosis translate to the human disease? Do the antibodies detected in Alzheimer’s disease target primary antigens, or do they target secondary antigens that might be unrelated to the pathogenesis of the disease? Knowing the identity of the relevant antigen in each disease is crucial for understanding disease pathogenesis and developing targeted therapies. Although not its primary objective, the peptoid assay could lead to the identification of endogenous targets of antibodies, if the antibodies captured in the peptoid assay are used to fish the complementary antigen out of serum.

But as Reddy et al. argue, for an antibody to be a useful biomarker, knowledge of its antigen is not necessary. In fact, antibodies of unknown specificity are already used in clinical diagnosis. For example, the cyclic citrullinated peptide test (CCP), which diagnoses rheumatoid arthritis, measures antibody binding to a collection of synthetic citrullinated peptides (Whiting et al., 2010). A big difference between the CCP test and a potential peptoid-based test, however, is this: Whereas the CCP test is the culmination of decades of research identifying citrullination as a key immunogenic process in rheumatoid arthritis, a peptoid-based test, requiring no prior knowledge of the disease at hand, could be developed in a fraction of that time. Thus, this new approach could prove to

be of tremendous value in clinically managing the many immune-mediated diseases whose pathogenesis is unclear.

Reddy, M.M., Wilson, R., Wilson, J., Connell, S., Gocke, A., Hynan, L., German, D., and Kodadek, T. (2010). Cell 144, this issue, 132–142. Rifai, N., Gillette, M.A., and Carr, S.A. (2006). Nat. Biotechnol. 24, 971–983.

REFERENCES Auger, I., Balandraud, N., Rak, J., Lambert, N., Martin, M., and Roudier, J. (2009). Ann. Rheum. Dis. 68, 591–594. Britschgi, M., Olin, C.E., Johns, H.T., Takeda-Uchimura, Y., LeMieux, M.C., Rufibach, K., Rajadas, J., Zhang, H., Tomooka, B., Robinson, W.H., et al. (2009). Proc. Natl. Acad. Sci. USA 106, 12145– 12150. Doyle, H.A., and Mamula, M.J. (2001). Trends Immunol. 22, 443–449. Horn, S., Lueking, A., Murphy, D., Staudt, A., Gutjahr, C., Schulte, K., Konig, A., Landsberger, M., Lehrach, H., Felix, S.B., et al. (2006). Proteomics 6, 605–613.

Robinson, W.H., DiGennaro, C., Hueber, W., Haab, B.B., Kamachi, M., Dean, E.J., Fournel, S., Fong, D., Genovese, M.C., de Vegvar, H.E., et al. (2002). Nat. Med. 8, 295–301. Vacirca, D., Delunardo, F., Matarrese, P., Colasanti, T., Margutti, P., Siracusano, A., Pontecorvo, S., Capozzi, A., Sorice, M., Francia, A., et al. (2010). Neurobiol. Aging. 10.1016/j.neurobiolaging.2010. 05.013. Whiting, P.F., Smidt, N., Sterne, J.A., Harbord, R., Burton, A., Burke, M., Beynon, R., Ben-Shlomo, Y., Axford, J., and Dieppe, P. (2010). Ann. Intern. Med. 152, 456–464, W155–W466. Wu, T., and Mohan, C. (2009). Autoimmun. Rev. 8, 595–598.

Cell 144, January 7, 2011 ª2011 Elsevier Inc. 15

Leading Edge

Review Epigenetics in Alternative Pre-mRNA Splicing Reini F. Luco,1 Mariano Allo,2 Ignacio E. Schor,2 Alberto R. Kornblihtt,2 and Tom Misteli1,* 1National

Cancer Institute, National Institutes of Health, Bethesda, MD 20892 USA de Fisiologı´a Biologı´a Molecular, LFBM and IFIBYNE-CONICET, Facultad de Ciencias Exactas Naturales, Universidad de Buenos Aires, Buenos Aires, Argentina *Correspondence: [email protected] DOI 10.1016/j.cell.2010.11.056 2Departamento

Alternative splicing plays critical roles in differentiation, development, and disease and is a major source for protein diversity in higher eukaryotes. Analysis of alternative splicing regulation has traditionally focused on RNA sequence elements and their associated splicing factors, but recent provocative studies point to a key function of chromatin structure and histone modifications in alternative splicing regulation. These insights suggest that epigenetic regulation determines not only what parts of the genome are expressed but also how they are spliced. Introduction The 10th anniversary of the publication of the first draft of the human genome sequence has sparked a renewed and expanded interest in alternative pre-mRNA splicing. Alternative splicing explains how the vast mammalian proteomic complexity can be achieved with the limited number of genes found in higher eukaryotes. Current estimates based on deep sequencing methodologies indicate that more than 90% of human genes undergo alternative splicing (Croft et al., 2000; Pan et al., 2008; Wang et al., 2008). Alternative splicing is an integral part of differentiation and developmental programs and contributes to cell lineage and tissue identity as indicated by the mapping of more than 22,000 tissue-specific alternative transcript events in a recent genome-wide sequencing study of tissue-specific alternative splicing (Wang et al., 2008). The importance of alternative splicing is dramatically highlighted by the numerous diseases that are caused by mutations in either cis-acting RNA elements or trans-acting protein splicing factors (Caceres and Kornblihtt, 2002; Cooper et al., 2009). Prominent splicing diseases include cystic fibrosis, frontotemporal dementia, Parkinsonism, retinitis pigmentosa, spinal muscular atrophy, myotonic dystrophy, premature aging, and cancer. Traditionally, alternative splicing has been thought to be predominantly regulated by splicing enhancers and silencers (Chasin, 2007). These short, conserved RNA sequences are typically 10 nt in length, are located either in exons or introns, acting either isolated or in clusters, and stimulate (enhancers) or inhibit (silencers) the use of splice sites through the specific binding of regulatory proteins such as SR proteins (serine/arginine-rich proteins) or heterogeneous nuclear ribonucleoproteins (hnRNPs) (Long and Caceres, 2009; Han et al., 2010). In addition, some silencers, instead of recruiting regulatory proteins, act by determining pre-mRNA secondary structure that hinders the recognition of a neighboring splicing enhancer by SR proteins (Buratti and Baralle, 2004). Disease mutations often affect the use of constitutive or alternative splice sites by cis-acting 16 Cell 144, January 7, 2011 ª2011 Elsevier Inc.

mutations that disrupt regulatory RNA sequence elements and by trans-acting mutations that affect the quality or quantity of alternative or constitutive splicing factors. It has long been clear that a full understanding of alternative splicing regulation will require the molecular characterization and structural modeling of the spliceosome and the analysis of RNA regulatory elements. However, the emerging complexity of alternative splicing regulation makes it apparent that information from those approaches will not be sufficient to decipher how alternative splicing is regulated. Here we discuss mechanisms and implications of the recently uncovered role of epigenetic components, such as chromatin structure and histone modifications, to alternative splicing regulation. Coupling of Transcription and Splicing More than 20 years ago, visualization of Drosophila embryo nascent transcripts by electron microscopy showed that splicing can occur cotranscriptionally (Beyer and Osheim, 1988) (Figure 1). Cotranscriptional splicing was later directly demonstrated for the human dystrophin gene (Tennyson et al., 1995), where it appears a very intuitive concept given that transcription of this 2400 kb gene would take 16 hr to complete. Furthermore, a quantitative study of the c-Src and fibronectin mRNAs, comparing chromatinbound and nucleoplasmic RNA fractions, shows that most introns are excised efficiently in the chromatin-bound fractions, with a gradient of cotranscriptional splicing efficiency from promoterproximal to promoter-distal introns, suggesting cotranscriptional splicing (Pandya-Jones and Black, 2009). However, cotranscriptionality of splicing is not strict, in the sense that introns are not necessarily removed in the exact order that they are transcribed (Attanasio et al., 2003; Bauren and Wieslander, 1994; Kessler et al., 1993; LeMaire and Thummel, 1990). If that were the case, the competition between splicing sites that leads to alternative splicing would be impossible. Splicing complexes are recruited to all introns and exons in a time window that begins when the target sequence

Figure 1. Coupling of Transcription and RNA Processing RNA polymerase II (green) recruits RNA-processing factors such as the 50 cap-binding complex (CAP) (yellow), splicing and pre-spliceosome factors (red), and the polyadenylation complex (blue) in the context of nucleosome-containing chromatin. Recruitment of RNA-processing factors occurs via the RNA Pol II C-terminal domain (CTD), and much of RNA processing occurs cotranscriptionally.

is transcribed and extends to the moment of splicing catalysis. For the entire splicing reaction to be cotranscriptional, both recruitment and catalysis must occur before transcription termination and transcript release. Alternatively, recruitment of some or all splicing factors may occur cotranscriptionally, but the catalysis itself may occur posttranscriptionally. Cotranscriptional pre-mRNA splicing appears to be a general rule for long mammalian genes. It is unclear how prevalent it is in organisms with shorter introns, such as yeast, although several studies support the notion that recruitment of spliceosomal components is also mostly cotranscriptional in this organism (Gornemann et al., 2005; Kotovic et al., 2003; Lacadie and Rosbash, 2005; Tardiff et al., 2006) (Figure 1). Completion of intron removal appears to be posttranscriptional in most cases, and only in genes with relatively long downstream exons does it occur prior to transcript release (Tardiff et al., 2006). The message from these studies is that cotranscriptional recruitment of splicing factors is largely preferred, but that cotranscriptional completion of intron removal is not mandatory and depends on the specific kinetics of transcription and splicing. In other words, the selective pressure in favor of cotranscriptional splicing acts on the association of splicing factors, which can be viewed as the ‘‘commitment to splice,’’ rather than on the catalysis itself. This might not apply to other RNA-processing events like capping and cleavage/polyadenylation (McCracken et al., 1997a, 1997b; Hirose and Manley, 1998; Maniatis and Reed, 2002; Moore and Proudfoot, 2009), wherein both the recruitment of the factors and enzymes involved as well as the catalysis appear to be cotranscriptional. Although cotranscriptionality of splicing is a prerequisite for coupling, it does not necessarily mean the two events are coupled. Cotranscriptionality simply means that splicing takes place, or is committed to occur, before the nascent RNA is released from RNA polymerase (Pol) II. Coupling implies that the transcription and splicing machineries interact with each other or that the kinetics of one process determines the outcome of the other. Efficient coordination between transcription and processing may be a specific feature of RNA Pol II and particularly of the carboxy-terminal domain (CTD) of its catalytic subunit given that a phosphorylated CTD is required for cotranscriptional splicing (Bird et al., 2004) (Figure 1). When protein-coding genes are placed under the control of either RNA Pol I,

RNA Pol III, or T7 RNA polymerase promoters, transcription takes place, but pre-mRNA processing is impaired and the resulting transcripts are poorly spliced (Dower and Rosbash, 2002; McCracken et al., 1998; Sisodia et al., 1987; Smale and Tjian, 1985). In fact, recruitment of splicing factors to sites of transcription is dependent on RNA Pol II CTD (Misteli and Spector, 1999) and deletion of the CTD causes defects in capping, cleavage/polyadenylation, and splicing of the b-globin transcript (McCracken et al., 1997b) (Figure 1). Many splicing factors are able to interact with RNA Pol II in vivo, including almost all known SR proteins and U1snRNP, and in nuclear extracts that support both transcription and splicing in vitro, SR proteins appear to be much more effective in promoting splicing when the latter is cotranscriptional than when it is posttranscriptional (Das et al., 2007). However, SR proteins are not delivered to splicing sites by RNA Pol II alone but rather require ongoing pre-mRNA synthesis (Sapra et al., 2009), demonstrating that recruitment is not dependent on preassembled SR-RNA Pol II complexes. Coupled in vitro transcription/splicing assays, although not necessarily reflecting functional coupling as it would occur in vivo (Lazarev and Manley, 2007), show that nascent pre-mRNA synthesized by RNA Pol II is stabilized and efficiently spliced (Hicks et al., 2006). This is likely because it is immediately and quantitatively directed into the spliceosome assembly pathway, instead of being assembled into nonspecific hnRNP complexes, which are inhibitory for spliceosome assembly (Das et al., 2006). Strong evidence for functional coupling between transcription and pre-mRNA processing comes from analyzing how modulation of transcription affects alternative splicing events. It has been demonstrated that the outcome of alternative splicing is influenced by the promoter used to drive transcription (Cramer et al., 1999, 1997; Pagani et al., 2003), by hormone-responsive elements (Auboeuf et al., 2002), and by recruitment of different transcription factors or coactivators to the promoter (Auboeuf et al., 2004a, 2004b; Nogue´s et al., 2002). The effects are not the trivial consequence of different mRNA levels produced by each promoter but depend on qualitative properties conferred by promoters to the transcription/RNA-processing machinery. Control of Alternative Splicing by Elongation Rate The standard experimental approach to study splicing mechanisms is by in vitro splicing assays. This methodology employs in vitro synthesized pre-mRNA substrates in splicing reactions carried out in cell-free nuclear extracts. Although these Cell 144, January 7, 2011 ª2011 Elsevier Inc. 17

conditions are appropriate to identify splicing factors and RNA intermediates, they are not ideally suited to obtain an accurate picture of the timing of splicing in relation to the generation of nascent RNA during transcription. These limitations can be overcome by in vivo experiments using either transfected reporter minigenes or endogenous genes as templates for splicing reactions. It was in fact differences in the behavior of a splicing event in vivo compared to in vitro that first hinted at a kinetic role for transcription on splicing. Eperon et al. (1988) found that the use of an alternative 50 splice site sequestered within a short stem of RNA secondary structure was determined by the length of the loop in vivo. Above a threshold loop length, the alternative site was used despite the potential structure. In contrast, the alternative site was not used during splicing in vitro with all lengths of loop tested (Eperon et al., 1988). The simplest interpretation of these experiments is that the rate of RNA synthesis affects its secondary structure, which in turn affects splicing. Further evidence for a kinetic link between transcription and splicing came from experiments in which a MAZ sequence, which leads to RNA Pol II pausing, inserted into the tropomyosin gene promoted higher inclusion of tropomyosin exon 3 (Roberts et al., 1998). Conclusive evidence for a role of elongation on alternative splicing regulation was the finding that the nature of the promoter affects alternative splicing outcome (Cramer et al., 1997, 1999; Kornblihtt, 2005). The original observation of this promoter effect involved transient transfection of mammalian cells with reporter minigenes for the alternatively spliced cassette exon 33 (E33, also referred to as EDI or EDA) of human fibronectin (FN) under the control of different RNA Pol II promoters. When transcription of the minigene was driven by the b-globin promoter, for example, E33 inclusion levels in the mature mRNA were about 10 times lower than when transcription was driven from the FN or cytomegalovirus (CMV) promoter. These effects were not the consequence of the promoter strength but depended on some qualitative properties conferred by promoters to the transcription/RNA-processing machinery. Two nonexclusive mechanisms could explain the promoter effect: differential promoter occupation could affect the recruitment of splicing factors by the transcription machinery (recruitment coupling) or determine different rates of RNA Pol II elongation (kinetic coupling). Several lines of evidence support the idea that RNA Pol II elongation can affect alternative splicing through kinetic coupling (Figure 2). Replication of reporter plasmids for alternative splicing in transiently transfected cells greatly stimulated E33 inclusion. This effect was counteracted by treating the cells with trichostatin A (TSA), a potent inhibitor of histone deacetylation and therefore a chromatin ‘‘opener,’’ allowing for the possibility that replication conveys a more compact chromatin structure to the template, thus slowing elongation and leading to higher E33 inclusion (Kadener et al., 2001). Furthermore, drugs that inhibit elongation, like DRB (Kadener et al., 2001; Nogue´s et al., 2002), flavopiridol, or camptothecin (de la Mata et al., 2010), favor E33 inclusion. On the other hand, activation of transcription by Sp1, a transcription factor that promotes initiation, has no effect on E33 inclusion, whereas activation by VP16, a factor that promotes both initiation and elongation, decreases E33 inclusion (Nogue´s et al., 2002). The strongest evidence for 18 Cell 144, January 7, 2011 ª2011 Elsevier Inc.

a kinetic role of RNA Pol II elongation comes, however, from a slow mutant of RNA Pol II, which increases E33 inclusion in human cells (de la Mata et al., 2003). Interestingly, the homologous mutation in Drosophila (C4 Pol II) is viable but shows changes in the alternative splicing pattern of ultrabithorax (Ubx) mRNA that are consistent with the only conspicuous phenotype of the C4 flies, which is an enlargement of the halteres that resembles the Ubx mutants. Why slowing elongation would only affect the Ubx gene is not known, but a clue might be that this gene has the longest introns in Drosophila (17 and 50 kb) flanking the alternative exons affected in the C4 genotype, suggesting that elongation becomes more critical when introns are long. Similar effects of elongation on splicing have been reported in yeast on an artificially created alternative exon when transcription is carried out by a slow RNA Pol II mutant or when the elongation factor TFIIS is mutated (Howe et al., 2003). Finally, DNA-damage signaling following irradiation of cells with UV light affects alternative splicing of fibronectin, caspase 9, Bcl-x, and other human genes as a consequence of the inhibition of RNA Pol II elongation caused by UV-dependent hyperphosphorylation of the CTD (Mun˜oz et al., 2009). These data support a ‘‘first come, first served’’ model for regulation of alternative splicing (Aebi and Weissmann, 1987) (Figure 2). In one version of this model, slow elongation favors removal of the intron upstream of an alternative cassette exon before removal of the downstream intron. In an alternative version, slow elongation favors recruitment of splicing factors to the upstream intron before the downstream intron is synthesized, which in turn would promote exon inclusion. Once commitment is achieved, the order of intron removal becomes irrelevant (Figure 2). The latter model is supported by recent evidence showing that there is a preferential removal of the intron downstream of the fibronectin cassette exon 33 before the upstream intron has been removed (de la Mata et al., 2010). Most importantly, whereas cis-acting mutations and trans-acting factors that upregulate E33 inclusion act by changing the relative order of intron removal, reduction of elongation, which also causes higher E33 inclusion, does not affect the order of intron removal, suggesting that slow elongation favors commitment to exon inclusion during spliceosome assembly (de la Mata et al., 2010). According to this, ‘‘first served’’ would not be equivalent to ‘‘first excised’’ but to ‘‘first committed,’’ in agreement with the observed preferential cotranscriptionality of spliceosome recruitment rather than catalysis. Chromatin and Histone Modifications as Regulators of Alternative Splicing As we delve deeper into the regulation of alternative splicing, it is becoming clear that control of splice site choice is far more complex than anticipated. Neither RNA-binding elements nor control by RNA Pol II elongation rate appear sufficient to fully explain the faithful regulation of alternative splicing. RNA-binding motifs are not always conserved between genes, and even when motifs are transcribed that contain errors, they often still accurately recruit the appropriate set of splicing factors to the exon (Fox-Walsh and Hertel, 2009). Similarly, although RNA Pol II elongation rate affects splicing outcome in different scenarios (de la Mata et al., 2003; Mun˜oz et al., 2009), it remains unclear

Figure 2. The RNA Polymerase II Kinetic Model for Alternative Splicing Rapid elongation of RNA polymerase II (Pol II) leads to simultaneous availability to the splicing machinery of a weak (red) and a strong (blue) splice site, which compete for the recruitment of splicing factors (purple, blue, and green ovals) resulting in skipping of the weaker exon (orange rectangle). Pausing or slowing down of the RNA Pol II favors the recruitment of the splicing machinery to the first transcribed, weaker exon leading to its subsequent inclusion in a ‘‘first served, first committed’’ model.

to what extent RNA pol II processivity can be modulated in vivo, how RNA Pol II elongation rate is controlled, and whether regulation of alternative splicing patterns through RNA Pol II kinetics is a commonly used mechanism in vivo. These considerations indicate that other mechanisms contribute to the control of alternative splicing. A major recent discovery is that chromatin structure and epigenetic histone modifications act as key regulators of alternative splicing. Chromatin Structure The first, albeit indirect, evidence that chromatin structure participates in the regulation of alternative splicing was the finding that fibronectin exon E33 inclusion was sensitive to replication-mediated chromatinization status of the plasmid and to the histone deacetylase inhibitor TSA (Kadener et al., 2001; Nogue´s et al., 2002). Further support came from the study of hormone-sensitive promoters that were tested for their effects on alternative splicing of a CD44 reporter gene (Auboeuf et al., 2002). Treatment with different steroid hormones induced changes in CD44 alternative splicing only if the minigene was under the control of the appropriate steroid-dependent promoter and in the presence of the specific hormone receptor, even though strong constitutive promoters were used (Auboeuf et al., 2002). Importantly, the effect on splicing was not due to changes in transcription rate, the density of the RNA Pol II, the strength of the promoter, or saturation of the splicing machinery but appeared mediated by the recruitment of specific hormone

receptor coregulators that remodeled chromatin (Auboeuf et al., 2002). Along the same lines, the histone acetyltransferase Gcn5 in yeast (Gunderson and Johnson, 2009) and STAGA in humans (Martinez et al., 2001) physically interact with U2 snRNPs, and the histone arginine methyltransferase CARM1 interacts with U1 snRNP proteins (Cheng et al., 2007; Ohkura et al., 2005), suggesting a role of chromatin complexes in facilitating the correct assembly of the pre-spliceosome on pre-mRNA. These effects are independent of elongation rate, arguing for a more direct role for chromatin structure on splicing factor recruitment (Gunderson and Johnson, 2009). Furthermore, chromatin remodelers of the SWI/SNF family in humans and Drosophila also have an effect on alternative splicing that is independent of their ATPase remodeling activity and dependent on physical interaction and recruitment of snRNPs U1 and U5 (Batsche et al., 2006; Tyagi et al., 2009). The recent advent of methods to map chromatin structure at a genome-wide scale further supports a role for chromatin structure in alternative splicing. Genome-wide mapping of nucleosome positioning by micrococcal nuclease (MNase) digestion from various species has shown that nucleosomes are positioned nonrandomly along genes and are particularly enriched at intron-exon junctions, thus marking exons (Andersson et al., 2009; Chodavarapu et al., 2010; Dhami et al., 2010; Kolasinska-Zwierz et al., 2009; Nahkuri et al., 2009; Ponts et al., 2008; Schwartz et al., 2009; Spies et al., 2009; Tilgner et al., 2009). Nucleosomes, defined as a stretch of 147 bp of DNA wrapped around an octamer of histone proteins, are structural units of chromatin that determine chromatin conformation and compaction. Intriguingly, the average size of a mammalian exon is similar to the length of DNA wrapped around a nucleosome, possibly pointing to a protective role of the nucleosome and a function in exon definition (Schwartz et al., 2009; Tilgner et al., 2009). Indeed, nucleosome enrichment around exons is conserved in evolution from plants to mammals and found both in somatic cells and gametes (Nahkuri et al., 2009), suggesting an essential role of nucleosome positioning in exon definition. The marking of exons by nucleosomes may Cell 144, January 7, 2011 ª2011 Elsevier Inc. 19

play a role in splicing regulation given that they are positioned irrespective of gene expression levels (Andersson et al., 2009; Tilgner et al., 2009). Moreover, isolated exons in the middle of long introns display higher nucleosome positioning than clustered exons separated by small introns (Spies et al., 2009), whereas pseudo-exons, which are nonincluded intronic sequences flanked by strong splice sites, are depleted of nucleosomes (Tilgner et al., 2009). More tellingly, included alternatively spliced exons are more highly enriched in nucleosomes than excluded ones (Schwartz et al., 2009) and nucleosome density varies according to splice site strength with stronger positioning at exons defined by weaker splice elements (Spies et al., 2009; Tilgner et al., 2009), arguing for a role of nucleosome positioning not only in exon definition but also in the regulation of splicing. Along with nucleosomes, RNA Pol II is also differentially distributed along genes in plants and humans with preferential accumulation at exons relative to introns (Brodsky et al., 2005; Chodavarapu et al., 2010; Schwartz et al., 2009). Nucleosomes have been shown to behave as barriers that can locally modulate RNA Pol II density by inducing its pausing (Hodges et al., 2009a). Together with the ability of RNA Pol II to interact with histone modifiers, such as the histone 3 lysine 36 (H3K36) methyltransferase Set2 (Xiao et al., 2003), and to recruit splicing regulators, such as SR proteins or U2 snRNP subunits (de la Mata and Kornblihtt, 2006; Listerman et al., 2006), nucleosome positioning may be modulating RNA Pol II density at exons and therefore splicing efficiency. In agreement, RNA Pol II is more highly enriched at alternatively spliced exons than at constitutive ones (Brodsky et al., 2005). Furthermore, overexpression of the ATPase-dependent chromatin-remodeling complex SWI/SNF subunit Brm in human cells induces accumulation of phospho-RNA Pol II in a central block of alternative exons of the CD44 gene and causes increased inclusion of these exons into mature mRNA (Batsche et al., 2006). Although these observations point to a role of nucleosome positioning and chromatin structure in alternative splicing regulation, a caveat of these studies is their correlative nature. Directed experiments to test the effect on alternative splice site selection upon modulation of chromatin and nucleosome positioning in a targeted fashion will be required to distinguish direct from indirect effects on alternative splicing. Histone Modifications in Alternative Splicing Regulation Histone modifications are emerging as major regulators of alternative splicing. Genome-wide analysis of the distribution of 42 histone modifications reveals that histone marks are nonrandomly distributed in the genome and that several modifications are enriched specifically in exons relative to their flanking intronic regions (Kolasinska-Zwierz et al., 2009; Spies et al., 2009; Andersson et al., 2009; Schwartz et al., 2009). Even though the enrichment of many histone modifications is a reflection of the higher density of nucleosomes at exons, some histone marks such as trimethylated H3K36 (H3K36me3), H3K4me3, and H3K27me2 are elevated even after normalization for nucleosome enrichment, whereas others, such as H3K9me3, are depleted (Dhami et al., 2010; Spies et al., 2009). In support of a splicing regulatory role of histone marks, histone modification levels do not correlate with transcriptional activity (Spies et al., 2009) and in active genes the transcription-associated H3K36me3 20 Cell 144, January 7, 2011 ª2011 Elsevier Inc.

modification is less prominently enriched in alternatively spliced exons than in constitutive exons (Andersson et al., 2009; Kolasinska-Zwierz et al., 2009). An additional indication for a role of histone modifications in alternative splicing is the observation that treatment of cells with the histone deacetylase inhibitor TSA induces skipping of the alternatively spliced fibronectin E33 and the neural cell adhesion molecule (NCAM) exon 18 (Nogue´s et al., 2002; Allo´ et al., 2009; Schor et al., 2009). In a more physiological context, depolarization of human neuronal cells increases H3K9 acetylation and H3K36 methylation locally around the alternatively spliced exon 18 of NCAM and induces exon skipping (Schor et al., 2009). Noticeably, no changes in histone acetylation are observed at the NCAM promoter. This reversible effect may be due to an intragenic and local modulation of the RNA Pol II elongation rate (Schor et al., 2009). Furthermore, targeting of an intronic sequence upstream of the alternatively spliced E33 of fibronectin with small-interfering RNAs induces local heterochromatinization and increased E33 inclusion without affecting general transcription levels (Allo´ et al., 2009). Consistently, inhibition of histone deacetylation, DNA methylation, H3K9 methylation, and downregulation of heterochromatin protein 1a (HP1a) abolishes the siRNA-mediated effect on exon E33 splicing (Allo´ et al., 2009), suggesting a role of these modifications in alternative splicing regulation. Further evidence for histone-mediated alternative splicing control comes from observations on the human fibroblast growth factor receptor 2 (FGFR2) gene. FGFR2 is alternatively spliced into two mutually exclusive and highly tissue-specific isoforms, FGFR2-IIIb and -IIIc. According to the pattern of splicing, the gene is enriched in a particular subset of histone modifications with H3K36me3 and H3K4me1 accumulating along the alternatively spliced region in mesenchymal cells, where exon IIIc is included, and H3K27me3 and H3K4me3 enriched in epithelial cells, where exon IIIb is used (Luco et al., 2010). Importantly, modulation of H3K36me3 or H3K4me3 levels by overexpression or downregulation of their respective histone methyltransferases changes the tissue-specific alternative splicing pattern in a predictable fashion (Luco et al., 2010). Taken together, these observations suggest that localized changes in chromatin conformation and histone modification signatures along an alternatively spliced region can change splicing outcome. Although there is no experimental evidence at present, it is possible that DNA methylation may also, directly or indirectly via histone modifications, affect splice site choice. DNA methylation patterns correlate better with histone methylation patterns than with genome sequence context (Meissner et al., 2008). Mapping in plants and human cells of DNA methylation levels by single-molecule whole-genome bisulfate sequencing reveals that DNA methylation is also nonrandomly distributed along the genome, specifically marking exons (Chodavarapu et al., 2010; Hodges et al., 2009b) and correlating well with H3K36me3 but inversely correlating with H3K4me2 levels (Hodges et al., 2009b). These observations point to a role for epigenetic modifications in the regulation of alternative splicing, and this regulation may involve the modulation of RNA Pol II elongation rate. However, an additional mechanism has recently emerged involving direct physical crosstalk between chromatin and the splicing

Figure 3. The Chromatin-Adaptor Model of Alternative Splicing Histone modifications along the gene determine the binding of an adaptor protein that reads specific histone marks and in turn recruits splicing factors. In the case of exons whose alternative splicing is dependent on polypyrimidine tractbinding protein (PTB) splicing factor, high levels of trimethylated histone 3 lysine 36 (H3K36me3, red) attract the chromatin-binding factor MRG15 that acts as an adaptor protein and by protein-protein interaction helps to recruit PTB to its weaker binding site inducing exon skipping. If the PTBdependent gene is hypermethylated in H3K4me3 (blue), MRG15 does not accumulate along the gene, and PTB is not recruited to its target premRNA, thus favoring exon inclusion.

machinery via an adaptor complex (Sims et al., 2007; Luco et al., 2010) (Figure 3). Chromatin-Splicing Adaptor Systems A hint toward a direct role for histone modifications in alternative splicing regulation came from comparative mapping of a set of histone modifications along several genes whose alternative splicing is dependent on the polypyrimidine tract-binding protein (PTB) splicing factor. These studies revealed a strong correlation between several histone modifications across the alternatively spliced regions and splicing outcome (Luco et al., 2010). PTBdependent genes were found to be enriched in H3K36me3 and depleted in H3K4me3 in the alternatively spliced regions. Modulation of these histone marks was sufficient to switch the pattern of PTB-dependent exon inclusion (Luco et al., 2010). The molecular mechanism by which H3K36me3 acts in this case does not appear to involve modulation of RNA Pol II elongation rate but rather the creation of a platform on chromatin for the recruitment of PTB to the nascent RNA (Figure 3) (Luco et al., 2010). This occurs via an adaptor protein, MRG15, that specifically binds to H3K36me3. The high levels of H3K36me3 along the alternatively spliced region of the gene attract MRG15, which in turn interacts with PTB recruiting it to the nascent RNA (Luco et al., 2010). In contrast, in cell types where H3K36me3 levels are low, the splicing repressor PTB is only poorly recruited to the newly forming RNA as a consequence favoring inclusion of the PTB-dependent exon (Luco et al., 2010) (Figure 3). H3K36me3, MRG15, and PTB thus establish a chromatin-splicing adaptor system. In line with this interpretation, increasing H3K36me3 levels in the absence of the MRG15 adaptor protein has no effect on FGFR2 alternative splicing (Luco et al., 2010). Although histone modifications clearly play a direct role in splicing regulation in this system, interestingly, the histone modifications do not appear to be the sole determinant of splicing outcome; they rather act as a modifier. Genome-wide analysis

of PTB-dependent alternative splicing patterns reveal that the splicing events that are most sensitive to changes in histone modifications rely on weak PTBbinding sites whereas alternative splicing events involving strong PTB-binding sites are not dependent on H3K36me3, suggesting that epigenetic modifications act in concert with RNA-binding elements to strengthen their effect (Luco et al., 2010). There is reason to believe that the combination of H3K36me3/ MRG15/PTB is not the only chromatin-splicing adaptor system in mammalian cells (Figure 4). It is known that H3K4me3 levels play a role in the recruitment of the early spliceosome to human cyclin D1 pre-mRNA via binding of the chromatin-adaptor protein CHD1 (Sims et al., 2007). CHD1 contains a chromodomain that specifically recognizes H3K4me3 and interacts with components of the U2 snRNP complex but not U1 snRNP (Sims et al., 2007). Consistent with a role in splicing regulation, downregulation of H3K4me3 or CHD1 alters the efficiency of pre-mRNA splicing and reduces association of splicing factor 3a (SF3a) subcomplexes and U2 snRNP with pre-mRNA in vitro and in vivo (Sims et al., 2007). Interestingly, CHD1 is also a component of the histone acetyltransferase SAGA complex (Pray-Grant et al., 2005) in which Gcn5, which binds to acetylated H3 (Li and Shogren-Knaak, 2009), also interacts and recruits U2 snRNP components to the exon (Gunderson and Johnson, 2009; Figure 4). Another example of a possible chromatin-splicing adaptor system is H3K9 trimethylation and HP1 proteins, which appear to recruit hnRNPs in Drosophila (Piacentini et al., 2009; Figure 4). Mass spectrometry analysis of proteins that bind to H3K9me identified the chromatin-binding protein HP1a/b and the splicing factors SRp20 and ASF/SF2 as interaction partners (Loomis et al., 2009). Coimmunoprecipitation experiments confirmed that HP1b interacts with ASF/SF2 in humans (Loomis et al., 2009) and HP1a with hnRNP proteins in Drosophila (Piacentini et al., 2009). These results point to a possible role for H3K9me3 in the regulation of recruitment of splicing factors mediated by the chromatinadaptor protein HP1, although their functional relevance to splice site selection remains to be determined (Figure 4). Finally, Cell 144, January 7, 2011 ª2011 Elsevier Inc. 21

Figure 4. Chromatin-Adaptor Complexes Several histone modification-binding chromatin proteins interact with splicing factors (Luco et al., 2010; Sims et al., 2007; Gunderson and Johnson, 2009; Piacentini et al., 2009; Loomis et al., 2009).

other combinations of interacting histone modifications, chromatin-binding proteins, and splicing factors may exist, possibly constituting a complex network of communication between chromatin and RNA. Genome-wide mapping of histone modifications and comparison to alternative splicing patterns should reveal such additional chromatin-splicing adaptor systems. An Integrated Model for Alternative Splice Site Selection Regulation of alternative splicing has long been thought to involve mostly cis-acting RNA elements. However, the picture becomes far more complex when one considers that RNA processing is coupled to transcription (Figure 5). The ultimate driving factor in determining splicing outcome is obviously the recruitment of splicing regulators to the target RNA (Barash et al., 2010). However, when and which factors are recruited is not only dependent on the combination of RNA motifs, the tissue- or developmental-specific pattern of expression of the splicing factors, or their posttranslational modifications as thought until now, but is also greatly influenced by chromatin architecture and histone

Figure 5. An Integrated Model for the Regulation of Alternative Splicing Alternative splicing patterns are determined by a combination of parameters including cis-acting RNA regulatory elements and RNA secondary structures (highlighted in orange) together with transcriptional and chromatin properties (highlighted in blue) that modulate the recruitment of splicing factors to the premRNA.

22 Cell 144, January 7, 2011 ª2011 Elsevier Inc.

modifications. The contribution of transcription regulators and histone modifiers to splicing regulation is likely 2-fold: On the one hand, they remodel and open chromatin for the recruitment of elongation factors that activate RNA Pol II elongation kinetics. On the other hand, the positioning of nucleosomes along exons, plus the enrichment in particular subsets of histone modifications, may modulate the recruitment of splicing regulators (Kolasinska-Zwierz et al., 2009; Schwartz et al., 2009; Spies et al., 2009). This could occur either through pausing of RNA Pol II, which favors the formation of the spliceosome by protein-protein interactions (de la Mata and Kornblihtt, 2006; Listerman et al., 2006), or through specific recruitment of splicing factors via adaptor complexes to weak RNA-binding sites (Luco et al., 2010; Sims et al., 2007; Gunderson and Johnson, 2009; Piacentini et al., 2009). These observations suggest an integrated model for the regulation of transcription and splicing in which the factors involved in the transcription control and chromatin maintenance also contribute to the recruitment and assembly of the spliceosome (Figure 5). The Role of Chromatin in Other RNA-Processing Events The role of chromatin and histone modifications likely goes beyond alternative splicing. There are some indications that other RNA-processing events are similarly modulated by chromatin and epigenetic modifications. In S. cerevisiae, the 30 region near the polyadenylation site is depleted of nucleosomes (Mavrich et al., 2008), and in human T cells nucleosome density dips noticeably within 200 nt of the canonical polyadenylation signal (Spies et al., 2009). Functional relevance for nucleosome density in polyadenylation is suggested by the fact that in genes containing multiple polyadenylation sites, the most highly used site preferentially falls within a nucleosomedepleted region (Spies et al., 2009). Bioinformatic analysis also suggests that nucleosome affinity is reduced near highly used polyadenylation sites but markedly increases just downstream of them (Spies et al., 2009). Altered nucleosome density may affect RNA Pol II elongation kinetics, which is known to affect polyadenylation, or the recruitment of the polyadenylation machinery to the nascent transcript (McCracken et al., 1997b). Further evidence for a role of chromatin in RNA processing comes from the surprising finding in Drosophila that several histone variants including the core histones H3.3A, H3.3B, H2a.V, and the H3-histone chaperone Asf1 are required for processing of the metazoan histone RNAs (Marzluff et al., 2008). Histone RNAs are unique in that they lack introns and are not polyadenylated and their 30 ends are processed in a single step by formation of a 30 stem-loop structure (Wagner et al., 2007). Loss of H2a.V, the functional ortholog of human H2A.X and H2A.Z,

Figure 6. The Epigenetics of Alternative Splicing The combination of histone modifications along a gene establishes and maintains tissue-specific transcription patterns (left panel), as well as heritable tissuespecific alternative splicing patterns (right panel).

results in readthrough and failure to process histone premRNAs. The reason appears to be failure to recruit histone RNA-processing factors to nascent histone RNAs (Marzluff et al., 2008). Although it is currently unknown how H2a.V affects processing factor recruitment, one intriguing possibility is that the altered chromatin structure at the histone loci interferes with the proper assembly of histone RNA-processing machinery, thus decreasing RNA-processing efficiency. Another RNA metabolic event that appears to be influenced by chromatin is RNA degradation. In an attempt to uncover novel functions for the S. pombe histone variant H2A.Z, Grewal and colleagues noted that absence of H2A.Z leads to an increase in antisense transcripts in 5%–8% of loci, whereas the level of sense transcripts is only modestly affected (Zofall et al., 2009). The accumulating antisense messages were mostly readthrough transcripts, and run-on experiments indicated that the elevated transcript levels are not due to increased transcription but are the consequence of failed degradation by the exosome (Zofall et al., 2009). In support of these conclusions, deletion of the exosome subunit rrp6 resulted in an antisense RNA profile similar to that observed upon loss of H2A.Z. These observations point to a role of chromatin structure in antisense RNA degradation by the exosome. The absence of H2A.Z may directly, or indirectly through loading of factors involved in maintaining chromatin structure, change chromatin conformation, which interferes with RNA Pol II progression, leading to RNA Pol II stalling and consequent degradation of transcripts. However this view is not fully consistent with the observation of readthrough transcripts generated in the absence of H2A.Z. Alternatively, H2A.Z may play a role in communicating to an RNA Pol II-associated exosome that readthrough transcripts have been generated. Intriguing questions are whether similar mechanisms are also at play in higher eukaryotes and whether H2A.Z, other histone variants, and/or chromatin structure in general play a more universal role in stability and controlled degradation of regular sense transcripts.

Chromatin as a Memory of Alternative Splicing Patterns Many alternative splicing events occur in a tissue- and/or cell type-specific fashion. How tissue- and cell type-specific alternative splicing patterns are established, propagated, and maintained is only poorly understood. One mechanism involves the tissue-specific expression of alternative splicing regulators, the classical example being the neuron-specific splicing factor NOVA-1/2, which regulates an extensive network of alternatively spliced target genes (Ule et al., 2005). However, given the scarcity of dedicated alternative splicing master controllers, such regulation is likely the exception rather than the rule. Cell- and tissue-specific alternative splicing patterns are obviously not determined solely by RNA-binding motifs, as they are present identically in all cell types. Although cell- and tissuespecific differences in expression of constitutive splicing factors have been reported (Hanamura et al., 1998), it is difficult to envision how global changes in abundance of general splicing factors in a cell type or tissue can account for the intricate regulation of individual exons, some of which need to be included, some skipped, and others requiring activation of cryptic splice sites, even within the same gene. Given the realization that chromatin structure and histone modifications can affect alternative splicing, it is attractive to speculate that, in analogy to the histone indexing mechanisms used to specify which genes are expressed and which ones are silenced, a histone-based system may also encode information that specifies alternative splicing patterns in cell types and tissues (Figure 6). Such a system may involve marking of chromatin stretches encoding alternatively spliced regions by histone modifications either to alter RNA Pol II elongation rate locally as the polymerase passes through or to recruit splicing factors via adaptor complexes. An advantage of such a histone-based alternative splicing regulatory system is that it would provide an epigenetic memory for splicing decisions that could be passed on during proliferation of a cell population and could be modified during differentiation without the requirement Cell 144, January 7, 2011 ª2011 Elsevier Inc. 23

to establish a new set of alternative splicing rules at each step of differentiation. Obviously, an epigenetic alternative splicing memory would still require the proper expression of splicing factors, a process that itself may be controlled by epigenetic mechanisms. Regardless of mechanisms, it appears that epigenetic regulation is not limited to controlling what regions of the genome are expressed, but also how they are spliced. Conclusions and Outlook The realization that chromatin and histone modifications contribute to RNA processing, particularly alternative premRNA splicing, has recently inspired many new avenues of investigation—but at the same time it raises many key questions. At this point, we do not even have a comprehensive view of how histone modifications relate to alternative splicing outcome. For that, histone modifications must be comprehensively mapped across the genome in as many cell types and tissues as possible and compared to genome-wide alternative splicing patterns. This information should be readily forthcoming from genomewide histone modification mapping projects and the parallel characterization of the transcriptome in those systems by deep sequencing. These approaches will also answer the fundamental question of whether histone modifications that have been implicated in alternative splicing regulation act alone or in a combinatorial fashion with other epigenetic marks. Systematic genome-wide studies should also resolve the issue of how extensive the regulation of alternative splicing by histone modifications is. Are all alternative splicing events sensitive to histone modifications, or is only a subset of exons affected? If so, what are their characteristics? An intriguing extension of these considerations is the possibility that noncoding RNAs might play a role in alternative splicing regulation (Kishore and Stamm, 2006; Khanna and Stamm, 2010). ncRNAs are now known to be involved in heterochromatin structure, and it is possible that some ncRNAs are specifically transcribed and associate with alternatively spliced regions of genes. Having established a role for histone modifications in alternative splicing, and given the intimate linkage between transcription and RNA processing, the question of whether splicing in turn also affects histone modifications must be asked. It is possible that in the same way histone modifications modulate recruitment of splicing factors, splicing regulators also modulate the recruitment of histone modifiers and chromatin remodelers to the nucleosomes regulating chromatin conformation in a feedback mechanism. In support of such crosstalk, inhibition of splicing abolishes transcription and splicing factors stimulate elongation (O’Keefe et al., 1994; Fong and Zhou, 2001; Lin et al., 2008), suggesting that transcription, chromatin, and splicing are intimately dependent on each other. Considering the still preliminary but tantalizing evidence that chromatin may also affect other RNA-processing events, it will be important to probe in more detail the effect of chromatin on 30 processing, RNA stability, and other RNA-processing steps. Finally, we must consider the possible physiological and pathological consequences and opportunities of histone-mediated RNA-processing effects. First and foremost, it will be important to determine whether histone modification effects on RNA processing are heritable and therefore truly epigenetic or whether 24 Cell 144, January 7, 2011 ª2011 Elsevier Inc.

they are merely transient modulators. To address this question, we will have to systematically analyze changes in RNA-processing patterns during differentiation and development and compare them to histone modification fingerprints of cells and tissues. We should also comprehensively probe for aberrant RNA processing in diseases caused by epigenetic defects. In addition, we may have to reconsider the expected effects of drugs targeting epigenetic mechanisms such as the clinically used histone deacetylase and DNA methyltransferase inhibitors, and we might want to think about designing new drugs targeting chromatin for the treatment of splicing diseases. Clearly the emerging role of epigenetics in RNA processing provides many new challenges, and even more opportunities. ACKNOWLEDGMENTS This work was supported by grants to A.R.K. from the Agencia Nacional de Promocio´n de Ciencia y Tecnologı´a of Argentina, the Universidad de Buenos Aires, and the European Alternative Splicing Network (EURASNET). I.E.S. is a recipient of a postdoctoral fellowships, and A.R.K. is a career investigator from the Consejo Nacional de Investigaciones Cientı´ficas y Te´cnicas of Argentina (CONICET). A.R.K. is an international research scholar of the Howard Hughes Medical Institute. Work in the Misteli laboratory is supported by the Intramural Research Program of the National Institutes of Health (NIH), NCI, Center for Cancer Research.

REFERENCES Aebi, M., and Weissmann, C. (1987). Precision and orderliness in splicing. Trends Genet. 3, 102–107. Allo´, M., Buggiano, V., Fededa, J.P., Petrillo, E., Schor, I., de la Mata, M., Agirre, E., Plass, M., Eyras, E., Abou Elela, S., et al. (2009). Control of alternative splicing through siRNA-mediated transcriptional gene silencing. Nat. Struct. Mol. Biol. 16, 717–724. Andersson, R., Enroth, S., Rada-Iglesias, A., Wadelius, C., and Komorowski, J. (2009). Nucleosomes are well positioned in exons and carry characteristic histone modifications. Genome Res. 19, 1732–1741. Attanasio, C., David, A., and Neerman-Arbez, M. (2003). Outcome of donor splice site mutations accounting for congenital afibrinogenemia reflects order of intron removal in the fibrinogen alpha gene (FGA). Blood 101, 1851–1856. Auboeuf, D., Honig, A., Berget, S.M., and O’Malley, B.W. (2002). Coordinate regulation of transcription and splicing by steroid receptor coregulators. Science 298, 416–419. Auboeuf, D., Dowhan, D.H., Kang, Y.K., Larkin, K., Lee, J.W., Berget, S.M., and O’Malley, B.W. (2004a). Differential recruitment of nuclear receptor coactivators may determine alternative RNA splice site choice in target genes. Proc. Natl. Acad. Sci. USA 101, 2270–2274. Auboeuf, D., Dowhan, D.H., Li, X., Larkin, K., Ko, L., Berget, S.M., and O’Malley, B.W. (2004b). CoAA, a nuclear receptor coactivator protein at the interface of transcriptional coactivation and RNA splicing. Mol. Cell. Biol. 24, 442–453. Barash, Y., Calarco, J.A., Gao, W.J., Pan, Q., Wang, X.C., Shai, O., Blencowe, B.J., and Frey, B.J. (2010). Deciphering the splicing code. Nature 465, 53–59. Batsche, E., Yaniv, M., and Muchardt, C. (2006). The human SWI/SNF subunit Brm is a regulator of alternative splicing. Nat. Struct. Mol. Biol. 13, 22–29. Bauren, G., and Wieslander, L. (1994). Splicing of Balbiani ring 1 gene premRNA occurs simultaneously with transcription. Cell 76, 183–192. Beyer, A.L., and Osheim, Y.N. (1988). Splice site selection, rate of splicing, and alternative splicing on nascent transcripts. Genes Dev. 2, 754–765. Bird, G., Zorio, D.A., and Bentley, D.L. (2004). RNA polymerase II carboxyterminal domain phosphorylation is required for cotranscriptional pre-mRNA splicing and 30 -end formation. Mol. Cell. Biol. 24, 8963–8969.

Brodsky, A.S., Meyer, C.A., Swinburne, I.A., Giles, H., Keenan, B.J., Liu, X.L.S., Fox, E.A., and Silver, P.A. (2005). Genomic mapping of RNA polymerase II reveals sites of co-transcriptional regulation in human cells. Genome Biol. 6, R64. Buratti, E., and Baralle, F.E. (2004). Influence of RNA secondary structure on the pre-mRNA splicing process. Mol. Cell. Biol. 24, 10505–10514. Caceres, J.F., and Kornblihtt, A.R. (2002). Alternative splicing: multiple control mechanisms and involvement in human disease. Trends Genet. 18, 186–193. Chasin, L.A. (2007). Searching for splicing motifs. Adv. Exp. Med. Biol. 623, 85–106. Cheng, D.H., Cote, J., Shaaban, S., and Bedford, M.T. (2007). The arginine methyltransferase CARM1 regulates the coupling of transcription and mRNA processing. Mol. Cell 25, 71–83. Chodavarapu, R.K., Feng, S., Bernatavichute, Y.V., Chen, P.Y., Stroud, H., Yu, Y., Hetzel, J.A., Kuo, F., Kim, J., Cokus, S.J., et al. (2010). Relationship between nucleosome positioning and DNA methylation. Nature 15, 388–392.

Han, S.P., Tang, Y.H., and Smith, R. (2010). Functional diversity of the hnRNPs: past, present and perspectives. Biochem. J. 430, 379–392. Hanamura, A., Caceres, J.F., Mayeda, A., Franza, B.R., and Krainer, A.R. (1998). Regulated tissue-specific expression of antagonistic pre-mRNA splicing factors. RNA 4, 430–444. Hicks, M.J., Yang, C.R., Kotlajich, M.V., and Hertel, K.J. (2006). Linking splicing to Pol II transcription stabilizes pre-mRNAs and influences splicing patterns. PLoS Biol. 4, e147. Hirose, Y., and Manley, J.L. (1998). RNA polymerase II is an essential mRNA polyadenylation factor. Nature 395, 93–96. Hodges, C., Bintu, L., Lubkowska, L., Kashlev, M., and Bustamante, C. (2009a). Nucleosomal fluctuations govern the transcription dynamics of RNA polymerase II. Science 325, 626–628.

Cooper, T.A., Wan, L., and Dreyfuss, G. (2009). RNA and disease. Cell 136, 777–793.

Hodges, E., Smith, A.D., Kendall, J., Xuan, Z.Y., Ravi, K., Rooks, M., Zhang, M.Q., Ye, K., Bhattacharjee, A., Brizuela, L., et al. (2009b). High definition profiling of mammalian DNA methylation by array capture and single molecule bisulfite sequencing. Genome Res. 19, 1593–1605.

Cramer, P., Pesce, C.G., Baralle, F.E., and Kornblihtt, A.R. (1997). Functional association between promoter structure and transcript alternative splicing. Proc. Natl. Acad. Sci. USA 94, 11456–11460.

Howe, K.J., Kane, C.M., and Ares, M., Jr. (2003). Perturbation of transcription elongation influences the fidelity of internal exon inclusion in Saccharomyces cerevisiae. RNA 9, 993–1006.

Cramer, P., Caceres, J.F., Cazalla, D., Kadener, S., Muro, A.F., Baralle, F.E., and Kornblihtt, A.R. (1999). Coupling of transcription with alternative splicing: RNA pol II promoters modulate SF2/ASF and 9G8 effects on an exonic splicing enhancer. Mol. Cell 4, 251–258.

Kadener, S., Cramer, P., Nogues, G., Cazalla, D., de la Mata, M., Fededa, J.P., Werbajh, S.E., Srebrow, A., and Kornblihtt, A.R. (2001). Antagonistic effects of T-Ag and VP16 reveal a role for RNA pol II elongation on alternative splicing. EMBO J. 20, 5759–5768.

Croft, L., Schandorff, S., Clark, F., Burrage, K., Arctander, P., and Mattick, J.S. (2000). ISIS, the intron information system, reveals the high frequency of alternative splicing in the human genome. Nat. Genet. 24, 340–341.

Kessler, O., Jiang, Y., and Chasin, L.A. (1993). Order of intron removal during splicing of endogenous adenine phosphoribosyltransferase and dihydrofolate reductase pre-mRNA. Mol. Cell. Biol. 13, 6211–6222.

Das, R., Dufu, K., Romney, B., Feldt, M., Elenko, M., and Reed, R. (2006). Functional coupling of RNAP II transcription to spliceosome assembly. Genes Dev. 20, 1100–1109.

Khanna, A., and Stamm, S. (2010). Regulation of alternative splicing by short non-coding nuclear RNAs. RNA Biol. 7, 480–485.

Das, R., Yu, J., Zhang, Z., Gygi, M.P., Krainer, A.R., Gygi, S.P., and Reed, R. (2007). SR proteins function in coupling RNAP II transcription to pre-mRNA splicing. Mol. Cell 26, 867–881. de la Mata, M., Alonso, C.R., Kadener, S., Fededa, J.P., Blaustein, M., Pelisch, F., Cramer, P., Bentley, D., and Kornblihtt, A.R. (2003). A slow RNA polymerase II affects alternative splicing in vivo. Mol. Cell 12, 525–532. de la Mata, M., and Kornblihtt, A.R. (2006). RNA polymerase II C-terminal domain mediates regulation of alternative splicing by SRp20. Nat. Struct. Mol. Biol. 13, 973–980. de la Mata, M., Lafaille, C., and Kornblihtt, A.R. (2010). First come, first served revisited: factors affecting the same alternative splicing event have different effects on the relative rates of intron removal. RNA 16, 904–912. Dhami, P., Saffrey, P., Bruce, A.W., Dillson, S.C., Chiang, K., Bonhoure, N., Koch, C.M., Bye, J., James, K., Foad, N.S., et al. (2010). Complex exon-intron marking by histone modifications is not determined solely by nucleosome distribution. PLoS ONE 5, e12339. Dower, K., and Rosbash, M. (2002). T7 RNA polymerase-directed transcripts are processed in yeast and link 30 end formation to mRNA nuclear export. RNA 8, 686–697. Eperon, L.P., Graham, I.R., Griffiths, A.D., and Eperon, I.C. (1988). Effects of RNA secondary structure on alternative splicing of pre-mRNA: is folding limited to a region behind the transcribing RNA polymerase? Cell 54, 393–401. Fong, Y.W., and Zhou, Q. (2001). Stimulatory effect of splicing factors on transcriptional elongation. Nature 414, 929–933. Fox-Walsh, K.L., and Hertel, K.J. (2009). Splice-site pairing is an intrinsically high fidelity process. Proc. Natl. Acad. Sci. USA 106, 1766–1771. Gornemann, J., Kotovic, K.M., Hujer, K., and Neugebauer, K.M. (2005). Cotranscriptional spliceosome assembly occurs in a stepwise fashion and requires the cap binding complex. Mol. Cell 19, 53–63. Gunderson, F.Q., and Johnson, T.L. (2009). Acetylation by the transcriptional coactivator Gcn5 plays a novel role in co-transcriptional spliceosome assembly. PLoS Genet. 5, e1000682.

Kishore, S., and Stamm, S. (2006). The snoRNA HBII-52 regulates alternative splicing of the serotonin receptor 2C. Science 311, 230–232. Kolasinska-Zwierz, P., Down, T., Latorre, I., Liu, T., Liu, X.S., and Ahringer, J. (2009). Differential chromatin marking of introns and expressed exons by H3K36me3. Nat. Genet. 41, 376–381. Kornblihtt, A.R. (2005). Promoter usage and alternative splicing. Curr. Opin. Cell Biol. 17, 262–268. Kotovic, K.M., Lockshon, D., Boric, L., and Neugebauer, K.M. (2003). Cotranscriptional recruitment of the U1 snRNP to intron-containing genes in yeast. Mol. Cell. Biol. 23, 5768–5779. Lacadie, S.A., and Rosbash, M. (2005). Cotranscriptional spliceosome assembly dynamics and the role of U1 snRNA:50 ss base pairing in yeast. Mol. Cell 19, 65–75. Lazarev, D., and Manley, J.L. (2007). Concurrent splicing and transcription are not sufficient to enhance splicing efficiency. RNA 13, 1546–1557. LeMaire, M.F., and Thummel, C.S. (1990). Splicing precedes polyadenylation during Drosophila E74A transcription. Mol. Cell. Biol. 10, 6059–6063. Li, S.S., and Shogren-Knaak, M.A. (2009). The Gcn5 bromodomain of the SAGA complex facilitates cooperative and cross-tail acetylation of nucleosomes. J. Biol. Chem. 284, 9411–9417. Lin, S., Coutinho-Mansfield, G., Wang, D., Pandit, S., and Fu, X.D. (2008). The splicing factor SC35 has an active role in transcriptional elongation. Nat. Struct. Mol. Biol. 15, 819–826. Listerman, I., Sapra, A.K., and Neugebauer, K.M. (2006). Cotranscriptional coupling of splicing factor recruitment and precursor messenger RNA splicing in mammalian cells. Nat. Struct. Mol. Biol. 13, 815–822. Long, J.C., and Caceres, J.F. (2009). The SR protein family of splicing factors: master regulators of gene expression. Biochem. J. 417, 15–27. Loomis, R.J., Naoe, Y., Parker, J.B., Savic, V., Bozovsky, M.R., Macfarlan, T., Manley, J.L., and Chakravarti, D. (2009). Chromatin binding of SRp20 and ASF/SF2 and dissociation from mitotic chromosomes is modulated by histone H3 serine 10 phosphorylation. Mol. Cell 33, 450–461.

Cell 144, January 7, 2011 ª2011 Elsevier Inc. 25

Luco, R.F., Pan, Q., Tominaga, K., Blencowe, B.J., Pereira-Smith, O.M., and Misteli, T. (2010). Regulation of alternative splicing by histone modifications. Science 327, 996–1000. Maniatis, T., and Reed, R. (2002). An extensive network of coupling among gene expression machines. Nature 416, 499–506.

euchromatic gene expression through RNA transcript association and interaction with hnRNPs in Drosophila. PLoS Genet. 5, e1000670. Ponts, N., Yang, J.F., Chung, D.W.D., Prudhomme, J., Girke, T., Horrocks, P., and Le Roch, K.G. (2008). Deciphering the ubiquitin-mediated pathway in apicomplexan parasites: A potential strategy to interfere with parasite virulence. PLoS ONE 3, e2386.

Martinez, E., Palhan, V.B., Tjernberg, A., Lymar, E.S., Gamper, A.M., Kundu, T.K., Chait, B.T., and Roeder, R.G. (2001). Human STAGA complex is a chromatin-acetylating transcription coactivator that interacts with pre-mRNA splicing and DNA damage-binding factors in vivo. Mol. Cell. Biol. 21, 6782–6795.

Pray-Grant, M.G., Daniel, J.A., Schieltz, D., Yates, J.R., 3rd, and Grant, P.A. (2005). Chd1 chromodomain links histone H3 methylation with SAGA- and SLIK-dependent acetylation. Nature 433, 434–438.

Marzluff, W.F., Wagner, E.J., and Duronio, R.J. (2008). Metabolism and regulation of canonical histone mRNAs: life without a poly(A) tail. Nat. Rev. Genet. 9, 843–854.

Roberts, G.C., Gooding, C., Mak, H.Y., Proudfoot, N.J., and Smith, C.W. (1998). Co-transcriptional commitment to alternative splice site selection. Nucleic Acids Res. 26, 5568–5572.

Mavrich, T.N., Ioshikhes, I.P., Venters, B.J., Jiang, C., Tomsho, L.P., Qi, J., Schuster, S.C., Albert, I., and Pugh, B.F. (2008). A barrier nucleosome model for statistical positioning of nucleosomes throughout the yeast genome. Genome Res. 18, 1073–1083.

Sapra, A.K., Anko, M.L., Grishina, I., Lorenz, M., Pabis, M., Poser, I., Rollins, J., Weiland, E.M., and Neugebauer, K.M. (2009). SR protein family members display diverse activities in the formation of nascent and mature mRNPs in vivo. Mol. Cell 34, 179–190.

McCracken, S., Fong, N., Rosonina, E., Yankulov, K., Brothers, G., Siderovski, D., Hessel, A., Foster, S., Shuman, S., and Bentley, D.L. (1997a). 50 -Capping enzymes are targeted to pre-mRNA by binding to the phosphorylated carboxy-terminal domain of RNA polymerase II. Genes Dev. 11, 3306–3318.

Schor, I.E., Rascovan, N., Pelisch, F., Allo, M., and Kornblihtt, A.R. (2009). Neuronal cell depolarization induces intragenic chromatin modifications affecting NCAM alternative splicing. Proc. Natl. Acad. Sci. USA 106, 4325–4330.

McCracken, S., Fong, N., Yankulov, K., Ballantyne, S., Pan, G., Greenblatt, J., Patterson, S.D., Wickens, M., and Bentley, D.L. (1997b). The C-terminal domain of RNA polymerase II couples mRNA processing to transcription. Nature 385, 357–361. McCracken, S., Rosonina, E., Fong, N., Sikes, M., Beyer, A., O’Hare, K., Shuman, S., and Bentley, D. (1998). Role of RNA polymerase II carboxy-terminal domain in coordinating transcription with RNA processing. Cold Spring Harb. Symp. Quant. Biol. 63, 301–309. Meissner, A., Mikkelsen, T.S., Gu, H.C., Wernig, M., Hanna, J., Sivachenko, A., Zhang, X.L., Bernstein, B.E., Nusbaum, C., Jaffe, D.B., et al. (2008). Genomescale DNA methylation maps of pluripotent and differentiated cells. Nature 454, 766–770. Misteli, T., and Spector, D.L. (1999). RNA polymerase II targets pre-mRNA splicing factors to transcription sites in vivo. Mol. Cell 3, 697–705. Moore, M.J., and Proudfoot, N.J. (2009). Pre-mRNA processing reaches back to transcription and ahead to translation. Cell 136, 688–700. Mun˜oz, M.J., Perez Santangelo, M.S., Paronetto, M.P., de la Mata, M., Pelisch, F., Boireau, S., Glover-Cutter, K., Ben-Dov, C., Blaustein, M., Lozano, J.J., et al. (2009). DNA damage regulates alternative splicing through inhibition of RNA polymerase II elongation. Cell 137, 708–720. Nahkuri, S., Taft, R.J., and Mattick, J.S. (2009). Nucleosomes are preferentially positioned at exons in somatic and sperm cells. Cell Cycle 8, 3420–3424. Nogue´s, G., Kadener, S., Cramer, P., Bentley, D., and Kornblihtt, A.R. (2002). Transcriptional activators differ in their abilities to control alternative splicing. J. Biol. Chem. 277, 43110–43114. Ohkura, N., Takahashi, M., Yaguchi, H., Nagamura, Y., and Tsukada, T. (2005). Coactivator-associated arginine methyltransferase 1, CARM1, affects premRNA splicing in an isoform-specific manner. J. Biol. Chem. 280, 28927–28935. O’Keefe, R.T., Mayeda, A., Sadowski, C.L., Krainer, A.R., and Specotr, D.L. (1994). Disruption of pre-mRNA splicing in vivo results in reorganization of splicing factors. J. Cell Biol. 124, 249–260. Pagani, F., Stuani, C., Zuccato, E., Kornblihtt, A.R., and Baralle, F.E. (2003). Promoter architecture modulates CFTR exon 9 skipping. J. Biol. Chem. 278, 1511–1517. Pan, Q., Shai, O., Lee, L.J., Frey, B.J., and Blencowe, B.J. (2008). Deep surveying of alternative splicing complexity in the human transcriptome by high-throughput sequencing. Nat. Genet. 40, 1413–1415. Pandya-Jones, A., and Black, D.L. (2009). Co-transcriptional splicing of constitutive and alternative exons. RNA 15, 1896–1908. Piacentini, L., Fanti, L., Negri, R., Del Vescovo, V., Fatica, A., Altieri, F., and Pimpinelli, S. (2009). Heterochromatin protein 1 (HP1a) positively regulates

26 Cell 144, January 7, 2011 ª2011 Elsevier Inc.

Schwartz, S., Meshorer, E., and Ast, G. (2009). Chromatin organization marks exon-intron structure. Nat. Struct. Mol. Biol. 16, 990–995. Sims, R.J., Millhouse, S., Chen, C.F., Lewis, B.A., Erdjument-Bromage, H., Tempst, P., Manley, J.L., and Reinberg, D. (2007). Recognition of trimethylated histone h3 lysine 4 facilitates the recruitment of transcription postinitiation factors and pre-mRNA splicing. Mol. Cell 28, 665–676. Sisodia, S.S., Sollner-Webb, B., and Cleveland, D.W. (1987). Specificity of RNA maturation pathways: RNAs transcribed by RNA polymerase III are not substrates for splicing or polyadenylation. Mol. Cell. Biol. 7, 3602–3612. Smale, S.T., and Tjian, R. (1985). Transcription of herpes simplex virus tk sequences under the control of wild-type and mutant human RNA polymerase I promoters. Mol. Cell. Biol. 5, 352–362. Spies, N., Nielsen, C.B., Padgett, R.A., and Burge, C.B. (2009). Biased chromatin signatures around polyadenylation sites and exons. Mol. Cell 36, 245–254. Tardiff, D.F., Lacadie, S.A., and Rosbash, M. (2006). A genome-wide analysis indicates that yeast pre-mRNA splicing is predominantly posttranscriptional. Mol. Cell 24, 917–929. Tennyson, C.N., Klamut, H.J., and Worton, R.G. (1995). The human dystrophin gene requires 16 hours to be transcribed and is cotranscriptionally spliced. Nat. Genet. 9, 184–190. Tilgner, H., Nikolaou, C., Althammer, S., Sammeth, M., Beato, M., Valcarcel, J., and Guigo, R. (2009). Nucleosome positioning as a determinant of exon recognition. Nat. Struct. Mol. Biol. 16, 996–1001. Tyagi, A., Ryme, J., Brodin, D., Farrants, A.K.O., and Visa, N. (2009). SWI/SNF associates with nascent pre-mRNPs and regulates alternative pre-mRNA processing. PLoS Genet. 5, e1000470. Ule, J., Ule, A., Spencer, J., Williams, A., Hu, J.S., Cline, M., Wang, H., Clark, T., Fraser, C., Ruggiu, M., et al. (2005). Nova regulates brain-specific splicing to shape the synapse. Nat. Genet. 37, 844–852. Wagner, E.J., Burch, B.D., Godfrey, A.C., Salzler, H.R., Duronio, R.J., and Marzluff, W.F. (2007). A genome-wide RNA interference screen reveals that variant histones, are necessary for replication-dependent histone pre-mRNA processing. Mol. Cell 28, 692–699. Wang, E.T., Sandberg, R., Luo, S., Khrebtukova, I., Zhang, L., Mayr, C., Kingsmore, S.F., Schroth, G.P., and Burge, C.B. (2008). Alternative isoform regulation in human tissue transcriptomes. Nature 456, 470–476. Xiao, T.J., Hall, H., Kizer, K.O., Shibata, Y., Hall, M.C., Borchers, C.H., and Strahl, B.D. (2003). Phosphorylation of RNA polymerase II CTD regulates H3 methylation in yeast. Genes Dev. 17, 654–663. Zofall, M., Fischer, T., Zhang, K., Zhou, M., Cui, B.W., Veenstra, T.D., and Grewal, S.I.S. (2009). Histone H2A.Z cooperates with RNAi and heterochromatin factors to suppress antisense RNAs. Nature 461, 419–422.

Massive Genomic Rearrangement Acquired in a Single Catastrophic Event during Cancer Development Philip J. Stephens,1 Chris D. Greenman,1 Beiyuan Fu,1 Fengtang Yang,1 Graham R. Bignell,1 Laura J. Mudie,1 Erin D. Pleasance,1 King Wai Lau,1 David Beare,1 Lucy A. Stebbings,1 Stuart McLaren,1 Meng-Lay Lin,1 David J. McBride,1 Ignacio Varela,1 Serena Nik-Zainal,1 Catherine Leroy,1 Mingming Jia,1 Andrew Menzies,1 Adam P. Butler,1 Jon W. Teague,1 Michael A. Quail,1 John Burton,1 Harold Swerdlow,1 Nigel P. Carter,1 Laura A. Morsberger,2 Christine Iacobuzio-Donahue,2 George A. Follows,3 Anthony R. Green,3,4 Adrienne M. Flanagan,5,6 Michael R. Stratton,1,7 P. Andrew Futreal,1 and Peter J. Campbell1,3,4,* 1Cancer

Genome Project, Wellcome Trust Sanger Institute, Hinxton, Cambridge CB10 1SA, UK of Pathology and Oncology, Johns Hopkins Medical Institutions, Baltimore, MD 21287, USA 3Department of Haematology, Addenbrooke’s Hospital, Cambridge CB2 0QQ, UK 4Department of Haematology, University of Cambridge, Cambridge CB2 0XY, UK 5Cancer Institute, University College London, London WC1E 6BT, UK 6Royal National Orthopaedic Hospital, Middlesex HA7 4LP, UK 7Institute for Cancer Research, Sutton, Surrey SM2 5NG, UK *Correspondence: [email protected] DOI 10.1016/j.cell.2010.11.055 2Departments

SUMMARY

Cancer is driven by somatically acquired point mutations and chromosomal rearrangements, conventionally thought to accumulate gradually over time. Using next-generation sequencing, we characterize a phenomenon, which we term chromothripsis, whereby tens to hundreds of genomic rearrangements occur in a one-off cellular crisis. Rearrangements involving one or a few chromosomes crisscross back and forth across involved regions, generating frequent oscillations between two copy number states. These genomic hallmarks are highly improbable if rearrangements accumulate over time and instead imply that nearly all occur during a single cellular catastrophe. The stamp of chromothripsis can be seen in at least 2%–3% of all cancers, across many subtypes, and is present in 25% of bone cancers. We find that one, or indeed more than one, cancer-causing lesion can emerge out of the genomic crisis. This phenomenon has important implications for the origins of genomic remodeling and temporal emergence of cancer. INTRODUCTION The textbook model of cancer development is of progression through a series of increasingly disordered clinical and pathological phases (Stratton et al., 2009). For example, invasive colorectal cancer often emerges from an antecedent benign adenomatous polyp; cervical cancer proceeds through intraepithelial neoplasia before breaching the basement membrane;

multiple myeloma frequently develops in individuals with a history of benign monoclonal plasma cell proliferation. Biologically, such stepwise clinical progression is underpinned by successive waves of clonal expansion as cells acquire the multiple genetic changes required for a fully malignant phenotype. Mutations are essentially random, occurring as independent events throughout the lifespan of an individual, potentially accelerated by exogenous carcinogens or DNA repair defects. Genetic variation generates phenotypic variation across the cells of an organ system, which are then subject to clonal selection through Darwinian competition. Variants that enhance a cell’s evolutionary fitness, so-called driver mutations, promote outgrowth of that clone and progression toward cancer (Stratton et al., 2009). The prevailing dogma of cancer evolution is therefore one of ‘‘gradualism’’ in which acquisition of driver mutations occurs cumulatively over years to decades, resulting in incremental progression through increasingly malignant phenotypes (Jones et al., 2008). There are, however, examples in which a more ‘‘punctuated equilibrium’’ evolutionary model may apply to development of cancers. Genome-wide telomere attrition in somatic cells, for example, may generate naked DNA ends that act as a nidus for on-going genomic rearrangement (Bardeesy and DePinho, 2002; O’Hagan et al., 2002; Sahin and Depinho, 2010). End-toend chromosome fusions resulting from telomere loss can lead to spiraling cycles of dsDNA breakage, aberrant repair and further chromosomal damage in both daughter cells (Artandi et al., 2000; Gisselsson et al., 2001). Iteration of this breakage and repair process can lead to extensive genomic remodeling in multiple competing subclones in only a few cell cycles (Bignell et al., 2007). Under these scenarios, bursts of somatic mutation may accrue in relatively short periods of chronological time. Such genomic rearrangements can drive the development of cancer through several mechanisms. They may result in copy Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc. 27

number changes, including deletion of tumor suppressor genes and increased copy number (amplification) of genes promoting malignant cellular processes. In addition, chromosomal rearrangements can juxtapose portions of coding sequence from two genes in the same orientation, leading to oncogenic fusion genes, or bring together an intact gene with the regulatory machinery of another gene, causing dysregulated gene expression. Here, we describe multiple cancer samples in which tens to hundreds of genomic rearrangements have been acquired in a single catastrophic event, a phenomenon we have termed chromothripsis (Greek, chromos for chromosome; thripsis, shattering into pieces). We characterize the genomic hallmarks of this process, its frequency across diverse cancers and how such cataclysmic genome disruption can promote the development of cancer. RESULTS Localized Genomic Rearrangement in a Patient with Chronic Lymphocytic Leukemia Advances in DNA sequencing have made it possible to identify the majority of somatically acquired genetic variants in cancer samples on a genome-wide basis (Ding et al., 2010; Mardis et al., 2009; Pleasance et al., 2010a, 2010b; Shah et al., 2009). In particular, paired-end sequencing allows discovery of genomic rearrangements (Campbell et al., 2008, 2010; Stephens et al., 2009), through sequencing both ends of 50–100 million genomic DNA fragments per sample. Alignment of the pairedend reads to the reference genome enables identification of putative genomic rearrangements. In a rearrangement screen of 10 patients with chronic lymphocytic leukemia (CLL), we identified one patient who had 42 somatically acquired genomic rearrangements involving the long arm of chromosome 4 (Figures 1A and 1B and Table S1 available online). The positions of these rearrangements relative to one another and to copy number changes on chromosome 4q reveal some striking patterns. First, the rearrangements show geographic localization within the genome. Apart from a separate 13q deletion in this patient, all rearrangements are confined to chromosome 4q and focal points on chromosomes 1, 12, and 15 (Figure 1C). This is different to the patterns of genomic instability we have typically seen in breast, lung, or pancreatic cancer where rearrangements tend to be either scattered genome-wide or, if localized, are associated with substantial genomic amplification (Campbell et al., 2008; Pleasance et al., 2010b; Stephens et al., 2009). Second, the copy number profile across the chromosome arm shows many positions at which copy number changes, but these changes alternate between just two states, namely one or two copies. Analysis of allelic ratios at germline single nucleotide polymorphism (SNP) positions on chromosome 4q demonstrated that regions of copy number 1 show loss of heterozygosity, but regions of copy number 2 retain heterozygosity (data not shown). Third, the many regions of copy number 1 are not caused by simple deletions. Instead, a series of complex rearrangements spanning the involved region generate the copy number changes, as can be seen by the distribution of rearrangements falling at 28 Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc.

change-points in copy number (Figure 1A). These have both inverted and noninverted orientation, with all four orientations of intrachromosomal breakpoints represented in approximately even numbers: deletion-type (8 rearrangements), tandem duplication-type (9), head-to-head inverted (6), and tail-to-tail inverted (10). Fourth, there is pronounced clustering of breakpoints across the chromosome arm with, for example, seven rearrangements involving the 30 kb region between 77.013 Mb and 77.043 Mb, and six rearrangements in the 25 kb between 170.620 Mb and 170.645 Mb. Fifth, although the locations of DNA breaks show clustering, the two conjoined fragments of chromosome at each breakpoint are not geographically proximate. That is, there are as many rearrangements joining regions of the chromosome normally separated by tens of megabases in the germline as there are junctions between close-by regions. Sixth, there are nine rearrangements joining the long arm of chromosome 4 to other chromosomes—breakpoints on these partner chromosomes also show clustering (Figure 1C). The sample analyzed was collected from a 62-year-old woman with CLL who had not previously received treatment. Her subsequent clinical course showed rapid deterioration, and she was treated with alemtuzumab, but unfortunately, she relapsed quickly. To assess whether the abnormalities seen in the pretreatment sample persisted in the relapsing cells or indeed showed further evolution, we sequenced a relapse specimen collected 31 months after the initial sample. All rearrangements present in the pretreatment sample were present in the later sample (Figures 1B–1D), and the striking copy number profile persisted. Furthermore, there were no new genomic rearrangements, suggesting that the process generating this complex regional remodeling had resolved before the patient was first diagnosed. Complex Rearrangement of Single Chromosomes Is Seen in At Least 2%–3% of All Cancers To assess whether the unusual genomic landscape observed in the patient with CLL could be seen in other cancer samples, we analyzed high-resolution copy number profiles of 746 cancer cell lines obtained using SNP arrays (Bignell et al., 2010). Of these, 96 cell lines have at least one chromosome with >50 positions at which copy number changes (Figure S1A), many of which are caused by amplicons or other complex clusters of rearrangements. Notably, 18/746 (2.4%; 95% confidence interval, 1.5%– 3.9%) cell lines have copy number profiles similar to that seen in the CLL patient, with frequent copy number changes confined to localized genomic regions rapidly alternating between one, two, or occasionally three different states (Figures S1B–S1T). Copy number changes could involve the entire chromosome (for example, SNU-C1, Figure S1G), a whole arm of a chromosome (SW982, Figure S1H), the telomeric portion of a chromosome (C32, Figure S1C), or an interstitial region of a chromosome (A172, Figure S1D). The pattern was seen in many different tumor types, including melanoma (4 cell lines), small cell lung cancer (3 cell lines), glioma (3 cell lines), hematological malignancies (2 cell lines), nonsmall cell lung cancer (1 cell line), synovial sarcoma (1 cell line), and esophageal (1 cell line), colorectal (1 cell line), renal (1 cell line), and thyroid (1 cell line) cancers. Furthermore, in segmented SNP array data from 2792 cancers,

A

Deletion-type Tandem dup-type Interchromosomal

PD3172a (CLL): Chr 4

Tail-to-tail inverted Head-to-head inverted

Copy number

Non-inverted orientation 3 2 1 0

Inverted orientation

80 100 120 Genomic location (Mb)

C

Initial presentation

D

9

160

10

ic at m

180

11

12

T1 T2 N

ar Se te q fa ct So m at ic So m at ic

T1 T2 N

e lin rm

at m

T1 T2 N

e

ic

e So

Ge

rm

at m

T1 T2 N

lin

ic

ic So

Rearrangements 1-12 6 7 8

ic

5

T1 T2 N

at m So

Ge

rm

lin

e

T1 T2 N

4

So

3

lin

2

Ge

1

140

at

B

m

60

So

40

rm

20

Ge

0

Relapse

Figure 1. Clustered Rearrangements on Chromosome 4q in a Patient with Chronic Lymphocytic Leukemia (A) Copy number between 70 Mb and 170 Mb of the chromosome oscillates between a copy number of 1 and 2, demarcated by back-and-forth intrachromosomal rearrangements of all four possible orientations, as well as several interchromosomal rearrangements. (B) PCR gel of 12 putative genomic rearrangements identified by sequencing. PCR across the breakpoint is performed for each rearrangement on tumor DNA for samples taken at initial presentation (T1) and relapse (T2) as well as germline DNA (N). (C) Genome-wide profile of rearrangements in a sample taken before chemotherapy. Chromosomes range round the outside of the circle, copy number changes are shown by the blue line in the inner ring, and somatically acquired genomic rearrangements are shown as arcs linking the two relevant genomic points. (D) Genome-wide profile of rearrangements from the same patient 31 months later, at relapse after therapy.

of which 80% were primary tumors (Beroukhim et al., 2010), we find evidence for chromothripsis in a similar proportion of cases (Figure S1T).

We selected four of these cell lines for further genomic analysis with massively parallel paired-end sequencing for rearrangements and cytogenetic studies: SNU-C1, 8505C, TK10, and Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc. 29

SCLC-21H (described later). In SNU-C1, derived from a colorectal cancer, we identified 239 rearrangements involving chromosome 15 (Figure 2A and Table S2). From 8505C, a thyroid cancer line, we mapped 77 rearrangements involving the short arm of chromosome 9 (Figure 2B and Table S2), and for TK10, a renal cancer, 55 rearrangements involving chromosome 5 (Figure 2C and Table S2). The distinctive genomic configuration observed in the CLL patient is stamped on these three cell lines. Striking geographic localization of rearrangements is evident in these samples. Although a few rearrangements were observed elsewhere in the genome (Figure S2), these are generally straightforward events such as deletions or tandem duplications and do not intersect with the regions of massive disruption shown in Figure 2. The localization is especially evident in 8505C (Figure 2B), in which rearrangements only involve the telomeric portion of chromosome 9p with sparing of the most centromeric bands of 9p and all 9q. As in the CLL patient, copy number oscillates rapidly between two states, with the lower copy number state showing loss of heterozygosity (LOH) and the higher copy number state retaining heterozygosity. One question that arises is whether the rearrangements are all found on a single parental copy of the chromosome or whether both copies are involved. We therefore performed spectral karyotyping on the three cell lines (Figure 3A and Figure S3). TK10, a hyperdiploid line, carries six copies of chromosome 5. Consistent with the observed copy number profile alternating between states of copy number 4 with LOH and copy number 6 with heterozygosity, the karyotype showed four grossly normal copies of chromosome 5 and two smaller derivative chromosomes. Similarly, in 8505C, two copies of chromosome 9 showed distinctly foreshortened p arms alongside two cytogenetically normal chromosomes. None of the three karyotypes indicated translocations involving the respective derivative chromosomes, confirming the impression from the paired-end sequencing data that the genomic remodeling of these regions was entirely intrachromosomal. Cytogenetic changes were consistently seen across all cells examined. The spectral karyotypes suggest that the rearrangements involve a single parental copy of the chromosome. To demonstrate this further, we designed FISH probes to five widely dispersed regions of chromosome 5 at copy number 6 in TK10 (Figure 3B). From the paired-end sequencing, we predicted that the two regions at 6 Mb and 172 Mb would be joined by a head-to-head inverted rearrangement, and the three regions at 32 Mb, 66 Mb, and 150 Mb would be joined by another head-to-head inverted rearrangement and a tandem duplication-type rearrangement. These FISH probes, labeled with different dyes, were hybridized to TK10 cells (Figure 3C). As expected, there were four copies of chromosome 5 per cell showing the correct genomic orientation and distribution of the five probes. In addition, each cell carried two copies of a derivative 5 chromosome in which all five probes were closely juxtaposed, as predicted by the sequencing data. These patterns were seen identically across all cells examined. Taken together, these data suggest that at least 2%–3% of all cancers show evidence for massive remodeling of a single chromosome, involving tens to hundreds of genomic rearrange30 Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc.

ments. The consistency of cytogenetic findings across the many cells examined implies that the clustering of genomic breakpoints cannot be explained by multiple, parallel rearrangements in different subclones. In the lines studied here, the genomic remodeling occurred when there were just the two parental copies of the relevant chromosome, preceding chromosomal duplication events. This explains why copy number states alternate between heterozygous and LOH and why more than one copy of the derivative chromosome is present. Chromothripsis Is Particularly Common in Bone Cancers and Can Involve More Than One Chromosome Alongside the rearrangement screen in CLL, we performed rearrangement screens in primary tumor samples from 20 patients with bone cancer, including 9 with osteosarcoma and 11 with chordoma, a rare type of cancer arising in the spinal column. Strikingly, five of these patients (25%; 95% confidence interval, 10%–49%), three with osteosarcoma and two with chordoma, also show large numbers of clustered rearrangements with the hallmarks of chromothripsis. In four of these five bone tumors, rearrangements affect localized regions of several chromosomes (Figure 4, Figure S4, Table S3, and Table S4). For example, we identified 147 somatically acquired genomic rearrangements in a chordoma sample, PD3808a, involving and linking together well-circumscribed regions of chromosomes 3q, 4q, 7q, 8p, and 9p (Figure 4A). Analogous to chromothripsis involving single chromosomes, copy number in each of these chromosomal regions cycles between two different states with retention of heterozygosity in the higher copy number state. Of the 147 rearrangements, 49 are intrachromosomal and show the same back-and-forth mixture of inverted and noninverted rearrangements described above. The numerous interchromosomal rearrangements link the various disrupted regions together, implying that the resulting genomic structure is a complex medley of fragments from different chromosomes jumbled together. In samples from three patients with osteosarcoma, PD3786a (Figure 4B), PD3791a (Figure S4A), and PD3799a (Figure S4B), we identified 88, 86 and 24 rearrangements respectively with similar overall patterns of copy number change and rearrangement. PD3807a, another chordoma sample, also had 38 rearrangements interlinking well-defined regions of four chromosomes (Figure 4C). Clinically, the patients ranged in age from 9 to 64 years and four of the samples were from resections of treatment-naive primary tumors, whereas one of the patients (PD3786a) had previously received neoadjuvant chemotherapy. In 1 of 13 pancreatic cancers we previously sequenced (Campbell et al., 2010), we identified 41 rearrangements involving chromosomes 1, 4, 10, and 14 with the hallmarks of chromothripsis (Figure S4H), suggesting that involvement of multiple chromosomes by this process is not restricted to bone tumors. The Vast Majority of Chromothripsis Rearrangements Occur in a Single Catastrophic Event There are two potential models for how such complex restructuring of a chromosome could develop. Under the progressive rearrangement model, the rearrangements occur sequentially and independently of one another over many cell cycles, leading

A

Tail-to-tail inverted Head-to-head inverted

Deletion-type Tandem dup-type

SNU-C1 (colorectal): Chr 15 Non-inverted orientation

4 Copy number 2 0 Allelic 1 ratio 0 Inverted orientation 15

B

20

25

30

35

40

45

50 55 60 65 70 Genomic location (Mb)

75

80

85

90

95 100

8505C (thyroid): Chr 9

4 2 0 1 0

0

C

10

20

30

40

50

60

70

80

90

100 110 120 130 140

TK10 (renal cell): Chr 5

6 3 0 1 0

0

20

40

60

80

100

120

140

160

180

Figure 2. Rearrangement Screens in Three Cancer Cell Lines Showing Evidence for Chromothripsis Copy number profiles derive from SNP6 microarray data and are shown as the upper panel of points for each cell line. Allelic ratios for each SNP are shown in the lower panel of dots: homozygous SNPs cluster at allelic ratios near 0 or 1, heterozygous SNPs cluster around 0.5. Intrachromosomal rearrangements of all four possible orientations are shown, with deletion-type events as blue lines, tandem duplication-type in red, tail-to-tail inverted rearrangements in green and head-to-head inverted rearrangements in yellow. (A) SNU-C1, a cell line from a colorectal cancer, carries 239 rearrangements involving chromosome 15. (B) 8505C, a thyroid cancer cell line, has 77 rearrangements involving chromosome 9p. (C) TK10, a renal cancer cell line, has 55 rearrangements involving chromosome 5.

Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc. 31

TK10 spectral karyotype

A

der(5) 1

2

3

6

7

8

13

14

15

19

20

4

9

21

5

10

11

12

16

17

18

22

X

C

B

Normal chr 5 Derivative chr 5

6 4

Normal chr 5

1 0

FISH colours 5.8

32.0 6.4

65.8 150.6 172.6 66.4 151.2 173.2 32.8

Derivative chr 5

Genomic location (Mb)

Normal chr 5

Normal chr 5 6 - 6.1 Mb 32.3 - 32.4 Mb 66 - 66.1 Mb 150.9 - 151 Mb 172.9 - 173 Mb

Normal chr 5

Figure 3. FISH Profiling of TK10 (A) The spectral karyotype of the TK10 genome. (B) Copy number profiles and genomic rearrangements for the five regions of chromosome 5 studied by multicolor FISH. (C) Multicolor FISH of TK10.

to increasingly disordered genomic structure (Figure 5A). This is the conventional view of how most complex regional clusters of rearrangements evolve, especially genomic amplification. Localization results either from rearrangement targeting a specific cancer gene or through regional abnormalities driving recurrent DNA breakage. The second model to explain the distinctive genomic structures described here is that the overwhelming majority of rearrangements occur in a single catastrophic event. 32 Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc.

In this scenario, the chromosome or chromosomal region shatters into tens to hundreds of pieces, some (but not all) of which are then stitched together by the DNA repair machinery in a mosaic patchwork of genomic fragments (Figure 5B). Several characteristics of the patterns we observe here make the progressive rearrangement model difficult to sustain, and give support to the catastrophe model. The first observation is that the number of copy number states observed in the final

A

PD3808a (chordoma)

Deletion-type Tandem dup-type Tail-to-tail inverted Head-to-head inverted Interchromosomal

Chr 9

Chr 3

Chr 8

Chr 7

B

Chr 4

PD3786a (osteosarcoma) Chr 14

C

PD3807a (chordoma)

Chr 8 Chr 14

Chr 1

Chr 8 Chr 12

Chr 3

Figure 4. Chromothripsis Involving More Than One Chromosome in Primary Samples from Patients with Bone Cancer For each case, the relevant chromosomes are shown with SNP6 microarray copy number profiles in the outer ring, allelic ratios in the inner ring, and somatically acquired genomic rearrangements shown as arcs in the center. (A) PD3808a, from a chordoma, shows 147 rearrangements interlinking chromosomes 3q, 4q, 7q, 8p, and 9p. (B) PD3786a, an osteosarcoma sample, carries 88 rearrangements involving chromosome 8, 12, and 14. (C) PD3807a, another chordoma sample, has 38 rearrangements involving chromosomes 1p, 3, 8, and 14.

configuration of the chromosome is restricted to two (occasionally three). With sequential, independent rearrangements, the number of different states observed would be expected to increase as the number of breakpoints rises (Figure 5A). Tandem

duplications increase copy number and, because many of the observed rearrangements with a tandem duplication pattern in these samples overlap with one another, we would anticipate a number of segments to have been sequentially amplified Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc. 33

A

B

Progressive rearrangements model Germline A B

C

D

E

F

G

H

I

J

Tandem duplication CDEF A B C D E F

C

D

E

F

Catastrophe model A

B

C

D

E

H

I

D

F

E

G

H

I

J

J

C

D

H

G

F

E

I

F

C

D

H

G

F

B

F

A

G

E

J

J

D

I

Deletion EI A B C

F

Catastrophic chromosome breakage G

C

Inversion EFGH A B C D

E

H

J Non-homologous end-joining

Tandem duplication BC A B C B C

J D

E

F

C

D

H

G

F

F

C

B

D

H

J I

(lost to cell) A

G

E

Resulting copy number & rearrangements graph

3 2 1 0

A

B

C

D

E

F

G

H

I

J

A

D

E

F

G

H

I

J

10 0

PD3172 TK10 0

50

SNU-C1

8505C 100

150

200

250

Number of breakpoints

p<0.0001

PD3172

SNU-C1

8505C

Ob se Exp rved ect ed

20

p=0.005

Ob se Exp rved ect ed

30

p<0.0001

Ob se Exp rved ect ed

40

p<0.0001 100,000 10,000 1,000 100 10 1 0.1 0.01

Ob se Exp rved ect ed

Num. copy number states

C

D Progressive rearrangement simulations

Distance between adjacent bkpts (kb)

C

B

TK10

Figure 5. Genomic Features of Chromothripsis Suggest that Most Rearrangements Occur in a Single Catastrophic Event (A) Example of a sequence of progressive rearrangements disrupting a model chromosome. The chromosomal configuration after each rearrangement is shown, together with the copy number and rearrangement plot that would result (in the style of Figure 2). (B) Example of how a chromosomal catastrophe might break the chromosome into many pieces that are then stitched back together haphazardly. (C) One thousand Monte Carlo simulations (black points) performed under the assumption that rearrangements accumulate progressively over time show the number of copy number states seen in the resultant derivative chromosome. Samples with chromothripsis, shown as red diamonds, fall well outside this spectrum. (D) Observed distances between adjacent breakpoints for each sample are shown beside the expected distribution if breaks occurred in entirely random locations.

several-fold under the progressive rearrangements model. Although deletion events would tend to counteract increases in copy number, the chances of these two processes being so balanced as to generate only two copy number states fall rapidly as the number of rearrangements increases. To demonstrate this, we performed Monte Carlo simulations of the progressive rearrangement model. Rearrangements were randomly sampled from the set of breakpoints found in SNU-C1, the resulting chromosome structure calculated, and the process repeated to generate different numbers of rearrangements (Figure 5C). As predicted, with increasing numbers of rearrangements, the observed number of different copy number states also rises. 34 Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc.

The observed profiles of the three cell lines and the CLL patient sit well outside the spectrum observed under simulations of the progressive rearrangement model. In contrast, the catastrophe model predicts exactly two copy number states. Those fragments that are retained in the eventual derivative chromosome will have the higher copy number state; those that are lost to the cell will be in the lower copy number state (Figure 5B). The second problem for the progressive rearrangements model is the retention of heterozygosity in regions with higher copy number. Once lost, heterozygosity cannot generally be regained. For example, the region around 66 Mb of chromosome

15 of SNU-C1 is heterozygous, but is encompassed in the span of no fewer than 21 rearrangements with the orientation of deletions, as well as 20 tandem duplication-type and 52 inverted rearrangements (Figure 2A). Under the progressive rearrangement model, a deletion that occurred early in the sequence of rearrangements would permanently remove heterozygosity between the breakpoints. Thus, deleting events can only occur late in the succession of rearrangements, once regions of retained heterozygosity have either been switched out of the region by inversion or copied by tandem duplication. When extended across all 239 rearrangements involving chromosome 15, there is major difficulty constructing a sequence of progressive rearrangements that would spare the heterozygosity found in over 20 separate segments. In contrast, alternating regions of heterozygosity and LOH is the natural consequence of the catastrophe model. With a normal parental chromosome and one shattered into many pieces, any fragment that is retained in the eventual derivative chromosome will be heterozygous; those that are lost to the cell will result in LOH in those regions (Figure 5B). A third feature arguing against the progressive rearrangement model is that breakpoints show significantly more clustering along the chromosome or chromosome arm than expected by chance (Figure 5D). A clean break across double-stranded DNA (dsDNA) generates two naked ends of which none, one or two may subsequently be repaired. Some of the clustering represents erroneous repair of both sides of a dsDNA break (see Figure 5B, for example). The extent of clustering observed in breakpoint locations, however, is much greater than explicable by this means alone. This presents some difficulties for the progressive rearrangements model because such nonrandom distribution of independently generated breaks would imply extensive regional variation in chromosomal fragility. Specific regions of increased propensity to rearrangement have been documented (Bignell et al., 2010), but not to the extent observed here. Under a catastrophe model, clustering among the prolific numbers of DNA breaks would perhaps be expected, depending on the process causing the DNA damage and repair. The limited overlap between sequences at the breakpoint junction suggests that the major mechanisms of DNA repair here are microhomology-mediated break repair and/or nonhomologous end-joining rather than homologous recombination (Figure S5). In conclusion, several distinctive genomic features imply that a major catastrophic event underpins the massive, but localized, genomic rearrangement in these samples. These arguments extend to cancers where we have observed involvement of several different chromosomes. We do not argue that absolutely every rearrangement was generated in one event—indeed, a later partial duplication of the derivative chromosome is likely to explain why some samples (such as C32, Figure S2C) oscillate across three copy number states rather than two. However, the majority of rearrangements seen in these examples almost certainly occurred in a single event. Chromothripsis Can Generate Genomic Consequences that Promote Cancer Development A cell suffering tens to hundreds of DNA breaks in a single cataclysmic event would be expected to undergo apoptosis. That

a cell can survive such an insult and progress to become cancerous suggests that the extensive remodeling of the genome may confer significant selective advantage to that clone. To explore this possibility, we analyzed the genomic data for evidence of changes that might promote the development of cancer. One small cell lung cancer cell line, SCLC-21H, demonstrates massive numbers of copy number changes on chromosome 8, mostly with the typical appearances of chromothripsis (Figure S2A). Interestingly, however, the SNP array data suggest that some segments of the chromosome might be heavily amplified. We mapped 170 breakpoints, all involving chromosome 8 and showing the expected patterns of rearrangements described above (Figure 6A and Table S2). Whereas most of the chromosome oscillates among low copy number states, there are 15 discrete segments of the chromosome present at markedly increased copy number, ranging from 50 to 200 copies per cell (Figure 6B). One of these segments contains the MYC oncogene, amplified in 10%–20% of small cell lung cancers (Sher et al., 2008). The rearrangement data demonstrate that the 15 regions are interwoven by a series of rearrangements, many of which demarcate the starts and ends of the massively amplified segments. Strikingly, we found no evidence for breakpoints linking these massively amplified regions to the other, nonamplified but rearranged, regions of chromosome 8. One potential mechanism for these findings is that at some stage while the cancer was evolving, chromosome 8 shattered into hundreds of pieces. Many of these were stitched together into a derivative chromosome 8, but 15 other fragments were joined to create a double minute chromosome of 1.1Mb in size (thick lines, Figure 6B). Containing MYC, it was of considerable selective advantage for daughter cells to carry extra copies of the double minute, and through further internal rearrangements (thin lines, Figure 6B) and overreplication, the massive amplification evolved. To assess this hypothesis, we performed multicolor FISH. First, we probed three nonamplified segments of chromosome 8 that the sequencing suggested were joined together through a head-to-head inverted rearrangement and a tandem duplication-type rearrangement. This revealed a single normal copy of chromosome 8 with the probes hybridized in the expected orientation and distance apart, and two derivative 8 chromosomes with the three probes closely juxtaposed (Figure 6C). Thus, the cells contain a cytogenetically normal chromosome 8 and a derivative chromosome 8 generated by chromothripsis that has subsequently undergone chromosomal duplication. Second, we probed three of the chromosome 8 regions that were heavily amplified (Figure 6D). This demonstrated huge numbers of extrachromosomal copies of the segments, with the probes closely abutting. In addition, there were two homogeneously staining regions identified by the probes, consistent with chromosomal integration of the double minutes. Probes for the double minute chromosomes were found in the correct orientation on the normal chromosome 8, but were absent from the two copies of the derivative chromosome 8 (Figure S6A). Taken together, these findings are consistent with the model that the catastrophic shattering of chromosome 8 has facilitated the creation of a double minute chromosome, which, in this Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc. 35

A Non-inverted orientation

6

Copy 3 number 0

Allelic 1 ratio 0 Inverted orientation 0

10

20

30

40

50

60 70 80 90 100 110 120 130 140 150 Genomic location (Mb) MYC

Copy number

B 250 200 150 100 50 0

128.9

116.4 127.5 127.6 128.7

99.4 116.2

90.1 99.2

66.6 84.0 84.2 89.9

11.2 11.4 29.6 29.7 40.3 40.4 40.5 40.6 41.3 41.4 55.5 55.6 63.2 63.3 63.5 63.7 66.4

FISH probes (for part D)

Genomic location (Mb)

C

D Homogeneously staining regions

13Mb 41Mb 49Mb

Derivative chr 8

Normal chr 8

Double minutes (x100s)

Figure 6. Generation of a Double Minute Chromosome Containing MYC by Chromothripsis in a Small Cell Lung Cancer Cell Line, SCLC-21H (A) Copy number profile, allelic ratio, and rearrangements of chromosome 8. (B) Copy number data from the rearrangement screen shows 15 discrete regions of chromosome 8 that are massively amplified, with 50–200 copies per cell. Each amplified region is demarcated by rearrangements linking to other heavily amplified segments (thick lines), with evidence for later internal rearrangements also found (thin lines). (C) Three color FISH for three regions of chromosome 8 (predicted to be linked by the rearrangement data, but not amplified; green, 13 Mb; red, 41 Mb; pale pink, 49 Mb). (D) FISH for three heavily amplified regions. The locations of the probes are shown in Figure 6B (red, 66.5 Mb; white, 99.3 Mb; green, 128.8 Mb).

36 Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc.

A

PD3808a: Chr 9p To chr 7q

*

Figure 7. Loss of Tumor Suppressor Genes through Chromothripsis

Chr 4q

Chr 8p

To chr 8p

To chr 7q

*

*

2 1 0 1 0

CDKN2A 17

B

20 23 Genomic location (Mb)

WRN 26 29

32 35 Genomic location (Mb)

C

8505C: Chr 9p

FBXW7 38 152

154 156 158 160 Genomic location (Mb)

(A) PD3808a, the chordoma sample shown in Figure 4A, shows clustered chromothripsis rearrangements around CDKN2A, leading to a loss of one copy of this tumor suppressor gene. The other copy is also lost, through a deletion, which presumably occurred on the other parental copy of chromosome 9p at a separate time-point (thick line). The same cluster of chromothripsis rearrangements causes loss of a second tumor suppressor gene, FBXW7, on chromosome 4q and a third cancer gene, WRN, on chromosome 8p. (B) Chromothripsis has also led to loss of one copy of CDKN2A in the thyroid cancer cell line, 8505C. (C) Loss of two tumor suppressor genes, CDKN2A and the microRNA cluster miR-15a/16-1, by clustered rearrangements involving chromosomes 4, 9, and 13 in a patient with CLL, PD3175a.

PD3175a (CLL) CDKN2A

the gene occurred during chromothripsis, although it is formally possible that an independent deletion of CDKN2A might 2 Chr 9 0 have occurred before chromothripsis. 18Mb 24Mb 4 The second copy of the gene was lost 2 To chr 10 through a focal deletion on the other 0 parental chromosome, which presumably 1 occurred as a temporally separate event 0 (thick blue line, Figure 7A). With so many rearrangements gener2 ated in a single genomic crisis, it is 0 feasible that more than one cancer45Mb 52Mb 98Mb 101Mb 168Mb 171Mb 21.5 21.75 22 22.25 22.5 causing lesion could occur in the same Chr 13 Chr 4 Genomic location (Mb) miR-15a / 16-1 event. In addition to the loss of CDKN2A described above, the chordoma sample PD3808a had a rearrangement that directly disrupted WRN, linking the 30 example, containing MYC, acts as a substrate for amplification, portion of this gene on chromosome 8 to an intergenic region on 9p just downstream of CDKN2A (thick purple line, Figure 7A). evolutionary selection and progression toward cancer. Chromothripsis may lead to the generation of other forms of WRN is a cancer gene in which germline mutation causes Werner marker chromosome also. We studied the spectral karyotype syndrome, a condition associated with markedly increased risk of the pancreatic cancer sample with evidence for chromothrip- of bone tumors, and in which somatic inactivating mutations sis involving multiple chromosomes (Figures S6B and S6C). have been documented in renal cancer (Dalgliesh et al., 2010). Even with the low resolution of SKY, a chromosome arm with This same patient also lost a copy of FBXW7 on chromosome at least six cytogenetically visible stripes could be seen, indi- 4q (Figure 7A). The rearrangements around this gene link to cating that the many interchromosomal rearrangements have chromosomes 3q, 7q, 8p, 9p, and elsewhere on 4q similar to intertwined segments from multiple different chromosomes those near CDKN2A and WRN, suggesting that loss of FBXW7 occurred during the same chromothripsis event. FBXW7 is into a distinctive marker chromosome. A second potential mechanism by which chromothripsis could inactivated in 6% of all cancers across many subtypes generate cancer-causing genomic changes is through loss or (Akhoondi et al., 2007; Kemp et al., 2005; Maser et al., 2007). disruption of tumor suppressor genes. In the chordoma, Inactivation is frequently heterozygous, supported by functional PD3808a, the CDKN2A gene is homozygously deleted (Fig- data suggesting it may be a haploinsufficient tumor suppressor ure 7A), with one of the copies probably lost through chromo- gene (Kemp et al., 2005; Mao et al., 2004). Thus, the single thripsis. The two rearrangements demarcating the copy number catastrophic event inducing chromothripsis in this patient has change from 2 to 1 around CDKN2A (marked with * in Figure 7A) resulted in disruption of three tumor suppressor genes. A number of other known cancer genes were affected by appear to be part of the network of interchromosomal rearrangements interlinking regions from chromosomes 3q, 4q, 7q, 8p, rearrangements across the samples described here (Table S5), and 9p seen in Figure 4A. This argues that loss of this copy of including ARID1A in PD3807a (chordoma). In 8505C, the CDKN2A

Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc. 37

chromothripsis involving chromosome 9p has led to loss of one copy of CDKN2A (Figure 7B); the other carries a deletion of the first exon of the gene. We also identified a second patient with CLL who showed evidence for loss of two tumor suppressor genes in a cluster of rearrangements involving chromosomes 4, 9, and 13 (Figure 7C). Here, single copies of both CDKN2A and miR-15a/16-1, the microRNA cluster deleted in >50% of CLL patients (Cimmino et al., 2005), were lost through interchromosomal rearrangements, whereas the other copy of the microRNA cluster was deleted in a presumably separate event (blue line, Figure 7C). Theoretically, chromothripsis rearrangements could juxtapose coding portions of two genes in the same orientation with an open reading frame, producing a potentially oncogenic fusion gene. Among chromothripsis rearrangements, we found 17 that could potentially create novel in-frame fusions (Table S5). None generates a classic cancer-associated fusion gene, such as BCR-ABL1 or EWS-FLI1, and the proportion of rearrangements generating novel in-frame fusions is similar to that observed for other types of rearrangements (Campbell et al., 2010; Stephens et al., 2009). This suggests that most are coincidental ‘‘passenger’’ events, unlikely to drive cancer development. Such dramatic restructuring of a genome will disrupt both coding sequences directly and the linkage between coding exons and regulatory elements of very many genes. We explored whether expression profiles of genes from chromosomes affected by chromothripsis differed from those of intact chromosomes. For SCLC-21H, genes from chromosomes that were not affected by chromothripsis showed an approximately normal distribution of expression levels relative to their expression in other SCLC cell lines (Figure S7), as expected. On chromosome 8, however, affected by chromothripsis, expression levels were decreased in 5% of genes in SCLC-21H relative to their expression in other SCLC cell lines (chromosome 7 versus chromosome 8, p = 0.001; chromosome 6 versus chromosome 8, p < 0.0001). Similar differences were observed for SNU-C1, in which chromothripsis affected chromosome 15 (chromosome 14 versus chromosome 15, p = 0.02; chromosome 13 versus chromosome 15, p < 0.0001). Taken together, these data exemplify the mechanisms by which chromothripsis can promote the development of cancer. In particular, more than one cancer-causing lesion can arise from a single catastrophe, and the chaotic genomic architecture that results can inactivate or disrupt the transcription of many more genes. DISCUSSION Here we describe a quite remarkable phenomenon whereby tens to hundreds of chromosomal rearrangements involving localized genomic regions can be acquired in an apparently one-off cellular catastrophe. Astoundingly, not only can a cell actually survive this crisis, it can emerge with a genomic landscape that confers a significant selective advantage to the clone, promoting the evolution toward cancer. Such an event appears to have occurred in 2%–3% of all cancers, across many subtypes, and may be particularly frequent in bone cancers. 38 Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc.

There are few documented examples of how catastrophic genomic change affects evolutionary processes. Reassortment of influenza virus genomes can lead to entirely novel strains with considerable pandemic potential (Neumann et al., 2009). In eukaryotes, ‘‘showers’’ of several point mutations in a localized genomic region in a single cell cycle have been described in murine models (Wang et al., 2007), with similar arguments extended to clustered mutations in humans with germline genetic diseases (Chen et al., 2009). We would predict that in the case of chromothripsis, the overwhelming majority of cells suffering such spectacular genomic damage would either die or acquire more detrimental than advantageous variants. However, very rarely, a cell might acquire one or more cancercausing lesions from such an event and this clone would then have taken a considerable leap along the road to cancer. There would still be the need for additional mutations in cancer genes, exemplified by the second hits in CDKN2A seen in the chordoma and thyroid cancer samples (Figure 7), but we might anticipate the emergent tumor having shorter latency. What causes such dramatic damage to the genome? The distinctive signature of the process gives some clues. The genomic regions involved in each example are sharply circumscribed, whether it be a whole chromosome, a chromosome arm or a region of just a few megabases within a chromosomal band. It seems likely that the insult occurs while the chromosomes are condensed for mitosis. During interphase, chromosomes are relaxed with long loops of DNA winding through the nucleus: although given chromosomes occupy general nuclear territories, these tend to be loosely defined and nonexclusive (Misteli, 2007). DNA damage acquired in interphase would seem unlikely to exhibit such intense clustering of breaks within such well-circumscribed genomic regions. The existence of rearrangements involving both sides of a DNA break, the potential to create both a derivative chromosome and a double minute chromosome in the same event and the seeming near-randomness of which fragment is joined to which fragment suggest that literally hundreds of shards of genomic DNA circulate unfettered in the nucleus during the catastrophe, that the DNA repair machinery is pasting them together in a helter-skelter tumult of activity. The agent of this physical chromosomal damage is unknown. One appealing possibility is ionizing radiation. Well-known to induce dsDNA breaks, a pulse of ionizing radiation could cut a swathe through a condensed chromosome and, depending on whether the angle of the path relative to the long axis of the chromosome is transverse, oblique or longitudinal, generate breaks involving a band, an arm or the whole chromosome. Such a model could potentially be tested by in vitro studies of cells surviving irradiation and by analysis of cancer genomes from patients with prior environmental or therapeutic radiation exposure. Another intriguing possibility is that the breakage-fusionbridge cycle associated with telomere attrition could induce the damage, especially because most examples of chromothripsis observed here involve regions extending to the telomeres (Figure S1). End-to-end chromosome fusions are a cytogenetic hallmark of telomere loss (Artandi et al., 2000; Gisselsson et al., 2001; O’Hagan et al., 2002), and the two centromeres of such dicentric chromosomes are pulled to opposite daughter

cells during anaphase, forming a so-called anaphase bridge (Bignell et al., 2007; McClintock, 1941; Sahin and Depinho, 2010). It is unclear how these bridges are resolved, but they appear to induce the formation of nuclear buds and micronuclei containing fragmented DNA in the daughter cells (Pampalona et al., 2010). It is therefore conceivable that the dramatic stretching and pinching of the chromosome bridge during the final stages of cytokinesis could be associated with catastrophic, but localized, genomic damage. If this hypothesis is true, cancer genomes from genetically engineered mouse models of telomerase deficiency (Artandi et al., 2000; Maser et al., 2007; O’Hagan et al., 2002) may demonstrate similar patterns of genomic rearrangement to those observed here. Whatever the mechanism of damage, the consequences are profound. Faced with hundreds of DNA breaks, the cell’s DNA repair machinery attempts to rescue the genome. The resultant hodgepodge bears little resemblance to its original structure, and the genomic disruption has wholesale and potentially oncogenic effects. EXPERIMENTAL PROCEDURES Samples Rearrangement screens were performed on genomic DNA from 10 patients with chronic B cell lymphocytic leukemia attending Addenbrooke’s Hospital, Cambridge, UK. Screens were also performed on genomic DNA samples from 20 patients with bone cancer (9 osteosarcoma, 11 chordoma) collected at the Royal National Orthopaedic Hospital, Middlesex, UK. From all 30 samples, we had germline DNA available. Informed consent was obtained from all patients or guardians and samples were collected and analyzed with approval from relevant Ethics Committees. The cell line set has previously been described (Bignell et al., 2010), and for the four samples presented here, germline DNA was not available. Massively Parallel Sequencing The protocols for massively parallel, paired-end sequencing to identify somatically acquired genomic rearrangements in cancer samples have been described in detail elsewhere (Campbell et al., 2008; Quail et al., 2008; Stephens et al., 2009). In brief, 5 mg of genomic DNA from the tumor sample was sheared to fragments 400–500 base pairs (bp) in size. Sequencing of 37 bp from either end was performed on the Illumina Genome Analyzer II platform. Reads were aligned to the reference human genome (NCBI build 36) using MAQ (Li et al., 2008). Putative genomic rearrangements were screened by PCR across the breakpoint in tumor DNA samples and, where available, germline DNA. SNP Array Analyses Tumor DNA samples from the 20 patients with bone cancer and the cell line set were also analyzed by Affymetrix SNP6 microarrays, as described (Bignell et al., 2010). Copy number and allelic ratio profiles were statistically processed using the PICNIC algorithm (Greenman et al., 2010). Multiplex-Fluorescence In Situ Hybridization Human 24 color M-FISH paint was made essentially following the ‘‘pooling’’ strategy described (Geigl et al., 2006). Briefly, individual human chromosome-specific DOP-PCR products were grouped into five re-amplifiable pools based on the fluorescence label and subsequently labeled with biotin-16dUTP, Texas Red-12-dUTP, Cy3-, Cy5-dUTP, and Green-dUTP. Labeled DNA was precipitated with human Cot-1 DNA. Where used, human fosmid clones were selected according to their positions in the hg17 reference assembly. Biotin-labeled probe was detected with one layer of Cy5.5-conjugated mouse anti-biotin. Metaphases were examined with either a Leica DM5000 or a Zeiss AxioIamger D1 fluorescence microscope.

Statistical Analysis Simulations of the progressive rearrangement model were performed 1000 times using the 239 rearrangements involving chromosome 15 identified in SNU-C1. Starting with a wild-type chromosome 15, rearrangements were randomly selected without replacement from the set of 239 events. At each step, the relevant rearrangement was applied to the current configuration of the chromosome: for example, a deletion-type rearrangement would lead to loss of intervening sequence between the breakpoints. Where the selected rearrangement was impossible (that is, one breakpoint occurred in a region already lost to the chromosome in that simulation), it was discarded and another selected. Where more than one copy of the breakpoint location existed in the current configuration (for example, the region had undergone tandem duplication in a previous rearrangement), which copy of the breakpoint location to use was chosen randomly. The number of unique copy number states across the chromosome was monitored for each simulation. To test whether the locations of genomic breakpoints showed more clustering than expected by chance, Kolmogorov-Smirnov tests were used to compare the observed distribution of distances between adjacent breaks and that expected under the null hypothesis (exponential distribution). For analysis of expression levels of genes from chromothripsis chromosomes compared to intact chromosomes, the expression levels of every gene on the relevant chromosomes were converted to Z-scores using the expression levels for other cell lines from the same tumor type. The distribution of Z-scores for the chromothripsis chromosome was then compared to the distribution for other chromosomes by the Kolmogorov-Smirnov test. The circle plots were generated with Circos (Krzywinski et al., 2009). ACCESSION NUMBERS Genome sequence data have been deposited at the European GenomePhenome Archive (http://www.ebi.ac.uk/ega/, hosted by the EBI) with accession number EGAD00001000002. SUPPLEMENTAL INFORMATION Supplemental Information includes seven figures and five tables and can be found with this article online at doi:10.1016/j.cell.2010.11.055. ACKNOWLEDGMENTS This work was supported by the Wellcome Trust (grant reference 077012/Z/ 05/Z) and the Chordoma Foundation. P.J.C. is personally funded through a Wellcome Trust Senior Clinical Research Fellowship (grant reference WT088340MA). I.V. is supported by a fellowship from The International Human Frontier Science Program Organization. We also acknowledge support for sample banking and processing from the Cambridge and UCL/UCLH NIHR Biomedical Research Centres and Skeletal Action Cancer Trust (SCAT), especially Dr. Anthony Bench, Dr. Wendy Erber, and Miss Dina Halai. We would like to thank George Vassiliou for the neologism ‘‘chromothripsis.’’ Received: August 20, 2010 Revised: November 3, 2010 Accepted: November 24, 2010 Published: January 6, 2011 REFERENCES Akhoondi, S., Sun, D., von der Lehr, N., Apostolidou, S., Klotz, K., Maljukova, A., Cepeda, D., Fiegl, H., Dafou, D., Marth, C., et al. (2007). FBXW7/hCDC4 is a general tumor suppressor in human cancer. Cancer Res. 67, 9006–9012. Artandi, S.E., Chang, S., Lee, S.L., Alson, S., Gottlieb, G.J., Chin, L., and DePinho, R.A. (2000). Telomere dysfunction promotes non-reciprocal translocations and epithelial cancers in mice. Nature 406, 641–645. Bardeesy, N., and DePinho, R.A. (2002). Pancreatic cancer biology and genetics. Nat. Rev. 2, 897–909.

Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc. 39

Beroukhim, R., Mermel, C.H., Porter, D., Wei, G., Raychaudhuri, S., Donovan, J., Barretina, J., Boehm, J.S., Dobson, J., Urashima, M., et al. (2010). The landscape of somatic copy-number alteration across human cancers. Nature 463, 899–905. Bignell, G.R., Greenman, C.D., Davies, H., Butler, A.P., Edkins, S., Andrews, J.M., Buck, G., Chen, L., Beare, D., Latimer, C., et al. (2010). Signatures of mutation and selection in the cancer genome. Nature 463, 893–898. Bignell, G.R., Santarius, T., Pole, J.C., Butler, A.P., Perry, J., Pleasance, E., Greenman, C., Menzies, A., Taylor, S., Edkins, S., et al. (2007). Architectures of somatic genomic rearrangement in human cancer amplicons at sequence-level resolution. Genome Res. 17, 1296–1303. Campbell, P.J., Stephens, P.J., Pleasance, E.D., O’Meara, S., Li, H., Santarius, T., Stebbings, L.A., Leroy, C., Edkins, S., Hardy, C., et al. (2008). Identification of somatically acquired rearrangements in cancer using genome-wide massively parallel paired-end sequencing. Nat. Genet. 40, 722–729. Campbell, P.J., Yachida, S., Mudie, L.J., Stephens, P.J., Pleasance, E.D., Stebbings, L.A., Morsberger, L.A., Latimer, C., McLaren, S., Lin, M.L., et al. (2010). The patterns and dynamics of genomic instability in metastatic pancreatic cancer. Nature 467, 1109–1113. Chen, J.M., Ferec, C., and Cooper, D.N. (2009). Closely spaced multiple mutations as potential signatures of transient hypermutability in human genes. Hum. Mutat. 30, 1435–1448. Cimmino, A., Calin, G.A., Fabbri, M., Iorio, M.V., Ferracin, M., Shimizu, M., Wojcik, S.E., Aqeilan, R.I., Zupo, S., Dono, M., et al. (2005). miR-15 and miR-16 induce apoptosis by targeting BCL2. Proc. Natl. Acad. Sci. USA 102, 13944–13949. Dalgliesh, G.L., Furge, K., Greenman, C., Chen, L., Bignell, G., Butler, A., Davies, H., Edkins, S., Hardy, C., Latimer, C., et al. (2010). Systematic sequencing of renal carcinoma reveals inactivation of histone modifying genes. Nature 463, 360–363. Ding, L., Ellis, M.J., Li, S., Larson, D.E., Chen, K., Wallis, J.W., Harris, C.C., McLellan, M.D., Fulton, R.S., Fulton, L.L., et al. (2010). Genome remodeling in a basal-like breast cancer metastasis and xenograft. Nature 464, 999–1005. Geigl, J.B., Uhrig, S., and Speicher, M.R. (2006). Multiplex-fluorescence in situ hybridization for chromosome karyotyping. Nat. Protoc. 1, 1172–1184. Gisselsson, D., Jonson, T., Petersen, A., Strombeck, B., Dal Cin, P., Hoglund, M., Mitelman, F., Mertens, F., and Mandahl, N. (2001). Telomere dysfunction triggers extensive DNA fragmentation and evolution of complex chromosome abnormalities in human malignant tumors. Proc. Natl. Acad. Sci. USA 98, 12683–12688. Greenman, C.D., Bignell, G., Butler, A., Edkins, S., Hinton, J., Beare, D., Swamy, S., Santarius, T., Chen, L., Widaa, S., et al. (2010). PICNIC: an algorithm to predict absolute allelic copy number variation with microarray cancer data. Biostatistics 11, 164–175. Jones, S., Chen, W.D., Parmigiani, G., Diehl, F., Beerenwinkel, N., Antal, T., Traulsen, A., Nowak, M.A., Siegel, C., Velculescu, V.E., et al. (2008). Comparative lesion sequencing provides insights into tumor evolution. Proc. Natl. Acad. Sci. USA 105, 4283–4288. Kemp, Z., Rowan, A., Chambers, W., Wortham, N., Halford, S., Sieber, O., Mortensen, N., von Herbay, A., Gunther, T., Ilyas, M., et al. (2005). CDC4 mutations occur in a subset of colorectal cancers but are not predicted to cause loss of function and are not associated with chromosomal instability. Cancer Res. 65, 11361–11366. Krzywinski, M., Schein, J., Birol, I., Connors, J., Gascoyne, R., Horsman, D., Jones, S.J., and Marra, M.A. (2009). Circos: an information aesthetic for comparative genomics. Genome Res. 19, 1639–1645.

40 Cell 144, 27–40, January 7, 2011 ª2011 Elsevier Inc.

Li, H., Ruan, J., and Durbin, R. (2008). Mapping short DNA sequencing reads and calling variants using mapping quality scores. Gen. Res. 18, 1851–1858. Mao, J.H., Perez-Losada, J., Wu, D., Delrosario, R., Tsunematsu, R., Nakayama, K.I., Brown, K., Bryson, S., and Balmain, A. (2004). Fbxw7/Cdc4 is a p53-dependent, haploinsufficient tumour suppressor gene. Nature 432, 775–779. Mardis, E.R., Ding, L., Dooling, D.J., Larson, D.E., McLellan, M.D., Chen, K., Koboldt, D.C., Fulton, R.S., Delehaunty, K.D., McGrath, S.D., et al. (2009). Recurring mutations found by sequencing an acute myeloid leukemia genome. N. Engl. J. Med. 361, 1058–1066. Maser, R.S., Choudhury, B., Campbell, P.J., Feng, B., Wong, K.K., Protopopov, A., O’Neil, J., Gutierrez, A., Ivanova, E., Perna, I., et al. (2007). Chromosomally unstable mouse tumours have genomic alterations similar to diverse human cancers. Nature 447, 966–971. McClintock, B. (1941). The stability of broken ends of chromosomes in Zea mays. Genetics 26, 234–282. Misteli, T. (2007). Beyond the sequence: cellular organization of genome function. Cell 128, 787–800. Neumann, G., Noda, T., and Kawaoka, Y. (2009). Emergence and pandemic potential of swine-origin H1N1 influenza virus. Nature 459, 931–939. O’Hagan, R.C., Chang, S., Maser, R.S., Mohan, R., Artandi, S.E., Chin, L., and DePinho, R.A. (2002). Telomere dysfunction provokes regional amplification and deletion in cancer genomes. Cancer Cell 2, 149–155. Pampalona, J., Soler, D., Genesca, A., and Tusell, L. (2010). Telomere dysfunction and chromosome structure modulate the contribution of individual chromosomes in abnormal nuclear morphologies. Mutat. Res. 683, 16–22. Pleasance, E.D., Cheetham, R.K., Stephens, P.J., McBride, D.J., Humphray, S.J., Greenman, C.D., Varela, I., Lin, M.L., Ordonez, G.R., Bignell, G.R., et al. (2010a). A comprehensive catalogue of somatic mutations from a human cancer genome. Nature 463, 191–196. Pleasance, E.D., Stephens, P.J., O’Meara, S., McBride, D.J., Meynert, A., Jones, D., Lin, M.L., Beare, D., Lau, K.W., Greenman, C., et al. (2010b). A small-cell lung cancer genome with complex signatures of tobacco exposure. Nature 463, 184–190. Quail, M.A., Kozarewa, I., Smith, F., Scally, A., Stephens, P.J., Durbin, R., Swerdlow, H., and Turner, D.J. (2008). A large genome center’s improvements to the Illumina sequencing system. Nat. Methods 5, 1005–1010. Sahin, E., and Depinho, R.A. (2010). Linking functional decline of telomeres, mitochondria and stem cells during ageing. Nature 464, 520–528. Shah, S.P., Morin, R.D., Khattra, J., Prentice, L., Pugh, T., Burleigh, A., Delaney, A., Gelmon, K., Guliany, R., Senz, J., et al. (2009). Mutational evolution in a lobular breast tumour profiled at single nucleotide resolution. Nature 461, 809–813. Sher, T., Dy, G.K., and Adjei, A.A. (2008). Small cell lung cancer. Mayo Clin. Proc. 83, 355–367. Stephens, P.J., McBride, D.J., Lin, M.L., Varela, I., Pleasance, E.D., Simpson, J.T., Stebbings, L.A., Leroy, C., Edkins, S., Mudie, L.J., et al. (2009). Complex landscapes of somatic rearrangement in human breast cancer genomes. Nature 462, 1005–1010. Stratton, M.R., Campbell, P.J., and Futreal, P.A. (2009). The cancer genome. Nature 458, 719–724. Wang, J., Gonzalez, K.D., Scaringe, W.A., Tsai, K., Liu, N., Gu, D., Li, W., Hill, K.A., and Sommer, S.S. (2007). Evidence for mutation showers. Proc. Natl. Acad. Sci. USA 104, 8403–8408.

The Cul4-Ddb1Cdt2 Ubiquitin Ligase Inhibits Invasion of a Boundary-Associated Antisilencing Factor into Heterochromatin Sigurd Braun,1 Jennifer F. Garcia,1 Margot Rowley,1 Mathieu Rougemaille,1,2 Smita Shankar,1 and Hiten D. Madhani1,* 1Department

of Biochemistry and Biophysics, University of California, San Francisco, 600 16th Street, GH-N372C, San Francisco, CA 94158, USA 2Present address: LEA Laboratory of Nuclear RNA metabolism, Centre de Ge ´ ne´tique Mole´culaire, Centre National de la Recherche Scientifique - UPR2167, 1, avenue de la Terrasse, 91190 Gif-sur-Yvette, France *Correspondence: [email protected] DOI 10.1016/j.cell.2010.11.051

SUMMARY

Partitioning of chromosomes into euchromatic and heterochromatic domains requires mechanisms that specify boundaries. The S. pombe JmjC family protein Epe1 prevents the ectopic spread of heterochromatin and is itself concentrated at boundaries. Paradoxically, Epe1 is recruited to heterochromatin by HP1 silencing factors that are distributed throughout heterochromatin. We demonstrate here that the selective enrichment of Epe1 at boundaries requires its regulation by the conserved Cul4-Ddb1Cdt2 ubiquitin ligase, which directly recognizes Epe1 and promotes its polyubiquitylation and degradation. Strikingly, in cells lacking the ligase, Epe1 persists in the body of heterochromatin thereby inducing a defect in gene silencing. Epe1 is the sole target of the Cul4-Ddb1Cdt2 complex whose destruction is necessary for the preservation of heterochromatin. This mechanism acts parallel with phosphorylation of HP1/Swi6 by CK2 to restrict Epe1. We conclude that the ubiquitin-dependent sculpting of the chromosomal distribution of an antisilencing factor is critical for heterochromatin boundaries to form correctly. INTRODUCTION Eukaryotic genomes are organized into active and inactive domains referred to as euchromatin and heterochromatin. This functional organization plays an important role in chromosome segregation, telomere maintenance, and genome stability. Given the repressive nature of heterochromatin, the regulation of its assembly is critical for genome homeostasis. The fission yeast Schizosaccharomyces pombe has become a powerful model system for dissecting mechanisms of eukaryotic heterochromatin control (for an extensive review, see Grewal, 2010). Its genome contains three distinct major heterochromatic domains:

pericentromeric otr repeats, subtelomeric regions, and the silent mating type locus mat2/mat3. As in metazoans and plants, a key step in heterochromatin assembly is the recruitment of a histone H3 lysine 9 (H3K9) methyltransferase (Clr4 in S. pombe) to chromatin. The methylation of H3K9 by Clr4 is required for the recruitment of the HP1 family proteins Swi6 and Chp2, which then appear to spread along the DNA fiber. It is thought that nucleation of heterochromatin is guided by RNA interference (RNAi)-dependent and -independent mechanisms. In the RNAidependent pathway, heterochromatic sequences are transcribed by RNA polymerase II (Pol II) during S phase, and siRNAs are subsequently generated by the RNAi machinery. Whereas pericentromeric heterochromatin requires the RNAidependent pathway for its establishment, both pathways act redundantly at the telomeres and the silent mating type locus. Initial H3K9 methylation and subsequent binding of HP1 proteins lead to the spreading of this repressive modification, but the phenomenon of heterochromatin spread is still poorly understood (reviewed in Talbert and Henikoff, 2006). HP1 proteins, whose chromatin association depends on H3K9 methylation, seem to be involved in recruiting the upstream H3K9 methyltransferase. This has been suggested to result in methylation of neighboring nucleosomes, thereby creating a positive feedback loop in assembly and spreading of heterochromatin over large distances in cis. Spreading is a stochastic process that can result in metastable but heritable silencing of neighboring euchromatic genes, a phenomenon known as position effect variegation (PEV) (Gaszner and Felsenfeld, 2006). The invasion of heterochromatin into adjacent euchromatic regions is prevented by boundary elements, which terminate the chain of events involved in spreading. The mechanisms by which boundaries are formed are complex, and a number of models have been proposed that include tethering of boundary elements to subnuclear regions and recruiting silencing-opposing activities (Gaszner and Felsenfeld, 2006). In S. pombe, the euchromatin-heterochromatin borders are characterized by sharp transitions of euchromatic and heterochromatic histone modifications (Cam et al., 2005). Specific boundary elements composed of inverted repeat (IR) sequences are found at the boundaries flanking the silent mating type locus and the pericentromeric regions of chromosomes 1 and 3 (Cam et al., 2005; Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc. 41

Noma et al., 2006). The left and right boundary elements at the silent mating type locus (IR-R/L) contain recruitment sites for transcription factor TFIIIC, which has been suggested to delineate heterochromatic domains by sequestering the boundary elements to the nuclear periphery (Noma et al., 2006). Conversely, the pericentromeric inverted repeat (IRC) elements are associated with Pol II-dependent transcription and show an enrichment of euchromatic marks (Cam et al., 2005; Noma et al., 2006). Other boundaries at pericentromeric regions are characterized by tRNA gene clusters, which seem to be critical for barrier function (Scott et al., 2006). Epe1 (enhancer of position effect) was previously identified in a screen for mutants in S. pombe that display propagation of heterochromatin beyond its natural borders. Mutants of epe1+ show enhanced PEV at the silent mating type locus and pericentromeric regions (Ayoub et al., 2003). Epe1 is critical for the boundary function of the pericentromeric IRC elements and mediates their Pol II-dependent transcription (Zofall and Grewal, 2006). The mechanism by which Epe1 antagonizes heterochromatin spread is unknown. Epe1 contains a JmjC domain that is present in many histone demethylases but lacks a conserved residue predicted to be involved in binding of a catalytic iron atom. Furthermore, no histone demethylase activity has been detected for Epe1 in vitro (Tsukada et al., 2006). Despite being an antisilencing factor, Epe1 interacts with the HP1 proteins Swi6 and Chp2 in vivo and in vitro and is itself recruited to heterochromatin in an HP1-dependent manner (Sadaie et al., 2008; Zofall and Grewal, 2006). In particular, Epe1 facilitates the recruitment of Pol II to heterochromatic regions (Zofall and Grewal, 2006). Perhaps due to this role in Pol II-dependent transcription, mutants of epe1+ have perturbed levels of heterochromatic siRNAs and are affected in the stability of heterochromatic domains (Trewick et al., 2007). In addition, Epe1 appears to compete for binding to heterochromatin with the HDAC effector complex SHREC (Shimada et al., 2009). These findings raised the important question of how heterochromatin is protected from the silencing-antagonizing activity of Epe1 that it recruits. Histones have long been known to be substrates for the ubiquitin system. Conjugation involves the transfer of ubiquitin to a lysine residue within the substrate by an enzymatic cascade comprising an activating enzyme (E1), a conjugating enzyme (E2), and an ubiquitin ligase (E3), the latter determining substrate specificity of ubiquitylation. Ubiquitylation plays a crucial role in the regulation of chromatin. For instance, monoubiquitylation of histone H2A is associated with silencing of the mammalian Hox gene cluster (Wang et al., 2004), whereas ubiquitylation of histone H2B is a prerequisite for methylation of H3K4 and H3K79 (Nakanishi et al., 2009; Sun and Allis, 2002). Methylation of H3K9 in S. pombe requires a multisubunit E3 that associates with the H3K9 methyltransferase Clr4 in the CLRC complex and is necessary for chromatin recruitment of Clr4 (Hong et al., 2005). This E3 enzyme, Cul4-Rik1Dos1/Dos2, is related to the cullin-RING finger family of ubiquitin ligases (CRLs), in particular the conserved Cul4-Ddb1DCAF complexes. Common to this family is a modular architecture that employs a cullin family scaffold, a RING finger protein that recruits the ubiquitin-conjugating E2 enzyme, and a substrate recognition factor (Jackson and Xiong, 2009). Ddb1 is a specific adaptor 42 Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc.

protein of Cul4 RING finger ligases (Cul4-Ddb1) and recruits the substrate recognition factor that confers specificity to the ubiquitylation reaction. Most of the identified substrate recognition factors (DCAFs, Ddb1/Cul4 associated factors) contain WD40 repeats (Lee and Zhou, 2007). However, in the Cul4-Rik1Dos1/Dos2 complex, the conserved Ddb1 adaptor is replaced by Rik1 and the substrate recognition DCAF subunit is replaced by Dos1/Dos2. As the Cul4-Rik1Dos1/Dos2 E3 seems to function particularly in silencing, it appears to be a specialized paralog of the conserved Cul4 CRLs. Despite its requirement for heterochromatin formation, the corresponding substrate has not been identified. Here, we report the identification of a regulatory mechanism essential for proper boundary formation and heterochromatic silencing in S. pombe, which unexpectedly requires the action of the canonical CRL Cul4-Ddb1Cdt2. We demonstrate that the Cul4-Ddb1Cdt2 complex directly recognizes and promotes ubiquitylation and degradation of the boundary factor Epe1. Strikingly, this pathway controls the distribution of this antisilencing factor within heterochromatic domains and restricts Epe1 to the heterochromatic boundaries. We show that this heterochromatin-sculpting function of Cul4-Ddb1Cdt2 is sufficient to explain its requirement for silencing. Our studies define a ubiquitindependent degradation event necessary for heterochromatin formation and demonstrate that it functions to shape heterochromatin. RESULTS A Targeted Knockout Screen Identifies Factors Required for Pericentromeric Silencing To identify factors required for heterochromatin formation, we disrupted candidate genes in fission yeast harboring a pericentromeric ura4+ reporter gene whose silencing can be assayed using the drug 5-FOA that counterselects for ura4+-expressing cells (Ekwall et al., 1999). In S. pombe, heterochromatin marked by the HP1 protein Swi6 colocalizes with the spindle pole body (SPB) during interphase (Appelgren et al., 2003). In fact, many other heterochromatic proteins display a similar SPB-like localization or dot-like staining within the nucleus (Matsuyama et al., 2006). A high-throughput study reported that 346 S. pombe proteins display such a localization pattern when fused to yellow fluorescent protein (YFP) and expressed from an inducible promoter (Matsuyama et al., 2006). We successfully deleted 166 of these genes in an imrL::ura4+ reporter strain. In addition, we deleted 23 other genes that display sequence motifs suggestive of a potential role in chromatin biology plus a few control genes encoding known silencing factors (Figure S1H and Table S1 available online). We screened this collection of 195 deletion mutants on 5-FOA media and isolated 12 mutants with a previously undescribed loss-of-silencing phenotype (Figure S1A). Among those mutants were 11 genes that encode SBP/nuclear dot proteins and 1 factor with a nucleoplasmic localization, Ddb1, the well-studied adaptor component of the canonical Cul4 CRLs. The ddb1D mutant showed a 5-FOA silencing phenotype comparable to cells lacking the histone H3K9 methyltransferase Clr4, which is essential for heterochromatin formation (Figure S1A, left panel). Among

the SPB/nuclear dot candidates, deletion of SPCC1393.05 also resulted in a strong silencing defect, and we have described an initial analysis of this gene, ers1+, elsewhere (Rougemaille et al., 2008). The remaining SBP/nuclear dot mutants exhibited weaker phenotypes, both in the original imrL::ura4+ strain and in a strain harboring a mat3M::ura4+ reporter gene that measures silencing at the mat2/3 silent cassette (Figures S1A and S1F). RT-qPCR analysis showed that many of these mutants accumulate silenced transcripts depending on the heterochromatic region assessed (Figures S1B–S1G). The silencing defect observed in ddb1D cell mutants suggested a critical function of this E3 ligase subunit in heterochromatin formation, and its role in silencing was investigated further. The Cul4-Ddb1Cdt2 Ubiquitin Ligase Promotes Silencing at Multiple Heterochromatic Domains Cells lacking Ddb1 show a silencing defect at the inner most repeat (imr) and outer repeat (otr) elements of the pericentromeric region but are also impaired in silencing at the mat3M locus of the silent mating type cassette and a subtelomeric region (Figure 1). This result distinguishes Ddb1 from factors directly involved in the RNAi pathway as they only impact silencing at centromeres. Analysis of steady-state levels of mRNAs originating from the imr1L::ura4+ and mat3M::ura4+ loci, as well as endogenous heterochromatic sequences, showed modest (particularly at pericentric regions) but reproducible increases upon deletion of ddb1+ (Figure 1E) compared to control strains lacking Clr4 or Rik1. The largest fold-change was observed at the mat3M locus (Figure 1E). These changes in transcript levels were nonetheless sufficient to interfere with reporter gene silencing (Figures 1C and 1E). Thus, Ddb1 is required for efficient silencing but is presumably not a core component of the heterochromatin formation machinery. To identify the relevant DCAF, we focused on the 105 WD40 repeat proteins present in S. pombe. We successfully knocked out 60 of the corresponding genes in the imrL::ura4+ reporter strain and screened this collection for mutants that phenocopy ddb1D. One mutant, cdt2D, displayed an identical phenotype to that of ddb1D cells in all assays (Figures 1C and 1E). Importantly, ddb1D cdt2D double mutants showed no additive silencing defect, indicating that ddb1+ and cdt2+ are epistatic and function in the same pathway (Figure 1C). Consistent with our findings, Cdt2 has previously been described as a substrate recognition factor of Cul4-Ddb1 involved in the degradation of chromatin-associated factors (Jin et al., 2006; Liu et al., 2005; Ralph et al., 2006). Methylation of lysine 9 of histone H3 is a hallmark of heterochromatin. To study whether methylation of histone H3K9 is affected by the absence of Ddb1, we determined the profile of dimethylated H3K9 (H3K9me2) by performing chromatin immunoprecipitation (ChIP) experiments at various heterochromatic regions. To control for nonspecific effects of Ddb1 on growth, we used a ddb1D spd1D double mutant in which a known cellcycle substrate of Ddb1, Spd1, is also absent—deletion of spd1+ suppresses the growth defect of the ddb1D mutant (see below). Consistent with the silencing defect seen at the pericentromeric and subtelomeric regions, we found a significant reduction in H3K9me2 levels at the cen-dg and the tlh1+/thl2+

loci (Figure 1F) but no changes in histone H3 levels (Figure 1G). In contrast, H3K9me2 levels were unaffected at the mating type locus in ddb1D spd1D cells (Figure 1F), despite the strong silencing defect seen at this heterochromatic locus (compare Figures 1C and 1E). Thus, the decrease in H3K9 methylation cannot generally explain the silencing defect of ddb1D cells, and the different heterochromatic domains seem to have distinct requirements for silencing. Silencing Is Not Inhibited by the Cul4-Ddb1Cdt2 Substrate Spd1 To date, only two substrates of Cul4-Ddb1Cdt2 have been identified in S. pombe: Cdt1, which is required for licensing of replication origins (Ralph et al., 2006), and Spd1, which is an inhibitor of ribonucleotide reductase (Bondar et al., 2004); both substrates are degraded during S phase. Accumulation of Spd1 causes cell-cycle delay, abnormal cellular size, and a substantial growth defect in ddb1D cells (Bondar et al., 2004; Holmberg et al., 2005). As two recent studies linked the onset of S phase to RNAi-mediated assembly of heterochromatin (Chen et al., 2008; Kloc et al., 2008), we sought to test the hypothesis of whether the accumulation of Spd1 may be the reason for the silencing defect. To this end, we knocked out spd1+ in ddb1D or cdt2D cells and examined the phenotypes of the corresponding double mutants. Consistent with previous reports (Bondar et al., 2004; Holmberg et al., 2005), we observed a suppression of the slow growth phenotype in ddb1D spd1D and cdt2D spd1D cells on nonselective media (Figure 1D). In contrast, the silencing defects of ddb1D and cdt2D were unaffected in the double mutants (Figures 1D and 1E), with the exception of a partial alleviation of the silencing defect at a subtelomeric locus (Figures 1D and 1E). Thus, Spd1 is not the major target of Cul4-Ddb1Cdt2 in heterochromatin formation. Because cdt1+ is an essential gene, we could not test genetically a requirement for Cdt1 degradation in silencing. To avoid potential secondary effects that may arise from abnormal cellular morphology and slow growth associated with increased levels of Spd1, we used ddb1D spd1D cells instead of the single ddb1D mutant for the experiments described below. Cul4-Ddb1Cdt2 Controls the Levels of the JmjC Protein Epe1 by Regulating Its Protein Turnover Considering the proteolytic role of Cul4-Ddb1Cdt2 in various systems (Jackson and Xiong, 2009), its requirement for proper heterochromatin formation may reflect the need for degrading an inhibitor of silencing that acts at pericentromeric regions, subtelomeric regions, and the silent mating type locus. Only one such antisilencing factor has been described in the S. pombe literature: Epe1 (Ayoub et al., 2003; Trewick et al., 2007; Zofall and Grewal, 2006). Having only a single obvious candidate to test, we therefore chose to focus on this factor as a possible target of Cul4-Ddb1Cdt2. Using homologous recombination, we epitope-tagged the endogenous epe1+ coding sequence with a CBP-23FLAG tag (Epe1-FLAG) and found that the steadystate level of the Epe1 protein was 3-fold higher in ddb1D spd1D and cdt2D spd1D mutants than in wild-type (WT) cells (Figures 2A and 2B). This increase in Epe1 protein levels was not due to changes in transcription or mRNA stability, as the mutants did not display a difference in epe1+ mRNA levels Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc. 43

IRC

centromere 1

otr1L

dh

imr1L

dg

otr1R

cnt imr1R

ura4 + IR-L

dg

IRC

B

dh

mat2P

cen-H

mat3M

E2

RING

ura4+

Cul4

A

IR-R

silent mating type locus Ddb1

Cdt2

substrate

ura4 +

subtelomeric region tel2L

heterochromatin

-URA +

imr1L::ura4

0

dh transcripts [relative to WT]

otr1R::ura4 mat3M::ura4 +

tel2L::ura4

WT ddb1Δ cdt2Δ rik1Δ

F +

WT spd1Δ ddb1Δ ddb1Δ spd1Δ cdt2Δ cdt2Δ spd1Δ

+ RT

rik1Δ 200

10

2 0

- RT + RT cen-dh

75 50 20 10 0

- RT

+ RT

100 4 2 0

- RT + RT cen-dg

20 16 12 8 4 0

- RT

+ RT

tlh1+/tlh2 +

1.25

-URA

H3K9me2 CHIP [relative to WT]

5-FOA

ddb1Δ spd1Δ

- RT

mat3M::ura4 +

imr1L::ura4

N/S

cdt2Δ

20

+

WT ddb1Δ cdt2Δ ddb1Δ cdt2Δ rik1Δ

D

ddb1Δ 2

imr::ura4 +

+

WT ddb1Δ cdt2Δ rik1Δ

spd1Δ

dg transcripts [relative to WT]

WT ddb1Δ cdt2Δ ddb1Δ cdt2Δ rik1Δ

ura4 + transcripts [relative to WT]

C

5-FOA

WT

10

tlh1+ /tlh2 + transcripts [relative to WT]

N/S

20

ura4 + transcripts [relative to WT]

E

ura4 +

WT

1

ddb1Δ spd1Δ

0.75 0.5 0.25

tel2L::ura4

+

WT spd1Δ ddb1Δ ddb1Δ spd1Δ

0

+ g M n-d + /tlh2 at3 m 1 tlh

ce

G

1.25

H3 CHIP [relative to WT]

mat3M::ura4

+

WT spd1Δ ddb1Δ ddb1Δ spd1Δ cdt2Δ cdt2Δ spd1Δ

otr1R::ura4

+

WT spd1Δ ddb1Δ ddb1Δ spd1Δ

1 0.75 0.5 0.25 0

+ g M n-d + /tlh2 at3 m ce 1 tlh

Figure 1. Cul4-Ddb1Cdt2 Promotes Silencing at Major Heterochromatic Loci Independently of the S Phase Inhibitor Spd1 (A) S. pombe heterochromatic domains with positions of the ura4+ reporter genes. (B) Cul4-Ddb1Cdt2 cullin-RING ubiquitin ligase architecture. (C and D) Reporter assays. N/S, nonselective; 5-FOA, 50 -fluoroorotic acid; URA, without uracil. (E) RT-qPCR analysis. Shown are transcript levels relative to wild-type (WT) ± standard error of the mean (SEM) of three independent experiments. (F) ChIP analysis of H3K9me2 levels. Shown are mean values relative to WT ± SEM of three independent ChIP samples. (G) ChIP analysis of histone H3 as in (F). Error bars represent variation from the mean of two independent experiments. See also Figure S1.

relative to WT cells (Figure 2C). These findings suggested that the degradation of Epe1 might be affected in the ddb1D and cdt2D mutants. 44 Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc.

To examine the half-life of Epe1 protein, we performed cycloheximide chase experiments. As shown in Figures 2D and 2E, we observed for WT cells that Epe1 is initially rapidly degraded,

followed by a slower turnover after 20–30 min. This degradation kinetics suggests that distinct pools of Epe1 exist in the cell, which may be turned over by different pathways. In agreement with the increased steady-state protein levels, we found that Epe1 is stabilized in ddb1D spd1D cells and in cdt2D spd1D cells (Figure 2E). This result implies that Cul4-Ddb1Cdt2 promotes degradation of Epe1 in vivo. Since Cdt2 is transcriptionally induced during S phase (Liu et al., 2005), we examined whether Epe1 levels decreased in a Cdt2-dependent fashion upon the induction of an S phase arrest using hydroxyurea (HU). Indeed, we observed that, upon addition of HU to asynchronous cultures, Epe1 levels decreased in WT cells but not in cdt2D cells (Figure 2F). epe1+ mRNA levels dropped modestly during the time course, but there was no difference in this phenotype between WT and cdt2D cells, indicating that the Cdt2-dependent drop in protein levels was due to turnover rather than an indirect effect of Cdt2 on epe1+ mRNA levels. Similarly, we found that the HU-induced turnover of Epe1 was blocked in cells lacking Ddb1 (Figure S2). To further analyze the role of Cul4-Ddb1Cdt2 in the regulation of Epe1, we examined whether Epe1 is ubiquitylated in vivo and whether ubiquitylation is diminished in the ddb1D spd1D mutant. In order to enrich for ubiquitylated Epe1 conjugates, we coexpressed N-terminally His-tagged ubiquitin (His-Ub) in WT and ddb1D spd1D cells, both expressing Epe1-FLAG, and performed pull-down experiments against the His-tag under denaturing conditions. When the precipitated His-Ub conjugates were analyzed by anti-FLAG immunoblots, we detected distinct, Epe1-FLAG-specific bands that show a slower migration pattern, indicating that a fraction of Epe1 is modified by ubiquitin (Figure 2G). Notably, whereas the levels of nonmodified Epe1 are increased in the ddb1D spd1D cells compared to WT cells, the corresponding ubiquitin conjugates are significantly decreased in the mutant (Figure 2G). To quantify the decrease in Cul4Ddb1Cdt2-dependent Epe1 ubiquitylation, we determined the ratio of Epe1-ubiquitin conjugates (pull-down samples) to nonmodified Epe1 (input) and found that the relative level of ubiquitylated Epe1 was about 3-fold reduced in the ddb1Dspd1D mutant compared to WT cells (Figure 2G). Collectively, these results demonstrate that Epe1 is ubiquitylated and degraded in a Cul4-Ddb1Cdt2-dependent manner. The remaining amount of Epe1-ubiquitin conjugates observed in ddb1D spd1D cells suggests that other ubiquitylation routes exist and is consistent with our findings that degradation of Epe1 is not entirely abrogated in cells lacking Cul4-Ddb1Cdt2. As Cdt2 is a substrate recognition component of Cul4-Ddb1 ubiquitin ligases, we tested whether it binds to Epe1. We first examined by two-hybrid analysis whether Cdt2 and Epe1 interacted. Indeed, we found that a Cdt2-lexA DNA-binding domain bait fusion interacted with an Epe1-B42 activation domain prey fusion but not a control prey fusion (Figures 3A and 3B). As might be expected, this interaction appeared to be weaker than the interaction between Epe1 and Swi6 (Figures 3A and 3B). Because the Epe1 DNA-binding fusion protein activated transcription strongly in the absence of a prey, it could not be used to examine interactions. We next generated S. pombe strains harboring epitope-tagged versions of Epe1 and Cdt2 expressed from their endogenous loci. Coimmunoprecipitation experiments on

whole-cell extracts derived from these strains confirmed a biochemical interaction between Epe1 and Cdt2 (Figure 3C). These data support the view that Cul4-Ddb1Cdt2 directly recognizes Epe1 to promote its ubiquitylation and degradation in vivo. Cul4-Ddb1Cdt2 Confines Epe1 to Heterochromatin Boundaries To determine whether Cul4-Ddb1Cdt2 affects the levels of Epe1 on chromatin, we performed extensive ChIP experiments in cells expressing Epe1-FLAG. In agreement with a previous study (Zofall and Grewal, 2006), we found that Epe1 can be detected in WT cells at sites within the pericentromeric region (Figure 4A), the silent mating type locus (Figure 4B), and the right telomeric end of chromosome 2 (tel2R, Figure 4C). In addition, Epe1 is present at a meiotic gene, mei4, but not at other nearby euchromatic genes (Figure 4D). It is important to note that, however, the pattern of Epe1 within heterochromatin is not uniform. In agreement with its function in boundary formation, Epe1 is enriched at the margins of heterochromatin with distinct peaks coinciding with the heterochromatic boundaries flanking the outer repeats (at the IRC elements) and inner most repeats of centromere 1, the left and right boundaries of the silent mating type locus (IR-R/L), and the telomere-distal side of the telomeric tlh2+ locus. When we explored the chromatin profile of Epe1 in the ddb1D spd1D mutant, we observed a strong accumulation of Epe1 at all heterochromatic domains as well as the meiotic mei4 gene (Figures 4A–4D). These increases in chromatin-associated Epe1 were also observed in cdt2D spd1D cells (Figure 4; lower panels) but absent in spd1D single mutants (Figure S3). Importantly, the accumulation of Epe1 in mutant cells was not confined to the boundaries but was seen in heterochromatic regions that are relatively depleted of Epe1 in WT cells. In particular, we observed for chromatin-bound Epe1 an increase up to 7-fold in the body of the mat2/3 silent locus but only 2-fold at the IR-R/L boundary elements (Figure 4B). These results demonstrate that the altered Epe1 levels on chromatin do not merely reflect the increase in cellular Epe1 levels but indicate a significant change in the chromosomal distribution of Epe1 in absence of Cul4-Ddb1Cdt2. To understand the mechanisms that determine the heterochromatic distribution of Epe1, we compared its chromatin profile with the pattern of H3K9me2. Previous work demonstrated that Epe1 is recruited to H3K9 methyl marks by the HP1 proteins Swi6 and Chp2 (Sadaie et al., 2008; Zofall and Grewal, 2006), which bind preferentially to di- and trimethylated H3K9 and show a virtually identical chromatin distribution to H3K9me2 (Noma et al., 2001; Sadaie et al., 2008). Surprisingly, we found that in WT cells the chromatin distributions of Epe1 and H3K9me2 are quite disparate at every heterochromatic domain tested (Figures 5A–5C), implying that the recruitment to heterochromatin is not sufficient to explain the specific chromatin profile of Epe1. In striking contrast, the profiles of Epe1 and H3K9me2 are nearly indistinguishable in the ddb1D spd1D mutant for the centromere and the silent mating type locus (Figures 5D and 5E); both profiles become similar for the subtelomeric tel2R region as well (Figure 5F). These findings strongly suggest that although H3K9me2 mediates the initial recruitment of Epe1 to heterochromatin via HP1 proteins, the distribution of Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc. 45

db

T

T

W

W

sp

d

2 dt

c

d1

Δ

Δ

sp

B

Epe1-FLAG anti-FLAG anti-RNAPII

Epe1-FLAG WT

3 2 1 0

Δ T 1Δ W pd1 pd s s 1Δ t2Δ b cd dd

0 7.5 15 22.5 30 45 60 min after CHX

D

C

4

90 % Epe1 remaining

Epe1-FLAG RNAPII Epe1-FLAG

cdt2Δ spd1Δ

RNAPII

4 3 2 1 0

Δ Δ T W pd1 pd1 s s Δ 2Δ t cd

1 db

d

** **

100

E

**

*

RNAPII

ddb1Δ spd1Δ

epe1+ mRNA levels relative to WT

d1

1Δ

Epe1 protein levels relative to WT

Δ

A

80

** cdt2Δ spd1Δ

*

**

70

**

ddb1Δ spd1Δ

**

60

* WT

50 40 5 0 0

0

0.5

1

2

3

4

F

15

30

45

60

min after CHX

hr after HU

Epe1-FLAG WT

Myc-Cdt2

input

G

His pull down Δ d1

RNAPII

Epe1-FLAG (cdt2Δ spd1Δ)

100

anti-FLAG

RNAPII

125

T

W

T

dd

*

Δ

d1

sp

W

sp

T

Δ b1

dd

W

Epe1-FLAG His-Ubiquitin

Epe1-HisUb[n] Epe1

Epe1

(non-modified)

75

Myc-Cdt2 (WT) 50

Epe1-FLAG (WT)

25 0

0

1

2

3

4

hr after HU

His

Ub[n]

anti-His

% protein level

T

W

Epe1-FLAG

cdt2Δ spd1Δ

Δ b1

75

Myc-cdt2 (WT)

50

epe1-FLAG (WT) epe1-FLAG (cdt2Δ spd1Δ)

25 0 0

1

2

3

4

hr after HU

His

Ub

Epe1-Ub levels relative to WT

% mRNA level

100

1 0.5 0

Δ

T 1 W spd 1Δ b d

d

Figure 2. Cul4-Ddb1Cdt2 Promotes the Ubiquitylation and Degradation of the JmjC Protein Epe1 (A) Western blot of C-terminally tagged Epe1 (Epe1-FLAG) expressed from its endogenous locus. Loading control: RNA polymerase II CTD repeat (RNAPII). (B) Quantification of protein levels. Epe1-FLAG protein levels were normalized to RNAPII. Shown are mean values relative to WT with SEM of five independent biological experiments. (C) epe1+ mRNA levels. Shown are transcript levels relative to WT with SEM from independent experiments (n = 4–5). (D) Cycloheximide (CHX) chase experiments. For the ddb1D spd1D and cdt2D spd1D samples, half of the total protein amount was loaded to better visualize changes in the decay rates of Epe1. Loading control: RNAPII. (E) Quantification of Epe1 decay. Epe1 protein levels were normalized to RNAPII and plotted versus time after CHX addition (time = 0 was set to 100%). Dataare represented as mean ± SEM of independent experiments (n = 7–14) and fitted for exponential decay. Single and double asterisks indicate p values of < 0.05 and < 0.01, respectively (Student’s t test). (F) Protein levels after treatment with hydroxyurea (HU). Epe1-FLAG and Myc-Cdt2 were expressed from their endogenous loci and analyzed at the designated time points after HU treatment (20 mM) for protein (top panels) and mRNA (lower graph) levels. Upper graph: levels of Epe1-FLAG and Myc-Cdt2, normalized to RNAPII and plotted as percentage of the relative maximum protein level. Lower graph: mRNA levels of epe1-FLAG and Myc-cdt2 plotted as percentage of the maximum of mRNA level. (G) In vivo ubiquitylation of Epe1-FLAG in WT and ddb1D spd1D cells expressing 6His-ubiquitin. Input fraction (0.005%) and precipitated 6His-ubiquitin conjugates were analyzed by anti-FLAG (upper panels) and anti-His (lower panels) immunoblotting. Negative control: WT cells expressing untagged Epe1.

46 Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc.

A prey

pty

em

e1

t2

i6

Ep

Cd

Sw

bait

Cdt2

Swi6

B bait

β-gal activity relative to empty prey

12

Cdt2 Swi6

8

4

0

pty

em

e1

t2

Cd

i6

Sw

Ep prey

C input

FLAG-IP Epe1-FLAG

anti-Myc anti-FLAG

Myc-Cdt2 Cdt2

endogenous

Epe1

Figure 3. Cdt2 Physically Interacts with Epe1 (A) Plate yeast two-hybrid analysis. Photographs of plates were taken 1 (bottom panel) or 2 days (top panel) after exposure to X-gal. (B) Quantitative yeast two-hybrid analysis. b-gal activity was normalized to the empty prey for each bait and plotted for Cdt2 (blue) and Swi6 (gray). Error bars: standard deviation (SD) of three replicates. (C) Coimmunoprepcipation of Cdt2 with Epe1. Strains expressing endogenous levels of Epe1-CBP-23FLAG, Myc13-Cdt2, or both were subjected to antiFLAG immunoprecipitation. Input and immunoprecipitated material were analyzed by anti-Myc (top panel) and anti-FLAG (bottom panel) immunoblots. Note that the anti-Myc antibody slightly crossreacts with an unspecific band that comigrates with Myc13-Cdt2 seen in the untagged anti-Myc control lane of the input fraction.

Epe1 within heterochromatic domains, and in particular its restriction to boundaries, is shaped by its removal from specific heterochromatic regions by the action of the Cul4-Ddb1Cdt2 complex. Regulation of Epe1 by Cul4-Ddb1Cdt2 Acts in Parallel with the CK2-Swi6 Pathway Because Epe1 is tethered to heterochromatin by silencing factors, we tested whether the modification state of heterochro-

matin influenced its turnover. Phosphorylation of Swi6 by CK2 has been shown recently to inhibit the association of Epe1 with heterochromatin and to promote the binding of the SHREC effector complex (Shimada et al., 2009). Because CK2 mutants and cells lacking Cul4-Ddb1Cdt2 both display increased association of Epe1 with heterochromatin, we considered the hypothesis that they function in a single pathway in which phosphorylation of Swi6 by CK2 triggers the turnover of Epe1. This hypothesis makes three predictions: (1) Epe1 protein should accumulate in mutants of CK2, e.g., cells lacking its regulatory subunit Ckb1, (2) double mutants lacking ckb1D and the ubiquitin ligase should show the same increase in Epe1 association with heterochromatin as the single mutants, and (3) mutants lacking Cul4-Ddb1Cdt2 should display a decrease in the binding of SHREC to heterochromatin seen in ckb1D mutants. As shown in Figures 6A–6C and Figure S4A, we obtained data that contradicted each of these predictions. Epe1 does not accumulate in ckb1D cells (Figure 6A), the double mutants show more Epe1 association with heterochromatin than the single mutants (Figure 6B and Figure S4A), and SHREC occupancy is unaffected in ligase-deficient cells (Figure 6C and Figure S4B). These data indicate that the two mechanisms operate in parallel (rather than in a single pathway) to regulate Epe1. Given that Swi6 phosphorylation by CK2 is not required for Epe1 turnover, we examined whether Swi6 was required for Epe1 regulation. We first confirmed and extended previous data demonstrating that Swi6 is required for the association of Epe1 with heterochromatin, finding that at boundaries, Epe1 association was either completely (IR-L/R) or nearly completely (IRC1) eliminated in swi6D cells (Figure 6D and Figures S4C and S4D). Next we tested whether Epe1 levels accumulate to those seen in ddb1D and cdt2D mutants when swi6+ is deleted. We found only a subtle increase in Epe1 levels in swi6D cells, indicating that Swi6 is not critical for Epe1 turnover (Figures 6E and 6F). These results demonstrate that heterochromatin association is not required for Epe1 turnover. Nonetheless, given that Epe1 is a chromatin-bound protein (Sadaie et al., 2008; Shimada et al., 2009; Zofall and Grewal, 2006), it seems likely that its ubiquitylation and its regulation occur in the context of chromatin (see Discussion). Regulation of Epe1 Is Sufficient to Explain the Role of Cul4-Ddb1Cdt2 in Heterochromatin Formation Next, by deleting epe1+ in a ddb1D spd1D strain, we examined whether the misregulation of Epe1 accounts for the defects in heterochromatin formation observed in cells lacking Ddb1. Indeed, by using silencing reporter assays, we found that the silencing defect of ddb1D spd1D cells was suppressed in the ddb1D spd1D epe1D triple mutant at the pericentromeric region and the mating type locus (Figure 7A). This suppression was specific for the Cul4-Ddb1Cdt2 pathway, as deletion of epe1+ did not suppress the silencing defect of cells lacking Rik1, the Ddb1 paralog in the Clr4-associated ubiquitin ligase Cul4-Rik1Dos1.

Note that a fraction of nonubiquitylated Epe1 can also be detected in the pull-down samples, probably due to the presence of several His-residue clusters within the Epe1 protein. Graph below shows the mean values of the ubiquitylation level of Epe1 relative to WT of three independent experiments (error bars = SEM). Total Epe1-ubiquitin conjugates (without the nonmodified Epe1 fraction) was quantified by densitometry of anti-FLAG western blots and normalized for the input level. See also Figure S2.

Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc. 47

A SPAP7G.06

IRC

otr1L dh

20

imr1L cnt imr1R

dg

WT

imr1R

SPAP4C9.01c otr1R IRC dg

ddb1Δ spd1Δ

IR-L mat2P

cen-H

C

SPBC1711.03 mat3M IR-R

D

tel2R LTR

tlh2

euchromatin cdk9

mei4

act1 mok12

12

7

15 10

15

6 5

8

10

4

10

6 3 4

5 5

0 3.76

3

Epe1-FLAG ChIP mut:WT ratio [log2]

mat locus

dh

+

Epe1-FLAG ChIP [fold enrichment to act1 ]

B

centromere 1

3.77 3.78 3.79 chromosome pos. [Mb]

ddb1Δ spd1Δ

cdt2Δ spd1Δ

0 2.13

2.14 2.15 chromosome pos. [Mb]

2

2

1

0

0 1.47 1.475 1.48 chromosome pos. [Mb]

4.52 4.53 chromosome pos. [Mb]

3

3

3

2

2

2

2

1

1

1

1

0

0

0

0

-1

-1

-1

-1

4.52 4.53 chromosome pos. [Mb]

1.47 1.475 1.48 chromosome pos. [Mb]

3.76

3.77 3.78 3.79 chromosome pos. [Mb]

2.13

2.14 2.15 chromosome pos. [Mb]

Figure 4. Deletion of ddb1+ Causes Accumulation of Epe1 within Heterochromatin Domains ChIP analysis of Epe1 at centromere 1 (A), the silent mating type region (B), the subtelomeric region of telomere 2 (C), and a meiotic gene locus (D) in WT (blue) and ddb1D spd1D cells (red). Upper panels: ChIP signals normalized to act1+. Lower panels: fold enrichment of Epe1 in ddb1D spd1D (red) and cdt2D spd1D (dark red) relative to WT. Data are represented as mean ± SEM of three independent experiments. See also Figure S3.

Furthermore, the suppression of the silencing defect of ddb1D spd1D was due to the loss of Epe1, as complementation of the epe1D mutation by reintroducing epe1+ completely reverted the suppression phenotype (Figure 7A). Consistent with these silencing reporter assay results, RT-qPCR measurements revealed that the levels of ura4+ transcripts originating from the mat3M::ura4+ locus were reduced in ddb1D spd1D epe1D cells to WT levels (Figure 7B). In agreement with a previous study (Trewick et al., 2007), we found that epe1D single mutants display a quantitative increase in centromeric transcripts (Figure S5), precluding a similar analysis at these regions. We instead probed the suppression of the ddb1D-associated silencing defects by investigating the level of H3K9me2 at pericentromeric and telomeric regions, which are decreased in cells lacking Ddb1 (Figure 1F). Remarkably, we observed that H3K9me2 levels were restored to WT levels in a ddb1D spd1D epe1D mutant at the pericentromeric region (Figure 7C). The H3K9me2 defect was also suppressed in this triple mutant at tel2R to levels seen in an epe1D single mutant. These results indicate that the reduced levels of H3K9me2 at these heterochromatic loci are caused by misregulation of Epe1 in cells lacking Cul4-Ddb1Cdt2. Collectively these 48 Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc.

findings demonstrate that degradation of Epe1 is sufficient to explain the requirement of Cul4-Ddb1Cdt2 for silencing. DISCUSSION Our study identified a regulatory mechanism required for proper boundary architecture and heterochromatic silencing in S. pombe. This mechanism involves the conserved ubiquitin ligase Cul4-Ddb1Cdt2, which targets the JmjC protein Epe1 for ubiquitin-dependent degradation. Epe1 antagonizes the spread of heterochromatin and has a potential role in boundary formation (Ayoub et al., 2003), yet it is found within heterochromatic domains and associates directly with the H3K9me-binding protein Swi6 (Zofall and Grewal, 2006). This paradoxical finding raises the fundamental question of how Epe1 is precluded from interfering with heterochromatin formation. Our findings demonstrate that Cul4-Ddb1Cdt2 controls the chromosomal landscape of Epe1 in a manner that substantially restricts its accumulation to heterochromatic boundaries by limiting its spreading into the bodies of heterochromatic domains. This heterochromatin-shaping function of Cul4-Ddb1Cdt2 is required for silencing.

IRC

otr1L dh

120

imr1L cnt imr1R imr1R

dg

dg

IR-L mat2P

cen-H

tel2R

SPBC1711.03 mat3M IR-R

LTR

120

120

100

100

100

80

80

80

60

60

60

40

40

40

20

20

20

H3-K9m2

3.76

3.77

Epe1

3.78

0 2.13

3.79

chromosome pos. [Mb]

0 2.14

4.52

2.15

chromosome pos. [Mb]

F

120

120

100

100

100

80

80

80

60

60

60

40

40

40

20

20

20

H3-K9m2

Epe1

0 3.76

3.77

3.78

3.79

4.53

chromosome pos. [Mb]

E

D 120

tlh2

dh

0

relative chromatin distribution [ % to maximum ]

C

mat locus

SPAP4C9.01c otr1R IRC

WT

SPAP7G.06

relative chromatin distribution [ % to maximum ]

B

centromere 1

0 2.13

chromosome pos. [Mb]

ddb1Δ spd1Δ

A

0 2.14

2.15

chromosome pos. [Mb]

4.52

4.53

chromosome pos. [Mb]

Figure 5. Epe1 Is Confined to Heterochromatic Boundaries in Wild-Type but Spreads through Entire Heterochromatin Domains in Cells Lacking Ddb1 Relative chromatin distribution of Epe1 (red) and H3K9me2 (green) within heterochromatic regions in WT (A–C) and ddb1D spd1D cells (D–F). ChIP for Epe1-FLAG and H3K9me2 were performed as described in Figure 1F and Figure 4. ChIP data were act1+ normalized and median centered. Data are represented as mean ± SEM of three independent experiments relative to the maximum (100%) of each heterochromatic region.

A Conserved Ubiquitin Ligase Promotes Silencing by Targeting a Silencing Inhibitor We identified Ddb1 and Cdt2 as silencing factors in targeted knockout screens for pericentromeric silencing and demonstrated their requirement for the integrity of other heterochromatic domains. Mutants of ddb1+ and cdt2+ are indistinguishable in their silencing defects and are epistatic to each other (Figure 1). Ddb1 and Cdt2 are highly conserved proteins (25% and 26% identity, 47% and 44% similarity, respectively, between the fission yeast and human homologs). Both proteins were originally identified as a heterodimeric factor recruited to DNA upon damage by ultraviolet irradiation (UV) (Dualan et al., 1995; Keeney et al., 1993), and mutations in the DCAFs DDB2 and CSA are associated with the human diseases Xeroderma pigmentosum complementation group E (XP-E) and the Cockayne Syndrome (CS), respectively (O’Connell and Harper, 2007). Although more than 50 different DCAFs have been identified (Lee and Zhou, 2007), the number of known substrates is

significantly smaller, reflecting the difficulty of identifying substrates of ubiquitin ligases. Notably, the known substrates are predominantly chromatin-associated proteins, suggesting a specialized role for Cul4-Ddb1 ligases in nuclear processes (O’Connell and Harper, 2007). Here we show that Cul4-Ddb1Cdt2 targets Epe1 in vivo (Figure 2) and that the putative substrate recognition subunit Cdt2 interacts with Epe1 (Figure 3). In WT cells, Epe1 is polyubiquitylated and degraded by an initial rapid and a late slow decay. Conversely, in cells lacking Ddb1 or Cdt2, Epe1 is significantly stabilized. Ubiquitylation of Epe1 is not completely abolished in ddb1D mutant cells, and only the rapid decay component is abrogated in the mutants, suggesting that other ligases likely also target Epe1. Nonetheless, the regulation of Epe1 by Cul4-Ddb1Cdt2 appears to be sufficient to explain the role of the ligase in silencing: The silencing defect at the mat3M locus and the decrease of H3K9 methylation at pericentromeric regions in ddb1D mutants are completely suppressed by removal of Epe1. Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc. 49

B

1

Epe1-FLAG T

W

T 6Δ W swi

2 1 0

IR-L/R

SPBCPT2R1.07c

(mat locus)

10

4

8

3

6

2

4

1

2

0

0 W

E

F

(subtelomeric)

5

T

T Δ T Δ W pd1 W pd1 s s Δ Δ b1 b1 dd dd

T 6Δ W swi

Epe1 protein levels [rel. to WT]

0

3

Clr1-FLAG

Δ T Δ Δ W pd1 kb1 pd1 s c s Δ Δ b1 b1 dd dd Δ b1 ck

IRC1

5

5

Epe1-FLAG

(centromeric)

10

10

0

T 1Δ 1Δ W ckb pd s Δ b1 dd

15

15

W

T

T 6Δ W swi

4 3 2 1 0

Δ Δ 1Δ T W wi6 pd1 pd s s Δs Δ 1 b1 db dd Δ d i6 sw

epe1+ mRNA levels [rel. to WT]

2

Clr1-FLAG ChIP [rel. to act1 + ]

3

D

FLAG ChIP [rel. to act1 + ]

mat2P mat3M 4

20

4

0

C

cen-dh/dg FLAG ChIP [rel. to act1+ ]

Epe1 protein levels [rel. to WT]

A

4 3 2 1 0

Δ 1Δ T W wi6 pd s s Δ b1 dd

Figure 6. Cul4-Ddb1Cdt2 -Dependent Degradation of Epe1 Acts Independently of HP1 Phoshorylation by Casein Kinase II (A) Epe1-FLAG protein levels from quantified western blots (normalized to RNAPII). Shown are mean values relative to WT with SEM from five independent experiments (except ddb1D spd1D with n = 2; error shows the variation from the mean). (B) ChIP analysis of Epe1-FLAG levels at centromere 1 (region between cen-dh and -dg). ChIP signals were normalized to act1+. Shown are mean values with SD of three parallel IP samples of one representative experiment. (C) ChIP analysis of Clr1-FLAG levels at the silent mating type region. Shown are mean values of two independent experiments with error bars representing the variation from the mean. (D) ChIP analysis of Epe1-FLAG levels at the outer boundary of centromere 1 (left panel), at inverted repeats of silent mating type region (middle panel), and at subtelomeric locus telomere-distal of tlh2 (right panel) in WT and swi6D cells. Shown are mean values with SD of three parallel IP samples of one representative experiment. (E) Epe1-FLAG protein levels. For comparison, the level of Epe1-FLAG in ddb1D spd1D (from Figure 2B) is also shown. Shown are mean values relative to WT with error bars (SEM) from independent experiments (n = 4–5). (F) epe1-FLAG mRNA levels. For comparison, the level of epe1-FLAG in ddb1D spd1D (from Figure 2C) is displayed. Shown are mean values relative to WT with error bars (SEM) from independent experiments (n = 4–8). See also Figure S4.

Sculpting Heterochromatin by Preventing the Internal Spread of a Silencing Inhibitor Cul4-Ddb1Cdt2 affects the Epe1 levels on chromatin consistent with its known role in regulating other chromatin-associated substrates. We observed that Epe1 is located predominantly at the heterochromatic boundaries in WT cells, in agreement with the notion that Epe1 plays a role in boundary formation (Ayoub et al., 2003; Zofall and Grewal, 2006). In striking contrast, Epe1 accumulates to high levels in the bodies of heterochromatic domains in cells lacking Cul4-Ddb1Cdt2. Although our results show that turnover of Epe1 does not require its association with heterochromatin, several pieces of evidence suggest that its regulation likely takes place on chromatin (Figures 7D and 7E). First, we confirmed previous findings that show that Epe1 does not appear to have affinity for boundaries in the absence of Swi6; thus, increasing Epe1 levels per se 50 Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc.

would not be expected to result in its enrichment at boundaries. Second, changes in Epe1 levels at chromatin are not uniform in cells lacking the ubiquitin ligase but instead show a distinct pattern: a strong accumulation of Epe1 within the bodies of the heterochromatic domains but only a modest increase of Epe1 at the boundaries (Figure 4). This is not because the association of Epe1 with boundary chromatin is saturated under these conditions, as we have found that cells also lacking ckb1D display even higher levels of Epe1 on chromatin (Figure S4A). Third, there is only a 3-fold increase of the total pool of Epe1 in ddb1D and cdt2D mutant cells, whereas the chromatin-bound Epe1 accumulates up to 7-fold (Figure 4). Fourth, whereas the distribution of Epe1 differs substantially from the chromatin profile of H3K9me2 in WT cells, its chromatin localization is nearly identical to the H3K9me2 pattern in absence of Ddb1 and no longer shows a preference to the

N/S

A

5-FOA

WT epe1Δ ddb1Δ spd1Δ ddb1Δ spd1Δ epe1Δ ddb1Δ spd1Δ epe1Δ + epe1 rik1Δ rik1Δ epe1Δ

imr1L::ura4

B

N/S

+

5-FOA

mat3M::ura4

+

centromere 1

C

SPAP7G.06

IRC

otr1L

imr1L

cnt

dg

dh

tel2R

imr1R

otr1R

IRC

tlh2

250

+

H3-K9me2 ChIP [fold enrichment to act1 ]

+

50

20

+

ura4 transcript level [rel. enchrichment to WT]

LTR

dh

dg

700

mat3M::ura4

100

SPAP4C9.01c

10 0

600 200 500 150

400 300

100

200 50

100

b1 dd W Δ b1 ep T sp Δ e d1 sp 1Δ Δ d1 ep Δ e1 rik 1Δ rik Δ ep 1Δ e1 Δ

0

0 3.76

3.77

3.78

3.79

chromosome pos. [Mb]

dd

WT epe1Δ

D

4.52

4.53

chromosome pos. [Mb] ddb1Δ spd1Δ ddb1Δ spd1Δ epe1Δ

E Epe1

Boundary Environment

Epe1

HP1 K9me

RING RING

Cul4

Cul4

Ddb1

Epe1

Ddb1

Cdt2 Ub

Epe1

Ub

Ub

Epe1

Cdt2

Epe1

Epe1 Epe1 HP1 HP1

Ub[n]

Epe1 Epe1

SILENCING

Epe1

CK2 SHREC

P HP1 EC

heterochromatin

P HP1

EC

Figure 7. Deletion of epe1+ Suppresses the Silencing Defect of Cells Lacking Ddb1

(A) Reporter gene assays. N/S, nonselective; 5-FOA, 50 -fluoroorotic acid. (B) RT-qPCR of ura4+ transcript levels derived from mat3M::ura4+. Shown are mean values relative to WT ± SEM of three independent experiments. (C) ChIP analysis of H3K9me2 at centromere 1 and the right arm subtelomeric region of chromosome 2. Shown are mean values ± SD of three parallel IP samples of one representative experiment. (D) Model for boundary formation through recruitment of Epe1 to heterochromatin by HP1 proteins and its subsequent removal from central heterochromatic domains by Cul4-Ddb1Cdt2. See text for details. (E) Independent pathways regulate Epe1 at chromatin. See text for details. See also Figure S5.

boundaries (Figure 5). Taken together, these findings strongly suggest that Epe1 by itself does not have any particular affinity to boundary elements, but rather that its removal from the body of heterochromatin explains its relative enrichment at boundaries.

A corollary to this model is that Epe1 must be protected from removal by Cul4-Ddb1Cdt2 at boundaries. Much evidence points to a role for nuclear envelope tethering as a requirement for boundary function (Ishii and Laemmli, 2003; Noma et al., 2006; Yusufzai et al., 2004). It is thus possible that subnuclear Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc. 51

localization of boundary regions limits their accessibility to Cul4Ddb1Cdt2 or the proteasome. Such a mechanism together with the ability of Swi6 to recruit Epe1 to heterochromatin could explain the enrichment of Epe1 observed at boundaries. Posttranslational modification or the presence of auxiliary factors could also play a role in directing Cul4-Ddb1Cdt2 to Epe1. Mutants defective in phosphorylation of Swi6 by CK2 display increased accumulation of Epe1 and decreased accumulation of the SHREC ATPase/HDAC complex on chromatin (Shimada et al., 2009). Together with our observations, these published data would be compatible with a model in which phosphorylation of Swi6 triggers Epe1 turnover. However, our analysis demonstrates decisively that Swi6 phosphorylation and ubiquitylation of Epe1 by Cul4-Ddb1Cdt2 act in different pathways to regulate the heterochromatin association of Epe1 (Figure 7E). That this protein is subjected to multiple layers of regulation is striking and emphasizes the concept that tightly regulating this antisilencing factor is critical for maintaining heterochromatic domains. Regulation of the Activity of Epe1 by Defining Its Distribution within Heterochromatin The barrier function of Epe1 correlates with its spatial restriction to the boundaries. Conversely, when Epe1 accumulates within heterochromatic domains due to the absence of Cul4-Ddb1Cdt2, lack of phosphorylation of Swi6, or overexpression of Epe1, it acts as an antagonist of silencing (Shimada et al., 2009; Zofall and Grewal, 2006). Interestingly, mutants of epe1+ affect Pol II-dependent transcription through heterochromatin and are perturbed in their levels of heterochromatic siRNAs (Trewick et al., 2007; Zofall and Grewal, 2006). These observations may point to an additional role of Epe1 besides its barrier function that is associated with the RNAi-dependent pathway of heterochromatin formation. Indeed, we observed within the body of heterochromatic regions detectable amounts of Epe1 above background levels (Figure 4). These low levels of chromatinbound Epe1 may represent the pool that is deposited prior to its removal by Cul4-Ddb1Cdt2. Considering that the processes of Pol II-dependent transcription through heterochromatin and siRNA formation are restricted to S phase (Chen et al., 2008; Kloc et al., 2008) and are also affected by Epe1 (Zofall and Grewal, 2006), it is possible that targeting of Epe1 by Cul4Ddb1Cdt2 is temporally controlled. This notion is supported by the finding that Cdt2, which itself is an unstable protein, is expressed only within a short time window during S phase (Liu et al., 2005; Oliva et al., 2005). In such a scenario, initial tethering of Epe1 to Swi6 would stimulate the binding of Pol II to heterochromatin and thus the formation of siRNAs during S phase; subsequent removal of Epe1 by Cul4-Ddb1Cdt2 would then allow assembly of heterochromatin. General Role of CRLs in Silencing The general significance of ubiquitylation in regulating heterochromatin formation is highlighted by the specialized CRL Cul4-Rik1Dos1/2, which is associated with the histone methyltransferase Clr4 in the CLRC complex and is required for silencing. The biologically relevant substrate of this E3 and its specific role in heterochromatin formation have not been eluci52 Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc.

dated. Orthologs of Rik1 have not been identified in other eukaryotes so far; however, the requirement of coupling E3 activity with H3K9 methylation seems to be conserved. A recent study demonstrated that mutants of Cul4 and Ddb1 homologs in N. crassa are completely deficient in H3K9 methylation analogous to rik1 mutants in S. pombe (Zhao et al., 2010). Moreover Cul4 is associated with the corresponding H3K9 histone methyltransferase, suggesting that a homologous Cul4-Ddb1DCAF complex replaces the role of Cul4-Rik1Dos1 in this fungal species (Zhao et al., 2010). Intriguingly, Ddb1 and Cullin-4A were also found to be components of the CEN-complex, which associates with the centromere-specific histone H3 CENP-A in human cells (Obuse et al., 2004), suggesting a conserved role in chromatin regulation. Whether CRLs of N. crassa and mammals target inhibitory substrates analogous to Epe1 remains to be investigated. EXPERIMENTAL PROCEDURES Yeast Strains, Plasmids, and Techniques Standard media and genome engineering methods were used. 5-FOA media contained 1 g/l 50 -fluoroorotic acid. Synthetic complete (SC) media minus the corresponding amino acid were used for drop-out media. EMM-leu media were used for growing strains harboring pREP1 plasmids. Strains are listed in Table S2. Library Construction and Screen Gene disruptions were performed in an imr1L(NcoI)::ura4 otr1R(SphI)::ade6K P(h+) reporter strain (Ekwall et al., 1999). Yeast Two-Hybrid Analysis Plasmids containing fusion proteins of Swi6, Cdt2, and Epe1 (described in Table S3) were transformed into EGY48 (Golemis et al., 2009). Cultures were grown overnight in SC-his-trp-ura +2% raffinose, plated onto SC-his-trpura +1% raffinose +2% galactose, and grown for 2 days at 30 C. Cells were permeabilized by chloroform and overlayed with top agar containing X-gal as described (Richter et al., 2007). For liquid assays, overnight cultures were diluted 1:20 and grown in SC-his-trp-ura +1% raffinose +2% galactose for another 4 hr. b-galactosidase liquid assays were performed as described (Shock et al., 2009), except that 20 ml each cell culture and permeabilization buffer were used. Chromatin Immunoprecipitation ChIP experiments were performed essentially as described (Nobile et al., 2009). Unless otherwise noted, cells were crosslinked with 1% formaldehyde for 20 min at 30 C. To increase the ChIP sensitivity, in Figure 6D and Figures S4B–S4F, crosslinking was performed by subsequent treatment of 10 mM dimethyl adipimidate and 1.5% formaldehyde as described (Kurdistani and Grunstein, 2003), except that formaldehyde crosslinking was restricted to 30 min. Epe1-FLAG, Clr1-FLAG, and anti-H3K9me2 were immunoprecipitated with 2–5 mg antibody (anti-FLAG, Sigma F3165; anti-H3K9me2, Abcam ab 1220) from lysates corresponding to 50–75 optical density 600 (OD600) (Epe1-FLAG, Clr1-FLAG) and 15–25 OD600 (H3K9me2) of cells. Immunoprecipitated DNA was quantified by real-time PCR (qPCR) with primers listed in Table S4 and normalized against act1+. RNA Extraction and RT-qPCR Analyses RT-qPCR experiments were carried out as previously described (Rougemaille et al., 2008), except that RNA samples were DNaseI-treated with DNA-free kit (Ambion). Ten micrograms of RNA was used in standard RT reactions using oligo[(dT)20-N] primers. cDNAs were quantified by qPCR with the primers listed in Table S4 and normalized against act1+. Immunotechniques For examination of protein levels, extracts were prepared under denaturing conditions (Knop et al., 1999). Cycloheximide (CHX) chase experiments

were performed as described (Braun et al., 2002) except that 0.15 mg/ml CHX was used as final concentration. Lysates corresponding to 1 OD600 of cells were analyzed by immunoblotting with anti-FLAG (Sigma, P3165) and antiRNA polymerase II carboxy-terminal domain (CTD) repeat (Abcam ab817) antibodies diluted 1:1000 and 1:8000, respectively, in blocking solution (LI-COR). For detection and quantification, an infrared imaging system (Odyssey, Li-COR) and the corresponding software were used. Details of coimmunoprecipitation experiments can be found in the Extended Experimental Procedures. Ubiquitin Pull-Down Experiments Expression of nmt1 promoter-driven 6His-ubiquitin (pREP1-6His-Ubi) was performed as described. Thirty to Forty-five minutes prior to harvest, cells were treated with 5 mM NEM added directly to the growth medium. Protein extraction and binding of ubiquitin conjugates were done under denaturing conditions essentially as described (Sacher et al., 2005). Further details can be found in the Extended Experimental Procedures. SUPPLEMENTAL INFORMATION Supplemental Information includes Extended Experimental Procedures, five figures, and four tables and can be found with this article online at doi:10.1016/j.cell.2010.11.051. ACKNOWLEDGMENTS We thank Michael Rape, Stefan Jentsch, Geeta Narlikar, David Morgan, members of their labs, and Ulrike Boettcher for critical reading of the manuscript. We are grateful to Danesh Moazed, Karl Ekwall, and Takashi Toda for strains, plasmids, and protocols. S.B., M. Rougemaille, and S.S. were supported by postdoctoral fellowships of the German Research Foundation (BR 3511/1-1), the Human Frontier Science Program, and the Leukemia and Lymphoma Society, respectively. J.F.G. was supported by an NIH/NIGMS IMSD predoctoral fellowship (R25-GM56847). This work was supported by a grant to H.D.M. from the National Institutes of Health (GM071801) and a scholar’s award to H.D.M. from the Leukemia and Lymphoma Society. S.B. and H.D.M. designed the study. S.B., M. Rowley, M. Rougemaille, and S.S. constructed and characterized the deletion library and strains. S.B. and M. Rowley performed the screen. J.F.G. designed and performed the experiments shown in Figure 3. S.B. performed the experiments shown in Figure 1, Figure 2, Figure 4, Figure 5, and Figure 6. S.B. and H.D.M. wrote the manuscript. All authors contributed to editing the manuscript. Received: April 17, 2010 Revised: October 14, 2010 Accepted: November 18, 2010 Published: January 6, 2011 REFERENCES Appelgren, H., Kniola, B., and Ekwall, K. (2003). Distinct centromere domain structures with separate functions demonstrated in live fission yeast cells. J. Cell Sci. 116, 4035–4042. Ayoub, N., Noma, K.-I., Isaac, S., Kahan, T., Grewal, S.I.S., and Cohen, A. (2003). A novel jmjC domain protein modulates heterochromatization in fission yeast. Mol. Cell. Biol. 23, 4356–4370. Bondar, T., Ponomarev, A., and Raychaudhuri, P. (2004). Ddb1 is required for the proteolysis of the Schizosaccharomyces pombe replication inhibitor Spd1 during S phase and after DNA damage. J. Biol. Chem. 279, 9937–9943.

Chen, E.S., Zhang, K., Nicolas, E., Cam, H.P., Zofall, M., and Grewal, S.I.S. (2008). Cell cycle control of centromeric repeat transcription and heterochromatin assembly. Nature 451, 734–737. Dualan, R., Brody, T., Keeney, S., Nichols, A.F., Admon, A., and Linn, S. (1995). Chromosomal localization and cDNA cloning of the genes (DDB1 and DDB2) for the p127 and p48 subunits of a human damage-specific DNA binding protein. Genomics 29, 62–69. Ekwall, K., Cranston, G., and Allshire, R.C. (1999). Fission yeast mutants that alleviate transcriptional silencing in centromeric flanking repeats and disrupt chromosome segregation. Genetics 153, 1153–1169. Gaszner, M., and Felsenfeld, G. (2006). Insulators: exploiting transcriptional and epigenetic mechanisms. Nat. Rev. Genet. 7, 703–713. Golemis, E.A., Serebriiskii, I., Finley, R.L., Kolonin, M.G., Gyuris, J., and Brent, R. (2009). Interaction trap/two-hybrid system to identify interacting proteins. Current Protocols in Protein Science, Chapter 19, Unit19.12. Grewal, S.I. (2010). RNAi-dependent formation of heterochromatin and its diverse functions. Curr. Opin. Genet. Dev. 20, 134–141. Holmberg, C., Fleck, O., Hansen, H.A., Liu, C., Slaaby, R., Carr, A.M., and Nielsen, O. (2005). Ddb1 controls genome stability and meiosis in fission yeast. Genes Dev. 19, 853–862. Hong, E.-J.E., Ville´n, J., Gerace, E.L., Gygi, S.P., and Moazed, D. (2005). A cullin E3 ubiquitin ligase complex associates with Rik1 and the Clr4 histone H3-K9 methyltransferase and is required for RNAi-mediated heterochromatin formation. RNA Biol. 2, 106–111. Ishii, K., and Laemmli, U.K. (2003). Structural and dynamic functions establish chromatin domains. Mol. Cell 11, 237–248. Jackson, S., and Xiong, Y. (2009). CRL4s: the CUL4-RING E3 ubiquitin ligases. Trends Biochem. Sci. 34, 562–570. Jin, J., Arias, E.E., Chen, J., Harper, J.W., and Walter, J.C. (2006). A family of diverse Cul4-Ddb1-interacting proteins includes Cdt2, which is required for S phase destruction of the replication factor Cdt1. Mol. Cell 23, 709–721. Keeney, S., Chang, G.J., and Linn, S. (1993). Characterization of a human DNA damage binding protein implicated in xeroderma pigmentosum E. J. Biol. Chem. 268, 21293–21300. Kloc, A., Zaratiegui, M., Nora, E., and Martienssen, R. (2008). RNA interference guides histone modification during the S phase of chromosomal replication. Curr. Biol. 18, 490–495. Knop, M., Siegers, K., Pereira, G., Zachariae, W., Winsor, B., Nasmyth, K., and Schiebel, E. (1999). Epitope tagging of yeast genes using a PCR-based strategy: more tags and improved practical routines. Yeast 15, 963–972. Kurdistani, S.K., and Grunstein, M. (2003). In vivo protein-protein and proteinDNA crosslinking for genomewide binding microarray. Methods 31, 90–95. Lee, J., and Zhou, P. (2007). DCAFs, the missing link of the CUL4-DDB1 ubiquitin ligase. Mol. Cell 26, 775–780. Liu, C., Poitelea, M., Watson, A., Yoshida, S.-h., Shimoda, C., Holmberg, C., Nielsen, O., and Carr, A.M. (2005). Transactivation of Schizosaccharomyces pombe cdt2+ stimulates a Pcu4-Ddb1-CSN ubiquitin ligase. EMBO J. 24, 3940–3951. Matsuyama, A., Arai, R., Yashiroda, Y., Shirai, A., Kamata, A., Sekido, S., Kobayashi, Y., Hashimoto, A., Hamamoto, M., Hiraoka, Y., et al. (2006). ORFeome cloning and global analysis of protein localization in the fission yeast Schizosaccharomyces pombe. Nat. Biotechnol. 24, 841–847.

Braun, S., Matuschewski, K., Rape, M., Thoms, S., and Jentsch, S. (2002). Role of the ubiquitin-selective CDC48(UFD1/NPL4)chaperone (segregase) in ERAD of OLE1 and other substrates. EMBO J. 21, 615–621.

Nakanishi, S., Lee, J.S., Gardner, K.E., Gardner, J.M., Takahashi, Y.-h., Chandrasekharan, M.B., Sun, Z.-W., Osley, M.A., Strahl, B.D., Jaspersen, S.L., et al. (2009). Histone H2BK123 monoubiquitination is the critical determinant for H3K4 and H3K79 trimethylation by COMPASS and Dot1. J. Cell Biol. 186, 371–377.

Cam, H.P., Sugiyama, T., Chen, E.S., Chen, X., Fitzgerald, P.C., and Grewal, S.I.S. (2005). Comprehensive analysis of heterochromatin- and RNAi-mediated epigenetic control of the fission yeast genome. Nat. Genet. 37, 809–819.

Nobile, C.J., Nett, J.E., Hernday, A.D., Homann, O.R., Deneault, J.-S., Nantel, A., Andes, D.R., Johnson, A.D., and Mitchell, A.P. (2009). Biofilm matrix regulation by Candida albicans Zap1. PLoS Biol. 7, e1000133.

Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc. 53

Noma, K., Allis, C.D., and Grewal, S.I. (2001). Transitions in distinct histone H3 methylation patterns at the heterochromatin domain boundaries. Science 293, 1150–1155.

Scott, K.C., Merrett, S.L., and Willard, H.F. (2006). A heterochromatin barrier partitions the fission yeast centromere into discrete chromatin domains. Curr. Biol. 16, 119–129.

Noma, K.-I., Cam, H.P., Maraia, R.J., and Grewal, S.I.S. (2006). A role for TFIIIC transcription factor complex in genome organization. Cell 125, 859–872.

Shimada, A., Dohke, K., Sadaie, M., Shinmyozu, K., Nakayama, J.-I., Urano, T., and Murakami, Y. (2009). Phosphorylation of Swi6/HP1 regulates transcriptional gene silencing at heterochromatin. Genes Dev. 23, 18–23.

O’Connell, B.C., and Harper, J.W. (2007). Ubiquitin proteasome system (UPS): what can chromatin do for you? Curr. Opin. Cell Biol. 19, 206–214. Obuse, C., Yang, H., Nozaki, N., Goto, S., Okazaki, T., and Yoda, K. (2004). Proteomics analysis of the centromere complex from HeLa interphase cells: UV-damaged DNA binding protein 1 (DDB-1) is a component of the CENcomplex, while BMI-1 is transiently co-localized with the centromeric region in interphase. Genes Cells 9, 105–120. Oliva, A., Rosebrock, A., Ferrezuelo, F., Pyne, S., Chen, H., Skiena, S., Futcher, B., and Leatherwood, J. (2005). The cell cycle-regulated genes of Schizosaccharomyces pombe. PLoS Biol. 3, e225. Ralph, E., Boye, E., and Kearsey, S.E. (2006). DNA damage induces Cdt1 proteolysis in fission yeast through a pathway dependent on Cdt2 and Ddb1. EMBO Rep. 7, 1134–1139. Richter, C., West, M., and Odorizzi, G. (2007). Dual mechanisms specify Doa4-mediated deubiquitination at multivesicular bodies. EMBO J. 26, 2454–2464. Rougemaille, M., Shankar, S., Braun, S., Rowley, M., and Madhani, H.D. (2008). Ers1, a rapidly diverging protein essential for RNA interference-dependent heterochromatic silencing in Schizosaccharomyces pombe. J. Biol. Chem. 283, 25770–25773. Sacher, M., Pfander, B., and Jentsch, S. (2005). Identification of SUMOprotein conjugates. Methods Enzymol. 399, 392–404. Sadaie, M., Kawaguchi, R., Ohtani, Y., Arisaka, F., Tanaka, K., Shirahige, K., and Nakayama, J.-I. (2008). Balance between distinct HP1 family proteins controls heterochromatin assembly in fission yeast. Mol. Cell. Biol. 28, 6973–6988.

54 Cell 144, 41–54, January 7, 2011 ª2011 Elsevier Inc.

Shock, T.R., Thompson, J., Yates, J.R., and Madhani, H.D. (2009). Hog1 mitogen-activated protein kinase (MAPK) interrupts signal transduction between the Kss1 MAPK and the Tec1 transcription factor to maintain pathway specificity. Eukaryot. Cell 8, 606–616. Sun, Z.-W., and Allis, C.D. (2002). Ubiquitination of histone H2B regulates H3 methylation and gene silencing in yeast. Nature 418, 104–108. Talbert, P.B., and Henikoff, S. (2006). Spreading of silent chromatin: inaction at a distance. Nat. Rev. Genet. 7, 793–803. Trewick, S.C., Minc, E., Antonelli, R., Urano, T., and Allshire, R.C. (2007). The JmjC domain protein Epe1 prevents unregulated assembly and disassembly of heterochromatin. EMBO J. 26, 4670–4682. Tsukada, Y.-i., Fang, J., Erdjument-Bromage, H., Warren, M.E., Borchers, C.H., Tempst, P., and Zhang, Y. (2006). Histone demethylation by a family of JmjC domain-containing proteins. Nature 439, 811–816. Wang, H., Wang, L., Erdjument-Bromage, H., Vidal, M., Tempst, P., Jones, R.S., and Zhang, Y. (2004). Role of histone H2A ubiquitination in Polycomb silencing. Nature 431, 873–878. Yusufzai, T.M., Tagami, H., Nakatani, Y., and Felsenfeld, G. (2004). CTCF tethers an insulator to subnuclear sites, suggesting shared insulator mechanisms across species. Mol. Cell 13, 291–298. Zhao, Y., Shen, Y., Yang, S., Wang, J., Hu, Q., Wang, Y., and He, Q. (2010). Ubiquitin ligase components Cullin4 and DDB1 are essential for DNA methylation in Neurospora crassa. J. Biol. Chem. 285, 4355–4365. Zofall, M., and Grewal, S.I.S. (2006). Swi6/HP1 recruits a JmjC domain protein to facilitate transcription of heterochromatic repeats. Mol. Cell 22, 681–692.

Crystal Structure and Allosteric Activation of Protein Kinase C bII 2 Layla F. Saidi,1 Gerhard Hummer,2 and James H. Hurley1,* _ Thomas A. Leonard,1 Bartosz Ro´zycki, 1Laboratory

of Molecular Biology of Chemical Physics, National Institute of Diabetes and Digestive and Kidney Diseases National Institutes of Health, Bethesda, MD 20892, USA *Correspondence: [email protected] DOI 10.1016/j.cell.2010.12.013 2Laboratory

SUMMARY

Protein kinase C (PKC) isozymes are the paradigmatic effectors of lipid signaling. PKCs translocate to cell membranes and are allosterically activated upon binding of the lipid diacylglycerol to their C1A and C1B domains. The crystal structure of full-length protein kinase C bII was determined at 4.0 A˚, revealing the conformation of an unexpected intermediate in the activation pathway. Here, the kinase active site is accessible to substrate, yet the conformation of the active site corresponds to a lowactivity state because the ATP-binding side chain of Phe629 of the conserved NFD motif is displaced. The C1B domain clamps the NFD helix in a lowactivity conformation, which is reversed upon membrane binding. A low-resolution solution structure of the closed conformation of PKCbII was derived from small-angle X-ray scattering. Together, these results show how PKCbII is allosterically regulated in two steps, with the second step defining a novel protein kinase regulatory mechanism. INTRODUCTION The formation of small-molecule second messengers is one of the most important consequences of the activation of cellsurface receptors. Many of the most prominent second messengers are lipids, and the paradigmatic lipid second messenger is sn-1,2 diacylglycerol (DAG). DAG activates a variety of cellular effectors, including kinases and GTPase-activating proteins (GAPs) and guanine nucleotide exchange factors (GEFs) for small G proteins (Yang and Kazanietz, 2003). By far the most widely distributed class of DAG effectors are the protein kinase C (PKC) isozymes (Rosse et al., 2010). Intensively studied for the past three decades, PKCs are the archetypal allosteric transducer of lipid second messenger signaling (Newton, 1995; Nishizuka, 1992). In keeping with their widespread tissue distribution, PKCs regulate a remarkable range of physiological pathways, including, but not limited to, T cell recognition, cell polarity, cell migration, proliferation and differentiation, neuronal signaling, and metabolism (Rosse et al., 2010).

PKCs are serine/threonine kinases of the AGC family (Pearce et al., 2010). The AGC family includes protein kinases A, B, C, D, and G and is characterized by a C-terminal extension of the kinase domain that contains one or two regulatory phosphorylation sites, important for kinase activity (Pearce et al., 2010). Like other kinases, PKCs require phosphorylation of a conserved Ser/Thr in the activation loop for activity (Pearce et al., 2010). PKCs are grouped into subclasses based on the domain composition of the regulatory portion and their respective cofactor requirements (Hurley and Grobler, 1997; Mellor and Parker, 1998; Newton, 1995). The conventional PKCs (a, bI, bII, g) are regulated via two DAG-binding C1 domains (Hurley et al., 1997) and a Ca2+- and phospholipid-binding C2 domain (Nalefski and Falke, 1996) (Figure 1A). The novel, Ca2+-independent PKCs (d, 3, h, q) have an N-terminal C2 domain that does not bind Ca2+ or phospholipids, and two typical DAG-binding C1 domains. The Ca2+/DAG-independent atypical isoforms (z, l/i) have a single atypical C1 domain that does not bind DAG and lack a C2 domain. All PKCs have a pseudosubstrate region, in which the phosphorylatable Ser/Thr is replaced by an Ala, which maintains the inactive state in the absence of an activating signal (Newton, 1995; Orr and Newton, 1994). PKCs are primed for activation by the phosphorylation of three residues (Tsutakawa et al., 1995). First, the activation loop Ser/Thr is phosphorylated by PDK1 (Chou et al., 1998; Le Good et al., 1998), and then the C-terminal turn and hydrophobic motif are phosphorylated by mTORC2 (Facchinetti et al., 2008; Ikenoue et al., 2008). Primed PKCs are activated to phosphorylate their substrates when their regulatory domains engage the appropriate combination of signals. Signal engagement triggers the release of the pseudosubstrate sequence from the active site, allowing access to substrates. In the case of the conventional PKCs, these signals are DAG, Ca2+, and phospholipids (Hurley and Grobler, 1997; Mellor and Parker, 1998; Newton, 1995). The question of how signal engagement triggers kinase activation at the structural level has preoccupied many laboratories, and a large body of fragmentary structural information is available for the isolated domains of PKCs. Structures have been solved of the catalytic domains of PKCs bII, q, and i (Grodsky et al., 2006; Messerschmidt et al., 2005; Takimura et al., 2010; Xu et al., 2004), the C2 domains of PKCs a, bII, d, h, and 3 (Guerrero-Valero et al., 2009; Littler et al., 2006; Ochoa et al., 2001; Pappa et al., 1998; Sutton and Sprang, 1998; Verdaguer et al., 1999), and the C1 domains from PKCs a (Hommel et al., 1994), Cell 144, 55–66, January 7, 2011 ª2011 Elsevier Inc. 55

Figure 1. Structure of PKCbII (A) Schematic of the domain structure of PKCbII. (B) Structure of the ordered portion of full-length PKCbII, comprising the C1B (blue), C2 (green), and kinase domains. Magenta denotes the C lobe and yellow the N lobe of the kinase domain. The NFD helix is colored cyan, calcium ions salmon, and AMPPNP orange. Phospho-Thr500, -Thr641, and -Ser660 are shown in a stick model. (C) Difference map (Fo-Fc) phased using a model subjected to a single pass of DEN-restrained refinement, prior to inclusion of residues 620–639. The map is contoured at 2s. See also Figure S1 and Figure S2.

g (Xu et al., 1997), and d (Zhang et al., 1995) alone or bound to the DAG-mimetic phorbol ester. This information has been difficult to integrate into a high-resolution picture of the activation pathway because of various technical challenges in the crystallization of full-length PKCs. In addition to the usual challenges of crystallization of multidomain proteins, suitable PKC samples for crystallization must be stoichiometrically phosphorylated at the activation loop, turn, and hydrophobic sites. Proteolysis in the highly labile V3 region connecting the regulatory and catalytic domains must be avoided. The Zn2+ ions required for the stability of the C1 domains must be retained. We optimized sample purification according to these criteria and were able to crystallize full-length rat PKCbII, an isozyme that has been the subject of especially intensive mechanistic analysis. Moreover, PKCbII is the target for the investigational diabetes drug ruboxistaurin (Das Evcimen and King, 2007). Crystals of PKCbII diffracted only to 4.0 A˚, but by taking advantage of improved methodology for low-resolution crystallographic refinement, the data proved adequate to map a previously unobserved conformation of a helix encompassing the conserved NFD motif of the AGC kinase family. Unexpectedly, lattice contacts between the C2 and catalytic domains led to the observation of what appeared to be a partially activated 56 Cell 144, 55–66, January 7, 2011 ª2011 Elsevier Inc.

conformation. We were able to confirm the structural inference that the observed conformation is part of the physiological activation pathway by mutational analysis of PKCbII translocation. Thus, this structure provides a snapshot of an intermediate in the lipid activation pathway of PKCbII. The analysis identified an unexpected mechanism of allosteric regulation through plasticity of the NFD motif region. In order to fill out the structural picture of the PKCbII activation pathway, small-angle X-ray scattering (SAXS) was used, in conjunction with constraints provided by the crystal structure, to determine a low-resolution structure of the closed, autoinhibited conformation. Together, these structural analyses allow us to map out a conformational activation pathway that is more complex than anticipated. RESULTS Structure of PKCbII PKCbII samples were confirmed for integrity with respect to intact primary structure, stoichiometric Zn binding, phorbol ester-dependent membrane binding and kinase activity, phosphorylation at the three canonical sites on the catalytic domain, and monodispersity (Figure S1 available online). Crystals of a maximum dimension of 20 mm were grown, and diffraction

Table 1. Crystallographic Data Collection and Refinement Statistics Data Collection Space group

P3221

Unit cell (a, b, c in A˚) Resolution range (A˚)

57.1–4.0

114.27, 114.27, 170.84

Observations

46215

Unique reflections

10117

Completeness (%)a

95.0 (88.0)

I/sI

3.5 (1.8)

Rsym (%)

19.2 (39.3)

Structure Refinement Resolution range (A˚)

57.1–4.0

Reflections used

10670

R factor, Rfreeb (%)

19.4 (25.0)

Model residues

101–292, 339–669

Rms Deviations Bond lengths (A˚)

0.012

Bond angles (deg)

0.878

a

Values in parentheses refer to the highest resolution shell. b Rfree = free R factor based on random 5% of all data.

data were collected to 4.0 A˚ (Table 1) on the GM/CA-CAT dedicated microbeam (ID23-B) at the Advanced Photon Source (APS) and the structure was solved by molecular replacement using the coordinates of the PKCbII kinase domain (Grodsky et al., 2006) (Protein Data Bank (PDB): 2I0E) and the PKCbII C2 domain (Sutton and Sprang, 1998) (PDB: 1A25). Electron density was visible for the kinase domain, the C2 domain, and a single C1 domain (Figures 1B and 1C; Figure S2A). The identity and orientation of the C1 domain was confirmed using an anomalous difference Fourier synthesis to locate the two native Zn atoms (Figure S2B). Following refinement, density for the C1-C2 linker became evident. The short, direct connection between the C1 and C2 domains identified the single ordered C1 domain as the C1B domain and confirmed that these domains belonged to the same PKCbII molecule. The use of deformable elastic network (DEN) (Schroder et al., 2010) restraints for the C2 and catalytic domains, representing the majority of the scattering matter, led to electron density maps of a substantially higher quality than expected at 4.0 A˚ (Figure 1C). The initial map calculated following DEN refinement with a model lacking residues 620–639 revealed a novel and wellordered conformation for this entire region (Figure 1C). The sequence could be assigned to this region, in spite of the limited resolution. Clear side-chain density was visible for Phe629, Phe632, and Phe633, providing an unambiguous set of markers to assign the primary sequence to the model. Residues 624– 634 are helical, and the register of these residues was therefore defined with high confidence. Because these residues include the conserved motif NFD (residues 628–630; see Figure 2E), we refer to them as the ‘‘NFD helix.’’ Side-chain features were less obvious for the sections 620–623 and 635–639. However, the main-chain density was nearly continuous and the start and end points of these sections were anchored either to landmarks

in the catalytic domain or to the assigned sequence in the new helix. Therefore these assignments also have a high confidence level. The C1B-C2 connector 151–158 has poorer density and its register is not assigned with high confidence. As described below, the configuration of the C1B-C2 linker in this structure is probably not physiologically relevant, so this limitation of the model does not affect its functional interpretation. The DEN refinement led to an excellent free R factor of 0.25. Nevertheless, the pseudosubstrate and C1A domains and the V3 (C2-kinase) linker were completely absent from the electron density. The kinase domain adopts the intermediate open conformation, as also observed in the bisindolymaleimide-bound isolated kinase domain structure (Figures 2A and 2B) (Grodsky et al., 2006), with minor exceptions noted below. The lattice is stabilized by contacts between the C2 domain of one PKC and two different instances of the catalytic domain (Figure S2A). The Ca2+-binding regions (CBRs) of the C2 span the two lobes of the first instance of the catalytic domain (Figure S2A, contact 1). The burial of the C2-bound Ca2+ ions in an interface with the catalytic domain would be inconsistent with the physiological function of these ions in bridging PKC to the membrane (Nalefski and Falke, 1996), therefore we dismissed this lattice contact as functionally irrelevant. The other major lattice contact involves nonconserved polar residues of both the C2 domain and the catalytic domain (Figure S2A, contact 2). Because the C2 domain is a conserved and functionally important feature of conventional and novel PKCs, we anticipated that functionally relevant contacts should include at least some conserved residues. This second contact is formed by some of the least conserved regions on both the C2 and catalytic domains, and therefore we judged its functional relevance to be implausible. Moreover, the conformation of the C2 domain in this structure is not judged to represent the closed, fully autoinhibited conformation of PKCbII. As described below, this conformation is not consistent with SAXS analysis of PKCbII in solution. Rather, it probably represents a snapshot of the behavior of the C2 domain in the activated state as trapped by lattice contacts. Strong difference density in a phosphate omit map is observed for the activation loop, confirming the phosphorylation of Thr500 (Figure S2C). Difference omit density is also observed for the C-terminal tail of the kinase, confirming the presence of the canonical turn and hydrophobic motif phospho-Thr641 and phospho-Ser660 (Figures S2D and S2E). The active site contains bound AMPPNP and Mg2+; however, the Mg2+ ions could not be positioned accurately into the model given the limitation of 4.0 A˚ resolution. The NFD Helix One of the unexpected features of this structure is the unusual positioning of the residues of the NFD motif (Figure 2), a conserved catalytic element of the AGC kinases. In the present structure, the NFD helix preceding residues in the C-terminal tail of PKCbII is in what we refer to as the ‘‘clamped’’ conformation (Figures 2A and 2B). The clamp keeps the NFD Phe629 out of the active site, but it is in intimate and extensive contact with both the C1B and catalytic domains. The Phe residue of this motif is normally part of the active site and binds the adenine of ATP, on the basis of the ATP-PKCi complex (Takimura et al., 2010), and it is not part of a helix. We refer to the active Cell 144, 55–66, January 7, 2011 ª2011 Elsevier Inc. 57

Figure 2. The Kinase Catalytic Domain and the NFD Helix (A) Comparison of the NFD helix in the isolated PKCbII catalytic domain structure (gray) and the present full-length PKCbII structure, where the NFD helix and adjacent regions are colored cyan. (B) Close-up of the NFD region. (C) Comparison of the NFD region in full-length PKCbII (cyan) and the ATP-bound kinase domain of PKCi (orange) (Takimura et al., 2010). (D) Phe629 of PKCbII corresponds to Phe327 of PKA and Phe543 of PKCi. (E) Sequence alignment of the NFD region of mammalian PKCs and PKCs from C. elegans and C. albicans (yeast), compared to AGC kinase family members PKA and Akt. An invariant proline corresponding to PKCbII Pro619 is at the start of the AGC-specific tail. The inactive NFD helix of our structure is indicated on the alignment (cyan box corresponds to helix in C), as are the rearranged, active NFD motifs of PKCi (orange boxes correspond to helices in C and D) and Akt2 (orange boxes).

58 Cell 144, 55–66, January 7, 2011 ª2011 Elsevier Inc.

Figure 3. The C1B Clamp (A and B) The interface between the C1B domain and the N lobe of the catalytic domain stabilizes the clamp by positioning the C1B appropriately with respect to the NFD helix. This interface is predominantly hydrophobic, with the N lobe side chains of Leu345, Leu358, Leu367, and Y422 (yellow, stick model) packing against Leu125, Ile126, and K141 (blue, surface model) of the C1B domain. Y430 is contributed by the C lobe of the kinase domain (magenta). (C) The C1B domain clamps the NFD helix and residues immediately N-terminal to the helix. Residues Pro619–Cys622 (cyan, stick model) are in an extended conformation that makes intimate contacts with the phorbol ester-binding cleft of the C1B domain. (D) Comparison of the C1B domain of PKCd bound to PMA (Zhang et al., 1995) with that of PKCbII shows a steric clash between phorbol ester and residues 619–622 running through the cleft.

conformation of the NFD motif as the ‘‘in’’ conformation (Figure 2C). Grodsky et al. (2006) first reported that this sequence could form a helix, in the context of the isolated PKCbII catalytic domain. In the isolated catalytic domain structure (Grodsky et al., 2006), the Phe is remote from the active site and forms part of a novel helix. We will refer to this as the ‘‘out’’ conformation (Figure 2A). The out conformation has no known regulatory relevance but does illustrate that this region has an intrinsic propensity to form a helix under appropriate conditions. C1B Contacts with the Catalytic Domain Mimic Membrane and Phorbol Ester The C1B and catalytic domain have an extensive interface made up mainly of conserved hydrophobic residues. On the catalytic domain side, the contact involves a conserved and mostly nonpolar region on the outer face of the N lobe (Figures 3A and 3B). On the C1B side, residues from the conserved hydrophobic tip of the domain, and adjoining residues, make contacts with the catalytic domain and NFD helix. In particular, C1B membranepenetrating side chains of residues Leu125 and Ile126 make

extensive contacts with Phe629 of the NFD adenine-binding motif. Finally, residues 619–622, immediately preceding the start of the NFD helix, fill the DAG-binding site of the C1B domain (Figure 3C). The Pro619 carbonyl appears to accept a hydrogen bond from the main-chain amide of Gly124, replacing the 3oxygen of phorbol ester as seen bound to the C1B domain of PKCd (Zhang et al., 1995). The Ala621 main-chain NH appears to donate a hydrogen bond to the main-chain carbonyl of Gly124, serving as a counterpart of the 4-hydroxyl of phorbol ester (Figure 3D). The C1B domain buries a total solvent-accessible area of 1073 A˚2 in all contacts with the rest of PKCbII. C1B Clamps the Novel NFD Helix of the Catalytic Domain in a Low-Activity Conformation In the isolated PKCbII catalytic domain structure, residues 627– 636 form the NFD helix, which is in the ‘‘out’’ conformation and has not been visualized in the structures of other kinase catalytic domains. Residues 625–626 are disordered in the isolated kinase domain. In the present full-length PKCbII structure, the NFD helix comprises a slightly different set of residues, 624–634, Cell 144, 55–66, January 7, 2011 ª2011 Elsevier Inc. 59

Figure 4. Mutational Analysis of the C1B Clamp in PKCbII Translocation (A) Plot of in vivo membrane translocation of PKCbII against PMA. Data points are the mean of two measurements, and error bars indicate the standard deviation from the mean. (B) PKCbII translocation was quantitated as the reduction in the cytosolic pool of PKC following exposure to PMA. Shown are the western blots for wild-type PKCbII and the indicated mutants. The intensity of the normalized bands is plotted in (A). EC50 values are reported for each mutant. Both soluble cytosolic (S) and membrane pellet (P) fractions are shown for L358D.

and the helix shifts by up to 16 A˚ to occupy the clamped conformation (Figures 2A and 2B). In the closed, catalytically competent conformation of AGC protein kinases, the central Phe of the NFD motif directly contacts the adenine of ATP (Figures 2C and 2D). In the present structure, Phe629 is 12 A˚ away from the adenine moiety (Figure 2C). Phe629 corresponds to Phe327 in PKA (Figure 2D). This Phe is conserved throughout the AGC kinases. In PKA, Phe327 contacts the adenine of ATP. For PKA, mutation of Phe327 weakens the interaction with ATP and lowers catalytic efficiency 50-fold by increasing the Michaelis constant (Km) for ATP (Yang et al., 2009). The only structure available of an isolated PKC catalytic domain bound to ATP is that of PKCi, where the equivalent residue, Phe543 (Figures 2C–2E), directly contacts the adenine (Takimura et al., 2010). In the PKCi-ATP complex, the NFD helix is unwound in the middle, with the Phe and Asp of the NFD motif both in an extended conformation. The residues immediately preceding them, 539–542, are in a 310-helical conformation oriented at a roughly 90 angle to the novel helix in full-length PKCbII (Figure 2C). Residues 534–542 are in a conformation that sterically overlaps with the PKCbII C1B conformation such that simultaneous occupancy of the fully active ATP-bound and C1B-engaged conformation is not possible. Downstream of the key ATP-binding Phe, one turn of the NFD helix persists in the ATP-bound conformation, but its path is different enough that the counterpart of the anchoring Phe633 in PKCbII, Phe547 of PKCi, moves 8 A˚ and forms a completely different set of contacts. Mutational Analysis of the C1B Clamp The physiological significance of the C1B clamp was investigated by analyzing the in vivo membrane translocation of wild60 Cell 144, 55–66, January 7, 2011 ª2011 Elsevier Inc.

type PKCbII and mutants designed to destabilize the interfaces between the kinase N lobe, C1B domain, and AGC-specific C-terminal tail containing the NFD helix. HEK293 cells expressing PKCbII and PDK1 were treated with phorbol 12-myristate 13-acetate (PMA). PMA-dependent membrane translocation was quantitated by the decrease in the cytosolic pool of PKCbII (Figures 4A and 4B). Deletion of two residues in the AGC linker that passes through the cleft of the C1B domain (D621–622) results in a 2-fold decrease in the concentration of PMA required to mediate half-maximal translocation (EC50, Figure 4B). Mutation of residues in the hydrophobic interface between the kinase N lobe and the C1B domain (L358, L367D, Y422D, Y430D) results in 5- to 14-fold decreases in the EC50 value. Within the NFD helix, F629D and F633D led to 2- to 5-fold decreases in EC50 (PMA). Mutation of Leu125 in the C1B domain, which is part of the C1B clamp interface, was the only mutation that increased EC50 (PMA). Indeed, the increase was dramatic, more than 10-fold. This is consistent with its primary role in membrane binding (Medkova and Cho, 1999; Xu et al., 1997). The destabilization of PKCbII by mutations that disrupt the C1B clamp confirms the physiological significance of this interface in maintaining a secondary mode of autoinhibition in addition to the known mechanism of pseudosubstrate engagement in the active site of the kinase domain. Solution Structure of PKCbII from Small-Angle X-Ray Scattering The solution structure of full-length PKCbII was analyzed using SAXS (Figure 5). Ab initio determination of the molecular envelope converged on a radius of gyration (RG) of 33 A˚ and a maximum particle dimension of 100 A˚ (Figure 5A).

The envelope did not, on its own, contain sufficient detail to position the four structural domains (C1A, C1B, C2, and kinase). Therefore, a new computational procedure, ensemble refine_ et al., 2011), was applied in order to ment of SAXS (Ro´zycki determine the best fit of the structural domains and interdomain linkers of PKCbII to the experimental SAXS curve, subject to the constraints imposed by the position of the C1B domain in the crystal structure. The pseudosubstrate was modeled in the conformation seen for the interaction between the catalytic and regulatory subunits of PKA (Kim et al., 2005, 2007). The position of the C1A was constrained by the shortness of the 8 residue pseudosubstrate-C1A and 15 residue C1A-C1B linkers. The C1B and kinase domains were held fixed in their crystallographic conformations. The C2 domain N terminus was constrained by the 6 residue C1B-C2 connector. The C2 domain C terminus was essentially unconstrained by the 46 residue V3 loop connecting it to the catalytic domain. The conformational space occupied by these linked domains was exhaustively sampled and the conformations that best fit the SAXS data were selected (Figure 5D). The conformations that provided acceptable fits to the SAXS I(q) curve (Figure 5B) also led to acceptable fits within the ab initio molecular envelope (Figure 5C). All had in common a conformation in which the C2 domain projected away from the rest of PKCbII. In each case, the C2 domain made contact with the C1B domain and/or associated linker regions but not with the catalytic or C1A domains. More compact conformations of the C2 domain relative to the rest of PKCbII did not lead to acceptable fits, nor did attempts to fit a single compact conformation combined with an ensemble of highly open structures. In solution, the conformation of PKCbII is best described as a single closed, autoinhibited state in which the C2 domain projects away from, and has limited contact with, the rest of the structure. DISCUSSION The central finding in this study is that the NFD helix is a linchpin of the regulation of PKCbII. The structural observation that the C1B domain clamps the NFD helix in an inactive conformation originated with the unexpected observation of a partially open conformation for PKCbII. The relevance of this mechanism for PKCbII in vivo was corroborated by mutational analysis of contact residues on both the C1B domain and the NFD helix. It was expected that destabilizing the C1B clamp would promote its disengagement from the NFD helix and the rest of the catalytic domain, lowering the energy barrier for translocation, and therefore lowering EC50 (PMA). Changing the register of the AGC linker in the cleft of the C1B domain, mutation of the hydrophobic side chains of Phe629 and Phe633 of the NFD helix, and mutations in the core of the N lobe:C1B interface all serve to destabilize PKCbII to such an extent that the EC50 values decrease by up to 14-fold. These EC50 shifts provide strong corroboration for the structural mechanism of regulation by NFD helix clamping. The phorbol ester- and DAG-binding site on the C1B domain is surrounded on three sides by an extremely hydrophobic rim formed by the exposed side chains of (PKCbII numbering) Pro112, Phe114, Tyr123, and Leu125. The counterparts of these residues in related PKCs have been shown to contact lipid tails by NMR in lipid micelles (Xu et al., 1997) and to penetrate the

hydrocarbon core of membrane bilayers by surface pressure studies (Medkova and Cho, 1999). The burial of Tyr123 and Leu125 against the catalytic domain in the present partially open structure renders them unavailable for membrane penetration. This situation is similar to the burial of the membranebinding surface of the C1 domain in the inactive conformation of b2-chimaerin (Canagarajah et al., 2004). The C1 domain of b2-chimaerin buries 1530 A˚2 of solvent-accessible surface area in contacts with the rest of the molecule, almost 50% more than for the PKCbII C1B. The more extensive C1 domain burial in b2-chimaerin helps explain why its EC50 for PMAinduced translocation is so much higher than for PKCbII, 1.2 mM as compared to 42 nM under otherwise similar conditions. It also may explain in part why mutations that disrupt the b2-chimaerin C1 domain interface yield larger EC50 decreases of up to 80-fold, as compared to a maximum of 14-fold for PKCbII. A full explanation of these differences will require the structure of the completely closed conformation of PKCbII and an assessment of the burial of the membrane-binding surface of the C1A domain in that conformation. It is notable that, in contrast to all other mutants tested, L125D shifts the PMA dose-response curve to the right. This mutant is expected to weaken membrane binding by the C1B domain at the same time as it destabilizes the clamp. The right-shift produced by L125D outweighs the potential left-shift caused by destabilization of the C1B clamp. The role of the membrane-binding site residues in clamping the inactive state highlights that dual and opposing effects are to be expected upon mutating these residues. This duality probably underlies the sometimes contradictory conclusions of various studies employing mutations to dissect the relative roles of PKC C1A and C1B domains in membrane translocation (Bogi et al., 1999; Szallasi et al., 1996) and lipid activation (Medkova and Cho, 1999). Upon activation, most of the NFD helix unwinds in order for Phe629 to insert itself into the active site. This highlights the NFD region as one of the most dynamic parts of the catalytic domain and one of the most important for regulation of PKCbII. The conservation of key residues involved in the clamp, such as Phe633, suggests that NFD helix regulation extends in evolution as far as the aPKCs and Akts. The aPKCs and Akts lack a C1B domain, and indeed the residues that insert into the phorbol ester-binding site of the C1B are not as well conserved. It is unclear what other domain or partner might take the place of the C1B domain of the cPKCs and nPKCs in this context. Conservation of the clamp residues is weaker still in PKA, where the present conformation of the NFD helix is not observed, even in the inactive holoenzyme (Kim et al., 2005). Observation of the NFD helix has only been reported once before (Grodsky et al., 2006), at which time its significance was not clear. These results beg the question as to why NFD helix-mediated regulation has not been reported in other kinases, and whether it is unique to the PKCs and perhaps their closest relatives. The aC helix of the PKCs appears to be unusually rigid, having been observed in the catalytically active conformation in all PKC structures to date (Grodsky et al., 2006; Messerschmidt et al., 2005; Takimura et al., 2010; Xu et al., 2004). Perhaps NFD helix regulation of PKC evolved as a substitute for what appears to be an inoperative aC helix mechanism in this kinase subfamily. There have been three Cell 144, 55–66, January 7, 2011 ª2011 Elsevier Inc. 61

Figure 5. SAXS of PKCbII (A) Pair distribution function (P(r)) for autoinhibited PKCbII. The radius of gyration is estimated to be 33.6 A˚ and 33.5 A˚ by Guinier analysis and the P(r) function, respectively. The maximum dimension of the particle is estimated to be 100 A˚. (B) Fit of the simulated scattering curves (I(q)) to the observed scattering of autoinhibited PKCbII. Experimental I(q) data points represent the mean of ten consecutive measurements of the same sample, and the error bars represent the standard error of the mean.

62 Cell 144, 55–66, January 7, 2011 ª2011 Elsevier Inc.

Figure 6. Model for Multistep Activation of PKCbII PKCbII translocates to the membrane upon Ca2+ release in the cell, where the calcium-binding regions of its C2 domain mediate bridging to the anionic phospholipid phosphatidylserine, with an adjoining site on the C2 domain-binding phosphatidylinositol (4,5)-bisphosphate (1). Subsequent binding of DAG to the C1A domain results in disengagement of the C1A domain (2), which in turn forces the removal of the pseudosubstrate (PS) from the catalytic cleft. Binding of a second molecule of DAG by the C1B domain results in unclamping of the kinase:C1B assembly (3) and rearrangement of the NFD helix into the catalytically competent state (4), triggering the phosphorylation of cellular targets (5) and subsequent downstream signaling. The C2 domain was docked to the membrane in the geometry described by Landgraf et al. (2008) and the C1 domains as modeled by Zhang et al. (1995).

major structural mechanisms described for regulation of kinase activity: steric blockage of the active site, modulation of the activation loop, and positioning of the aC helix (Huse and Kuriyan, 2002). To this list, we propose adding a fourth mechanism for the PKCs and perhaps extending to the Akt/PKBs, regulation by clamping of the NFD helix. Solution structural analysis by SAXS is consistent with the long-standing model for PKC activation, in which displacement of the pseudosubstrate sequence is coupled to lipid binding by the C1 domains and their displacement from contacts with the catalytic domain (Hurley and Grobler, 1997; Newton, 1995; Orr and Newton, 1994). In models for the activation of the Ca2+dependent cPKCs a, bI, bII, and g, the C2 domain rapidly engages with the membrane following Ca2+ binding to the C2 domain CBRs (Nalefski and Newton, 2001; Oancea and Meyer, 1998). C2 binding initiates two-dimensional diffusion of the cPKC on the membrane surface until the cPKC dissociates or encounters a DAG molecule. The exposure of the Ca2+-binding CBRs of the C2 domain in the SAXS structure of PKCbII is consistent with such a function for the C2 domain. The current structural results suggest that the next step in cPKC activation is DAG engagement with the C1A domain, which both stabilizes

membrane residency and pulls the pseudosubstrate out of the active site (Figure 6). However, this is still insufficient for full activation, which requires yet another step. This last step is the binding of the C1B domain to the membrane, unclamping of the NFD helix, and engagement of Phe629 into the active site (Figure 6). Taken together, the crystallographic and SAXS analysis leads to a picture of PKC activation that is fully consistent with long-standing models for pseudosubstrate displacement (Newton, 1995; Orr and Newton, 1994), while at the same time revealing a novel and unexpected role for the NFD motif region as an additional mode of allosteric regulation. EXPERIMENTAL PROCEDURES Protein Expression and Purification Homo sapiens PDK-1 and Rattus norvegicus PKCbII containing a TEV-cleavable N-terminal glutathione S-transferase (GST) tag were cloned into the P10 and PH cassettes, respectively, of pFastBac Dual (Invitrogen). Wild-type and a mutant PKCbII in which three surface-exposed Cys had been mutated to Ser (C70S/C217S/C622S, the ‘‘SEC’’ mutant) were subcloned. Recombinant bacmid was generated by transforming DH10 MultiBac cells (Berger et al., 2004), and the Sf9 cells were transfected with the bacmid to generate baculovirus. Hi5 cells were infected with the baculovirus and grown in Express Five

(C) Superimposition of the best fit solution to the SAXS curve with an ab initio molecular envelope generated by DAMMIF, in two views rotated by 90 degrees from one another. The assembly of PKCbII kinase and regulatory domains fits inside the envelope. (D) A sample of the top solutions to the simulation of full-length, autoinhibited PKCbII indicating the uncertainty in the placement of the C1A domain (shown in blue). See also Figure S1.

Cell 144, 55–66, January 7, 2011 ª2011 Elsevier Inc. 63

(Invitrogen) medium supplemented with 40 mM ZnCl2 for 72 hr at 28 C. For purification, cells were lysed in 50 mM Tris, pH 7.4, 300 mM NaCl, 50 mM NaF, 5 mM sodium pyrophosphate, 10 mM b-glycerophosphate, 1 mM TCEP, 2 mM benzamidine, 2 mg/ml leupeptin, 0.5 mM sodium orthovanadate, 0.5% (w/v) CHAPS, and 1:100 protease inhibitor cocktail (Sigma P8849). All steps were performed at 4 C. The lysate was spun at 43,152 3 g, 45 min to pellet-insoluble material. The supernatant was incubated with glutathione sepharose (GS-4B; GE Healthcare), 2 hr, then washed with 100 ml lysis buffer, 1 hr, followed by 100 ml of TEV cleavage buffer (50 mM Tris, pH 7.4, 50 mM NaF, 5 mM sodium pyrosphophate, 10 mM b-glycerophosphate, 1 mM TCEP, 0.5 mM sodium orthovanadate, 0.25% (w/v) CHAPS). GST- PKCbII was cleaved on-resin with TEV protease overnight. The supernatant from the TEV cleavage reaction was applied to a Q-sepharose anion exchange column (GE Healthcare) and PKCbII eluted with a linear NaCl gradient in 50 mM Tris, pH 8.0, 1 mM TCEP. Fractions containing PKCbII were pooled, concentrated, and loaded onto a Superdex 200 (10/300 GL; GE Healthcare) gel-filtration column equilibrated in 20 mM Tris, pH 8.0, 100 mM NaCl, 2 mM MgCl2, 1 mM TCEP. PKCbII was eluted as a single peak with a retention time consistent with a globular, monomeric protein of 78 kDa. Crystallization and Structure Determination Crystals of wild-type and SEC mutant PKCbII grew over 2 days from fresh protein preparations. SDS-PAGE analysis of the drops from which crystals were harvested showed no indication of PKCbII degradation (Figure S1A). PKCbII was combined with 1 mM AMPPNP (Sigma) and concentrated to 2 mg/ml. Both wild-type and SEC proteins crystallized; however the best diffractors were obtained for the SEC mutant. Hexagonal crystals of the SEC mutant were grown in 1%–4% (w/v) PEG 8000, 100 mM Tris, pH 8.5, either attached to the plate or to a skin of denatured protein on the surface of the drop. Multiple crystals were mounted in a loop, cryoprotected in 20% ethylene glycol, and drop frozen in liquid nitrogen. Data were collected on the microfocus beamline GM/CA-CAT ID23-B at the Advanced Photon Source (APS). Crystals in the loop were detected and centered in the beam by using the RASTER software to scan the loop for diffraction. A complete dataset from a single crystal with unit cell dimensions a = b = 114.27 A˚, c = 170.84 A˚ was collected to 4.0 A˚ resolution in spacegroup P3221. The structure of PKCbII was solved by molecular replacement using the program Phaser (McCoy et al., 2007). The catalytic domain of PKCbII (PDB ID: 2I0E) and the C2 domain of PKCbII (PDB ID: 1A25) were used as search models. The C1B domain was located in the map using the interatomic distance between the zinc ions to help position it, and this was followed by rigid body refinement. No density was observed for the N-terminal pseudosubstrate segment and C1A domain (residues 1–100) and the V3 linker (residues 293–338). Refinement was carried out using the deformable elastic network (DEN) model as implemented in CNS 1.3 (Schroder et al., 2010) with the catalytic and C2 domains as separate restraint groups. The DEN parameters g and k were set at 0.5 and 0.1, respectively. Residues 353 and 619–639 of the catalytic domain were omitted from the restraint group. Density syntheses following DEN refinement were instrumental in determining the position of the novel helix. The refined structure has no residues in disallowed regions of the Ramachandran plot and only ten residues in generously allowed regions (Laskowski et al., 1993). The MolProbity (Davis et al., 2007) all atom clash score was 41.6, placing the structure in the 64th percentile (with 100th best) of structures refined at 3.0 A˚ or lower resolution. The MolProbity protein geometry score was 3.5, placing the structure in the 70th percentile of structures at 3.25–4.25 A˚ resolution. Clashes and stereochemical outliers were more common in the C1B domain, the C1B-C2 linker, and the NFD helix and adjacent residues, which were not constrained by the DEN model. In Vivo Translocation Assay Wild-type and mutant PKCbII proteins were analyzed for their ability to translocate to membranes in a phorbol ester-dependent manner in vivo. HEK293 GnTI cells were cotransfected with mammalian expression vectors containing PDK-1 and PKCbII and grown at 37 C for 48 hr. Cells were stimulated with 80 pM–1 mM phorbol 12-myristate 13-acetate (PMA) (Sigma) for 15 min at 37 C, harvested, and snap frozen on dry ice. Cells were resuspended in 500 ml PKC lysis buffer, sonicated on low power on ice, 15 s, 1 s pulses followed by 1 s off.

64 Cell 144, 55–66, January 7, 2011 ª2011 Elsevier Inc.

The lysate was spun at 900 3 g, 5 min, 4 C to discard the organelles and enrich plasma membrane. The supernatant was then transferred to a fresh tube and spun at 100,000 3 g, 30 min, 4 C. The supernatant (cytosolic fraction) was transferred to a fresh tube on ice. Ten microliters of a 3:1 mixture of cytosolic or membrane fraction with 43 LDS gel loading buffer (Invitrogen) containing 50 mM b-mercaptoethanol was loaded on a 4%–12% polyacrylamide gel (Invitrogen). The gel was blotted onto PVDF and PKCbII detected using an anti-PKCbII primary antibody (BD Biosciences BD610128) and Alexa Fluor 488-conjugated goat anti-rabbit IgG secondary antibody (Invitrogen). Cytosolic PKCbII fractions were quantitated using a Typhoon scanner at 519 nm. Small-Angle X-Ray Scattering Wild-type PKCbII samples were prepared with and without the nonhydrolyzable nucleotide analog AMPPNP. Data were collected on 25 mM samples in 20 mM Tris, pH 8.0, 100 mM NaCl, 2 mM MgCl2, 1 mM TCEP ± 1 mM AMPPNP at SSRL beamline BL4-2. Data reduction and analysis were performed using the beamline software SAStool. The program AutoGNOM of the ATSAS suite (Petoukhov et al., 2007) was used to generate P(r) curves and to determine maximum dimension (Dmax) and RG from the scattering intensity curve (I(q) versus q) in an automatic, unbiased manner, although rounds of manual fitting in GNOM (Svergun, 1992) were used to verify these values. Ab initio molecular envelopes were computed by the programs DAMMIF (Franke and Svergun, 2009) and GASBOR (Svergun et al., 2001). Multiple iterations of DAMMIF and GASBOR were averaged using DAMAVER (Volkov and Svergun, 2003), the core residues fixed, and the model subjected to a further cycle of DAMMIN (Svergun, 1999) refinement. SAXS Structural Refinement Monte Carlo simulations of PKCbII were performed using an energy function and simulation model developed to study multiprotein complexes (Kim and _ Hummer, 2008; Kim et al., 2008; Ren et al., 2009; Ro´zycki et al., 2011). In this model, folded protein domains are represented as rigid bodies. Here, the catalytic domain together with the bound C1B domain and the pseudosubstrate peptide modeled on the basis of PKA (PDB entries 3FHI and 2QCS) (Kim et al., 2005, 2007) are treated as one rigid body. C1A and C2 domains are modeled as two separate rigid bodies. The interactions between the domains are treated at the residue level with amino-acid-dependent pair potentials and Debye-Hu¨ckel-type electrostatic interactions. Flexible linker peptides connecting the rigid domains are represented as amino acid beads on an excluded-volume random-coil polymer. We adapted the CRYSOL algorithm (Svergun et al., 1995) to calculate SAXS intensity I(q) for all simulated structures. CRYSOL default parameters were used in the calculation. To estimate the uncertainty of computed intensity I(q), we varied the hydration layer electron density between 0.015 e/A˚3 and 0.03 e/A˚3 (Yang et al., 2009). To gain insights into the protein-relevant configurations, we selected several structures whose SAXS profiles I(q) best fit the experimental data (Figures 5B–5D). We used chi-square as a measure of discrepancy between the computed SAXS curves and experimental data. The error term that enters the c2 formula contains the statistical error of experiment and the uncertainty of computed intensity I(q). Only structures with c2 < 1.5 were considered. ACCESSION NUMBERS Crystallographic coordinates have been deposited in the Protein Data Bank with accession code 3PFQ. SUPPLEMENTAL INFORMATION Supplemental Information includes Extended Experimental Procedures and two figures and can be found with this article online at doi:10.1016/j.cell. 2010.12.013. ACKNOWLEDGMENTS We thank J. Chang for technical assistance, J. Lloyd and D.E. Anderson for assistance with mass spectrometry, J.F. Mushinski and A. Toker for DNAs,

E. Boura for assistance with SAXS data collection, H. Tsuruta and T. Weiss for user support at BL4-2, SSRL, M. Becker for user support at GM/CA-CAT, APS, H. Mischak and M. Pearson for early efforts, A. Newton for discussions, and A. Bru¨nger for a prerelease version of CNS 1.3. GM/CA-CAT has been funded in whole or in part with Federal funds from the NCI (Y1-CO-1020) and the NIGMS (Y1-GM-1104). Use of the Advanced Photon Source was supported by the U.S. Department of Energy, Basic Energy Sciences, Office of Science, under contract No. DE-AC02-06CH11357. SAXS data were collected at the Stanford Synchrotron Radiation Laboratory, a national user facility operated by Stanford University on behalf of the U.S. Department of Energy, Office of Basic Energy Sciences. The SSRL Structural Molecular Biology Program is supported by the Department of Energy, Office of Biological and Environmental Research, and by the National Institutes of Health, National Center for Research Resources, Biomedical Technology Program. T.A.L. was supported in part by an EMBO long term fellowship. B.R. was supported by a Marie Curie International Outgoing Fellowship within the 7th European Community Framework Programme. This research was supported by the Intramural Program of the National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health to J.H.H. and G.H. Received: August 27, 2010 Revised: October 29, 2010 Accepted: December 9, 2010 Published: January 6, 2011

REFERENCES Berger, I., Fitzgerald, D.J., and Richmond, T.J. (2004). Baculovirus expression system for heterologous multiprotein complexes. Nat. Biotechnol. 22, 1583– 1587. Bogi, K., Lorenzo, P.S., Acs, P., Szallasi, Z., Wagner, G.S., and Blumberg, P.M. (1999). Comparison of the roles of the C1a and C1b domains of protein kinase C a in ligand induced translocation in NIH 3T3 cells. FEBS Lett. 456, 27–30. Canagarajah, B., Collucio Leskow, F., Ho, Y.S.J., Mischak, H., Saidi, L., Kazanietz, M.G., and Hurley, J.H. (2004). Structural mechanism for lipid activation of the rac-specific GAP, b2-chimaerin. Cell 119, 407–418. Chou, M.M., Hou, W.M., Johnson, J., Graham, L.K., Lee, M.H., Chen, C.S., Newton, A.C., Schaffhausen, B.S., and Toker, A. (1998). Regulation of protein kinase C z by PI 3-kinase and PDK-1. Curr. Biol. 8, 1069–1077. Das Evcimen, N., and King, G.L. (2007). The role of protein kinase C activation and the vascular complications of diabetes. Pharmacol. Res. 55, 498–510. Davis, I.W., Leaver-Fay, A., Chen, V.B., Block, J.N., Kapral, G.J., Wang, X., Murray, L.W., Arendall, W.B., Snoeyink, J., Richardson, J.S., et al. (2007). MolProbity: all-atom contacts and structure validation for proteins and nucleic acids. Nucleic Acids Res. 35, W375–W383. Facchinetti, V., Ouyang, W.M., Wei, H., Soto, N., Lazorchak, A., Gould, C., Lowry, C., Newton, A.C., Mao, Y.X., Miao, R.Q., et al. (2008). The mammalian target of rapamycin complex 2 controls folding and stability of Akt and protein kinase C. EMBO J. 27, 1932–1943.

Hurley, J.H., Newton, A.C., Parker, P.J., Blumberg, P.M., and Nishizuka, Y. (1997). Taxonomy and function of C1 protein kinase C homology domains. Protein Sci. 6, 477–480. Huse, M., and Kuriyan, J. (2002). The conformational plasticity of protein kinases. Cell 109, 275–282. Ikenoue, T., Inoki, K., Yang, Q., Zhou, X., and Guan, K.L. (2008). Essential function of TORC2 in PKC and Akt turn motif phosphorylation, maturation and signalling. EMBO J. 27, 1919–1931. Kim, C., Xuong, N.H., and Taylor, S.S. (2005). Crystal structure of a complex between the catalytic and regulatory (RI alpha) subunits of PKA. Science 307, 690–696. Kim, C., Cheng, C.Y., Saldanha, S.A., and Taylor, S.S. (2007). PKA-I holoenzyme structure reveals a mechanism for cAMP-dependent activation. Cell 130, 1032–1043. Kim, Y.C., and Hummer, G. (2008). Coarse-grained models for simulations of multiprotein complexes: application to ubiquitin binding. J. Mol. Biol. 375, 1416–1433. Kim, Y.C., Tang, C., Clore, G.M., and Hummer, G. (2008). Replica exchange simulations of transient encounter complexes in protein-protein association. Proc. Natl. Acad. Sci. USA 105, 12855–12860. Landgraf, K.E., Malmberg, N.J., and Falke, J.J. (2008). Effect of PIP2 binding on the membrane docking geometry of PKC a C2 domain: An EPR sitedirected spin-labeling and relaxation study. Biochemistry 47, 8301–8316. Laskowski, R.A., Macarthur, M.W., Moss, D.S., and Thornton, J.M. (1993). Procheck - A program to check the stereochemical quality of protein structures. J. Appl. Crystallogr. 26, 283–291. Le Good, J.A., Ziegler, W.H., Parekh, D.B., Alessi, D.R., Cohen, P., and Parker, P.J. (1998). Protein kinase C isotypes controlled by phosphoinositide 3-kinase through the protein kinase PDK1. Science 281, 2042–2045. Littler, D.R., Walker, J.R., She, Y.M., Finerty, P.J., Newman, E.M., and DhePaganon, S. (2006). Structure of human protein kinase C hPKC h C2 domain and identification of phosphorylation sites. Biochem. Biophys. Res. Commun. 349, 1182–1189. McCoy, A.J., Grosse-Kunstleve, R.W., Adams, P.D., Winn, M.D., Storoni, L.C., and Read, R.J. (2007). Phaser crystallographic software. J. Appl. Cryst. 40, 658–674. Medkova, M., and Cho, W.H. (1999). Interplay of C1 and C2 domains of protein kinase C-a in its membrane binding and activation. J. Biol. Chem. 274, 19852– 19861. Mellor, H., and Parker, P.J. (1998). The extended protein kinase C superfamily. Biochem. J. 332, 281–292. Messerschmidt, A., Macieira, S., Velarde, M., Badeker, M., Benda, C., Jestel, A., Brandstetter, H., Neuefeind, T., and Blaesse, M. (2005). Crystal structure of the catalytic domain of human atypical protein kinase C-i reveals interaction mode of phosphorylation site in turn motif. J. Mol. Biol. 352, 918–931. Nalefski, E.A., and Falke, J.J. (1996). The C2 domain calcium-binding motif: Structural and functional diversity. Protein Sci. 5, 2375–2390.

Franke, D., and Svergun, D.I. (2009). DAMMIF, a program for rapid ab-initio shape determination in small-angle scattering. J. Appl. Cryst. 42, 342–346.

Nalefski, E.A., and Newton, A.C. (2001). Membrane binding kinetics of protein kinase C b II mediated by the C2 domain. Biochemistry 40, 13216–13229.

Grodsky, N., Li, Y., Bouzida, D., Love, R., Jensen, J., Nodes, B., Nonomiya, J., and Grant, S. (2006). Structure of the catalytic domain of human protein kinase C b II complexed with a bisindolylmaleimide inhibitor. Biochemistry 45, 13970– 13981.

Newton, A.C. (1995). Protein kinase C: Structure, function, and regulation. J. Biol. Chem. 270, 28495–28498.

Guerrero-Valero, M., Ferrer-Orta, C., Querol-Audi, J., Marin-Vicente, C., Fita, I., Gomez-Fernandez, J.C., Verdaguer, N., and Corbalan-Garcia, S. (2009). Structural and mechanistic insights into the association of PKC a-C2 domain to PtdIns(4,5)P-2. Proc. Natl. Acad. Sci. USA 106, 6603–6607.

Oancea, E., and Meyer, T. (1998). Protein kinase C as a molecular machine for decoding calcium and diacylglycerol signals. Cell 95, 307–318.

Nishizuka, Y. (1992). Intracellular signaling by hydrolysis of phospholipids and activation of protein kinase C. Science 258, 607–614.

Hommel, U., Zurini, M., and Luyten, M. (1994). Solution structure of a cysteinerich domain of rat protein kinase C. Nat. Struct. Biol. 1, 383–387.

Ochoa, W.F., Garcia-Garcia, J., Corbalan-Garcia, I.F.S., Verdaguer, N., and Gomez-Fernandez, J.C. (2001). Structure of the C2 domain from novel protein kinase C epsilon. A membrane binding model for Ca2+-independent C2 domains. J. Mol. Biol. 311, 837–849.

Hurley, J.H., and Grobler, J.A. (1997). Protein kinase C and phospholipase C: Bilayer interactions and regulation. Curr. Opin. Struct. Biol. 7, 557–565.

Orr, J.W., and Newton, A.C. (1994). Intrapeptide regulation of protein kinase C. J. Biol. Chem. 269, 8383–8387.

Cell 144, 55–66, January 7, 2011 ª2011 Elsevier Inc. 65

Pappa, H., Murray-Rust, J., Dekker, L.V., Parker, P.J., and McDonald, N.Q. (1998). Crystal structure of the C2 domain from protein kinase C d. Structure 6, 885–894. Pearce, L.R., Komander, D., and Alessi, D.R. (2010). The nuts and bolts of AGC protein kinases. Nat. Rev. Mol. Cell Biol. 11, 9–22. Petoukhov, M.V., Konarev, P.V., Kikhney, A.G., and Svergun, D.I. (2007). ATSAS 2.1: Towards automated and web-supported small-angle scattering data analysis. J. Appl. Cryst. 40, S223–S228. Ren, X., Kloer, D.P., Kim, Y.C., Ghirlando, R., Saidi, L.F., Hummer, G., and Hurley, J.H. (2009). Hybrid structural model of the complete human ESCRT0 complex. Structure 17, 406–416. Rosse, C., Linch, M., Kermorgant, S., Cameron, A.J.M., Boeckeler, K., and Parker, P.J. (2010). PKC and the control of localized signal dynamics. Nat. Rev. Mol. Cell Biol. 11, 103–112. _ Ro´zycki, B., Kim, Y.C., and Hummer, G. (2011). SAXS ensemble refinement of ESCRT-III CHMP3 conformational transitions. Structure, in press. Published online January 11, 2011. 10.1016/j.str.2010.10.006. Schroder, G.F., Levitt, M., and Bru¨nger, A.T. (2010). Super-resolution biomolecular crystallography with low-resolution data. Nature 464, 1218–1222. Sutton, R.B., and Sprang, S.R. (1998). Structure of the protein kinase C b phospholipid-binding C2 domain complexed with Ca2+. Structure 6, 1395–1405. Svergun, D.I. (1992). Determination of the regularization parameter in indirect: Transform methods using perceptual criteria. J. Appl. Cryst. 25, 495–503.

Szallasi, Z., Bogi, K., Gohari, S., Biro, T., Acs, P., and Blumberg, P.M. (1996). Non-equivalent roles for the first and second zinc fingers of protein kinase C d: Effect of their mutation on phorbol ester-induced translocation in NIH 3T3 cells. J. Biol. Chem. 271, 18299–18301. Takimura, T., Kamata, K., Fukasawa, K., Ohsawa, H., Komatani, H., Yoshizumi, T., Takahashi, I., Kotani, H., and Iwasawa, Y. (2010). Structures of the PKC-i kinase domain in its ATP-bound and apo forms reveal defined structures of residues 533-551 in the C-terminal tail and their roles in ATP binding. Acta Crystallogr. D Biol. Crystallogr. 66, 577–583. Tsutakawa, S.E., Medzihradszky, K.F., Flint, A.J., Burlingame, A.L., and Koshland, D.E. (1995). Determination of in vivo phosphorylation sites in protein kinase C. J. Biol. Chem. 270, 26807–26812. Verdaguer, N., Corbalan-Garcia, S., Ochoa, W.F., Fita, I., and Gomez-Fernandez, J.C. (1999). Ca2+ bridges the C2 membrane-binding domain of protein kinase C a directly to phosphatidylserine. EMBO J. 18, 6329–6338. Volkov, V.V., and Svergun, D.I. (2003). Uniqueness of ab initio shape determination in small-angle scattering. J. Appl. Cryst. 36, 860–864. Xu, R.X., Pawelczyk, T., Xia, T.H., and Brown, S.C. (1997). NMR structure of a protein kinase C-gamma phorbol-binding domain and study of protein-lipid micelle interactions. Biochemistry 36, 10709–10717. Xu, Z.B., Chaudhary, D., Olland, S., Wolfrom, S., Czerwinski, R., Malakian, K., Lin, L., Stahl, M.L., McCarthy, D.J., Benander, C., et al. (2004). Catalytic domain crystal structure of protein kinase C-theta PKC q. J. Biol. Chem. 279, 50401–50409.

Svergun, D.I. (1999). Restoring low resolution structure of biological macromolecules from solution scattering using simulated annealing. Biophys. J. 76, 2879–2886.

Yang, C., and Kazanietz, M.G. (2003). Divergence and complexities in DAG signaling: looking beyond PKC. Trends Pharmacol. Sci. 24, 602–608.

Svergun, D., Barberato, C., and Koch, M.H.J. (1995). CRYSOL - A program to evaluate x-ray solution scattering of biological macromolecules from atomic coordinates. J. Appl. Cryst. 28, 768–773.

Yang, J., Kennedy, E.J., Wu, J., Deal, M.S., Pennypacker, J., Ghosh, G., and Taylor, S.S. (2009). Contribution of non-catalytic core residues to activity and regulation in protein kinase A. J. Biol. Chem. 284, 6241–6248.

Svergun, D.I., Petoukhov, M.V., and Koch, M.H.J. (2001). Determination of domain structure of proteins from X-ray solution scattering. Biophys. J. 80, 2946–2953.

Zhang, G., Kazanietz, M.G., Blumberg, P.M., and Hurley, J.H. (1995). Crystal structure of the Cys2 activator-binding domain of protein kinase C d in complex with phorbol ester. Cell 81, 917–924.

66 Cell 144, 55–66, January 7, 2011 ª2011 Elsevier Inc.

Amyloid-like Aggregates Sequester Numerous Metastable Proteins with Essential Cellular Functions Heidi Olzscha,1,4 Sonya M. Schermann,1,4 Andreas C. Woerner,1 Stefan Pinkert,1 Michael H. Hecht,2 Gian G. Tartaglia,3,5 Michele Vendruscolo,3 Manajit Hayer-Hartl,1,* F. Ulrich Hartl,1,* and R. Martin Vabulas1,* 1Department

of Cellular Biochemistry, Max Planck Institute of Biochemistry, Am Klopferspitz 18, D-82159 Martinsried, Germany of Chemistry, Princeton University, Princeton, NJ 08544, USA 3Department of Chemistry, University of Cambridge, Lensfield Road, Cambridge CB2 1EW, UK 4These authors contributed equally to this work 5Present address: Bioinformatics & Genomics Program, CRG Centre for Genomic Regulation, Dr. Aiguader 88, Barcelona 08003, Spain *Correspondence: [email protected] (M.H-H.), [email protected] (F.U.H.), [email protected] (R.M.V.) DOI 10.1016/j.cell.2010.11.050 2Department

SUMMARY

Protein aggregation is linked with neurodegeneration and numerous other diseases by mechanisms that are not well understood. Here, we have analyzed the gain-of-function toxicity of artificial b sheet proteins that were designed to form amyloid-like fibrils. Using quantitative proteomics, we found that the toxicity of these proteins in human cells correlates with the capacity of their aggregates to promote aberrant protein interactions and to deregulate the cytosolic stress response. The endogenous proteins that are sequestered by the aggregates share distinct physicochemical properties: They are relatively large in size and significantly enriched in predicted unstructured regions, features that are strongly linked with multifunctionality. Many of the interacting proteins occupy essential hub positions in cellular protein networks, with key roles in chromatin organization, transcription, translation, maintenance of cell architecture and protein quality control. We suggest that amyloidogenic aggregation targets a metastable subproteome, thereby causing multifactorial toxicity and, eventually, the collapse of essential cellular functions. INTRODUCTION The majority of proteins must fold into well-defined three-dimensional structures in order to fulfill their biological functions. This fundamental process is aided by a complex cellular machinery of molecular chaperones, which act to prevent misfolding and aggregation (Frydman, 2001; Hartl and Hayer-Hartl, 2002; Morimoto, 2008). Failure of a protein to fold properly, or to retain its folded state, has emerged as the cause of numerous diseases. Aberrant folding is often the result of destabilizing mutations

and may cause the loss of critical functions. However, in a growing number of diseases, misfolding and aggregation results predominantly in a toxic gain of function (Stefani and Dobson, 2003; Winklhofer et al., 2008). In these disorders, specific proteins, differing substantially in size and sequence, typically self-assemble into amyloid-like fibrils with cross-b structure which are deposited within or outside of cells. This phenomenon underlies some of the most debilitating neurodegenerative disorders, including Parkinson’s, Huntington’s, and Alzheimer’s disease. Amyloidogenic aggregation is observed with many protein sequences (Chiti and Dobson, 2006; Goldschmidt et al., 2010) and is often associated with the accumulation of soluble, oligomeric species that precede fibril formation and are thought to be responsible for toxicity (Campioni et al., 2010; Chiti and Dobson, 2006; Jahn and Radford, 2008). The underlying mechanisms are only poorly understood but a prominent hypothesis suggests that the aggregates, in particular the more heterogeneous oligomers, may expose flexible hydrophobic surfaces that can mediate aberrant interactions with other proteins, resulting in their functional impairment and sequestration (Bolognesi et al., 2010; Chiti and Dobson, 2006). In another model, misfolding proteins, by engaging the chaperone machinery, are thought to interfere with central protein quality control and clearance mechanisms, possibly resulting in a propagation of folding defects (Balch et al., 2008; Bence et al., 2001; Gidalevitz et al., 2006). Finally, based on experiments with model membranes, oligomeric aggregation intermediates can compromise the integrity of lipid membranes (Lashuel and Lansbury, 2006). Importantly, these different routes of toxic action are not mutually exclusive but may operate in parallel. To investigate the toxicity mechanisms of amyloid-like aggregation, we have established a cellular model based on the expression of artificial proteins that were designed to form b sheet structures, and shown previously to self-assemble into fibrils in vitro (West et al., 1999). The sequences of these proteins were explicitly designed to contain b strands with an alternating pattern of polar and nonpolar residues, while the exact identities of the side chains were varied combinatorially. Similar bipolar Cell 144, 67–78, January 7, 2011 ª2011 Elsevier Inc. 67

segments occur in 30% of human proteins, including several neurodegenerative disease proteins, but are usually buried within the folded structure (Tartaglia et al., 2008). Because the model proteins were designed de novo, they are not biased by the evolutionary burden of natural proteins and thus allowed us to study the gain-of-function toxicity caused by aggregation without interference either from loss-of-function alterations or from an augmentation of the biological activities of natural disease proteins (Cooper et al., 2006; Lam et al., 2006). Here, we tested specifically the hypothesis that the aggregates engage in widespread aberrant protein interactions. We found that expression of the model proteins in human cells results in aggregate formation and toxicity. Quantitative proteomic analysis reveals that the aggregates interact with and sequester multiple preexistent and newly synthesized proteins. Interestingly, these interactions can be explained in terms of specific sequence features of the coaggregating proteins, such as their multidomain character and their enrichment in disordered regions, properties that are strongly linked with multifunctionality and the occupancy of hub positions in the cellular protein network. We suggest that aberrant interactions with numerous proteins having key cellular functions contribute to aggregate toxicity. RESULTS Designed b Sheet Proteins Are Cytotoxic To investigate the gain-of-function cytotoxicity associated with amyloid-like aggregation, we used several model polypeptides from a combinatorial library rationally designed to form crossb fibrils (West et al., 1999). These proteins, henceforth designated as b proteins, contain six b strands connected by 4-amino acid linker segments, with each strand comprising seven amino acids in a polar-nonpolar alternating pattern. An N-terminal c-Myc-epitope was attached to facilitate detection (Figure 1A). The three proteins chosen for analysis, b4, b17, and b23, differ in sequence (pairwise identities of b strands 35%), with b23 having the highest hydrophobic volume and b sheet propensity, due to its higher isoleucine content (Figure 1A) (Tartaglia et al., 2008). As a control, we used the designed a-helical protein, a-S824, which is similar to the b proteins in amino acid composition but folds into a 4-helix bundle structure (Wei et al., 2003) (Figure 1A). Upon dilution from denaturant into physiological buffer, the purified b proteins adopted b sheet conformation as determined by CD and rapidly assembled into aggregates detectable with the amyloid-binding dyes thioflavin T (ThT) and NIAD-4 (Nesterov et al., 2005) (Figures 1B, 1C, and Figures S1A and S1B available online). The intensity of ThT and NIAD-4 binding was highest for b23, followed by b17 and b4 (Figures 1B and 1C), consistent with the relative b aggregation propensity of these proteins calculated with the sequence-based Zagg method (Tartaglia et al., 2008) (Zagg scores are: b4, 0.79; b17, 0.83; b23, 0.93; a-S824, and 0.30). ANS fluorescence, a probe for exposed hydrophobic regions, suggested the presence of hydrophobic surfaces on the aggregates, in particular for b23 and b17 (Figure 1D). As shown via electron microscopy, the b proteins formed mostly relatively short protofilaments (2–3 nm in diameter) as well as more 68 Cell 144, 67–78, January 7, 2011 ª2011 Elsevier Inc.

heterogeneous, globular aggregates (Figure 1E). Formation of globular species was most pronounced with b23 (Figure 1E). Similar prefibrillar aggregates are also observed with natural amyloidogenic proteins and correlate with cytotoxicity (Bolognesi et al., 2010; Campioni et al., 2010; Chiti and Dobson, 2006). In contrast, in low-salt buffer (pH 6), the b proteins formed thicker (10–12 nm diameter) and longer fibrils (Figure 1F) with FTIR spectroscopic properties characteristic of amyloid (Figure S1C). Thus, the model proteins undergo amyloid-like aggregation in vitro, and in physiological buffer populate prefibrillar species. Upon expression in HEK293T cells for 3 days, the b proteins, but not a-S824, reduced cell viability substantially, as measured by the MTT assay (Figure 2A), and induced cell death in the order b23 > b17 > b4 (Figures S2A and S2B). Cell viability was less impaired after 24 hr of expression, without a significant difference in toxicity between the three b proteins (Figure 2A). Upon cell fractionation, the b proteins were largely recovered in the insoluble fraction, whereas a-S824 was soluble (Figure 2B). Note that b4 and b17 migrated on SDS-PAGE more slowly than expected, but this difference was not observed in urea/ SDS gels (Figure S1A). Confocal immunofluorescence microscopy with anti-Myc antibody showed that the b-protein-expressing cells adopted a collapsed shape lacking filopodia (Figure 2C). Aggregates accumulated mostly in the perinuclear space and the nuclei were often deformed. The aggregates were NIAD-4 positive (Figure 2D), suggesting the presence of amyloid-like material. To detect oligomeric aggregation intermediates, cell extracts were fractionated by size exclusion chromatography followed by dot blot analysis with the A11 antibody, which was raised against the Alzheimer Ab peptide and preferentially recognizes amyloid oligomers associated with cytotoxicity, independent of amino acid sequence (Kayed et al., 2003). b23 expression generated substantially higher levels of A11 reactive material than expression of b4 and b17 (Figure 2E and Figure S2C), consistent with the greater toxicity of b23 and its pronounced tendency to form prefibrillar aggregates in vitro (Figures 1E and 1F). In summary, the designed b proteins resemble amyloidogenic disease proteins in terms of aggregation properties and toxicity and allow us to investigate the mechanism of gain-of-function toxicity independently from evolved biological interactions. Identification of the b Protein Interactome Gain-of-function toxicity of aggregation may arise, at least in part, from aberrant interactions of the aggregates with cellular proteins. To test this hypothesis, we performed a sensitive, quantitative proteomic analysis of the b protein interactome using SILAC (stable isotope labeling with amino acids in cell culture) (Ong and Mann, 2006) and peptide identification by tandem mass spectrometry (LC-MS/MS). These experiments were performed at 24 hr after b protein transfection when cell viability was not yet severely impaired (Figure 2A). In one set of experiments, cells labeled with light (L), medium (M) or heavy (H) arginine and lysine isotopes were transfected with empty vector, a-S824 and b23, respectively. In another set-up, a-S824, b4, and b17 were expressed in L-, M-, and H-labeled cells, respectively (Figure 3A). Preferential interactions with b4,

A

Myc epitope 1 80 β4: MCEQKLISEEDL GMQISMDYQLEIEGNDNKVELQLNDSGGEVKLQIRGPGGRVHFNVHSSGSNLEVNFNNDGGEVQFHMH β17: MCEQKLISEEDLGMQISMDYEIKFHGDGDNFDLNLDDSGGDLQLQIRGPGGRVHVHIHSSSGKVDFHVNNDGGDVEVKMH β23: MCEQKLISEEDL GMQISMDYNIQFHNNGNEIQFEIDDSGGDIEIEIRGPGGRVHIQLNDGHGHIKVDFNNDGGELQIDMH

1

α-S824: MCEQKLISEEDL GMYGKLNDLLEDLQEVLKHVNQHWQGGQKNMNKVDHHLQNVIEDIHDFMQGG

64

65 GSGGKLQEMMKEFQQVLDEIKQQLQGGDNSLHNVHENIKEIFHHLEELVHR 105

C

ThT 15 10 5 0

100 80 60 40 20 0

460

480

500

520

550

β4

600

650

700

β4

β17

β17

100 nm

β4

10 8 6 4 2

400

Wavelength (nm) α-S824

F

ANS

0

500

540

Wavelength (nm)

E

D

NIAD-4 120

Fluorescence (AU)

Fluorescence (AU)

20

Fluorescence (AU)

B

500

550

600

β23

β23

100 nm

β17

100 nm

450

Wavelength (nm)

100 nm

β23

100 nm

100 nm

Figure 1. Amyloidogenic Aggregation of Model Proteins In Vitro (A) Sequences of the model proteins, b4, b17, and b23, designed to form b-sheet fibrils, and a-S824 designed to form a 4-helical bundle. Polar and nonpolar amino acids are indicated in gray and yellow, respectively, b strands and a helices by blue arrows and rods respectively. N-terminal c-Myc tags are shown in red. (B–D) Tinctorial properties of b protein aggregates. The purified proteins indicated (3 mM) were analyzed in 25 mM HEPES buffer (pH 7.5), 150 mM KCl, 0.5 mM MgCl2-containing 20 mM Thioflavin T (B), 1 mM NIAD-4 (C), or 20 mM ANS (D). Fluorescence was recorded as described in Experimental Procedures. (E and F) Transmission electron microscopy of aggregates formed by b4, b17 and b23, as above, at pH 7.5 (E) or 10 mM potassium phosphate (pH 6.0) (F). Proteins were negatively stained and observed at a magnification of 55,0003. See also Figure S1.

b17, or b23 were explored in a third type of experiment. Total cell lysates were prepared essentially without removal of aggregate material and combined 1:1:1 (Figure 3A and Figure S3A). The expressed proteins were quantitatively isolated using anti-Myc antibody coupled to magnetic beads, followed by SDS-PAGE, in-gel digestion, and LC-MS/MS analysis. It seemed plausible that initial coaggregation may be driven by relatively weak interactions, which might introduce a stochastic element in the proteomic analysis. To overcome this problem we based our analysis on extensive biological repetitions of the

experiments. Three proteomic experiments were performed, each consisting of three biological repeats (independent transfections). A protein was identified as b protein interactor when its isotope-labeled peptides were either enriched relative to the a-S824 control or relative to one of the other b proteins with > 95% confidence in at least two of the three repeats of a set (see Extended Experimental Procedures and Figures S3B–S3D). A total of 94 interactors of b23, 73, of b17 and 57 of b4 were identified in experiments of equivalent sampling size, consistent with the relative toxicity of the proteins (Figure 3B Cell 144, 67–78, January 7, 2011 ª2011 Elsevier Inc. 69

Viability (% of control)

A 100

24 h 72 h

Figure 2. Cytotoxicity and Aggregation of Model Proteins in HEK293T Cells

B

(A) Viability of HEK293T cells expressing b4, b17, b23, and a-S824 measured by MTT assay 24 hr and 72 hr after transfection. MTT reduction by kDa T T S P T S P T S P T S P 60 control cells transfected with empty vector, C, was 17 set to 100%. Standard deviations were derived 40 from at least three independent experiments. 10 (B) Solubility of b4, b17, b23, and a-S824 analyzed 20 by fractionation of lysates from cells 24 hr after transfection by centrifugation and immunoblotting 0 C β4 β17 β23 α-S824 with anti-Myc antibody. T, total lysate; S, soluble fraction; P, pellet fraction; C, empty vector control. α-S824 β4 β17 β23 C E Representative results from at least three independent experiments. (C and D) Protein distribution and aggregation in DIC 5 intact cells expressing a-S824, b4, b17, and b23. After 24 hr, proteins were detected by immuno4 fluorescence with anti-Myc antibodies (C). DIC, differential interference contrast images. Amyloid3 Myc like aggregates were detected by staining with 2 NIAD-4 (D) (see Experimental Procedures). Nuclei were counterstained with DAPI. The scale bar 1 D represents 20 mm. Representative images of three independent experiments. 0 NIAD-4 α-S824 β4 β17 β23 (E) Quantification of A11 antibody reactivity in extracts from b-protein-expressing cells 24 hr after transfection. The cumulative dot blot signal of fractions from size exclusion chromatography was corrected for the cumulative anti-Myc signal, indicating the amounts of a-S824, b4, b17, and b23, and expressed relative to the A11 reactivity in a-S824 expressing cells (set to 1) (see Figure S2C for original data). Averages and standard deviations represent at least three independent experiments. See also Figure S2.

C

β4

β17

β23

α-S824

A11 reactivity (AU)

80

and Tables S1–S3). Only four proteins were marginally enriched on a-S824 relative to the vector only control, including two ribosomal proteins (Figure S3B). Approximately 60% of the b4 and b17 interactors were also found to interact with b23, indicating a high degree of overlap in interaction profiles (Figure 3B). Western blotting of pulldowns and immunofluorescence analysis of cells confirmed the results from SILAC/MS for several interactors (Figures S3E and S3F). Thus, interactions of the b protein aggregates with multiple endogenous proteins precede the strong decrease in cell viability observed at 72 hr after transfection (Figure 2A). As summarized for b23, most of the proteins associated with the aggregates have their primary location in the cytoplasm, nucleus and mitochondria (Figure 3C and Table S1). Proteins involved in chromatin regulation, RNA processing, transcription, translation, cytoskeletal function, vesicle transport, and protein quality control were highly represented. These proteins are generally of average cellular abundance (Su et al., 2002) and for several of them between 10% and 45% of total was associated with the aggregates, based on depletion from supernatant fractions after pulldown as measured by SILAC/ MS (Table S1 and Extended Experimental Procedures). Note that this analysis probably underestimates the extent of sequestration, since coaggregates may partially dissociate during isolation. Interestingly, 12 different translation initiation factors interacted directly or indirectly with the aggregates, including 9 of the 13 subunits of the eIF3 complex and 3 subunits of eIF4 (Figure 3C). b17 aggregates contained 10 and b4 aggregates 9 of these proteins (Tables S2 and S3). 70 Cell 144, 67–78, January 7, 2011 ª2011 Elsevier Inc.

Immunofluorescence analysis demonstrated extensive colocalization of eIF3D and eIF4GII with the aggregates and western blotting of pulldowns confirmed that at least 10% of cellular eIF3D coaggregated with b23 and b17, compared to 6% with b4 (Figures S3E and S3F, and data not shown). Indeed, labeling experiments showed that cells expressing b23 for 24 hr had a 35% reduced protein synthesis capacity (Figure S3G). Similarly, the altered morphology of b-proteinexpressing cells revealed by actin staining (Figure S3H) may be attributed to the association of filamins A, B, C, (FLNA, FLNB, FLNC), and the giant protein plectin-1 (PLEC1) (500 kDa) with the aggregates (Figure 3C), proteins that are critical for the formation and maintenance of cytoskeletal architecture. These results show that many different proteins, involved in a range of essential cellular functions, are affected by the b protein aggregates. Aberrant Stress Response in b-Protein-Expressing Cells The proteomic analysis identified several cytosolic chaperones and chaperone regulators to be associated with the aggregates, including Hsc70 (Hsc71) and its cochaperones Hsp110 (Hsp105), Hdj1/2 and Bag2, as well as the nascent chain associated complex, NAC (Figure 3C). Hsp110 was enriched in the aggregates in a manner correlating with the relative toxicity of b4, b17 and b23, as confirmed by western blotting and immunofluorescence (Figures 4A and 4B). Indeed, overexpression of Hsp110 (Figure S4A) partially suppressed b4 and b17 aggregation and toxicity but was inefficient in mitigating the toxic effects of b23 (Figures 4C and Figures S4B and S4C).

A

L: Arg0,Lys0

M: Arg6,Lys4 H: Arg10,Lys8

α-S824

β23

β4

β17

β17

β23

Expt. II: α-S824 Expt. III:

β4

mix lysates 1:1:1

Intensity

Expt. I

Expt. I: Vector only

LC-MS/MS

IP

β23

Expt. II β17

α-S C 824

C

β4 (57)

Miscellaneous ABCE1

m/z

SBDS

BAG2

HSPA8

CHORDC1

HSPH1

Translation EEF1A1

EIF3I

DNAJA1

NACA

EIF3A

EIF3L

CACYBP

UBC

DNAJB1

NACB

EIF3C

EIF3M

ERLIN2

UBR4

EIF3D

EIF4A1

EIF3E

EIF4A3

EIF3F

EIF4G3

SLC25A6

VDAC1

EIF3G

IGF2BP3

CHCHD3

VDAC2

IMMT

VDAC3

Vesicle transport AP1B1

CLTC

AP1G1

SEC16A

AP1M1

VAPA

tRNA synthetases EPRS

WARS YARS

SSBP1

Metabolism

31 15

RNA processing Ribosome biogenesis

ALDH18A1

23

RBM8A

DIMT1L

GEMIN4

SMN1

CHERP

NVL

GEMIN5

SR140

HDLBP

PDCD11

Nuclear structure Cytoskeleton AKAP12

β17 (73)

BAT1

CAD

AHNAK

β23 (94)

Mitochondria

4

1 47

m/z

Protein degradation

SUMO2

21

β17

Molecular chaperones

OGT

FAM120A

β23

β4

α-S β4 824

m/z

B

Expt. III

LMNB1

PLEC1

ARHGDIA

Nuclear transport SEPT2

CCDC88A

SEPT7

KPNA2

DIAPH1

SEPT9

RANBP1

FLNA

TUBA1A

FLNB

TUBB2C

FLNC

VIM

KIF5B

ZYX

MARCKS

NUMA1

THOC2

STRAP

WDR3

Chromatin structure RUVBL2

H1F0

SMC4

CAND1

PNN

SMARCA4

CNOT1

PURA

SMARCA5

GTF2I

SND1

RBBP4 RUVBL1

SMARCC1

Unspecified function BAT2D1 HCFC1

Transcription

CHD4

PCNP

DNA maintenance

PRKDC TNKS1BP1

Figure 3. Interactome Analysis of b proteins (A) Design of SILAC experiments to identify b protein interactors by LC-MS/MS. L, M, and H, light, medium, and heavy isotope media. C, vector only control. (B) Overlap between the interactor sets of b4, b17, and b23. Total numbers of identified interactors are given in parentheses. (C) The b23 interacting proteins are grouped according to cellular location and function. See also Figure S3 and Tables S1–S3.

Remarkably, expression of the b proteins did not induce the cytosolic stress response or heat-shock response (HSR), as no increase in the levels of Hsp110, Hsp70, or Hsp27 was observed (Figures 4A and data not shown). A possible defect in the HSR was further analyzed using a luciferase reporter gene under control of the HSF1-dependent Hsp70 promoter (Williams et al., 1989). While inhibition of proteasome function by MG132 in control cells resulted in a 5-fold induction of the reporter, this induction was completely abolished in cells expressing the b proteins for 24 hr (Figure 4D). The phorbol-12-myristate-13acetate (PMA)-mediated induction of a luciferase reporter under the NF-kB promoter was also impaired, but to a lesser extent than the inhibition of the stress response (Figure 4E). Thus, expression of the model b sheet proteins leads to a deficiency of the normal cytosolic stress response, thereby limiting the capacity of cells to mount an effective defense.

Structural Features of b Protein Interactors A bioinformatic analysis of the physicochemical properties of the b protein interactors was conducted to see whether these proteins share certain structural features. We focused our initial analysis on the interactors of b23 (Table S1). Compared to a set of 3055 control proteins identified by LC-MS/MS in a total cell lysate (Table S4), the b23 interactors are shifted to higher molecular weight, with a significantly greater fraction of proteins above 150 kDa (p < 0.005) (Figure 5A). In addition, the interactors have a lower average hydrophobicity and a bimodal hydrophobicity distribution (Kyte and Doolittle, 1982) (p < 0.005) (Figure 5B). The occurrence of domain folds among the b23 interactors, as classified in SCOP, was generally similar to that of lysate proteins. However, the b23 interactors contained significantly more proteins with all beta domains (SCOP class b) (p < 0.05) (Figure S5A), including the beta-barrel VDAC proteins of the Cell 144, 67–78, January 7, 2011 ª2011 Elsevier Inc. 71

C

Hsp110

C

β4 β17 β23 α-S824

IP Lysate Immunoblot

B

Myc

Hsp110

Merge

Viability (% of control)

A

Figure 4. Impairment of the Stress Response in b-Protein-Expressing Cells

120 100 80 60 40 20 0 −

C

D

10

Promoter activity (fold induction)

α-S824

β4

β17 − +

β23 α-S824 − + − +

HSP70-Luc

1 C

10

1 C

outer mitochondrial membrane and the filamins which have Ig-domain repeats (Table S1 and Figure S3F). The lower hydrophobicity of many interactors suggested that these proteins may be rich in intrinsically unstructured regions (IURs). Indeed, compared to lysate proteins, the b23 interactors have a significantly greater fraction of total amino acids in IURs (p < 0.05), based on the DisoDB database (Pentony and Jones, 2009) and the DisEMBL and IUPred prediction tools for unstructured regions (Dosztanyi et al., 2005; Linding et al., 2003) (Figure 5C, Tables S1 and S4, and data not shown). Pronounced differences emerged when considering the fraction of proteins with IURs longer than 30 or 50 residues (Figure 5D). For example, 60% of the b23 interactors are predicted to contain at least one unstructured segment of 30 amino acids, compared to 45% in lysate proteins or the complete proteome (p < 0.005) (Figure 5D). The corresponding numbers for IURs > 50 amino acids are 40% and 25% (p < 0.005), respectively. Moreover, the b23 interactors contain on average 3 disordered segments of 30 amino acids (compared to 1.8 for the lysate proteins, p < 0.005). The predicted IURs of the interactors are shifted to greater length (p < 0.005) (Figure S5B and Tables S1 and S4), with 21% of the proteins containing IURs > 80 amino acids 72 Cell 144, 67–78, January 7, 2011 ª2011 Elsevier Inc.

β4 β17 β23 α-S824

NF-κB-Luc

100

Promoter activity (fold induction)

E

β23

β4 − +

Hsp110 overexpression

C

β17

+

β4 β17 β23 α-S824

(A and B) Association of Hsp110 with b protein aggregates. HEK293T cells were transfected as indicated (C, empty vector control). 24 hr after transfection, Myc-tagged proteins b4, b17, b23, and a-S824 were immunoprecipitated from cell lysates and analyzed by immunoblotting with antiHSP110 antibody (A). Lysate samples correspond to 8% of input used for IP. b4, b17, and b23 in pulldowns was associated with 5%, 9%, and 16% of total cellular Hsp110, respectively (also see Figure S3F). For immunofluorescence analysis (B), cells were fixed and costained with anti-Myc antibodies and anti-Hsp110 antibodies. Nuclei were stained with DAPI. Representative examples of three independent experiments are shown. (C) Partial rescue of b protein toxicity by Hsp110 overexpression. Cells were transfected with empty vector or the expression vector for human Hsp110. 24 hr later, cells were electroporated with empty vector, C, or expression vectors for b4, b17, b23, and a-S824. Three days after the second transfection, MTT assays were performed. Empty vector control was set to 100% viability. Standard deviations of three independent experiments are shown. (D and E) Inhibitory effect of b proteins on cellular stress response pathways. Cells were cotransfected with HSP70-luciferase reporter (D) or NF-kB-luciferase reporter constructs (E) and the b-protein-expressing plasmids. 6 hr later, 5 mM MG132 (D) or 16 mM PMA (E) were added to induce the respective promoter. Luciferase activity was measured 24 hr after transfection. The promoter activity in cells transfected with control vector, C, without inducer was set to 1. Standard deviations of three independent experiments are shown. See also Figure S4.

and 10% of 100 to 429 residues. The IURs are enriched in polar amino acids and in amino acids that have a high propensity to form coil and turn regions, such as M, K, R, E, S, Q, and P, and are depleted in aromatic and hydrophobic amino acids W, Y, F, C, I, and V (Dunker et al., 2008). Such sequences are structurally flexible and populate a range of conformational states from extended disordered to collapsed, molten globule-like structures (Dunker et al., 2008; Pentony and Jones, 2009). Using the Zagg algorithm to predict aggregation propensities, the b23 interactors have higher aggregation scores than lysate proteins (see Figure S6B below). A comparison of the proteins that interact preferentially with b4, b17, and b23 (18, 28, and 27 proteins, respectively), as defined by the SILAC experiments (Figure S3D and Extended Experimental Procedures), revealed that a gradual increase in molecular weight and decrease in hydrophobicity of the interactors, along with a slight increase in their disorder, correlated with the differential cytotoxicity of the three b proteins (Figure 5E). This trend was also observed when comparing the complete interactor sets of the three b proteins (Tables S1–S3). The prominent association of proteins with low hydrophobicity and high intrinsic disorder with the aggregates was

30 Molecular weight (kDa)

% of identified proteins

A

20

10

p<0.005

350 300 250 200 150 100 50 0

Lysate proteins

β23 interactors

640

600

560

520

480

320

280

240

200

160

120

80

40

0

0

Molecular weight (kDa) p<0.005

14

0.2 0.0

12

Hydrophobicity

% of identified proteins

B

10 8

Figure 5. Structural b Protein Interactors

Properties

of

the

(A and B) Distribution of molecular weight (A) and average hydrophobicity (B) of lysate proteins (gray) and b23 interactors (red) (Tables S1 and S4). Box plots (insets) indicate the distribution of the data. Dashed horizontal line indicates the median, whisker caps and circles indicate 10th/90th and 5th/95th percentiles, respectively. P values are based on Mann-Whitney test. (C and D) Disorder analysis of b protein interactors. Percentage of disordered residues in interactor sequences (C). p value based on Mann-Whitney test. Fraction of proteins with disordered stretches longer than 30 or 50 amino acids (D). **p < 0.005 based on a chi-square test. Disorder was determined using DisoDB. (E) Structural properties of lysate proteins and of proteins interacting preferentially with b4, b17, or b23. See also Figure S5 and Tables S1–S6.

-0.2 -0.4 -0.6 -0.8

(Lee et al., 2006). The Arctic mutant formed visible aggregates more readily 4 and showed substantially greater toxicity -1.2 Lysate β23 than WT Ab142 (Figures S5D and S5E). 2 proteins interactors Analysis by SILAC/MS revealed that 0 the Ab interactome is comparable in complexity to that of the b proteins, with a direct overlap of 25%, promiHydrophobicity Erlin nently including translation initiation C D E Preferred interactors factors, chromatin regulators, RNA prop<0.05 cessing proteins, mitochondrial mem24 ** brane proteins and chaperones (Table 60 Lysate proteins 20 70 S5). We also identified 31 proteins which 50 β4 interactors 60 16 were enriched on the Arctic mutant relaβ17 interactors 40 50 ** tive to the less toxic Ab142 WT (Table 40 β23 interactors 12 30 30 S6). Notably, these proteins resemble 8 20 20 -0.55 the b protein interactors in physico10 -0.50 4 10 -0.45 0 chemical properties and are significantly -0.40 20 40 0 0 60 -0.35 Disord enriched in IURs (p < 0.05) (Figures 80 a.a. >30 >50 er 100 [prote >30 a.a. S5F–S5I). in s (%)] Lysate proteins From these results, we conclude that β23 interactors cells contain a subpopulation of metastable proteins that are prone to interact with and potentially become sequestered unexpected. To test whether such proteins are targeted more by toxic species populated in the process of amyloid-like generally by amyloid-like aggregation, we performed an initial aggregation. analysis of interactors of wild-type Ab1-42 and its Arctic mutant (E22G), which causes early-onset Alzheimer’s disease. The b Protein Interactors Include Pre-Existent and Newly latter was included because it is known to populate higher Synthesized Proteins levels of prefibrillar aggregates and toxic oligomers exposing While the results above suggested that structural flexibility is crithydrophobic surfaces (Bolognesi et al., 2010; Nilsberth et al., ical in facilitating the interaction of endogenous proteins with the 2001). To allow a comparison with the model b aggregates, b aggregates, we noted that for 40% of the b23 interactors no the Ab proteins were also expressed in the cytosol, using IURs > 30 amino acids are predicted (Figure 5D and Table S1). GFP fusions (Kim et al., 2006). In contrast to the artificial We therefore considered the possibility that some of these b proteins, the Ab constructs were degraded but accumulated proteins may succumb to coaggregation upon synthesis upon partial proteasome inhibition with MG132 (Figure S5C) before adopting stably folded structures. To test this idea, we 6

1.0

0.8

0.6

0.4

0.2

0.0

-0.2

Hy

dr

op ho

bi cit

y

t (kDa) Molecular weigh

-0.4

-0.6

-0.8

-1.0

-1.4

-1.2

Proteins with disordered stretches (%)

Disordered residues (%)

-1.6

-1.0

Cell 144, 67–78, January 7, 2011 ª2011 Elsevier Inc. 73

A

L: Arg0,Lys0

Figure 6. Newly Synthesized and Pre-Existent b Protein Interactors

B 50

M: Arg6,Lys4

45 40 35

β23

Transfection : α-S824

H: Arg10,Lys8

M: Arg6,Lys4

L: Arg0,Lys0

New

30 25

β23 interactors

20

Disordered residues (%)

C 35 30 25

p<0.05

Intensity

α-S 824 β23 m/z Preexistent (Old)

β23 α-S 824 β23

m/z Newly-synthesized (New)

D

Old

10 5

-0.6 -0.3 0.0 0.3 0.6 0.9 1.2 1.5

Log(H/Mβ23)/(H/Mlysate)

E 80 70 60 50 40 30 20 10 0

** **

Lysate proteins Old β23 interactors New β23 interactors

t (kDa) Molecular weigh

β23 interactor:

β23

Proteins with disordered stretches (%)

Intensity

15

(A) Design of SILAC based mass spectrometric analysis to identify proteins preferentially interacting with b23 as newly synthesized (new) or preexistent (old) proteins (Pulse-SILAC). (B) Ratios of heavy to medium isotopes (H/M) of b23 interactors relative to the H/M ratios for the same proteins in the total cell lysate (see Extended Experimental Procedures). The log of this ratio of ratios increases with the tendency of a protein to interact with b23 as a new protein. (C and D) Disorder analysis of new and old b23 interactors. Percentage of disordered residues in interactor sequences (C). p value is based on Mann-Whitney test. Fraction of proteins with continuous disordered stretches > 30 or > 50 amino acids (aa) (D). **p < 0.005 for old interactors relative to lysate (see Figure 5D), based on Chisquare test. (E) Molecular weight, disorder and hydrophobicity of old and new b23 interactors relative to lysate proteins. See also Figure S6 and Tables S4 and S7.

70 60

translation initiation factors (Table S7), consistent with impairment of translation 15 efficiency being an early consequence 10 -0.7 of b protein toxicity (Figures S3E–S3G). -0.6 5 -0.5 Interestingly, the old and new interac-0.4 20 40 0 60 -0.3 Disord tors from the ends of the distribution (15 80 a.a. >30 >50 er 100 [prote >30 a.a. proteins each) differ markedly in their ins (% Old β23 interactors )] structural properties. The old interactors New β23 interactors contain a significantly greater fraction of amino acids in IURs than the new interactors (Figure 6C). They are strongly enpulse-labeled HEK293T cells expressing b23 or a-S824 with riched in continuous disordered regions (Figure 6D) and are of 35 S-methionine, followed by immunoisolation of the proteins. low average hydrophobicity (Figure 6E). In contrast, the new Around 7% of the proteins labeled within 15 min were coisolated proteins are similar to lysate proteins in terms of hydrophobicity, with b23, compared to only 1% with a-S824 (Figure S6A), sug- but are lower in disorder and substantially larger in size (Figgesting that a substantial fraction of newly synthesized polypep- ure 6E). Their folding pathways may be complex and kinetically slow, possibly resulting in the prolonged exposure of hydrotides can interact with b23. To identify such proteins, we performed pulse-SILAC experi- phobic residues during folding. Based on analysis using the ments. Cells were cultured with medium amino acid isotopes Zagg algorithm, the new b23 interactors show high intrinsic (M) to label preexistent proteins, followed by transfection with aggregation scores only in their unfolded states (Figure S6B). b23. The culture was divided and one half was immediately In contrast, the old interactors are highly flexible and are unable shifted to media containing heavy amino acid isotopes (H). to bury aggregation-prone regions in their native structures. Control cells were cultured with light amino acids (L) and trans- Thus, these proteins have high aggregation scores both in their fected with a-S824. After 24 hr, the cells from the three condi- unfolded and folded states (Tartaglia et al., 2008) (Figure S6B). tions were combined and subjected to anti-Myc pulldown and Some of them may coaggregate as preexistent or newly syntheLC-MS/MS analysis (Figure 6A). The H/M isotope ratio of the sized proteins, consistent with their lower peak values in the b23 interactors in the pulldown relative to their H/M ratios in isotope labeling ratios compared to the new proteins (Figure 6B). In summary, the b23 interactors can be divided into two overthe lysate was used to indicate whether they interact with b23 preferentially as newly synthesized (New) or pre-existent (Old) lapping subsets of relatively aggregation-prone proteins: One proteins (Figure 6A). H/M labeling ratios were obtained for 50 group is enriched in IURs, which would be prone to aggregate b23 interactors, and a number of these showed a clear prefer- even in their post-folding state. This group is highly represented ence for interaction soon after synthesis (Figure 6B and Table among the old proteins. The second group contains an abunS7). In contrast, fewer proteins interacted preferentially as old dance of large and/or multidomain proteins, which require longer proteins. These interactors include Hsp110 as well as several times for synthesis and may fold slowly. Consequently, in dr

op

ho

bi

cit

y

50 40 30 20 10 0

Hy

20

74 Cell 144, 67–78, January 7, 2011 ª2011 Elsevier Inc.

Figure 7. Mechanism of b Aggregation Toxicity

B

12.4

80 7.5

7.2

Disease proteins

β4 β17 β23

60

4.9 2.8

HPRD

Lysate

20

β23

1000 850 700 550

to r

0.8

0.4

40

s

1.5

0 200

All interactions Essential Disease

Esse

250

ntial

ac

1.6

300

intera

In te r

13 12 11 10 9 8 7 6 5 4 3 2 1 0

Average interactions /Protein

A

350

ctors

C

Proteins or protein complexes containing disordered regions

Chaperones

co-aggregation

Structured proteins

Chaperones

Newlysynthesized

co-aggregation

Toxic β-aggregates/ oligomers

Amyloidogenic regions (red)

Proteins not yet folded and/or assembled

conditions of limited chaperone capacity, they would be prone to aggregate during and shortly after synthesis. This group is enriched among the newly synthesized proteins. Finally, some proteins occupy a transition zone, combining physicochemical features of both groups. Aggregate Interactors Have Critical Network Functions The structural flexibility and relatively large size of the aggregate interactors suggests that these proteins may normally be involved in numerous functional protein interactions. To address this possibility, we analyzed how the b23 interactors are linked with the cellular protein network. A query of the Human Proteome Reference Data Base (HPRD) (Keshava Prasad et al., 2009) revealed that each of these proteins functionally interacts with 12 different proteins on average, compared to 7 per lysate protein and 7.5 per protein in HPRD (19,651 entries) (Figure 7A). Notably, most of the b23 interactors have no or only few interactions with any of the other b23 interactors, suggesting that coaggregation may disrupt their functional complexes. For example, the microfilament protein vimentin interacts with more than 100 different proteins according to HPRD, but only three of those are among the identified b23 interactors, although 49 potential vimentin interactors were detected in the lysate or background of the pulldowns (data not shown). Essential proteins often occupy critical ‘‘hub’’ positions in the network (Haynes et al., 2006; Jeong et al., 2001). Each b23 target protein interacts on average with 5 different essential proteins, compared to only 3 per lysate protein and 1.5 per entry in HPRD (Figure 7A). Moreover, the b23 interactors are more frequently linked than lysate proteins, through direct interactions, with proteins that have been found in association with

(A and B) Functional context of b protein interactors within the protein interaction network. Shown in (A) are the average number of functional interactions of the b23 interactors in comparison to proteins in the HPRD database and in the experimentally determined cell lysate (3000 proteins). These functional interactions are categorized into total interactions, interactions with essential proteins and interactions with proteins involved in neurodegenerative disease (Raychaudhuri et al., 2009). In (B), the complete sets of b4, b17, and b23 interactors are compared in terms of these functional properties. (C) Model for the interaction of b aggregates with pre-existent and newly synthesized proteins. Preexistent proteins are structurally flexible in their functional state and are involved in multiple protein-protein interactions, which may be disrupted by their association with the b aggregates. The newly synthesized proteins are structurally vulnerable to coaggregation during folding and assembly. Interaction of both the preexistent and newly synthesized proteins with the b aggregates is facilitated by the limiting capacity of chaperones to shield aggregate surfaces and by the failure of the cells to mount an efficient stress response.

neurodegenerative disease proteins (Raychaudhuri et al., 2009) (Figure 7A and Table S1). Assuming that a disturbance of functional protein interactions contributes critically to b aggregation toxicity, b23 would be expected to differ in this regard from the less cytotoxic proteins b4 and b17. We found that the b4, b17 and b23 interactors are physically linked to a total of 600, 643, and 912 different proteins, including 216, 213, and 340 essential proteins and 53, 56, and 84 proteins associated with neurodegenerative disease networks, respectively (Figure 7B). Thus, the capacity of the b protein aggregates to interact with and sequester highly connected cellular proteins correlates well with their relative cytotoxicity. DISCUSSION Widespread Coaggregation of Metastable Proteins A key finding of this study is that amyloidogenic aggregation can result in the sequestration of numerous proteins that share distinct physicochemical properties: They are relatively large in size and exhibit high structural flexibility, with a significant enrichment in disordered regions, features that are strongly linked with multifunctionality (Figure 7C). The artificial b sheet proteins used as a model were designed to assemble into fibrils (West et al., 1999). Like natural amyloidogenic proteins, they populate a range of prefibrillar aggregation intermediates, which are likely to represent the primary toxic agents in aggregation diseases (Chiti and Dobson, 2006; Jahn and Radford, 2008). Based on recent findings, the proteotoxicity of such species correlates with the exposure of ANS-binding hydrophobic surfaces (Bolognesi et al., 2010; Campioni et al., 2010) and reactivity with the A11 anti-oligomer antibody Cell 144, 67–78, January 7, 2011 ª2011 Elsevier Inc. 75

(Kayed et al., 2003), properties that are reproduced by the model proteins. Flexible hydrophobic surfaces and unpaired backbone structure that is not yet integrated into a stable cross-b core (Mossuto et al., 2010) may endow oligomers and protofilaments with the capacity to engage in widespread aberrant interactions with metastable proteins. Whether oligomeric species with similar interaction properties also occur during nonamyloidogenic aggregation, leading to amorphous structures rather than fibrils, remains to be determined. The b protein aggregates were found to interact with preexistent and newly synthesized polypeptides (Figure 7C). The former are strongly enriched in intrinsically unstructured regions (IURs) and are of lower average hydrophobicity. A similar trend was observed for interactors of the toxic aggregates of the Arctic mutant of Ab1-42, which is known to transiently populate high concentrations of prefibrillar aggregates (Bolognesi et al., 2010) (Figures S5C–S5I). Proteins rich in structural disorder are considered to be adaptable to multiple interaction partners (Dunker et al., 2008; Pentony and Jones, 2009). On the other hand, local structural fluctuations in these proteins are expected to give rise to the exposure of sequence elements with a higher propensity to form aggregates, consistent with the relatively high Zagg scores of the b protein interactors (Tartaglia et al., 2008). Indeed, some of the best known neurodegenerative disease proteins, such as a-synuclein or tau, are thought to be almost entirely unstructured. In contrast, the proteins that interact with the aggregates during or soon after synthesis have average hydrophobicity and disorder. These proteins tend to be large in size and are likely to populate nonnative states which expose hydrophobic surfaces during their folding, assembly or transport that must be shielded by molecular chaperones (Figure 7C). For example, among the b protein interactors are mitochondrial membrane proteins such as VDAC and the ADP/ATP translocase which require chaperone protection during post-translational sorting (Young et al., 2003). By targeting flexible regions and hydrophobic surfaces of preexistent and newly synthesized proteins, the b protein aggregates may act in a ‘chaperone-like’ manner but cannot promote folding through regulated release. Consequently, more and more proteins are recruited, which in turn may generate new interaction surfaces, thereby magnifying the toxic potential of the aggregates (‘snowball effect’). Interference with Multiple Key Cellular Functions The b protein interactors include many proteins with key cellular functions in transcription and translation, chromatin regulation, vesicular transport, cell motility and architecture, as well as protein quality control (Figure 3C). Similar proteins were also found to interact with aggregates of Ab (Table S5), suggesting that these pathways may be more generally at risk in aggregation disorders. Bioinformatic analysis showed that most of the coaggregating proteins have numerous functional interactors, consistent with their preferential role as network hubs (Haynes et al., 2006; Jeong et al., 2001). The number of functional interactions of the sequestered proteins correlates with the relative cytotoxicity of the b protein aggregates (Figure 7B). It is thus likely that the aggregates compete for binding to disordered regions with a protein’s normal interactors and the more toxic forms may be 76 Cell 144, 67–78, January 7, 2011 ª2011 Elsevier Inc.

able to compete more effectively. Based on our proteomic and biochemical measurements, 10%–45% of total may be sequestered for several of the interacting proteins (Figures S3E–S3F and Table S1). Moreover, certain proteins may misfold upon interaction with the aggregates but remain in solution. Thus, dependent on the interaction strength of the aggregates, an increasing number of key functions may be affected, eventually resulting in fatal network collapse. We estimate that the human proteome contains 2000 proteins that structurally resemble the experimentally identified b aggregate interactors. Dependent on cell type and the exact structural properties of the causative aggregate, different subsets of these proteins may be affected, which may help to explain different patterns of pathobiology. It will be interesting to see which of these proteins are preferentially targeted by b aggregates in neuronal cells. Our results also lend support to the recent view that protein misfolding and aggregation disturbs proteostasis by compromising the cellular folding environment (Morimoto, 2008). We suggest that the association of endogenous proteins with the aggregates is facilitated by the failure of the affected cells to mount an efficient stress response, a phenomenon that was previously observed during prion infection (Tatzelt et al., 1995) and may be particularly serious in postmitotic cells, such as neurons. Inhibition of the stress response may be due to the sequestration by the aggregates of multiple chromatin regulators, which interact with numerous transcription factors, including HSF1 (Erkina et al., 2010; Sullivan et al., 2001). As a consequence of limiting proteostasis capacity, newly synthesized polypeptides with a high chaperone requirement for folding may become increasingly vulnerable to sequestration by disease protein aggregates (Figure 7C). This fatal chain of events may be further enhanced during aging, which is associated with a decline of proteostasis and thus would result in a reduced capacity of cells to protect their more vulnerable proteins against coaggregation (Balch et al., 2008; Morimoto, 2008). EXPERIMENTAL PROCEDURES Protein Purification and In Vitro Analysis of Aggregates Proteins a-S824, b4, b17 and b23 were expressed in E. coli BL21 cells and purified as described in Extended Experimental Procedures. Fluorescence analysis, circular dichroism, FTIR spectroscopy and negative stain electron microscopy of the aggregates were performed using standard methods (see Extended Experimental Procedures). Cell Culture, Immunoblotting and Reporter Assays Human HEK293T cells were cultured under standard conditions (see Extended Experimental Procedures). Transient transfections were performed by electroporation with 30 mg expression vector or by Lipofectamin (Invitrogen) transfection for overexpression of Hsp110. Immunoblots were developed using the chemiluminescence kit Rodeo ECL (USB) and analyzed using a LAS-3000 image reader (Fujifilm) and the AIDA software (Raytest). For luciferase reporter assays, cells were lysed in Lysis Buffer (Promega) and luciferase activity measured using a Lumat LB9507 (EG&G Berthold). Cell Viability Cell viability was analyzed by measuring the capacity of cells to reduce 3-(4,5Dimethylthiazol-2-yl)-2,5-diphenyltetrazoliumbromide (MTT) to formazan at different times after transfection with a-S824, b protein or Ab42-GFP constructs (Shearman, 1999).

Solubility Analysis and Oligomer Quantification Cells were lysed in Triton X-100/Na deoxycholate-containing PBS with protease inhibitors. Benzonase was used to hydrolyze DNA. Raw debris was removed at 2000 3 g for 5 min and the supernatant was fractionated by centrifugation (100.000 3 g, 30 min) into pellet and soluble fractions or by gel filtration on a Superose 6 column (Amersham Bioscience), followed by dot blot analysis with anti-oligomer antibody A11 (Kayed et al., 2003) (see Extended Experimental Procedures for details). Immunofluorescence and Fluorescence Imaging Transfected cells were fixed with paraformaldehyde, permeabilized with Triton X-100 and stained with antibodies as indicated. Images were recorded with a Leica TCS SP2 confocal laser scanning microscope. Protein aggregates were analyzed by staining with NIAD-4 (ICX Nomadics) (see Extended Experimental Procedures). SILAC and Sample Preparation for LC-MS/MS Analysis Labeling of cells was performed in custom medium supplemented with light (L), medium (M) or heavy (H) arginine and lysine isotopes (see Extended Experimental Procedures). In pulse-SILAC experiments, M-labeled cells were shifted to H-medium, as indicated in Figure 6A. Cells were lysed and cell debris removed by low-speed centrifugation (2000 3 g, 5 min). Lysates from L, M and H cells were adjusted to equal protein concentration and mixed at a 1:1:1 ratio. An aliquot of this mix was set aside as ‘‘lysate’’ control. Anti-Myc or anti-GFP MicroBeads (Miltenyi Biotech) were used to isolate the Myc-tagged proteins or GFP-fusion proteins and their interactors. The bound proteins were eluted and processed as described (Ong and Mann, 2006). The spectra were interpreted using MaxQuant version 1.0.12.31 (Cox and Mann, 2008) combined with Mascot version 2.2 (Matrix Science, www.matrixscience. com). See Extended Experimental Procedures for details. The raw MS data along with a full list of identified proteins and quantitations is available at https://proteomecommons.org/tranche, entering the following hash: +Ff0/ p8lSBrrzCKZfzAwYS3+Bqw5fonokB679f136te2iklhHtFMUpeT5SM/I3XuufTyr Xj0ycVVC6G4Li/L02 dA4jcAAAAAAABVfg = =. Bioinformatic Analysis Average hydrophobicity was calculated according to Kyte and Doolittle (1982), protein disorder using the DisoDB database (Pentony and Jones, 2009) and aggregation propensities according to Tartaglia et al. (2008). Protein fold prediction and the analysis of functional protein interactions are described in Extended Experimental Procedures. Student’s t test and Mann-Whitney test were used to compare groups. Chi-square tests were used to determine significant differences between categorical data. SUPPLEMENTAL INFORMATION

REFERENCES Balch, W.E., Morimoto, R.I., Dillin, A., and Kelly, J.W. (2008). Adapting proteostasis for disease intervention. Science 319, 916–919. Bence, N.F., Sampat, R.M., and Kopito, R.R. (2001). Impairment of the ubiquitin-proteasome system by protein aggregation. Science 292, 1552–1555. Bolognesi, B., Kumita, J.R., Barros, T.P., Esbjorner, E.K., Luheshi, L.M., Crowther, D.C., Wilson, M.R., Dobson, C.M., Favrin, G., and Yerbury, J.J. (2010). ANS Binding Reveals Common Features of Cytotoxic Amyloid Species. ACS Chem. Biol. 5, 735–740. Campioni, S., Mannini, B., Zampagni, M., Pensalfini, A., Parrini, C., Evangelisti, E., Relini, A., Stefani, M., Dobson, C.M., Cecchi, C., et al. (2010). A causative link between the structure of aberrant protein oligomers and their toxicity. Nat. Chem. Biol. 6, 140–147. Chiti, F., and Dobson, C.M. (2006). Protein misfolding, functional amyloid, and human disease. Annu. Rev. Biochem. 75, 333–366. Cooper, A.A., Gitler, A.D., Cashikar, A., Haynes, C.M., Hill, K.J., Bhullar, B., Liu, K.N., Xu, K.X., Strathearn, K.E., Liu, F., et al. (2006). Alpha-synuclein blocks ER-Golgi traffic and Rab1 rescues neuron loss in Parkinson’s models. Science 313, 324–328. Cox, J., and Mann, M. (2008). MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteome-wide protein quantification. Nat. Biotechnol. 26, 1367–1372. Dosztanyi, Z., Csizmok, V., Tompa, P., and Simon, I. (2005). IUPred: web server for the prediction of intrinsically unstructured regions of proteins based on estimated energy content. Bioinformatics 21, 3433–3434. Dunker, A.K., Silman, I., Uversky, V.N., and Sussman, J.L. (2008). Function and structure of inherently disordered proteins. Curr. Opin. Struct. Biol. 18, 756–764. Erkina, T.Y., Zou, Y., Freeling, S., Vorobyev, V.I., and Erkine, A.M. (2010). Functional interplay between chromatin remodeling complexes RSC, SWI/ SNF and ISWI in regulation of yeast heat shock genes. Nucleic Acids Res. 38, 1441–1449. Frydman, J. (2001). Folding of newly translated proteins in vivo: The role of molecular chaperones. Annu. Rev. Biochem. 70, 603–647. Gidalevitz, T., Ben-Zvi, A., Ho, K.H., Brignull, H.R., and Morimoto, R.I. (2006). Progressive disruption of cellular protein folding in models of polyglutamine diseases. Science 311, 1471–1474. Goldschmidt, L., Teng, P.K., Riek, R., and Eisenberg, D. (2010). Identifying the amylome, proteins capable of forming amyloid-like fibrils. Proc. Natl. Acad. Sci. USA 107, 3487–3492. Hartl, F.U., and Hayer-Hartl, M. (2002). Molecular chaperones in the cytosol: from nascent chain to folded protein. Science 295, 1852–1858.

Supplemental information includes Extended Experimental Procedures, six figures, and seven tables and can be found with this article online at doi:10.1016/j.cell.2010.11.050.

Haynes, C., Oldfield, C.J., Ji, F., Klitgord, N., Cusick, M.E., Radivojac, P., Uversky, V.N., Vidal, M., and Iakoucheva, L.M. (2006). Intrinsic disorder is a common feature of hub proteins from four eukaryotic interactomes. PLoS Comput. Biol. 2, e100.

ACKNOWLEDGMENTS

Jahn, T.R., and Radford, S.E. (2008). Folding versus aggregation: Polypeptide conformations on competing pathways. Arch. Biochem. Biophys. 469, 100–117.

We thank C.G. Glabe for providing the A11 antibody and H. Wagner for plasmids HSP70-Luc and NF-kB-Luc. We acknowledge the help of H. Engelhardt with FTIR measurements, technical assistance by O. Mihalache and V. Marcus with electron microscopy, and R. Zenke (MPIB core facility) for confocal microscopy. Financial support from EU Framework 7 Integrated Project PROSPECTS, the Deutsche Forschungsgemeinschaft (SFB 596), the ErnstJung Foundation, and the Ko¨rber Foundation is acknowledged. H.O. and A.W. have a fellowship from the Fonds der Chemischen Industrie, and H.O. is supported by the Elite Graduate Network of Bavaria. Received: April 15, 2010 Revised: September 6, 2010 Accepted: November 11, 2010 Published: January 6, 2011

Jeong, H., Mason, S.P., Barabasi, A.L., and Oltvai, Z.N. (2001). Lethality and centrality in protein networks. Nature 411, 41–42. Kayed, R., Head, E., Thompson, J.L., McIntire, T.M., Milton, S.C., Cotman, C.W., and Glabe, C.G. (2003). Common structure of soluble amyloid oligomers implies common mechanism of pathogenesis. Science 300, 486–489. Keshava Prasad, T.S., Goel, R., Kandasamy, K., Keerthikumar, S., Kumar, S., Mathivanan, S., Telikicherla, D., Raju, R., Shafreen, B., Venugopal, A., et al. (2009). Human Protein Reference Database–2009 update. Nucleic Acids Res. 37, D767–D772. Kim, W., Kim, Y., Min, J., Kim, D.J., Chang, Y.T., and Hecht, M.H. (2006). A high-throughput screen for compounds that inhibit aggregation of the Alzheimer’s peptide. ACS Chem. Biol. 1, 461–469.

Cell 144, 67–78, January 7, 2011 ª2011 Elsevier Inc. 77

Kyte, J., and Doolittle, R.F. (1982). A simple method for displaying the hydropathic character of a protein. J. Mol. Biol. 157, 105–132. Lam, Y.C., Bowman, A.B., Jafar-Nejad, P., Lim, J., Richman, R., Fryer, J.D., Hyun, E.D., Duvick, L.A., Orr, H.T., Botas, J., et al. (2006). ATAXIN-1 interacts with the repressor Capicua in its native complex to cause SCA1 neuropathology. Cell 127, 1335–1347. Lashuel, H.A., and Lansbury, P.T., Jr. (2006). Are amyloid diseases caused by protein aggregates that mimic bacterial pore-forming toxins? Q. Rev. Biophys. 39, 167–201. Lee, E.K., Park, Y.W., Shin, D.Y., Mook-Jung, I., Yoo, Y.J., Lee, E.K., Park, Y.W., Shin, D.Y., Mook-Jung, I., and Yoo, Y.J. (2006). Cytosolic amyloidbeta peptide 42 escaping from degradation induces cell death. Biochem. Biophys. Res. Commun. 344, 471–477. Linding, R., Jensen, L.J., Diella, F., Bork, P., Gibson, T.J., and Russell, R.B. (2003). Protein disorder prediction: implications for structural proteomics. Structure 11, 1453–1459. Morimoto, R.I. (2008). Proteotoxic stress and inducible chaperone networks in neurodegenerative disease and aging. Genes Dev. 22, 1427–1438. Mossuto, M.F., Dhulesia, A., Devlin, G., Frare, E., Kumita, J.R., de Laureto, P.P., Dumoulin, M., Fontana, A., Dobson, C.M., and Salvatella, X. (2010). The Non-Core Regions of Human Lysozyme Amyloid Fibrils Give Rise to Cytotoxicity. J. Mol. Biol. 402, 783–796. Nesterov, E.E., Skoch, J., Hyman, B.T., Klunk, W.E., Bacskai, B.J., and Swager, T.M. (2005). In vivo optical imaging of amyloid aggregates in brain: design of fluorescent markers. Angew. Chem. Int. Ed. 44, 5452–5456. Nilsberth, C., Westlind-Danielsson, A., Eckman, C.B., Condron, M.M., Axelman, K., Forsell, C., Stenh, C., Luthman, J., Teplow, D.B., Younkin, S.G., et al. (2001). The ‘Arctic’ APP mutation (E693G) causes Alzheimer’s disease by enhanced Abeta protofibril formation. Nat. Neurosci. 4, 887–893. Ong, S.E., and Mann, M. (2006). A practical recipe for stable isotope labeling by amino acids in cell culture (SILAC). Nat. Protoc. 1, 2650–2660.

Shearman, M.S. (1999). Toxicity of protein aggregates in PC12 cells: 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide assay. Methods Enzymol. 309, 716–723. Stefani, M., and Dobson, C.M. (2003). Protein aggregation and aggregate toxicity: new insights into protein folding, misfolding diseases and biological evolution. J. Mol. Med. 81, 678–699. Su, A.I., Cooke, M.P., Ching, K.A., Hakak, Y., Walker, J.R., Wiltshire, T., Orth, A.P., Vega, R.G., Sapinoso, L.M., Moqrich, A., et al. (2002). Large-scale analysis of the human and mouse transcriptomes. Proc. Natl. Acad. Sci. USA 99, 4465–4470. Sullivan, E.K., Weirich, C.S., Guyon, J.R., Sif, S., and Kingston, R.E. (2001). Transcriptional activation domains of human heat shock factor 1 recruit human SWI/SNF. Mol. Cell. Biol. 21, 5826–5837. Tartaglia, G.G., Pawar, A.P., Campioni, S., Dobson, C.M., Chiti, F., and Vendruscolo, M. (2008). Prediction of aggregation-prone regions in structured proteins. J. Mol. Biol. 380, 425–436. Tatzelt, J., Zuo, J., Voellmy, R., Scott, M., Hartl, F.U., Prusiner, S.B., and Welch, W.J. (1995). Scrapie prions selectively modify the stress response in neuroblastoma cells. Proc. Natl. Acad. Sci. USA 92, 2944–2948. Wei, Y.N., Liu, T., Sazinsky, S.L., Moffet, D.A., Pelczer, I., and Hecht, M.H. (2003). Stably folded de novo proteins from a designed combinatorial library. Protein Sci. 12, 92–102. West, M.W., Wang, W.X., Patterson, J., Mancias, J.D., Beasley, J.R., and Hecht, M.H. (1999). De novo amyloid proteins from designed combinatorial libraries. Proc. Natl. Acad. Sci. USA 96, 11211–11216. Williams, G.T., McClanahan, T.K., and Morimoto, R.I. (1989). E1a transactivation of the human HSP70 promoter is mediated through the basal transcriptional complex. Mol. Cell. Biol. 9, 2574–2587.

Pentony, M.M., and Jones, D.T. (2009). Modularity of intrinsic disorder in the human proteome. Proteins 78, 212–221.

Winklhofer, K.F., Tatzelt, J., and Haass, C. (2008). The two faces of protein misfolding: gain- and loss-of-function in neurodegenerative diseases. EMBO J. 27, 336–349.

Raychaudhuri, S., Dey, S., Bhattacharyya, N.P., and Mukhopadhyay, D. (2009). The role of intrinsically unstructured proteins in neurodegenerative diseases. PLoS ONE 4, e5566.

Young, J.C., Hoogenraad, N.J., and Hartl, F.U. (2003). Molecular chaperones Hsp90 and Hsp70 deliver preproteins to the mitochondrial import receptor Tom70. Cell 112, 41–50.

78 Cell 144, 67–78, January 7, 2011 ª2011 Elsevier Inc.

The Cell-Non-Autonomous Nature of Electron Transport Chain-Mediated Longevity Jenni Durieux,1 Suzanne Wolff,2 and Andrew Dillin1,* 1The Howard Hughes Medical Institute, The Glenn Center for Aging Research, The Salk Institute for Biological Studies, 10010 North Torrey Pines Road, La Jolla, CA 92037, USA 2The Scripps Research Institute, Department of Molecular and Experimental Medicine, 10550 North Torrey Pines Road, La Jolla, CA 92037, USA *Correspondence: [email protected] DOI 10.1016/j.cell.2010.12.016

SUMMARY

The life span of C. elegans can be increased via reduced function of the mitochondria; however, the extent to which mitochondrial alteration in a single, distinct tissue may influence aging in the whole organism remains unknown. We addressed this question by asking whether manipulations to ETC function can modulate aging in a cell-non-autonomous fashion. We report that the alteration of mitochondrial function in key tissues is essential for establishing and maintaining a prolongevity cue. We find that regulators of mitochondrial stress responses are essential and specific genetic requirements for the electron transport chain (ETC) longevity pathway. Strikingly, we find that mitochondrial perturbation in one tissue is perceived and acted upon by the mitochondrial stress response pathway in a distal tissue. These results suggest that mitochondria may establish and perpetuate the rate of aging for the whole organism independent of cell-autonomous functions. INTRODUCTION An aging organism exhibits correlated and recognizable changes to its physiology over time. These changes occur coordinately across multiple tissues and organs, in concordance with theories that posit a strong role for the participation of the endocrine system in the regulation of age-related phenotypes (Russell and Kahn, 2007; Tatar et al., 2003). Within invertebrate model organisms such as C. elegans and Drosophila, evidence strongly suggests that tissue-specific manipulations of endocrine pathway components affect the aging process of the entire organism. These include alteration of signals from the somatic germline which control the aging of nonmitotic tissues (Arantes-Oliveira et al., 2002; Hsin and Kenyon, 1999); restoration or reduction of insulin/IGF-1 signaling (IIS) in neuronal or fat tissues (Broughton et al., 2005; Hwangbo et al., 2004; Kapahi

et al., 2004; Libina et al., 2003; Wolkow et al., 2000); and genetic manipulations to specific neurons which then alter the capacity for the entire animal to respond to dietary restriction (Bishop and Guarente, 2007). These systems have offered the simplicity of studying tissue-specific expression in organisms in which single-gene mutations can affect longevity, and have been extended to mammalian model systems (Bluher et al., 2003; Conboy et al., 2005; Taguchi et al., 2007). Such evidence indicates that there are key tissues that transmit longevity signals to modulate the aging process. Moreover, these adaptations may have evolved to provide the animal with a mechanism by which an environmental, extrinsic signal could be sensed and then amplified across the entire animal to coordinate the appropriate onset of reproduction, senescence and/or aging. Life span can be increased by reduced function of the mitochondria. Mutation or reduced function in nuclear genes encoding electron transport chain (ETC) components in yeast, C. elegans, Drosophila, and mice delay the aging process (Copeland et al., 2009; Dell’Agnello et al., 2007; Dillin et al., 2002b; Feng et al., 2001; Hansen et al., 2008; Kirchman et al., 1999; Lapointe et al., 2009; Lee et al., 2002; Liu et al., 2005). In C. elegans, mitochondria undergo a period of dramatic proliferation during the L3/L4 stage (Tsang and Lemire, 2002). After this larval molt is completed, mitochondrial DNA proliferation in post-mitotic tissues is minimal (Tsang and Lemire, 2002). Intriguingly, the L3/L4 larval developmental period has proven to be a critical period in which the ETC modulates the aging process (Dillin et al., 2002b). Thus, the sensing and monitoring of key events during the L3/L4 transition by the ETC longevity pathway initiates and maintains the rate of aging of the animal for the rest of its life. Reduction of the ETC longevity pathway during adulthood can not result in increased longevity, even as ATP synthesis becomes impaired (Dillin et al., 2002a; Rea et al., 2007). How the mitochondrial signaling pathway modulates the aging process and the identity of the pathway constituents that transmit these longevity signals remain unknown. One possible suggestion for the observation of increased longevity under conditions of reduced mitochondrial function originates from the ‘‘rate of living’’ theory of aging, in existence for over a hundred years, which suggests that the metabolic expenditures of an organism ultimately determine its life span Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc. 79

(Pearl, 1928; Rubner, 1908). A modification to this theory has suggested alternatively that, because reactive oxygen species (ROS) are generated as a byproduct of the metabolic activity of the mitochondrial ETC during the production of ATP (Harman, 1956), a decrease in ROS production is the major contributing factor to the long-lived phenotypes of ETC mutants (Feng et al., 2001; Rea and Johnson, 2003). Recent evidence (Copeland et al., 2009; Gems and Doonan, 2008; Van Raamsdonk and Hekimi, 2009; Yang et al., 2007) does not support a linear relationship between ROS production and life span. With the increased skepticism toward the oxidative stress theory of aging comes the question: if not by manipulation of ROS in a cellautonomous manner, then by what mechanism does reduction of mitochondrial function affect aging? We attempted to address this question by asking whether manipulations to ETC function could modulate aging in a cellnon-autonomous fashion in the nematode C. elegans. We asked whether key tissues could govern increases in longevity when components of the mitochondrial ETC are inactivated. We also reasoned that if we could identify the crucial tissues from which the ETC longevity pathway functions, we could identify the origin of the longevity signal and perhaps potential mediators of this signal. RESULTS cco-1 Functions in Specific Tissues to Affect the Aging Process To ascertain whether tissue-specific ETC knockdown could alter the life span of an organism, we created transgenic worms carrying an inverted repeat hairpin (HP) directed toward the nuclear-encoded cytochrome c oxidase-1 subunit Vb/ COX4 (cco-1). cco-1 was chosen because knockdown of this gene results in intermediate phenotypes compared to knockdown of the other ETC genes by RNAi, allowing both positive and negative modulation of longevity to be identified (Dillin et al., 2002b; Lee et al., 2002; Rea et al., 2007). Furthermore, cco-1 RNAi does not result in the detrimental phenotypes observed when bacterial feeding RNAi against other components of the ETC is administered undiluted, such as severe developmental delay and lethality (Copeland et al., 2009; Gems and Doonan, 2008; Van Raamsdonk and Hekimi, 2009; Yang et al., 2007). In worms and plants, RNAi can have a systemic effect due to spreading of the dsRNA molecules. For example, exposure of the intestine to bacterially expressed dsRNA results in the dsRNA entering through the intestinal lumen but eliciting knockdown in other cells, such as the muscle and hypodermis (Jose et al., 2009). To remove the systemic nature of RNAi from our experimental design, we used systemic RNAi deficient (sid-1(qt9)) mutant worms (Figure 1A). sid-1 encodes a transmembrane protein predicted to serve as a channel for dsRNA entry. While defective for systemic RNAi, the sid-1(qt9) mutants are fully functional for cell-autonomous RNAi (Winston et al., 2002). Lines were generated in the sid-1(qt9) mutant background using an inverted repeat of the cco-1 cDNA under the control of well-characterized promoters expressed in neurons (unc-119 and rab-3) (Maduro and Pilgrim, 1995; Nonet et al., 1997), intes80 Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc.

tine (ges-1) (Aamodt et al., 1991), and body-wall muscle cells (myo-3) (Miller et al., 1986; Okkema et al., 1993). Knockdown of cco-1 in the intestine using the ges-1 intestinespecific promoter driving a cco-1 hairpin construct significantly increased life span (Figure 1B, representative line of 13, Table S1, available online), whereas the myo-3 muscle-specific promoter driving a cco-1 hairpin in the body-wall muscle either had no effect or even decreased life span (Figure 1C, representative line of 6, Table S1). The rab-3 neuron-specific promoter driving a cco-1 hairpin also increased life span (Figure 1D, representative line of 2, Table S1). Because the life-span extension in the neuronal promoter line was not as great as that observed in intestinal hairpin lines, we tested another neuronal promoter, the pan-neural unc-119 promoter. Consistent with the rab-3 promoter, we observed a moderate increase in life span across multiple unc-119 promoter transgenic lines (Figure 1E, representative line of 8, Table S1). The results of these experiments suggest a primary requirement for ETC knockdown in intestinal and neuronal tissues for increased longevity, albeit the neuronal derived ETC knockdown was consistently less robust compared to the intestinal knockdown. We tested an alternative method for tissue-specific RNAi to verify our hairpin apporach. Tissue-specific RNAi can also be achieved by feeding dsRNA to rde-1 mutant animals in which the wild-type rde-1 gene has been rescued using tissue-specific promoters (Figure 2A) (Qadota et al., 2007). rde-1 encodes an essential component of the RNAi machinery encoding a member of the PIWI/STING/Argonaute family of proteins. Life-span analyses were performed with rde-1(ne219) mutant animals in which rde-1 was restored by tissue-specific expression of wild-type rde-1 cDNA (Qadota et al., 2007). rde-1 was rescued in transgenic lines under the control of the lin-26 hypodermal promoter, the hlh-1 body wall muscle promoter, and the nhx-1 intestine expressing promoter (Qadota et al., 2007). These lines were then tested for their effects on life span when animals were fed cco-1 dsRNA producing bacteria. As expected, feeding rde-1(ne219) mutant animals cco-1 dsRNA producing bacteria did not extend life span, since these animals fail to perform RNAi due to the lack of rde-1 (Figure 2B and Table S1). Consistent with the cco-1 hairpin approach, knockdown of cco-1 in the intestine, by restoring rde-1 using the intestine-specific nhx-1 promoter and feeding cco-1 dsRNA bacteria significantly increased life span. In fact, intestinal cco-1 dsRNA fed to rde-1(ne219);nhx-1p::rde-1 worms was able to completely recapitulate the life-span extension generated by feeding cco-1 dsRNA to wild-type animals (Figure 2C and Table S1). Furthermore, cco-1 knockdown in the body wall muscle decreased life span (Figure 2D and Table S1), similar to results obtained from the muscle-specific cco-1 RNAi hairpin experiments. Hypodermal knockdown of cco-1 had no significant effect on life span (Figure 2E and Table S1). Consistent with cco-1 feeding RNAi increasing longevity in an insulin/IGF-1 pathway independent manner, we found the life-span extension of intestinal cco-1 RNAi animals to be daf-16 independent (Figure 2F and Table S1). Because cco-1 knockdown in two distinct tissues increased longevity, we tested whether combination of the intestinal and nervous system cco-1 knockdown could result in an even further

wildtype

sid-1 mutant

dsRNA

dsRNA

A

no dsRNA import

systemic RNAi knockdown

B

no knockdown

C

1.0

Intestine

0.8

0.7

0.7

0.6 0.5 0.4

0.6 0.5 0.4 0.3

0.2

0.2

ges-1p::cco-1HP control

0.1

myo-3p::cco-1HP control

0.1

0.0

0.0

0

10

20

30

40

Days

0

5

10

15

20

25

E 1.0

Neurons

0.9 0.7

0.7

Surviving

0.8

0.5 0.4

Neurons

0.9

0.8 0.6

30

Days

1.0

Surviving

Muscle

0.9

0.3

D

1.0

0.8

Surviving

Surviving

0.9

0.6 0.5 0.4 0.3

0.3 0.2

0.2

rab-3p::cco-1HP control

0.1

unc-119p::cco-1HP control

0.1 0.0

0.0 0

5

10

15

20

25

30

Days

0

10

20

30

40

Days

Figure 1. Life-Span Analysis of cco-1 Hairpin Transgenic Animals (A) Wild-type worms allow import of dsRNA from surrounding tissues, but sid-1(qt9) mutant worm can not import dsRNA and RNAi knockdown is no longer systemic but is maintained locally within the tissue in which the dsRNA is produced (Winston et al., 2002). (B) Intestine-specific knockdown of cco-1 results in life-span extension. sid-1(qt9)/rol-6 control (black line, mean 18.8 ± 0.7 days), ges-1p::cco-1hairpin (green line, 23.9 ± 0.8 days, p < .0001). (C) Body-wall muscle knockdown of cco-1 does not significantly affect life span. sid-1(qt9)/rol-6 control (black line, mean 18.6 ± 0.5 days), myo-3p::cco-1 hairpin (blue line, mean 16.6 ± 0.5 days, p = 0.0574). (D) Neuronal knockdown of cco-1 extends life span. sid-1(qt9)/rol-6 control (black line, 18.2 ± 0.2 days), rab-3p::cco-1 hairpin (purple line, 21.7 ± 0.5 days, p < .0001). (E) Neuronal knockdown of cco-1 driven by the unc-119 promoter also extends life span. sid-1(qt9)/rol-6 control (black line, mean 19.8 ± 0.7 days), unc119p::cco-1 hairpin (red line, mean 23.8 ± 0.8 days, p = .0001). Please see Table S1 for all statistical analyses and also Figure S1 and Movie S1 for additional experiments.

increase in longevity compared to knockdown in each individual tissue. We crossed our long lived rab-3::cco-1HP (neuronal) lines to the ges-1::cco-1HP (intestinal) lines and tested the life span of the double transgenic animal. Animals with cco-1 knocked down in the nervous system and the intestine did not live longer than knockdown in either the nervous system or the intestine (Figure 2G and Table S1). To test this via a second method, we

used worms in which a gly-19p::sid-1 transgene could specifically restore the capacity for feeding RNAi within the intestine of sid-1 mutant animals. We again found that feeding worms bacteria expressing cco-1 dsRNA did not further extend the life span of our rab-3::cco-1 HP animals (Figure S1A). Therefore, there does not appear to be synergy among the tissues in which cco-1 knockdown is required to extend longevity. This finding Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc. 81

A

B

1.0

rde-1 rescued tissue

0.9

feeding dsRNA

0.7

rde-1(ne219)

Surviving

0.8

RNAi knockdown

rde-1 mutant tissue

0.6 0.5 0.4 0.3 0.2

vector only cco-1 RNAi

0.1

No RNAi initiation

0.0 0

C

1.0

0.8

0.7

0.7

0.6 0.5 0.4 0.3

25

0.6 0.5 0.4 0.3

0.2

vector only cco-1 RNAi

0.1

0.2

0

5

10

vector only cco-1 RNAi

0.1

0.0 15

20

25

30

0.0

35

0

Days

F Hypodermis resuced rde-1

0.9

5

0.7

Surviving

0.7 0.6 0.5 0.4

0.6 0.5 0.4 0.3

0.3

0.1 0.0

0.0 5

vector only daf-16/EV RNAi (50/50) daf-16/cco-1 RNAi (50/50)

0.2

vector only cco-1 RNAi 0

20

Intestine rescued rde-1

0.9 0.8

0.1

15

1.0

0.8

0.2

10

Days

1.0

10

15

20

25

0

5

10

15

20

25

30

Days

Days

G

20

Body wall muscle rescued rde-1

0.9

Surviving

Surviving

15

1.0

0.8

Surviving

10

Days

Intestine rescued rde-1

0.9

E

5

D

1.0 0.9

Surviving

0.8 0.7 0.6 0.5 0.4

rab-3HP ges-1HPxrab-3HP ges-1HP sid-1

0.3 0.2 0.1 0.0

0

10

20

30

40

Figure 2. Life-Span Analysis of Tissue-Specific Complementation of rde-1 with cco-1 Feeding RNAi (A) Tissues exposed to dsRNA from feeding RNAi initiate knockdown if rde-1 has been rescued in the corresponding tissue. Neighboring tissues are unable to initiate RNAi if rde-1 is absent. (B) rde-1(ne219) mutants do not respond to cco-1 feeding RNAi. Animals fed bacteria harboring an empty vector (black line, mean 18.0 ± 0.3 days), cco-1 RNAi (red line, mean 18.16 ± 0.4 days, p < 0.4043). (C) rde-1 rescued in the intestine (VP303) extends life span when fed cco-1 dsRNA producing bacteria. Animals fed vector only bacteria (black line, mean 14.7 ± 0.6 days), cco-1 RNAi (green line, mean 22.0 ± 0.2 days, p < .0001). (D) rde-1 rescued in the body wall muscle (NR350) decreases life span when fed cco-1 dsRNA bacteria. Animals fed bacteria harboring empty vector (black line, mean 13.5 ± 0.3 days), cco-1 RNAi (blue line, mean 11.8 ± 0.3 days, p < 0.0002.

82 Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc.

suggests that the nervous system and the intestinal cells communicate with each other to modulate aging in response to reduced cco-1 function.

developmental and behavioral deficits of ETC RNAi, contributions from specific tissues must also play an important role in these life history traits.

Tissue-specific ETC Knockdown Uncouples Multiple Correlates of Longevity Resistance to oxidative stress, UV damage, and heat stress is associated with multiple forms of increased longevity. We tested whether the increased longevity of the tissue-specific cco-1 RNAi animals was due to resistance to any of these stresses. Tissue-specific knockdown of cco-1 did not affect the response of animals to oxidative stress induced by paraquat in a manner correlated with their longevity phenotype (Table S2), consistent with recent results in worms and flies (Copeland et al., 2009; Doonan et al., 2008; Van Raamsdonk and Hekimi, 2009) (Lee et al., 2002). We next tested whether resistance to UV damage correlated with increased longevity. Again, none of the long-lived tissue-specific hairpin lines were more resistant to UV damage than wild-type animals (Table S3). Finally, we tested whether the long-lived tissue-specific cco-1 hairpin lines were more resistant to heat stress than control animals and found that they were not (Table S4). Collectively, the increased longevity of tissue specific cco-1 hairpin animals did not correlate with the known stress resistance phenotypes associated with other pathways that modulate the aging process (Arantes-Oliveira et al., 2002; Larsen et al., 1995; Lee et al., 1999; Lee et al., 2003; McElwee et al., 2004). RNAi of cco-1 slows development, growth, movement and reduces fecundity (Dillin et al., 2002b). Through an RNAi dilution approach, many of these side effects could be uncoupled from longevity, an observation that suggested a quantitative model for ETC function upon these life history traits (Rea et al., 2007). We tested whether a qualitative difference among the mitochondrial ETC from different tissues could also explain these observed side effects. We found that many of these traits could be uncoupled from increased longevity conferred by simply reducing mitochondrial function in a particular tissue. For example, long lived animals in which cco-1 was reduced in neuronal cells produced worms of nearly identical length to their control counterparts (Figure S1B), reached adulthood at the similar rates (data not shown) and had similar numbers of progeny (Figure S1C). These results are consistent with ETC reduction in all tissues of Drosophila increased life span and decreased fertility, while knockdown in neurons increased longevity without affecting fertility (Copeland et al., 2009). Additionally, reduction of cco-1 in the intestine or nervous system did not result in slowed movement; however, reduction in the body wall muscles did (Movie S1). Therefore, in addition to the quantitative model proposed by Rea et al. to explain the

The Mitochondrial Unfolded Protein Response Is Required for ETC-Mediated Longevity In response to a mitochondrial perturbation there exists a stress response mechanism that is communicated to the nucleus to increase the expression of mitochondrial associated protein chaperones, such as HSP-6 and HSP-60, referred to as the mitochondria-specific unfolded protein response (UPRmt) (Benedetti et al., 2006; Yoneda et al., 2004; Zhao et al., 2002). hsp-6 is the mitochondrial hsp70 heat shock protein family member and hsp-60 is the mitochondrial GroE/hsp60/hsp10 chaperonin family member. The UPRmt is activated upon different forms of mitochondrial stress including the misfolding of mitochondria-specific proteins or stoichiometric abnormalities of large multimeric complexes, such as ETC complexes (Yoneda et al., 2004). Disrupting subunits of ETC complexes by either RNAi or mutation activates the mitochondrial stress response (Benedetti et al., 2006; Yoneda et al., 2004). cco-1 RNAi is a potent inducer of hsp-6 and hsp-60 (Yoneda et al., 2004). Intrigued by this discovery, we tested whether the UPRmt might play a central and specific role in the increased longevity generated by ETC RNAi. We tested whether other well-known pathways that regulate the aging process also induced the UPRmt. Unlike cco-1 RNAitreated animals (Figure 3A), animals treated with RNAi toward daf-2, the IIS receptor, or eat-2 mutant animals, a genetic surrogate for diet restriction induced longevity, did not induce the UPRmt (Figures 3B and 3C) even though each of these interventions increase longevity. Therefore, induction of the UPRmt appears specific to the ETC longevity pathway and not other longevity pathways. In addition to the UPRmt, the unfolded protein response in the endoplasmic reticulum (UPRER) is also induced under conditions of protein misfolding, although confined to the ER (Ron and Walter, 2007). We tested whether mitochondrial reduction resulted in a general upregulation of all protein misfolding pathways by treating hsp-4p::GFP reporter worm strains (Calfon et al., 2002) with cco-1 RNAi. HSP-4 is the worm ortholog of the ER chaperone, BiP, which is transcriptionally induced by the UPRER. Unlike the UPRmt, cco-1 RNAi did not induce expression of the UPRER (Figure 3D), although ER stress induced by tunicamycin did. Furthermore, cco-1 RNAi did not inhibit the ability of cells to induce the UPRER by treatment with tunicamycin. We also tested if cco-1 RNAi induced a marker of cytosolic protein misfolding by treating animals containing the hsp-16.2p::GFP reporter strain (Link et al., 1999) with cco-1 RNAi. HSP-16.2 is

(E) rde-1 rescued in the hypodermis (NR222) has no effect on life span when fed cco-1 dsRNA producing bacteria. Vector only (black line, mean 13.5 ± 0.3 days), cco-1 RNAi (purple line, mean 14.3 ± 0.4 days, p = 0.148). (F) Life-span extension by cco-1 feeding RNAi in the intestine of ges-1:;rde-1 rescued animals is independent of daf-16. Intestinal rde-1 rescued worms were fed empty vector (black line), mean 16.4 ± 0.6 days, daf-16 RNAi diluted 50% with empty vector (gold line, mean 16.3 ± 0.4 days) or daf-16 diluted 50% with cco-1 RNAi (green19.9 ± 0.5 days, p < .0001) .Please see Table S1 for all statistical analysis. (G) Double transgenic animals carrying rab-3:;cco-1HP and ges-1:cco-1HP (blue line, mean life span 23.4 ± 0.6 days) did not live longer than either the rab-3::cco1HP (red line, mean life span 23.2 ± 0.6 days, p = .64) or the the ges1::cco-1HP (green line, mean life span 22.9 ± 0.6 days, p = .75) animals. sid-1 control (black line, mean life span 19.2 ± 0.5 days).

Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc. 83

hsp-6p::GFP (UPRmt)

hsp-6p::GFP

400 350

GFP relative fluoresence Day 1

A

i B

cco-1 RNAi

EV

ii

hsp-6p::GFP (UPRmt)

cco-1 RNAi

EV

300 250 200 150 100

C 50

iii

0 hsp-6p::GFP on EV

hsp-6p::GFP on cco-1 RNAi

hsp-6p::GFP GFP relative fluoresence Day 1

400

350

300

250

200

150

100

i

ii

daf-2 RNAi

hsp-6p::GFP (UPRmt)

0

EV

hsp-6p::GFP

eat-2(ad1116);hsp-6p:GFP

hsp-6p::GFP EV

hsp-6p::GFP on daf-2 RNAi

hsp-6p::GFP on EV

eat-2;hsp-6p::GFP on EV

iii

daf-2 RNAi

hsp-6p::GFP on cco-1 RNAi

eat-2(ad1116);hsp-6p:GFP GFP relative fluoresence Day 1

C

EV

50

450 400 350 300 250 200 150 100 50

i

overlay

ER D hsp-4p::GFP (UPR )

0

ii

GFP

hsp-4p::GFP

EV

cco-1 RNAi

ii

i hsp-16.2p::GFP (HSR) 1

E

EV

EV

cco-1 RNAi

EV

cco-1 RNAi

ii

84 Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc.

cco-1 RNAi

GFP relative fluoresence Day 1

80 70 60 50 40 30 20 10 0 hsp-4p::GFP on EV

hsp-4p::GFP on cco-1RNAi

hsp16.2 on EV

hsp16.2 on cco-1

hsp-4p::GFP on tunicamycin

120

heat shocked

EV

90

iii

hsp-16.2p::GFP

heat shocked

i

EV

cco-1 RNAi

EV

cco-1 RNAi GFP

GFP relative fluoresence Day 1

EV

cco-1 RNAi

eat-2;hsp-6p::GFP on cco-1 RNAi

100

witn tunicamycin

witn tunicamycin

cco-1 RNAi

iii

100

80

60

40

20

0

iii

RNAi

hsp16.2 with heat shock

a small heat shock protein of the hsp20/alpha-B crystallin family and is under transcriptional control of the heat shock response (HSR) regulated by HSF-1. Much like the UPRER, cco-1 RNAi was unable to induce this reporter associated with cytosolic misfolding (Figure 3E). As positive controls, heat shock could induce the HSR reporter and cco-1 RNAi did not block this response. Thus, it appears that knockdown of cco-1 specifically induces the UPRmt, and not other protein misfolding pathways. The UPRmt Is a Potent Transducer of the ETC-Longevity Pathway We tested whether the UPRmt is a key component of the ETC longevity pathway since there appeared to be a positive and specific correlation of induction of the UPRmt and ETC mediated longevity. If the UPRmt is indeed a regulator of the ETC longevity pathway, we predicted that loss of the UPRmt would specifically suppress the extended longevity of ETC reduced animals and not other longevity pathways. The UPRmt consists of a signaling cascade that results in upregulation of nuclear-encoded genes to alleviate the stress sensed in the mitochondria. Perception of misfolding in the mitochondria requires the nuclear localized ubiquitin-like protein UBL-5, which acts as an essential and specific coactivator of the homeodomain transcription factor, DVE-1. Together, UBL-5 and DVE-1 respond to mitochondrial perturbation to increase expression of mitochondrial chaperones, including hsp-6 and hsp-60 (Benedetti et al., 2006). ClpP is the homolog of the E.coli ClpP protease located in the mitochondria that plays a role in generating the mitochondrial derived signal to activate DVE-1/UBL-5 stress responsive genes (Haynes et al., 2007). We treated long-lived ETC mutant animals with RNAi directed toward the known pathway components of the UPRmt and tested the resulting life span. Because cco-1 RNAi is extremely sensitive to dilution and is not efficiently knocked down in combination with a second RNAi (Figures S2A and S2B), we first chose to examine the requirement for UPRmt genes on several long-lived mitochondrial mutants. RNAi of ubl-5, the dve-1 transcriptional co-regulator, specifically blocked the extended life span of the mitochondrial mutants, isp-1 (qm150) and clk-1(e2519) (Figure 4A, Figure S2C, and Table S1) compared to the life span of wild-type animals. RNAi of ubl-5 did not suppress the extended life span of long-lived daf-2 or eat-2 mutant animals (Figures 4B and 4C; Table S1). Furthermore, ubl-5 RNAi did not shorten the life span of wild-type animals (Figure 4D and Table S1). Taken together, ubl-5 appears essential and specific for the extended longevity of mitochondrial mutants.

RNAi of dve-1 suppressed the life span of all long-lived animals and shortened the life span of wild-type animals (Figures S2D–S2G). This result is not surprising given the roles of dve-1 in growth and development and the embryonic lethality observed for homozygous dve-1 mutant animals (Burglin and Cassata, 2002; Haynes et al., 2007). Furthermore, RNAi of hsp-6, hsp-60 or clpp-1 suppressed longevity in the same manner as dve-1, suggesting that these RNAi treatments were pleiotropic and simply made the animals sick (data not shown). Thus, our results indicate that ubl-5 is specific for the longevity response, possibly by specifying the transcriptional activity of DVE-1, to mitochondrial ETC mediated longevity and dve-1, hsp-6, hsp-60 and clpp-1 have more broad roles in development making their specific roles in mitochondrial ETC mediated longevity difficult to discern at this time. The Temporal Requirements of the UPRmt and ETC-Mediated Longevity Overlap The life-span extension by ETC RNAi has a distinct temporal requirement during the L3/L4 stages of larval development (Dillin et al., 2002b; Rea et al., 2007). Furthermore, markers of the UPRmt, namely hsp-6p::GFP, have their greatest activation late in larval development at the L4 stage when challenged with mitochondrial stress (Yoneda et al., 2004). We verified these findings by following the activation of the UPRmt of animals treated with cco-1 RNAi and tested whether the timing requirement of cco-1 mediated longevity could be uncoupled from the induction of the UPRmt. Worms carrying the hsp-6p::GFP UPRmt reporter were transferred onto bacteria expressing cco-1 dsRNA at every developmental stage from embryo to day 2 of adulthood (Figures 5A–5C). Worms transferred to cco-1 dsRNA at the L1 stage induced hsp-6p::GFP throughout development, and this signal perpetuated itself into adulthood (Figure S3). Worms could induce the UPRmt if transferred to the cco-1 RNAi treatment before the L4 larval stage (Figures 5B and 5C). After the L4 larval stage, worms transferred to bacteria expressing cco-1 dsRNA were unable to induce the hsp-6p::GFP marker (Figures 5B and 5C) and were not long lived (Dillin et al., 2002b; Rea et al., 2007). Thus, inactivation of cco-1 must be instituted before the L3/L4 larval stage to initiate induction of the UPRmt. Inactivation in adulthood does not induce the UPRmt and does not result in increased longevity. Inactivation of ETC components during larval development is sufficient to confer increased longevity on adult animals even though the knocked-down ETC component can be restored in adulthood (Dillin et al., 2002b; Rea et al., 2007). We tested

Figure 3. Induction of the UPRmt Is Specific to the ETC Longevity Pathway (A) hsp-6p::GFP reporter worms fed empty vector (EV) containing bacteria have low levels of background GFP (i) overlay; (ii) GFP. hsp-6p::GFP reporter worms fed cco-1 RNAi upregulate the UPRmt. Relative fluorescence was quantified using a fluorescence plate reader (iii). (B) daf-2 RNAi does not induce hsp-6p::GFP (i) overlay; (ii) GFP. hsp-6p::GFP reporter worms were hatched on empty vector, cco-1, or daf-2 dsRNA expressing bacteria and allowed to grow to day 1 of adult hood. Relative fluorescence was quantified (iii). (C) Dietary restricted eat-2(ad1116) mutant worms do not upregulate hsp-6p::GFP reporter (i) overlay; (ii) GFP. Relative fluorescence was quantified (iii). (D) The UPRER is not induced by cco-1 RNAi, (i) overlay; (ii) GFP. hsp-4p::GFP transgenic reporter worms were fed empty vector containing bacteria or cco-1 dsRNA bacteria. No fluorescence upregulation was detected (iii). Both EV and to a lesser extent cco-1 RNAi fed worms were able to upregulate the UPRER upon treatment with tunicamycin, (i and ii) which is known induce UPRER. Relative fluorescence of was quantified (iii). (E) cco-1 RNAi does not induce a marker of cytosolic protein misfolding stress, (i) overlay; (ii) GFP. hsp16.2p::GFP reporter worms were fed EV or cco-1 dsRNA bacteria. No fluorescence upregulation was detected (iii). As positive controls, heat shock for 6 hr at 31 C could induce the heat shock response (HSR) and cco-1 RNAi did not block this response (i and ii). In all panels, error bars indicate standard deviations (SD).

Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc. 85

A

B 1.0

1.0

isp-1(qm150)

0.9

0.8

0.7

0.7

Surviving

Surviving

0.8 0.6 0.5 0.4 0.3

0.6 0.5 0.4 0.3

0.2

vector only ubl-5 RNAi

0.1

N2 wildtype

0.1

0.2

0.0

vector only ubl-5 RNAi

0.0 0

5

10

15

20

25

30

35

40

45

0

Days

10

20

30

40

50

Days

D

C

1.0

1.0

eat-2(ad1116)

0.9

0.8

0.7

0.7

0.6 0.5 0.4 0.3

vector only ubl-5 RNAi N2 wildtype

0.2 0.1

10

0.6 0.5 0.4 0.3 0.2

ubl-5 RNAi vector only

0.1 0.0

0.0 0

N2 wildtype

0.9

0.8

Surviving

Surviving

daf-2 (e1370)

0.9

20

30

40

Days

0

5

10

15

20

25

30

Days

Figure 4. ubl-5 Is Necessary and Specific for ETC-mediated Longevity (A) The long life span of isp-1(qm150) mutant animals is dependent upon ubl-5. isp-1(qm150) (empty vector, black line, mean 25.8 ± 1.0 days), isp-1(qm150) fed ubl-5 dsRNA bacteria (orange line, mean 15.5 ± 0.7 days, p < .0001), N2 wild-type (gray line, mean 19 ± 0.5 days). (B) daf-2(e1370) mutant life span is unaffected by ubl-5 knockdown. daf-2(e1370) mutant animals grown on empty vector bacteria (black line, mean 40.1 ± 1.2 days), daf-2(e1370) fed ubl-5 dsRNA bacteria (orange line, mean 39.9 ± 1.2 days, p = .327). (C) Dietary restricted eat-2(ad1116) mutant life span is not dependent upon ubl-5. N2 on empty vector (gray line, mean life span 18.2 ± 0.4 days); eat-2(ad1116) on empty vector (black line, mean 26.4 ± 0.6 day)s; eat-2(ad1116) fed ubl-5 dsRNA bacteria (orange line, mean 23.3 ± 0.7 days, p < 0.0004). (D) N2 wild-type life span is unaffected by ubl-5 knockdown. N2 grown on empty vector bacteria (black line, mean 18.2 ± 0.4 days), N2 fed ubl-5 dsRNA bacteria (orange line, mean 20.3 ± 0.4 days, p = 0.0834). All statistical data can be found in Table S1. See also Figure S2 for additional experiments.

whether developmental inactivation of cco-1 could not only induce, but whether it could also maintain activation of the UPRmt during adulthood, even though adult inactivation of cco-1 was unable to induce the UPRmt. Worms treated with cco-1 RNAi during larval development and then moved to dicer (dcr-1) RNAi (a key component of the RNAi machinery) to block further RNAi activity on day 1 of adulthood have an extended life span (Dillin et al., 2002b). Similarly, hsp-6p::GFP worms treated with cco-1 RNAi during larval development and moved onto dcr-1 RNAi maintained the induced response of the UPRmt (Figures 5D and 5E). Therefore, inactivation during larval development of cco-1 is sufficient to initiate and maintain a signal to increase longevity and induce the UPRmt in adult animals. The results of these experiments match the timing requirements of the life-span extension for ETC RNAi-treated worms and support the idea that the signals for increased longevity and induction/ maintenance of the UPRmt are not separable. The UPRmt Responds to Cell-Non-Autonomous Cues from ETC Knockdown Intrigued by the tissue-specific nature by which cco-1 depletion can modulate the aging process of the entire animal, the specific 86 Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc.

role of the UPRmt in the longevity response in ETC mutant animals and the overlapping timing requirements for both ETC RNAi and induction of the UPRmt, we hypothesized that the induction of the UPRmt may be able to act cell-non-autonomously in a multicellular organism. If so, we reasoned that induction of the UPRmt in one tissue by cco-1 reduction might lead to the UPRmt being upregulated in a distal tissue that has not experienced cco-1 reduction (Figure 6A). Consistent with this hypothesis, transgenic worms with the cco-1 hairpin expressed in all neurons (either the rab-3 or the unc-119 promoter driven cco-1 hairpin) were able to induce hsp-6p::GFP expression in the intestine (Figure 6B). In fact, neuronal RNAi of cco-1 induced the UPRmt reporter to the same extent as animals with intestinal cco-1 RNAi (Figures 6B and 6C). We were unable to ascertain whether mitochondrial ETC knockdown in the intestine could signal to the nervous system to induce the UPRmt due to the low expression of the hsp-6p::GFP reporter in neuronal cells. Because ubl-5 RNAi could block the long life span of mitochondrial mutants and ubl-5 is required for induction of the UPRmt, we tested whether reduction of ubl-5 in the intestinal cells could block the induction of the cell-non-autonomous signal for the UPRmt that originated from the nervous system.

B

A

hsp-6p::GFP (UPRmt)

Stage of transfer to cco-1 RNAi L1

L2

green

L3

green green

L4

Young Adult

not green

not green

Stage of tranfer to cco-1 RNAi:

C Developmental transfer to cco-1 RNAi induces hsp-6p::GFP

Control on EV

Hatch

L2

L3

L4

YA

600 500 400 300 200 100 0 Control

Hatch

L2

L3

L4

YA

Stage at which transferred to RNAi

D

Transfer to dcr-1 RNAi L1

L2

L3

L4

fed cco-1 RNAi reduced CCO-1

green

green green

Young Adult

E

hsp-6p::GFP (UPRmt)

Day2

dcr-1 RNAi CCO-1 restored

green

still green

still green

cco-1 RNAi cco-1 RNAi moved to dicer RNAi at L4

Figure 5. The Temporal Activation of ETC-generated Longevity Signal Is Coincident with Induction of the UPRmt (A) hsp-6p::GFP reporter worms were transferred to cco-1 RNAi at each larval developmental stage and early adulthood. GFP fluorescent measurements were taken 16 hr after reaching young adulthood in all cases. (B) hsp-6p::GFP is upregulated if transfer occurs before the L4 stage of development. (C) Quantification of hsp-6p::GFP in (A); error bars represent standard deviation (SD). (D) cco-1 knockdown during larval development is sufficient to induce the hsp-6p::GFP reporter in adulthood. hsp-6p::GFP reporter worms we grown on cco-1 dsRNA bacteria during development and then moved to dcr-1 dsRNA producing bacteria at the L4 larval stage, to disrupt the RNAi machinery allowing CCO-1 levels to return to normal. UPRmt remains induced. (E) hsp-6p::GFP fluorescence 48 hr after transfer to dcr-1 RNAi as described by schematic D. See also Figure S3.

We created lines expressing the rab-3p::cco-1HP in conjunction with a gly-19p::ubl-5HP containing the hsp-6p::GFP reporter. Surprisingly, we found that ubl-5 reduction in the intestinal cells could not block the induction of the UPRmt from signals generated in the nervous system (Figure 6D). Furthermore, intestine-specific depletion of ubl-5 was not sufficient to block the long life span of animals with reduced cco-1 expression in the

nervous system (Figure 6E). However, intestine-specific knockdown of ubl-5 did block induction of the UPRmt in the intestinal cells of animals fed bacteria expressing dsRNA of cco-1 (Figure 6F and Figure S4A). Neuronal cells are not able to induce a RNAi response to feeding dsRNA, indicating that ubl-5 is required for induction of the UPRmt in response to cco-1 reduction in nonneuronal cells, but other, yet to be defined factors, Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc. 87

A Cell autonomous x

Cell non-autonomous

x

x

x

x

x

x

hsp-6p::GFP ges-1p::cco-1HP

hsp-6p::GFP rab-3p::cco-1HP

B

i

unc-119HP x hsp-6p::GFP

iv

ges-1HP x hsp-6p::GFP

D

rab-3p::cco-1HP gly-19p::ubl-5-KD x hsp-6p::GFP

i Empty vector

E

iii

rab-3HP x hsp-6p::GFP

GFP relative fluorescence

C

ii

hsp-6p::GFP control

cco-1 RNAi

rab-3p::cco-1HP x hsp-6p::GFP

ii

ges-1HP rab-3 HP x x i hsp-6p:GFP hsp-6p:GFP

F

1.0

feeding cco-1 RNAi

0.9 0.8

Surviving

0.7 0.6 0.5 0.4 sid-1(qt9) EV sid-1(qt9); gly-19p::ubl-5-KD EV - 5 KD sid-1(qt9); rab-3p::cco-1-HP; gly-19p::ublsid-1(qt9); rab-3p::cco-1-HP

0.3 0.2 0.1 0.0 0

5

10

15

20

25

30

Days

feeding cco-1 RNAi gly-19p::ubl-5-KD hsp-6p::GFP

feeding cco-1 RNAi hsp-6p::GFP

Figure 6. Cell-Non-Autonomous Upregulation of the UPRmt (A) Representation of cell-autonomous and non-autonomous upregulation of UPRmt. ‘‘X’s’’ depict tissue where cco-1 is knocked down (intestine or neurons). Green indicates location of upregulation of hsp-6p::GFP reporter (intestine upon knockdown in intestine or neurons). (B) hsp-6p::GFP reporter worms were crossed to tissue-specific cco-1 hairpin lines. Control hsp-6p::GFP shows only background GFP (i). Neuron-specific cco-1 hairpin results in upregulation of hsp-6p::GFP in the intestine (rab-3 (ii) and unc-119 (iii) lines shown). Intestine-specific ges-1p::cco-1 hairpin (iv) also results in upregulation of the hsp-6p::GFP reporter in the intestine. (C) Fluorescent quantification of B; error bars represent standard deviation (SD). (D) Intestinal knockdown of ubl-5 does not block UPRmt induction caused by neuronal cco-1 reduction. Worm strains were created with rab-3::cco-1HP; hsp-6p::gfp with gly-19p::ubl-5KD. (E) Intestinal reduction of ubl-5 does not block the life-span extensions of rab-3::cco-1HP animals. sid-1(qt9) (blue line, mean = 19.4 ± 0.6 days); sid-1(qt9); gly19p::ubl-5-KD (red line, mean = 20.1 ± 0.7 ± 0.7 days); rab-3p::cco-1-HP; gly-19p::ubl-5-KD (yellow line, mean = 24.1 ± 0.7 days); rab-3p::cco-1HP (green line, mean = 24.5 ± 0.7 days); N2 on cco-1 RNAi (mean = 27.3 ± 0.6 days. p > 0.66 (green versus yellow)). (F) Intestinal reduction of ubl-5 (gly-19p::ubl-5HP) blocks UPRmt induction of animals fed bacteria expressing cco-1 dsRNA. See also Figure S4.

88 Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc.

are required to respond to the cell-non-autonomous signals from the nervous system to induce the UPRmt.

This work identified key tissues, genes essential and specific for mitochondrial longevity and at least one mechanism necessary for increased longevity in response to altered mitochondrial function in a metazoan. The UPRmt was essential for the extended longevity of ETC mutant animals and has been previously reported to be upregulated in response to RNAi of cco-1 (Yoneda et al., 2004). Consistent with the temporal requirements of the ETC to modulate longevity during the L3/ L4 larval stages, the UPRmt could only be induced when cco-1 RNAi was administered before the L3/L4 larval stage, but not in adulthood. Therefore, induction of the UPRmt mirrored the temporal requirements of the ETC to promote longevity when reduced. More importantly, the fact that induction of the UPRmt can be maintained long into adulthood, well after the mitochondrial insult had been given in larval development, indicates that the animal might possess an epigenetic mechanism to ensure increased resistance to future mitochondrial perturbations. One of the most surprising findings is the UPRmt can be activated in a cell-non-autonomous manner. Because the hsp-6p::GFP reporter is primarily limited to expression in the intestine, we were well poised to ask if perturbation of cco-1 in the nervous system could induce the UPRmt in the intestine. Therefore, neuronal limited knockdown of cco-1 could profoundly induce the hsp-6 reporter indicates that a cue from the nervous system must travel to the intestine to induce the UPRmt (Figure 7). It is not clear whether the factor is proteinacious, nucleic acid based or a small molecule, but it is clear that its production in a limited number of cells can profoundly influence the survival of the entire organism. Because this signal is the product of perceived mitochondrial stress that results in increased survival, we have termed this cell-non-autonomous signal a ‘‘mitokine.’’ While many of these perturbations have pleiotropic effects that result in their short life span, their ability to upregulate the UPRmt is not sufficient to overcome these potentially harmful side effects. We found that muscle-specific cco-1 RNAi could also induce the intestinal hsp-6p::GFP reporter, yet these animals were not long lived (Figure S4B). We also find that short-lived mev-1 mutant animals also induce the UPRmt (data not shown). Finally, many of the nuclear-encoded mitochondrial genes discovered to induce the UPRmt when inactivated using RNAi (Yoneda et al., 2004) are not long lived (data not shown). Therefore, ectopic induction of the UPRmt is required but is not sufficient in the establishment of the prolongevity cue from mitochondria in these settings. Of the currently identified UPRmt pathway members, the ubiquitin like protein, UBL-5, which provides transcriptional specificity for the homeobox transcription factor DVE-1 in response to unfolded proteins in the mitochondria, is essential for the increased longevity of ETC mutant animals. Knockdown of ubl-5 specifically in the intestine was not sufficient to block mitokine signaling from the nervous system, but was able to block induction of animals fed dsRNA of cco-1, which can not reduce cco-1 in the nervous system. Therefore, it appears that ubl-5

I

II

IV

III mitochondria

sending cells

Mitokine

receiving cells

UP R

mt

te d

DISCUSSION

Electron Transport Chain

a upregul

Figure 7. Model for the Cell-Non-Autonomous Nature of the UPRmt Cells experiencing mitochondrial stress, in this scenario neuronal cells (circles) marked within the yellow box, produce a signal that is transmitted from the mitochondria to the nucleus to regulate the expression of genes regulated by UBL-5 and possibly DVE-1. These cells serve as sending cells and produce an extracellular signal (mitokine) that can be transmitted to distal, receiving cells, in this case intestinal cells marked in the green box. Receiving cells perceive the mitokine and induce the mitochondrial stress response. See also Figure S5.

either functions exclusively in a cell-autonomous fashion, or the signals generated to induce the UPRmt in the nervous system are very different from the signal generated in other tissues as ubl-5 was not required for neuronal induction. It is intriguing to speculate why reduced mitochondrial ETC in only a few tissues are able to send a prolongevity cue, or mitokine, but others do not. Because the intestine and the sensory neurons (amphids and phasmids) are the only cells that are in direct contact with the worm’s environment (the hypodermis/ skin is wrapped in a protective, dense cuticle), perhaps these cells are fine-tuned to perceive mitochondrial insults that might be present in the environment. Alternatively, mitochondrial metabolism in the nervous system and intestine might have different requirements than other tissues making disruptions in these tissues more susceptible to perturbation and subsequent UPRmt upregulation. As another possibility, there is a growing body of research emphasizing the importance of ROS, not as damaging agents, but as crucial components of cell signaling. It remains a possibility that ROS may act as signaling molecules and potentially serve as the mitokine or intermediary to elicit a nuclear response. However, animals treated with high doses of the antioxidants N-acetyl-L-cysteine (NAC) or ascorbic acid (vitamin C) were not able to block mitokine signaling (Figure S5), a thorough investigation of mitochondrial function from each tissue will be essential to test these hypotheses. In the future it will be important to understand how mitochondrial stress initiates the UPRmt in a cell-autonomous fashion and how this stress is then transmitted throughout the organism to Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc. 89

induce the UPRmt in cells that have yet to possess mitochondrial stress (i.e., a cell-non-autonomous fashion). Furthermore, the identity and mode of action of the mitokine will provide an avenue to explore treatment of mitochondrial diseases in a tissue and cell-type-specific manner if conserved from worm to man. EXPERIMENTAL PROCEDURES Strains HC114 (sid-1(qt9)), MQ887 (isp-1(qm150)), CB4876 (clk-1(e2519)), CF1041 (daf-2(e1370)), TK22 (mev-1(kn1)III), WB27 (rde-1(ne219)), NR222 (rde-1 (ne219)V; kzIs9, NR350 (rde-1(ne219) V; kzIs20), SJ4100 (zcIs13[hsp6p::GFP]), SJ4058 (zcIs9[hsp-60p::GFP]), CL2070 (dvIs[hsp-16.2::GFP]) and N2 wild-type were obtained from the Caenorhabditis Genetics Center. VP303 was a generous gift from the Strange lab. The myo-3 promoter hairpin RNAi transgene was created by inserting PCR amplified cco-1 cDNA with no stop codon into pPD97.86 (Addgene). The reverse complement cco-1 cDNA was inserted into pGEX2T after the GST linker to be used as the hairpin loop as described. PCR amplifications were used to add an AgeI site to the 30 end of the cco-1 cDNA and NgoMIV (compatible and nonrecleavable with AgeI) to the 50 end of the GST linker. Ligation of the PCR products in the presence of AgeI enzyme and NgoMIV were followed by gel extraction of the promoter hairpin fragment as described (Hobert 2002). The ges-1 and unc-119 promoters were PCR amplified from genomic DNA and cloned in place of the myo-3 promoter driving cco-1. The rab-3 promoter was a gift from Kang Shen, Stanford University, and sequence verified. Transgenic tissue-specific RNAi hairpin expressing strains were generated by microinjecting gel extracted hairpin RNAi constructs (40-60ng/ml) mixed with an equal concentration of pRF4(rol-6) co-injection marker or myo-2::GFP into sid-1(qt9) worms. Control lines were generated by injecting sid-1(qt9) with 50ng/ml pRF4(rol-6). Extrachromosomal arrays were integrated and backcrossed five times as described. The gly-19 intestinal promoter driving wild-type sid-1 was injected into rab-3p::cco-1HP transgenic worms to enable knockdown of cco-1 in the neurons by the hairpin transgene and in the intestine by feeding RNAi. ubl-5 knockdown strains were generated with constructs as described (Esposito et al., 2007; Hobert, 2002). Forward and reverse orientation ubl-5 constructs were driven by the gly-19 intestinal promoter and co-injected with the myo-3p::tdTomato marker. Life-Span Analyses Life-span analyses were performed as described previously (Dillin et al., 2002a). 80-100 animals were used per condition and scored every day or every other day. All life-span analyses were conducted at 20 C and repeated at least twice. JMP IN 8 software was used for statistical analysis. In all cases, P-values were calculated using the log-rank (Mantel–Cox) method. GFP Expression and Quantification SJ4100 hsp-6::GFP were bleached to collect synchronous eggs and grown on cco-1 RNAi. At each stage from larval stage 1 to Day 1 of adulthood, worms were assayed for GFP expression. Alternatively, SJ4100 worms were grown on empty vector and transferred to cco-1 RNAi at each developmental stage at which time GFP was assayed at Day 1 or 2 of adulthood. Integrated hairpin RNAi worm lines were crossed to SJ4100 hsp-6p::GFP reporter lines. GFP was monitored in Day 1 adults. Fluorimetry assays were performed using a Tecan fluorescence plate reader. 100 roller worms were picked at random (25 into 4 wells of a black walled 96-well plate) and each well was read three times and averaged. Each experiment was repeated three times. SUPPLEMENTAL INFORMATION Supplemental Information includes Extended Experimental Procedures, five figures, four tables, and one movie and can be found with this table online at doi:10.1016/j.cell.2010.12.016.

90 Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc.

ACKNOWLEDGMENTS We thank Drs. W. Mair, S. Panowski, M. Raices, P. Douglas, and N. Baird for thoughtful editing and scientific insight. We thank Z. Liu for cloning and strain integration. We are grateful to Drs. C. Hunter, C. Haynes, D. Ron, and K. Shen for strains and reagents; and anonymous reviewers for their insight. This work was supported by NIA R01 AG024365, NIH Developmental Biology Training Grant, and HHMI. A.D. is cofounder of Proteostasis Therapeutics, Inc. and declares no financial interest related to this work. Received: January 7, 2010 Revised: October 5, 2010 Accepted: November 30, 2010 Published: January 6, 2011 REFERENCES Aamodt, E.J., Chung, M.A., and McGhee, J.D. (1991). Spatial control of gutspecific gene expression during Caenorhabditis elegans development. Science 252, 579–582. Arantes-Oliveira, N., Apfeld, J., Dillin, A., and Kenyon, C. (2002). Regulation of life-span by germ-line stem cells in Caenorhabditis elegans. Science 295, 502–505. Benedetti, C., Haynes, C.M., Yang, Y., Harding, H.P., and Ron, D. (2006). Ubiquitin-like protein 5 positively regulates chaperone gene expression in the mitochondrial unfolded protein response. Genetics 174, 229. Bishop, N.A., and Guarente, L. (2007). Two neurons mediate diet-restrictioninduced longevity in C. elegans. Nature 447, 545–549. Bluher, M., Kahn, B.B., and Kahn, C.R. (2003). Extended longevity in mice lacking the insulin receptor in adipose tissue. Science 299, 572–574. Broughton, S.J., Piper, M.D.W., Ikeya, T., Bass, T.M., Jacobson, J., Driege, Y., Martinez, P., Hafen, E., Withers, D.J., and Leevers, S.J. (2005). Longer lifespan, altered metabolism, and stress resistance in Drosophila from ablation of cells making insulin-like ligands. Proc. Natl. Acad. Sci. USA 102, 3105. Burglin, T.R., and Cassata, G. (2002). Loss and gain of domains during evolution of cut superclass homeobox genes. Int. J. Dev. Biol. 46, 115–124. Calfon, M., Zeng, H., Urano, F., Till, J.H., Hubbard, S.R., Harding, H.P., Clark, S.G., and Ron, D. (2002). IRE1 couples endoplasmic reticulum load to secretory capacity by processing the XBP-1 mRNA. Nature 415, 92–96. Conboy, I.M., Conboy, M.J., Wagers, A.J., Girma, E.R., Weissman, I.L., and Rando, T.A. (2005). Rejuvenation of aged progenitor cells by exposure to a young systemic environment. Nature 433, 760–764. Copeland, J.M., Cho, J., Lo, T., Hur, J.H., Bahadorani, S., Arabyan, T., Rabie, J., Soh, J., and Walker, D.W. (2009). Extension of Drosophila life span by RNAi of the mitochondrial respiratory chain. Current Biology. Dell’Agnello, C., Leo, S., Agostino, A., Szabadkai, G., Tiveron, C., Zulian, A., Prelle, A., Roubertoux, P., Rizzuto, R., and Zeviani, M. (2007). Increased longevity and refractoriness to Ca2+-dependent neurodegeneration in Surf1 knockout mice. Hum. Mol. Genet. 16, 431. Dillin, A., Crawford, D.K., and Kenyon, C. (2002a). Timing requirements for insulin/IGF-1 signaling in C. elegans. Science 298, 830–834. Dillin, A., Hsu, A.-L., Arantes-Oliveira, N., Lehrer-Graiwer, J., Hsin, H., Fraser, A.G., Kamath, R.S., Ahringer, J., and Kenyon, C. (2002b). Rates of behavior and aging specified by mitochondrial function during development. Science 298, 2398–2401. Doonan, R., McElwee, J.J., Matthijssens, F., Walker, G.A., Houthoofd, K., Back, P., Matscheski, A., Vanfleteren, J.R., and Gems, D. (2008). Against the oxidative damage theory of aging: superoxide dismutases protect against oxidative stress but have little or no effect on life span in Caenorhabditis elegans. Genes Dev. 22, 3236. Esposito, G., Di Schiavi, E., Bergamasco, C., and Bazzicalupo, P. (2007). Efficient and cell specific knock-down of gene function in targeted C. elegans neurons. Gene 395, 170–176.

Feng, J., BussiE¨re, F., and Hekimi, S. (2001). Mitochondrial electron transport is a key determinant of life span in Caenorhabditis elegans. Dev. Cell 1, 633–644. Gems, D., and Doonan, R. (2008). Oxidative stress and aging in the nematode Caenorhabditis elegans. Oxidative Stress in Aging: From Model Systems to Human Diseases, 81. Hansen, M., Chandra, A., Mitic, L.L., Onken, B., Driscoll, M., and Kenyon, C. (2008). A role for autophagy in the extension of lifespan by dietary restriction in C. elegans. PLoS Genet. 4, e24.

McElwee, J.J., Schuster, E., Blanc, E., Thomas, J.H., and Gems, D. (2004). Shared transcriptional signature in Caenorhabditis elegans Dauer larvae and long-lived daf-2 mutants implicates detoxification system in longevity assurance. J. Biol. Chem. 279, 44533. Miller, D.M., Stockdale, F.E., and Karn, J. (1986). Immunological identification of the genes encoding the four myosin heavy chain isoforms of Caenorhabditis elegans. Proc. Natl. Acad. Sci. USA 83, 2305–2309.

Harman, D. (1956). Aging: A Theory Based on Free Radical and Radiation Chemistry. J. Gerontol. 11, 298–300.

Nonet, M.L., Staunton, J.E., Kilgard, M.P., Fergestad, T., Hartwieg, E., Horvitz, H.R., Jorgensen, E.M., and Meyer, B.J. (1997). Caenorhabditis elegans rab-3 Mutant Synapses Exhibit Impaired Function and Are Partially Depleted of Vesicles. J. Neurosci. 17, 8061–8073.

Haynes, C.M., Petrova, K., Benedetti, C., Yang, Y., and Ron, D. (2007). ClpP mediates activation of a mitochondrial unfolded protein response in C. elegans. Dev. Cell 13, 467–480.

Okkema, P.G., Harrison, S.W., Plunger, V., Aryana, A., and Fire, A. (1993). Sequence Requirements for Myosin Gene Expression and Regulation in Caenorhabditis elegans. Genetics 135, 385–404.

Hobert, O. (2002). PCR fusion-based approach to create reporter gene constructs for expression analysis in transgenic C. elegans. Biotechniques 32, 728–730. Hsin, H., and Kenyon, C. (1999). Signals from the reproductive system regulate the lifespan of C. elegans. Nature 399, 362–366. Hwangbo, D.S., Gersham, B., Tu, M.-P., Palmer, M., and Tatar, M. (2004). Drosophila dFOXO controls lifespan and regulates insulin signalling in brain and fat body. Nature 429, 562–566. Jose, A.M., Smith, J.J., and Hunter, C.P. (2009). Export of RNA silencing from C. elegans tissues does not require the RNA channel SID-1. Proc. Natl. Acad. Sci. USA 106, 2283. Kapahi, P., Zid, B.M., Harper, T., Koslover, D., Sapin, V., and Benzer, S. (2004). Regulation of lifespan in Drosophila by modulation of genes in the TOR signaling pathway. Curr. Biol. 14, 885–890. Kirchman, P.A., Kim, S., Lai, C.Y., and Jazwinski, S.M. (1999). Interorganelle signaling is a determinant of longevity in Saccharomyces cerevisiae. Genetics 152, 179–190. Lapointe, J., Stepanyan, Z., Bigras, E., and Hekimi, S. (2009). Reversal of the mitochondrial phenotype and slow development of oxidative biomarkers of aging in long-lived Mclk1+/mice. J. Biol. Chem., M109. Larsen, P.L., Albert, P.S., and Riddle, D.L. (1995). Genes that regulate both development and longevity in Caenorhabditis elegans. Genetics 139, 1567– 1583. Lee, C.K., Klopp, R.G., Weindruch, R., and Prolla, T.A. (1999). Gene expression profile of aging and its retardation by caloric restriction. Science 285, 1390. Lee, S.S., Kennedy, S., Tolonen, A.C., and Ruvkun, G. (2003). DAF-16 target genes that control C. elegans life-span and metabolism. Science 300, 644–647. Lee, S.S., Lee, R.Y.N., Fraser, A.G., Kamath, R.S., Ahringer, J., and Ruvkun, G. (2002). A systematic RNAi screen identifies a critical role for mitochondria in C. elegans longevity. Nature Genetics 33, 40–48. Libina, N., Berman, J.R., and Kenyon, C. (2003). Tissue-specific activities of C. elegans DAF-16 in the regulation of lifespan. Cell 115, 489–502. Link, C.D., Cypser, J.R., Johnson, C.J., and Johnson, T.E. (1999). Direct observation of stress response in Caenorhabditis elegans using a reporter transgene. Cell Stress Chaperones 4, 235–242. Liu, X., Jiang, N., Hughes, B., Bigras, E., Shoubridge, E., and Hekimi, S. (2005). Evolutionary conservation of the clk-1-dependent mechanism of longevity: loss of mclk1 increases cellular fitness and lifespan in mice. Genes Dev. 19, 2424–2434. Maduro, M., and Pilgrim, D. (1995). Identification and Cloning of unc-119, a Gene Expressed in the Caenorhabditis elegans Nervous System. Genetics 141, 977–988.

Pearl, R. (1928). The Rate of Living (London, UK: University of London Press). Qadota, H., Inoue, M., Hikita, T., K^ppen, M., Hardin, J.D., Amano, M., Moerman, D.G., and Kaibuchi, K. (2007). Establishment of a tissue-specific RNAi system in C. elegans. Gene 400, 166–173. Rea, S., and Johnson, T.E. (2003). A metabolic model for life span determination in Caenorhabditis elegans. Dev. Cell 5, 197–203. Rea, S.L., Ventura, N., and Johnson, T.E. (2007). Relationship between mitochondrial electron transport chain dysfunction, development, and life extension in Caenorhabditis elegans. PLoS Biol. 5, e259. Ron, D., and Walter, P. (2007). Signal integration in the endoplasmic reticulum unfolded protein response. Nat. Rev. Mol. Cell Biol. 8, 519–529. Rubner, M. (1908). Das Problem der Lebensdauer und seine Be-ziehungen zu Wachstum und ErnaE` hrung (Muenchen, Germany: R. Oldenburg). Russell, S.J., and Kahn, C.R. (2007). Endocrine regulation of ageing. Nat. Rev. Mol. Cell Biol. 8, 681–691. Taguchi, A., Wartschow, L.M., and White, M.F. (2007). Brain IRS2 signaling coordinates life span and nutrient homeostasis. Science 317, 369. Tatar, M., Bartke, A., and Antebi, A. (2003). The endocrine regulation of aging by insulin-like signals. Science 299, 1346–1351. Tsang, W.Y., and Lemire, B.D. (2002). Mitochondrial genome content is regulated during nematode development. Biochem. Biophys. Res. Commun. 291, 8–16. Van Raamsdonk, J.M., and Hekimi, S. (2009). Deletion of the Mitochondrial Superoxide Dismutase sod-2 Extends Lifespan in Caenorhabditis elegans. PLoS Genet. 5. Winston, W.M., Molodowitch, C., and Hunter, C.P. (2002). Systemic RNAi in C. elegans requires the putative transmembrane protein SID-1. Science 295, 2456–2459. Wolkow, C.A., Kimura, K.D., Lee, M.-S., and Ruvkun, G. (2000). Regulation of C. elegans life-span by insulin-like signaling in the nervous system. Science 290, 147–150. Yang, W., Li, J., and Hekimi, S. (2007). A Measurable increase in oxidative damage due to reduction in superoxide detoxification fails to shorten the life span of long-lived mitochondrial mutants of Caenorhabditis elegans. Genetics 177, 2063–2074. Yoneda, T., Benedetti, C., Urano, F., Clark, S.G., Harding, H.P., and Ron, D. (2004). Compartment-specific perturbation of protein handling activates genes encoding mitochondrial chaperones. J. Cell Sci. 117, 4055–4066. Zhao, Q., Wang, J., Levichkin, I.V., Stasinopoulos, S., Ryan, M.T., and Hoogenraad, N.J. (2002). A mitochondrial specific stress response in mammalian cells. EMBO J. 21, 4411–4419.

Cell 144, 79–91, January 7, 2011 ª2011 Elsevier Inc. 91

Dynamics between Stem Cells, Niche, and Progeny in the Hair Follicle Ya-Chieh Hsu,1 H. Amalia Pasolli,1 and Elaine Fuchs1,* 1Howard Hughes Medical Institute, Laboratory of Mammalian Cell Biology & Development, The Rockefeller University, New York, NY 10065, USA *Correspondence: [email protected] DOI 10.1016/j.cell.2010.11.049

SUMMARY

Here, we exploit the hair follicle to define the point at which stem cells (SCs) become irreversibly committed along a differentiation lineage. Employing histone and nucleotide double-pulse-chase and lineage tracing, we show that the early SC descendents en route to becoming transit-amplifying cells retain stemness and slow-cycling properties and home back to the bulge niche when hair growth stops. These become the primary SCs for the next hair cycle, whereas initial bulge SCs become reserves for injury. Proliferating descendents further en route irreversibly lose their stemness, although they retain many SC markers and survive, unlike their transit-amplifying progeny. Remarkably, these progeny also home back to the bulge. Combining purification and gene expression analysis with differential ablation and functional experiments, we define critical functions for these non-SC niche residents and unveil the intriguing concept that an irreversibly committed cell in an SC lineage can become an essential contributor to the niche microenvironment. INTRODUCTION Adult stem cells (SCs) govern tissue homeostasis and wound repair. They reside in a specific niche, defined as the microenviroment, that hosts and maintains SCs (Spradling et al., 2008). Most SCs are infrequently cycling, a feature thought to preserve their stemness, namely their ability to self-renew and remain undifferentiated over the animal’s lifetime. During normal homeostasis, they often exit from their niches and progress to become transit-amplifying (TA) cells, undergoing a series of rapid divisions before committing to terminal differentiation (Fuchs, 2009; Morrison and Kimble, 2006). Determining the point in a lineage hierarchy where SCs lose long-term self-renewing capacity and become irreversibly committed represents a fundamental and challenging question in SC biology. Transitioning from a slow-cycling to more fastcycling state is not indicative, as hematopoietic stem cells (HSCs) and hair follicle (HF) SCs can reversibly switch from dormancy to cycling during normal homeostasis and wound 92 Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc.

repair (Blanpain et al., 2004; Foudi et al., 2009; Nowak et al., 2008; Taylor et al., 2000; Waghmare et al., 2008; Wilson et al., 2008). Merely exiting their niche is also not a reliable measure, as some HSCs circulate, trafficking between their bone marrow niche and extramedullary tissues (Cao et al., 2004). Even embarking along a differentiation pathway may not be an unequivocal indicator of loss of stemness; studies in Drosophila and mouse testis show that germline SC niche vacancies can be filled by early spermatogonial cells that dedifferentiate when returned to the niche (Brawley and Matunis, 2004; Kai and Spradling, 2004; Brinster and Avarbock, 1994). The murine HF offers an excellent system for monitoring an SC lineage and exploring plasticity of SC progenies. During homeostasis, the lower HF cycles through bouts of active hair growth (anagen), destruction (catagen), and rest (telogen) (Lavker et al., 2003; Paus and Cotsarelis, 1999). When the new HF emerges, it grows next to the old hair, which persists into the next cycle. This creates a protrusion or ‘‘bulge,’’ first described >100 years ago (Unna, 1876). In 1990, nucleotide pulse-chase experiments revealed the existence of slow-cycling, label-retaining cells (LRCs) in the bulge (Cotsarelis et al., 1990). A decade later, these cells were isolated, characterized, and shown to self-renew longterm and contribute to HF lineages and wound repair (Blanpain et al., 2004; Claudinot et al., 2005; Ito et al., 2005; Morris et al., 2004; Tumbar et al., 2004; Zhang et al., 2009). These findings established the bulge as a bona fide HF-SC niche. Hair growth is fueled by bulge SCs, which are activated at the start of anagen by the dermal papilla (DP), a cluster of underlying mesenchymal cells. Upon activation, SCs exit the bulge and proliferate downward, creating a long linear trail of cells, the outer root sheath (ORS) (Ito et al., 2005; Zhang et al., 2009). In mature HFs, the ORS extends from bulge to matrix. Enveloping the DP at the HF base, matrix cells cycle rapidly but transiently before differentiating upward to generate the hair and its channel (Figure 1A and Figure S1A available online). Catagen illuminates an unambiguous distinction between long-lived HF-SCs and short-lived matrix progeny that undergo massive apoptosis. The remaining epithelial strand retracts, drawing the DP upward. Current evidence suggests that at the catagen/telogen transition, a few bulge SCs migrate to meet the DP, generating the hair germ (HG) (Ito et al., 2004; Zhang et al., 2009). Bearing closer resemblance to bulge than matrix, HG cells are activated prior to bulge at the start of anagen (Greco et al., 2009). Prior to activation, HF-SCs undergo an extended rest period that can last for months.

Figure 1. Dynamics of Slow- and Fast-Cycling Cells throughout the Hair Cycle (A) Cycling portions of a mature HF. (B–D) Tet-Off H2BGFP mice were chased from P21 to the days/stages noted (P35–P37 corresponds to AnaVI). Before harvesting skin, some mice were given 236 hr pulses of BrdU. Schematic depicts H2BGFP cells (green) that will be label-retaining (LRCs) when the chase is stopped at each stage. Cell #s are counting down from the bulge base (= 0). In (C), the right panels of each pair are duplicates of GFP monochromes of the left panels. Scale bars, 30 mm. Bu, bulge. DAPI in blue. In (D), note that S phase cells in AnaVI are mainly in ORSlow and matrix. %HFs with BrdU+ cells in bulge or in different ORS segments were quantified from P29–P35. For each stage, n = 2 mice and R23 HFs/mouse were counted. Data are mean ± standard deviation SD.

Whereas extensive studies have been performed on bulge SCs and TA-matrix cells, the properties and fates of ORS cells are less clear. Although these cells do not express high levels of CD34, a marker of their SC bulge predecessors, they display many HF-SC markers not found in matrix, including Lgr5, Sox9, Lhx2, and TCF3 (Fuchs, 2009). To date, functional studies on ORS cells have been limited to cultures of microdissected rat whisker HFs, where long-term clones capable of engraftment were obtained not only from bulge and ORS but also from matrix (Claudinot et al., 2005; Oshima et al., 2001; Rochat et al., 1994). Whether ORS cells in vivo possess stemness like their predecessors or are committed like their TA progeny remains unknown. The unique regenerative aspects of HF homeostasis and the transient, but spatially and temporally well-defined ORS offer an unparalleled opportunity to examine the transition between a SC and a TA cell. Where does the transition from slow cycling to fast cycling occur along the ORS trail? What happens to ORS cells during catagen? Do they undergo apoptosis like TA cells, or do they survive like SCs? If they survive, do they home back to the bulge, and if they do so, do they function as SCs? Answering these questions could provide fundamental insights into defining not only the importance of bulge to SC survival but also the point where an SC loses its stemness and embarks upon a terminal differentiation pathway. If all ORS cells either die or differentiate during catagen, this would imply that HF-SCs lose stemness once they leave the bulge. If on the other hand, some ORS cells survive and return back to the bulge as functional SCs, this would imply that at least at some point along the progression from bulge to TA compartments, an SC that has left the bulge still possesses intrinsic features of stemness. In this report, we address these key questions in the unperturbed in vivo confines of the normal hair cycle, as well as in response to injury. In so doing, we unearth surprising and dynamic features of the HF-SC niche and its progeny. First, we

show that during anagen, the upper ORS cells remain slow cycling, like their SC predecessors, whereas lower ORS cells cycle more rapidly. Moreover, slow-cycling ORS cells survive catagen and contribute mightily to both HG and a new bulge: in fact, they become the main source of SCs used during the next hair cycle. Lacking a DP, the initial bulge ceases to play a major role in homeostasis but can respond to injury. Finally and perhaps most surprisingly, some actively cycling lower ORS cells not only survive catagen but also home back to the bulge. Like their upper ORS predecessors, these cells retain many HF-SC markers. However, they irreversibly lose their ability to proliferate in normal homeostasis or upon wounding. Instead, they function decisively in hair anchorage during the resting phase and in providing the quiescent signaling cues that control the hair cycle. Our findings define a point along the ORS where cells lose stemness and become irreversibly fated to differentiate or die and illuminate how downstream SC progeny can provide negative feedback to the niche and restrict SC self-renewal and tissue formation. RESULTS Bulge Descendents in Anagen ORS Can Be Subdivided into Slow Cycling and Faster Cycling To dissect the properties of bulge SC descendents along the ORS, we first addressed whether these cells remain slow-cycling like bulge or become faster-cycling like matrix. We used bi-transgenic mice expressing Doxycycline (Doxy)-repressible histone H2BGFP controlled by a keratin 5 (K5) promoter. With this TetOff system, skin epithelium is uniformly labeled until Doxy exposure, when new H2BGFP synthesis is tightly repressed and cells deplete existing GFP by 23/division (Tumbar et al., 2004). Doxy chase was begun at postnatal day (P)21, just before the start of the 1st hair cycle (Figure 1B). At the last stage of anagen (AnaVI, based on Muller-Rover et al., 2001; P35–P37 in these mice), the bulge but not matrix contained bright H2BGFP LRCs (Figures 1C and 1D). Interestingly, an LRC trail extended from the bulge base along the ORS. To quantify, we standardized image acquisition for GFP intensities by setting the detector so Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc. 93

Figure 2. Label-Retaining HF-SC Descendents in the ORS Are Spared from Apoptosis during Catagen and Form a New Bulge and HG at Telogen Tet-Off H2BGFP mice were chased from P21 to the HF stages indicated. Prior to analysis, some mice were given a 4 hr BrdU pulse. (A) H2BGFP LRCs maintain the same distribution from anagen VI to catagen’s end. (B) BrdU labeling in early catagen (Cat-II) shows proliferation only in matrix and ORSlow and not ORSGFP+ or ORSmid. (C–E) TUNEL and CP3 immunolabeling in catagen VI–VIII reveals cell death in the retracting epithelial strand (ES) but rarely in LRCs. (F) Fate of H2BGFP LRCs at catagen to 2nd telogen transition. Like the old bulge, the emerging new bulge and HG contain LRCs. Quantifications: n = 3 mice; 42 total HFs analyzed for bulge; 87 for HG. (G) HF-SC bulge marker CD34 is upregulated in upper ORSGFP+ beginning at catagen V. The shape of the future new bulge and HG can already be distinguished at Cat VIII. HFs are outlined by dotted white line. White dots mark autofluorescent hair shafts. Scale bars, 30 mm. Data are mean ± SD.

(#17–40) remained proliferative 2 days after ORSGFP+; and ORSlow cells (#40–80) and matrix cycled continuously.

that brightest bulge cells were maximal but not saturated in a 12 bit (= 4096 shades) image. Under this condition, bulge was composed almost exclusively of cells with GFP intensities >256 (GFPbright). Within the first 16 positions counting down along the ORS trail in a planar section, >85% of cells were GFP+ (intensity > 16) with the majority of GFPbright cells in #1–12. By contrast, few ORS cells in sites 16–30 were GFP+, and below #30, GFP was not detected (Figure 1C and Figure S1B). To further probe proliferation heterogeneities during anagen, we performed a series of short BrdU pulse experiments in TetOff H2BGFP mice. When labelings were conducted at intervals between P27 and P29, i.e., after HFs displayed an emerging inner root sheath (IRS) and typical anagen shape, BrdU+ cells were detected in bulge, ORS, and matrix. By P31, however, most bulge cells had ceased proliferation. Although occasional BrdU+ cells were still seen in ORS positions 1–16, proliferation was mostly beneath this zone in ORS and matrix. By P33, most BrdU+ ORS cells were below #30, and from P35–P37, BrdU+ cells were largely restricted to matrix, with few labeled ORS cells <#40 (Figure 1D and Figure S1C). The transition between ORS and matrix was at #75–85. To summarize: ORSGFP+(#1–16) behaved similarly to bulge, displaying slow-cycling characteristics and returning to quiescence 2 days after bulge proliferation ceased; ORSmid 94 Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc.

SCs that Exit the Bulge during the Growth Phase and Remain Slow Cycling (ORSGFP+) Are Spared From Apoptosis during the Destructive Phase Because ORSGFP+ cells divide only a few times during anagen, we next wondered what happens to these cells during the destructive phase. Throughout catagen, the ORSGFP+ trail showed no signs of migration or expansion (Figure 2A). Three possible lineage behaviors could explain this result: (1) GFP+ ORS cells are static in catagen; (2) the influx of bulge cells to ORSGFP+ is compensated by an efflux of ORSGFP+ cells, proliferating and moving toward matrix; (3) ORSGFP+ cells undergo apoptosis and bulge cells migrate downward to replace them. To distinguish between these models, we monitored apoptotic and proliferative events during this period. Early catagen was marked by dramatic reduction in matrix proliferation (Figure 2B). Cells within bulge and ORS were mostly quiescent during catagen. These data rule out model 2. Soon thereafter, cell death was detected by DNA fragmentation (TUNEL) and activated caspase 3 (CP3). Apoptosis began within matrix and expanded upward into the retracting epithelial strand (Figure 2C). However, even toward catagen’s end, only a few ORSGFP+ and bulge LRCs scored positive for TUNEL or CP3, and ultrastructural analysis suggested that these were likely healthy cells that had engulfed apoptotic debris (Figure 2D and Figures S2A–S2D). These data rule out model 3. Together, these differences revealed a striking molecular boundary between the TUNEL/CP3-positive epithelial strand and the apparently unscathed LRCs in bulge and ORS. These data are consistent with model 1.

ORSGFP+ Cells Form a New Bulge and Hair Germ at the End of Catagen As most ORSGFP+ cells survive catagen, we next addressed where they go. Near the end of catagen, the LRC trail was still located below the initial bulge. When HFs were followed into telogen and DP adopted its characteristic position directly below the HG, it became apparent that the ORSGFP+ trail had adopted a form resembling the HG and a new bulge (Figure 2E). Imaging and quantifications coupled with fluorescenceactivated cell sorting (FACS) analysis confirmed that 98% of CD34+ cells within this new bulge and 80%–90% of those in HG were GFP+ (Figure 2F and Figure S3C). Because in the TetOff H2BGFP system, a GFP cell from an earlier time cannot be the source for a GFP+ cell at a later point, this leaves only ORSGFP+ and initial bulge as possible sources for this structure. Given that from catagen to telogen, the initial bulge showed no major change in size or cell number (Figure S2E), our collective evidence favored ORSGFP+ as the major source of both this new bulge and HG. These results were surprising, however, because anagen ORSGFP+ cells did not show strong immunostaining for either bulge marker CD34 or HG marker P-cadherin. After first confirming that new bulge and HG express these markers (Figure 2F), we traced the temporal roots of these differences. Toward catagen’s end, the first 8 cells in the ORSGFP+ zone acquired strong CD34 immunostaining, whereas the lower part of ORSGFP+ remained CD34 but became P-cadherin-bright (Figure 2G; data not shown). To further monitor these events, we performed a lineagetracing experiment with Lgr5-EGFP-IRES-CreER/Rosa-stopLacZ-stop mice (Barker et al., 2007). At full anagen, Lgr5 was expressed by both bulge and ORS; however, as ORS is >153 larger than bulge, we could preferentially mark ORS cells by giving a low dose of tamoxifen (Figure 3A; Figures S3A and S3B). Indeed when examined by thick sections (90 mm) at anagen’s end, >70% of HFs had LacZ+ ORS cells, whereas only 4% had LacZ+ bulge cells. When traced into telogen, >20% of HFs had LacZ+ cells in new bulge or HG, whereas HFs with LacZ+ cells in the old bulge remained at 5%. These data add to the evidence that ORSGFP+ is the main source for both new bulge and HG. Based upon our proliferation analyses of ORS during anagen, we next began with pulse-chased Tet-Off H2BGFP mice and superimposed a series of BrdU pulse-chases that allowed us to differentiate old bulge and ORSGFP+ and selectively label different ORS segments. As proliferation within the initial bulge occurs during early anagen (P25–P29), we preferentially labeled ORS by administering BrdU at P30–P32 (Figure 3B). Moreover, because cells below ORSGFP+ cycle frequently, a 6 day chase at anagen’s end and into catagen (P38) restricted most BrdU+ cells to the ORSGFP+ zone. When chased further into telogen, these GFP/BrdU double+ cells were found in new bulge and HG, thereby demonstrating that the ORSGFP+ of the prior cycle contributes substantially to the new bulge and HG. Finally, we used epifluorescence intensity analyses to estimate the average number of divisions that GFP+ LRCs within each compartment undergo during a hair cycle. Because cell divisions are silenced during catagen and telogen, GFP intensi-

ties within each compartment are retained, and thus, their intensities in telogen serve as a guide to their origin nearing anagen’s end. Analyses of GFP+ cells chased from the start of the first postnatal anagen to telogen revealed that cells that maintained their residence in the initial bulge divided the least (23) during the hair cycle: their epifluorescence corresponded to the old bulge (Figure 3C). By contrast, cells that resided in the ORS and acquired CD34 during catagen corresponded in epifluorescence to new bulge and on average divided 1–23 more than the old bulge. LRCs that resided in the ORS, but remained CD34, displayed epifluorescence corresponding to HG and reflective of 1–23 more divisions than new bulge. Quantifications in control mice showed that without Doxy, HG and new and old bulges displayed uniform levels of the highest H2BGFP signal, underscoring tight Tet-Off regulation and validating the efficacy of the data (Figure S3C). Together, these results establish that cells in ORSGFP+ are the major source of HG and new bulge. Furthermore, they unveil a hitherto unrecognized relation between HF-SCs that exit the bulge during the growth phase and the new bulge and HG that form during the destructive phase. Some Mid-Zone ORS Cells Contribute to the HG The ability of cells within the ORSGFP+ to survive catagen led us to wonder whether more fast-cycling bulge progeny located in the middle or lower ORS might also be spared and possibly be the source of the 10%–20% of HG cells with very low GFP. To test this hypothesis, we delayed our BrdU pulse to P34– P36, a time when most bulge and ORSGFP+ cells had ceased proliferation. Although ORSmid cells were proliferative at this time, they became quiescent soon thereafter, as evidenced by their label retention after a 4 day chase. These BrdU+ LRCs were uniformly low for GFP and could be found in the HG following a chase into telogen (Figure S3D). Thus although most of the HG was comprised of ORSGFP+ cells, the mid-zone was the source of HG cells undergoing the most divisions in the prior hair cycle. This cut-off was further defined by delaying the BrdU pulse so that cells below the mid-zone were preferentially labeled: under these conditions, no BrdU+ cells were detected in the HG (see below). Together, these experiments delineate the ORSGFP+ zone as the point at which an HF-SC can exit the initial bulge and still be recycled to the CD34+ new bulge. The point at which HF-SCs can be recycled to the HG extends to the mid-zone. Moreover, these data reveal a strong inverse correlation between the number of divisions progeny have undergone and whether they are recycled to bulge versus HG. A Potential Conundrum: The Return of Some Fast-Cycling Bulge Descendents below the ORSmid Zone to the Niche During our double-label pulse-chase studies, we noticed that some chased BrdU+ cells started to show up in the new bulge when BrdU was pulsed at P34–P36. Their number increased dramatically when the BrdU pulse was delayed to P36–P38 (end of anagen) and then chased through catagen, suggestive that these cells came from a position below the ORS midzone (Figure 4A). Closer inspection revealed that in contrast Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc. 95

96 Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc.

to ORSGFP+, these BrdU+ cells were restricted to the innermost new bulge. Immunolabeling and FACS analysis showed that the cells were CD34 and hence distinct from suprabasal CD34+ bulge cells described previously (Blanpain et al., 2004). Instead, these cells were positive for keratin 6 (K6) (Figure 4A; Figures S4A–S4C). K6 is also known to mark the terminally differentiated companion layer, which is derived from matrix during anagen and sandwiched between ORS and IRS (Figure S1A). The K6+ bulge cells differed from companion layer by their expression of key HF-SC transcription factors (Figure 4B). Ultrastructurally, K6+ bulge cells also more closely resembled CD34+ bulge SCs than companion layer cells (Figures S4D–S4F). However, they differed from bulge SCs in making unusual adhesive contacts with the club hair and extensive desmosomal contacts with their neighbors. Thus, the cells fated to enter the innermost layer of the new bulge at the end of catagen were distinct from any of the known HF residents. The K6+ Bulge Layer Is Derived from Actively Cycling Cells in the Lower ORS A priori, the K6+ bulge layer could be a bona fide companion layer, derived from matrix but displaying distinct molecular features at different hair cycle stages. Alternatively, it could be derived from actively cycling lower ORS cells, which face a barrier to lineage progression when matrix apoptoses (catagen IV). The latter possibility was intriguing given that the ORS expresses many HF-SC markers that matrix and anagen companion layers do not (Rendl et al., 2005) (Figure S5A). To identify the source of K6+ bulge cells, we first revisited our Lgr5-CreER/Rosa-LacZ lineage tracings, which labeled cells in both upper and lower ORS, but not matrix when induced near the end of anagen (Figure 3A). In telogen, LacZ+ cells were present in both CD34+ and K6+ bulge layers (Figure 3A and Figure S5B). Similar results were obtained when lineage tracings were conducted with K14creER/Rosa26-LacZ mice (Figure S5C). Together with the BrdU pulse results, these data suggest that the K6+ bulge layer comes from lower ORS and not matrix. To further demonstrate that the lower ORS/matrix cell step that occurs during anagen is bypassed in catagen, we used K14-Tet-On/H2BGFP mice, which, upon Doxy, turn on H2BGFP in ORS (Nguyen et al., 2006). When Doxy was given at mid-anagen and HFs were examined at anagen’s end, H2BGFP was nicely expressed in ORS (including its CD34+ bulge layer)

but not matrix or companion layer (Figure 4C). Given that H2BGFP is stable and ORS cells don’t divide during catagen, the fate of label could be monitored. Through catagen IV, bright H2BGFP persisted only in ORS. At catagen V after the matrix apoptosed, GFP+ cells appeared in the K6+ layer at the tip of the retracting club hair (shown). These cells survived catagen, remained K6+ and GFP+, expressed HF-SC/ORS markers, and wound up in the telogen bulge (Figure 4C; Figures S5F and S5G). Moreover, because when chased into telogen, K6+ cells of new bulge were GFP+ whereas those of old bulge were GFP, the GFP labeling reflected lineage fate mapping and not ectopic K14 promoter activity (shown). We also confirmed this by inducing H2BGFP in telogen and verifying absence of GFP+ cells in the K6+ bulge layer (Figure 4C; Figure S5D). These results exclude both matrix and anagen companion layer as possible precursors of the K6+ bulge. Moreover, the catagen-imposed bypassing of the ORS/matrix transition explains the dramatically different biology between cells of the K6+ bulge (ORS-derived) and companion layer (matrix-derived). To specifically follow the fate and movement of ORSlow cells, we combined a BrdU pulse-chase scheme (pulse at the end of anagen) with this Tet-On H2BGFP strategy. ORSlow was the only population with BrdU/H2BGFP double+ cells in every HF. As expected, starting from catagen V, double+ cells appeared in the K6+ layer around the newly formed club hair and retracted upward with it as the new bulge formed. During catagen, a few double+ cells also appeared in the epithelial strand and in the external layer flanking newly formed K6+ cells (Figure 4C; Figure S5G). In sharp contrast to these layers, however, no TUNEL+ or CP3+ cells were detected in the ORS-derived K6+ layer (Figure S5E, 0/352 HF examined). Thus, although some ORSlow cells survived catagen and wound up in new bulge, they took up residence in the K6+CD34 layer. K6+ Bulge Cells Do Not Participate in Normal Homeostasis We next wondered whether niche environment might restore HF-SC function to lower ORS-derived bulge cells, enabling them to utilize the remaining proliferative capacity they possessed prior to catagen. The importance of testing this possibility was heightened when we discovered that the K6+ layer displayed not only HF-SC markers Sox9, Lhx2, and TCF3, still expressed by lower ORS, but also nuclear NFATc1, previously found only in bulge SCs (Horsley et al., 2008) (Figure S6A).

Figure 3. ORSGFP+ LRCs Form the HG and CD34+ SCs of the New Bulge Schematics summarize experiments and results. (A) Lineage tracing with Lgr5CreER/Rosa-LacZ to monitor ORS cell fate. Tamoxifen was given in full-anagen. Three days later (AnaVI), most HFs had LacZ+ cells in ORS (both up and low) but not bulge or matrix. When chased to telogen, LacZ+ cells showed up mainly in new bulge (both CD34+ and K6+ layers). n = 2 mice, 176 HFs. Quantifications were on 90 mm sections; images shown are 20 mm sections (hematoxylin counterstain). (B) H2BGFP/BrdU double-label, double-pulse-chase experiment to monitor fate of upper ORSGFP cells. Tet-Off H2BGFP mice were chased from P21 and BrdU pulses during mid-anagen (P30–P32) preferentially labeled cells in the upper ORSGFP trail. When chased to the anagen/catagen transition (P38), many HFs had BrdU+ LRCs in their ORSGFP+ (arrowhead); BrdU+ LRCs were rare elsewhere (n = 3 mice, 42 HFs). When chased to telogen (P43), BrdU+/GFP+ cells (arrowhead) were mainly in the new bulge or HG (n = 3 mice, 58 HFs). Scale bars, 30 mm. new Bu and K6+ Bu denote CD34+ and K6+ layers of new bulge. (C) Relative epifluorescence intensities of cells from different LRC populations of Tet-Off H2BGFP mice chased from the 1st telogen through one hair cycle. Raw datasets are plotted to right of each box-and-whisker diagram: median, 25th and 75th percentiles are denoted by notch, bottom and top boxes; 5th and 95th percentile are whiskers; minimum and maximum measurements are x’s. Asterisks indicate significant differences between datasets (**p < 0.01). Data are mean ± SD.

Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc. 97

CD34+ bulge cells incorporated BrdU, no proliferation was detected in K6+ bulge cells of >190 HFs examined (Figure 5A). Although the K6+ bulge layer was quiescent, it was possible that some of its cells might migrate to the CD34+ layer or ORS and contribute to the hair cycle. We ruled this out by administering BrdU at the end of anagen and then chasing into telogen to selectively label K6+ bulge cells (Figure 5B). When chased throughout the following hair cycle, no migration of BrdU-labeled cells was observed. In the next telogen, BrdU label remained within the K6+ layer of what became the old bulge (Figure 5B; Figure S6B). Our data infer that even though highly proliferative cells in anagen can wind up in the bulge, slow-cycling CD34+ cells are the source of new HFs. To further substantiate this, we took Tet-Off H2BGFP mice that were chased from P21 to the 2nd telogen, so that H2BGFP labeled CD34+ old and new bulge and CD34 HG but not K6+ bulge cells or HF cells above the bulge (Figure 5C). When the 2nd adult anagen began, the emerging follicle was composed of GFP+ cells. Even though proliferation rapidly diluted out label, the fluorescence intensity was still higher than all other HF compartments except CD34+ bulge and HG. Thus although prior lineage tracings have shown contributions of CD34+ bulge and HG cells to the new HF (Morris et al., 2004; Zhang et al., 2009), these new data further suggest that cells in the K6+ bulge layer, junctional zone, sebaceous gland, and infundibulum do not contribute to normal HF homeostasis. Based upon these data, CD34+ slow-cycling SCs appear to be the sole source of SCs used for HF homeostasis.

Figure 4. K6+ Bulge Cells Express HF-SC Markers and Are Derived from Actively Cycling Lower ORS during Catagen (A) H2BGFP/BrdU double-label, double pulse-chase scheme to monitor fate of cells below the mid-zone ORS. BrdU was administered to Tet-Off H2BGFP mice at anagen’s end (P36–P38) and analyzed directly, or after chasing to telogen. At P38, many lower ORS and matrix cells are BrdU+ (n = 2 mice, 38 HFs). When chased to P43, BrdU+ LRCs are restricted to the K6+ layer of the new bulge (n = 4 mice, 129 HFs). (B) K6+ bulge cells express key transcription factors characteristic of bulge SCs and their ORS progeny. (C) Lineage tracing with K14Tet-On/H2BGFP coupled with BrdU pulse-chase. Doxy was administered to turn on H2BGFP in the entire ORS, but not K6+ companion layer or matrix, in late anagen. From catagen V through telogen, H2BGFP is detected in K6+ cells at the tip of the newly formed club hair. This layer retracts with the club hair during catagen and winds up in the new bulge by telogen. Note that K6+ old bulge cells are not GFP+. If BrdU is given at anagen’s end, BrdU+/H2BGFP+ cells are found in ORSlow. At catagen, double+ cells are in the K6+ layer, which by telogen has moved inside new bulge. Scale bars, 30 mm. K6+ club = K6+ layer enclosing new club hair. new Bu and K6+ Bu denote CD34+ and K6+ layers of new bulge (n = 2 mice; >23 HFs per stage). Data are mean ± SD.

We first addressed whether K6+ bulge cells proliferate in the next hair cycle. To capture proliferation if it existed, we pulsed with BrdU throughout the 2 day window of early anagen when HF-SCs are most active. Under conditions where >60% of 98 Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc.

CD34+ HF-SCs in Both Old and New Bulge Participate in Wound Repair, whereas K6+ Bulge Cells Do Not Respond to Either Wounding or CD34+ Bulge Cell Ablation The new HF always emerged from new and not old bulge, leading us to wonder whether the old bulge functions in normal homeostasis. Careful inspection revealed an occasional proliferative cell at the base of the old bulge, but only when BrdU was pulsed at anagen III, i.e., the height of new bulge activity (Figure 6A). However, when quantified, this constituted <15% of combined bulge activity, indicating that the CD34+ new bulge is primarily responsible for fueling new HF downgrowth. CD34+ bulge cells are known to contribute to wound repair (Blanpain et al., 2004; Ito et al., 2005; Morris et al., 2004). Two additional questions now surfaced: (1) As the old bulge still contains CD34+ HF-SCs, does it contribute to wound repair, and if so, does it do so equivalently with the new bulge? (2) While refractile to normal homeostasis, do K6+ bulge cells participate in wound repair? To address these questions, we introduced punch wounds during the extended 2nd telogen, administered BrdU, and then examined the skin 2 days later (Figure 6B). BrdU+ cells were found in the CD34+ outer layer of new bulge and its associated HG. Proliferating CD34+ cells were also detected in old bulge. Quantifications revealed similar contributions from the bulges, but only from CD34+ and not K6+ cells. Next, we tested responses under more extreme conditions: HF-SC ablation (Figure 6C). K15-CrePGR mice were used to induce expression of diphtheria toxin receptors (DTRs) (Buch et al., 2005) in CD34+ bulge cells. RU486 was given to activate CrePGR, followed by DT to ablate CD34+ bulge cells. Activated

Figure 5. CD34+ Bulge SCs and HG Are the Only Cells Used in Normal HF Homeostasis (A) Representative example and quantifications of HFs in early anagen (AnaII–AnaIII) with 2 day BrdU pulse. Note BrdU in some CD34+ HF-SCs (arrowheads) but not inner K6+ bulge cells. White dots denote autofluorescent club hair (n = 2 mice, 61 HFs for AnaII; n = 3 mice, 133 HFs for AnaIII). (B) A 2 day BrdU pulse given at anagen’s end is followed through two hair cycles. Although not proliferative, K6+ bulge cells retain label from late anagen predecessors. K6+/BrdU+ cells persist through the next hair cycle and become the K6 layer of old bulge. (C) Lineage tracing using Tet-Off H2BGFP mice shows that the new HF comes from bulge LRCs. Chase was begun at 1st telogen; analysis began at end of 2nd telogen when H2BGFP selectively labels old bulge, new bulge, and HG. Developing HFs are composed solely of H2BGFP+ cells. IF, infundibulum; JZ, junctional zone; SG, sebaceous gland. Scale bars, 30 mm. Box-and-whisker plots measure H2BGFP epifluorescence intensity within different HF populations. Note paucity of GFP intensity in IF, JZ, and K6+ bulge, revealing their lack of contribution to GFP+ newly forming HF. **p < 0.01. Data are mean ± SD.

CP3 was prevalent in CD34+ but not K6+ bulge layers, verifying selective targeting. Death was equivalent for both new and old bulges, and a 2 day BrdU pulse prior to analyses further revealed proliferation in CD34+ bulge cells that had not been ablated (Figure 6C; data not shown). By contrast, K6+ bulge cells did not respond even when the HF-SC reservoir was depleted. K6+ Bulge Cells Fail to Grow In Vitro To address K6+ bulge cell potential, we devised a strategy to isolate them by FACS and then carried out colony formation assays (Figure 6D). An Lhx2-GFP transgene behaved similarly to endogenous Lhx2, displaying GFP in both K6+ and CD34+ bulge layers of telogen HFs (Figure S6C). K6+ and CD34+ cells in the GFP+ pool were then further fractionated by differential CD34, E-cadherin, and integrin levels: K6+ bulge cells were gated as GFP+CD34a6-b1Ecadhigh, whereas CD34+ cells were sorted as GFP+ and CD34+. Post-sort marker analysis re-

vealed >94% purity, and consistent with immunofluorescence, RT-PCR showed that K6+ and CD34+ cells expressed similar levels of Sox9, Lhx2, Tcf3, and Nfatc1 (Figures S6C and S6D). In vitro, even though both populations attached to their substratum, only CD34+ and not K6+ bulge cells formed colonies (Figure 6D). Collectively, these data suggest that lower ORS cells that home back to the bulge have irreversibly lost their proliferative and regenerative potential. K6+ Bulge Cells Function to Retain the Hair Coat and Maintain HF-SC Quiescence If K6+ bulge cells do not function as HF-SCs, what are their functions in the bulge niche? The first clues came from hair-plucking studies (Figure S7A). When telogen hairs were plucked, the entire K6 layer came with the club hair, demonstrating a tight association. By contrast, most CD34+ cells remained in the bulge. Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc. 99

Figure 6. CD34+ but Not K6+ Bulge Cells Are Activated in Wound Repair and upon HF-SC Ablation (A) Homeostasis. Mice given a 2 day pulse with BrdU at start of 2nd anagen were monitored to AnaIII. Arrowhead marks BrdU+ cell in old bulge, which is >33 less active than new bulge (n = 3 mice, 62 HFs). (B) Wounding. Punch biopsies were from 2nd telogen. BrdU given over 2 days shows that old and new bulges participate comparably in injury responses (n = 3 mice, 31 HFs). (C) SC ablation. K15-CrePGR was used to express DTR in CD34+ bulge during 2nd telogen. After DT and BrdU, mice were analyzed (n = 3 mice, 64 HFs). Note that when bulge SCs are ablated, remaining CD34+ cells from old and new bulges proliferate, whereas K6+ bulge cells are refractory. Scale bars, 30 mm. (D) K6+ and CD34+ bulge cells were isolated by FACS from telogen HFs of Lhx2-EGFP mice and subjected to culture experiments, performed in triplicate. One thousand cells of each population were assayed for attachment 16 hr post-plating. K6+ and CD34+ cells were also plated, and at 3 weeks, colonies were fixed and stained with 1% Rhodamine B. Data are mean ± SD.

To test whether the K6+ layer functions to anchor the club hair to the bulge, we devised a strategy to preferentially ablate K6+ versus CD34+ bulge cells (Figure 7A). Sox9-CreER (Soeda et al., 2010) is expressed in both K6+ and CD34+ bulge layers. However, in Sox9-CreER/Rosa-iDTR mice, when hairs were first trimmed and tamoxifen was applied topically, drug entered the hair pores and preferentially induced DTR in K6+ cells. By contrast, RU486 induced DTR in CD34+ bulge cells of K15CrePGR/Rosa-iDTR mice (Figure 7D and data not shown). Following DTR induction, mice were then treated with DT. Within 4 days, Sox9-iDTR mice began to show hair loss and by day (D) 6 100 Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc.

the coat was gone (Figure 7B; 5 of 5 mice tested). During this time, K15-iDTR mice kept their hair coat even when slightly more bulge cells were ablated (Figure S7B). Interestingly, 2 days after hair loss, skins of Sox9-iDTR mice became dark (Figure 7B). This indicator of anagen entry (Muller-Rover et al., 2001) was substantiated by histological analyses (shown). It occurred R50 days earlier than in K15iDTR or control mice (Figure 7C). By mechanisms unknown, hair plucking also triggers anagen reentry. Interestingly, Sox9iDTR mice were >4 days quicker to enter anagen compared to mice undergoing hair plucking, but K6+ bulge cells were

also lost by plucking (Figure S7A). These collective findings led us to consider the possibility that it is loss of K6+ bulge that triggers this precocious anagen. We therefore assessed whether K6+ bulge cells actively participate in maintaining telogen. To focus specifically on the effects of K6+ cell loss rather than hair loss or mechanical trauma, we identified conditions wherein sufficient K6+ bulge cells survived in our Sox9-iDTR mice so that club hairs were retained longer. We then adjusted RU486 so that total bulge cell death events were similar between Sox9- and K15-iDTR mice (Figure S7B). Under these conditions, CP3+ cells were detected in CD34+ (K15-iDTR) and K6+ (Sox9-iDTR) layers beginning at D2, peaking at D4–D6 and waning by D10 (Figure 7D; data not shown). In K15-iDTR mice, apoptosis was accompanied by brief proliferation from surviving CD34+ bulge cells, as detected by sequential BrdU pulses over this interval. This appeared to be a repair response, as HFs thereafter returned to quiescence and anagen was not induced (Figure 7E; Figure S7C). By contrast, despite club hair retention (Figure 7E, inset), Sox9iDTR HFs not only displayed more robust CD34+ proliferation but soon afterwards entered anagen. Throughout these treatments, K6+ cells showed no signs of proliferation. These effects were not observed with RU486, tamoxifen, or DT alone (Figure 7C; data not shown), confirming that K6+ bulge cells contribute markedly in maintaining the resting state of the hair cycle. Intriguingly, the effects of K6+ cell loss were transient: following anagen entry, CD34+ bulge cells gradually returned to quiescence (data not shown). To explore what potential signals from the K6+ bulge might influence CD34+ cell behavior, we focused on Fgf18 and Bmp6. These genes were previously shown to be upregulated in CD34+ telogen bulge cells in vivo, and their encoded factors inhibit HF-SC cycling in vitro (Blanpain et al., 2004; Greco et al., 2009). Remarkably, purified K6+ bulge cells displayed enormous enrichment of these genes compared to other epidermal or dermal cells that might impact on bulge microenvironment. Moreover, fgf18 and bmp6 expression levels were >253 in K6+ versus CD34+ bulge cells (Figure 7F). To test whether these factors account for the ability of K6+ bulge cells to maintain CD34+ bulge quiescence, we ablated the K6+ layer in Sox9-iDTR mice and injected these growth factors intradermally using fluorescent beads as a guiding reference. In contrast to buffer and beads alone, each of these factors potently suppressed CD34+ bulge activation upon ablation of the K6+ layer (Figure 7G; Figure S7D). DISCUSSION Recycling SCs Caught in Transit from Their Niche to The TA Compartment In this study, we examined the fate of HF-SCs that departed from their niche but had not yet reached the TA pool when the destructive phase began. In contrast to the prevailing view, we discovered that many ORS cells in transit between bulge and matrix survive the massive apoptosis that follows anagen. By using H2BGFP as a sensitive and stable label for lineage tracing, we not only tracked the fate of CD34+ HF-SCs but also distinguished descendents on the basis of cell divisions that occur

after departing the niche. By combining this powerful approach with classical lineage tracing and nucleotide pulse-chase, we discovered that descendents closest to the bulge and that undergo the fewest cell divisions are recycled and contribute to the long-term SC pool that fuels the subsequent hair cycle. Bulge descendents a little further en route become activated SCs of the HG (Greco et al., 2009). Fast-cycling ORS cells nearer to matrix return to the bulge and function, but not as bona fide SCs (Figure S7F). Recycled SCs Make a New Niche It was hitherto unrecognized that HF-SCs that depart their niche in one cycle become CD34+ bulge SCs for the next cycle. Our findings further imply that HF-SCs with fewer divisions are set aside in reserve within the old bulge, whereas those undergoing more divisions are recycled into a new bulge for homeostasis. In this regard, it is intriguing that in hematopoiesis, the most quiescent HSCs participate in injury repair whereas less quiescent ones are used in homeostasis (Foudi et al., 2009; Wilson et al., 2008). It will be interesting in the future to probe deeper into the impact of cell divisions on HF-SC stemness and to explore how melanoblast and other SCs in the bulge migrate and function in this dynamic niche environment. The Origin of the HG Two major theories propose how the HG originates: (1) the lateral disc hypothesis posits that the HG comes from Shh+ cells within matrix (Panteleyev et al., 2001); and (2) the bulge migration hypothesis suggests that HG cells migrate from the bulge at the catagen/telogen transition (Ito et al., 2004; Zhang et al., 2009). Both models are attractive, and although Shh+ matrix cells are not the source of HF-SCs or HG (Greco et al., 2009), we show that cells outside the bulge clearly survive the destructive phase, as initially speculated (Panteleyev et al., 2001). Moreover, our study extends the similarities between HG and bulge (Greco et al., 2009; Ito et al., 2004; Zhang et al., 2009) even though HG appears to be derived from ORS-SCs that exited the bulge during the prior growth phase. Our findings agree with prior studies emphasizing close proximity to DP stimuli in explaining why HG is activated so quickly at the start of a new cycle (Greco et al., 2009). However, our new data suggest that this enhanced sensitivity may also rely upon the intrinsic feature of having traveled further along the lineage and divided 1–23 more than new bulge cells. Moreover, as the ORS mid-zone is at the nexus along the lineage at which long-term self-renewing potential begins to wane, this could explain why HG cells do not sustain stemness in vitro (Greco et al., 2009). Unexpected Origins and Properties of the K6+ Bulge: Insights into Hair Cycle Control Our findings suggest that when catagen is initiated and matrix undergoes apoptosis, proliferative cells in the lower ORS are short-circuited and return to the bulge. This finding helps to resolve the paradox as to why there are fast-cycling cells from anagen that return to the bulge (Jaks et al., 2008). Since these cells do express many HF-SC markers, it is tempting to conclude that they are SCs coming home to roost. However, our data Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc. 101

Figure 7. K6+ Bulge Cells Anchor the Club Hair and Serve as a Signaling Center for the HF-SC Niche (A) Scheme to differentially express DTR in K6+ and CD34+ bulge cells and selectively DT-ablate each layer during extended 2nd telogen. D1 = day 1 of DT injection. Sox9-iDTRhigh was treated with Tamoxifen for 4 days while Sox9-iDTRlow was treated for 2 days (see text). (B) In contrast to K15-iDTR mice, Sox9-iDTR mice lose their hair coat at D6 (anchoring function) and skin turns black at D8, reflective of precocious entry into anagen and confirmed by histology. (C) Quantifications of anagen entry (signaling function). Graph shows #days before HFs entered anagen III for various genotypes/treatments. Horizontal bars denote median. (D and E) DT-treated K15- and Sox9-iDTR mice were given a 1 day BrdU pulse at times indicated prior to immunodetection of apoptotic and proliferation markers on skin sections. In D10 Sox9-iDTR image, K6 marks the anagen companion layer. Scale bars, 30 mm. Graphs quantify %bulges positive for CP3, BrdU, and anagen. Box-and-whisker plots indicate #events per HF/bulge: mid-line, median; box, 25th to 75th percentiles; whiskers, minimum and maximum. (F) Real-time PCR on FACS-isolated populations. Values are normalized to total skin (epidermis+dermis) mRNAs. Bmp4 is higher in dermis (derm) and a6+ cells as previously reported (Plikus et al., 2008). Note high Fgf18 and Bmp6 expression in K6+ bulge cells. (G) Sox9-iDTR mice were treated 3 days with DT and growth factors and then 1 day with BrdU before analysis of proliferation (>24 HFs/experiment in duplicate).

102 Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc.

clearly show that once catagen sets in, the proliferative potential of these cells ceases and cannot be reactivated for either homeostasis or wound repair. The most interesting facet of these surviving lower ORS cells is that despite their failure to regain proliferative potential, they perform essential non-SC functions in the bulge. We posit that their special ability to sense the bulge niche and thereby halt upward protrusion of hair from the skin surface could originate from retention of HF-SC markers when normal lineage progression is bypassed. Additional environmental cues (Figure S7E) along the lineage may endow them with differentiation features that enable their anchorage to the hair shaft and protect the animal against cyclical alopecia. Perhaps even more remarkable is that lower ORS-derived bulge cells act as a signaling center in the niche (Figure 7H). FGF18 and BMP6 were previously identified as autocrine factors able to maintain bulge SC quiescence (Blanpain et al., 2004; Greco et al., 2009). Our findings confirm these data but also point to a paracrine loop that likely transmits a burst of these key factors to bulge SCs at the catagen/telogen transition when the K6+ layer forms. By imposing potent quiescence signals to the niche, the K6 bulge layer counterbalances the activating role provided by the DP at this time. In this way, the K6 bulge layer establishes the need for prolonged crosstalk between HF-SCs and DP to generate the blocking factors that enable HF-SCs to overcome this threshold (Greco et al., 2009). Moreover, at anagen, the DP moves away from the niche toward the matrix but the K6 layer remains with the bulge, thereby launching a regulatory switch between bulge SCs and TA matrix. We posit that the smaller autocrine FGF18/BMP6 signals may further provide a means for HF-SCs along the upper ORS to retain their slow-cycling status until the new bulge is formed. Our studies now pave the way to test such hypotheses in the future. Lessons Learned about Stem Cell Biology By taking advantage of the distinct cycling properties of the HF lineage, we designed unambiguous lineage-tracing experiments to determine the sources and fates of HF-SCs and their descendents throughout homeostasis. By superimposing a temporal series of pulse-chase experiments, we adopted the principles of a movie to monitor HF-SC movement during the hair cycle. This strategy could be useful for many other SCs that display proliferative differences with their progeny. Although the system has its advantages, the timing of labeling can determine whether the slow-cycling SCs retain the label (e.g., Cotsarelis et al., 1990; Morris and Potten, 1999) or their committed fast-cycling progeny do (e.g., Jaks et al., 2008). Our studies also reveal an example of SC progeny that have become irreversibly committed and yet that still express SC markers and home back to their SC niche. These revelations may help to resolve some controversies that have arisen from label-retaining

and conventional lineage-tracing experiments in this and other SC systems. Several other concepts derived from our study could have broad implications for SC biology: First, our data show that a downstream lineage of SCs can be a critical component of the niche microenvironment and regulate the rate of SC divisions. However, our data also illustrate the power of intrinsic factors on stemness. As exemplified by the K6+ bulge layer and in contrast to the Drosophila germline (Brawley and Matunis, 2004; Kai and Spradling, 2004), once cells lose these features, regaining stemness is not assured by returning to an unperturbed niche, nor by depleting the niche of its SCs. Thus, although the niche can markedly affect SC behavior, it may not be sufficient to reset the stemness clock once SCs have passed the point of no return in a lineage. Future studies will be valuable in further defining the distinctions and the parallels among different SCs and their niches. EXPERIMENTAL PROCEDURES Mice and Labeling Experiments Lhx2-EGFP was generated by the GENSAT project (Heintz, 2004). See Extended Experimental Procedures for all other strains used. Tet-Off and Tet-On were activated by continuously feeding mice with Doxy (2 mg/kg) starting at P21 or times specified. CreER was activated by intraperitoneal injection (150 mg/g tamoxifen in corn oil) or topical application (10 mg/ml in ethanol) as specified. CrePGR activation was by topical application of RU486 (1% in ethanol). For 5-Bromo-20 -deoxyuridine (BrdU) pulse-chase experiments, mice were injected intraperitoneally (50 mg/g) (Sigma-Aldrich) and chased for times specified. To test for rare proliferations, injections were supplemented with dietary BrdU water (0.8 mg/ml). Hair Cycle Timing Subdivisions of the hair cycle into 6 anagen and 8 catagen stages were based on Muller-Rover et al. (2001). Because hair cycles vary among strains and sexes, stages instead of mouse ages were usually evaluated. For K5-tTA/ pTRE-H2BGFPK5-tTA (Tet-Off) mice, both are provided. Typically 3–4 mice of matched sex were analyzed. Wounding and Ablation 0.6 cm punch biopsy wounds were created on backs of anesthetized mice. For ablation of bulge cells, K15-CrePGR or Sox9-CreER X Rosa-iDTR mice were first treated topically to induce Cre, then injected intraperitoneally (i.p.) with diphtheria toxin (200 ng DT/injection, Sigma) 13/day for 5 days. Quantifications of H2BGFP Intensities Different stages of Tet-Off H2BGFP HFs were costained for CD34 and imaged by LSM510 laser-scanning confocal (same laser input and gain). Fluorescence intensities were measured using MetaMorph 7 (Universal Imaging Corp.). Cell division numbers were estimated based on H2BGFP epifluorescence intensity (brightest cells assigned 0 divisions). To reflect cell proliferation histories, native H2BGFP intensities were measured. The exception is Figure 5C where signals were enhanced with GFP antibody and laser input and gain were increased as necessary to detect weakly fluorescent cells. Individual data points were plotted, and statistic analyses (Student’s t test) were performed using OriginLab 7.5 or Prism 5 software. Box-and-whisker plots are used to describe the entire population without assumptions on statistical distribution (Schober et al., 2007).

(H) Model summarizes different outcomes of ablating K6+ versus CD34+ bulge cells in telogen. When CD34+ cells are ablated, remaining HF-SCs become activated briefly but then return to quiescence. When K6+ cells are ablated, quiescent signals (FGF18, BMP6) from K6+ bulge are lost, greatly reducing the threshold for anagen activation. The K6+ bulge layer also functions to anchor the club hair. **p < 0.01. Data are mean ± SD.

Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc. 103

SUPPLEMENTAL INFORMATION

Heintz, N. (2004). Gene expression nervous system atlas (GENSAT). Nat. Neurosci. 7, 483.

Supplemental Information includes Extended Experimental Procedures and seven figures and can be found with this article online at doi:10.1016/j.cell. 2010.11.049.

Horsley, V., Aliprantis, A.O., Polak, L., Glimcher, L.H., and Fuchs, E. (2008). NFATc1 balances quiescence and proliferation of skin stem cells. Cell 132, 299–310.

ACKNOWLEDGMENTS We are grateful to colleagues who generously donated mice, especially B. de Crombrugghe (UT MD Anderson) and H. Akiyama (Kyoto University) for sharing Sox9-CreER mice prior to publication; S. Mazel, C. Bare, and RU’s FCRC for FACS sorting; A. North (Bioimaging Resource Center) for advice in image acquisition; Comparative Biology Center (AAALAC-accredited) for health care to our mice; and members of the Fuchs’ lab, in particular: T. Chen, D. Devenport, E. Ezhkova, and B. Keyes for comments on the manuscript; M. Schober for advice on image analyses and quantifications; N. Stokes and D. Oristian for assistance in mouse research. Y.-C. H. is a Starr Stem Cell Scholars Postdoctoral Fellow. This work was supported by grants from NIH (R01AR050452), Starr Foundation, and NYSTEM (N09G074) to E.F., who is an HHMI Investigator. RU FCRC is supported by the Empire State Stem Cell fund through NYSDOH Contract #C023046. Received: May 3, 2010 Revised: October 1, 2010 Accepted: November 17, 2010 Published: January 6, 2011

Ito, M., Kizawa, K., Hamada, K., and Cotsarelis, G. (2004). Hair follicle stem cells in the lower bulge form the secondary germ, a biochemically distinct but functionally equivalent progenitor cell population, at the termination of catagen. Differentiation 72, 548–557. Ito, M., Liu, Y., Yang, Z., Nguyen, J., Liang, F., Morris, R.J., and Cotsarelis, G. (2005). Stem cells in the hair follicle bulge contribute to wound repair but not to homeostasis of the epidermis. Nat. Med. 11, 1351–1354. Jaks, V., Barker, N., Kasper, M., van Es, J.H., Snippert, H.J., Clevers, H., and Toftgard, R. (2008). Lgr5 marks cycling, yet long-lived, hair follicle stem cells. Nat. Genet. 40, 1291–1299. Kai, T., and Spradling, A. (2004). Differentiating germ cells can revert into functional stem cells in Drosophila melanogaster ovaries. Nature 428, 564–569. Lavker, R.M., Sun, T.T., Oshima, H., Barrandon, Y., Akiyama, M., Ferraris, C., Chevalier, G., Favier, B., Jahoda, C.A., Dhouailly, D., et al. (2003). Hair follicle stem cells. J. Investig. Dermatol. Symp. Proc. 8, 28–38. Morris, R.J., and Potten, C.S. (1999). Highly persistent label-retaining cells in the hair follicles of mice and their fate following induction of anagen. J. Invest. Dermatol. 112, 470–475. Morris, R.J., Liu, Y., Marles, L., Yang, Z., Trempus, C., Li, S., Lin, J.S., Sawicki, J.A., and Cotsarelis, G. (2004). Capturing and profiling adult hair follicle stem cells. Nat. Biotechnol. 22, 411–417.

REFERENCES

Morrison, S.J., and Kimble, J. (2006). Asymmetric and symmetric stem-cell divisions in development and cancer. Nature 441, 1068–1074.

Barker, N., van Es, J.H., Kuipers, J., Kujala, P., van den Born, M., Cozijnsen, M., Haegebarth, A., Korving, J., Begthel, H., Peters, P.J., and Clevers, H. (2007). Identification of stem cells in small intestine and colon by marker gene Lgr5. Nature 449, 1003–1007.

Muller-Rover, S., Handjiski, B., van der Veen, C., Eichmuller, S., Foitzik, K., McKay, I.A., Stenn, K.S., and Paus, R. (2001). A comprehensive guide for the accurate classification of murine hair follicles in distinct hair cycle stages. J. Invest. Dermatol. 117, 3–15.

Blanpain, C., Lowry, W.E., Geoghegan, A., Polak, L., and Fuchs, E. (2004). Self-renewal, multipotency, and the existence of two cell populations within an epithelial stem cell niche. Cell 118, 635–648.

Nguyen, H., Rendl, M., and Fuchs, E. (2006). Tcf3 governs stem cell features and represses cell fate determination in skin. Cell 127, 171–183.

Brawley, C., and Matunis, E. (2004). Regeneration of male germline stem cells by spermatogonial dedifferentiation in vivo. Science 304, 1331–1334. Brinster, R.L., and Avarbock, M.R. (1994). Germline transmission of donor haplotype following spermatogonial transplantation. Proc. Natl. Acad. Sci. USA 91, 11303–11307. Buch, T., Heppner, F.L., Tertilt, C., Heinen, T.J., Kremer, M., Wunderlich, F.T., Jung, S., and Waisman, A. (2005). A Cre-inducible diphtheria toxin receptor mediates cell lineage ablation after toxin administration. Nat. Methods 2, 419–426. Cao, Y.A., Wagers, A.J., Beilhack, A., Dusich, J., Bachmann, M.H., Negrin, R.S., Weissman, I.L., and Contag, C.H. (2004). Shifting foci of hematopoiesis during reconstitution from single stem cells. Proc. Natl. Acad. Sci. USA 101, 221–226.

Nowak, J.A., Polak, L., Pasolli, H.A., and Fuchs, E. (2008). Hair follicle stem cells are specified and function in early skin morphogenesis. Cell Stem Cell 3, 33–43. Oshima, H., Rochat, A., Kedzia, C., Kobayashi, K., and Barrandon, Y. (2001). Morphogenesis and renewal of hair follicles from adult multipotent stem cells. Cell 104, 233–245. Panteleyev, A.A., Jahoda, C.A., and Christiano, A.M. (2001). Hair follicle predetermination. J. Cell Sci. 114, 3419–3431. Paus, R., and Cotsarelis, G. (1999). The biology of hair follicles. N. Engl. J. Med. 341, 491–497. Plikus, M.V., Mayer, J.A., de la Cruz, D., Baker, R.E., Maini, P.K., Maxson, R., and Chuong, C.M. (2008). Cyclic dermal BMP signalling regulates stem cell activation during hair regeneration. Nature 451, 340–344. Rendl, M., Lewis, L., and Fuchs, E. (2005). Molecular dissection of mesenchymal-epithelial interactions in the hair follicle. PLoS Biol. 3, e331.

Claudinot, S., Nicolas, M., Oshima, H., Rochat, A., and Barrandon, Y. (2005). Long-term renewal of hair follicles from clonogenic multipotent stem cells. Proc. Natl. Acad. Sci. USA 102, 14677–14682.

Rochat, A., Kobayashi, K., and Barrandon, Y. (1994). Location of stem cells of human hair follicles by clonal analysis. Cell 76, 1063–1073.

Cotsarelis, G., Sun, T.T., and Lavker, R.M. (1990). Label-retaining cells reside in the bulge area of pilosebaceous unit: implications for follicular stem cells, hair cycle, and skin carcinogenesis. Cell 61, 1329–1337.

Schober, M., Raghavan, S., Nikolova, M., Polak, L., Pasolli, H.A., Beggs, H.E., Reichardt, L.F., and Fuchs, E. (2007). Focal adhesion kinase modulates tension signaling to control actin and focal adhesion dynamics. J. Cell Biol. 176, 667–680.

Foudi, A., Hochedlinger, K., Van Buren, D., Schindler, J.W., Jaenisch, R., Carey, V., and Hock, H. (2009). Analysis of histone 2B-GFP retention reveals slowly cycling hematopoietic stem cells. Nat. Biotechnol. 27, 84–90. Fuchs, E. (2009). The tortoise and the hair: Slow-cycling cells in the stem cell race. Cell 137, 811–819. Greco, V., Chen, T., Rendl, M., Schober, M., Pasolli, H.A., Stokes, N., Dela Cruz-Racelis, J., and Fuchs, E. (2009). A two-step mechanism for stem cell activation during hair regeneration. Cell Stem Cell 4, 155–169.

104 Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc.

Soeda, T., Deng, J.M., de Crombrugghe, B., Behringer, R.R., Nakamura, T., and Akiyama, H. (2010). Sox9-expressing precursors are the cellular origin of the cruciate ligament of the knee joint and the limb tendons. Genesis 10. 1002/dvg.20667. Spradling, A.C., Nystul, T., Lighthouse, D., Morris, L., Fox, D., Cox, R., Tootle, T., Frederick, R., and Skora, A. (2008). Stem cells and their niches: integrated units that maintain Drosophila tissues. Cold Spring Harb. Symp. Quant. Biol. 73, 49–57.

Taylor, G., Lehrer, M.S., Jensen, P.J., Sun, T.T., and Lavker, R.M. (2000). Involvement of follicular stem cells in forming not only the follicle but also the epidermis. Cell 102, 451–461.

Waghmare, S.K., Bansal, R., Lee, J., Zhang, Y.V., McDermitt, D.J., and Tumbar, T. (2008). Quantitative proliferation dynamics and random chromosome segregation of hair follicle stem cells. EMBO J. 27, 1309–1320.

Tumbar, T., Guasch, G., Greco, V., Blanpain, C., Lowry, W.E., Rendl, M., and Fuchs, E. (2004). Defining the epithelial stem cell niche in skin. Science 303, 359–363.

Wilson, A., Laurenti, E., Oser, G., van der Wath, R.C., Blanco-Bose, W., Jaworski, M., Offner, S., Dunant, C.F., Eshkind, L., Bockamp, E., et al. (2008). Hematopoietic stem cells reversibly switch from dormancy to self-renewal during homeostasis and repair. Cell 135, 1118–1129.

Unna, P.G. (1876). Beitrage zur histologie und entwicklungsgeschichte der menschlichen oberhaut und ihrer anhangsgebilde. Arch. Mikroskop. Anat. Entwicklungsmech. 12, 665–741.

Zhang, Y.V., Cheong, J., Ciapurin, N., McDermitt, D.J., and Tumbar, T. (2009). Distinct self-Renewal and differentiation phases in the niche of infrequently dividing hair follicle stem cells. Cell Stem Cell 5, 267–278.

Cell 144, 92–105, January 7, 2011 ª2011 Elsevier Inc. 105

Presenilin-Dependent Receptor Processing Is Required for Axon Guidance Ge Bai,1 Onanong Chivatakarn,1 Dario Bonanomi,1 Karen Lettieri,1 Laura Franco,1 Caihong Xia,2 Elke Stein,3 Le Ma,2 Joseph W. Lewcock,1,4 and Samuel L. Pfaff1,* 1Howard Hughes Medical Institute and Gene Expression Laboratory, The Salk Institute for Biological Studies, 10010 North Torrey Pines Road, La Jolla, CA 92037, USA 2Zilkha Neurogenetic Institute, Department of Cell and Neurobiology, Keck School of Medicine, University of Southern California, 1501 San Pablo Street, Los Angeles, CA 90089, USA 3Department of Molecular, Cellular, and Developmental Biology, Yale University, 266 Whitney Avenue, New Haven, CT 06520, USA 4Present Address: Department of Neurobiology, Genentech, Inc., 1 DNA Way, South San Francisco, CA 94080, USA *Correspondence: [email protected] DOI 10.1016/j.cell.2010.11.053

SUMMARY

The Alzheimer’s disease-linked gene presenilin is required for intramembrane proteolysis of amyloidb precursor protein, contributing to the pathogenesis of neurodegeneration that is characterized by loss of neuronal connections, but the role of Presenilin in establishing neuronal connections is less clear. Through a forward genetic screen in mice for recessive genes affecting motor neurons, we identified the Columbus allele, which disrupts motor axon projections from the spinal cord. We mapped this mutation to the Presenilin-1 gene. Motor neurons and commissural interneurons in Columbus mutants lacking Presenilin-1 acquire an inappropriate attraction to Netrin produced by the floor plate because of an accumulation of DCC receptor fragments within the membrane that are insensitive to Slit/Robo silencing. Our findings reveal that Presenilin-dependent DCC receptor processing coordinates the interplay between Netrin/DCC and Slit/Robo signaling. Thus, Presenilin is a key neural circuit builder that gates the spatiotemporal pattern of guidance signaling, thereby ensuring neural projections occur with high fidelity. INTRODUCTION Normal behavioral functions rely on complex neural circuits comprised of large ensembles of precisely connected neurons. During embryonic development, the growth of axonal and dendritic processes is tightly regulated to ensure that proper synaptic connections are formed. In the mature nervous system, the growth potential of neurons is generally far less robust; however, some regions of the CNS display ongoing neurogenesis and high levels of plasticity throughout adulthood (Kandel et al., 2000). Although abnormal circuit development and neurodegenerative diseases are both detrimental to the performance of the 106 Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc.

nervous system, our understanding of the pathways that link the establishment and maintenance of neural circuitries is limited. Considerable progress has been made in identifying extracellular cues that influence axonal growth cone dynamics, including members of the four classic guidance families: Netrins, Slits, Semaphorins and Ephrins; and their respective neuronal receptors: DCC, Robo, Neuropilins/Plexins and Ephs (Dickson, 2002). Although other guidance factors and receptors continue to be identified, a remarkable feature of these signaling proteins is their recurrent usage throughout the developing nervous system. Moreover, projection neurons extend axons over long distances to their final targets with a series of intermediate guidance cues along their pathway. While this strategy simplifies the problem of locating targets separated by vast distances, it imposes the need for dynamic control of signal responsiveness so that axons do not stall at their intermediate targets (O’Donnell et al., 2009; Tessier-Lavigne and Goodman, 1996; Yu and Bargmann, 2001). The expression and localization of guidance cues and receptors is exquisitely tailored to allow growth cones to rapidly switch their responsiveness at specific times and places throughout development. Several mechanisms have been identified to ensure the correct presentation and receipt of guidance signals, including regulation of receptor membrane trafficking, endocytosis, proteolytic processing, and localized mRNA transport and translation (Brittis et al., 2002; O’Donnell et al., 2009). Another important strategy for modulating axonal responsiveness derives from the coordinated interplay between different guidance-signaling pathways. For example, Netrin is a chemoattractant for commissural axons until they reach the floor plate (FP) at the midline, where they encounter repulsive Slit ligands. Here Robo becomes activated, triggering repulsion from the midline and silencing the attractive response toward Netrin through direct receptor interaction (Stein and Tessier-Lavigne, 2001). An important remaining challenge has been to understand how different regulatory strategies are coordinated to control the spatial and temporal activity of guidance signaling during embryonic development. To identify new modulatory components that regulate the spatiotemporal pattern of axon guidance signaling, we performed an ENU mutagenesis screen in transgenic mice using a GFP reporter to visualize embryonic motor neurons (MNs)

(Lewcock et al., 2007). These cells develop within the spinal cord then grow axons into the periphery in a highly stereotypical pattern to form connections with muscles in order to relay locomotor commands. Here, we characterize the Columbus mutant, which exhibits a motor axon midline-crossing phenotype, whereby numerous motor axons fail to even exit the neural tube. The Columbus mutation disrupts the expression of Presenilin-1 (PS1), which encodes a 467-residue protein with a nine-transmembrane domain topology. PS1 is an essential component of the g-secretase complex that cleaves amyloidb precursor protein (APP), leading to the formation of toxic plaques that disrupt neuronal connections and contribute to Alzheimer’s disease (AD) pathogenesis. Although multiple embryonically expressed genes including the Notch and DCC receptors have been identified as PS1 substrates, it remains unclear whether PS1 has a role in establishing neuronal connections during development (Parks and Curtis, 2007). In this study, we show that PS1 function is required to control the spatiotemporal pattern of axonal responses to Netrin by coordinating the activity of different signaling pathways. Our findings reveal an important molecular link between neural circuit formation and disorders causing degeneration. RESULTS Columbus Mutants Exhibit Multiple Errors in Motor Axon Guidance To identify genes involved in motor axon navigation, we conducted a forward genetic screen in GFP reporter mice mutagenized with N-ethyl-N-nitrosourea (ENU) (Lewcock et al., 2007). The embryonic MN-specific transgenic reporter ISLMN:GFP-F was crossed with heterozygous ENU mutants and the offspring were intercrossed to generate homozygous mutants. We identified a mutant we called Columbus that displays a severe defect in ventral root formation (Figures S1A–S1D available online) (Lewcock et al., 2007). Normally, motor axons preferentially grow through the anterior half of the somite, whereas Columbus motor axons exhibited no preference for the anterior- or posterior-somite, leading to a loss of segmentally organized ventral roots. Transverse sections of Columbus mutants also revealed that a subset of MNs had extended axons into the FP rather than out the ventral roots (Figures 1A, 1B, 1D, and 1E). To characterize the midline axon growth defect in more detail, we imaged spinal cords using an open book preparation at E13 and found MN misprojections at all levels of the spinal cord (Figures 1G, 1H, 1J, and 1K and data not shown). Some motor axons crossed the midline and projected to the contralateral side while others stalled in the FP and formed bundles (Figures S1E and S1F). Likewise, midline motor axon guidance defects were observed with other MN reporters such as Hb9:GFP (Figures S1G and S1H). Presenilin-1 (PS1) Is Mutated in Columbus SNP markers for the mutagenized DBA mouse strain were used to map the Columbus mutation to a 16.7 Mb segment on chromosome 12. The ventral root segmentation and midline-crossing motor axon phenotypes did not segregate during the outcrosses. Next, genomic DNA for candidate genes was sequenced. We

identified a Thymine to Adenine base conversion in intron 11 of the mouse presenilin-1 gene (PS1), which has 12 exons and 11 introns in the genome (Figure 1M and Figures S1I and S1J). To examine how this mutation affects PS1 expression, we first performed RT-PCR on PS1 mRNA from Columbus mutants using primers that flank the 11th intron. We found that the point mutation shifted the PCR product from 164 bp to 401 bp in Columbus mutants (Figure 1N). Sequencing Columbus mutant PS1 transcripts showed that the T/A base conversion disrupted the normal splice site for the 12th exon of PS1. The mutation unmasked a cryptic splice acceptor within intron 11 that resulted in a 237 bp insertion, which introduces two premature stop codons (Figure 1M). Western-blot analysis revealed that PS1 protein was undetectable in Columbus mutants using both antiN and -C-terminal antibodies. These findings demonstrate that the Columbus mutation alters PS1 splicing, which severely disrupts PS1 protein expression (Figure 1O). To confirm that the motor axon guidance defects observed in Columbus mutants were due to PS1 rather than another mutation, we crossed the ISLMN:GFP-F MN reporter into PS1 knockout mice (Shen et al., 1997). Embryos with a targeted disruption of the PS1 gene displayed a similar combination of pathfinding errors to those observed in the Columbus mutant, including failure to form discrete ventral roots and midline-crossing of motor axons (Figures 1C, 1F, 1I, 1L; Figures S1K–S1P; and data not shown). Next we analyzed the distribution of PS1 protein in mouse embryos using immunofluorescence and found it was expressed at high levels by MNs and interneurons in the spinal cord as well as peripheral tissues (Figures S2A–S2C). In contrast, progenitor cells in the ventricular zone expressed much lower levels of PS1. Within MNs PS1 was detected as the cells became postmitotically born and began axonogenesis, and both cell bodies and axons were labeled. PS1 Function in Cell Fate Specification Based on the expression of PS1, it remained unclear whether it was required for spinal neuron differentiation, motor axon guidance, and/or the proper development of peripheral tissues. Notch-delta signaling is required for both spinal cord neurogenesis and somite development, and cleavage of the Notch receptor by g-secretase is required to generate the notch intracellular domain involved in gene regulation (Selkoe and Kopan, 2003). To determine which tissues require PS1 activity, we crossed floxed PS1 mice with Nestin-Cre transgenic mice to generate a neural cell-specific PS1 conditional knockout (PS1 cKO) mouse. In PS1 cKO embryos, the segmentally repeated ventral roots developed normally (data not shown), whereas the inappropriate midline-crossing motor axon tracts still formed (Figures S2D and S2E). These data suggest that nonneuronal expression of PS1 is required for proper ventral root formation, likely because PS1 is necessary for anteroposterior somite patterning (Shen et al., 1997). In contrast, the midline motor axon-crossing defect in PS1 mutants arises from defects within the neural tube. Next, we examined progenitor cell growth and neural differentiation in developing spinal cords from Columbus mutants using immunofluorescent staining. We found that loss of PS1 function did not change progenitor cell patterning, MN specification, or MN subtype diversification (Figures S2F–S2K), presumably because PS2 can partially compensate for PS1 inactivation in Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc. 107

Ctrl ISLMN:GFP-F

A

B

C

D

E

F

FP

FP

Col ISLMN:GFP-F

PS1 KO ISLMN:GFP-F

FP Ctrl ISLMN:GFP-F

G

I

H

K

J

WT PS1

M

L

Col Mutant PS1

TCTGT

DNA

PS1 KO ISLMN:GFP-F

Col ISLMN:GFP-F

Exon 11

mRNA

Exon 12

Exon 11 Exon 12

WT Col

N

TCAGT Exon 11

Exon 11

164bp

401bp

Exon 12

164bp

O

Exon 12

WT Col

401bp

(kDa) 51

FL

Exon 11

Anti-PS1 N-ter

39

28 a

19

NTF

CTF

Anti-PS1 C-ter

Anti-GAPDH Exon 12

108 Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc.

maintaining Notch signaling (Donoviel et al., 1999). Taken together, our results indicate that the midline motor axon growth defect is due to a neural-tube-intrinsic function of PS1 that is unrelated to MN fate specification. Motor Axons Grow Toward the Floor Plate in PS1 Mutants Since the inappropriate midline-growth of motor axons was intrinsic to the neural tube, we tested whether the FP and/or MNs behaved abnormally in PS1 mutants. The FP and GFPlabeled MNs were dissected from ISLMN:GFP-F E12 embryos. Tissue was taken separately from either PS1 knockout mutants or controls (a mix of heterozygous and wild-type embryos) and cocultured in 3D collagen/matrigel matrices for 24 hr. A semiquantitative assessment of axon outgrowth was performed by counting GFP-positive neurite numbers on the side facing toward (proximal) and away (distal) from the FP (Figure 2A). In the presence of control FP explants, MNs from wild-type embryos extended more axons from the distal side of the explant, indicating the FP either represses or repels motor axon growth (Figures 2C and 2G). Likewise, control MNs failed to extend axons toward FP explants dissected from PS1 mutant embryos (Figures 2E and 2G). On the other hand, MNs dissected from PS1 mutant embryos extended numerous axons toward FP explants regardless of whether the FP was dissected from control or PS1-deficient embryos (Figures 2D, 2F, and 2G). Thus, PS1 inactivation in FP cells does not alter motor axon growth compared to controls, whereas the loss of PS1 function in neural tissue leads to aberrant growth of MNs in explant cultures. These findings indicate that PS1-deficient MNs either lose responsiveness to a chemorepellent from the FP or acquire responsiveness to a chemoattractant. PS1-Deficient Motor Neurons Are Attracted to Netrin-1 To examine whether PS1 mutant MNs displayed abnormal responsiveness to guidance signals, we cocultured GFP-labeled MN explants from ISLMN:GFP-F embryos with Cos cell aggregates that had been transfected with cDNAs encoding known guidance signals. Since Semaphorins are well-established repellents in the midline, we tested whether MN responsiveness to Sema3A was altered in PS1 mutants. We found that both control and PS1-deficient MNs were repelled by Cos cell aggregates expressing Sema3A after 24 hr in culture (Figures 2H–2L). Neither wild-type nor PS1-deficient spinal MNs from E12 embryos were repelled by Slit2-secreting cell clusters (Figure 2L),

suggesting that midline-growth of PS1 mutant MNs is not due to loss of Slit responsiveness. To assess whether MN attraction to guidance signals was altered, we shortened our coculture assay to 15–18 hr to minimize axonal outgrowth in the absence of growth-promoting signals (Figure 2B). We first tested Shh and Netrin-1, which are both known chemoattractants expressed by the FP (Charron et al., 2003; Kennedy et al., 1994). Neither control nor PS1-deficient MNs were attracted to Cos cells secreting Shh (Figures 2O and 2P). Likewise, Netrin-1 failed to promote axonal outgrowth from control MNs (Figure 2Q). Interestingly, there was a marked increase in axonal outgrowth from PS1 mutant explants when cocultured with Netrin-1 secreting Cos cells (Figure 2R). To exclude the possibility that Netrin-1 acted indirectly by altering the Cos cell aggregates, we tested whether purified recombinant Netrin-1 was active. We found that MN explants from control embryos were unresponsive, whereas explants from PS1 mutants extended axons in a dose-dependent manner to bath-applied recombinant Netrin-1 (Figure 2T). These results suggest that PS1 mutant MNs acquire abnormal responsiveness to Netrin-1. Inhibition of Netrin-1/DCC Signaling Rescues the Midline-Motor Axon Phenotype To test whether abnormal Netrin responsiveness caused motor axon guidance defects, we crossed PS1 knockout mice to Netrin-1 hypomorphic mutants to generate PS1/Netrin-1 double-mutant embryos (Serafini et al., 1996). We found that the inappropriate growth of motor axons within the FP was significantly reduced by disruption of Netrin-1 expression (Figures 3D–3I). Next, we tested whether DCC was required for PS1-deficient MN chemoattraction to the FP. The axon outgrowth from PS1 mutant MNs toward FP explants was inhibited by function blocking antibodies to DCC (Figures 3A– 3C). Furthermore, we found that MNs in PS1/DCC doubleknockout embryos avoided the FP (Figures 3J–3O). These data suggest that MNs lacking PS1 become attracted to Netrin-1 because of abnormal signaling from the DCC receptor, which leads to inappropriate motor axon growth into the FP. Inhibition of g-Secretase Activity Confers Responsiveness to Netrin PS1 encodes the catalytic component of the multisubunit g-secretase, but it was unclear whether protease activity was required for regulating MN axon growth. We added g-secretase antagonist L-685458 to wild-type explant cultures and assayed motor

Figure 1. Columbus Mutants Display Midline Motor Axon Guidance Defects (A–F) Motor axons in transverse sections of E12.5 mouse embryos at the brachial level labeled with ISLMN:GFP-F transgenic reporter. Boxed regions in (A–C) are enlarged in (D–F), respectively. n > 8 embryos for each genotype. (G–L) Flat-mount images of E13 mouse spinal cords at lumbar levels with anterior on top. Boxed regions in (G–I) are enlarged in (J–L), respectively. Dotted line marks medial edge of MN cell bodies. Note that motor columns are slightly disorganized at the lumbar level of PS1 mutants, leading to an increased distance between the motor columns. n > 10 embryos for each genotype. (M) Schematic of Columbus mutation in PS1 gene. (N) RT-PCR analysis of PS1 mRNA using primers flanking intron 11 (arrowheads in M). (O) Western-blot analysis of PS1 protein in Columbus mutants. Full-length PS1 protein is proteolytically processed in vivo to a 30 kDa N-terminal fragment (NTF) and a 20 kDa C-terminal fragment (CTF). Antibody against PS1 N terminus recognized the full-length holoprotein (FL) and NTF in wild-type but not in the Columbus mutant. Likewise, an antibody against the C terminus of PS1 recognized the 20 kDa CTF in wild-type but not Columbus mutants. GAPDH was used as an internal control. The scale bars represent 100 mm (A–C), 40 mm (D–F), 75 mm (G–I), and 30 mm (J–L). See also Figure S1.

Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc. 109

A

Repulsion Assay Proximal

FP

Distal

PS1 KO MN

C

G

P:D ratio

Ctrl FP

2.5

*

*

2.0 1.5 1.0 0.5

PS1 KO FP

E

F

H

0.0 Ctrl FP Ctrl FP KO FP KO FP + + + + Ctrl MN KO MN Ctrl MN KO MN

PS1 KO MN

Ctrl MN

L

I

2.0 Ctrl MN

Sema3A

M

*

*

KO MN

1.0 0.5

K

0.0 Ctrl

PS1 KO MN

Ctrl MN

1.5

N Ctrl

Sema3A

Slit2

5

S

Ctrl MN

KO MN

*

4

P:D ratio

Sema3A

P:D ratio

Ctrl

Ctrl

J

Ctrl

Distal

MN

Cell Agg

D

Ctrl FP PS1 KO FP

Proximal

MN

Ctrl MN

Figure 2. PS1-Deficient Motor Neurons Are Attracted to Netrin

Attraction Assay

B

3 2 1 0

P Shh Netrin-1

Netrin-1

T

R

Q

Ctrl

Shh

Netrin-1

15

Neurite numbers relative to ctrl

Shh

O

Ctrl MN

KO MN

12

*

9

*

6 3 0

Netrin-1 (ng/ml)

0

axon responses to recombinant or Cos cell produced Netrin-1. Netrin-mediated motor axon growth was stimulated when L685458 was added to the medium (Figures 4A and 4B and 110 Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc.

50

(A) Schematic of motor explant repulsion assay. MN explants are cocultured with floor plate or Cos cell aggregates in 3D collagen/matrigel matrices for 24 hr. Motor axons are visualized with the transgenic ISLMN:GFP-F reporter. (B) Schematic of motor explant attraction assay. MN explants are cocultured with Cos cell aggregates (Cell Agg) in 3D collagen/matrigel matrices for 15 hr, which is insufficient time for motor axon outgrowth unless the Cos cells express a chemoattractant. (C–F) Motor explant repulsion assay. GFP-labeled mouse motor explants were cocultured with FP (FP was to the left of explants). FPs and motor explants from PS1 heterozygous or wild-type littermates were used as controls. (G) Histogram showing quantification (proximal:distal [P:D] ratio) of outgrowth from explants in culture with FPs. n = 8 (Ctrl FP + Ctrl MN), n = 7 (Ctrl FP + KO MN), n = 7 (KO FP + Ctrl MN), n = 5 explants (KO FP + KO MN). Data are presented as the mean ± standard error of the mean (SEM) (*p < 0.05). The following abbreviations are used: Ctrl, control; KO, knockout. (H–K) Motor explant repulsion assay. GFP-labeled mouse motor explants were cocultured with Cos cell aggregates and transfected as indicated (cell aggregates were to the left of explants). (L) P:D ratio of outgrowth from explants in the presence of cell aggregates. For control (wildtype/heterozygous [WT/Het]) motor explants, n = 13 (control), n = 10 (Sema3A), n = 17 (Slit2); for PS1 KO motor explants, n = 9 (control), n = 11 (Sema3A), n = 9 (Slit2). Data are presented as the mean ± SEM (*p < 0.05). (M–R) Motor explant attraction assay. GFPlabeled mouse motor explants were cocultured with Cos cell aggregates transfected as indicated (cell aggregates were to the left of explants). (S) P:D ratio of outgrowth from explants in the presence of Cos cell aggregates. For control (WT/Het) motor explants, n = 11 (control), n = 10 (Shh), n = 12 (Netrin); for PS1 KO motor explants, n = 15 (control), n = 8 (Shh), n = 10 (Netrin). Data are presented as the mean ± SEM (*p < 0.05). (T) Histogram showing quantification (neurite numbers relative to control) of outgrowth from motor explants in the presence of recombinant Netrin-1. For control (WT/Het) motor explants, n = 21 (0 ng/ml), n = 25 (50 ng/ml), n = 19 (250 ng/ml); for PS1 KO motor explants, n = 13 (0 ng/ml), n = 15 (50 ng/ml), n = 16 (250 ng/ml). Data are presented as the mean ± SEM (*p < 0.05). See also Figure S2.

250

data not shown), indicating that the catalytic activity of PS1 is normally required to prevent MNs from responding to Netrin-1. Furthermore, the finding that MNs become responsive to

B

C

PS1 KO MN

PS1 KO MN + DCC Ab

Ctrl FP

Ctrl FP

Ctrl

2.0 1.5

P:D ratio

A

1.0

*

0.5 0.0

Ctrl DCC Ab

E13 ISLMN:GFP-F Netrin-/-PS1+/-

Netrin-/-PS1-/-

Netrin+/+PS1-/-

D

E

F

G

H

I

E13 ISLMN:GFP-F DCC-/-PS1+/-

J

M

DCC+/-PS1-/-

K

N

DCC-/-PS1-/-

L

O

Figure 3. Inhibition of Netrin/DCC Signaling Rescues the MidlineCrossing Phenotype (A and B) Motor explant repulsion assay. GFP-labeled PS1 KO motor explants were cocultured with control (WT/Het) FPs in the presence of (A) vehicle control or (B) DCC function-blocking antibody. (C) P:D ratio of outgrowth from explants in the presence of FPs. n = 9 (control), n = 7 (DCC Ab). Data are presented as the mean ± SEM (*p < 0.05). (D–O) Flat-mount images of GFP-positive MNs in E13 lumbar mouse spinal cords. Boxed regions in (D–F) and (J–L) are enlarged in (G–I) and (M–O), respectively. Anterior is to the top for all the images. Note that motor columns are slightly disorganized in PS1/Netrin-1 and PS1/DCC double mutants, similar to those observed in PS1 single mutants. At least three embryos were assayed for each genotype. The scale bars represent 50 mm (D–F), 30 mm (G–I), 50 mm (J–L), and 30 mm (M–O).

Netrin-1 following the acute inhibition of PS1 activity with L-685458 suggests that the axon guidance defects in PS1 mutants are not indirectly due to developmental defects that arise from the chronic loss of g-secretase activity throughout embryonic development. Next we tested whether isolated MNs lacking g-secretase activity turn in response to Netrin-1. We established a Dunn chamber assay to create a Netrin-1 gradient (Yam et al., 2009). In this assay, the outer annular well was filled with media containing Netrin-1, and the inner well was filled with control media (Figure 4D). A gradient was formed over the annular bridge through diffusion of the Netrin from the outer to the inner well that was stable for many hours (Yam et al., 2009) (data not shown). ISLMN:GFP-positive (wild-type) chick MNs were dissociated and cultured on a coverslip, which was then inverted over the Dunn chamber (Figures 4D and 4E). In the control condition, the direction of motor axonal growth remained unchanged (Figure 4F). In contrast, when g-secretase inhibitor L-685458 was added axons turned toward the source of Netrin-1 (Figure 4G). To quantify the extent of turning, we measured the initial angle (a), defined as the angle between the initial orientation of the axon and the gradient, and the angle turned (b), defined as the angle between the initial and final trajectories of the axon (positive for turns up the gradient and negative for turns down the gradient) (Yam et al., 2009) (Figure 4H). Scatter plots of these angles showed that for the control condition, no net turning occurred. However, in the presence of g-secretase inhibitor, there was a significant bias toward positive turning angles (Figures 4I and 4J). Together, these data indicate that g-secretase activity is required cell autonomously for newly generated MNs to prevent the Netrin receptor expressed by MNs from responding to Netrin-1 as a chemoattractant. DCC Stubs Promote Axon Growth Next we explored how g-secretase was involved in regulating MN responsiveness to Netrin. Previous studies have found that DCC is the target of protease cleavage (Taniguchi et al., 2003). First, DCC is cleaved by metalloprotease(s) that lead to shedding of the ectodomain segment, generating a membrane-tethered DCC stub. This DCC stub is the substrate of g-secretase, which releases the intracellular domain (ICD) from the membrane (Figure 5A) (Taniguchi et al., 2003). Although the full-length DCC receptor is typically viewed as the primary Netrin-signaling component, it has been shown that the DCC stub and DCCICD also have signaling properties (Gitai et al., 2003; Parent et al., 2005; Taniguchi et al., 2003). We found that the levels of DCC stubs and DCC-ICD fragments were very low in wild-type neurons, whereas DCC stubs accumulated to high levels in PS1 mutants (Figures 5B and 5C and data not shown). As expected, g-secretase inhibition also caused the DCC stub to accumulate (see below). To examine whether cleavage of DCC played a role in Netrin signaling in MNs we performed collagen assays with wild-type MN explants and Netrin-1-producing cells in the presence or absence of metalloprotease and/or g-secretase inhibitors. Application of metalloprotease inhibitor GM6001 failed to stimulate motor axon outgrowth, while the g-secretase inhibitor L-685458 caused motor axons to become responsive to Netrin-1 (Figures 5D–5F and 5H). Interestingly, the Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc. 111

B

MN

+ γ-sec inh

FP

FP

+DMSO

Figure 4. Inhibition of g-Secretase Activity Switches MNs to a Netrin-Responsive State

1.5

C

MN

P:D ratio

A

*

1.0 0.5 0.0

DMSO γ-sec inh

E

D

coverslip

coverslip

Netrin

bridge (gradient) outer well inner well (no netrin) (netrin) inner well outer well bridge

F Netrin

0.5 h

1.5 h

DMSO

0h

ISLMN:GFP 0h

0.5 h

ISLMN:GFP

J

α

-5 0

β 5 10 15 20

α = initial angle α 175 5 β = angle turned

100

100

50

50

0

0

-50

60 120 180-50

-100

-100

-150

-150

γ-sec inh

60 120 180

β > 0 for turns up gradient β < 0 for turns down gradient

addition of GM6001 to the culture blocked the ability of L-685458 to trigger MN chemoattraction to Netrin-1, indicating that metalloprotease-mediated cleavage is a prerequisite for acquiring Netrin responsiveness (Figures 5G and 5H). Thus, full-length DCC is unlikely to be sufficient to cause MN chemoattraction. Taken together, these data suggest that DCC in MNs is normally the target of a sequential protease pathway. Blocking g-secretase activity should have two effects on DCC processing: (1) cause an accumulation of membrane-tethered DCC stubs, and (2) reduce the generation of DCC-ICD fragments (Figure 5A and 5B). To test whether one or both of these DCC fragments influenced motor axon growth, we made two constructs, one containing the intracellular domain of DCC (DCC-ICD) and the other containing a myristoylated form of the intracellular domain of DCC (Myr-DCC-ICD) to mimic membrane-tethered DCC stubs. First, Myr-DCC-ICD was 112 Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc.

40

*

30 20 10 0 -10

in h

0

150

SO

attraction

5

repulsion

10

Angle turned ( )

150

DMSO

coelectroporated into Hamburger-Hamilton (HH) stage 14 chick embryo neural tubes with a MN-specific reporter Hb9:DsRed. After 2 days incubation, the spinal cords were flat mounted and imaged. We found that the Myr-DCC-ICD construct mimicking DCC stubs induced DsRed-positive motor axons to cross the midline and project contralaterally like MNs in PS1 mouse mutants (Figures 5I and 5J). In contrast, expression of: (1) DCC full-length receptor, (2) the DCC extracellular domain, and (3) the DCC-ICD each failed to induce motor axon growth toward the FP (Figures 5K–5M). Next we tested whether a loss of DCC-ICD signaling caused MNs lacking g-secretase activity to become responsive to Netrin-1. MN explants were dissected from Hb9:DsRed electroporated chick embryos and cocultured with Cos cells expressing Netrin-1. Wild-type DsRed-positive MNs were not attracted to Netrin-1, whereas explants treated with the g-secretase inhibitor L-685458 became Netrin-1 responsive (Figures 5N, 5P, and 5R). This finding suggests that Netrin receptor regulation by g-secretase is evolutionarily conserved in chick. Likewise, MNs were γ-s ec

I 20

DM

H 15

1.5 h

Mean angle turned ( º )

Netrin

γ-sec inh

G

(A and B) Motor explant repulsion assay. GFPlabeled motor explants were cocultured with FPs in the presence of (A) DMSO vehicle or (B) g-secretase inhibitor L-685458. (C) P:D ratio of outgrowth from explants in the presence of FPs. n = 7 (DMSO), n = 9 (g-sec inh). Data are presented as the mean ± SEM (*p < 0.05). (D) Side view and (E) top view schematic of an assembled Dunn chamber. ISLMN:GFP-labeled chick MNs cultured on a coverslip have been inverted over the chamber. The inner well is filled with control media, whereas the outer well is filled with media containing 50 ng/ml of Netrin-1. A gradient forms across the annular bridge because of diffusion of Netrin-1 from the outer to the inner well. (F) In the presence of DMSO, growing chick MNs did not change their trajectories in the Netrin gradient. (G) In the presence of g-secretase inhibitor, motor axons turned toward increasing concentrations of Netrin. Note that all images have been rotated such that the gradient increases along the y axis. The scale bar represents 10 mm. (H) Definition of the initial angle, a, the angle between the initial axon position and the gradient, and angle turned, b, the angle between the vectors representing the initial and final positions of the axon. The first and last 10 mm of the axons were tracked to obtain trajectories. (I) Scatter plot of the angle turned versus the initial angle for motor axons in the presence of dimethyl sulfoxide (DMSO) or g-secretase inhibitor (g-sec inh) L-685458. n = 41 (DMSO), n = 31 (g-sec inh). (J) Histogram showing the mean angle turned (± SEM) for axons in the presence of DMSO or g-secretase inhibitor L-685458. (*p < 0.05; Kolmogorov-Smirnov test).

A Netrin

WT

Metalloprotease

B Netrin

Metalloprotease

PS1 KO

Figure 5. DCC Stubs Cause Motor Neuron Chemoattraction to Netrin

P:D ratio

DM S M O P in h

Netrin-1

Netrin-1

Neurite numbers relative to ctrl

Netrin-1

Netrin-1

Netrin-1

Netrin-1

Netrin-1

Netrin-1

(A) In wild-type embryos, DCC is first cleaved by a metalloprotease that leads to shedding of the ectodomain segment, generating membranetethered DCC stubs. DCC stub is subsequently γ-secretase processed by g-secretase to generate ICD. (B) In PS1 KO embryos, the production of DCC-ICD is DCC DCC stub DCC DCC stub disrupted, and DCC stubs accumulate to high DCC-ICD levels on the cell membrane. (C) Western-blot analysis of DCC protein in mouse spinal cords. Protein extracts from the spinal cords C E12.5 mouse spinal cord H WT MN WT MN of Columbus mutants (Col, lane 2), PS1 knockouts (KO, lane 4), or their control littermates (Ctrl, lane 1 + MP inh + DMSO ctrl Col ctrl KO 4 D E (kDa) and 3) were analyzed by immunoblotting with the * * 3 191ˉ DCC intracellular domain-specific antibody. High DCC FL 2 levels of DCC stub were apparent in Columbus and 97ˉ 1 IB: PS1 KO embryos. Glyceraldehyde 3-phosphate Anti-DCC 0 dehydrogenase (GAPDH) was used as an internal 64ˉ (C-ter) h h inh in c in F + γ-sec inh G +& MP γ-sec inh standard. ec se s DCC stub γ γ51ˉ (D–G) Motor explant attraction assay. GFP-labeled + P M mouse motor explants were cocultured with IB: Netrin-1 cell aggregates in the presence of (D) Anti-GAPDH DMSO vehicle, (E) metalloprotease inhibitor GM6001, (F) g-secretase inhibitor L-685458, or (G) both GM6001 and L-685458. (H) P:D ratio of outgrowth from mouse motor DCC stub MN Ctrl MN Ctrl DCC-FL explants in the presence of Netrin-1 cell aggreN O gates. n = 16 (DMSO), n = 8 (metalloprotease K I inhibitor [MP inh]), n = 20 (g-sec inh), and n = 19 explants (MP inh + g-sec inh). Data are presented as the mean ± SEM (*p < 0.05). (I–M) Flat-mount images of chick spinal cords that DCC-ICD MN Ctrl MN DCC-ECD have been electroporated with plasmids encoding + γ-sec inh + γ-sec inh P Q myristoylated DCC intracellular domain (DCC FP L stub), DCC full-length (DCC-FL), DCC extracellular DCC stub domain (DCC-ECD), or DCC intracellular domain (DCC-ICD) as indicated. Hb9-DsRed reporter (red) J was coelectroporated to label MNs. n > 8 embryos DCC-ICD for each plasmid. The scale bar represents 100 mm. 8 M R (N–Q) Chick motor explant attraction assay. Chick * 6 * spinal cords were electroporated with plasmids 4 * encoding myristoylated DCC intracellular domain FP 2 (DCC stub), DCC intracellular domain (DCC-ICD), or control plasmids as indicated. Hb9:DsRed 0 l l reporter (red) was coelectroporated to label MNs. CD Ctr tub Ctr C-I Cs DC DC Motor explants were dissected from electro+ porated spinal cords and cocultured with the γ-sec inhibitor Netrin-1 cell aggregates in 3D collagen/matrigel matrices for 15 hr. Minimal axon outgrowth was observed from MNs in control conditions. (R) Histogram showing quantification (neurite numbers relative to control) of outgrowth from chick motor explants in the presence of Netrin-1 cell aggregates. n = 16 (control), n = 13 (DCC stub), n = 9 (control + L-685458), and n = 14 (DCC-ICD + L-685458). Data are presented as the mean ± SEM (*p < 0.05).

attracted to Netrin-1 when electroporated with Myr-DCC-ICD (DCC stub mimic) (Figures 5O and 5R). Electroporation of the DCC-ICD failed to prevent L-685458-treated MNs from responding to Netrin-1 (Figures 5Q and 5R), suggesting that inappropriate Netrin attraction is not due to an inability to generate the intracellular domain fragments in PS1 mutants. Rather, these data indicated that the accumulation of DCC stubs in PS1 mutants cause newly generated MNs to become responsive to Netrin-1 in the FP.

Inhibition of Slit/Robo Signaling Switches Motor Neurons to a Netrin-1 Responsive State Since MNs express high levels of DCC (Figures S3A and S3B and data not shown) (Keino-Masu et al., 1996), we wondered what normally prevents these neurons from growing to the Netrinpositive FP. Growth cone turning assays suggest that commissural neurons silence Netrin signaling when their axons encounter the Slit-expressing FP, which activates the Robo receptor and leads to Robo-DCC interactions that prevent Netrin Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc. 113

A

1

2

Netrin

Netrin

Slit

1

? DCC

DCC

DCC

Robo

+ + + 2

C

Ctrl MN

Myr-Robo1 MN

+Robo1-Fc

E

*

*

4

*

3 2 1 0

Ctrl DN-Robo1 Ctrl Myr-Robo1

H

3

I P:D ratio

WT MN Netrin-1

Netrin-1

G

+Robo1-Fc

5

+ Robo1-Fc

WT MN Ctrl

F Neurite numbers relative to ctrl

D

Netrin-1

B

DN-Robo1 MN

Netrin-1

Netrin-1

Netrin-1

Ctrl MN

Silencing

Attraction

+ Netrin Slit

+Robo1-Fc

2

*

1 0 BSA Robo1-Fc

Robo1/2 Het

J

Robo1/2 KO

Hb9:GFP

K

attraction (Figure 6A) (Stein and Tessier-Lavigne, 2001). Interestingly, MNs coexpress both Slit and Robo (Figures S3C–S3F) (Brose et al., 1999), leading to the possibility of a self-silencing mechanism that inhibits Netrin responsiveness (Figure 6A). To test this hypothesis, we cocultured chick DsRed-positive MN explants with Netrin-1 expressing Cos cell aggregates. Wildtype chick MNs are normally not attracted to Netrin, but electroporation of a dominant-negative form of Robo1 (Robo1-ectodomain + transmembrane domain) into the explants induced motor axon growth toward the Netrin-1 source (Figures 6C and 6F). A similar effect was observed when Robo1-Fc was bath-applied to the explants to sequester Slits (Figures 6D and 6F). Next we generated a gain-of-function variant of Robo1 by combining a myristoylation signal to the intracellular domain of Robo1 114 Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc.

Hb9:GFP

Figure 6. Inhibition of Slit/Robo Signaling Induces Motor Axon Growth Toward Netrin-1 (A) Spinal cord diagram showing Slit/Robo-mediated silencing mechanism. (1) Pre-crossing commissural axons (red) are attracted to the FP (gray triangle) by Netrin (+) receptor DCC; (2) the midline Slit () activates Robo which blocks Netrin/DCC attraction. Slit is expressed in the motor column (green circle), leading to the possibility of a self-silencing mechanism that inhibits Netrin responsiveness in MNs. (B–E) Chick motor explant attraction assay. Chick spinal cords have been electroporated with plasmids encoding Robo1 extracellular domain (DNRobo1), myristoylated Robo1 intracellular domain (Myr-Robo1), or control plasmids as indicated. Hb9:DsRed (red) reporter was coelectroporated with above plasmids to label MNs. (F) Histogram showing quantification (neurite numbers relative to control) of outgrowth from chick motor explants in the presence of Netrin-1 cell aggregates. n = 13 (control), n = 8 (DN-Robo1), n = 10 (control + Robo1-Fc), and n = 10 (MyrRobo1 + Robo1-Fc). Data are presented as the mean ± SEM (*p < 0.05). (G and H) Motor explant attraction assay. Mouse motor explants were cocultured with Netrin-1 cell aggregates in the presence of vehicle control or Robo1-Fc. (I) P:D ratio of outgrowth from motor explants in the presence of Netrin-1 cell aggregates. n = 16 (BSA) and n = 19 (Robo1-Fc). Data are presented as the mean ± SEM (*p < 0.05). (J and K) Flat-mount images of E13.5 mouse spinal cords at lumbar levels with anterior on the left. Dotted line marks medial edge of MN cell bodies. At least five embryos were assayed for each genotype. See also Figure S3.

(Myr-Robo1), which has been demonstrated to constitutively interact with the DCC cytoplasmic domain (Stein and Tessier-Lavigne, 2001). Electroporation of Myr-Robo1 into chick MNs markedly inhibited the attraction to Netrin-1 caused by Robo1-Fc (Figures 6E and 6F). A similar effect was observed in mouse explants in which bath-applied Robo1-Fc induced motor axon growth toward Netrin-1 (Figures 6G and 6H). To confirm these observations in vivo, we crossed Robo1/2 double mutants to MN transgenic reporter Hb9:GFP and found a subset of MNs extended axons into the FP (Figures 6J and 6K). These findings suggest MNs normally use a self-silencing mechanism based on Slit/Robo coexpression to prevent DCC from causing attraction to Netrin produced by the FP. DCC Stubs Are Immune to Robo-Mediated Silencing Next we examined why the Slit/Robo silencing of Netrin-chemoattraction was ineffective in PS1 mutants. Since DCC stubs accumulate in PS1-deficient embryos, first we tested whether

A

Figure 7. DCC Stubs Do Not Interact with Robo

B

+Netrin-1 & Slit-2

(A) Interaction of Robo and DCC in 293-T cells cotransfected with plasmids encoding Robo1Myc and DCC. Sixteen hours after transfection, DCC cells were treated with vehicle DMSO or g-secreγ-sec inhibitor tase inhibitor for an additional 10 hr, then incubated for 20 min with recombinant Netrin-1 and Robo1 Anti-Myc Slit2 and subjected to immunoprecipitation (IP) with antibodies to Myc or DCC extracellular DCC-FL DCC Stub DCC Anti-DCC domain (N-ter). Immunoprecipitates were Robo (C-ter) DCC stub analyzed by immunoblot with antibodies to Myc or DCC intracellular domain (C-ter). Crude lyses Anti-DCC Anti-Myc IP: (B) Diagram showing the interaction of DCC and (N-ter) Robo receptor in the presence of DCC stubs. Activation of Robo by Slit leads to interaction of Ctrl PS1 KO Robo with full-length DCC; DCC stubs are C excluded from the heteroreceptor complex formed between DCC-FL and Robo, but the DCC stubs can associate with DCC-FL complex lacking Robo. (C) Immunostaining of TAG-1-positive commissural axons in transverse sections of E12.5 mouse spinal cords. TAG-1 staining in PS1 KO embryos was thicker at the midline (red bracket), and the ventral funiculus (yellow bracket) was absent D WT PS1 MUT compared to controls (WT or Het). At least six embryos were assayed for each genotype. The scale bar represents 100 mm. (D) Model for PS1 function in axon navigation. Cellular steps in upper panels contain numbers 2 2 1 corresponding to receptor signaling in lower 1 1 1,3 + + + + 3 + + panels. In wild-type spinal cords, commissural + + + + + + 1,3? axons (red) are initially attracted to the FP by Netrin 2 2,3 until they reach the midline and encounter repul+ Netrin + Netrin Slit Slit sive Slit ligands (1). Robo becomes activated, triggering repulsion from the midline and silencing 2 3 1 Netrin Netrin Netrin the attractive response toward Netrin through Slit Slit interaction with DCC (2). Thus, commissural neurons are first attracted to the FP but then grow through the midline and enter the contralateral ventral funiculus. MNs (green) differ from commissural neurons in that they also express DCC DCC DCC Robo Stub DCC Slits. Slit/Robo interactions in MNs prevent these Robo cells from acquiring responsiveness to FP-derived Netrin (2). MNs may acquire responsiveness to Silencing Attraction Attraction Netrin later in development when their axons reach the periphery and Slit levels decline (1). Concomitantly, peripheral sources of Netrin could induce DCC stub production to overcome residual Slit/Robo silencing, thus triggering a response to Netrin. In PS1 mutants, commissural axons (red) fail to exit the FP and motor axons (green) misproject toward the midline because of abnormal attraction to Netrin (3). In the absence of PS1, the sequential cleavage of DCC is disrupted leading to the accumulation of DCC stubs on the membrane that are resistant to Slit/Robo silencing, thus triggering attraction to Netrin (3). See also Figure S4.

+ + -

+ + +

+ + -

+ + +

+ + -

+ + +

Netrin

Netrin

Slit

TAG-1

Robo1-Myc

DCC stubs dominantly bind to Robo and release full-length DCC for signaling attraction to Netrin. We performed coimmunoprecipitation assays to examine the interactions between DCC and Robo1 in the absence or presence of DCC stubs. Full-length DCC was coexpressed with myc epitope-tagged Robo1 (Robo1-Myc). g-Secretase inhibitor was added into the cell culture bath to induce the accumulation of DCC stubs. When DCC was immunoprecipitated with an antibody to the extracellular domain of DCC (N terminus), similar amounts of Robo1 coimmunoprecipitated with DCC, regardless of whether or not DCC stubs were present (Figure 7A). Likewise, using anti-Myc

antibody to immunoprecipitate Robo1, we found that coimmunoprecipitated full-length DCC pull-down was similar in the presence or absence of DCC stubs (Figure 7A). Interestingly, Robo1 immunoprecipitation did not pull down DCC stubs, suggesting these fragments of DCC are not capable of high affinity interactions with Robo1 (Figures 7A and 7B). We coprecipitated DCC stubs with full-length DCC, however, suggesting Robo1 is excluded from full-length DCC complexes containing the DCC stub (Figure 7B). Taken together, these findings indicate that the receptor complexes containing DCC stubs are insensitive to silencing because they do not interact with Robo. Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc. 115

Ligand Binding Induces the Accumulation of DCC Stubs Since DCC stubs possess potent activity, we tested whether the generation and/or lifespan of this Netrin receptor fragment was regulated. After screening a variety of growth factors and neurotrophins (data not shown), we discovered that Netrin itself induced the accumulation of DCC stubs in a dose-dependent manner (Figures S4B and S4C and data not shown). A slight increase in DCC-ICD was also detected when Netrin was present, but the overall levels of full-length DCC were not substantially changed under these conditions (Figures S4B and S4C). Because DCC is sequentially cleaved by proteases, accumulation of the DCC stub intermediate could occur through one or more mechanisms, including enhancement of the metalloprotease cleavage to generate the stub and/or inhibition of g-secretase activity involved in clearing the stub (Figure S4A). We first isolated the metalloprotease activity in this process by inhibiting g-secretase with L-685458. Under this condition, we found DCC stub levels increased upon Netrin stimulation (Figure S4E). Next, we examined g-secretase cleavage by comparing the ratio of DCC stub (substrate) to DCC-ICD (product) and found that DCC-ICD:DCC stub levels declined in the presence of Netrin (Figure S4D). These findings indicate that Netrin regulates the accumulation of DCC stub, likely through influencing the sequential processing of DCC mediated by metalloprotease and g-secretase, thereby fine tuning DCC signaling (Figure S4F). Commissural Neurons Require PS1 Although commissural neurons are initially attracted to the midline source of Netrin-1, they use Slit/Robo silencing to switch off their responsiveness in order to cross the midline. To test whether PS1 function is required for this switch, we first performed TAG-1 immunostaining of E12.5 spinal cord sections from control and PS1 mutant embryos. TAG-1-positive commissural axons project ventrally toward the FP in both control and PS1 mutant embryos; however, instead of forming a tightly bundled commissure at the FP, the axons appeared disorganized and defasciculated in the mutants (Figure 7C). In addition, the ventral commissure was thicker and the TAG-1 signal in the ventral funiculus was largely absent, suggesting that commissural axons had failed to exit the FP (Figure 7C). DiI labeling of dorsal commissural neurons revealed that commissural axons either stall in the FP of PS1 mutants or inappropriately recross the midline (Figures S4G–S4L). These findings suggest that PS1 is required to switch the guidance properties of commissural neurons once they encounter the FP. In summary, PS1 plays a critical role in establishing neuronal connectivity in both efferent-motor and afferent-sensory pathways. DISCUSSION Long-axoned neurons grow to their targets using a series of intermediate guideposts. A classic example of an intermediate navigational target is the ventral midline of the neural tube. Commissural axons are first attracted to the FP by Netrin but then rapidly lose their responsiveness when axons encounter Slit at the midline. Slit activation of Robo receptors expressed by commissural neurons has two effects: it repels the neurons from the midline and it silences Netrin attraction (Stein and Tessier-Lav116 Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc.

igne, 2001) (Figure 7D). Here, we show that the interplay between Robo and DCC signaling also regulates Netrin responsiveness in MNs, although there are important distinctions from the way signaling is modulated in commissural neurons. While MNs express DCC and Robo, they differ from commissural neurons in that they also express Slits (Brose et al., 1999). Slit/Robo interactions in MNs prevent these cells from acquiring responsiveness to FP-derived Netrin (Figure 7D). Thus, MNs appear to employ a self-silencing mechanism to regulate their responsiveness to Netrin, whereas commissural neurons respond to Slit produced by the FP. Here, the autonomous versus nonautonomous expression of Slit in MNs and commissural neurons, respectively, appears to underlie the difference in when and where these neurons respond to Netrin. Inside the early neural tube, Slit-positive MNs are initially insensitive to Netrin (VarelaEchavarrı´a et al., 1997), presumably to allow their axons to extend into the periphery without becoming attracted to the midline, whereas Slit-negative dorsal commissural neurons are initially responsive to Netrin (Serafini et al., 1996). As development proceeds, Netrin expression is established in several peripheral targets of MNs, including the dermomyotome and the dorsal limb bud, and Slit expression is reduced in MNs (Holmes et al., 1998; Kennedy et al., 1994; Pu¨schel, 1999; Serafini et al., 1996), raising the possibility that older MNs acquire responsiveness to Netrin when their axons extend to the periphery and are no longer at risk of being inappropriately attracted to the FP. By contrast, commissural neurons start to lose their Netrin responsiveness once their axons encounter Slit at the midline (Shirasaki et al., 1996; Stein and Tessier-Lavigne, 2001). The timing and MN subtype regulation of Slit ligand expression in MNs may have a profound role in controlling guidance, counter to the typical view whereby selective guidance receptor expression is the primary determinant. Dorsally projecting cranial motor axons (cranial branchial motor neuron/cranial visceral motor neuron) not expressing Slit, show repulsion to exogenous Slit, while spinal MNs with Slit2 expression have no response (Hammond et al., 2005). This is consistent with the finding that addition of Robo1-Fc to cocultures of spinal motor explants and FP tissue does not block FP-mediated repulsion, suggesting that Slit proteins are not midline repellents for spinal MNs (Patel et al., 2001). Likewise, we found that newly generated Slit-positive MNs are insensitive to Slit repulsion, whereas more mature MNs appear to become sensitive when their Slit levels decline (Brose et al., 1999). These findings suggest that the Slits are dynamically regulated in MNs and have a variety of effects, ranging from masking repulsive responses to silencing Netrin attraction. Although PS1 is widely expressed, it directly influences Netrin/ DCC signaling with cell type precision and spatiotemporal specificity. We found that Netrin stimulation appears to activate metalloprotease activity and inhibit g-secretase leading to accumulation of the intermediate product, DCC stubs. Although the regulatory mechanism controlling protease activity remains unclear, ligand binding to Notch receptor is thought to induce conformational changes that facilitate metalloprotease cleavage (Gordon et al., 2008). In addition, several interacting proteins and kinase pathways have been reported to modulate g-secretase activity (De Strooper and Annaert, 2010; Kim et al., 2006). Moreover, we found that the amounts of DCC stub induced by Netrin

in MNs are normally insufficient to overcome the intrinsic silencing mediated by Slit/Robo coexpression in these cells. However, when Slit silencing is partially released, ligand-induced DCC stubs can trigger Netrin attraction (data not shown). Thus, regulation of DCC processing might help to fine tune the responsiveness of commissural neurons and MNs to Netrin (Figure 7D). The full-length DCC receptor and membrane bound DCC stubs exhibit different protein interactions. We found that Robo interacts with the full-length DCC receptor but not the truncated DCC stubs. Thus, the differences in protein interactions of the DCC full-length and DCC stub appear to influence Slit/Robosilencing of Netrin chemoattraction. Interestingly, a previous study showed that the myristoylated form of DCC can be coimmunoprecipitated with Robo when they are cotransfected into cells, indicating that DCC stubs might interact with Robo under some conditions (Stein and Tessier-Lavigne, 2001). We probed DCC-Robo interactions using immunoprecipitations where fulllength DCC was expressed and DCC stubs were forced to accumulate with g-secretase inhibitor. Thus, it appears that Robo interactions with DCC and its cleaved fragments are hierarchical. Full-length DCC appears to be the preferred partner of Robo, whereas the DCC stubs that accumulate with full-length receptors in PS1 mutants escape Slit/Robo interactions that mediate silencing (Figure 7D). Dominantly inherited mutations in the genes encoding presenilins and the amyloid precursor protein are the major causes of familial Alzheimer’s disease (FAD). The prevailing view of Alzheimer’s pathogenesis posits that accumulation of b-amyloid (Ab) peptides, particularly Ab42, is the central event triggering neurodegeneration and that FAD arises from mutations in PS1 that alter or reduce protease activity (Hardy and Higgins, 1992; Wolfe, 2007). Our findings may provide further insight into understanding the pathogenic mechanisms that underlie FAD and help to identify treatments. First, abnormal axon guidance signaling may have roles in AD pathogenesis by affecting the maintenance and repair of neuronal circuits. Beyond guiding brain wiring during fetal development, many guidance molecules persist in the adult central nervous system and participate in maintenance, repair, and plasticity of neural circuits (Saxena and Caroni, 2007). Recent findings show that Netrin can regulate Ab peptide production and improve Alzheimer’s phenotypes in AD mouse models, which directly link Netrin signaling to neurodegeneration (Lourenc¸o et al., 2009). Therefore, deregulation of guidance signaling by abnormal Presenilin activity may contribute to the pathogenesis and dysfunction seen in AD. Second, our observation that PS1 mutants exhibit impaired axon growth suggests that g-secretase inhibitors used to block toxic Ab production in AD might disrupt axon growth leading to undesirable side effects such as impaired regeneration and repair. In the future, analysis of possible guidance abnormalities in AD mouse models may help to reveal more direct links between abnormal circuit development and the age-dependent loss of neuronal connections. EXPERIMENTAL PROCEDURES DNA Constructs DCC-FL, Myr-DCC-ICD, DCC-ECD, Robo1-FL, and DN-Robo1 expression plasmids were described (Stein and Tessier-Lavigne, 2001). To generate

DCC-ICD, we amplified the ICD by PCR with EcoRI and XhoI sites and cloned into HA-pcDNA3. To generate Myr-Robo1, the ICD was amplified by PCR with NheI and XmaI sites and cloned into pCAGGS-ES. Mice The generation of Tg (Hb9:GFP), Tg (ISLMN:GFP-F), Tg (Nestin:Cre), PS1flox/flox, Netrin-1 mutant, DCC mutant, and Robo1/2 mice was described (Fazeli et al., 1997; Lee et al., 2004; Lewcock et al., 2007; Ma and Tessier-Lavigne, 2007; Serafini et al., 1996; Shen et al., 1997; Shirasaki et al., 2006). PS1 knockouts were generated by crossing PS1flox/flox animals to mice expressing Cre in the germline. ENU screen and mapping were performed as previously described (Lewcock et al., 2007). Explant Cultures MNs were dissected from the spinal cords of E12 mouse embryos or electroporated chick embryos in cold neurobasal media (Invitrogen). Explants were embedded in the rat tail collagen+matrigel (1:1; BD Biosciences) and cocultured with E12 mouse FP or COS cell aggregates in MN media (Shirasaki et al., 2006) for 15–24 hr, then fixed. Where indicated, Netrin-1 proteins (R&D), Robo1-Fc chimera (1–2 mg/ml, R&D), L-685458 (1 mM, Calbiochem), and/or GM 6001 (2.5 mM, Calbiochem) were added to the culture media. Bovine serum albumin (BSA) or DMSO were used as controls. Experiments were done R3 times and the Student’s t test was used to calculate significance. Dunn Chamber Assay Dunn chamber guidance assays were performed as described (Yam et al., 2009). Briefly, ISLMN:GFP-labeled chick MNs were dissected, dissociated, and plated on Laminin-coated coverslips with DMSO or L-685458 (1 mM). After MNs attached (2–3 hr) and formed visible growth cones, the chambers were assembled. SUPPLEMENTAL INFORMATION Supplemental Information includes Supplemental Experimental Procedures and four figures and can be found with this article online at doi:10.1016/j. cell.2010.11.053. ACKNOWLEDGMENTS We would like to thank Edward Koo for critical reading of the manuscript, Shane Andrews for assistance with imaging, and members of the Pfaff lab for advice and discussions on the experiments and manuscript. We are also grateful to Marc Tessier-Lavigne and Jie Shen for providing Netrin-1, DCC and PS1fl/fl mutant mice. We thank Roman Giger and Geetha Suresh for plasmids and Developmental Studies Hybridoma Bank for antibodies. These experiments were supported by National Institutes of Health Grant R37NS037116 through National Institute of Neurological Disorders and Stroke and the Howard Hughes Medical Institute. G.B., D.B., and K.L. are supported by the Howard Hughes Medical Institute. O.C. is supported by NINDS Grant R37NS037116. L.F. is supported by NINDS Grant P01NS031249. S.L.P. is a Howard Hughes Medical Institute Investigator. Received: May 20, 2010 Revised: September 20, 2010 Accepted: November 8, 2010 Published: January 6, 2011 REFERENCES Brittis, P.A., Lu, Q., and Flanagan, J.G. (2002). Axonal protein synthesis provides a mechanism for localized regulation at an intermediate target. Cell 110, 223–235. Brose, K., Bland, K.S., Wang, K.H., Arnott, D., Henzel, W., Goodman, C.S., Tessier-Lavigne, M., and Kidd, T. (1999). Slit proteins bind Robo receptors

Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc. 117

and have an evolutionarily conserved role in repulsive axon guidance. Cell 96, 795–806. Charron, F., Stein, E., Jeong, J., McMahon, A.P., and Tessier-Lavigne, M. (2003). The morphogen sonic hedgehog is an axonal chemoattractant that collaborates with netrin-1 in midline axon guidance. Cell 113, 11–23. De Strooper, B., and Annaert, W. (2010). Novel research horizons for presenilins and g-secretases in cell biology and disease. Annu. Rev. Cell Dev. Biol. 26, 235–260. Dickson, B.J. (2002). Molecular mechanisms of axon guidance. Science 298, 1959–1964. Donoviel, D.B., Hadjantonakis, A.K., Ikeda, M., Zheng, H., Hyslop, P.S., and Bernstein, A. (1999). Mice lacking both presenilin genes exhibit early embryonic patterning defects. Genes Dev. 13, 2801–2810. Fazeli, A., Dickinson, S.L., Hermiston, M.L., Tighe, R.V., Steen, R.G., Small, C.G., Stoeckli, E.T., Keino-Masu, K., Masu, M., Rayburn, H., et al. (1997). Phenotype of mice lacking functional Deleted in colorectal cancer (Dcc) gene. Nature 386, 796–804. Gitai, Z., Yu, T.W., Lundquist, E.A., Tessier-Lavigne, M., and Bargmann, C.I. (2003). The netrin receptor UNC-40/DCC stimulates axon attraction and outgrowth through enabled and, in parallel, Rac and UNC-115/AbLIM. Neuron 37, 53–65. Gordon, W.R., Arnett, K.L., and Blacklow, S.C. (2008). The molecular logic of Notch signaling—a structural and biochemical perspective. J. Cell Sci. 121, 3109–3119. Hammond, R., Vivancos, V., Naeem, A., Chilton, J., Mambetisaeva, E., Mambitisaeva, E., Andrews, W., Sundaresan, V., and Guthrie, S. (2005). Slit-mediated repulsion is a key regulator of motor axon pathfinding in the hindbrain. Development 132, 4483–4495. Hardy, J.A., and Higgins, G.A. (1992). Alzheimer’s disease: the amyloid cascade hypothesis. Science 256, 184–185. Holmes, G.P., Negus, K., Burridge, L., Raman, S., Algar, E., Yamada, T., and Little, M.H. (1998). Distinct but overlapping expression patterns of two vertebrate slit homologs implies functional roles in CNS development and organogenesis. Mech. Dev. 79, 57–72.

Ma, L., and Tessier-Lavigne, M. (2007). Dual branch-promoting and branchrepelling actions of Slit/Robo signaling on peripheral and central branches of developing sensory axons. J. Neurosci. 27, 6843–6851. O’Donnell, M., Chance, R.K., and Bashaw, G.J. (2009). Axon growth and guidance: receptor regulation and signal transduction. Annu. Rev. Neurosci. 32, 383–412. Parent, A.T., Barnes, N.Y., Taniguchi, Y., Thinakaran, G., and Sisodia, S.S. (2005). Presenilin attenuates receptor-mediated signaling and synaptic function. J. Neurosci. 25, 1540–1549. Parks, A.L., and Curtis, D. (2007). Presenilin diversifies its portfolio. Trends Genet. 23, 140–150. Patel, K., Nash, J.A., Itoh, A., Liu, Z., Sundaresan, V., and Pini, A. (2001). Slit proteins are not dominant chemorepellents for olfactory tract and spinal motor axons. Development 128, 5031–5037. Pu¨schel, A.W. (1999). Divergent properties of mouse netrins. Mech. Dev. 83, 65–75. Saxena, S., and Caroni, P. (2007). Mechanisms of axon degeneration: from development to disease. Prog. Neurobiol. 83, 174–191. Selkoe, D., and Kopan, R. (2003). Notch and Presenilin: regulated intramembrane proteolysis links development and degeneration. Annu. Rev. Neurosci. 26, 565–597. Serafini, T., Colamarino, S.A., Leonardo, E.D., Wang, H., Beddington, R., Skarnes, W.C., and Tessier-Lavigne, M. (1996). Netrin-1 is required for commissural axon guidance in the developing vertebrate nervous system. Cell 87, 1001–1014. Shen, J., Bronson, R.T., Chen, D.F., Xia, W., Selkoe, D.J., and Tonegawa, S. (1997). Skeletal and CNS defects in Presenilin-1-deficient mice. Cell 89, 629–639. Shirasaki, R., Mirzayan, C., Tessier-Lavigne, M., and Murakami, F. (1996). Guidance of circumferentially growing axons by netrin-dependent and -independent floor plate chemotropism in the vertebrate brain. Neuron 17, 1079– 1088.

Kandel, E.R., Schwartz, J.H., and Jessell, T.M. (2000). Principles of neural science, Fourth Edition (New York: McGraw-Hill, Health Professions Division).

Shirasaki, R., Lewcock, J.W., Lettieri, K., and Pfaff, S.L. (2006). FGF as a target-derived chemoattractant for developing motor axons genetically programmed by the LIM code. Neuron 50, 841–853.

Keino-Masu, K., Masu, M., Hinck, L., Leonardo, E.D., Chan, S.S., Culotti, J.G., and Tessier-Lavigne, M. (1996). Deleted in Colorectal Cancer (DCC) encodes a netrin receptor. Cell 87, 175–185.

Stein, E., and Tessier-Lavigne, M. (2001). Hierarchical organization of guidance receptors: silencing of netrin attraction by slit through a Robo/DCC receptor complex. Science 291, 1928–1938.

Kennedy, T.E., Serafini, T., de la Torre, J.R., and Tessier-Lavigne, M. (1994). Netrins are diffusible chemotropic factors for commissural axons in the embryonic spinal cord. Cell 78, 425–435.

Taniguchi, Y., Kim, S.H., and Sisodia, S.S. (2003). Presenilin-dependent ‘‘gamma-secretase’’ processing of deleted in colorectal cancer (DCC). J. Biol. Chem. 278, 30425–30428.

Kim, S.K., Park, H.J., Hong, H.S., Baik, E.J., Jung, M.W., and Mook-Jung, I. (2006). ERK1/2 is an endogenous negative regulator of the gamma-secretase activity. FASEB J. 20, 157–159.

Tessier-Lavigne, M., and Goodman, C.S. (1996). The molecular biology of axon guidance. Science 274, 1123–1133.

Lee, S.K., Jurata, L.W., Funahashi, J., Ruiz, E.C., and Pfaff, S.L. (2004). Analysis of embryonic motoneuron gene regulation: derepression of general activators function in concert with enhancer factors. Development 131, 3295– 3306. Lewcock, J.W., Genoud, N., Lettieri, K., and Pfaff, S.L. (2007). The ubiquitin ligase Phr1 regulates axon outgrowth through modulation of microtubule dynamics. Neuron 56, 604–620. Lourenc¸o, F.C., Galvan, V., Fombonne, J., Corset, V., Llambi, F., Mu¨ller, U., Bredesen, D.E., and Mehlen, P. (2009). Netrin-1 interacts with amyloid precursor protein and regulates amyloid-beta production. Cell Death Differ. 16, 655–663.

118 Cell 144, 106–118, January 7, 2011 ª2011 Elsevier Inc.

Varela-Echavarrı´a, A., Tucker, A., Pu¨schel, A.W., and Guthrie, S. (1997). Motor axon subpopulations respond differentially to the chemorepellents netrin-1 and semaphorin D. Neuron 18, 193–207. Wolfe, M.S. (2007). When loss is gain: reduced presenilin proteolytic function leads to increased Abeta42/Abeta40. Talking Point on the role of presenilin mutations in Alzheimer disease. EMBO Rep. 8, 136–140. Yam, P.T., Langlois, S.D., Morin, S., and Charron, F. (2009). Sonic hedgehog guides axons through a noncanonical, Src-family-kinase-dependent signaling pathway. Neuron 62, 349–362. Yu, T.W., and Bargmann, C.I. (2001). Dynamic regulation of axon guidance. Nat. Neurosci. Suppl. 4, 1169–1176.

Tunable Signal Processing in Synthetic MAP Kinase Cascades Ellen C. O’Shaughnessy,1 Santhosh Palani,2 James J. Collins,1,4,* and Casim A. Sarkar2,3,* 1Howard Hughes Medical Institute and Department of Biomedical Engineering, Center for BioDynamics and Center for Advanced Biotechnology, Boston University, Boston, MA 02215, USA 2Department of Bioengineering 3Department of Chemical and Biomolecular Engineering University of Pennsylvania, Philadelphia, PA 19104, USA 4Wyss Institute for Biologically Inspired Engineering, Harvard University, Boston, MA 02115, USA *Correspondence: [email protected] (J.J.C.), [email protected] (C.A.S.) DOI 10.1016/j.cell.2010.12.014

SUMMARY

The flexibility of MAPK cascade responses enables regulation of a vast array of cell fate decisions, but elucidating the mechanisms underlying this plasticity is difficult in endogenous signaling networks. We constructed insulated mammalian MAPK cascades in yeast to explore how intrinsic and extrinsic perturbations affect the flexibility of these synthetic signaling modules. Contrary to biphasic dependence on scaffold concentration, we observe monotonic decreases in signal strength as scaffold concentration increases. We find that augmenting the concentration of sequential kinases can enhance ultrasensitivity and lower the activation threshold. Further, integrating negative regulation and concentration variation can decouple ultrasensitivity and threshold from the strength of the response. Computational analyses show that cascading can generate ultrasensitivity and that natural cascades with different kinase concentrations are innately biased toward their distinct activation profiles. This work demonstrates that tunable signal processing is inherent to minimal MAPK modules and elucidates principles for rational design of synthetic signaling systems. INTRODUCTION MAPK pathways are ubiquitous, versatile signaling modules found in all eukaryotic cells. They transmit and process signals regulating a broad array of cell fate decisions, including proliferation, differentiation, motility, stress responses, and apoptosis (Avruch, 2007). Though several MAPK families have been elucidated, all consist of three sequential kinases activated by dual, nonprocessive phosphorylation events. The biological design principles underlying a three-tiered cascade structure have been the subject of vigorous debate for many years. A number of hypotheses have been put forth to account for the

use of multistep signaling pathways, including signal amplification, increased signaling speed, multiple points of regulation, noise tolerance, and the generation of switch-like responses (Chen and Thorner, 2005). Though it is likely that this heavily conserved pathway has evolved to serve multiple purposes, its structure clearly enables great flexibility of system response. Extensive experimental work on MAPK cascades has shown that responses can be graded or switch-like (ultrasensitive), transient or sustained, and monostable or bistable (Bhalla et al., 2002; Huang and Ferrell, 1996; Poritz et al., 2001; Santos et al., 2007). A growing body of research has demonstrated that the systems-level properties of MAPK activation vary greatly depending on cellular context and that the characteristics of this activation can determine cell fate. Seminal work on the activation of p42 MAPK in X. laevis demonstrated steep ultrasensitivity (Ferrell and Machleder, 1998; Huang and Ferrell, 1996), whereas studies of MAPK cascades in S. cerevisiae revealed graded activation profiles (Poritz et al., 2001). A series of studies in mammalian systems showed that different cell types exhibit either switch-like (Bagowski et al., 2003; Harding et al., 2005) or proportional (Mackeigan et al., 2005; Whitehurst et al., 2004) activation responses. Even within the same cell type, an individual MAPK cascade can exhibit different dynamic responses that lead to distinct cell fates. In PC12 cells, MAPK activation is transient and graded when stimulated with epidermal growth factor, leading to proliferation. In contrast, nerve growth factor stimulation results in cell differentiation through sustained and ultrasensitive MAPK activation (Marshall, 1995). Recent work has shown that altering the activation profile of MAPK in these cells is sufficient to reverse the stimulus-phenotype relationship (Santos et al., 2007). Further, in BHK cells, tethering the MAPK module to the plasma membrane can lower the threshold of activation, and consequently, the percentage of cells that differentiate depends upon both the stimulus magnitude and the location of the cascade in the cell (Harding et al., 2005). Finally, it has been shown that activation of MAPK in NIH 3T3 fibroblasts is dependent upon temporal dynamics of the stimulus (Bhalla et al., 2002). Naive cells exhibit bistable MAPK activation in response to plateletderived growth factor, whereas previously stimulated cells show proportional, monotonic activation. It has been proposed Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc. 119

that this desensitization enables the downregulation of MAPK activity, a molecule whose prolonged stimulation can be deleterious. It is clear from both experimental and theoretical work that the behavior of MAPK cascades in vivo is dictated, in part, by multiple levels of feedback regulation. In particular, bistability, an essential characteristic in many cell fate decisions, arises through feedback. However, feedback alone is not sufficient to achieve bistability. Bistable responses require a nonlinearity in the system, feedback to reinforce the nonlinearity, and proper balance between the system components (Ferrell and Xiong, 2001). Ultrasensitivity in MAPK activation is a critical, systemslevel property, as it can serve as the fundamental nonlinearity required to achieve stable, potentially irreversible cellular decisions. Indeed, several experimental studies of bistable MAPK activation report ultrasensitive activation profiles as well (Bhalla et al., 2002; Ferrell and Machleder, 1998; Santos et al., 2007). Dissecting the contributions of ultrasensitivity and feedback to a specific bistable response is often untenable in the context of complex endogenous signaling networks. Many biological systems are organized into functional modules that perform key steps in a larger process (Hartwell et al., 1999). It is often difficult to characterize the potential behaviors of these component subsystems in vivo because of interconnectivity and complex layers of regulation found in biological systems. This is particularly true in signaling networks, which exhibit multifaceted regulation such as feedback, localization, and extensive, functional crosstalk (Natarajan et al., 2006). Nonetheless, interpreting, influencing, and predicting how complex networks carry out biological functions depends critically upon our understanding of their component modules. Here, we use a synthetic biology approach to obviate the challenges of interconnectivity in uncovering the inherent capabilities of the MAPK module; we do so by systematically applying intrinsic and extrinsic perturbations to tune the activation dynamics of a minimal, well-insulated cascade. By studying an isolated module, we develop a mechanistic understanding of the effects of these perturbations both experimentally and computationally. In the present study, we built an exogenous, minimal MAPK cascade to investigate the effects of extrinsic and intrinsic regulators on the plasticity of this isolated signaling module. We expressed the mammalian Raf-MEK-ERK cascade in the yeast S. cerevisiae and applied the intrinsic perturbation of concentration variation between cascade members and the extrinsic perturbations of scaffolding and negative regulation. We show that varying the relative concentrations of MEK and ERK confers great flexibility of the system response and may prime the cascade for either low or high ultrasensitivity. Strikingly, the X. laevis p42 MAPK cascade that shows sharp ultrasensitivity and the S. cerevisiae pheromone pathway that is more graded both fit well within the theoretical framework generated by this intrinsic perturbation. In addition, we identify cascading itself as a concentration-dependent mechanism for generating ultrasensitivity. We further demonstrate that, in contrast to a biphasic dependence on scaffold concentration (prozone effect), expression of a two-member scaffold results in a monotonic dependence of signal strength on scaffold concentration, with a sharp 120 Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc.

reduction observed at high concentration. Finally, introducing negative regulation of either MEK or ERK leads to a reduction in ultrasensitivity and an increase in threshold, even in the absence of feedback. We used a computational model to integrate these intrinsic and extrinsic perturbations over a large parameter space and found regions in which signal characteristics can be tuned independently. Experimentally, we demonstrate the decoupling of signal strength from the ultrasensitivity and threshold of the response. Thus, through the use of a synthetic signaling module, we identified mechanisms for generating ultrasensitivity, elucidated cellular strategies for tuning the activation of the system, and highlighted regulatory principles that can be used in designing artificial signaling networks. RESULTS The Basic Synthetic Cascade We constructed a basic synthetic cascade using the mammalian pathway of Raf-MEK-ERK (Figure 1A) in which the Raf level of the cascade is a hormone-binding domain fusion protein, eGFP:DRaf-1:ER-DD (Raf:ER), whose kinase activity is modulated by the addition of b-estradiol (McMahon, 2001). Consequently, the entire cascade is cytosolic (Figure S1 available online) and does not depend upon receptor-mediated membrane recruitment for activation. Single integrated copies of wild-type, epitope-tagged Raf:ER, c-myc-MEK1 (MEK), and His6-ERK2 (ERK) were coexpressed from different auxotrophic loci by a reverse tTA-driven expression system (PtetO7) (Bellı´ et al., 1998) induced with anhydrotetracycline (aTC). The system was stimulated with estradiol and assayed for steady-state activated ERK directly with quantitative western blotting (Figure 1B) to eliminate additional signal processing steps such as transcriptional activation (Mackeigan et al., 2005). As designed, our system is not bistable, as it does not fulfill the requirements of nonlinearity, feedback, and proper balance, and therefore the population mean reflects the unique activation state of the system. When the three kinases were expressed at equal concentration levels, the system reached a half-maximal response (EC50) at 32 ± 1.4 nM and exhibited moderate ultrasensitivity (Figure 1C). The apparent Hill coefficient attained by this basic, constant-expression cascade, nH = 1.8 ± 0.13, fits well within the range of sensitivities observed in natural systems. An ordinary differential equation (ODE) model of the basic cascade based on mass action kinetics (Huang and Ferrell, 1996) accurately captures the response (Figure 1C). All parameters and initial conditions were refined from literature values by globally fitting the Hill coefficient, signal strength, and EC50 to the experimental steady-state response curves of the basic and variable expression cascades (Extended Experimental Procedures). Parameter values were then held constant in simulating subsequent perturbations. The model assumes nonprocessive, dual-step phosphorylation events between sequential kinases, as has been demonstrated experimentally (Burack and Sturgill, 1997). Insulation of this exogenous cascade from the host was assayed by a variety of methods. We identified possible interaction pathways by homology (Avruch, 2007) and previous findings in

A

Figure 1. The Basic Synthetic Cascade

B Δ

Estradiol

C Normalized Fold Increase

1.2 1.0 0.8 0.6 0.4 0.2 0.0 -5.0

-2.5

0.0

2.5

Estradiol (log μM)

D

E

25

N

o

Fa

Su

ga

or ct

ol

Po

α-

di tra

si Es nd tiv tr e ad C i Ba on ol ck tro gr l Po o si Es und tiv tr e ad C i Ba on ol ck tro gr l Po o si Es und tiv tr e ad C io on l tro l

ou gr ck

r

0

0

Ba

20

The Variable Expression Cascades In natural systems, Raf, MEK, and ERK are present at different concentrations, frequently increasing for sequential kinases. Further, the relative concentrations of cascade members vary significantly across different experimental systems (Ferrell, 1996). To understand the systems-level effects of this intrinsic perturbation, we constructed a cascade increasing the kinase concentration of each subsequent step, achieving 10-fold difference across the entire cascade (Figure 2A). Concentration was varied by expressing a single, integrated copy of Raf:ER with PtetO7, highcopy MEK with PtetO7, and high-copy ERK from the Gal1 promoter (PGal) induced with galactose. The concentration range of this synthetic system was initially determined computationally by a global fit to the basic and variable cascades and was verified experimentally by quantitative ELISA against ERK (Figure S2). Under this perturbation, we found that the ultrasensitivity increased from nH = 1.8 ± 0.13 to 2.8 ± 0.19, whereas the threshold was reduced from 32 ± 1.4 nM to 6.6 ± 0.14 nM (Figure 2B). Qualitatively, this response is captured well by our computational model (Figure 2C). Interestingly, for this experimentally attainable range of parameter values, the model predicts that the MEK concentration will have a greater effect than that of ERK on both the ultrasensitivity and threshold of the response. We computationally varied MEK and ERK from 10 nM to 100 nM and determined the Hill coefficient and EC50 of each activation response (Figures 2D and 2E). The slope of the resulting surfaces is steeper when MEK is varied for a fixed value of ERK than when ERK is varied for a fixed MEK value. Therefore, we hypothesized that reducing the concentration of MEK in our variable cascade would increase the threshold and reduce the ultrasensitivity of the response. To test this hypothesis, we built a variable expression cascade with a concentration of MEK in between those in the basic and e

1

40

the literature (Atienza et al., 2000) and consequently designed pathway-specific transcriptional reporters for the pheromone, invasive growth, and cell wall integrity pathways in S. cerevisiae. The synthetic cascade was stimulated with estradiol, and expression of tdTomato from each native pathway-specific promoter was determined by flow cytometry. We found minimal transcriptional activation from each of the endogenous pathways assayed when compared with pathway-specific positive controls (Figure 1D). In addition, we stimulated cells with agonists of endogenous MAPK pathways, including a factor (Bashor et al., 2008), glucose depletion (Cullen and Sprague, 2000), and caffeine (Jung et al., 2002), and determined the activation level of the synthetic MAPK cascade. We found almost no phosphorylation of ERK in response to glucose starvation and caffeine and 15% activation in response to a factor despite the significant impact that these stimuli have on diverse cellular processes (Figure 1E). Further, we assayed for dephosphorylation of the synthetic cascade by endogenous phosphatases and found very modest interaction with these negative regulators (Figure S1). Taken together, these data indicate that the synthetic cascade is well insulated from the host, and the dynamic responses that we observe are a result of the minimal system itself and do not arise from unknown endogenous effectors.

in

2

60

fe

3

80

af

4

C

Percent Response

5

Es

Fold Increase

100

(A) Schematic of the basic synthetic cascade. Raf:ER, MEK, and ERK were coexpressed by PtetO7. The kinase activity of Raf:ER is modulated by estradiol. See also Figure S1A. (B) Representative western blot of steady-state ERK activation with increasing estradiol concentration. (C) ERK activation in response to estradiol titration (black squares). The experimental data are the mean ± SEM normalized to the fitted baselines. Data were fit with a modified Hill equation (black line). The model simulation results of the steady-state response profile for the basic cascade are shown as a red line. The system response is moderately ultrasensitive, with a Hill coefficient of 1.8 ± 0.13 and an EC50 of 32 ± 1.4 nM. (D) Transcriptional activation of endogenous promoters by the basic synthetic cascade. The data are the mean ± SEM tdTomato expression for the mating (red), invasive growth (orange), and cell wall integrity (yellow) pathways normalized to background. (E) Activation of ERK by endogenous stimuli. Data are the mean ± SEM phosphorylation level normalized to the estradiol stimulated cascade. See also Figure S1B.

Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc. 121

A

B

C High-MEK Variable Cascade 1.0

High-MEK Variable Cascade Normalized Fold Increase

kinase concentration

Normalized Fold Increase

Δ

Expt

0.8 0.6 0.4 0.2 0.0 -3

-2

-1

1.0 Model 0.8 0.6 0.4 0.2 0.0

0

-3

Estradiol (log μM)

-2

-1

0

Estradiol (log μM)

E

EC50

Hill Coefficient

D

[ERK2]

F

)

] (nM)

1.0

Expt

0.8 0.6 0.4 0.2 0.0 -3

G Normalized Fold Increase

Normalized Fold Increase

Low-MEK Variable Cascade

-2

-1

Estradiol (log μM)

0

)

1] (nM

[ERK2

1] (nM

[MEK

(nM)

[MEK

Low-MEK Variable Cascade 1.0 Model 0.8 0.6 0.4 0.2 0.0 -3

-2

-1

0

Estradiol (log μM)

Figure 2. The Variable Expression Cascades (A) Schematic of variable expression cascades. The kinase concentration increases at each tier in the cascade. (B) Normalized fold increase of the high-MEK variable cascade (dark purple squares) and the basic cascade (black squares). The high-MEK variable cascade: single-copy Raf:ER expressed by PtetO7, high-copy MEK expressed by PtetO7, and high-copy ERK expressed by PGal (see also Figure S2). The data are the mean ± SEM, normalized to the fitted baselines. Data were fit with a modified Hill equation (solid lines). (C) Model simulations of the high-MEK variable cascade (dark purple) and the basic cascade (black). The fitted concentrations were Raf:ER = 10 nM, MEK = 80 nM, and ERK = 100 nM. (D) Simulated Hill coefficient for the variable cascade in which both MEK and ERK were varied from 10 nM to 100 nM. (E) Simulated EC50 for the variable cascade for the same parameters as in (D). (F) Normalized fold increase of the low-MEK variable cascade (light purple squares) and the basic cascade (black squares). The low-MEK variable cascade: single-copy Raf:ER expressed by PtetO7, single-copy MEK expressed by PGal, and high-copy ERK expressed by PGal. The data are the mean ± SEM, normalized to the fitted baselines. Data were fit with a modified Hill equation (solid lines). (G) Model simulations of the low-MEK variable cascade (light purple) and the basic cascade (black). The fitted concentrations were Raf:ER = 10 nM, MEK = 30 nM, and ERK = 100 nM.

122 Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc.

A

Figure 3. The Basic Cascade with Scaffolding

B 100

Percent Response

Δ

Expt Model

80 60 40 20 0 no scaffold

C 1.0 0.8 0.6 0.4 0.2

Expt

Normalized Fold Increase

Normalized Fold Increase

D 1.2

1.2 1.0 0.8

low scaffold

Model

high scaffold

(A) Schematic of the basic cascade coexpressed with a twomember scaffold pax*. See also Figure S3A. (B) Percent response of the basic cascade alone and coexpressed with single-copy or high-copy pax*. The experimental data (blue) are the mean ± SEM, and the computational data (gray) are single values. All are normalized to the maximum activation of the basic cascade in the absence of scaffold. (C) Normalized fold increase of the basic cascade (black squares) coexpressed with single-copy (red squares) or highcopy (orange squares) pax*. The data are the mean ± SEM, normalized to the fitted baselines (basic and single-copy pax*) or the minimum and maximum values (high-copy pax*) for each condition. The data were fit with a modified Hill equation (black and red lines) or biphasic dose-response equation (orange line). (D) A compartmental model of no (black), low (red), or high (orange) scaffold expression. Simulations are normalized to the maximum value of each condition. See also Figure S3B.

0.6 0.4 0.2

folding to effectively propagate a signal. This is in contrast to some native yeast MAPK cascades, -3 -2 -1 0 1 2 -3 -2 -1 0 1 2 such as the mating and high-osmolarity pathways, Estradiol (log μM) Estradiol (log μM) that require scaffolding (Qi, 2005). Furthermore, our findings show that, when expressed near stoichiometric ratios that are optimal for signal propagation (Levchenko et al., 2000), scaffolding original variable cascades (Figures 2F and 2G). In this low-MEK has a negligible effect on signal strength compared to that of variable cascade, MEK was expressed from a single integrated the basic cascade. copy by PGal, whereas the Raf:ER and ERK levels were kept In addition to a reduction in signal strength, we observed the same in the two variable cascades. As predicted, lowering a two-phase dose-response in the presence of high scaffold MEK expression resulted in both a reduction in ultrasensitivity concentrations (Figure 3C). Though paxillin is not homologous from nH = 2.8 ± 0.19 to 2.0 ± 0.19 and an increase in threshold to native yeast MAPK scaffolds, exogenous paxillin has previfrom EC50 = 6.6 ± 0.14 nM to 15 ± 0.78 nM when compared ously been shown to localize strongly to sites of polarized growth with the high-MEK cascade (Figures 2B and 2F). in yeast (Gao et al., 2004). Therefore, we hypothesized that this complex behavior arises from sequestration of the MAPK cascade, and we tested this hypothesis by explicitly including The Basic Cascade with Scaffolding compartmentalization in our model of the basic cascade to Scaffolding has been shown to be critical to signal transduction generate a scaffold model. We found a single-phase ERK activafor some MAPK pathways (Elion, 2001), whereas, in others, the tion profile at moderate scaffold levels and a biphasic doseeffects of colocalization remain unclear (Kolch, 2005). Furtherresponse at high scaffold concentrations (Figure 3D). This more, it is evident that scaffold concentration can be a strong multiphase response could not be generated with either a welldeterminant of MAPK activation (Burack and Shaw, 2000; mixed model or a compartmental model in which paxillin cannot Levchenko et al., 2000; Locasale et al., 2007). Therefore, we bind ERK (Figure S3), suggesting that sequestration of the coexpressed the basic cascade with a modified two-member cascade is necessary to elicit the observed response. Y118D (pax*), and varied the (MEK and ERK) scaffold, paxillin scaffold concentration from 30 nM to 100 nM by expression The Basic Cascade with Negative Regulation from single-copy and high-copy PGal (Figure 3A). Association In natural MAPK cascades, sequential activation by kinases is of ERK and wild-type paxillin is regulated by phosphorylation balanced by inactivation through negative regulation. Both theoof Tyr118 (Ishibe et al., 2003), which we mimicked with mutation retical (Heinrich et al., 2002) and experimental (Hornberg et al., to Asp and verified by coimmunoprecipitation (Figure S3). We 2005) work has demonstrated that activation and deactivation found that moderate expression of pax* from single-copy PGal processes control distinct aspects of the MAPK response. results in a slight reduction in ultrasensitivity and a marginal Further, it has been shown that high concentrations of phosphaeffect on signal strength (Figure 3B). In contrast, we observed tase can convert an ultrasensitive, bistable MAPK response into a marked decrease in signal strength to 25% unscaffolded a proportional, transient activation profile via the disruption of levels when pax* is coexpressed from high-copy PGal. These positive feedback (Bhalla et al., 2002). We hypothesized that, results demonstrate that the Raf-MEK-ERK cascade is even in the absence of feedback, the presence of negative reguinherently catalytically efficient and does not depend on scaf- lators would tune the activation profile of the MAPK response. 0.0

0.0

Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc. 123

Δ

P

C Normalized Fold Increase

B Normalized Fold Increase

A

1.0 Expt 0.8 0.6 0.4 0.2 0.0

P

-3

-2

-1

0

1.0 Model 0.8 0.6 0.4 0.2 0.0

1

-3

Estradiol (log μM)

Δ

P P

-1

0

1

F 1.0

Normalized Fold Increase

E Normalized Fold Increase

D

-2

Estradiol (log μM)

Expt

0.8 0.6 0.4 0.2 0.0 -3

-2

-1

0

Estradiol (log μM)

1

1.0 Model 0.8 0.6 0.4 0.2 0.0 -3

-2

-1

0

1

Estradiol (log μM)

Figure 4. Negative Regulation of the Basic Cascade (A) Schematic of the basic cascade coexpressed with the ERK phosphatase MKP1-cyt. (B) Normalized fold increase of the basic cascade (black squares) coexpressed with high-copy MKP1-cyt (green squares) expressed by PGal. The data are the mean ± SEM, normalized to the fitted baselines. The data were fit with a modified Hill equation (solid lines). (C) Model simulations with (green) and without (black) MKP1-cyt coexpression. Data are normalized to the maximum value of each condition. (D) Schematic of the basic cascade with the MEK inhibitor CI-1040. (E) Normalized fold increase of the basic cascade (black squares) pretreated with 50 nM CI-1040 for 30 min prior to estradiol stimulation (blue squares). The data are the mean ± SEM, normalized to the fitted baselines. The data were fit with a modified Hill equation (solid lines). (F) Model simulations with (blue) and without (black) inhibitor pretreatment. Data are normalized to the maximum value of each condition.

We independently introduced two modes of negative regulation to a purely kinase-based cascade to test this hypothesis. First, we coexpressed the basic cascade with a weak, cytosolically targeted ERK phosphatase, MKP-1L16A/L17A (MKP1-cyt) (Wu et al., 2005), from single-copy PGal (Figure 4A). The presence of this dual-specificity phosphatase affected not only the threshold of the system, which increased from 32 ± 1.4 nM to 110 ± 9.3 nM, but also the ultrasensitivity, which was reduced from nH = 1.8 ± 0.13 to 1.2 ± 0.11 (Figure 4B). A model of the basic cascade with this additional enzymatic reaction predicted both of these features qualitatively (Figure 4C). In addition to enzymatic negative regulation of ERK, we perturbed the system with the small molecule MEK inhibitor, CI-1040 (Figure 4D). CI-1040 is a strong noncompetitive but reversible binder that targets the unphosphorylated form of MEK and renders the kinase catalytically inactive (Ohren et al., 2004). We found similar results to coexpression of MKP1-cyt in that the Hill coefficient decreased from 1.8 ± 0.13 to 1.1 ± 0.12 in the presence of 50 nM CI-1040, and the threshold increased from 32 ± 1.4 nM to 120 ± 10 nM (Figure 4E). A model of the basic 124 Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc.

cascade with an additional inhibitor-binding reaction captured both features of this response (Figure 4F). Whereas these results are similar to those observed under ERK phosphatase coexpression, the mechanisms behind enzymatic and binding-mediated regulation are different. CI-1040 effectively lowers the concentration of available MEK, thereby reducing the ultrasensitivity and increasing the threshold. The converse effect is seen in the variable cascade, in which increasing MEK concentration results in greater steepness and a lower response threshold. Integrating Variable Expression Cascades with Negative Regulation Our computational model captures all of the qualitative trends seen in the experimental data for individual perturbations to the basic synthetic cascade. We therefore used the model to probe the effects of combining different perturbations over a broader range of parameters. Initially, we explored the intrinsic perturbation of concentration variation within the basic cascade (Figure 5A), as MEK and ERK were varied from 10 nM to 2 mM and 10 nM to 10 mM, respectively, and the Hill coefficient, EC50, and

A

B

C MEK1 Inhibition

ERK2 Phosphatase

X

ERK2 Concentration (nM)

Y

Signal Strength

EC50

Hill Coefficient

No Regulation

V

B

V

B MEK1 Concentration (nM)

Figure 5. Computational Analysis of Concentration Variation and Negative Regulation (A) No regulation: Raf:ER was held constant at 10 nM, and MEK and ERK were varied from 10 nM to 2 mM and 10 nM to 10 mM, respectively. The top panel shows the Hill coefficient, the middle panel shows the EC50, and the bottom panel shows the normalized signal strength. (Top) The point marked ‘‘X’’ indicates the Xenopus cascade, and the point marked ‘‘Y’’ indicates the yeast pheromone cascade. Each symbol represents a distinct class of system response (Table S1). For example, the square identifies systems for which the Hill coefficient is low, the EC50 is high, and the normalized signal strength is low. (Bottom) The point marked ‘‘B’’ denotes the basic cascade, and the point marked ‘‘V’’ denotes the high-MEK variable cascade. See also Figure S4 and Figure S5. (B) MEK inhibition: The same parameter space as in (A) modeled with 50 nM CI-1040. (Top) The star represents a distinct system response not seen in the absence of negative regulation. (Bottom) ‘‘B’’ indicates the basic cascade, and ‘‘V’’ indicates the high-MEK variable cascade. (C) ERK phosphatase: The same parameter space as in (A) modeled with 100 nM MKP1-cyt coexpression. The star represents the system response in (B) achieved in a different region of parameter space.

signal strength were determined over the entire parameter space. Strikingly, the Hill coefficient is strongly biphasic, whereas the EC50 and signal strength are monotonic. This enables the decoupling of ultrasensitivity from other signal characteristics simply by adjusting the relative concentrations of cascade members. For example, at high MEK concentrations,

the EC50 is low and the signal strength is high, whereas the Hill coefficient can be independently tuned from 2 to 4 by varying the ERK concentration. Conversely, at low MEK levels, the Hill coefficient is low for all ERK concentrations, whereas the EC50 and signal strength vary considerably as a function of ERK level. The decoupling of signal characteristics in the absence of Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc. 125

extrinsic perturbation highlights the inherent flexibility in a threetiered kinase cascade. We then applied the extrinsic perturbation of MEK inhibition (Figure 5B) and found that the Hill coefficient is also biphasic with ERK, yet the transitions become steeper with respect to MEK concentration. The threshold is shifted so that, for low MEK, the EC50 is high for a larger range of ERK concentration than in the unregulated case. Not surprisingly, the region of maximal signal strength is smaller, showing a marked reduction in gain for low MEK and low ERK concentrations. We show in Figure 4 that two modes of inhibition were similar for given parameter values; however, modeling negative regulation with ERK phosphatase reveals very different behavior from binding inhibition of MEK when surveyed over a wider parameter space (Figure 5C). Though experimentally we see a decrease in ultrasensitivity upon phosphatase coexpression (Figures 4B and 4C), our model predicts that, in a different region of parameter space, the Hill coefficient will be greater than that of the basic cascade alone. Furthermore, in the presence of phosphatase, the EC50 becomes biphasic, and the signal strength exhibits a broad range of intermediate values. These nonmonotonic profiles increase the possible combinations of signal characteristics and further enable the decoupling of signal aspects. Theoretical Analysis of Cascade-Generated Ultrasensitivity Several mechanisms leading to MAPK ultrasensitivity have been posited including zeroth-order ultrasensitivity (Goldbeter and Koshland, 1981), dual-step phosphorylation, and competitive inhibition (Ferrell, 1996). Our synthetic cascade is not subject to either zeroth-order ultrasensitivity or competitive inhibition, but it is affected by dual-step phosphorylation of MEK and ERK. Interestingly, our experimental and computational results both indicate that kinase concentration itself, even in the absence of zeroth-order effects, contributes to MAPK ultrasensitivity (Figure 2 and Figure 5). Therefore, we simulated dualstep and single-step mass action kinetic models over a wide range of concentrations to dissect the contributions of multistep processes and relative kinase concentration to the overall ultrasensitivity of ERK activation (Figure 6A). In addition, we generated a Hill equation that matched the dose-response of active MEK from these explicit models and used this aggregate equation as the stimulus to a final, explicit ERK step. The relative kinase concentrations chosen correspond to distinct sets of behaviors identified in Figure S4 and Figure S5. We found that, in all cases, the ultrasensitivity arising from explicit cascading is greater than simple multiplicative accumulation (Figure 6B). Furthermore, when each level in the cascade is modeled as a single phosphorylation process (i.e., when dualstep phosphorylation is eliminated), the supermultiplicative generation of ultrasensitivity by cascading is maintained and is strongly influenced by concentration, though the overall ultrasensitivity is considerably lower than that of dual-step cascades. To explore the mechanisms underlying this generation of ultrasensitivity by cascading, we analyzed the dose-response curves of each species accounted for by mass action kinetics (Table S2). When the cascade is modeled explicitly, active MEK exists in either a free or complex-bound state, and the behavior of 126 Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc.

these distinct species varies dramatically depending on the relative enzyme and substrate concentrations. For example, when the ultrasensitivity of ERK is high, this steepness arises specifically from the complex-bound MEK species even though the total MEK ultrasensitivity is low (Figure 6C). In contrast, when the ERK ultrasensitivity is low, there is a negligible difference between the dose-responses of each MEK species, as all have low ultrasensitivity. Notably, this phenomenon occurs for both dual-step and single-step kinase activation. Furthermore, when the Raf and MEK levels of the cascade are aggregated into a single Hill equation that accurately mimics the dose-response of total active MEK in the explicit model, the complex-bound MEK species has a lower ultrasensitivity than in the explicit model alone (Figure 6D). Therefore, the multitiered structure of MAPK cascades does not simply serve to accumulate ultrasensitivity multiplicatively, but it can actually generate ultrasensitivity de novo in a concentration-dependent manner through a distribution of intermediate species. An explicit model that accounts for all physiologically relevant species, without simplifying assumptions (i.e., Hill equation and Michaelis-Menten approximations) or the inclusion of phosphatases, captures this mechanism of ultrasensitivity. MEK Inhibition of the Variable Cascade Based on our computational model, we predicted that applying 50 nM CI-1040 to the high-MEK variable cascade would decrease the ultrasensitivity, increase the threshold, and leave the signal strength unaffected. This is in contrast to the basic cascade in which the signal strength is predicted to be significantly lower upon MEK inhibition. As shown in Figure 5, both the basic and variable cascades are in the high signal strength region in the absence of MEK inhibition, yet with inhibition, they are separated by a sharp transition from low to high signal strength. In this parameter regime, the Hill coefficient and EC50 become decoupled from the signal strength of the system. Therefore, we tested MEK inhibition in the variable cascade and observed a reduction in ultrasensitivity from nH = 2.8 ± 0.19 to 1.7 ± 0.17 and an increase in EC50 from 6.6 ± 0.14 nM to 16 ± 1.1 nM (Figure 7A) as predicted by the model (Figure 7B). Strikingly, the basic cascade shows a significant reduction in signal strength, to 25% uninhibited levels (Figures 7C and 7D), whereas the response of the variable cascade is unaffected by MEK1 inhibition (Figures 7E and 7F). These computational and experimental results demonstrate both the ability to decouple signal characteristics and the great degree of plasticity exhibited by the MAPK topology itself through the use of multiple relatively simple perturbations. DISCUSSION In this study, we used a synthetic biology approach to explore factors regulating the systems-level properties of a minimal MAPK module. Synthetic biology is a rapidly evolving discipline that has both engineered novel biological functions and contributed to the current understanding of natural processes (Grunberg et al., 2010; Hasty et al., 2002; Khalil and Collins, 2010; Kiel et al., 2010; Sprinzak and Elowitz, 2005). By reconstructing a biological system, applying perturbations, and measuring the

A

Mass-Action Kinetic Model Single-step Reaction

nH2

Raf*

MEK

Raf + E

MEKp

nH3

ERK

MEKpp

ERKp

Raf*

MEKpp

Dual-step Reaction nH1 Raf + E

Hill Equation Coupled to Dual-step Tier

MEK

ERKpp

MEK*

ERK

nH_Total > nH1 * nH2 * nH3

E

ERK*

nH_Total > nH1 * nH2 * nH3

ERK

ERKp

ERKpp

nH_Total (Hill Eq.) < nH_Total (Mass Action) (same MEKpp dose-response)

B

Low Ultrasensitivity

High Ultrasensitivity

10, 10, 1000 (Raf, MEK, ERK nM) Dual Single

10, 1000, 1000 (Raf, MEK, ERK nM) Dual Single

Raf MEK ERK

C

1.0 1.2 1.3

1.00 1.00 1.05

1.0 1.1 3.8

MEK Complex (Low nH)

MEK Complex (High nH)

MEK Total (Low nH)

MEK Total (High nH)

0.8 0.6 0.4 0.2 0.0 -2

-1

0

Estradiol (log μM)

1

Normalized Response

1.0

-3

MEK Total (Input) MEK Complex (Hill Eq.) MEK Complex (Mass Action)

D

Single-step Kinetics Normalized Response

Normalized Response

Dual-step Kinetics

-4

0.85 0.91 1.90

1.0 0.8 0.6 0.4 0.2 0.0 -4

-3

-2

-1

0

Estradiol (log μM)

1

1.0 0.8 0.6 0.4 0.2 0.0 -4

-3

-2

-1

0

1

Estradiol (log μM)

Figure 6. Theoretical Analysis of Cascade-Generated Ultrasensitivity (A) Schematic of dual-step and single-step mass action kinetic models and a lumped Hill equation model coupled to a dual-step tier. (B) The Hill coefficients of each level of the cascade for both dual- and single-step mass action kinetic models for representative low- and high-ultrasensitivity systems. See also Table S2. (C) Simulations of distinct active MEK species for dual- and single-step cascades at the relative concentrations indicated in (B). (D) Comparison of simulated active MEK complex species with the Hill equation and mass action kinetic models for the same input function (total active MEK) in the high-ultrasensitivity system.

response, we can test and expand our understanding of how the system works and gain insight into how to modify it. Well-defined synthetic systems obviate some of the difficulties of natural systems that have an inherent degree of uncertainty with respect to molecules and topology. We used yeast as an ex vivo system to build a purely exogenous protein interaction network based on the mammalian ERK1/2 pathway. We also used mechanistic computational modeling to directly demonstrate conceptually new behaviors, such as cascade-based generation and concentration-based tuning of ultrasensitivity, independent of any experimental system.

Our simple synthetic cascade was used to investigate the role of scaffolding in shaping the activation profile of ERK. In contrast to the prozone effect, we see a monotonic dependence on scaffold concentration in which the signal strength is strong when the scaffold is either absent or expressed at optimal stoichiometric ratios and greatly attenuated at high scaffold concentrations. Our data show that the synthetic cascade is catalytically efficient, and therefore this monotonic response occurs because the cascade is not aided by the presence of scaffold. The disruption of signal strength seen at high scaffold concentrations can arise through multiple mechanisms, Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc. 127

1.2

Figure 7. MEK Inhibition of the Variable Cascade

B Normalized Fold Increase

Normalized Fold Increase

A

Expt

1.0 0.8 0.6 0.4 0.2 0.0 -3

-2

-1

0

1.2 1.0 0.8 0.6 0.4 0.2 0.0

1

-3

-2

D 100 80 60 40 20

120

Expt Percent Response

Percent Response

120

-1

0

Estradiol (log μM)

Estradiol (log μM)

C

Model

Model

100 80 60

1

(A) Normalized fold increase for the high-MEK variable cascade with (gray squares) and without (dark purple squares) pretreatment with 50 nM CI-1040 and for the basic cascade with (blue squares) and without (black squares) this inhibitor. The data are the mean ± SEM normalized to the fitted baselines for each condition. The data were fit with a modified Hill equation (solid lines). (B) Model simulations of the experimental conditions described in (A). (C) Percent response of the basic cascade with (blue squares) and without (black squares) 50 nM CI-1040 pretreatment. The data are the mean ± SEM normalized to the maximum activation of the basic cascade in the absence of inhibitor. (D) Model simulations of the experimental conditions described in (C). (E) Percent response of the high-MEK variable cascade with (gray squares) and without (dark purple squares) 50 nM CI-1040 pretreatment. The data are the mean ± SEM normalized to the maximum activation of the high-MEK variable cascade in the absence of inhibitor. (F) Model simulations of the experimental conditions described in (E).

40 20

istics can be decoupled under certain parameter regimes (Figure 5 and Figure 7). This suggests -3 -2 -1 0 1 -3 -2 -1 0 1 a potential cellular strategy for tuning the response Estradiol (log μM) Estradiol (log μM) of the MAPK cascade in distinct biological contexts without changing the module topology itself. E F Controlling protein expression levels through either 120 120 Expt Model gene expression or posttranslational modification 100 100 presents an effective and robust method for varying the activation profile of MAP kinases. Ground80 80 breaking work on ultrasensitivity in the Xenopus 60 60 oocyte Mos-MEK-p42 MAPK cascade revealed a sharply switch-like, irreversible response (Xiong 40 40 and Ferrell, 2003). Interestingly, the relative 20 20 concentrations of these proteins (Mos, 3 nM; 0 0 MEK, 1200 nM; p42 MAPK, 330 nM) (Ferrell, 1996) map the system to the high-ultrasensitivity -3 -2 -1 0 1 -3 -2 -1 0 1 region of our theoretical heat map (Figure 5, Estradiol (log μM) Estradiol (log μM) marked X). In contrast, the Ste11-Ste7-Fus3 MAPK cascade of the pheromone pathway in yeast has been shown to exhibit a less ultrasensitive actiincluding combinatorial inhibition or dampened amplification. vation profile (Hao et al., 2008; Poritz et al., 2001). Strikingly, the Combinatorial inhibition occurs when individual kinases become relative concentrations of these proteins (Ste11, 41 nM; Ste7, sequestered on separate scaffold molecules, as has been 37 nM; Fus3, 470 nM) (Ghaemmaghami et al., 2003) place this posited experimentally for the KSR1 and JIP1/2 scaffolds cascade in the low-ultrasensitivity region of our theoretical (Burack and Shaw, 2000) and computationally for a generic framework (Figure 5, marked Y). Though it is clear that factors two-member scaffold (Levchenko et al., 2000). More recent extrinsic to the module, such as positive feedback, scaffolding, theoretical work on the MAPK cascade found that scaffolds and subcellular compartmentalization, play important roles in lower the amplification and magnitude of a signal under condi- determining the activation profile of these natural signaling tions conducive to signal transduction in the absence of scaffold cascades, these systems may be primed for their respective (Locasale et al., 2007). This effect was mediated entirely by behaviors through the intrinsic parameter of kinase dampening of amplification through the cascade and did not concentration. We also identified kinase cascading as a mechanism for de involve combinatorial inhibition. We demonstrated that varying the relative concentrations of novo generation of ultrasensitivity, and we showed that this proteins intrinsic to the three-tiered MAPK module enables great effect is strongly concentration dependent, though it is not mediflexibility of the system response and that key signal character- ated by zeroth-order ultrasensitivity or solely by dual-step 0

Percent Response

Percent Response

0

128 Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc.

phosphorylation. We used a mass action kinetic model, based largely on the seminal work of Huang and Ferrell (1996), to explore the role of cascading in generating ultrasensitivity. The model presented in their work specifically accounted for dualstep dephosphorylation at each level of the cascade and was therefore subject to heightened ultrasensitivity through an additional dual-step process and through zeroth-order ultrasensitivity by creation of an activation-deactivation cycle. We systematically deconstructed this model, first removing phosphatases and then simulating activation of each kinase as a single-step reaction; by eliminating these additional sources of ultrasensitivity, we were able to analyze the inherent contributions of the kinase-cascading architecture in generating MAPK ultrasensitivity. We found that, over a broad range of relative kinase concentrations, the steepness of each subsequent tier increases in a greater than multiplicative manner. This result was surprising, given that previous theoretical studies (Brown et al., 1997; Ferrell, 1997) suggest that the ultrasensitivity of sequential levels in a cascade accumulates in a multiplicative or submultiplicative manner. In contrast, our analysis shows that, instead of simply passing steepness from one level to the next, cascading can lead to an output response whose Hill coefficient is greater than the product of the coefficients of the previous levels (Figure 6). The difference between our analysis and previous work arises because earlier models relied essentially on independent, though sequential, Hill equations in which there is one output species for a given input stimulus concentration. In our analysis, each level of the cascade contains a distribution of species (i.e., different phosphorylation and complexation states), and the concentrations of these species depend on both the kinetic parameters of each reaction and the total concentrations of Raf, MEK, and ERK. For a given input stimulus, this explicit apportioning of species results in a different output than the corresponding Hill equation approximation (Figure 6D), even at steady state, and the behaviors of these distinct species vary considerably as a function of concentration (Figure 6C). Furthermore, a biphasic Hill coefficient is a prominent feature of concentration variation seen under all conditions (Figure 5). Ultrasensitivity is low when either MEK or ERK is limited, and the extent of limitation for a given set of concentrations can be determined by the specific MEK species generated (Figure S4 and Figure S5). The optimum Hill coefficient will depend not only on the molarity of the individual kinases, but on parameters of activation as well. Previous work has shown that introducing compensatory alterations in the association and dissociation rate constants of Ras-Raf binding has a significant impact on the activation profile of ERK (Kiel and Serrano, 2009). Moreover, the influence of these different parameters is cell type specific, indicating that the plasticity of MAPK signaling is multifaceted. The steepness and magnitude of the biphasic Hill coefficient become enhanced through the addition of a phosphatase (Figure 5C) because of the introduction of an activation/deactivation cycle. The mechanism of dephosphorylation is also nonprocessive (Zhao and Zhang, 2001) and can contribute to the ultrasensitivity of the response. In addition, the generation of a cycle may further increase ultrasensitivity through zeroth-order effects in this region of parameter space. Interestingly, the addition of a cycle also results in a biphasic profile for the EC50 that is

in contrast to MEK inhibition, in which no enzymatic cycle is created. Under MEK inhibition, the response is similar to that of the unregulated system, but the region boundaries become shifted, creating further opportunities to independently tune certain signal characteristics. We demonstrate experimentally that the signal strength of the system can be decoupled from the EC50 and Hill coefficient (Figure 7) and identify additional possibilities for isolating distinct signal features through this theoretical analysis. Looking across each signal characteristic for different modes of regulation (Figure 5), it becomes apparent that some combinations of features are possible, whereas others do not arise (Table S1). We find that, for a low Hill coefficient, any combination of other signal aspects is attainable. However, to achieve a steep ultrasensitivity, the cascade must signal efficiently, which requires that the EC50 be low and the signal strength be high. To build a MAPK cascade with a sharply switch-like response and a high threshold, additional modes of regulation must be employed. From a synthetic biology perspective, this type of analysis is useful in designing novel signaling cascades. For example, several research groups have successfully engineered endogenous receptors with altered specificity (Dwyer et al., 2003; Looger et al., 2003) or expressed exogenous receptors (Chen and Weiss, 2005) that recognize novel ligands and relay the signal through endogenous pathways. We have identified mechanisms for tuning the threshold and ultrasensitivity of a ubiquitous relay module, and therefore these principles may be useful in altering biosensor sensitivity, building kinase-based logic gates, or tuning native signaling responses (such as differentiation, survival, or apoptosis) in engineering cells and tissues. Synthetic biology at the level of posttranslational modification is currently at the cusp of rapid discovery and growth (Grunberg et al., 2010; Kiel et al., 2010). In this study, we have shown how the key signal characteristics of strength, threshold, and ultrasensitivity in a ubiquitous and essential signaling module can be controlled through simple and accessible perturbations. EXPERIMENTAL PROCEDURES Plasmid and Strain Construction Complete tables of plasmids and strains used in this study can be found in Table S3 and Table S4, respectively. All plasmid construction was done using standard cloning methods for PCR amplification, restriction digest, ligation, and bacterial transformation. Integration vectors were derived from pRS4036, and episomal vectors were derived from pESC-leu and pESC-ura (all from Stratagene). The anhydrotetracycline-regulated tetO7 promoter and reverse tTA cDNA were cloned from pCM252 (EuroScarf). MKP-1L16A/L17A cDNA was obtained from Anton Bennett through Addgene plasmid 13478. Additional cDNAs were kindly provided by the following: His6-ERK and c-myc-MEK from Natalie Ahn, eGFP:DRaf-1:ER-DD from Steen Hansen, c-myc-paxillin from Lloyd Cantley, the pTy1 promoter from Gerald Fink, and the YIL117c promoter from David Levin. PaxillinY118D was generated by site-directed mutagenesis with standard PCR techniques and sequence verified. Yeast transformation was done with the LiAc method in selective synthetic complete dropout media. Single integration was confirmed by PCR screening of genomic DNA. Growth and Induction All strains were grown from a fresh colony in synthetic complete (SC) dropout media (0.67% yeast nitrogen base without amino acids, 2% glucose, 0.14% His-Leu-Trp-Ura dropout mix supplemented with appropriate amino acids) for 24 hr. Cells were grown with shaking at 30 C. Liquid culture was diluted

Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc. 129

into SC media with 2% galactose (no glucose) and 0.5 mg/ml aTC (Sigma Aldrich) and grown for 16 hr. One sample per estradiol titration was grown without aTC to determine the background. Cells were stimulated with b-estradiol (Sigma Aldrich) for 45 min for basic cascades and 120 min for variable cascades. For inhibitor experiments, cells were pretreated for 30 min with CI-1040 (Axon Medchem) prior to estradiol stimulation. One milliliter cultures (OD660 = 0.7–0.8) were lysed for analysis by incubating them in 1 ml fresh 0.1 M NaOH at room temperature for 5 min. They were then resuspended in 50 ml lysis buffer (120 mM Tris-HCl [pH 6.8], 10% glycerol, 4% SDS, 5% b-mercaptoethanol, and 0.004% bromophenol blue), incubated at 95 C for 3 min, and frozen at 20 C. Estradiol titration was done in triplicate for each strain.

Quantitative Western Blotting and Analysis Standard SDS-PAGE electrophoresis and transfer procedures were used. Primary antibodies used include anti-phospho-p44/42 MAP kinase (Cell Signaling Technology), anti-ERK (BD Biosciences), anti-a-tubulin (AbD Serotech), and anti-MEK (BD Biosciences). Secondary antibodies used with the Odyssey Infrared Imaging System (Li-Cor Biosystems) include goat-amouse-IRDye700DX (Rockland), goat-a-rabbit-IRDye800CW (Rockland), and donkey-a-rat-IRDye800CW (Rockland). Standard western blotting procedures were used with Odyssey blocking buffer (Li-Cor). PhosphoERK intensity was normalized by OD660 and total ERK signal or a-tubulin as loading controls. Each individual experiment was log transformed and fit to a modified Hill equation (1) with a nonlinear least-squares method using Prism 4 software (GraphPad Software) and then normalized to the fitted baselines (Ro and Rmax): R = Ro +

ðRmax  Ro Þ  LnH ðR  Ro Þ   max = Ro +  LnH + EC50 nH 1 + 10ðlogEC50 logLÞ nH

(1)

The left-hand equality is the standard Hill equation with lower and upper baselines Ro and Rmax, respectively; R is the system response (signal strength), L is the ligand concentration, nH is the Hill coefficient, and EC50 is the halfmaximal concentration of activation (threshold). The right-hand equality is a rearrangement of the Hill equation that was used to fit log-transformed data. The normalized data for each strain were pooled, and the mean and standard error of the mean were calculated with Prism 4.

Transcriptional Activation of Endogenous Pathways The minimal promoter of Fus1 (266 to +1) fused to tdTomato was custom synthesized by Geneart with NotI and BamHI sites flanking the promoter region. Additional promoters, a modified Ty1 promoter amplified from BMH261 (Madhani and Fink, 1997) and the YIL117c promoter excised directly from p1366 (Jung et al., 2002), were subcloned into NotI and BamHI. Transformed strains were grown as above and stimulated with either b-estradiol for 2 hr or positive controls: 2.5 mM a factor for 2 hr (pheromone and cell wall integrity), glucose starvation for 4 hr (invasive growth), and 40 mM caffeine for 6 hr (cell wall integrity). tdTomato expression was quantified by flow cytometry using a Guava flow cytometer (Millipore).

Model Formulation We used an ordinary differential equation (ODE)-based, deterministic approach to model the basic cascade and its variations. In the basic cascade, protein synthesis, binding reactions, dual-step phosphorylation, and protein degradation reactions were explicitly included. All binding and activation reactions were modeled with mass action kinetics and did not rely on approximations of rapid equilibrium or pseudosteady state (Michaelis-Menten). The series of ODEs for each model was solved using the numerical stiff solver ode23s in MATLAB (The Mathworks). Steady-state response plots, as well as 3D surface and 2D phase plots, were generated in MATLAB. Further descriptions of the various model formulations are given in the Extended Experimental Procedures, and model definitions, reactions, equations, parameters, and initial conditions are given in Table S5, Table S6, Table S7, Table S8, and Table S9, respectively.

130 Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc.

SUPPLEMENTAL INFORMATION Supplemental Information includes Extended Experimental Procedures, five figures, and nine tables and can be found with this article online at doi:10. 1016/j.cell.2010.12.014. ACKNOWLEDGMENTS This work was supported by the National Institutes of Health (NIH) through the NIH Director’s Pioneer Award Program, grant number DP1 OD00364, and by the Ellison Medical Foundation and the Howard Hughes Medical Institute (to J.J.C.); and by the American Heart Association, grant number 0835132N, and startup funds from the University of Pennsylvania (to C.A.S.). We also thank Natalie Ahn, Lloyd Cantley, Gerald Fink, Steen Hansen, and David Levin for kindly providing cDNA. Received: September 1, 2009 Revised: October 1, 2010 Accepted: December 10, 2010 Published: January 6, 2011 REFERENCES Atienza, J.M., Suh, M., Xenarios, I., Landgraf, R., and Colicelli, J. (2000). Human ERK1 induces filamentous growth and cell wall remodeling pathways in Saccharomyces cerevisiae. J. Biol. Chem. 275, 20638–20646. Avruch, J. (2007). MAP kinase pathways: the first twenty years. Biochim. Biophys. Acta 1773, 1150–1160. Bagowski, C.P., Besser, J., Frey, C.R., and Ferrell, J.E., Jr. (2003). The JNK cascade as a biochemical switch in mammalian cells: ultrasensitive and allor-none responses. Curr. Biol. 13, 315–320. Bashor, C.J., Helman, N.C., Yan, S., and Lim, W.A. (2008). Using engineered scaffold interactions to reshape MAP kinase pathway signaling dynamics. Science 319, 1539–1543. Bellı´, G., Garı´, E., Piedrafita, L., Aldea, M., and Herrero, E. (1998). An activator/ repressor dual system allows tight tetracycline-regulated gene expression in budding yeast. Nucleic Acids Res. 26, 942–947. Bhalla, U.S., Ram, P.T., and Iyengar, R. (2002). MAP kinase phosphatase as a locus of flexibility in a mitogen-activated protein kinase signaling network. Science 297, 1018–1023. Brown, G.C., Hoek, J.B., and Kholodenko, B.N. (1997). Why do protein kinase cascades have more than one level? Trends Biochem. Sci. 22, 288. Burack, W.R., and Sturgill, T.W. (1997). The activating dual phosphorylation of MAPK by MEK is nonprocessive. Biochemistry 36, 5929–5933. Burack, W.R., and Shaw, A.S. (2000). Signal transduction: hanging on a scaffold. Curr. Opin. Cell Biol. 12, 211–216. Chen, M.T., and Weiss, R. (2005). Artificial cell-cell communication in yeast Saccharomyces cerevisiae using signaling elements from Arabidopsis thaliana. Nat. Biotechnol. 23, 1551–1555. Chen, R.E., and Thorner, J. (2005). System biology approaches in cell signaling research. Genome Biol. 6, 235. Cullen, P.J., and Sprague, G.F., Jr. (2000). Glucose depletion causes haploid invasive growth in yeast. Proc. Natl. Acad. Sci. USA 97, 13619–13624. Dwyer, M.A., Looger, L.L., and Hellinga, H.W. (2003). Computational design of a Zn2+ receptor that controls bacterial gene expression. Proc. Natl. Acad. Sci. USA 100, 11255–11260. Elion, E.A. (2001). The Ste5p scaffold. J. Cell Sci. 114, 3967–3978. Ferrell, J.E. (1996). Tripping the switch fantastic: how a protein kinase cascade can convert graded inputs into switch-like outputs. Trends Biochem. Sci. 21, 460–466. Ferrell, J.E. (1997). How responses get more switch-like as you move down a protein kinase cascade. Trends Biochem. Sci. 22, 288–289.

Ferrell, J.E., and Machleder, E.M. (1998). The biochemical basis of an all-ornone cell fate switch in Xenopus oocytes. Science 280, 895–898. Ferrell, J.E., and Xiong, W. (2001). Bistability in cell signaling: How to make continuous processes discontinuous, and reversible processes irreversible. Chaos 11, 227–236. Gao, X.D., Caviston, J.P., Tcheperegine, S.E., and Bi, E. (2004). Pxl1p, a paxillin-like protein in Saccharomyces cerevisiae, may coordinate Cdc42p and Rho1p functions during polarized growth. Mol. Biol. Cell 15, 3977–3985. Ghaemmaghami, S., Huh, W.K., Bower, K., Howson, R.W., Belle, A., Dephoure, N., O’Shea, E.K., and Weissman, J.S. (2003). Global analysis of protein expression in yeast. Nature 425, 737–741. Goldbeter, A., and Koshland, D.E. (1981). An amplified sensitivity arising from covalent modification in biological systems. Proc. Natl. Acad. Sci. USA 78, 6840–6844. Grunberg, R., Ferrar, T.S., van der Sloot, A.M., Constante, M., and Serrano, L. (2010). Building blocks for protein interaction devices. Nucleic Acids Res. 38, 2645–2662.

Levchenko, A., Bruck, J., and Sternberg, P.W. (2000). Scaffold proteins may biphasically affect the levels of mitogen-activated protein kinase signaling and reduce its threshold properties. Proc. Natl. Acad. Sci. USA 97, 5818–5823. Locasale, J.W., Shaw, A.S., and Chakraborty, A.K. (2007). Scaffold proteins confer diverse regulatory properties to protein kinase cascades. Proc. Natl. Acad. Sci. USA 104, 13307–13312. Looger, L.L., Dwyer, M.A., Smith, J.J., and Hellinga, H.W. (2003). Computational design of receptor and sensor proteins with novel functions. Nature 423, 185–190. Mackeigan, J.P., Murphy, L.O., Dimitri, C.A., and Blenis, J. (2005). Graded mitogen-activated protein kinase activity precedes switch-like c-Fos induction in mammalian cells. Mol. Cell. Biol. 25, 4676–4682. Madhani, H.D., and Fink, G.R. (1997). Combinatorial control required for the specificity of yeast MAPK signaling. Science 275, 1314–1317. Marshall, C.J. (1995). Specificity of receptor tyrosine kinase signaling: transient versus sustained extracellular signal-regulated kinase activation. Cell 80, 179–185.

Hao, N., Nayak, S., Behar, M., Shanks, R., Nagiec, M., Errede, B., Hasty, J., Elston, T., and Dohlman, H. (2008). Regulation of Cell Signaling Dynamics by the Protein Kinase-Scaffold Ste5. Mol. Cell 30, 649–656.

McMahon, M. (2001). Steroid receptor fusion proteins for conditional activation of Raf-MEK-ERK signaling pathway. Methods Enzymol. 332, 401–417.

Harding, A., Tian, T., Westbury, E., Frische, E., and Hancock, J.F. (2005). Subcellular localization determines MAP kinase signal output. Curr. Biol. 15, 869–873.

Natarajan, M., Lin, K., Hsueh, R., Sternweis, P., and Ranganathan, R. (2006). A global analysis of cross-talk in a mammalian cellular signalling network. Nat. Cell Biol. 8, 571–580.

Hartwell, L.H., Hopfield, J.J., Leibler, S., and Murray, A.W. (1999). From molecular to modular cell biology. Nature 402, C47–C52.

Ohren, J.F., Chen, H., Pavlovsky, A., Whitehead, C., Zhang, E., Kuffa, P., Yan, C., McConnell, P., Spessard, C., Banotai, C., et al. (2004). Structures of human MAP kinase kinase 1 (MEK1) and MEK2 describe novel noncompetitive kinase inhibition. Nat. Struct. Mol. Biol. 11, 1192–1197.

Hasty, J., Mcmillen, D., and Collins, J. (2002). Engineered gene circuits. Nature 420, 224–230. Heinrich, R., Neel, B.G., and Rapoport, T.A. (2002). Mathematical models of protein kinase signal transduction. Mol. Cell 9, 957–970. Hornberg, J.J., Bruggeman, F.J., Binder, B., Geest, C.R., de Vaate, A.J., Lankelma, J., Heinrich, R., and Westerhoff, H.V. (2005). Principles behind the multifarious control of signal transduction. ERK phosphorylation and kinase/phosphatase control. FEBS J. 272, 244–258. Huang, C.Y., and Ferrell, J.E. (1996). Ultrasensitivity in the mitogen-activated protein kinase cascade. Proc. Natl. Acad. Sci. USA 93, 10078–10083. Ishibe, S., Joly, D., Zhu, X., and Cantley, L.G. (2003). Phosphorylation-dependent paxillin-ERK association mediates hepatocyte growth factor-stimulated epithelial morphogenesis. Mol. Cell 12, 1275–1285. Jung, U.S., Sobering, A.K., Romeo, M.J., and Levin, D.E. (2002). Regulation of the yeast Rlm1 transcription factor by the Mpk1 cell wall integrity MAP kinase. Mol. Microbiol. 46, 781–789. Khalil, A.S., and Collins, J.J. (2010). Synthetic biology: applications come of age. Nat. Rev. Genet. 11, 367–379. Kiel, C., and Serrano, L. (2009). Cell type-specific importance of ras-c-raf complex association rate constants for MAPK signaling. Sci. Signal. 2, ra38. Kiel, C., Yus, E., and Serrano, L. (2010). Engineering signal transduction pathways. Cell 140, 33–47. Kolch, W. (2005). Coordinating ERK/MAPK signalling through scaffolds and inhibitors. Nat. Rev. Mol. Cell Biol. 6, 827–837.

Poritz, M.A., Malmstrom, S., Kim, M.K., Rossmeissl, P.J., and Kamb, A. (2001). Graded mode of transcriptional induction in yeast pheromone signalling revealed by single-cell analysis. Yeast 18, 1331–1338. Qi, M. (2005). MAP kinase pathways. J. Cell Sci. 118, 3569–3572. Santos, S.D., Verveer, P.J., and Bastiaens, P.I. (2007). Growth factor-induced MAPK network topology shapes Erk response determining PC-12 cell fate. Nat. Cell Biol. 9, 324–330. Sprinzak, D., and Elowitz, M. (2005). Reconstruction of genetic circuits. Nature 438, 443–448. Whitehurst, A., Cobb, M.H., and White, M.A. (2004). Stimulus-coupled spatial restriction of extracellular signal-regulated kinase 1/2 activity contributes to the specificity of signal-response pathways. Mol. Cell. Biol. 24, 10145–10150. Wu, J.J., Zhang, L., and Bennett, A.M. (2005). The noncatalytic amino terminus of mitogen-activated protein kinase phosphatase 1 directs nuclear targeting and serum response element transcriptional regulation. Mol. Cell. Biol. 25, 4792–4803. Xiong, W., and Ferrell, J. (2003). A positive-feedback-based bistable ‘memory module’ that governs a cell fate decision. Nature 426, 460–465. Zhao, Y., and Zhang, Z.Y. (2001). The mechanism of dephosphorylation of extracellular signal-regulated kinase 2 by mitogen-activated protein kinase phosphatase 3. J. Biol. Chem. 276, 32382–32391.

Cell 144, 119–131, January 7, 2011 ª2011 Elsevier Inc. 131

Resource

Identification of Candidate IgG Biomarkers for Alzheimer’s Disease via Combinatorial Library Screening M. Muralidhar Reddy,1,2 Rosemary Wilson,1 Johnnie Wilson,1 Steven Connell,3 Anne Gocke,3 Linda Hynan,4 Dwight German,5 and Thomas Kodadek2,* 1Opko

Health Laboratories, 130 Scripps Way, Jupiter, FL 33458, USA of Chemistry & Cancer Biology, The Scripps Research Institute, Scripps Florida, 130 Scripps Way, #3A2, Jupiter, FL 33458, USA 3Division of Translational Research, Department of Internal Medicine 4Department of Clinical Sciences 5Department of Psychiatry University of Texas Southwestern Medical Center, 5323 Harry Hines Boulevard, Dallas, TX 75390, USA *Correspondence: [email protected] DOI 10.1016/j.cell.2010.11.054 2Departments

SUMMARY

The adaptive immune system is thought to be a rich source of protein biomarkers, but diagnostically useful antibodies remain unknown for a large number of diseases. This is, in part, because the antigens that trigger an immune response in many diseases remain unknown. We present here a general and unbiased approach to the identification of diagnostically useful antibodies that avoids the requirement for antigen identification. This method involves the comparative screening of combinatorial libraries of unnatural, synthetic molecules against serum samples obtained from cases and controls. Molecules that retain far more IgG antibodies from the case samples than the controls are identified and subsequently tested as capture agents for diagnostically useful antibodies. The utility of this method is demonstrated using a mouse model for multiple sclerosis and via the identification of two candidate IgG biomarkers for Alzheimer’s disease. INTRODUCTION There is great interest in the discovery of disease-specific protein biomarkers in easily accessible biological fluids such as serum. A particularly interesting subproteome in this regard is the IgG antibody population (Anderson and LaBaer, 2005). The adaptive immune system is known to react specifically to many different disease states, in part through the amplification of particular antibodies that recognize disease-specific antigens. Thus, it should be possible to devise diagnostic tests for many different diseases based on the measurement of the levels of these antibodies in serum. However, this has proven difficult. Because antibodies are highly specific receptors for their 132 Cell 144, 132–142, January 7, 2011 ª2011 Elsevier Inc.

cognate antigens, the general thinking is that a diagnostic test designed to monitor the level of a disease-specific antibody would require immobilized antigen as a ‘‘capture agent.’’ Unfortunately, there are many pathogenic conditions, including autoimmune diseases, neurological conditions, and cancers, for which the antigens that trigger the primary immune response are unknown, and thus, a definitive blood test is not available. To address this problem powerful proteomics technologies have been employed to screen large collections of expressed proteins, peptides, or other biomolecules in an attempt to discover native antigens recognized by disease-specific antibodies. Some notable successes have been achieved (Fathman et al., 2005; Frulloni et al., 2009; Gibson et al., 2010; Hudson et al., 2007; Kanter et al., 2006; Lueking et al., 2003; Robinson et al., 2002b; Steller et al., 2005; Wang et al., 2005). However, none of these techniques appears to represent a general route to the rapid discovery of antibody biomarkers of real diagnostic utility. It is reasonable to suspect that a limitation of screens that employ collections of unmodified peptides, proteins, or lipids is that they are unlikely to contain the primary autoantigens that trigger the earliest and most disease-specific autoimmune response. It seems more likely that these primary antigens are biomolecules that are chemically modified in unusual ways due to the pathogenic chemistry involved in that particular disease state. In other words, it may be that collections of unmodified biomolecules represent the wrong region of ‘‘chemical space’’ in which to be looking for autoantigens or mimics thereof. With this hypothesis in mind, we were interested in testing a fundamentally different approach in which a combinatorial library of unnatural synthetic molecules is screened for ligands that bind antibodies abundant in the serum of animals or patients with a particular disease, but not healthy controls. The idea behind this approach is that unnatural molecules will simply represent a ‘‘shape library’’ that occupies regions of chemical space outside of that represented by unmodified biomolecules. A few of these molecules might, by chance, recognize the antigen-binding pocket of disease-specific antibodies well

enough to retain them from the blood, though they would almost certainly not bind as well as the (unknown) native antigens. This is the thinking behind almost any high-throughput screen of synthetic molecule libraries or collections against protein drug targets of pharmaceutical interest. Moreover, whereas antibodies are generally not considered drug targets, it is known that antibody ligands with structures quite different from that of the native antigen can be isolated through library screening. For example, peptide libraries have been screened successfully for ‘‘mimotopes’’ that bind to carbohydrate-binding antibodies, and these peptides can even be used as vaccines to raise antibodies against the native carbohydrate antigen (Knittelfelder et al., 2009). However, to the best of our knowledge, all such mimotope screens have employed a single, well-defined antibody target and have not been utilized in de novo searches for diagnostically useful antibody biomarkers. We demonstrate here that microarrays displaying thousands of peptoids (N-substituted oligoglycines; Simon et al., 1992) can be used along with a differential screening strategy for the simultaneous isolation of candidate IgG antibody biomarkers and selective peptoid ligands able to pull them out of the blood. In two mouse models it is shown that these peptoids are antigen surrogates in the sense that they bind selectively to the antibodies raised against the antigen employed to trigger the disease state. This methodology provides an unbiased approach to the discovery of IgG serum biomarkers that does not require prior knowledge of native antigens. In this report we describe the development of this technology and its application to a mouse model for multiple sclerosis. We also demonstrate that the approach is applicable to the discovery of potentially useful diagnostic biomarkers in humans through the discovery of compounds that bind antibodies that are present at high levels in the serum of patients with Alzheimer’s disease (AD).

RESULTS

Figure 1. Schematic Representation of the Strategy Employed to Identify Synthetic Molecules that Capture Antibody Biomarkers The Y-shaped figures represent IgG antibodies. The figure depicts hypothetical binding of an antibody present at high levels in an autoimmune serum sample, but not in a healthy serum sample, binding to two compounds on a microarray. After subsequent probing with a fluorescently labeled secondary antibody, this would produce a much higher intensity at these two spots on the

Identification of Synthetic Ligands for EAE-Specific Antibodies in Mice Our strategy for the unbiased isolation of synthetic moleculeantibody complexes of diagnostic utility is depicted in Figure 1. We hypothesized that if one exposed crude serum containing the entire complement of circulating antibodies to a large library of synthetic molecules immobilized on a microscope slide, each antibody would, by chance, bind specifically to a few molecules in the library with sufficient affinity and specificity to be retained on the slide. This is the philosophy behind any protein-targeted combinatorial library screening experiment. IgG antibody binding to the array could be visualized by subsequent addition of a fluorescently labeled secondary antibody. Thus, the fluorescence intensity at each feature on the microarray would reflect the amount of antibody retained by each compound. Because the adaptive immune response results in the amplification of B cells that recognize ‘‘foreign’’ antigens, the antibodies they produce will be present at much higher levels in the blood of array (indicated in red scale) after exposure to the autoimmune serum sample than the healthy serum sample.

Cell 144, 132–142, January 7, 2011 ª2011 Elsevier Inc. 133

animals exposed to a particular antigen or with a particular disease than in the blood of matched control animals. Therefore, the features on the array to which they bind should ‘‘light up’’ to a far greater extent than would be the case when serum obtained from a control animal is hybridized to the array. If so, this would allow the unbiased identification of synthetic ligands for antibodies that are amplified in a particular immunological state. This approach does not require any foreknowledge of the nature of the antigens that drive the immune response. Note that this strategy is not an attempt to obtain a ‘‘fingerprint’’ or ‘‘signature’’ of a disease state. Rather it is an unbiased high throughput screen for a few specific antigen surrogates of diagnostic utility in a large synthetic combinatorial library. To explore this idea we employed microarrays comprised of two copies each of 4608 octameric peptoids along with various markers and control spots. The peptoid library (see Figure S1 available online) was synthesized using the split and pool method (Alluri et al., 2003; Figliozzi et al., 1996). Methods employed to construct the peptoid microarrays have been reported previously (Reddy and Kodadek, 2005). Although these molecules share the a-amino acid backbone of peptides, they are quite different otherwise, in that the side chains protrude from the main chain nitrogen (sp2 hybridized) rather than the a carbon (sp3 hybridized), giving them a completely different shape. Moreover, we employed a library of peptoids in which many of the side chains did not resemble any of the 20 naturally occurring amino acids (Figure S1). Therefore, the peptoid library represents a collection of molecular shapes that could not possibly have been ‘‘seen’’ by the immune system in vivo and could not mimic a native antigen closely. Two C57BL/6 mice were immunized with complete Freund’s adjuvant (CFA) and a peptide derived from myelin oligodendrocyte glycoprotein (Mog), whereas two additional mice were injected with CFA alone as a control. The Mog-immunized animals develop a syndrome called experimental autoimmune encephalomyelitis (EAE) that resembles human multiple sclerosis in some respects and is one of the most commonly used animal models for this disease (Hauser, 2008; McFarland and Martin, 2007). Serum samples were collected from the mice 36 days following immunization. They were then diluted several thousand-fold to provide a final total serum protein concentration of 15 mg/ml and hybridized to the peptoid microarray. After incubation and washing, the IgG-binding pattern was visualized by subsequent incubation with an Alexa 647-labeled secondary antibody. As a control, the secondary antibody alone was exposed to the array, and any features that bound significant amounts of the labeled secondary antibody were ignored in subsequent analyses. Figure 2A shows raw data from such an experiment. As expected, the Mog peptide employed as the antigen, when spotted onto the array as a positive control, captures large amounts of IgG antibodies from the serum collected from Mog/CFA-immunized mice, but not mice immunized with CFA alone. Several peptoids were identified that were reproducibly (n = 3 arrays) much brighter when exposed to the Mog/CFA-immunized serum than the CFA-immunized serum (an intensity of >40,000 versus <10,000 at this particular serum protein concentration at these instrument settings). One of these is highlighted in Figure 2A. 134 Cell 144, 132–142, January 7, 2011 ª2011 Elsevier Inc.

The peptoids, named anti-Mog antibody-binding peptoids 1–3 (AMogP1–3), that provided the highest level of discrimination were sequenced by tandem mass spectrometry using soluble peptoid from the well storing the solution used to spot onto the glass slide. Their structures bear no obvious resemblance to that of the peptide antigen (Figure 2 and Figure S2) and, in fact, could not have because many of the building blocks employed to construct the peptoid library were not analogs of natural amino acids (Figure S1). Validation of AMogP1–3 as Specific Diagnostic Biomarkers for EAE A critical issue in any biomarker discovery effort is to test the putative marker in a blinded fashion on subjects that were not employed in the ‘‘training set’’ used to discover the marker in the first place (Ransohoff, 2005). Therefore, the ability of peptoids AMogP1–3 to distinguish seven other Mog/CFA-immunized mice from seven other CFA-injected animals was analyzed in a series of blinded experiments. None of these mice had been used in the training set. As shown in Figure 3, all three peptoids performed perfectly in distinguishing the Mog peptide-immunized mice from the control mice. Figure 3A shows the quantification of data obtained in three independent experiments using ‘‘subarrays’’ (Figure 3B) that display AMogP1–3, the Mog peptide antigen, and a negative control peptide using serum obtained from mice 36 days following immunization with CFA or CFA + Mog peptide (EAE). The negative control peptide, which was used in all subsequent experiments, represents residues 1–11 of mouse myelin basic protein (MBP) (ASQKRPSQRSK). This peptide was chosen as a control for antigen-specific capture of antibodies because immunization of mice with it can also lead to EAE but through the production of different antibodies and T cells (Hauser, 2008). To test the selectivity of these peptoids for antibodies amplified in the Mog peptide-immunized mice, animals of the same genetic background were immunized with CFA and a peptide antigen derived from Ovalbumin (Ova; see Figure S2 for the peptide sequence). Serum collected from these mice 36 days after immunization displayed strong reactivity with the Ova peptide antigen (Figure 3C). However, only background levels of IgG antibody were captured by AMogP1–3 from this serum sample (Figure 3C). To further test their selectivity, serum was collected from B6 mice with systemic lupus erythematosus (SLE) (Liu and Wakeland, 2001) and exposed to these peptoids. Again, only background levels of antibodies were retained (Figure 3C). These data argue that the AMogP1–3 peptoids are selective capture agents for antibodies produced in Mog/CFAimmunized mice. We next asked if the peptoids were capable of tracking the development of the adaptive immune response that leads to EAE. Serum samples were collected from a Mog peptide/CFAimmunized mouse at various time points following antigen injection, and these serum samples were analyzed on a small subarray displaying peptoids AMogP1–3, the Mog peptide antigen, and the MBP-derived control peptide. As shown in Figure 4A, little or no signal above background was observed on any of the peptoids from the serum samples collected immediately prior to injection or 7 days later. However, by 14 days

postinjection, significant signal was seen on the peptoid features, and by 21 days all three EAE-specific peptoids evinced a strong signal (note that an intensity of 60,000–65,000 U represents saturation of the detector at these instrument settings). This time course is in line with the expected kinetics of the development of an adaptive immune response. The behavior of the peptoids as ligands for the EAE-specific antibodies was similar to that of the Mog peptide antigen, whereas the control peptoid did not capture significant amounts of antibody at any time point. We conclude that peptoids AMogP1–3 are capable of monitoring the development of an adaptive immune response over time.

Characterization of the Peptoid-Binding Antibodies Given that the antibodies captured by the AMogP1–3 peptoids arise over a time frame of 2–3 weeks following immunization with the Mog peptide, it seems likely that they indeed capture anti-Mog peptide antibodies, rather than antibodies directed against some other antigen that might arise as a secondary response to the developing autoimmune disease. To determine if this is indeed the case, serum from Mog/CFA-immunized mice was passed over an excess of Mog peptide coupled to sepharose or a control peptide column. These sera were then hybridized to a subarray that included AMogP1–3, Mog peptide, and the control peptide. As shown in Figure 4B, depletion of the anti-Mog antibodies from the serum of the Mog/CFA-immunized mice abolished IgG antibody binding to the immobilized peptoids, whereas passage of the serum over the control column resulted in strong binding of IgG antibody to the peptoids. As expected, the same result was observed when the level of antibodies captured by the Mog peptide was analyzed. We conclude that these peptoids bind to anti-Mog IgG antibodies.

Identification of Selective Ligands for Anti-Ova Peptide Antibodies To determine if this approach is useful to identify antibodies and cognate peptoid ligands unique to other immunological states, or if the EAE mice somehow represent a special case, we turned to the Ova peptide/CFA-immunized mice. These mice are healthy and are not afflicted with an autoimmune inflammatory disease, as is the case in EAE. Using exactly the same methods described above for the analysis of the Mog/CFA-immunized mice, we identified three peptoids that captured much higher levels of IgG antibodies from the serum of two Ova/CFAimmunized animals than from two CFA-immunized controls (Figure S3). These peptoids, called anti-Ova antibody-binding

Figure 2. Identification and Characterization of Peptoids that Capture Antibodies Present at High Levels in Mog Peptide-Immunized Mice (A) Raw images of peptoid arrays hybridized with serum obtained from CFA- or CFA +Mog peptide-immunized mice. About half of two arrays are shown at the top. The sections of the arrays boxed in blue are blown up to highlight a region displaying one of the peptoids (AMogP3) that clearly distinguished the CFAand CFA + Mog peptide-immunized mice. Images were obtained by incubating serum from immunized mice with the array followed by addition of fluorescently labeled (Alexa 647) Goat-anti Mouse IgG antibody. The intensity of the fluorescence at each spot is displayed in a false-colored red scale in

which a white spot means the intensity is beyond the linear range of the detector. The structure of AMogP3, the compound that is highlighted in the pink box, is shown as its free form. The molecule was tethered covalently to the array via the cysteine sulfur that is included in all of the molecules in the library. (B) Quantitation of the fluorescence intensity measured at each of the three peptoid (AMogP1–3) features on the array that discriminate CFA + Mog peptide- from CFA-immunized mice. The error bars indicate the standard deviation from the mean for three independent experiments. The general structure of the library employed to make the array is shown in Figure S1. The structures of the other two peptoids, AMogP2 and AMogP3, that distinguish control and EAE mice are shown in Figure S2.

Cell 144, 132–142, January 7, 2011 ª2011 Elsevier Inc. 135

Figure 3. Validation of Peptoids AMogP1–3 as Capture Agents for EAE-Specific Antibodies (A) ‘‘Subarrays’’ containing only AMogP1–3, the Mog peptide, and a MBP-derived negative control peptide were created. Serum from seven Mog/ CFA-immunized and seven CFA-injected mice not used in the previous experiments were analyzed in a blinded fashion. The fluorescence intensity observed at each feature is shown after unblinding the sample identities. Shown is the mean ± standard deviation (SD) for samples run in triplicate. (B) Raw images of subarrays containing the AMogP1–3, control peptide, and Mog peptide that were incubated with serum from a Mog + CFAimmunized mouse (left) or a CFA-immunized mouse (right). (C) Selectivity of peptoids for antibodies present in Mog peptide-immunized mice. Subarrays containing the AMogP1–3 peptoids, the Mog peptide, the Ova peptide, and a control peptide were exposed to serum from three mice immunized with Ova peptide (Ova1–3) or three mice with SLE (SLE1–3), followed by a fluorescently labeled secondary antibody. The fluorescence intensities at each feature are shown. Mean ± SD for samples run in triplicate is shown. Figure S2 displays the peptide sequences and peptoid structures.

peptoids 1–3 (AOvaP1–3), were sequenced by tandem mass spectrometry. Their structures are shown in Figure S2. To further explore the utility of these putative biomarkers for Ova peptide immunoreactivity, subarrays containing AOvaP1–3, the Ova peptide antigen, and the control peptide were employed in blinded experiments using serum obtained from eight OVA/ CFA-immunized and eight CFA-immunized mice not employed in the training set 21 days after immunization. As shown in Figures 5A and 5B, all three peptoids exhibited much higher signals when exposed to serum collected from the OVA/CFAimmunized mice than when exposed to serum for the CFAimmunized mice. The selectivity of the peptoids was tested by exposing them to serum derived from Mog peptide-immunized animals or mice with SLE. As shown in Figure 5C, very little cross-reactivity 136 Cell 144, 132–142, January 7, 2011 ª2011 Elsevier Inc.

between peptoids AOvaP1–3 and IgGs in these serum samples was observed. A time course experiment using serum samples collected 0, 7, 14, or 21 days after immunization with Ova peptide + CFA or CFA alone was conducted. This showed a buildup of antibodies that recognize both the peptoids and the Ova peptide over time. In this case 21 days were required to observe a robust response (Figure S4A). Finally, as shown in Figure S4B, depletion of the anti-Ova peptide antibodies from the serum of the Ova/CFA-immunized mice by passage over immobilized Ova peptide largely abolished IgG antibody binding to the peptoids, whereas passage of the serum over an immobilized control peptide did not. This argues that the peptoids recognize anti-Ova peptide antibodies. Discovery of Candidate IgG Biomarkers for Alzheimer’s Disease It was important to determine if the protocol described above is capable of identifying potentially useful diagnostic antibodypeptoid pairs for a human disease state. To address this point we undertook an effort to search for IgG antibodies that are enriched in the serum of patients with AD. AD is the most common form of dementia and is a rapidly growing problem in the developed world due to the aging of the population. Although careful clinical examination conducted by an expert, combined with

Figure 4. Peptoids AMogP1–3 Capture Anti-Mog Peptide Antibodies Resulting from an Adaptive Immune Response (A) Level of IgG antibody captured by the peptoids as a function of time after immunization. Subarrays displaying the molecules indicated were incubated with serum collected from mice at the indicated times after immunization with Mog peptide + CFA, followed by fluorescently labeled secondary antibody. The amount of fluorescence captured at each feature is shown. (B) Effect of depletion of anti-Mog peptide antibodies on the amount of IgG antibodies captured by the peptoids. Serum from Mog peptide + CFAimmunized mice was passed over columns displaying either excess Mog peptide or a control peptide. These Mog-depleted or mock-depleted serum samples were then hybridized to a subarray displaying AMogP1–3, Mog peptide, and a control molecule. After subsequent hybridization with labeled secondary antibody, the signal intensities were recorded and plotted. Mean ± SD for samples run in triplicate is shown.

radiological scans, is a reasonably effective method with which to diagnose the disease, the only unequivocal protocol for the diagnosis of AD is a postmortem autopsy of the brain (Hampel et al., 2010). To our knowledge, no blood test for this disease has yet been reported (Blennow et al., 2010), though an increase in the serum protein clusterin has recently been reported to accompany AD (Thambisetty et al., 2010). Using the same procedure described above for the EAE mice, serum samples from six patients with AD (McKhann et al., 1984) (three of which were autopsy confirmed; see Table S1 for clinical parameters) were analyzed using peptoid microarrays displaying approximately 15,000 peptoids. Each sample was analyzed in triplicate. The same procedure was carried out with serum samples obtained from six age-matched, nondemented control individuals (NC23-28, see Table S1). To attempt to ensure specificity for AD, we also analyzed six serum samples obtained from patients with Parkinson’s disease (PD) (Gelb et al., 1999) as well.

After measuring the signal level on each spot of each array, three peptoids were chosen that best distinguished the patients with AD from the controls. These peptoids captured at least 3-fold more IgG antibody from all six of the patients with AD than any of the control subjects (Figure 6) or patients with PD (Figure 7A). The structures of the three peptoids were deduced by tandem mass spectrometry and are shown in Figure 6. They were named AD peptoids (ADP) 1–3. To further test these three peptoids as potential capture agents for AD-specific antibodies, subarrays displaying resynthesized and HPLC-purified ADPs1–3 as well as controls were employed to analyze serum samples obtained from 16 different patients with AD, 16 new controls, and samples from six patients with lupus. These experiments were conducted in a blinded fashion. When the samples were unblinded, the data showed that peptoids ADP1–3 captured much more IgG antibodies from the patients with AD than 14 of the control individuals (Figure 6) or the six patients with lupus (Figure 7A). Two of the control individuals (NC31 and NC41) displayed an unusually high level of antibodies that cross-react with each of the three ADPs. This level was comparable to that found in the serum of patient 15 with AD, who exhibited the lowest levels of peptoid-reactive antibodies from the AD set. Overall, for the AD versus NC subjects, the sensitivity was 93.7% for each of the peptoids, and specificity was from 93.7% to 100%. The positive and negative predictive values were greater than 93% for each of the peptoids, and the accuracy ranged from 93% to 96%. Finally, the area under the curve for each of the three peptoids was 0.99 ± 0.01 (see Figure S5, and Table S2, Table S3, Table S4, Table S5, and Table S6 for a full statistical analysis). The fluorescent intensities of the ADP1–3 spots on the subarrays employed for these experiments were at or near the saturation point for the detector. To ensure that we were close to the linear range, measurements were redone at different serum protein concentrations. As shown in Figure S6A, the sample from a patient with AD demonstrated concentration-dependent signals that saturated at a total serum protein concentration of 15 mg/ml, the concentration employed for all of the experiments whose results are shown in Figure 6 and Figure 7. The intensity of the signal was also linear at different photomultiplier tube gains on the scanner (Figure S6B). We conclude that the ratios of the peptoid-binding antibodies in case and control samples presented in Figure 6 represent a valid quantitative comparison. To determine if peptoids ADP1, 2, and 3 bind the same antibodies, serum from one of the patients with autopsy-confirmed AD was passed repeatedly over a column containing immobilized ADP1 in order to completely deplete the sample of antibodies that bind to this peptoid. As a control, the same procedure was done using a column displaying a control peptide (MBP Ac1-11; see Figure S2 for sequence). The serum was then exposed to a subarray displaying ADP1–3, and the level of IgG antibodies captured by each peptoid from the ADP1depleted and mock-depleted samples was measured. As shown in Figure 7B, immobilized ADP1 bound high levels of antibody from the mock-depleted AD sample, but almost no IgG was captured from the ADP1-depleted sample, as expected. Depletion of the ADP1-binding antibodies also reduced the amount of IgG captured by ADP3 to background levels, arguing that these Cell 144, 132–142, January 7, 2011 ª2011 Elsevier Inc. 137

Figure 5. Validation of Peptoids Identified as Biomarkers of Ova Immunoreactivity (A) Subarrays’’ containing only AOvaP1–3, the Ova peptide, and a control peptide were created. Serum from seven Ova peptide + CFA-immunized and seven CFA-injected mice not used in the previous experiments were analyzed in a blinded fashion. The fluorescence intensity observed at each feature is shown after unblinding the sample identities. (B) Raw images of subarrays containing the control peptide, AOvaP1–3, and Ova peptide that were incubated with serum from Ova + CFAimmunized mouse (left) or a CFA-immunized mouse (right). (C) Selectivity of peptoids for antibodies present in Ova peptide-immunized mice. Subarrays containing the AOvaP1–3 peptoids, the Mog peptide, the Ova peptide, and a control peptide were exposed to serum from three mice immunized with Mog peptide (Mog1–3) or three mice with SLE (SLE1–3). The fluorescence intensities at each feature observed after probing with the fluorescently labeled secondary antibody are shown. Error bars represent the mean ± SD for samples run in triplicate. The structures of the Ova peptide antigen and the peptoids that distinguish Ova-immunized from control mice (AOvaP1–3) are shown in Figure S2. Figure S3 displays some of the primary data that led to the identification of AOvaP1–3 as discriminators of mice that were and were not immunized with Ova peptide. Figure S4 demonstrates that peptoids AOvaP1–3 bind anti-Ova peptide antibodies.

two peptoids capture the same antibodies. In stark contrast depletion of ADP1-binding antibodies had no effect on the amount of IgG captured by ADP2, arguing that this peptoid binds different antibodies than ADP1 and ADP3.

DISCUSSION We have described here a technology with which the entire complement of serum IgG antibodies can be screened against a peptoid library in order to identify complexes of diseasespecific antibodies and individual peptoids. This approach to the discovery of serum antibody biomarkers differs from other efforts of which we are aware in that it makes no attempt to directly identify the native antigen or a close mimic thereof by screening libraries of peptides, lipids, proteins, nucleic acids, or other naturally occurring molecules. Instead, the idea behind 138 Cell 144, 132–142, January 7, 2011 ª2011 Elsevier Inc.

this approach is that it should be possible to use high-throughput screening to identify a synthetic, unnatural molecule that happens to have the right shape and chemical functionality to bind the antigen recognition pocket of the antibody of interest well enough to pull it out of serum, even if said molecule does not bind as well as the native antigen. Although peptoids and peptides share an a-amino acid backbone, they are otherwise quite different in shape and chemical properties. For example the side chain in peptoids protrudes from the main chain nitrogen, which is sp2 hybridized, whereas the peptide a carbon is sp3 hybridized. In addition peptoids lack the N-H group in the main chain, which is often a contributor to hydrogen-bond interactions that stabilize either peptide secondary structures or interactions with a partner-binding protein. Finally, many of the side chains in the library of peptoids used in this study did not resemble the side chains of the 20 common amino acids (see Figure S1). Thus, this approach is quite distinct from previously reported screens of peptide libraries (Robinson et al., 2002a; Wang et al., 2005), which aim to identify a native epitope or at least a close relative. The peptoid molecules cannot possibly mimic a native peptide antigen closely.

Figure 6. Peptoids that Retain Antibodies from the Serum of Patients with Alzheimer’s Disease A peptoid library was screened for ligands to AD-specific IgG antibodies. The structures of the three best peptoids that were found to discriminate age-matched controls and patients with AD are shown in the top right. The levels of antibodies retained from the indicated serum samples in subsequent subarray experiments are shown on the left. The numbers indicate a patient identifier (e.g., AD1 or NC9; only every other number is shown). The samples employed in the training sets are labeled as such (AD Train and NC Train), as are the samples employed in blinded test studies. NC, normal control. The error bars indicate the mean ± SD for samples run in triplicate. See text for details. Table S2, Table S3, Table S4, Table S5, and Table S6, and Figure S5 present a detailed statistical analysis of these data as well as those shown in Figure 7A. Figure S6 demonstrates that the intensities shown in this figure represent the high end of the linear range of the assay.

As shown schematically in Figure 1, this type of screening was done using a microarray format that allowed comparison of the binding of antibodies from case and control serum samples to

thousands of peptoids. Molecules that retained far more IgG antibody from the case samples were considered candidate capture agents for IgG antibodies highly enriched in the disease Cell 144, 132–142, January 7, 2011 ª2011 Elsevier Inc. 139

Figure 7. Peptoids ADP1–3 Bind Two Different Antibodies that Are Present in the Serum of Patients with Alzheimer’s Disease, but Not Patients with PD or Lupus (SLE) (A) Comparison of levels of IgG antibodies captured by peptoids ADP1–3 from serum samples collected from a patient with AD (individual 1), a normal control (individual 23), or patients with PD or lupus (SLE). (B) Serum from a patient with autopsy-confirmed AD was passed repeatedly over immobilized ADP1 or, as a control, an irrelevant peptide. The serum samples were then diluted and hybridized to subarrays displaying peptoids ADP1–3. The amount of antibody captured by each peptoid was measured. Shown is the mean ± SD for samples run in triplicate. Table S2, Table S3, Table S4, Table S5, and Table S6, and Figure S5 present a detailed statistical analysis of these data as well as those shown in Figure 6.

state of interest. This protocol was first employed to test if peptoids could be identified that capture antibodies that distinguish between healthy mice and animals with EAE. The same analysis was applied to mice immunized with a peptide derived from Ova. In each case three peptoids, called AMogP1–3 and AOvaP1–3, respectively, were identified that captured high levels of IgG antibody from the serum of the immunized animals, but not the control animals. As must have been the case (see above), the peptoids bear no obvious resemblance to the Mog and Ova peptide antigens, respectively (Figure S2), even though it was demonstrated that these molecules indeed bind the antibodies raised against the peptide antigens (Figure 4 and Figure S4). Subsequent validation studies with samples obtained from mice not used in the training set validated these peptoids as excellent capture agents for antibodies unique to the EAE and Ova peptideimmunized mice, respectively (Figure 3 and Figure 5). 140 Cell 144, 132–142, January 7, 2011 ª2011 Elsevier Inc.

Of course the more important question is whether this approach is relevant to the discovery of peptoid-antibody complexes that might be of utility in medical diagnostics. Although the mouse work was encouraging and proved the principle of using libraries of unnatural molecules to search for autoantibody ligands, this study employed relatively homogeneous laboratory mice and simple, single-antigen immunization models. The greater immunological diversity between different people than between different laboratory mice might complicate the application of this technology to the discovery of biomarkers for human disease. To address this important question, we carried out a preliminary study of serum samples collected from patients with AD. It has been reported that patients with AD have lower levels of serum anti-amyloid antibodies than healthy individuals (Weksler et al., 2002). Although this difference is not sufficient to act as the basis of a diagnostic test, it does suggest the possibility of finding more useful antibody markers of the disease. Peptoid libraries were screened using serum samples from six patients with AD (three autopsy confirmed), six matched control individuals, and six patients with PD. Three peptoids were identified that captured at least 3-fold higher levels of IgG antibodies from all six of the patients with AD than any of the controls or patients with PD (Figure 6 and Figure 7). The structures of these peptoids are shown in Figure 6. Depletion of ADP1-binding antibodies from the serum of AD samples demonstrated that ADP1 and ADP3 bind the same IgG antibodies, whereas ADP2 binds different antibodies (Figure 7B). Thus, we have discovered at least two candidate autoantibody biomarkers for AD. Subsequent blinded studies were conducted using samples from more patients with AD, controls (see Table S1), and patients with a different disease (lupus) to test the utility of the peptoids identified in the original screen. These ‘‘validation samples’’ were not employed in the training set and, thus, constitute a fair and critical test of the utility of the peptoid-antibody complexes as biomarkers (Ransohoff, 2005). Once unblinded, the data (Figure 6 and Figure 7) showed that these peptoid antibody complexes are indeed highly promising biomarkers for the diagnosis of AD (see Figure S5 and Table S2, Table S3, Table S4, Table S5, and Table S6 for the results of a comprehensive statistical analysis). Two of the control individuals displayed a relatively high level of the AD antibodies (NC31 and 41), similar to that seen in the patient with AD with the lowest levels of antibodies (AD15) (Figure 6). The individual from whom the NC31 sample was collected is a 75-year-old female with a mini-mental state examination (MMSE) score (McKhann et al., 1984) of 29 out of a possible 30 and without obvious clinical signs of AD (Table S1). The NC41 sample was from a 65-year-old female with the same MMSE score. These may represent examples of false positives, for example due to the cross-reaction of non-AD associated antibodies with these peptoids, or could represent presymptomatic detection of developing disease. Because these samples contained high levels of antibodies that bind to ADP2 as well as antibodies that bind to ADP1 and ADP3, we favor the latter hypothesis, but this cannot be concluded with certainty. The development of a simple blood test for AD is an important unrealized goal (Blennow et al., 2010). This preliminary study is

promising in that it represents a high level of diagnostic sensitivity and specificity (Saah and Hoover, 1997), at least within the relatively limited range of samples analyzed. However, it is important to point out that more work will be required before it is clear whether the peptoids ADP1–3 will be useful reagents for the clinical diagnosis of AD. First, the analysis of a larger number of patient samples derived from a more diverse population will be required. Second, it will be important to test samples collected from patients with mild cognitive impairment (MCI) that subsequently progressed to AD because early detection of developing disease is an important clinical goal. Third, all of the measurements done in this study were conducted on a microarray platform that may not be easily employed in a clinical setting, so optimized conditions for using the peptoids on other analytical platforms will have to be developed. Fourth, if these biomarkers are indeed validated, then it will be of great interest to identify the native antigens that they recognize. Studies to address all of these issues are in progress. The experiments in this paper were designed solely to address the issue of whether this technology is applicable to the discovery of biomarker candidates for human disease. We conclude that this is indeed the case. In summary, we have developed and validated a technology based on parallel screens of synthetic combinatorial libraries for the discovery of IgG biomarkers and simple, synthetic capture agents capable of retaining them from serum. We believe that this technology will have a significant impact on the development of diagnostic tests for a variety of important diseases. EXPERIMENTAL PROCEDURES General Remarks All chemicals and solvents were purchased from commercial suppliers and used without further purification. Secondary antibodies were obtained from Molecular Probes (Goat anti-mouse IgG—Alexa 647) and Goat anti-mouseRPE (Invitrogen). The slides were scanned using GenePix Autoloader 4200AL scanner (Molecular Devices, Sunnyvale, CA, USA) at 10 mm resolution using 635 nm laser at 100% power and 650 photomultiplier tube gain. All the scanned images were analyzed by the GenePix Pro 6.0 software (Axon Instruments, Union City, CA, USA). Peptoid Library Synthesis and Microarray Construction General protocols for the creation of peptoid libraries and peptoid-displaying microarrays have been published previously (Alluri et al., 2003; Figliozzi et al., 1996; Olivos et al., 2002; Reddy and Kodadek, 2005). A detailed protocol for the creation of the particular libraries and arrays used in this study is provided in the Extended Experimental Procedures. Animal Experiments C57BL/6 mice were purchased from Jackson Laboratory (Bar Harbor, ME, USA) and bred and maintained in a federally approved animal facility at the University of Texas Southwestern Medical Center (Dallas, TX, USA) in accordance with the regulations of the Institutional Animal Care and Use Committee. All mice were between 7 and 10 weeks of age when the experiments were initiated. For actively induced EAE and OVA immunizations, 30 female C57BL/6 mice were injected subcutaneously (s.c.) at four sites over the flanks and shoulders with 200 mg MOG 35–55 peptide (MEVGWYRSPFSRVVHLYRNGK) (CS Bio, Menlo Park, CA, USA) or OVA 323–339 peptide (ISQAVHAAHAEINEAGR) (University of Texas Southwestern Medical Center, Dallas, TX, USA) in an emulsion with CFA (Difco, Detroit, MI, USA). For control experiments, mice

were injected with CFA alone. Pertussis toxin (200 ng/mouse) (List Biological Laboratories) was injected i.p. at the time of immunization and 48 hr later to enhance the upregulation of adhesion molecules and to facilitate blood brain barrier breakdown. Mice were scored on a scale of 0–6: 0, no clinical disease; 1, limp/flaccid tail; 2, moderate hind limb weakness; 3, severe hind limb weakness; 4, complete hind limb paralysis; 5, quadriplegia or premoribund state; 6, death. Retro-orbital bleeds were performed every 7 days following immunization, starting at day 0 and continuing to day 50 post-immunization on mice in all groups (OVA, MOG, and CFA). Serum was collected by centrifugation and frozen at 20 C until used. Array Analysis of Serum Samples Peptoid slides were covered with a hybridization chamber and equilibrated with TBST (50 mM Tris [pH 8.0], 150 mM NaCl, 0.1% Tween 20) for 15 min. The slides were then blocked with 1 ml of blocking buffer for 1 hr at 4 C. The blocking buffer was removed, and the slides were incubated with 1 ml of serum (adjusted to 15 mg/ml of total protein) for 18 hr at 4 C with gentle shaking. Microarrays were then washed three times with TBST and hybridized with Alexa 647-labeled Goat anti-mouse antibody (1:400 dilution) for 2 hr on an orbital shaker at 4 C. The chamber cassettes were removed from the microarray slides and washed with TBST (3 3 15 ml), followed by 0.13 TBST (1 3 10 ml). The slides were then dried on a centrifuge (5 min at 1500 rpm) and scanned on a microarray scanner (GenePix Autoloader 4200 from TeleChem International, Inc., Sunnyvale, CA, USA) by using the 635 nm laser at 100% power and a 650 photomultiplier tube gain. All the scanned images were analyzed using GenePix Pro 6.0 (Axon Instruments, Union City, CA, USA) and GeneSpring software (Silicon Genetics, Redwood City, CA, USA). The experiments were done in triplicate, and each group of three included slides printed in different batches to avoid bias due to batch-to-batch differences in the slides. The GenePix Results (GPRs) were generated by using GenePix Pro 6.0 software. Local background subtracted median (F635 MedianB635) spot intensities were used for further analysis. These signal intensities were used for downstream analysis using GeneSpring software. All the GPRs were loaded onto the GeneSpring, and only features that gave greater than 40,000 signal intensity in immunized mice and less than 10,000 in control mice were selected and saved to the results’ folder. These results were transferred to the Excel, and peptoids with the highest signal intensity differential and reproducibility in all of the experiments were selected for further testing. The same criteria were used to analyze all the test experiments on a subarray. Analysis of Human Serum Samples The human serum samples were analyzed in exactly the same way as described for the mouse serum samples, except that the serum was diluted to a total protein concentration of 15 mg/ml prior to hybridization to the array. A mouse anti-human secondary antibody was employed to monitor the level of human IgG captured on a spot. Depletion of Mog Peptide-Binding Antibodies from the Serum All the reagents were allowed to equilibrate to room temperature. The resin from the SulfoLink immobilization kit for peptides (Thermo Scientific Inc.; 44999) was suspended by rocking and then the column was placed in a 15 ml tube and centrifuged at 1000 rpm for 1 min to remove the storage buffer. The column was equilibrated by adding 2 ml of coupling buffer followed by centrifugation. Peptide (1 mg) was dissolved in 2 ml of coupling buffer and added to the column. The top and bottom caps were replaced, and the resin was mixed by rotating the column end over end at room temperature for 15 min. The column was placed upright and incubated for 30 min at room temperature without mixing. The top and bottom caps were removed, and the column was placed into a new 15 ml tube and centrifuged at 1000 rpm for 1 min to collect the nonbound peptide. The column was washed by adding 2 ml of wash solution, followed by centrifugation. This step was repeated three times for a total of four washes. The column was washed again by adding 2 ml of coupling buffer, and the column was then centrifuged. This was repeated once for a total of two washes. L-cysteine-HCL (15.8 mg) was dissolved in 2 ml of coupling buffer (50 mM cysteine), and the solution was added to the column. The resin was mixed for 15 min at room temperature, then incubated

Cell 144, 132–142, January 7, 2011 ª2011 Elsevier Inc. 141

for 30 min without mixing. The top cap was removed first and then the bottom cap to allow the column to drain. The column was centrifuged at 1000 rpm for 1 min to remove the nonbound cysteine. The column was washed with 2 ml of 13 TBST, and serum sample was added to the column. After the serum sample entered the resin bed, the top and bottom caps were replaced, and the column was incubated for 60 min at room temperature. The top and bottom caps were removed, and the column was centrifuged to collect the flowthrough serum. Finally, the resin was washed by adding 2 ml of binding/ wash buffer.

Hauser, S.L. (2008). Multiple lessons for multiple sclerosis. N. Engl. J. Med. 359, 1838–1841.

SUPPLEMENTAL INFORMATION

Knittelfelder, R., Riemer, A.B., and Jensen-Jarolim, E. (2009). Mimotope vaccination—from allergy to cancer. Expert Opin. Biol. Ther. 9, 493–506.

Supplemental Information includes Extended Experimental Procedures, six figures, and six tables and can be found with this article online at doi:10. 1016/j.cell.2010.11.054.

Liu, K., and Wakeland, E.K. (2001). Delineation of the pathogenesis of systemic lupus erythematosus by using murine models. Adv. Exp. Med. Biol. 490, 1–6.

ACKNOWLEDGMENTS We thank Drs. Mike Racke, Amy Lovett-Racke, and Ward Wakeland for contributing EAE and SLE samples in the early phase of this project. We thank Kristin Martin-Cook of the UT Southwestern Medical Center’s Alzheimer’s Disease Center for selecting the normal and AD serum samples for analysis and Dr. Padraig O’Suilleabhain for diagnosis and collection of the PD serum samples. This work was supported by an NIH Director Pioneer Award to T.K. (DP1OD000663) and the NHLBI Proteomics Initiative of the National Heart, Lung & Blood Institute, National Institutes of Health (contract No. NO1-HV28185). Human serum sample collection was supported by grant NIH grant AG12300. Received: February 25, 2010 Revised: August 4, 2010 Accepted: November 19, 2010 Published: January 6, 2011

Hudson, M.E., Pozdnyakova, I., Haines, K., Mor, G., and Snyder, M. (2007). Identification of differentially expressed proteins in ovarian cancer using high-density protein microarrays. Proc. Natl. Acad. Sci. USA 104, 17494– 17499. Kanter, J.L., Narayana, S., Ho, P.P., Catz, I., Warren, K.G., Sobel, R.A., Steinman, L., and Robinson, W.H. (2006). Lipid microarrays identify key mediators of autoimmune brain inflammation. Nat. Med. 12, 138–143.

Lueking, A., Possling, A., Huber, O., Beveridge, A., Horn, M., Eickoff, H., Schuchardt, J., Lehrach, H., and Cahill, D.J. (2003). A non-redundant human protein chip for antibody screening and serum profiling. Mol. Cell. Proteomics 2, 1342– 1349. McFarland, H.F., and Martin, R. (2007). Multiple sclerosis: a complicated picture of autoimmunity. Nat. Immunol. 8, 913–919. McKhann, G., Drachman, D., Folstein, M., Katzman, R., Price, D., and Stadlan, E.M. (1984). Clinical diagnosis of Alzheimer’s disease: report of the NINCDSADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer’s Disease. Neurology 34, 939–944. Olivos, H.J., Alluri, P.G., Reddy, M.M., Saloney, D., and Kodadek, T. (2002). Microwave-assisted solid phase synthesis of peptoids. Org. Lett. 4, 4057– 4059. Ransohoff, D.F. (2005). Lessons from controversy: ovarian cancer screening and serum proteomics. J. Natl. Cancer Inst. 97, 315–319. Reddy, M.M., and Kodadek, T. (2005). Protein ‘‘fingerprinting’’ in complex mixtures with peptoid microarrays. Proc. Natl. Acad. Sci. USA 102, 12672– 12677.

REFERENCES

Robinson, W.H., Steinman, L., and Utz, P.J. (2002a). Protein and peptide array analysis of autoimmune disease. Biotechniques, 66–69.

Alluri, P.G., Reddy, M.M., Bacchawat-Sikder, K., Olivos, H.J., and Kodadek, T. (2003). Isolation of protein ligands from large peptoid libraries. J. Am. Chem. Soc. 125, 13995–14004.

Robinson, W.H., DiGennaro, C., Hueber, W., Haab, B., Kamachi, M., Dean, E., Fournel, S., Fong, D., Genovese, M.C., de Vegvar, H., et al. (2002b). Antigen arrays for multiplex characterization of autoantibody responses. Nat. Med. 8, 295–301.

Anderson, K.S., and LaBaer, J. (2005). The sentinel within: exploiting the immune system for cancer biomarkers. J. Proteome Res. 4, 1123–1133. Blennow, K., Hampel, H., Weiner, M., and Zetterberg, H. (2010). Cerebrospinal fluid and plasma biomarkers in Alzheimer disease. Nat. Rev. Neurol. 6, 131–144. Fathman, C.G., Soares, L., Chan, S.M., and Utz, P.J. (2005). An array of possibilities for the study of autoimmunity. Nature 435, 605–611. Figliozzi, G.M., Goldsmith, R., Ng, S.C., Banville, S.C., and Zuckermann, R.N. (1996). Synthesis of N-substituted glycine peptoid libraries. Methods Enzymol. 267, 437–447. Frulloni, L., Lunardi, C., Simone, R., Dolcino, M., Scattolini, C., Falconi, M., Benini, L., Vantini, I., Corrocher, R., and Puccetti, A. (2009). Identification of a novel antibody associated with autoimmune pancreatitis. N. Engl. J. Med. 361, 2135–2142.

Saah, A.J., and Hoover, D.R. (1997). ‘‘Sensitivity’’ and ‘‘specificity’’ reconsidered: the meaning of these terms in analytical and diagnostic settings. Ann. Intern. Med. 126, 91–94. Simon, R.J., Kania, R.S., Zuckermann, R.N., Huebner, V.D., Jewell, D.A., Banville, S., Ng, S., Wang, L., Rosenberg, S., Marlowe, C.K., et al. (1992). Peptoids: a modular approach to drug discovery. Proc. Natl. Acad. Sci. USA 89, 9367– 9371. Steller, S., Angenendt, P., Cahill, D.J., Heuberger, S., Lehrach, H., and Kreutzberger, J. (2005). Bacterial protein microarrays for identification of new potential diagnostic markers for Neisseria meningitidis infections. Proteomics 5, 2048–2055.

Gelb, D.J., Oliver, E., and Gilman, S. (1999). Diagnostic criteria for Parkinson disease. Arch. Neurol. 56, 33–39.

Thambisetty, M., Simmons, A., Velayudhan, L., Hye, A., Campbell, J., Zhang, Y., Wahlund, L.O., Westman, E., Kinsey, A., Guntert, A., et al. (2010). Association of plasma clusterin concentration with severity, pathology, and progression in Alzheimer disease. Arch. Gen. Psychiatry 67, 739–748.

Gibson, D.S., Banha, J., Penque, D., Costa, L., Conrads, T.P., Cahill, D.J., O’Brien, J.K., and Rooney, M.E. (2010). Diagnostic and prognostic biomarker discovery strategies for autoimmune disorders. J. Proteomics 73, 1045–1060.

Wang, X., Yu, J., Sreekumar, A., Varambally, S., Shen, R., Giachero, D., Mehra, R., Montie, J.E., Pienta, K.J., Sanda, M.G., et al. (2005). Autoantibody signatures in prostate cancer. N. Engl. J. Med. 353, 1224–1235.

Hampel, H., Frank, R., Broich, K., Teipel, S.J., Katz, R.G., Hardy, J., Herholz, K., Bokde, A.L., Jessen, F., Hoessler, Y.C., et al. (2010). Biomarkers for Alzheimer’s disease: academic, industry and regulatory perspectives. Nat. Rev. Drug Discov. 9, 560–574.

Weksler, M.E., Relkin, N., Turkenich, R., LaRusse, S., Zhou, L., and Szabo, P. (2002). Patients with Alzheimer disease have lower levels of serum antiamyloid peptide antibodies than healthy elderly individuals. Exp. Gerontol. 37, 943–948.

142 Cell 144, 132–142, January 7, 2011 ª2011 Elsevier Inc.

Resource

Phenotypic Landscape of a Bacterial Cell Robert J. Nichols,1,2 Saunak Sen,3 Yoe Jin Choo,2 Pedro Beltrao,4 Matylda Zietek,2 Rachna Chaba,2 Sueyoung Lee,2 Krystyna M. Kazmierczak,5 Karis J. Lee,2,8 Angela Wong,2,9 Michael Shales,4 Susan Lovett,6 Malcolm E. Winkler,5 Nevan J. Krogan,4 Athanasios Typas,2,* and Carol A. Gross2,7,* 1Oral and Craniofacial Sciences Graduate Program, University of California, San Francisco, 513 Parnassus Avenue, San Francisco, CA 94143, USA 2Department of Microbiology and Immunology, University of California, San Francisco, 600 16th Street, San Francisco, CA 94158, USA 3Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, CA 94107, USA 4Department of Cellular and Molecular Pharmacology and The California Institute for Quantitative Biomedical Research, University of California, San Francisco, San Francisco, CA 94158, USA 5Department of Biology, Indiana University, Bloomington, IN 47405, USA 6Department of Biology and Rosenstiel Basic Medical Sciences Research Center, Brandeis University, Waltham, MA 02454-9110, USA 7Department of Cell and Tissue Biology, University of California, San Francisco, San Francisco, CA 94158, USA 8Present address: UCLA School of Dentistry, 10833 Le Conte Avenue, Los Angeles, CA 90095-1762, USA 9Present address: Massachusetts General Hospital, 50 Staniford Street, Boston, MA 02114, USA *Correspondence: [email protected] (A.T.), [email protected] (C.A.G.) DOI 10.1016/j.cell.2010.11.052

SUMMARY

The explosion of sequence information in bacteria makes developing high-throughput, cost-effective approaches to matching genes with phenotypes imperative. Using E. coli as proof of principle, we show that combining large-scale chemical genomics with quantitative fitness measurements provides a high-quality data set rich in discovery. Probing growth profiles of a mutant library in hundreds of conditions in parallel yielded > 10,000 phenotypes that allowed us to study gene essentiality, discover leads for gene function and drug action, and understand higher-order organization of the bacterial chromosome. We highlight new information derived from the study, including insights into a gene involved in multiple antibiotic resistance and the synergy between a broadly used combinatory antibiotic therapy, trimethoprim and sulfonamides. This data set, publicly available at http://ecoliwiki.net/ tools/chemgen/, is a valuable resource for both the microbiological and bioinformatic communities, as it provides high-confidence associations between hundreds of annotated and uncharacterized genes as well as inferences about the mode of action of several poorly understood drugs. INTRODUCTION Before the physical basis of genes was understood, associating phenotypes with a heritable unit laid the foundation of modern genetics. Following discovery of the genetic code, linking a

phenotype to the responsible gene remained the most expeditious way to unravel gene function. With the explosion of sequence information, the balance has shifted. We now have many genes of unknown function. To capitalize on the burgeoning sequence bank, it is imperative to develop high-throughput technologies that link genes to phenotypes and expedite discovery of gene function. This is particularly true for prokaryotes, which represent a major fraction of the sequenced genomes and are in the forefront of metagenomic efforts (Qin et al., 2010). Chemical and environmental perturbations have traditionally linked phenotypes to genotypes through forward genetic screens, but reverse genetic approaches are being increasingly utilized (Barker et al., 2010). Phenotype microarrays utilize a high-resolution readout of cellular respiration to evaluate fitness of a strain in hundreds of conditions (Bochner, 2009). This approach is appropriate for studying a few strains but is difficult to expand to genome-scale screens. In pooled growth competitions, thousands of strains are assayed in a single culture environment. Fitness values are derived from measuring strain abundance in a test relative to control condition (Giaever et al., 2004; Girgis et al., 2009; Hillenmeyer et al., 2008; Hoon et al., 2008; Lee et al., 2005; Pan et al., 2004; Parsons et al., 2006; Warner et al., 2010; Xu et al., 2007). These approaches are very efficient, but competition between strains in each condition makes it difficult to determine relative strain growth across conditions (Girgis et al., 2009), especially for strains that grow slowly even in the absence of perturbation (Lee et al., 2005). Arraying mutant strains on solid media allows independent evaluation of strain fitness but has been used only for low-resolution measurements of entire libraries (Liu et al., 2010; Tamae et al., 2008) or for essential genes (Pathania et al., 2009). Highthroughput genetic interaction studies, pioneered in yeast (Schuldiner et al., 2005; Tong et al., 2001), are complementary to chemical genomics approaches. Such analyses quantitatively Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc. 143

measure colony growth of double mutants in high-density format on agar surfaces and have led to numerous successes in identifying gene function and network organization (Beltrao et al., 2010). Similar methodology has been developed for E. coli (Butland et al., 2008; Typas et al., 2008). We use E. coli to illustrate the power of applying the high-resolution quantitative fitness measurements of genetic interaction analysis to high-throughput phenotypic analysis of culturable microbes. ‘‘Phenomic profiling’’ provides a quantitative description of the response of all single-gene deletions to physiologically relevant stresses and drug challenges. By profiling 4000 genes in > 300 perturbations, we identified thousands of phenotypes and a diverse suite of conditionally essential genes. This approach provides new insights into the chromosome organization, functional landscape, and evolutionary trajectory of E. coli. It facilitates high-confidence association of genes of unknown function to those of known function, as highlighted by discovery of the role of a gene involved in multiple antibiotic resistance in this manuscript and identification of two lipoproteins essential for peptidoglycan synthesis (Typas et al., 2010). Finally, the degree to which various gene deletions alter toxic drug effects has led to powerful insights regarding drug mode of action (Kohanski et al., 2008), and we demonstrate that our analysis generates numerous leads concerning drug function. RESULTS AND DISCUSSION Phenomic Profiling of E. coli K12 Yields a Robust, High-Quality Data Set We determined quantitative growth scores for the Keio singlegene deletion library (Baba et al., 2006); essential gene hypomorphs (C-terminally tandem-affinity tagged [Butland et al., 2005] or specific alleles); and a small RNA/small protein knockout library (Hobbs et al., 2010) in conditions representing the range of stresses that E. coli encounters. Mutant strains arrayed in high density on agar plates (1536 colonies/plate) were grown in 324 conditions covering 114 unique stresses (Figure 1A and Figure S1A and Table S1 available online). Colony sizes were analyzed and converted to a drug-gene score using an approach developed for quantifying genetic interactions (see Extended Experimental Procedures). More than half of the conditions were antibiotic/antimicrobial treatments (Figure 1A). By using a subinhibitory concentration series that maximally inhibited growth of the wild-type (WT) strain % 50%, we were able to search for specific drug-gene interactions (Figure S1A) and reduce the ability of spontaneous suppressor mutations to overtake the colony. Two independently derived clones of each mutant strain were analyzed (for sRNA mutants, a single isolate was arrayed twice), and screens were performed at least twice, enabling scores to be based on four to six independent measurements. Correlation between replicate colony size measurements was very high (r = 0.77; Figure 1B). The final data set (Table S2) was comprised of scores for the 3979 mutant strains passing quality control (e.g., proper normalized colony size distribution and replicate reproducibility; see Extended Experimental Procedures). The entire data set is available in an interactive, searchable format and as a flat file on the E. coli wiki website (at http://ecoliwiki.net/tools/chemgen/). 144 Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc.

The entire matrix (3979 mutants 3 324 conditions) was subjected to two-dimensional (2D) hierarchical clustering (Figure 1C). Drugs with similar effects cluster on the x axis; mutants that behaved similarly cluster on the y axis. Notably, concentrations of the same drug, drugs of the same family, and/or similar conditions clustered together, as did mutants of genes known to be part of the same operon, biological pathway, and/or protein complex. Zoomed insets of our clustergram illustrate examples. Genes in the rfa operon (rfaG, rfaP, rfaQ, rfaB and rfaI), which encodes enzymes that synthesize the inner- and outer-lipopolysaccharide (LPS) core, strongly cluster together with three/four genes that are responsible for the synthesis of one of the sugar building blocks, ADP-L glycero-b-D-manno heptose (rfaD, rfaE, and lpcA). Importantly, clustering reflects their shared sensitivities to a concentration series of compounds known to perturb the envelope integrity of the cell, consistent with the role of LPS. dsbA and dsbB, encoded in different operons, also cluster. The DsbA/DsbB complex generates disulfide bridges in the periplasm. The response of each mutant strain across all conditions is denoted as its ‘‘phenotypic signature.’’ High correlation between two phenotypic signatures is highly predictive of known indicators of functional connection between genes. Gene pairs with correlation coefficients (r) between 0.6 and 0.8 (p < 1034) are more than 100-fold enriched for genes sharing common operon membership and 150-fold enriched for genes with known protein interactions determined from low-throughput experiments (http://www.ecocyc.org; Figure 1D). This benchmarking analysis indicates that our phenomics data set is biologically meaningful. Correlated phenotypic signatures also reproduce connections between curated biological pathways (Figure S1B). For example, electron transfer components cluster tightly (e.g., nuo genes encoding NADH dehydrogenase I complex; Figure 1C). Their clustering reflects high sensitivity to membrane-perturbing stresses, including detergents, dyes, and metals, and increased resistance to aminoglycosides, in agreement with early studies that illustrate decreased aminoglycoside uptake in the absence of a fully functional electron transport chain (Girgis et al., 2009; Taber et al., 1987). All three examples described in Figure 1C are consistent with the expectation that highly correlated phenotypic signatures are biologically meaningful (r R 0.60.8). Phenomic Profiling Defines Responsive and Conditionally Essential Genes A central goal of this study was to systematically evaluate the impact of every gene deletion on E. coli fitness in diverse environments, as few gene deletions in E. coli have robust reported growth phenotypes, and only 8% of the genes are essential in rich media (Baba et al., 2006; Yamamoto et al., 2009). We used a statistical method to define a reliable phenotype. In brief, we standardized the interquartile range of the distribution of scores for each screen and then determined the probability that each condition-gene interaction represented a true phenotype using a normal cumulative distribution function (see Extended Experimental Procedures). Using a 5% probability that these phenotypes arose by chance as a cutoff (false discovery rate [FDR] % 5%), 49% of all strains tested (1957/ 3979 strains; Figure 2A) had one or more phenotypes. We refer

A

Stresses/Conditions Tested

C

Drug/Antibiotic Targets

Antibiotic/Drug (14) Antimicrobial/Antifungal (4) Antimicrobial (13)

Drug (39)

Multiple/unknown/poorly defined target (34)

Peptidoglycan (48)

Dye (15)

324

Membrane (15)

BILE-0.5% BILE-1.0% BILE-2.0% DEOXYCHOLATE-2.0% DEOXYCHOLATE-0.5% DEOXYCHOLATE-0.1% SDS0.5%/EDTA0.5 SDS1.0%/EDTA0.5 SDS0.5%/EDTA0.1 SDS-0.5% SDS-1.0% SDS-2.0% SDS-3.0% SDS-4.0% BENZALKONIUM-10 BENZALKONIUM-25 DIBUCAINE-0.4 DIBUCAINE-0.8 DIBUCAINE-1.2 NOVOBIOCIN-30 TRICLOSAN-0.05 ACRIFLAVINE-10 ETHIDIUMBROMIDE-50 ACRIFLAVINE-2 ETHIDIUMBROMIDE-10 ETHIDIUMBROMIDE-2 PROPIDIUMIODIDE-20 PROPIDIUMIODIDE-50 MINOCYCLINE-0.2 MINOCYCLINE-0.5 MINOCYCLINE-1.0 PUROMYCIN-25 PUROMYCIN-5 PYOCYANIN-10.0 MITOMYCINC-0.1 STREPTONIGRIN-0.5

Detergent (28) LPS (9) Fatty acid metabolism (8)

Antibiotic (139) Chemical stress (26)

DNA (23)

C/N source (8)

Protein synthesis (48)

Environmental stress (24) Hormone (5) Metal stress (9)

B

Division (4) Transcription (8) THF biosynthesis (12)

ECK3622-RFAQ ECK3617-RFAI ECK3618-RFAB ECK3620-RFAP ECK3621-RFAG ECK3610-RFAF ECK3609-RFAD ECK0223-LPCA ECK3042-RFAE

1000

750

3,979

Colony Size - Replicate 2

r = 0.77

140

500

120

ECK2272-NUOL ECK2281-NUOB ECK2273-NUOK ECK2282-NUOA ECK2274-NUOJ ECK2276-NUOH ECK2270-NUON ECK2278-NUOF ECK2271-NUOM ECK2279-NUOE

100 80 60

250

40

0

250

500

750

1000

Colony Size - Replicate 1

D

200

Fold Enrichment

Protein-Interactions Common Operon

150

Score 10.0 7.5 5.0 2.5 0.0 -2.5 -5.0 -7.5 -10.0

100

MECILLINAM-0.03 CEFSULODIN-6.0,MECILLINAM-0.03 FUSIDICACID-1 FUSIDICACID-5 FUSIDICACID-20 FUSIDICACID-50 NALIDIXICACID-0.5 NALIDIXICACID-1.0 NALIDIXICACID-1.5 NALIDIXICACID-2.0 AZIDOTHYMIDINE-0.5 AZIDOTHYMIDINE-1.0 AZIDOTHYMIDINE-2.5 ISONIAZID-1.0 ISONIAZID-1.5 MMS-0.05% CIPROFLOXACIN-0.006 CIPROFLOXACIN-0.008 CIPROFLOXACIN-0.004 NORFLOXACIN-0.02 NORFLOXACIN-0.01 NORFLOXACIN-0.04 MECILLINAM-0.06 MECILLINAM-0.09 MECILLINAM-0.12 CEFSULODIN-18.0 CEFSULODIN-24.0 CEFTAZIDIME-0.05 CEFTAZIDIME-0.075 EDTA-0.1 EDTA-0.5 EDTA-1.0 PH10 PH9.5 PH9 PH8

20

0

ECK3852-DSBA ECK1173-DSBB

50

0 <0

0 - 0.2

0.2 - 0.4

0.4 - 0.6

0.6 - 0.8

0.8 - 1

Correlation Coefficient

Figure 1. Phenomic Profiling of the Enhanced Keio Collection Yields a Robust and Rich Data Set (A) Classification of the 324 stresses screened (left) and cellular targets of the antibiotic/antimicrobial/drug classes (right). (B) Heat map representation of scatter plot comparing normalized colony sizes in pixels of plate replicates 1 and 2 across the entire data set. Bins indicate the square root of the number of replicate pairs within a 10310 pixel window as depicted by color scale. Note that the vast majority of the replicates have highly correlated colony sizes. (C) Clustergram of fitness scores for 3979 mutant strains in response to all 324 conditions. Zoomed insets demonstrate coclustering of conditions (x axis) and genes (y axis) for a common pathway (rfa cluster) and protein complexes encoded in the same operon (nuo) or in different operons (dsbA and dsbB). Gray boxes indicate missing data. (D) High correlation between a pair of phenotypic signatures is predictive of shared protein interaction and/or operon membership. See also Figure S1, Table S1, and Table S2.

to these genes as the ‘‘responsive genome.’’ This responsive genome is a work in progress, as it is limited to genes whose removal causes growth phenotypes in response to the stresses tested. Expanding the stresses tested and/or the readout (e.g., motility) will certainly increase this number (Girgis et al., 2007). A cumulative plot of the number of individual phenotypes per strain shows that very few genes have many phenotypes. Multi-stress responsive (MSR) strains (R30 phenotypes; Table S3) participate in many cellular processes, suggesting that our stresses encompassed diverse cellular challenges (Figure 2B). With a stringent cutoff of 5% FDR, the maximum number of phenotypes from a single screen was 173 (4% strains; Figure S2A), and the total number of phenotypes (13497) represents 1% of all condition-gene pairs tested. Overall, 80% of the phenotypes were negative (gene deletion more sensitive) and 20% positive (gene deletion more resistant), consistent with

recent genetic interaction analyses in S. cerevisiae (Fiedler et al., 2009) and S. pombe (Roguev et al., 2008). This suggests that removal of a gene product is more likely to decrease than enhance resistance to stress (Figure S2B). In summary, our analysis captured numerous highly specific condition-gene responses. Clearly, this data set can be used to assign more phenotypes at a lower confidence level. Indeed, a recent chemical genomics data set in S. cerevisiae reported phenotypes for more than 95% of gene deletions tested, many stemming from a handful of severe stresses (Hillenmeyer et al., 2008). Conditionally essential (CE) genes are essential for growth in a particular condition. Deletions of such genes shows very small colony sizes and high-confidence negative scores in particular conditions (see Extended Experimental Procedures). We identified 197 CE genes, comprised of auxotrophs, which exhibit no growth in minimal media, and rich-media CE genes, which Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc. 145

B

A

C

D

Figure 2. Identification of Responsive and Conditionally Essential Genes (A) Using a 5% false discovery rate (FDR), 49% of strains tested had at least one phenotype (open circle on the red line). As the FDR is relaxed, more phenotypes are identified (red line). At 5% FDR, some strains have several phenotypes (black), and very few (2.3%; 94 strains) have 30 or more phenotypes (green, multi-stress responsive [MSR] genes). (B) MSR genes participate in a wide variety of cellular processes, particularly those related to metabolism and the cell envelope. Genes were manually curated to COG-based functions; each gene was allowed to belong only to a single function. (C) 196 genes are conditionally essential (CE) in this study. Of these, roughly half have been previously described as CE due to auxotrophy. Note that some auxotrophic genes also display a no growth phenotype in at least one rich medium condition and are classified jointly as auxotroph and rich media CE. (D) CE gene products are enriched in the outer cell envelope (periplasm and outer membrane) relative to Keio essential genes (p = 0.00026), highlighting the importance of this compartment in tolerating stress. The cytoplasmic gene category is not displayed here but is not enriched for CE gene products. See also Figure S2, Table S3, Table S4, and Table S5.

exhibit no growth in at least one rich media-based stress (Table S4). Importantly, our data set had 70% overlap with a previous study of Keio Collection auxotrophs (Figure 2C; Joyce et al., 2006) despite significant experimental differences (e.g., growth in liquid versus solid media). Many of the remaining 30% were extremely sick but above the stringent threshold that we used to define no growth. We also identified 23 additional auxotrophs specific to alternative carbon/nitrogen sources that were not tested in the Joyce et al. study. Genes that are essential for survival in natural environments are likely to extend beyond those required for laboratory growth and could be targets for new antimicrobials (D’Elia et al., 2009). The 116 rich-media CE genes that we identified (Figure 2C) result from physiologically relevant stresses and increase the current number of essential genes by roughly 30%. Interestingly, many of these gene products are located in the outer cell envelope 146 Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc.

(Figure 2D), a selective permeability barrier for Gram-negative bacteria that is severely underrepresented for known essential genes (Figure 2D). Many of the stresses generating CE phenotypes are part of the natural environment of E. coli, e.g., bile salts (Table S5), indicating that these genes are likely indispensable for E. coli to survive in vivo. Similarly, using the largest metagenomic data set to date, Qin et al. reported that envelope-specific functions, such as adhesion, were commonly required for life in the gut (Qin et al., 2010). Phenomic Profiling Helps Assign Function to Uncharacterized Genes A key motivation for our study was to provide phenotypes for mutants of genes without functional annotation. Using a recently assembled compendium of such ‘‘orphan genes’’ in E. coli (Hu et al., 2009), we find that the fraction of mutant orphan genes

B

A 1

% Annotated / % Ophans

All genes Annotated Orphans

Fraction of Genes

0.8

0.6

7 6 5 4 3 2 1 0

0.4

0

5

0.8

Fraction of Group Correlating to Annotated

Fold Enrichment for Annotated Genes 8

10 15 20 25 30 35 40 Number of Phenotypes

0.2

0.6 0.5 0.4 0.3

10-22

0.2

10-34

0.1

10-50

10-75 10-120

0

0 0

1

2

5

10

20

30

0.4

40

0.5

Number of Phenotypes Functions of Annotated Genes with High Correlation to Orphan Genes

Protein quality control RNA/tRNA/rRNA processing-modification Signal transduction Toxin-antiitoxin Transcription

D

Post-translational modification/enzyme inhibitor

Metabolism

Extracellular structures/proteins Transport

Energy production/conversion DNA replication/recombination/repair

General function prediction only Unknown function Other major processes

0.6

0.7

0.8

0.9

1

Correlation Coefficient

log2 (Broadly Distributed/ Proteobacteria-Only)

C

Annotated Orphans

10-14

0.7

Cell motility & chemotaxis Cell envelope biogenesis/outer membrane Cell division/cell cycle

3 Annotated Orphans 2

1

0

-1

-2

-3 0

5

10

15

20

25

30

35

Number of Phenotypes

Figure 3. Phenomic Profiling Identifies Phenotypes for Orphan Gene Mutants (A) Cumulative distribution of phenotypes indicating the fraction of gene mutants in each class having at least the number of phenotypes shown on the x axis. The plot reveals that orphan gene mutants have phenotypes but tend to have fewer phenotypes than annotated gene mutants. The insert quantifies phenotype deficit of orphan mutants. (B) Cumulative distribution of highly correlated pairs identifies many orphan genes that correlate highly to an annotated gene, providing high-confidence clues to the function of the orphan gene. Values shown above each pair of bars are the p values associated with pairwise correlation of any two strains at the indicated correlation coefficients. (C) High-confidence correlations between orphans and annotated genes (r R 0.5) provide leads that are related to many different cellular functions. Procedure for functional assignment is described in Figure 2. Note that several ‘‘annotated genes’’ were classified as genes of ‘‘unknown function’’ or ‘‘general function prediction only’’ after manual curation. (D) Annotated genes that are responsible for many phenotypes tend to be broadly conserved, whereas the most responsive orphan genes tend to be restricted to g-proteobacteria. For a list of the orphans-annotated pairs (r > 0.5), see Table S6.

with phenotypes is close to that of annotated genes (Figure 3A), but the former tend to have fewer phenotypes, indicating the power of phenomic analysis for identifying their phenotypes. Importantly, the phenotypic profiles of > 25% of all orphan genes correlate strongly with those of annotated genes (r R 0.5; Figure 3B and Table S6), providing high-confidence leads (p < 1022) for discovery of their function. As these orphans are tied to a wide variety of cellular processes (Figure 3C), the data set will be of broad utility.

A small fraction of orphan gene knockouts have many phenotypes. Whereas annotated genes that are responsible for many phenotypes are broadly distributed among bacteria, the most responsive orphans tend to be narrowly distributed (Figure 3D). This result suggests that evolutionary conservation is not a reliable indicator of the importance of an orphan gene to the organism and that annotating them solely by homology has limitations. Such orphans may have evolved to fulfill an important but specialized function required by the niche of the organism. Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc. 147

marR

marB

marA

acrB

acrA

A

C

6

5

acrB marA

Correlation Coefficient 1.00 0.67 0.33 0.00 -0.33 -0.67 -1.00

marB

-0.50 -0.46

marR

0.64

4

3

2

1

300,000

Strains per Bin

log2(relative marA transcript level)

acrA

250,000 0

200,000

WT

150,000

marB:kan

marR-

marR-/ marB:kan

100,000 50,000 0

-1

-0.8 -0.6 -0.4 -0.2

0

0.2

0.4

0.6

0.8

1

Correlation Coefficient

B marR

marA

marB

MarR MarR

MarA

MarB

Figure 4. A Function for marB

(A) marR and marB phenotypic signatures are highly correlated with each other and are highly anticorrelated with that of acrB (top). The bottom graph positions these correlations in a histogram showing all pair-wise correlation coefficients between the 3979 mutants. (B) Schematic of the E. coli multiple antibiotic resistance (mar) operon. marB is a gene of unknown function, but our results suggest that it encodes a protein that inhibits MarA. (C) RT-PCR analysis shows that marA transcription is derepressed in marB cells. Derepression is independent of and additive with that of marR.

In support of this idea, a multiresponsive orphan identified in this study (lpoB) is restricted to enterobacteria and regulates peptidoglycan synthesis, a conserved process that is ubiquitous among bacteria (Typas et al., 2010). Using Phenotypic Signatures to Identify Gene Function Both correlated phenotypic signatures (Figures 1C and 1D; Typas et al., 2010) and anticorrelated phenotypic signatures have functional significance. For example, the phenotypic signatures of deletions of a transcriptional repressor and important target genes are likely to be anticorrelated. We find that marR and marB were highly anticorrelated with acrB, whereas marR and marB were highly correlated (Figure 4A). marB is a gene of unknown function in the multiple antibiotic resistance 148 Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc.

operon (marRAB), which also includes the operon repressor marR and its activator, marA. MarA also activates genes involved in antibiotic resistance, most importantly acrAB, encoding the major antibiotic efflux pump in E. coli (Figure 4B) (Martin and Rosner, 2002). We explored whether MarB, like MarR, repressed MarA. Because of the inherent problems of high-throughput collections (suppressors, gene duplications, and cross-contamination), we always apply stringent quality control procedures to any follow-up investigations, including PCR validation of Keio isolates and verification that retransduced strains maintain their phenotype. As mar is a hot spot for adaptive mutations, we also sequenced the entire operon and promoter region of single deletions and the double mutants that we constructed.

Deletion of either marB or marR resulted in higher MarA levels, and the double-marRB mutant showed additive effects on MarA transcript level (Figure 4C) and protein level (data not shown). These effects were observed in both Kan-marked and clean deletions (Kan cassette excised, leaving an 82 nt scar). The DmarR strain exhibits 23 more increase in MarA transcript levels than marR::kan (data not shown), arguing for a small polar effect of the cassette. Both marB::kan and DmarB exhibit the same 2-fold increase in MarA levels (data not shown). These data suggest that MarB represses MarA independently of MarR. MarB does not have the signature of a DNA-binding protein, suggesting that it acts posttranscriptionally. MarA level is controlled by the Lon protease (Griffith et al., 2004), but lon and marB effects are additive, indicating that MarB does not function through Lon (data not shown). MarA has been proposed to scan for activation sites while bound to RNA polymerase; by direct binding to either partner, MarB could disrupt complex formation. Alternatively, MarB may function in the periplasm. As MarB has a predicted periplasmic signal sequence, it could titrate an activating ligand for mar (e.g., salicylate). Although mar is highly studied (200 primary publications; PubMed), our screen provided the first lead for MarB function. MarA targets 40 genes, many of which are also coregulated by the SoxS and Rob activators, with similar DNA-binding motifs as MarA (Martin et al., 2008; Martin and Rosner, 2002). The rules of engagement are poorly understood, but each activator responds to different environmental cues, and overexpression of each leads to distinct phenotypes (Warner and Levy, 2010). It is likely that tight control of each activator has an impact on the final gene expression output, which is crucial for cellular proliferation. MarB may be an important player in fine-tuning the expression of MarA, especially because it is a conserved member of the mar operon, which has only recently emerged in selected enterobacteria. Strong evidence for the importance of mar operon regulation in these organisms is that mar is a hot spot for mutations conferring higher drug resistance in E. coli (Nicoloff et al., 2006, 2007). Phenomic Profiling Reveals Metabolic Network Behaviors under Antifolate Drug Stress Tetrahydrofolate (THF) and its methyl/formylated derivatives are key molecules in all kingdoms of life for one-carbon metabolism. THF is used to synthesize glycine, methionine, purines, and dTTP in a process that leads to recycling of the THF species back to THF or dihydrofolate (DHF) (Figure 5A). The bacterial THF biosynthesis pathway is targeted by two drugs: sulfonamides (Sulfa) target FolP, and trimethoprim (TMP) targets FolA (Figure 5A). Dual inhibition by Sulfa and TMP is strongly synergistic, and therefore these drugs are almost exclusively administered in combination for treatment of ear, urinary tract, and bronchial infections. Despite extensive clinical use and years of laboratory investigation, we lack a complete mechanistic understanding of why these drugs are strongly synergistic. A network feature identified by phenomic profiling could contribute to synergy. We find that the two drug classes have major phenotypic differences. Sulfa and TMP treatments are highly correlated within their class (r = 0.57 for Sulfa, 0.67 for TMP) but poorly

correlated with each other (r = 0.15 ± 0.04), just slightly more than the correlation observed between all screens (r = 0.025 ± 0.12). Thus, subinhibitory TMP and Sulfa treatments have fundamentally different effects on the cell even though both partially block THF biosynthesis. Importantly, removing enzymes acting directly downstream of THF production resulted in opposite drug sensitivities: the serine hydroxymethyltransferase mutant (glyA::kan) was sensitive only to TMP; conversely, glycine cleavage (GCV) mutants (gcvP::kan, gcvH::kan, and gcvT::kan) were sensitive only to Sulfa (Figure 5B). The mutant results were reproduced in liquid culture (Figure 5D), in which glyA TMP sensitivity is manifested as a growth rate phenotype (left), and gcvP sulfamethizole (SMT) sensitivity is registered as a low stationary phase density (right). GlyA and GCV lie on opposite sides of a branched pathway that converts THF to 5,10-methylene THF (5,10-mTHF; Figure 5A). As glyA and gcv mutants exhibit synthetic lethality, they are the only routes to production of this essential metabolite (Figure 5C). A simple explanation for the differential responses of glyA and gcvP is that 5,10-mTHF is predominantly produced via different branches under each drug treatment. A corollary is that combination drug treatment inhibits both branches, resulting in synergistic limitation for 5,10-mTHF, before the pools of THF are depleted. In support of this idea, despite the increased sensitivity of glyA and gcvP to single drugs, these strains grew no more poorly than WT under the drug combination (Figure 5D). Thus, genetically eliminating either branch of the pathway reduced but did not eliminate synergy. The downstream biosynthetic reactions are also differentially affected by TMP and Sulfa (Figure S3A), and we are currently testing whether they partially account for the residual synergy. Streptococcus pneumoniae lacks the GCV system and exhibits significantly less drug synergy than E. coli across different growth conditions (Figure 5E and data not shown). We performed our comparison using concentrations of TMP and SMT that caused the same relative growth defect in each species (Figure 5E). These data together support the hypothesis that simultaneous inhibition of the branched pathway for production of 5,10-mTHF contributes to the observed antifolate synergy in E. coli. Our data do not indicate whether differential effects of TMP and Sulfa on GlyA and GCV result from differential inhibition of expression or activity or the intrinsic properties of each enzyme. We favor the idea that differential metabolite accumulation and subsequent feed-forward enzymatic regulation make a contribution to the distinct cellular responses to these two drugs. Recent metabolomic flux analyses indicate that high doses of TMP lead to accumulation and depletion of select metabolites, as well as to protein-level regulation of portions of the network (Kwon et al., 2008, 2010). Although a comparable analysis has not been performed for Sulfa drugs, deletion of the predicted 5-formyl-THF cycloligase, ygfA, which likely degrades 5-formyl-THF (Jeanguenin et al., 2010), clusters tightly with the gcv mutants and exhibits sensitivity only to Sulfa drugs (Figure S3A). That 5-formyl-THF degradation is critical only under Sulfa stress suggests differential accumulation (or requirement) of THF species under Sulfa and TMP treatments. 5-formyl-THF is a known inhibitor of several enzymes in the THF network of other organisms (Field et al., 2006; Stover and Schirch, 1993) Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc. 149

A

chorismate

GTP

PABA

6-hydroxymethyldihydropterin-pyrophosphate

folP

C

Sulfa Drugs

7,8-dihydropteroate 7,8-dihydrofolate (DHF) folA Trimethoprim thymidine

THF purines gcvPHT methionine

D

5,10-methylene THF

5-methyl THF

5,10-methenyl THF

10-formyl THF

140

Doubling Time in Minutes

glycine

B

4.0

wild type glyA

120

Overnight Culture OD 600

glyA

100 80 60 40 20

3.0 2.5 2.0 1.5 1.0 0.5

0

0 No Drug TMP 0.1 SMT 50 TMP 0.1 TMP 0.1 + SMT 50 + SMT 50 expected observed

TMP

gcvP gcvT gcvH glyA

Drug-Gene Score 8.00 5.33 2.67 0.00 -2.67 -5.33 -8.00

Normalized Doublings per Hour

SMT + TMP

No Drug TMP 0.1 SMT 50 TMP 0.1 TMP 0.1 + SMT 50 + SMT 50 expected observed

7,8-DHF

E SULFA

wild type gcvP

3.5

7,8-DHF

5,10-mTHF

5,10-mTHF

E. coli

S. pneumoniae

1.2

1.2

1.0

1.0

0.8

0.8

0.6

0.6

0.4

0.4

0.2

0.2

0

0 No Drug

TMP

SMT

TMP + TMP + SMT SMT expected observed

No Drug

TMP

SMT TMP + TMP + SMT SMT expected observed

Figure 5. A New Network Feature Contributing to Antifolate Drug Synergy (A) Schematic of the E. coli tetrahydrofolate (THF) biosynthesis pathway and the enzymatic steps inhibited by Sulfa and TMP. (B) Clustergram of genes that respond to Sulfa, TMP, or the combination. Zoomed image indicates that gcv mutants are sensitive to Sulfa, that glyA is sensitive to TMP, and that these four mutants exhibit essentially wild-type growth in response to the drug combination. (C) glyA and gcvP are a synthetic lethal pair. Image of a plate mating between the donor Hfr gcvP::cat and 24 kanR recipients (arrayed in boxes of 838 colonies) grown on kanamycin/chloramphenicol medium to select for double-mutant strains; position of the glyA::kan and gcvP::kan recipients is highlighted. (D) Liquid culture experiments verify growth phenotypes on agar plates shown in Figure 5B. The deviation of the observed from the expected value for the TMP/ SMT combination denotes the degree of synergy of the two drugs, which is lower for glyA and gcvP cells compared to wild-type cells. Concentrations shown for TMP and SMT are in mg/ml. (E) Quantification of synergy in E. coli and S. pneumoniae, which lacks the branched pathway for generating 5,10-mTHF present in E. coli. Comparisons were performed using single drug concentrations giving equivalent inhibition of both organisms. S. pneumoniae has reduced synergy compared to E. coli. See also Figure S3.

and could act as an effective protein-level regulator. Similarly, a strain lacking a predicted alanine racemase, yggS, is sensitive only to Sulfa; D-alanine is known to inactivate GlyA (Schirch 150 Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc.

et al., 1985), and yggS and glyA form a synthetic lethal pair (Figure S3B). Thus, the different cellular responses to these two drugs may be due, in part, to metabolite-based enzymatic

A

Group

Strand Bias

Essential Genes Responsive Genes (weighted) Responsive Genes Conditionally-Essential Genes CE Rich-Media Genes CE Auxotrophic Genes Non-Responsive Genes

B

oriC

+ + + + + none -

C

p-value < 10-5 3 x 10-5 3 x 10-3 0.04 0.01 3 x 10-3

oriC Responsive Genes

-Proteobacteria Genes

Responsive Genes CE Genes Essential Genes

Proteobacteria Genes Broadly Conserved Genes

ter

ter

Figure 6. Phenomic Profiling Generates Insights into Genome Organization (A) Essential and responsive genes are biased to the plus strand of DNA (transcription direction coincident with replication), and the nonresponsive genes are biased to the minus strand of DNA. (B and C) For each panel, circular plots depict gene position, adjusting coordinates so that the chromosome starts at the origin of replication (oriC = 0 bp); the terminus region (ter) is opposite the oriC. Each trace represents spatial enrichment for the variable plotted based on a 100 kb sliding window. Three dashed lines of the same color accompany each trace, indicating the minimum permutation threshold, the baseline representing zero enrichment, and the maximum permutation threshold (inside to outside). Permutation thresholds are the result of 1000 randomizations of gene class assignments (see Extended Experimental Procedures) and indicate significant negative and positive spatial enrichment at a p value of 0.05. (B) Responsive and CE genes are concentrated around the oriC and scarce around the terminus. (C) The terminus is positively enriched for genes restricted to the g-proteobacteria and negatively enriched for broadly conserved genes. See also Figure S4.

regulation. An extension is that the synergy of combination therapy could rest primarily on complementary inhibition of different one-carbon biosynthesis reactions and therefore recycling of THFs. This model would allow for synergy even with the expected additive limitation of THF production. In summary, our results illustrate the power of phenomic profiling to yield insights into drug action and the ability of a networks view to provide new paradigms for analysis of drug interaction mechanisms, which can facilitate hypothesis-driven research on drug interactions (Bollenbach et al., 2009). This type of analysis may be generally useful in predicting drug synergies and in explaining variable drug-drug interactions across species. Phenomic Profiling Gives Insights into Genomic Organization The E. coli genome is encoded on a single, circular chromosome, with a single origin of replication, oriC. Essential genes are biased to the plus (+) strand, where transcription proceeds in the same direction as DNA replication. This may avert head-on collisions between RNA and DNA polymerases that would result in aborted transcripts or truncated or frame-shifted proteins (Rocha and Danchin, 2003). Here, we show that responsive and CE genes, which are important for optimal growth of the organism, also show + strand bias (Figure 6A). Indeed, the weighted

responsive genome (responsive genes weighted by number of phenotypes identified) is heavily biased to the + strand, indicating great selective pressure to place genes that are important for rapid growth on the + strand. Conversely, the nonresponsive genome is biased to the minus () strand. As our approaches expand to incorporate additional phenotypic readouts that are more important for cells with reduced division and DNA replication (e.g., biofilm formation), the + strand bias of responsive genes will presumably be reduced. The chromosome is massively compacted in the cell to create the nucleoid, which is thought to contribute significantly to the organization of gene expression (Travers and Muskhelishvili, 2005; Vora et al., 2009). Chromosomal loci have spatial addresses in the cell, corresponding closely to their chromosomal position (Toro and Shapiro, 2010). In addition, highly expressed genes that are associated with transcription and translation are located near the origin of replication (oriC), presumably to benefit from the ‘‘gene dosage’’ effect created when rapidly growing cells initiate multiple rounds of DNA replication per division (Couturier and Rocha, 2006). Projecting the spatial distribution of the responsive genes onto the circular chromosome (Figure 6B, black trace) provides us with a snapshot of the E. coli genome from a functional perspective. This projection is based on a 100 kb sliding window and therefore captures organization above the operon level (Figure S4A and Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc. 151

Extended Experimental Procedures). A pattern of alternating peaks and valleys is clearly evident, indicating that responsive genes cluster spatially into large chromosomal regions that are separated by regions generally devoid of responsive genes. ‘‘Valleys’’ are comprised of spatially separated operons that are often transcribed from different strands, indicating that low responsiveness is a regional characteristic rather than an artifact due to large nonresponsive operons. Our finding of clustering above the operon level is in accord with other studies showing that gene expression is broadly correlated across certain regions of the chromosome (Carpentier et al., 2005; Jeong et al., 2004). The responsive genome is most enriched around oriC, which has the highest concentration of responsive genes (Figure 6B, black trace). This area is also enriched for the most responsive genes (Figure S4B) and for CE genes (Figure 6B, red trace), providing strong support for the idea that the E. coli chromosome tends to store genes of high functional importance near the oriC. In contrast, the terminus region is relatively devoid of responsive genes (Figure 6B, black trace) and has a paucity of broadly conserved genes (Figure 6C, red trace) and a corresponding enrichment for genes restricted to g-proteobacteria (Figure 6C, blue trace). We postulate that the terminus region contains newly acquired genes that have yet to fully integrate into the cellular network and tend to lack phenotypes. This could enable cells to avoid unnecessarily high expression of such genes as a consequence of the gene dosage effect. Should this result prove true across bacterial species, it could point to a general organizing principle of circular chromosomes. Phenomic Profiling Describes Drug Action ‘‘Drug-centric’’ analyses are more complex than ‘‘gene-centric’’ analyses. Whereas genes mostly participate in a single biological process, many parameters are required to describe drug action: uptake, primary/secondary targets, and efflux. Therefore, pairwise relationships between drugs are more complex than those between genes. For example, two drugs may cluster based on drug uptake even though their primary targets differ. In addition, drug signatures are an order of magnitude larger than gene signatures (3979 versus 324). To reduce the complexity of drug signatures, we calculated drug-gene ontology (GO) scores, which represent the probability that a given GO group specifically interacts with a given drug (i.e., number of phenotypes associated with genes in the GO group versus across the entire data set). We used these drug-GO scores to explore drug mode of action through a network-based clustering strategy (see Extended Experimental Procedures). The position of drugs in the network (Figure 7A) is based both on the magnitude of their drug-gene ontology (GO) scores (gray) and on drug-drug correlations (yellow). Of the 719 significant drug-GO interactions (p % 103), which include 64 drugs and 218 GO groups, 657 were negative, and only 62 were positive. Thus, disrupting a linked biological process was very likely to increase drug sensitivity (Table S7). Drug-drug correlations increased the resolution of the network and captured drug similarities that escaped the drug-GO analysis. We found that drugs with the same cellular target tend to cluster. For example, drugs targeting DNA (orange) fall in the lower right, those targeting THF-biosynthesis (light green) fall 152 Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc.

on the bottom edge, and those targeting peptidoglycan (PG; purple) predominantly cluster in the upper left. Interestingly, b-lactams cover the center of the PG cluster, whereas drugs targeting other steps of PG synthesis are located at the periphery. The correlation coefficients between b-lactams reveal that the similarity of their phenotypic signatures is related to their respective primary target penicillin-binding proteins (Figure S5A). Interestingly, known synergistic double-drug combinations (TMP/sulfonamides and mecillinam/cefsulodin) occupy spaces that are distinct from either individual drug, arguing that the combination elicits a different cellular response from the individual drugs. It will be interesting to determine whether this holds true for antagonistic or neutral interactions or whether these combinations elicit responses closer to one or both drugs. Importantly, specific drug-GO interactions suggest hypotheses for mechanism of action even for well-studied drugs. Quinolones inhibit DNA gyrase by trapping it as a quinolone adduct, whose mechanism of resolution is poorly understood (Drlica et al., 2008). One GO category, ‘‘cellular DNA catabolic process’’ (xseAB), selectively and specifically (p = 106) interacted negatively with all four quinolones screened (Figure 7B and Figure S5B), expanding on a previous report of xseAB mutant sensitivity to fluoroquinolones (Tamae et al., 2008). We suggest that XseAB (exonuclease VII) is the enzyme that cleaves quinolonebound DNA gyrase from the DNA to allow repair to proceed, a possibility that we are currently exploring. Our drug network also provides clues for the mode of action of poorly described drugs and, conversely, suggests that additional factors are required to explain the action of other drugs. Nitrofurantoin (NTF; Figure 7B) is reported to have a multifaceted impact on cells (McOsker and Fitzpatrick, 1994; Tu and McCalla, 1975), but our data suggest that its cytotoxic effects reflect DNA damage, as it causes lesions requiring nucleotide excision repair (NER) and activates the SOS response. NTF is the only DNA-targeting drug requiring NER, but not double-strand break repair, suggesting that its primary toxic lesion is associated with the replication fork (Figure S5B). In addition, our network analysis validates the idea that indolicidin, a neutrophil antimicrobial peptide, mediates its effects by compromising the inner-membrane permeability of E. coli in a manner that is similar to the proton motive force uncoupler, CCCP (Falla et al., 1996). Finally, phleomycin and bleomycin do not cluster with DNA response drugs, suggesting that they have broader cellular impact (Hecht, 2000; Yeh et al., 2006) from inducing DNA scissions (Giloni et al., 1981). These insights suggest that this phenomic data set is a rich source for discovery of drug function and interrelationships. Perspectives To keep pace with exploding sequence information, cost-effective, high-throughput phenotyping technologies must be developed. Here, we show that phenomic profiling in E. coli fulfills this goal. Our data set is of great utility in identifying the function of orphan genes. Three cases (marB, lpoA, and lpoB) were investigated here or in a study based on this data set (Typas et al., 2010), and we are actively pursuing functional discovery of numerous (>20) orphan genes, as well as annotated genes with previously unsuspected roles in collaboration with others. Because > 25% of the orphan genes are highly correlated to

A

B

Target DNA Division LPS THF biosynthesis Fatty acid metabolism Membrane Multiple/unknown/poorly defined Peptidoglycan Protein synthesis Transcription

Interaction negative positive correlation

Figure 7. Network View Reveals New Insights into Drug Action (A) Colored nodes represent all drugs profiled in this study that were found to have significant interactions with gene ontology (GO) biological process groups (gray nodes). Connections between nodes represent significant drug-GO interactions (p % 103, gray) or high drug-drug correlation (r R 0.32; p % 1097, yellow). Drug node size is based on the number of connections associated with that node, i.e., larger nodes have more drug-GO interactions. Spatial clustering is driven by the p values of drug-GO interactions and drug-drug correlations, resulting in drugs with similar cellular action lying near each other in the network. Drugs with multiple, unknown, or poorly defined targets are shown in dark blue. (B) Zoomed view of subnetwork shadowed by light blue box in (A). All four quinolones screened (orange) interact negatively with xseAB (exonuclease VII) and are the only drugs that require the exonuclease; p = 106. NTF is found to activate the SOS response and create lesions requiring nucleotide-excision repair (Figure S5B). Connections between nodes represent significant drug-GO interactions (p < 103, gray). For all drug-GO interactions, see Table S6. See also Figure S5.

an annotated gene (r R 0.5), this data set provides a rapid method for function discovery. An important finding is that the most responsive orphan genes tend to be narrowly distributed among bacteria. Interestingly, our results mirror initial observations from human microbiome studies. These studies found that: (1) roughly half of the functions encoded in the minimal gut metagenome (ubiquitously present in all 124 individuals screened) are both unknown and of limited evolutionary conservation (Qin et al., 2010), and (2) across four pan-genome species analyzed, the vast majority of noncommon genes were of either unknown function (70%) or were unique family members of functions that were part of the core gene set (Nelson et al., 2010). The latter are probably species-specific additions to conserved biological processes of the pan-genome. Together, these studies argue that, when computational methods based on gene conservation fail, large-scale phenomic analyses can be a second tier for assigning function. To make this approach a reality, low-cost methods for developing deletion

libraries must be developed (Goodman et al., 2009). Singlegene deletion ordered libraries are currently available for only a handful of organisms (Cameron et al., 2008; de Berardinis et al., 2008; Gallagher et al., 2007; Goodman et al., 2009; Kim et al., 2010; Liu et al., 2008; Noble et al., 2010; references in Barker et al., 2010), but advances in transposon mutagenesis make it feasible to create ordered mutant libraries in most organisms. In E. coli, expansion of this work will rest on the ability to assess additional phenotypes through deeper exploration of phenotypic space. The greatest potential resides at the intersection of screening more diverse stresses and incorporating additional cellular readouts. Colorimetric readouts would enable measurement of transcriptional activity or biofilm formation on solid agar surfaces and represent an immediate potential advance for phenomic profiling. High-throughput microscopy would provide a new avenue for such approaches (Werner et al., 2009). Our data set provides information on a substantial collection of antibiotics/antimicrobial compounds that cover a broad Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc. 153

spectrum of drug targets, structural classes, and drug generations, providing a platform for future studies focused on natural products or antimicrobials with unknown targets. Our data set can also provide a platform for studying the mechanism behind drug interactions (Yeh et al., 2009), as shown here for the case of sulfonamides and TMP. Understanding the mechanism underlying known drug interactions may help to predict novel interactions and manipulate existing drug combinations to increase their effectiveness in the clinic. In summary, we have generated a valuable resource for microbiologists studying a wide range of biology and demonstrated the numerous and diverse applications of this data set to infer information on both gene and drug function. As, to our knowledge, the most comprehensive prokaryotic chemical genomic study to date (3979 strains 3 324 conditions), our data set will serve as a base for future studies that aim to increase information and/or resolution on both the gene and drug fronts. We hope that the usefulness of this resource will trigger analogous studies in other organisms, bringing us a step nearer to closing the gene sequence-function gap. EXPERIMENTAL PROCEDURES

REFERENCES Baba, T., Ara, T., Hasegawa, M., Takai, Y., Okumura, Y., Baba, M., Datsenko, K.A., Tomita, M., Wanner, B.L., and Mori, H. (2006). Construction of Escherichia coli K-12 in-frame, single-gene knockout mutants: the Keio collection. Mol. Syst. Biol. 2, 2006.0008. Barker, C.A., Farha, M.A., and Brown, E.D. (2010). Chemical genomic approaches to study model microbes. Chem. Biol. 17, 624–632. Beltrao, P., Cagney, G., and Krogan, N.J. (2010). Quantitative genetic interactions reveal biological modularity. Cell 141, 739–745. Bochner, B.R. (2009). Global phenotypic characterization of bacteria. FEMS Microbiol. Rev. 33, 191–205. Bollenbach, T., Quan, S., Chait, R., and Kishony, R. (2009). Nonoptimal microbial response to antibiotics underlies suppressive drug interactions. Cell 139, 707–718. Butland, G., Peregrı´n-Alvarez, J.M., Li, J., Yang, W., Yang, X., Canadien, V., Starostine, A., Richards, D., Beattie, B., Krogan, N., et al. (2005). Interaction network containing conserved and essential protein complexes in Escherichia coli. Nature 433, 531–537. Butland, G., Babu, M., Dı´az-Mejı´a, J.J., Bohdana, F., Phanse, S., Gold, B., Yang, W., Li, J., Gagarinova, A.G., Pogoutse, O., et al. (2008). eSGA: E. coli synthetic genetic array analysis. Nat. Methods 5, 789–795. Cameron, D.E., Urbach, J.M., and Mekalanos, J.J. (2008). A defined transposon mutant library and its use in identifying motility genes in Vibrio cholerae. Proc. Natl. Acad. Sci. USA 105, 8736–8741.

Experimental procedures are partially elaborated upon in the text and figure legends and are fully explicated in the Supplemental Information.

Carpentier, A.S., Torre´sani, B., Grossmann, A., and He´naut, A. (2005). Decoding the nucleoid organisation of Bacillus subtilis and Escherichia coli through gene expression data. BMC Genomics 6, 84.

SUPPLEMENTAL INFORMATION

Couturier, E., and Rocha, E.P. (2006). Replication-associated gene dosage effects shape the genomes of fast-growing bacteria but only for transcription and translation genes. Mol. Microbiol. 59, 1506–1518.

Supplemental Information includes Extended Experimental Procedures, five figures, and seven tables and can be found with this article at doi:10.1016/ j.cell.2010.11.052.

ACKNOWLEDGMENTS We thank R. Kishony, J. Weissman, A. Hochschild, and R.G. Martin for critically reading this manuscript; J. Hu and P. Thomas for hosting these data on E. coli Wiki; C. Raetz for CHIR-090 and M. Gottesman for Bicyclomycin; H. Mori for the Keio Collection; G. Storz for sharing the sRNA deletion library prior to publication; J.Greenblatt and A. Emili for SPA-tagged alleles; W. Margolin, R. Misra, T. Silhavy, and B. Palsson for mutants; and T. Baker and R. Sauer for controllable degradation plasmids. This work was supported by NIH R01 GM085697 and ARRA GM085697-01S1 to C.A.G. and N.J.K.; NIH R01 GM036278 to C.A.G.; NIH K99GM092984 to A.T.; NIH AI060744 to M.E.W.; NIH GM078338 to S.S.; NIH F31 DE020206-01 and NIH T32 DE007306 (R.J.N. support); European Molecular Biology Organization long-term fellowship (to A.T.); and Human Frontier Science Program Long-Term Postdoctoral Fellowship (to P.B.). In loving memory of S.R. Bartlett and J.T. Blair. A.T., R.J.N., N.J.K., and C.A.G. conceived this study; A.T. and R.J.N. designed research; A.T., R.J.N., Y.J.C., S.L., J.L., and A.W. performed the screen; R.J.N. did the data processing; R.J.N., S.S., P.B., and A.T. analyzed the data; A.T., R.J.N., C.A.G., and S.L. interpreted the data; R.J.N., A.T., and M.Z. performed the follow-up work in the antifolate drug story, except for the work in Streptococcus pneumoniae, which was conducted by K.M.K.; R.J.N., R.C., A.T., and M.Z. performed the follow-up work in the mar story; R.J.N., A.T., and C.A.G. wrote the manuscript; R.J.N. and M.S. prepared the figures; P.B., R.C., M.E.W., K.M.K., S.L., S.S., and N.J.K. edited the manuscript; A.T. and C.A.G. supervised all aspects of this project. Received: August 31, 2010 Revised: November 7, 2010 Accepted: November 24, 2010 Published online: December 23, 2010

154 Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc.

D’Elia, M.A., Pereira, M.P., and Brown, E.D. (2009). Are essential genes really essential? Trends Microbiol. 17, 433–438. de Berardinis, V., Vallenet, D., Castelli, V., Besnard, M., Pinet, A., Cruaud, C., Samair, S., Lechaplais, C., Gyapay, G., Richez, C., et al. (2008). A complete collection of single-gene deletion mutants of Acinetobacter baylyi ADP1. Mol. Syst. Biol. 4, 174. Drlica, K., Malik, M., Kerns, R.J., and Zhao, X. (2008). Quinolone-mediated bacterial death. Antimicrob. Agents Chemother. 52, 385–392. Falla, T.J., Karunaratne, D.N., and Hancock, R.E. (1996). Mode of action of the antimicrobial peptide indolicidin. J. Biol. Chem. 271, 19298–19303. Fiedler, D., Braberg, H., Mehta, M., Chechik, G., Cagney, G., Mukherjee, P., Silva, A.C., Shales, M., Collins, S.R., van Wageningen, S., et al. (2009). Functional organization of the S. cerevisiae phosphorylation network. Cell 136, 952–963. Field, M.S., Szebenyi, D.M., and Stover, P.J. (2006). Regulation of de novo purine biosynthesis by methenyltetrahydrofolate synthetase in neuroblastoma. J. Biol. Chem. 281, 4215–4221. Gallagher, L.A., Ramage, E., Jacobs, M.A., Kaul, R., Brittnacher, M., and Manoil, C. (2007). A comprehensive transposon mutant library of Francisella novicida, a bioweapon surrogate. Proc. Natl. Acad. Sci. USA 104, 1009–1014. Giaever, G., Flaherty, P., Kumm, J., Proctor, M., Nislow, C., Jaramillo, D.F., Chu, A.M., Jordan, M.I., Arkin, A.P., and Davis, R.W. (2004). Chemogenomic profiling: identifying the functional interactions of small molecules in yeast. Proc. Natl. Acad. Sci. USA 101, 793–798. Giloni, L., Takeshita, M., Johnson, F., Iden, C., and Grollman, A.P. (1981). Bleomycin-induced strand-scission of DNA. Mechanism of deoxyribose cleavage. J. Biol. Chem. 256, 8608–8615. Girgis, H.S., Hottes, A.K., and Tavazoie, S. (2009). Genetic architecture of intrinsic antibiotic susceptibility. PLoS ONE 4, e5629. Girgis, H.S., Liu, Y., Ryu, W.S., and Tavazoie, S. (2007). A comprehensive genetic characterization of bacterial motility. PLoS Genet. 3, 1644–1660.

Goodman, A.L., McNulty, N.P., Zhao, Y., Leip, D., Mitra, R.D., Lozupone, C.A., Knight, R., and Gordon, J.I. (2009). Identifying genetic determinants needed to establish a human gut symbiont in its habitat. Cell Host Microbe 6, 279–289. Griffith, K.L., Shah, I.M., and Wolf, R.E., Jr. (2004). Proteolytic degradation of Escherichia coli transcription activators SoxS and MarA as the mechanism for reversing the induction of the superoxide (SoxRS) and multiple antibiotic resistance (Mar) regulons. Mol. Microbiol. 51, 1801–1816. Hecht, S.M. (2000). Bleomycin: new perspectives on the mechanism of action. J. Nat. Prod. 63, 158–168.

of molecular genetics and informatics to microarray data. Mol. Microbiol. 44, 1611–1624. Martin, R.G., Bartlett, E.S., Rosner, J.L., and Wall, M.E. (2008). Activation of the Escherichia coli marA/soxS/rob regulon in response to transcriptional activator concentration. J. Mol. Biol. 380, 278–284. McOsker, C.C., and Fitzpatrick, P.M. (1994). Nitrofurantoin: mechanism of action and implications for resistance development in common uropathogens. J. Antimicrob. Chemother. 33 Suppl A, 23–30.

Hillenmeyer, M.E., Fung, E., Wildenhain, J., Pierce, S.E., Hoon, S., Lee, W., Proctor, M., St Onge, R.P., Tyers, M., Koller, D., et al. (2008). The chemical genomic portrait of yeast: uncovering a phenotype for all genes. Science 320, 362–365.

Nelson, K.E., Weinstock, G.M., Highlander, S.K., Worley, K.C., Creasy, H.H., Wortman, J.R., Rusch, D.B., Mitreva, M., Sodergren, E., Chinwalla, A.T., et al; Human Microbiome Jumpstart Reference Strains Consortium. (2010). A catalog of reference genomes from the human microbiome. Science 328, 994–999.

Hobbs, E.C., Astarita, J.L., and Storz, G. (2010). Small RNAs and small proteins involved in resistance to cell envelope stress and acid shock in Escherichia coli: analysis of a bar-coded mutant collection. J. Bacteriol. 192, 59–67.

Nicoloff, H., Perreten, V., McMurry, L.M., and Levy, S.B. (2006). Role for tandem duplication and lon protease in AcrAB-TolC- dependent multiple antibiotic resistance (Mar) in an Escherichia coli mutant without mutations in marRAB or acrRAB. J. Bacteriol. 188, 4413–4423.

Hoon, S., Smith, A.M., Wallace, I.M., Suresh, S., Miranda, M., Fung, E., Proctor, M., Shokat, K.M., Zhang, C., Davis, R.W., et al. (2008). An integrated platform of genomic assays reveals small-molecule bioactivities. Nat. Chem. Biol. 4, 498–506.

Nicoloff, H., Perreten, V., and Levy, S.B. (2007). Increased genome instability in Escherichia coli lon mutants: relation to emergence of multiple-antibioticresistant (Mar) mutants caused by insertion sequence elements and large tandem genomic amplifications. Antimicrob. Agents Chemother. 51, 1293– 1303.

Hu, P., Janga, S.C., Babu, M., Dı´az-Mejı´a, J.J., Butland, G., Yang, W., Pogoutse, O., Guo, X., Phanse, S., Wong, P., et al. (2009). Global functional atlas of Escherichia coli encompassing previously uncharacterized proteins. PLoS Biol. 7, e96. Jeanguenin, L., Lara-Nunez, A., Pribat, A., Hamner Mageroy, M., Gregory, J.F., III, Rice, K.C., de Crecy-Lagard, V., and Hanson, A.D. (2010). Moonlighting glutamate formiminotransferases can functionally replace 5-formyltetrahydrofolate cycloligase. J. Biol. Chem., in press. Published online October 15, 2010. 10.1074/jbc.M110.190504. Jeong, K.S., Ahn, J., and Khodursky, A.B. (2004). Spatial patterns of transcriptional activity in the chromosome of Escherichia coli. Genome Biol. 5, R86. Joyce, A.R., Reed, J.L., White, A., Edwards, R., Osterman, A., Baba, T., Mori, H., Lesely, S.A., Palsson, B.O., and Agarwalla, S. (2006). Experimental and computational assessment of conditionally essential genes in Escherichia coli. J. Bacteriol. 188, 8259–8271. Kim, D.U., Hayles, J., Kim, D., Wood, V., Park, H.O., Won, M., Yoo, H.S., Duhig, T., Nam, M., Palmer, G., et al. (2010). Analysis of a genome-wide set of gene deletions in the fission yeast Schizosaccharomyces pombe. Nat. Biotechnol. 28, 617–623. Kohanski, M.A., Dwyer, D.J., Wierzbowski, J., Cottarel, G., and Collins, J.J. (2008). Mistranslation of membrane proteins and two-component system activation trigger antibiotic-mediated cell death. Cell 135, 679–690. Kwon, Y.K., Higgins, M.B., and Rabinowitz, J.D. (2010). Antifolate-induced depletion of intracellular glycine and purines inhibits thymineless death in E. coli. ACS Chem. Biol. 5, 787–795. Kwon, Y.K., Lu, W., Melamud, E., Khanam, N., Bognar, A., and Rabinowitz, J.D. (2008). A domino effect in antifolate drug action in Escherichia coli. Nat. Chem. Biol. 4, 602–608. Lee, W., St Onge, R.P., Proctor, M., Flaherty, P., Jordan, M.I., Arkin, A.P., Davis, R.W., Nislow, C., and Giaever, G. (2005). Genome-wide requirements for resistance to functionally distinct DNA-damaging agents. PLoS Genet. 1, e24. Liu, O.W., Chun, C.D., Chow, E.D., Chen, C., Madhani, H.D., and Noble, S.M. (2008). Systematic genetic analysis of virulence in the human fungal pathogen Cryptococcus neoformans. Cell 135, 174–188. Liu, A., Tran, L., Becket, E., Lee, K., Chinn, L., Park, E., Tran, K., and Miller, J.H. (2010). Antibiotic sensitivity profiles determined with an Escherichia coli gene knockout collection: generating an antibiotic bar code. Antimicrob. Agents Chemother. 54, 1393–1403. Martin, R.G., and Rosner, J.L. (2002). Genomics of the marA/soxS/rob regulon of Escherichia coli: identification of directly activated promoters by application

Noble, S.M., French, S., Kohn, L.A., Chen, V., and Johnson, A.D. (2010). Systematic screens of a Candida albicans homozygous deletion library decouple morphogenetic switching and pathogenicity. Nat. Genet. 42, 590–598. Pan, X., Yuan, D.S., Xiang, D., Wang, X., Sookhai-Mahadeo, S., Bader, J.S., Hieter, P., Spencer, F., and Boeke, J.D. (2004). A robust toolkit for functional profiling of the yeast genome. Mol. Cell 16, 487–496. Parsons, A.B., Lopez, A., Givoni, I.E., Williams, D.E., Gray, C.A., Porter, J., Chua, G., Sopko, R., Brost, R.L., Ho, C.H., et al. (2006). Exploring the modeof-action of bioactive compounds by chemical-genetic profiling in yeast. Cell 126, 611–625. Pathania, R., Zlitni, S., Barker, C., Das, R., Gerritsma, D.A., Lebert, J., Awuah, E., Melacini, G., Capretta, F.A., and Brown, E.D. (2009). Chemical genomics in Escherichia coli identifies an inhibitor of bacterial lipoprotein targeting. Nat. Chem. Biol. 5, 849–856. Qin, J., Li, R., Raes, J., Arumugam, M., Burgdorf, K.S., Manichanh, C., Nielsen, T., Pons, N., Levenez, F., Yamada, T., et al; MetaHIT Consortium. (2010). A human gut microbial gene catalogue established by metagenomic sequencing. Nature 464, 59–65. Rocha, E.P., and Danchin, A. (2003). Essentiality, not expressiveness, drives gene-strand bias in bacteria. Nat. Genet. 34, 377–378. Roguev, A., Bandyopadhyay, S., Zofall, M., Zhang, K., Fischer, T., Collins, S.R., Qu, H., Shales, M., Park, H.O., Hayles, J., et al. (2008). Conservation and rewiring of functional modules revealed by an epistasis map in fission yeast. Science 322, 405–410. Schirch, V., Hopkins, S., Villar, E., and Angelaccio, S. (1985). Serine hydroxymethyltransferase from Escherichia coli: purification and properties. J. Bacteriol. 163, 1–7. Schuldiner, M., Collins, S.R., Thompson, N.J., Denic, V., Bhamidipati, A., Punna, T., Ihmels, J., Andrews, B., Boone, C., Greenblatt, J.F., et al. (2005). Exploration of the function and organization of the yeast early secretory pathway through an epistatic miniarray profile. Cell 123, 507–519. Stover, P., and Schirch, V. (1993). The metabolic role of leucovorin. Trends Biochem. Sci. 18, 102–106. Taber, H.W., Mueller, J.P., Miller, P.F., and Arrow, A.S. (1987). Bacterial uptake of aminoglycoside antibiotics. Microbiol. Rev. 51, 439–457. Tamae, C., Liu, A., Kim, K., Sitz, D., Hong, J., Becket, E., Bui, A., Solaimani, P., Tran, K.P., Yang, H., and Miller, J.H. (2008). Determination of antibiotic hypersensitivity among 4,000 single-gene-knockout mutants of Escherichia coli. J. Bacteriol. 190, 5981–5988.

Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc. 155

Tong, A.H., Evangelista, M., Parsons, A.B., Xu, H., Bader, G.D., Page´, N., Robinson, M., Raghibizadeh, S., Hogue, C.W., Bussey, H., et al. (2001). Systematic genetic analysis with ordered arrays of yeast deletion mutants. Science 294, 2364–2368. Toro, E., and Shapiro, L. (2010). Bacterial chromosome organization and segregation. Cold Spring Harb. Perspect. Biol. 2, a000349.

antimicrobial peptides (CAMPs): Rob-dependent CAMP induction of the marRAB operon. Microbiology 156, 570–578. Werner, J.N., Chen, E.Y., Guberman, J.M., Zippilli, A.R., Irgon, J.J., and Gitai, Z. (2009). Quantitative genome-scale analysis of protein localization in an asymmetric bacterium. Proc. Natl. Acad. Sci. USA 106, 7858–7863.

Travers, A., and Muskhelishvili, G. (2005). DNA supercoiling - a global transcriptional regulator for enterobacterial growth? Nat. Rev. Microbiol. 3, 157–169.

Warner, J.R., Reeder, P.J., Karimpour-Fard, A., Woodruff, L.B., and Gill, R.T. (2010). Rapid profiling of a microbial genome using mixtures of barcoded oligonucleotides. Nat. Biotechnol. 28, 856–862.

Tu, Y., and McCalla, D.R. (1975). Effect of activated nitrofurans on DNA. Biochim. Biophys. Acta 402, 142–149.

Xu, D., Jiang, B., Ketela, T., Lemieux, S., Veillette, K., Martel, N., Davison, J., Sillaots, S., Trosok, S., Bachewich, C., et al. (2007). Genome-wide fitness test and mechanism-of-action studies of inhibitory compounds in Candida albicans. PLoS Pathog. 3, e92.

Typas, A., Nichols, R.J., Siegele, D.A., Shales, M., Collins, S.R., Lim, B., Braberg, H., Yamamoto, N., Takeuchi, R., Wanner, B.L., et al. (2008). Highthroughput, quantitative analyses of genetic interactions in E. coli. Nat. Methods 5, 781–787. Typas, A., Banzhaf, M., van den Berg van Saparoea, B., Verheul, J., Biboy, J., Nichols, R.J., Zietek, M., Beilharz, K., Kannenberg, K., von Rechenberg, M., et al. (2010). Regulation of peptidoglycan synthesis by outer-membrane proteins. Cell 143, 1097–1109. Vora, T., Hottes, A.K., and Tavazoie, S. (2009). Protein occupancy landscape of a bacterial genome. Mol. Cell 35, 247–253. Warner, D.M., and Levy, S.B. (2010). Different effects of transcriptional regulators MarA, SoxS and Rob on susceptibility of Escherichia coli to cationic

156 Cell 144, 143–156, January 7, 2011 ª2011 Elsevier Inc.

Yamamoto, N., Nakahigashi, K., Nakamichi, T., Yoshino, M., Takai, Y., Touda, Y., Furubayashi, A., Kinjyo, S., Dose, H., Hasegawa, M., et al. (2009). Update on the Keio collection of Escherichia coli single-gene deletion mutants. Mol. Syst. Biol. 5, 335. Yeh, P., Tschumi, A.I., and Kishony, R. (2006). Functional classification of drugs by properties of their pairwise interactions. Nat. Genet. 38, 489–494. Yeh, P.J., Hegreness, M.J., Aiden, A.P., and Kishony, R. (2009). Drug interactions and the evolution of antibiotic resistance. Nat. Rev. Microbiol. 7, 460–466.

Scientific Editor, Cell

Cell is seeking a scientist to join its editorial team. The minimum qualification for this position is a PhD in a relevant area of biomedical research, although additional postdoctoral or editorial experience is preferred. This is a superb opportunity for a talented individual to play a critical role in promoting science by helping researchers disseminate their findings to the wider community. As an editor, you would be responsible for assessing submitted research papers and overseeing the refereeing process, and you would commission and edit material for Cell's Leading Edge. You would also travel frequently to scientific conferences to follow developments in research and to establish and maintain close ties with the scientific community. You would have opportunities to pioneer and contribute to new trends in scientific publishing. The key qualities we look for are breadth of scientific interest and the ability to think critically about a wide range of scientific issues. The successful candidate will be highly motivated and creative and able to work independently as well as in a team. This is a full-time, in-house editorial position, based at the Cell Press office in Kendall Square near MIT in Cambridge, Massachusetts. Cell Press offers an attractive salary and benefits package and a stimulating working environment that encourages innovation.

To apply, please visit http://reedelsevier.taleo.net/careersection/51/jobdetail.ftl?lang=en&job=SCI0005W.

Scientific Editor, Cell Press Cell Press seeks to appoint three Scientific Editors with dual roles covering scientific editing and the review material. These positions will be associated with the Cell Press titles Cancer Cell, Current Biology, Developmental Cell, and Neuron, and expertise in any of the relevant areas covered by these journals will be considered. Working closely with the research community, you will be acquiring, managing, and developing new editorial content for the Cell Press research titles. These positions will also work closely with other aspects of the business, including production, business development, marketing, and commercial sales, and, therefore, provide an excellent entry opportunity to science publishing. You will work as part of a highly dynamic and collaborative editorial group in the Cambridge, MA office. These positions are an exciting opportunity to stay at the forefront of the latest scientific advances while developing a new career in an exciting publishing environment. Minimum qualifications are a PhD in a relevant life science discipline, and additional postdoctoral or other experience is a plus. Ideal candidates would have a strong scientific background and broad research interests, excellent writing and communication skills, strong organizational and interpersonal skills, as well as creative energy and enthusiasm for science and science communication. Prior publishing or editorial experience is an advantage but is not a requirement.

To apply Please submit to the url below a CV and cover letter explaining your interest in an editorial position and describing your qualifications, research interests, and reasons for pursuing a career in scientific publishing. Applications will be accepted on an ongoing basis through January 10, 2011. http://reedelsevier.taleo.net/careersection/51/jobdetail.ftl?lang=en&job=SCI00063.

No phone inquiries. Elsevier-Cell Press is an Equal Opportunity Employer.

Scientific Editor, Molecular Cell

Molecular Cell is seeking a full-time scientific editor to join its editorial team. We will consider qualified candidates with scientific expertise in any area that the journal covers. The minimum qualification for this position is a PhD in a relevant area of biomedical research, although additional experience is preferred. This is a superb opportunity for a talented individual to play a critical role in the research community away from the bench. As a scientific editor, you would be responsible for assessing submitted research papers, overseeing the refereeing process, and choosing and commissioning review material. You would also travel frequently to scientific conferences to follow developments in research and establish and maintain close ties with the scientific community. The key qualities we look for are breadth of scientific interest and the ability to think critically about a wide range of scientific issues. The successful candidate will also be highly motivated and creative and able to work independently as well as in a team. This is a full-time in-house editorial position, based at the Cell Press office in Cambridge, Massachusetts. Cell Press offers an attractive salary and benefits package and a stimulating working environment. Applications will be held in the strictest of confidence and will be considered on an ongoing basis until the position is filled.

To apply Please submit a CV and cover letter describing your qualifications, research interests, and reasons for pursuing a career in scientific publishing, as soon as possible, to our online jobs site: http://www.elsevier.com/wps/find/job_search.careers. Click on “search for US jobs” and select “Massachusetts.” Or: http://reedelsevier.taleo.net/careersection/51/jobdetail.ftl?lang=en&job=SCI0005X.

No phone inquiries, please. Cell Press is an equal opportunity/affirmative action employer, M/F/D/V.

Scientific Editor, Cell Metabolism Cell Metabolism is seeking a full-time scientific editor to join its editorial team. Cell Metabolism publishes metabolic research with an emphasis on molecular mechanisms and translational medicine. The minimum qualification for this position is a PhD in a relevant area of biomedical research, although additional postdoctoral and/or editorial experience is preferred. This is a superb opportunity for a talented individual to play a critical role in promoting science by helping researchers shape and disseminate their findings to the wider community. The scientific editor is responsible for assessing submitted research papers, overseeing the refereeing process, and choosing, commissioning, and editing review material. The scientific editor frequently travels to scientific conferences to follow developments in research and establish and maintain close ties with the scientific community. The key qualities we look for are breadth of scientific interest, the ability to think critically about a wide range of scientific issues, and strong communication skills. The successful candidate will also be highly motivated and creative and able to work independently as well as in a team and should have opportunities to pioneer and contribute to new trends in scientific publishing. This is a full-time in-house editorial position, based at the Cell Press office in Cambridge, Massachusetts. Cell Press offers an attractive salary and benefits package and a stimulating working environment that encourages innovation. Please submit a CV and cover letter describing your qualifications, general research interests, and motivation for pursuing a career in scientific publishing. Applications will be considered on an ongoing basis until the closing date of January 20, 2011.

To apply, visit http://reedelsevier.taleo.net/careersection/51/jobdetail.ftl?lang=en&job=SCI0005Y.

No phone inquiries. Elsevier-Cell Press is an Equal Opportunity Employer.

Scientific Editor, Neuron

Neuron is seeking an additional full-time scientific editor to join its editorial team based in Cambridge, Massachusetts. Neuron publishes across a range of disciplines including developmental, molecular, cellular, systems, and cognitive neuroscience. As a scientific editor, you would be responsible for assessing submitted research manuscripts, overseeing the review process, and commissioning and editing review material for the journal. You would also travel frequently to scientific conferences to follow developments in research and to establish and maintain close ties with the scientific community. The minimum qualification for this position is a PhD in a relevant area of biomedical research, although previous editorial experience is beneficial. This is a superb opportunity for a talented individual to play a critical role in the research community away from the bench. The key qualities we are looking for are breadth of scientific interest and the ability to think critically about a wide range of scientific issues. The successful candidate will also be highly motivated and creative, possess strong communication skills, and be able to both work independently and as part of a team. This is a full-time, in-house editorial position, based at Cell Press headquarters in Cambridge, Massachusetts. Cell Press offers an attractive salary and benefits package and a stimulating work environment. Applications will be held in the strictest of confidence and will be considered on an ongoing basis.

To apply Please submit a cover letter describing your background, interests, and a candid appraisal of the strengths and weaknesses of Neuron, along with your CV, to http://reedelsevier.taleo.net/careersection/51/jobdetail.ftl?lang=en&job=SCI0006F. Applications will be accepted through January 10, 2011.

Cell Press Business Project Editor Position Available Cell Press is seeking a Business Project Editor to plan, develop, and implement projects that have commercial or sponsorship potential. By drawing on existing content or developing new material, the Editor will work with Cell Press’s commercial sales group to create collections of content in print or online that will be attractive to readers and sponsors. The Editor will also be responsible for leveraging new online opportunities for engaging the readers of Cell Press journals. The successful candidate will have a PhD in the biological sciences, broad scientific interests, a fascination with technology, good commercial instincts, and a true passion for both science and science communication. They should be highly organized and dedicated, with excellent written and oral communication skills, and should be willing to work to tight deadlines. The position is full time and based in Cambridge, MA. Cell Press offers an attractive salary and benefits package and a stimulating work environment. Applications will be considered on a rolling basis. For consideration, please apply online and include a cover letter and resume. To apply, visit the career page at http://www.elsevier.com and search on keywords “Business Project Editor.”

Positions Available

OPPORTUNITY AT THE UNIVERSITY OF GENEVA THE FACULTY OF MEDICINE of GENEVA is seeking applications for 3 positions of:

ASSISTANT PROFESSOR or ASSOCIATE PROFESSOR in the Section of Fundamental Medicine These full-time positions at the level of assistant or associate professor involve pre- and postgraduate teaching and research in the fields of metabolism or cardiovascular disease or biology of cancer or host pathogen interactions or stem cells. Candidates should have demonstrated experience in one of these specific fields. They must be able to develop and lead research programs in a particular aspect of the field and to coordinate research projects in collaboration with other medical specialties. They must be willing to participate in interdisciplinary projects as well as assuming all pertinent administrative tasks. A Doctorate of Science or Medicine (MD) is required and some knowledge of French is expected at term. The starting date for the position is June 1st 2011, or according to agreement. Applications must be sent before March 1st 2011, to: The Dean of the Faculty of Medicine Centre médical universitaire 1 rue Michel-Servet 1211 Genève 4 - Switzerland Information concerning applications and job description are available from [email protected] Tel. +41 22 379 50 26 Women are encouraged to apply

Look Again. Discover More. • Access to the 14 Cell Press primary research journals and 14 Trends reviews title, all on the same platform • Improved, more robust article and author search • Easy to navigate home page, articles pages and archive

www.cell.com

T H E N E W E D I T I O N O F F U N D A M E N TA L N E U R O S C I E N C E I S N O W AVA I L A B L E !

FUNDAMENTAL NEUROSCIENCE THIRD EDITION EDITED BY:

Larry R. Squire, VA Medical Center and University of California, San Diego, La Jolla, California, USA February 2008 Hardback

SENIOR EDITORS: Darwin Berg, Floyd E. Bloom, Sascha du Lac, Anirvan Ghosh, and Nicholas C. Spitzer

1,280 pp. 9780123740199 Print list price: $99.95

This comprehensive text reference presents the entire spectrum of modern neuroscience. Addressing feedback on the previous edition, Fundamental Neuroscience, Third Edition is now more concise and reader-friendly than ever before. Each chapter is heavily illustrated and provides clinical boxes describing experiments, disorders, and methodological approaches and concepts. A companion site contains an image bank of the figures for use in poster presentations, slides, and handouts. Capturing the promise and excitement of this fast-moving field, Fundamental Neuroscience, Third Edition is the book that you will reference throughout your neuroscience career! NEW TO THIS EDITION: • 30% new material including new chapters on Dendritic Development and Spine Morphogenesis; Chemical Senses; Cerebellum; Eye Movements; Circadian Timing; Sleep and Dreaming; and Consciousness • Companion Web site with all figures and Web links to additional material • Multiple model system coverage beyond rats, mice, and monkeys • Extensively expanded index for easier referencing WHY BUY THIS EDITION? • It presents the entire spectrum of modern neuroscience as a cohesive reference rather than just a collection of review articles • It is written at a level appropriate for graduate-level students and researchers with various science backgrounds who need a solid foundation in neuroscience principles • Clinical conditions are discussed in neuroscience to help readers establish the connection between the basic biology and the neurological symptoms and syndromes • It includes more than 650 illustrations, including color photographs, imaging, and micrographs, also available on the companion Web site

SECURE ONLINE ORDERING AT ELSEVIERDIRECT.COM

.%7


Reach Your Ideal Candidate! så0OSTåONEåORåMULTIPLEåå JOBåOPENINGSå så0ROMOTEåANåUPCOMINGåå CONFERENCEåORåEVENTå så!NNOUNCEåAåGRANTAWARDå så%LEVATEåYOURåå ORGANIZATIONSåPROlLE

careers.cell.com

http://www.eupa.org/

Editor in Chief: Juan J. Calvete, Valencia, Spain Executive Editors: Proteomics in Cell Biology Jean-Jacques Diaz, Lyon, France

EuPA now has its own journal!

Covered by PubMed

Proteomics in Microbiology Concha Gil, Madrid, Spain Proteomics in Plant Systems Jesus V. Jorrín, Córdoba, Spain Proteomics in Animal Models Dario Neri, Zürich, Switzerland Proteomics in Protein Science Jasna Peter-Katalinic, Münster, Germany Biomedical Applications of Proteomics and Congress Proceedings Jean-Charles Sanchez, Geneva, Switzerland Proteomics of Body Fluids and Proteomic Technologies Pier Giorgio Righetti, Milan, Italy Bioinformatics in Proteomics Peter Højrup, Odense, Denmark

Submitting Authors Manuscripts can be submitted to the Journal of Proteomics at http://ees.elsevier.com/jprot

For a complete listing of the editorial board, visit the journal’s homepage

To receive more information register at:

http://www.elsevier.com/locate/jprot

For the past 40 years FEBS Letters has published award-winning research papers crafted by distinguished experts in the field.

www.febsletters.com

3TAYå#ONNECTED *OINå!3('å4ODAY
"ENElTS¬OF¬-EMBERSHIP så3UBSCRIPTIONåTOå!*(' åNOWåPUBLISHEDåå BYå#ELLå0RESSå så2EDUCEDåREGISTRATIONåFEESåFORåTHEåå !NNUALå-EETINGå så$ISCOUNTEDåSUBSCRIPTIONSåTOå#ELLåå 0RESSåJOURNALS så!CCESSåTOåEDUCATIONALåMATERIALS så)NmUENCEåONåPUBLICåPOLICYåPROCESS

WWWASHGORG

#ALLåFORå0APERS 3UBMITåYOURåMANUSCRIPTåTODAY
)MPACT¬ &ACTOR 

3OURCEåå4HOMSONå2EUTERSå*OURNALå#ITATIONå2EPORTS

4OPICSåCOVEREDåINCLUDE

å

¬#HANNELS ¬2ECEPTORS ¬AND¬%LECTRICAL¬3IGNALING

¬3PECTROSCOPY ¬)MAGING ¬/THER¬4ECHNIQUES

¬0ROTEINS¬

¬-USCLE¬AND¬#ONTRACTILITY

¬"IOPHYSICAL¬4HEORY¬AND¬-ODELING¬

¬3UPRAMOLECULAR¬!SSEMBLIES

¬#ELL¬"IOPHYSICS

¬"IOENERGETICS

¬0HOTOBIOPHYSICS

¬.UCLEIC¬!CIDS¬

¬-EMBRANES¬

¬%LECTROPHYSIOLOGY

HTTPSUBMITBIOPHYSJORG

4HEå"IOPHYSICALå3OCIETYå ANDå#ELLå0RESS !åNEWåPARTNERSHIP


3EEåWHATSåNEWåAT WWWBIOPHYSJORG

Scopus is the largest abstract and citation database of peer-reviewed literature and quality web sources with smart tools to track, analyze and visualize research.

enrich your experience

www.scopus.com

SnapShot: Chromatin Remodeling: INO80 and SWR1 Yunhe Bao and Xuetong Shen Department of Molecular Carcinogenesis, M.D. Anderson Cancer Center, Smithville, TX 78957, USA SWR1

Arp8

Taf14

Arp4 les3

Ino80

Nhp10

Top view

Side view

les4 les5

les6

Rvb1 Rvb2

Arp6 Swc6

Act1 Arp4

les1 les2

Arp5

Swc3

Swc2

Act1

Act1

Rvb1, 2

INO80

Swc7 Swc5

Yaf9

Bdf1

Swr1 Rvb1

Swc4

Rvb2

Remodeling mechanism ATP ADP+Pi

Ino80 Ino80 Actin Arp4 Actin Arp4

Nucleosome sliding AA

AZ

ZZ

ATP ADP+Pi

Swc2

Swr1

Arp4

First H2A.Z-H2B dimer exchange

Actin

Second H2A.Z-H2B dimer exchange Actin Arp4 Swc2

Swc2

Swr1

Swc2 H2A.Z H2B

Actin Arp4

H2A H2B

Swr1

Swr1

Actin Arp4

Actin Arp4

ATP ADP+Pi

INO80 biological functions

Swr1

Swc2

SWR1 biological functions

Transcription repression and activation

H2A.Z deposition flanking nucleosome-free region around transcription site

X H2A.Z H2B

Transcription factors

SWR1

INO80

Transcription factors

Loading of H2A.Z around DSB

DNA repair and checkpoint regulation P

P

P

MRX

P

INO80 Replication fork stabilization DNA damage tolerance

RPA

P

P

P

P

P

SWR1 P

Replisome Checkpoint factors

Stress

INO80 condition

P

Mec1

Replication fork progression

Replisome

P

INO80

Checkpoint factors

Normal condition

Repair factors

H2A.Z deposition at the boundary of euchromatin and heterochromatin SIR

? SWR1

Rad18 PCNA

158

Cell 144, January 7, 2011 ©2011 Elsevier Inc.

DOI 10.1016/j.cell.2011.12.024

See online version for legend and references.

Expand your stem cell library and save today on the latest books on stem cells and regenerative medicine Stem Cells

Stem Cell Anthology

Scientific Facts and Fiction

From Stem Cell Biology, Tissue Engineering, Regenerative Medicine, Cloning and Stem Cell Methods

Christine Mummery, Ian Wilmut, Anja Van de Stolpe and Bernard Roelen November 2010 | 400 pages | Paperback | $79.95 | €57.95 | £48.99 | ISBN: 9780123815354

Principles of Regenerative Medicine, 2nd Edition

Bruce M. Carlson October 2009 | 450 pp. | Hardback | $150.00 | €100.00 | £95.00 |AU$222.00 | ISBN: 9780123756824

Essential Stem Cell Methods

Anthony Atala and Robert Lanza November 2010 | 1400 pages | Hardback | $199.95 | €143.00 | £125.00 | ISBN: 9780123814227

A Volume in the Reliable Lab Solutions Series

Heart Development and Regeneration, 2-Volume Set

Tissue Engineering

Robert Lanza and Irina Klimanskaya April 2009 | 628 pp. | Paperback | $75.00 | €50.95 | £45.99 |AU$111.00 | ISBN: 9780123750617

Nadia Rosenthal and Richard P. Harvey June 2010 | 1072 pp. | Hardback | $199.95 | €143.00 | £125.00 | AU$296.00 | ISBN: 9780123813329

Clemens van Blitterswijk, Peter Thomsen, Jeffrey Hubbell, Ranieri Cancedda, Anders Lindahl Sahlgrenska, Jerome Sohier and David F. Williams March 2008 | 760 pp. | Hardback | $115.00 | €76.95 | £69.99 |AU$170.00 | ISBN: 9780123708694

Essentials of Stem Cell Biology, 2nd Edition

Human Stem Cell Manual

Robert Lanza, Roger Pedersen, John Gearhart, E. Donnall Thomas, Brigid Hogan, James Thomson, Douglas Melton and Sir Ian Wilmut June 2009 | 600 pp. | Hardback | $199.95 | €134.00 | £125.00 | AU$302.00 | ISBN: 9780123747297

Jeanne F. Loring, Robin L. Wesselschmidt and Philip H. Schwartz June 2007 | 488 pp. | Spiral bound | $88.95 | €59.95 | £53.99 |AU$132.00 | ISBN: 9780123704658

Foundations of Regenerative Medicine Clinical and Therapeutic Applications Anthony Atala, Robert Lanza, James Thomson and Robert Nerem September 2009 | 750 pp. | Hardback | $99.95 | €66.95 | £60.99|AU$148.00 | ISBN: 9780123750853

A Laboratory Guide

Handbook of Stem Cells 2-Volume Set with CD-ROM Vol. 1–2 Vol. 1 – Embryonic Stem Cells Vol. 2 – Adult & Fetal Stem Cells Robert Lanza, Roger Pedersen, Helen Blau, E. Donnall Thomas, John Gearhart, James Thomson, Brigid Hogan, Catherine Verfaillie, Douglas Melton, Irving Weissman, Malcolm Moore and Michael West September 2004 | 1,760 pp. | Hardback | $566.00 | €380.00 | £345.00 | AU$817.00 | ISBN: 9780124366435

Cell Stem Cell subscribers save 25% on their book order Secure ordering online at elsevierdirect.com Enter promo code 28024 at check out Prices and publication dates subject to change without notice.

SAN TA CR UZ B I OT ECH N O LO GY, I N C .

Leading (#1 worldwide) supplier of specialty biochemicals for proteomics research.

SANTA CRUZ

®

BIOTECHNOLOGY The Power to Question

Products include over 71,300 unique biochemicals sorted alphabetically, by CAS number, by formula and on the basis of functional activity. See our web site for product details and ordering information. www.scbt.com

HEADQUARTERS TOLL FREE: 800.457.3801 TEL: 831.457.3800 FAX: 831.457.3801 E-MAIL: [email protected]

EUROPEAN SUPPORT TOLL FREE: +00800.4573.8000 PHONE: +49.6221.4503.0 FAX: +49.6221.4503.45 E-MAIL: [email protected]

ASIAN SUPPORT JAPAN TOLL FREE: (010) 800.40402026 S. KOREA TOLL FREE: 00798.1.1.002.0297 N. CHINA TOLL FREE: 10.800.711.0752 S. CHINA TOLL FREE: 10.800.110.0694 FAX: 831.457.3801 E-MAIL: [email protected]

ChemCruz

™